2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
34 #include "mac-learning.h"
38 #include "ofp-print.h"
40 #include "ofproto-sflow.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
44 #include "openvswitch/datapath-protocol.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto
)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER
= 1
73 struct netdev
*netdev
;
74 struct ofp_phy_port opp
; /* In host byte order. */
77 static void ofport_free(struct ofport
*);
78 static void hton_ofp_phy_port(struct ofp_phy_port
*);
80 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
81 const flow_t
*flow
, struct ofproto
*ofproto
,
82 const struct ofpbuf
*packet
,
83 struct odp_actions
*out
, tag_type
*tags
,
84 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
89 uint64_t flow_cookie
; /* Controller-issued identifier.
90 (Kept in network-byte order.) */
91 uint16_t idle_timeout
; /* In seconds from time of last use. */
92 uint16_t hard_timeout
; /* In seconds from time of creation. */
93 bool send_flow_removed
; /* Send a flow removed message? */
94 long long int used
; /* Last-used time (0 if never used). */
95 long long int created
; /* Creation time. */
96 uint64_t packet_count
; /* Number of packets received. */
97 uint64_t byte_count
; /* Number of bytes received. */
98 uint64_t accounted_bytes
; /* Number of bytes passed to account_cb. */
99 tag_type tags
; /* Tags (set only by hooks). */
100 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
102 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
103 * exact-match rule (having cr.wc.wildcards of 0) generated from the
104 * wildcard rule 'super'. In this case, 'list' is an element of the
107 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
108 * a list of subrules. A super-rule with no wildcards (where
109 * cr.wc.wildcards is 0) will never have any subrules. */
115 * 'n_actions' is the number of elements in the 'actions' array. A single
116 * action may take up more more than one element's worth of space.
118 * A subrule has no actions (it uses the super-rule's actions). */
120 union ofp_action
*actions
;
124 * A super-rule with wildcard fields never has ODP actions (since the
125 * datapath only supports exact-match flows). */
126 bool installed
; /* Installed in datapath? */
127 bool may_install
; /* True ordinarily; false if actions must
128 * be reassessed for every packet. */
130 union odp_action
*odp_actions
;
134 rule_is_hidden(const struct rule
*rule
)
136 /* Subrules are merely an implementation detail, so hide them from the
138 if (rule
->super
!= NULL
) {
142 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
143 * (e.g. by in-band control) and are intentionally hidden from the
145 if (rule
->cr
.priority
> UINT16_MAX
) {
152 static struct rule
*rule_create(struct ofproto
*, struct rule
*super
,
153 const union ofp_action
*, size_t n_actions
,
154 uint16_t idle_timeout
, uint16_t hard_timeout
,
155 uint64_t flow_cookie
, bool send_flow_removed
);
156 static void rule_free(struct rule
*);
157 static void rule_destroy(struct ofproto
*, struct rule
*);
158 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
159 static void rule_insert(struct ofproto
*, struct rule
*,
160 struct ofpbuf
*packet
, uint16_t in_port
);
161 static void rule_remove(struct ofproto
*, struct rule
*);
162 static bool rule_make_actions(struct ofproto
*, struct rule
*,
163 const struct ofpbuf
*packet
);
164 static void rule_install(struct ofproto
*, struct rule
*,
165 struct rule
*displaced_rule
);
166 static void rule_uninstall(struct ofproto
*, struct rule
*);
167 static void rule_post_uninstall(struct ofproto
*, struct rule
*);
168 static void send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
169 long long int now
, uint8_t reason
);
171 /* ofproto supports two kinds of OpenFlow connections:
173 * - "Controller connections": Connections to ordinary OpenFlow controllers.
174 * ofproto maintains persistent connections to these controllers and by
175 * default sends them asynchronous messages such as packet-ins.
177 * - "Transient connections", e.g. from ovs-ofctl. When these connections
178 * drop, it is the other side's responsibility to reconnect them if
179 * necessary. ofproto does not send them asynchronous messages by default.
182 OFCONN_CONTROLLER
, /* An OpenFlow controller. */
183 OFCONN_TRANSIENT
/* A transient connection. */
186 /* An OpenFlow connection. */
188 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
189 struct list node
; /* In struct ofproto's "all_conns" list. */
190 struct rconn
*rconn
; /* OpenFlow connection. */
191 enum ofconn_type type
; /* Type. */
193 /* OFPT_PACKET_IN related data. */
194 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
195 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
196 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
197 int miss_send_len
; /* Bytes to send of buffered packets. */
199 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
200 * requests, and the maximum number before we stop reading OpenFlow
202 #define OFCONN_REPLY_MAX 100
203 struct rconn_packet_counter
*reply_counter
;
205 /* type == OFCONN_CONTROLLER only. */
206 enum nx_role role
; /* Role. */
207 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
208 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
209 struct status_category
*ss
; /* Switch status category. */
210 enum ofproto_band band
; /* In-band or out-of-band? */
213 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
214 * "schedulers" array. Their values are 0 and 1, and their meanings and values
215 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
216 * case anything ever changes, check their values here. */
217 #define N_SCHEDULERS 2
218 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
219 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
220 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
221 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
223 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
225 static void ofconn_destroy(struct ofconn
*);
226 static void ofconn_run(struct ofconn
*, struct ofproto
*);
227 static void ofconn_wait(struct ofconn
*);
228 static bool ofconn_receives_async_msgs(const struct ofconn
*);
229 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
231 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
232 struct rconn_packet_counter
*counter
);
234 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
235 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
239 uint64_t datapath_id
; /* Datapath ID. */
240 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
241 char *mfr_desc
; /* Manufacturer. */
242 char *hw_desc
; /* Hardware. */
243 char *sw_desc
; /* Software version. */
244 char *serial_desc
; /* Serial number. */
245 char *dp_desc
; /* Datapath description. */
249 struct netdev_monitor
*netdev_monitor
;
250 struct port_array ports
; /* Index is ODP port nr; ofport->opp.port_no is
252 struct shash port_by_name
;
256 struct switch_status
*switch_status
;
257 struct fail_open
*fail_open
;
258 struct netflow
*netflow
;
259 struct ofproto_sflow
*sflow
;
261 /* In-band control. */
262 struct in_band
*in_band
;
263 long long int next_in_band_update
;
264 struct sockaddr_in
*extra_in_band_remotes
;
265 size_t n_extra_remotes
;
268 struct classifier cls
;
269 bool need_revalidate
;
270 long long int next_expiration
;
271 struct tag_set revalidate_set
;
272 bool tun_id_from_cookie
;
274 /* OpenFlow connections. */
275 struct hmap controllers
; /* Controller "struct ofconn"s. */
276 struct list all_conns
; /* Contains "struct ofconn"s. */
277 enum ofproto_fail_mode fail_mode
;
278 struct pvconn
**listeners
;
280 struct pvconn
**snoops
;
283 /* Hooks for ovs-vswitchd. */
284 const struct ofhooks
*ofhooks
;
287 /* Used by default ofhooks. */
288 struct mac_learning
*ml
;
291 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
293 static const struct ofhooks default_ofhooks
;
295 static uint64_t pick_datapath_id(const struct ofproto
*);
296 static uint64_t pick_fallback_dpid(void);
298 static void update_used(struct ofproto
*);
299 static void update_stats(struct ofproto
*, struct rule
*,
300 const struct odp_flow_stats
*);
301 static void expire_rule(struct cls_rule
*, void *ofproto
);
302 static void active_timeout(struct ofproto
*ofproto
, struct rule
*rule
);
303 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
304 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
306 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
308 static void handle_openflow(struct ofconn
*, struct ofproto
*,
311 static void refresh_port_groups(struct ofproto
*);
313 static void update_port(struct ofproto
*, const char *devname
);
314 static int init_ports(struct ofproto
*);
315 static void reinit_ports(struct ofproto
*);
318 ofproto_create(const char *datapath
, const char *datapath_type
,
319 const struct ofhooks
*ofhooks
, void *aux
,
320 struct ofproto
**ofprotop
)
322 struct odp_stats stats
;
329 /* Connect to datapath and start listening for messages. */
330 error
= dpif_open(datapath
, datapath_type
, &dpif
);
332 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
335 error
= dpif_get_dp_stats(dpif
, &stats
);
337 VLOG_ERR("failed to obtain stats for datapath %s: %s",
338 datapath
, strerror(error
));
342 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
344 VLOG_ERR("failed to listen on datapath %s: %s",
345 datapath
, strerror(error
));
349 dpif_flow_flush(dpif
);
350 dpif_recv_purge(dpif
);
352 /* Initialize settings. */
353 p
= xzalloc(sizeof *p
);
354 p
->fallback_dpid
= pick_fallback_dpid();
355 p
->datapath_id
= p
->fallback_dpid
;
356 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
357 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
358 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
359 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
360 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
362 /* Initialize datapath. */
364 p
->netdev_monitor
= netdev_monitor_create();
365 port_array_init(&p
->ports
);
366 shash_init(&p
->port_by_name
);
367 p
->max_ports
= stats
.max_ports
;
369 /* Initialize submodules. */
370 p
->switch_status
= switch_status_create(p
);
376 /* Initialize flow table. */
377 classifier_init(&p
->cls
);
378 p
->need_revalidate
= false;
379 p
->next_expiration
= time_msec() + 1000;
380 tag_set_init(&p
->revalidate_set
);
382 /* Initialize OpenFlow connections. */
383 list_init(&p
->all_conns
);
384 hmap_init(&p
->controllers
);
390 /* Initialize hooks. */
392 p
->ofhooks
= ofhooks
;
396 p
->ofhooks
= &default_ofhooks
;
398 p
->ml
= mac_learning_create();
401 /* Pick final datapath ID. */
402 p
->datapath_id
= pick_datapath_id(p
);
403 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
410 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
412 uint64_t old_dpid
= p
->datapath_id
;
413 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
414 if (p
->datapath_id
!= old_dpid
) {
415 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
417 /* Force all active connections to reconnect, since there is no way to
418 * notify a controller that the datapath ID has changed. */
419 ofproto_reconnect_controllers(p
);
424 is_discovery_controller(const struct ofproto_controller
*c
)
426 return !strcmp(c
->target
, "discover");
430 is_in_band_controller(const struct ofproto_controller
*c
)
432 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
435 /* Creates a new controller in 'ofproto'. Some of the settings are initially
436 * drawn from 'c', but update_controller() needs to be called later to finish
437 * the new ofconn's configuration. */
439 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
441 struct discovery
*discovery
;
442 struct ofconn
*ofconn
;
444 if (is_discovery_controller(c
)) {
445 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
446 ofproto
->dpif
, ofproto
->switch_status
,
455 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_CONTROLLER
);
456 ofconn
->pktbuf
= pktbuf_create();
457 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
459 ofconn
->discovery
= discovery
;
461 char *name
= ofconn_make_name(ofproto
, c
->target
);
462 rconn_connect(ofconn
->rconn
, c
->target
, name
);
465 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
466 hash_string(c
->target
, 0));
469 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
470 * target or turn discovery on or off (these are done by creating new ofconns
471 * and deleting old ones), but it can update the rest of an ofconn's
474 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
476 struct ofproto
*ofproto
= ofconn
->ofproto
;
480 ofconn
->band
= (is_in_band_controller(c
)
481 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
483 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
485 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
486 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
488 if (ofconn
->discovery
) {
489 discovery_set_update_resolv_conf(ofconn
->discovery
,
490 c
->update_resolv_conf
);
491 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
494 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
495 struct pinsched
**s
= &ofconn
->schedulers
[i
];
497 if (c
->rate_limit
> 0) {
499 *s
= pinsched_create(c
->rate_limit
, c
->burst_limit
,
500 ofproto
->switch_status
);
502 pinsched_set_limits(*s
, c
->rate_limit
, c
->burst_limit
);
505 pinsched_destroy(*s
);
512 ofconn_get_target(const struct ofconn
*ofconn
)
514 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
517 static struct ofconn
*
518 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
520 struct ofconn
*ofconn
;
522 HMAP_FOR_EACH_WITH_HASH (ofconn
, struct ofconn
, hmap_node
,
523 hash_string(target
, 0), &ofproto
->controllers
) {
524 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
532 update_in_band_remotes(struct ofproto
*ofproto
)
534 const struct ofconn
*ofconn
;
535 struct sockaddr_in
*addrs
;
536 size_t max_addrs
, n_addrs
;
540 /* Allocate enough memory for as many remotes as we could possibly have. */
541 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
542 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
545 /* Add all the remotes. */
547 HMAP_FOR_EACH (ofconn
, struct ofconn
, hmap_node
, &ofproto
->controllers
) {
548 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
550 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
554 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
555 if (sin
->sin_addr
.s_addr
) {
556 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
559 if (ofconn
->discovery
) {
563 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
564 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
567 /* Create or update or destroy in-band.
569 * Ordinarily we only enable in-band if there's at least one remote
570 * address, but discovery needs the in-band rules for DHCP to be installed
571 * even before we know any remote addresses. */
572 if (n_addrs
|| discovery
) {
573 if (!ofproto
->in_band
) {
574 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
577 if (ofproto
->in_band
) {
578 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
580 ofproto
->next_in_band_update
= time_msec() + 1000;
582 in_band_destroy(ofproto
->in_band
);
583 ofproto
->in_band
= NULL
;
591 update_fail_open(struct ofproto
*p
)
593 struct ofconn
*ofconn
;
595 if (!hmap_is_empty(&p
->controllers
)
596 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
597 struct rconn
**rconns
;
601 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
605 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
606 HMAP_FOR_EACH (ofconn
, struct ofconn
, hmap_node
, &p
->controllers
) {
607 rconns
[n
++] = ofconn
->rconn
;
610 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
611 /* p->fail_open takes ownership of 'rconns'. */
613 fail_open_destroy(p
->fail_open
);
619 ofproto_set_controllers(struct ofproto
*p
,
620 const struct ofproto_controller
*controllers
,
621 size_t n_controllers
)
623 struct shash new_controllers
;
624 struct ofconn
*ofconn
, *next
;
628 shash_init(&new_controllers
);
629 for (i
= 0; i
< n_controllers
; i
++) {
630 const struct ofproto_controller
*c
= &controllers
[i
];
632 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
633 if (!find_controller_by_target(p
, c
->target
)) {
634 add_controller(p
, c
);
639 HMAP_FOR_EACH_SAFE (ofconn
, next
, struct ofconn
, hmap_node
,
641 struct ofproto_controller
*c
;
643 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
645 ofconn_destroy(ofconn
);
647 update_controller(ofconn
, c
);
653 shash_destroy(&new_controllers
);
655 update_in_band_remotes(p
);
659 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
660 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
661 struct ofconn
, hmap_node
);
662 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
663 rconn_status_cb
, ofconn
->rconn
);
668 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
670 p
->fail_mode
= fail_mode
;
674 /* Drops the connections between 'ofproto' and all of its controllers, forcing
675 * them to reconnect. */
677 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
679 struct ofconn
*ofconn
;
681 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &ofproto
->all_conns
) {
682 rconn_reconnect(ofconn
->rconn
);
687 any_extras_changed(const struct ofproto
*ofproto
,
688 const struct sockaddr_in
*extras
, size_t n
)
692 if (n
!= ofproto
->n_extra_remotes
) {
696 for (i
= 0; i
< n
; i
++) {
697 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
698 const struct sockaddr_in
*new = &extras
[i
];
700 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
701 old
->sin_port
!= new->sin_port
) {
709 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
710 * in-band control should guarantee access, in the same way that in-band
711 * control guarantees access to OpenFlow controllers. */
713 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
714 const struct sockaddr_in
*extras
, size_t n
)
716 if (!any_extras_changed(ofproto
, extras
, n
)) {
720 free(ofproto
->extra_in_band_remotes
);
721 ofproto
->n_extra_remotes
= n
;
722 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
724 update_in_band_remotes(ofproto
);
728 ofproto_set_desc(struct ofproto
*p
,
729 const char *mfr_desc
, const char *hw_desc
,
730 const char *sw_desc
, const char *serial_desc
,
733 struct ofp_desc_stats
*ods
;
736 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
737 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
738 sizeof ods
->mfr_desc
);
741 p
->mfr_desc
= xstrdup(mfr_desc
);
744 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
745 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
746 sizeof ods
->hw_desc
);
749 p
->hw_desc
= xstrdup(hw_desc
);
752 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
753 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
754 sizeof ods
->sw_desc
);
757 p
->sw_desc
= xstrdup(sw_desc
);
760 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
761 VLOG_WARN("truncating serial_desc, must be less than %zu "
763 sizeof ods
->serial_num
);
765 free(p
->serial_desc
);
766 p
->serial_desc
= xstrdup(serial_desc
);
769 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
770 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
771 sizeof ods
->dp_desc
);
774 p
->dp_desc
= xstrdup(dp_desc
);
779 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
780 const struct svec
*svec
)
782 struct pvconn
**pvconns
= *pvconnsp
;
783 size_t n_pvconns
= *n_pvconnsp
;
787 for (i
= 0; i
< n_pvconns
; i
++) {
788 pvconn_close(pvconns
[i
]);
792 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
794 for (i
= 0; i
< svec
->n
; i
++) {
795 const char *name
= svec
->names
[i
];
796 struct pvconn
*pvconn
;
799 error
= pvconn_open(name
, &pvconn
);
801 pvconns
[n_pvconns
++] = pvconn
;
803 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
811 *n_pvconnsp
= n_pvconns
;
817 ofproto_set_listeners(struct ofproto
*ofproto
, const struct svec
*listeners
)
819 return set_pvconns(&ofproto
->listeners
, &ofproto
->n_listeners
, listeners
);
823 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
825 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
829 ofproto_set_netflow(struct ofproto
*ofproto
,
830 const struct netflow_options
*nf_options
)
832 if (nf_options
&& nf_options
->collectors
.n
) {
833 if (!ofproto
->netflow
) {
834 ofproto
->netflow
= netflow_create();
836 return netflow_set_options(ofproto
->netflow
, nf_options
);
838 netflow_destroy(ofproto
->netflow
);
839 ofproto
->netflow
= NULL
;
845 ofproto_set_sflow(struct ofproto
*ofproto
,
846 const struct ofproto_sflow_options
*oso
)
848 struct ofproto_sflow
*os
= ofproto
->sflow
;
851 struct ofport
*ofport
;
852 unsigned int odp_port
;
854 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
855 refresh_port_groups(ofproto
);
856 PORT_ARRAY_FOR_EACH (ofport
, &ofproto
->ports
, odp_port
) {
857 ofproto_sflow_add_port(os
, odp_port
,
858 netdev_get_name(ofport
->netdev
));
861 ofproto_sflow_set_options(os
, oso
);
863 ofproto_sflow_destroy(os
);
864 ofproto
->sflow
= NULL
;
869 ofproto_set_stp(struct ofproto
*ofproto OVS_UNUSED
, bool enable_stp
)
873 VLOG_WARN("STP is not yet implemented");
881 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
883 return ofproto
->datapath_id
;
887 ofproto_has_controller(const struct ofproto
*ofproto
)
889 return !hmap_is_empty(&ofproto
->controllers
);
892 enum ofproto_fail_mode
893 ofproto_get_fail_mode(const struct ofproto
*p
)
899 ofproto_get_listeners(const struct ofproto
*ofproto
, struct svec
*listeners
)
903 for (i
= 0; i
< ofproto
->n_listeners
; i
++) {
904 svec_add(listeners
, pvconn_get_name(ofproto
->listeners
[i
]));
909 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
913 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
914 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
919 ofproto_destroy(struct ofproto
*p
)
921 struct ofconn
*ofconn
, *next_ofconn
;
922 struct ofport
*ofport
;
923 unsigned int port_no
;
930 /* Destroy fail-open and in-band early, since they touch the classifier. */
931 fail_open_destroy(p
->fail_open
);
934 in_band_destroy(p
->in_band
);
936 free(p
->extra_in_band_remotes
);
938 ofproto_flush_flows(p
);
939 classifier_destroy(&p
->cls
);
941 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
943 ofconn_destroy(ofconn
);
945 hmap_destroy(&p
->controllers
);
948 netdev_monitor_destroy(p
->netdev_monitor
);
949 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
952 shash_destroy(&p
->port_by_name
);
954 switch_status_destroy(p
->switch_status
);
955 netflow_destroy(p
->netflow
);
956 ofproto_sflow_destroy(p
->sflow
);
958 for (i
= 0; i
< p
->n_listeners
; i
++) {
959 pvconn_close(p
->listeners
[i
]);
963 for (i
= 0; i
< p
->n_snoops
; i
++) {
964 pvconn_close(p
->snoops
[i
]);
968 mac_learning_destroy(p
->ml
);
973 free(p
->serial_desc
);
976 port_array_destroy(&p
->ports
);
982 ofproto_run(struct ofproto
*p
)
984 int error
= ofproto_run1(p
);
986 error
= ofproto_run2(p
, false);
992 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
994 if (error
== ENOBUFS
) {
995 reinit_ports(ofproto
);
997 update_port(ofproto
, devname
);
1002 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1003 * means that 'ofconn' is more interesting for monitoring than a lower return
1006 snoop_preference(const struct ofconn
*ofconn
)
1008 switch (ofconn
->role
) {
1009 case NX_ROLE_MASTER
:
1016 /* Shouldn't happen. */
1021 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1022 * Connects this vconn to a controller. */
1024 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1026 struct ofconn
*ofconn
, *best
;
1028 /* Pick a controller for monitoring. */
1030 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &ofproto
->all_conns
) {
1031 if (ofconn
->type
== OFCONN_CONTROLLER
1032 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1038 rconn_add_monitor(best
->rconn
, vconn
);
1040 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1046 ofproto_run1(struct ofproto
*p
)
1048 struct ofconn
*ofconn
, *next_ofconn
;
1053 if (shash_is_empty(&p
->port_by_name
)) {
1057 for (i
= 0; i
< 50; i
++) {
1061 error
= dpif_recv(p
->dpif
, &buf
);
1063 if (error
== ENODEV
) {
1064 /* Someone destroyed the datapath behind our back. The caller
1065 * better destroy us and give up, because we're just going to
1066 * spin from here on out. */
1067 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1068 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1069 dpif_name(p
->dpif
));
1075 handle_odp_msg(p
, buf
);
1078 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1079 process_port_change(p
, error
, devname
);
1081 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1082 &devname
)) != EAGAIN
) {
1083 process_port_change(p
, error
, devname
);
1087 if (time_msec() >= p
->next_in_band_update
) {
1088 update_in_band_remotes(p
);
1090 in_band_run(p
->in_band
);
1093 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
1095 ofconn_run(ofconn
, p
);
1098 /* Fail-open maintenance. Do this after processing the ofconns since
1099 * fail-open checks the status of the controller rconn. */
1101 fail_open_run(p
->fail_open
);
1104 for (i
= 0; i
< p
->n_listeners
; i
++) {
1105 struct vconn
*vconn
;
1108 retval
= pvconn_accept(p
->listeners
[i
], OFP_VERSION
, &vconn
);
1110 struct rconn
*rconn
;
1113 rconn
= rconn_create(60, 0);
1114 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1115 rconn_connect_unreliably(rconn
, vconn
, name
);
1118 ofconn_create(p
, rconn
, OFCONN_TRANSIENT
);
1119 } else if (retval
!= EAGAIN
) {
1120 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1124 for (i
= 0; i
< p
->n_snoops
; i
++) {
1125 struct vconn
*vconn
;
1128 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1130 add_snooper(p
, vconn
);
1131 } else if (retval
!= EAGAIN
) {
1132 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1136 if (time_msec() >= p
->next_expiration
) {
1137 COVERAGE_INC(ofproto_expiration
);
1138 p
->next_expiration
= time_msec() + 1000;
1141 classifier_for_each(&p
->cls
, CLS_INC_ALL
, expire_rule
, p
);
1143 /* Let the hook know that we're at a stable point: all outstanding data
1144 * in existing flows has been accounted to the account_cb. Thus, the
1145 * hook can now reasonably do operations that depend on having accurate
1146 * flow volume accounting (currently, that's just bond rebalancing). */
1147 if (p
->ofhooks
->account_checkpoint_cb
) {
1148 p
->ofhooks
->account_checkpoint_cb(p
->aux
);
1153 netflow_run(p
->netflow
);
1156 ofproto_sflow_run(p
->sflow
);
1162 struct revalidate_cbdata
{
1163 struct ofproto
*ofproto
;
1164 bool revalidate_all
; /* Revalidate all exact-match rules? */
1165 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
1166 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
1170 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1172 if (p
->need_revalidate
|| revalidate_all
1173 || !tag_set_is_empty(&p
->revalidate_set
)) {
1174 struct revalidate_cbdata cbdata
;
1176 cbdata
.revalidate_all
= revalidate_all
;
1177 cbdata
.revalidate_subrules
= p
->need_revalidate
;
1178 cbdata
.revalidate_set
= p
->revalidate_set
;
1179 tag_set_init(&p
->revalidate_set
);
1180 COVERAGE_INC(ofproto_revalidate
);
1181 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
1182 p
->need_revalidate
= false;
1189 ofproto_wait(struct ofproto
*p
)
1191 struct ofconn
*ofconn
;
1194 dpif_recv_wait(p
->dpif
);
1195 dpif_port_poll_wait(p
->dpif
);
1196 netdev_monitor_poll_wait(p
->netdev_monitor
);
1197 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
1198 ofconn_wait(ofconn
);
1201 poll_timer_wait_until(p
->next_in_band_update
);
1202 in_band_wait(p
->in_band
);
1205 fail_open_wait(p
->fail_open
);
1208 ofproto_sflow_wait(p
->sflow
);
1210 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1211 poll_immediate_wake();
1213 if (p
->need_revalidate
) {
1214 /* Shouldn't happen, but if it does just go around again. */
1215 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1216 poll_immediate_wake();
1217 } else if (p
->next_expiration
!= LLONG_MAX
) {
1218 poll_timer_wait_until(p
->next_expiration
);
1220 for (i
= 0; i
< p
->n_listeners
; i
++) {
1221 pvconn_wait(p
->listeners
[i
]);
1223 for (i
= 0; i
< p
->n_snoops
; i
++) {
1224 pvconn_wait(p
->snoops
[i
]);
1229 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1231 tag_set_add(&ofproto
->revalidate_set
, tag
);
1235 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1237 return &ofproto
->revalidate_set
;
1241 ofproto_is_alive(const struct ofproto
*p
)
1243 return !hmap_is_empty(&p
->controllers
);
1247 ofproto_send_packet(struct ofproto
*p
, const flow_t
*flow
,
1248 const union ofp_action
*actions
, size_t n_actions
,
1249 const struct ofpbuf
*packet
)
1251 struct odp_actions odp_actions
;
1254 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1260 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1262 dpif_execute(p
->dpif
, flow
->in_port
, odp_actions
.actions
,
1263 odp_actions
.n_actions
, packet
);
1268 ofproto_add_flow(struct ofproto
*p
,
1269 const flow_t
*flow
, uint32_t wildcards
, unsigned int priority
,
1270 const union ofp_action
*actions
, size_t n_actions
,
1274 rule
= rule_create(p
, NULL
, actions
, n_actions
,
1275 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */,
1277 cls_rule_from_flow(flow
, wildcards
, priority
, &rule
->cr
);
1278 rule_insert(p
, rule
, NULL
, 0);
1282 ofproto_delete_flow(struct ofproto
*ofproto
, const flow_t
*flow
,
1283 uint32_t wildcards
, unsigned int priority
)
1287 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1291 rule_remove(ofproto
, rule
);
1296 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1298 struct rule
*rule
= rule_from_cls_rule(rule_
);
1299 struct ofproto
*ofproto
= ofproto_
;
1301 /* Mark the flow as not installed, even though it might really be
1302 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1303 * There is no point in uninstalling it individually since we are about to
1304 * blow away all the flows with dpif_flow_flush(). */
1305 rule
->installed
= false;
1307 rule_remove(ofproto
, rule
);
1311 ofproto_flush_flows(struct ofproto
*ofproto
)
1313 COVERAGE_INC(ofproto_flush
);
1314 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1315 dpif_flow_flush(ofproto
->dpif
);
1316 if (ofproto
->in_band
) {
1317 in_band_flushed(ofproto
->in_band
);
1319 if (ofproto
->fail_open
) {
1320 fail_open_flushed(ofproto
->fail_open
);
1325 reinit_ports(struct ofproto
*p
)
1327 struct svec devnames
;
1328 struct ofport
*ofport
;
1329 unsigned int port_no
;
1330 struct odp_port
*odp_ports
;
1334 svec_init(&devnames
);
1335 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
1336 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1338 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1339 for (i
= 0; i
< n_odp_ports
; i
++) {
1340 svec_add (&devnames
, odp_ports
[i
].devname
);
1344 svec_sort_unique(&devnames
);
1345 for (i
= 0; i
< devnames
.n
; i
++) {
1346 update_port(p
, devnames
.names
[i
]);
1348 svec_destroy(&devnames
);
1352 refresh_port_group(struct ofproto
*p
, unsigned int group
)
1356 struct ofport
*port
;
1357 unsigned int port_no
;
1359 assert(group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
);
1361 ports
= xmalloc(port_array_count(&p
->ports
) * sizeof *ports
);
1363 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
1364 if (group
== DP_GROUP_ALL
|| !(port
->opp
.config
& OFPPC_NO_FLOOD
)) {
1365 ports
[n_ports
++] = port_no
;
1368 dpif_port_group_set(p
->dpif
, group
, ports
, n_ports
);
1375 refresh_port_groups(struct ofproto
*p
)
1377 size_t n_flood
= refresh_port_group(p
, DP_GROUP_FLOOD
);
1378 size_t n_all
= refresh_port_group(p
, DP_GROUP_ALL
);
1380 ofproto_sflow_set_group_sizes(p
->sflow
, n_flood
, n_all
);
1384 static struct ofport
*
1385 make_ofport(const struct odp_port
*odp_port
)
1387 struct netdev_options netdev_options
;
1388 enum netdev_flags flags
;
1389 struct ofport
*ofport
;
1390 struct netdev
*netdev
;
1394 memset(&netdev_options
, 0, sizeof netdev_options
);
1395 netdev_options
.name
= odp_port
->devname
;
1396 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1398 error
= netdev_open(&netdev_options
, &netdev
);
1400 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1401 "cannot be opened (%s)",
1402 odp_port
->devname
, odp_port
->port
,
1403 odp_port
->devname
, strerror(error
));
1407 ofport
= xmalloc(sizeof *ofport
);
1408 ofport
->netdev
= netdev
;
1409 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1410 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1411 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1412 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1413 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1415 netdev_get_flags(netdev
, &flags
);
1416 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1418 netdev_get_carrier(netdev
, &carrier
);
1419 ofport
->opp
.state
= carrier
? 0 : OFPPS_LINK_DOWN
;
1421 netdev_get_features(netdev
,
1422 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1423 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1428 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1430 if (port_array_get(&p
->ports
, odp_port
->port
)) {
1431 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1434 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1435 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1444 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1446 const struct ofp_phy_port
*a
= &a_
->opp
;
1447 const struct ofp_phy_port
*b
= &b_
->opp
;
1449 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1450 return (a
->port_no
== b
->port_no
1451 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1452 && !strcmp((char *) a
->name
, (char *) b
->name
)
1453 && a
->state
== b
->state
1454 && a
->config
== b
->config
1455 && a
->curr
== b
->curr
1456 && a
->advertised
== b
->advertised
1457 && a
->supported
== b
->supported
1458 && a
->peer
== b
->peer
);
1462 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1465 /* XXX Should limit the number of queued port status change messages. */
1466 struct ofconn
*ofconn
;
1467 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
1468 struct ofp_port_status
*ops
;
1471 if (!ofconn_receives_async_msgs(ofconn
)) {
1475 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1476 ops
->reason
= reason
;
1477 ops
->desc
= ofport
->opp
;
1478 hton_ofp_phy_port(&ops
->desc
);
1479 queue_tx(b
, ofconn
, NULL
);
1481 if (p
->ofhooks
->port_changed_cb
) {
1482 p
->ofhooks
->port_changed_cb(reason
, &ofport
->opp
, p
->aux
);
1487 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1489 uint16_t odp_port
= ofp_port_to_odp_port(ofport
->opp
.port_no
);
1490 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1492 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1493 port_array_set(&p
->ports
, odp_port
, ofport
);
1494 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1496 ofproto_sflow_add_port(p
->sflow
, odp_port
, netdev_name
);
1501 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1503 uint16_t odp_port
= ofp_port_to_odp_port(ofport
->opp
.port_no
);
1505 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1506 port_array_delete(&p
->ports
, odp_port
);
1507 shash_delete(&p
->port_by_name
,
1508 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1510 ofproto_sflow_del_port(p
->sflow
, odp_port
);
1515 ofport_free(struct ofport
*ofport
)
1518 netdev_close(ofport
->netdev
);
1524 update_port(struct ofproto
*p
, const char *devname
)
1526 struct odp_port odp_port
;
1527 struct ofport
*old_ofport
;
1528 struct ofport
*new_ofport
;
1531 COVERAGE_INC(ofproto_update_port
);
1533 /* Query the datapath for port information. */
1534 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1536 /* Find the old ofport. */
1537 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1540 /* There's no port named 'devname' but there might be a port with
1541 * the same port number. This could happen if a port is deleted
1542 * and then a new one added in its place very quickly, or if a port
1543 * is renamed. In the former case we want to send an OFPPR_DELETE
1544 * and an OFPPR_ADD, and in the latter case we want to send a
1545 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1546 * the old port's ifindex against the new port, or perhaps less
1547 * reliably but more portably by comparing the old port's MAC
1548 * against the new port's MAC. However, this code isn't that smart
1549 * and always sends an OFPPR_MODIFY (XXX). */
1550 old_ofport
= port_array_get(&p
->ports
, odp_port
.port
);
1552 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1553 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1554 "%s", strerror(error
));
1558 /* Create a new ofport. */
1559 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1561 /* Eliminate a few pathological cases. */
1562 if (!old_ofport
&& !new_ofport
) {
1564 } else if (old_ofport
&& new_ofport
) {
1565 /* Most of the 'config' bits are OpenFlow soft state, but
1566 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1567 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1568 * leaves the other bits 0.) */
1569 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1571 if (ofport_equal(old_ofport
, new_ofport
)) {
1572 /* False alarm--no change. */
1573 ofport_free(new_ofport
);
1578 /* Now deal with the normal cases. */
1580 ofport_remove(p
, old_ofport
);
1583 ofport_install(p
, new_ofport
);
1585 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1586 (!old_ofport
? OFPPR_ADD
1587 : !new_ofport
? OFPPR_DELETE
1589 ofport_free(old_ofport
);
1591 /* Update port groups. */
1592 refresh_port_groups(p
);
1596 init_ports(struct ofproto
*p
)
1598 struct odp_port
*ports
;
1603 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1608 for (i
= 0; i
< n_ports
; i
++) {
1609 const struct odp_port
*odp_port
= &ports
[i
];
1610 if (!ofport_conflicts(p
, odp_port
)) {
1611 struct ofport
*ofport
= make_ofport(odp_port
);
1613 ofport_install(p
, ofport
);
1618 refresh_port_groups(p
);
1622 static struct ofconn
*
1623 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1625 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1626 ofconn
->ofproto
= p
;
1627 list_push_back(&p
->all_conns
, &ofconn
->node
);
1628 ofconn
->rconn
= rconn
;
1629 ofconn
->type
= type
;
1630 ofconn
->role
= NX_ROLE_OTHER
;
1631 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1632 ofconn
->pktbuf
= NULL
;
1633 ofconn
->miss_send_len
= 0;
1634 ofconn
->reply_counter
= rconn_packet_counter_create ();
1639 ofconn_destroy(struct ofconn
*ofconn
)
1641 if (ofconn
->type
== OFCONN_CONTROLLER
) {
1642 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1644 discovery_destroy(ofconn
->discovery
);
1646 list_remove(&ofconn
->node
);
1647 switch_status_unregister(ofconn
->ss
);
1648 rconn_destroy(ofconn
->rconn
);
1649 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1650 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1651 pktbuf_destroy(ofconn
->pktbuf
);
1656 ofconn_run(struct ofconn
*ofconn
, struct ofproto
*p
)
1661 if (ofconn
->discovery
) {
1662 char *controller_name
;
1663 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1664 discovery_question_connectivity(ofconn
->discovery
);
1666 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1667 if (controller_name
) {
1668 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1669 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1672 rconn_disconnect(ofconn
->rconn
);
1677 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1678 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1681 rconn_run(ofconn
->rconn
);
1683 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1684 /* Limit the number of iterations to prevent other tasks from
1686 for (iteration
= 0; iteration
< 50; iteration
++) {
1687 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1692 fail_open_maybe_recover(p
->fail_open
);
1694 handle_openflow(ofconn
, p
, of_msg
);
1695 ofpbuf_delete(of_msg
);
1699 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1700 ofconn_destroy(ofconn
);
1705 ofconn_wait(struct ofconn
*ofconn
)
1709 if (ofconn
->discovery
) {
1710 discovery_wait(ofconn
->discovery
);
1712 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1713 pinsched_wait(ofconn
->schedulers
[i
]);
1715 rconn_run_wait(ofconn
->rconn
);
1716 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1717 rconn_recv_wait(ofconn
->rconn
);
1719 COVERAGE_INC(ofproto_ofconn_stuck
);
1723 /* Returns true if 'ofconn' should receive asynchronous messages. */
1725 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1727 if (ofconn
->type
== OFCONN_CONTROLLER
) {
1728 /* Ordinary controllers always get asynchronous messages unless they
1729 * have configured themselves as "slaves". */
1730 return ofconn
->role
!= NX_ROLE_SLAVE
;
1732 /* Transient connections don't get asynchronous messages unless they
1733 * have explicitly asked for them by setting a nonzero miss send
1735 return ofconn
->miss_send_len
> 0;
1739 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1740 * and 'target', suitable for use in log messages for identifying the
1743 * The name is dynamically allocated. The caller should free it (with free())
1744 * when it is no longer needed. */
1746 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1748 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1751 /* Caller is responsible for initializing the 'cr' member of the returned
1753 static struct rule
*
1754 rule_create(struct ofproto
*ofproto
, struct rule
*super
,
1755 const union ofp_action
*actions
, size_t n_actions
,
1756 uint16_t idle_timeout
, uint16_t hard_timeout
,
1757 uint64_t flow_cookie
, bool send_flow_removed
)
1759 struct rule
*rule
= xzalloc(sizeof *rule
);
1760 rule
->idle_timeout
= idle_timeout
;
1761 rule
->hard_timeout
= hard_timeout
;
1762 rule
->flow_cookie
= flow_cookie
;
1763 rule
->used
= rule
->created
= time_msec();
1764 rule
->send_flow_removed
= send_flow_removed
;
1765 rule
->super
= super
;
1767 list_push_back(&super
->list
, &rule
->list
);
1769 list_init(&rule
->list
);
1771 rule
->n_actions
= n_actions
;
1772 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1773 netflow_flow_clear(&rule
->nf_flow
);
1774 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->created
);
1779 static struct rule
*
1780 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1782 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1786 rule_free(struct rule
*rule
)
1788 free(rule
->actions
);
1789 free(rule
->odp_actions
);
1793 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1794 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1795 * through all of its subrules and revalidates them, destroying any that no
1796 * longer has a super-rule (which is probably all of them).
1798 * Before calling this function, the caller must make have removed 'rule' from
1799 * the classifier. If 'rule' is an exact-match rule, the caller is also
1800 * responsible for ensuring that it has been uninstalled from the datapath. */
1802 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1805 struct rule
*subrule
, *next
;
1806 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
1807 revalidate_rule(ofproto
, subrule
);
1810 list_remove(&rule
->list
);
1816 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1818 const union ofp_action
*oa
;
1819 struct actions_iterator i
;
1821 if (out_port
== htons(OFPP_NONE
)) {
1824 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1825 oa
= actions_next(&i
)) {
1826 if (action_outputs_to_port(oa
, out_port
)) {
1833 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1834 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1836 * The flow that 'packet' actually contains does not need to actually match
1837 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1838 * the packet and byte counters for 'rule' will be credited for the packet sent
1839 * out whether or not the packet actually matches 'rule'.
1841 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1842 * the caller must already have accurately composed ODP actions for it given
1843 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1844 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1845 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1846 * actions and apply them to 'packet'. */
1848 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1849 struct ofpbuf
*packet
, const flow_t
*flow
)
1851 const union odp_action
*actions
;
1853 struct odp_actions a
;
1855 /* Grab or compose the ODP actions.
1857 * The special case for an exact-match 'rule' where 'flow' is not the
1858 * rule's flow is important to avoid, e.g., sending a packet out its input
1859 * port simply because the ODP actions were composed for the wrong
1861 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
1862 struct rule
*super
= rule
->super
? rule
->super
: rule
;
1863 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
1864 packet
, &a
, NULL
, 0, NULL
)) {
1867 actions
= a
.actions
;
1868 n_actions
= a
.n_actions
;
1870 actions
= rule
->odp_actions
;
1871 n_actions
= rule
->n_odp_actions
;
1874 /* Execute the ODP actions. */
1875 if (!dpif_execute(ofproto
->dpif
, flow
->in_port
,
1876 actions
, n_actions
, packet
)) {
1877 struct odp_flow_stats stats
;
1878 flow_extract_stats(flow
, packet
, &stats
);
1879 update_stats(ofproto
, rule
, &stats
);
1880 rule
->used
= time_msec();
1881 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->used
);
1886 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
1889 struct rule
*displaced_rule
;
1891 /* Insert the rule in the classifier. */
1892 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
1893 if (!rule
->cr
.wc
.wildcards
) {
1894 rule_make_actions(p
, rule
, packet
);
1897 /* Send the packet and credit it to the rule. */
1900 flow_extract(packet
, 0, in_port
, &flow
);
1901 rule_execute(p
, rule
, packet
, &flow
);
1904 /* Install the rule in the datapath only after sending the packet, to
1905 * avoid packet reordering. */
1906 if (rule
->cr
.wc
.wildcards
) {
1907 COVERAGE_INC(ofproto_add_wc_flow
);
1908 p
->need_revalidate
= true;
1910 rule_install(p
, rule
, displaced_rule
);
1913 /* Free the rule that was displaced, if any. */
1914 if (displaced_rule
) {
1915 rule_destroy(p
, displaced_rule
);
1919 static struct rule
*
1920 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
1923 struct rule
*subrule
= rule_create(ofproto
, rule
, NULL
, 0,
1924 rule
->idle_timeout
, rule
->hard_timeout
,
1926 COVERAGE_INC(ofproto_subrule_create
);
1927 cls_rule_from_flow(flow
, 0, (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
1928 : rule
->cr
.priority
), &subrule
->cr
);
1929 classifier_insert_exact(&ofproto
->cls
, &subrule
->cr
);
1935 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
1937 if (rule
->cr
.wc
.wildcards
) {
1938 COVERAGE_INC(ofproto_del_wc_flow
);
1939 ofproto
->need_revalidate
= true;
1941 rule_uninstall(ofproto
, rule
);
1943 classifier_remove(&ofproto
->cls
, &rule
->cr
);
1944 rule_destroy(ofproto
, rule
);
1947 /* Returns true if the actions changed, false otherwise. */
1949 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
1950 const struct ofpbuf
*packet
)
1952 const struct rule
*super
;
1953 struct odp_actions a
;
1956 assert(!rule
->cr
.wc
.wildcards
);
1958 super
= rule
->super
? rule
->super
: rule
;
1960 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
1961 packet
, &a
, &rule
->tags
, &rule
->may_install
,
1962 &rule
->nf_flow
.output_iface
);
1964 actions_len
= a
.n_actions
* sizeof *a
.actions
;
1965 if (rule
->n_odp_actions
!= a
.n_actions
1966 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
1967 COVERAGE_INC(ofproto_odp_unchanged
);
1968 free(rule
->odp_actions
);
1969 rule
->n_odp_actions
= a
.n_actions
;
1970 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
1978 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
1979 struct odp_flow_put
*put
)
1981 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
1982 put
->flow
.key
= rule
->cr
.flow
;
1983 put
->flow
.actions
= rule
->odp_actions
;
1984 put
->flow
.n_actions
= rule
->n_odp_actions
;
1985 put
->flow
.flags
= 0;
1987 return dpif_flow_put(ofproto
->dpif
, put
);
1991 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
1993 assert(!rule
->cr
.wc
.wildcards
);
1995 if (rule
->may_install
) {
1996 struct odp_flow_put put
;
1997 if (!do_put_flow(p
, rule
,
1998 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
2000 rule
->installed
= true;
2001 if (displaced_rule
) {
2002 update_stats(p
, displaced_rule
, &put
.flow
.stats
);
2003 rule_post_uninstall(p
, displaced_rule
);
2006 } else if (displaced_rule
) {
2007 rule_uninstall(p
, displaced_rule
);
2012 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
2014 if (rule
->installed
) {
2015 struct odp_flow_put put
;
2016 COVERAGE_INC(ofproto_dp_missed
);
2017 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
2019 rule_install(ofproto
, rule
, NULL
);
2024 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
2026 bool actions_changed
;
2027 uint16_t new_out_iface
, old_out_iface
;
2029 old_out_iface
= rule
->nf_flow
.output_iface
;
2030 actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
2032 if (rule
->may_install
) {
2033 if (rule
->installed
) {
2034 if (actions_changed
) {
2035 struct odp_flow_put put
;
2036 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
2037 | ODPPF_ZERO_STATS
, &put
);
2038 update_stats(ofproto
, rule
, &put
.flow
.stats
);
2040 /* Temporarily set the old output iface so that NetFlow
2041 * messages have the correct output interface for the old
2043 new_out_iface
= rule
->nf_flow
.output_iface
;
2044 rule
->nf_flow
.output_iface
= old_out_iface
;
2045 rule_post_uninstall(ofproto
, rule
);
2046 rule
->nf_flow
.output_iface
= new_out_iface
;
2049 rule_install(ofproto
, rule
, NULL
);
2052 rule_uninstall(ofproto
, rule
);
2057 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
2059 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
2061 if (ofproto
->ofhooks
->account_flow_cb
2062 && total_bytes
> rule
->accounted_bytes
)
2064 ofproto
->ofhooks
->account_flow_cb(
2065 &rule
->cr
.flow
, rule
->tags
, rule
->odp_actions
, rule
->n_odp_actions
,
2066 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
2067 rule
->accounted_bytes
= total_bytes
;
2072 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
2074 assert(!rule
->cr
.wc
.wildcards
);
2075 if (rule
->installed
) {
2076 struct odp_flow odp_flow
;
2078 odp_flow
.key
= rule
->cr
.flow
;
2079 odp_flow
.actions
= NULL
;
2080 odp_flow
.n_actions
= 0;
2082 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2083 update_stats(p
, rule
, &odp_flow
.stats
);
2085 rule
->installed
= false;
2087 rule_post_uninstall(p
, rule
);
2092 is_controller_rule(struct rule
*rule
)
2094 /* If the only action is send to the controller then don't report
2095 * NetFlow expiration messages since it is just part of the control
2096 * logic for the network and not real traffic. */
2100 && rule
->super
->n_actions
== 1
2101 && action_outputs_to_port(&rule
->super
->actions
[0],
2102 htons(OFPP_CONTROLLER
)));
2106 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
2108 struct rule
*super
= rule
->super
;
2110 rule_account(ofproto
, rule
, 0);
2112 if (ofproto
->netflow
&& !is_controller_rule(rule
)) {
2113 struct ofexpired expired
;
2114 expired
.flow
= rule
->cr
.flow
;
2115 expired
.packet_count
= rule
->packet_count
;
2116 expired
.byte_count
= rule
->byte_count
;
2117 expired
.used
= rule
->used
;
2118 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
2121 super
->packet_count
+= rule
->packet_count
;
2122 super
->byte_count
+= rule
->byte_count
;
2124 /* Reset counters to prevent double counting if the rule ever gets
2126 rule
->packet_count
= 0;
2127 rule
->byte_count
= 0;
2128 rule
->accounted_bytes
= 0;
2130 netflow_flow_clear(&rule
->nf_flow
);
2135 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2136 struct rconn_packet_counter
*counter
)
2138 update_openflow_length(msg
);
2139 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2145 send_error(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2146 int error
, const void *data
, size_t len
)
2149 struct ofp_error_msg
*oem
;
2151 if (!(error
>> 16)) {
2152 VLOG_WARN_RL(&rl
, "not sending bad error code %d to controller",
2157 COVERAGE_INC(ofproto_error
);
2158 oem
= make_openflow_xid(len
+ sizeof *oem
, OFPT_ERROR
,
2159 oh
? oh
->xid
: 0, &buf
);
2160 oem
->type
= htons((unsigned int) error
>> 16);
2161 oem
->code
= htons(error
& 0xffff);
2162 memcpy(oem
->data
, data
, len
);
2163 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2167 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2170 size_t oh_length
= ntohs(oh
->length
);
2171 send_error(ofconn
, oh
, error
, oh
, MIN(oh_length
, 64));
2175 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2177 opp
->port_no
= htons(opp
->port_no
);
2178 opp
->config
= htonl(opp
->config
);
2179 opp
->state
= htonl(opp
->state
);
2180 opp
->curr
= htonl(opp
->curr
);
2181 opp
->advertised
= htonl(opp
->advertised
);
2182 opp
->supported
= htonl(opp
->supported
);
2183 opp
->peer
= htonl(opp
->peer
);
2187 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2189 struct ofp_header
*rq
= oh
;
2190 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2195 handle_features_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2196 struct ofp_header
*oh
)
2198 struct ofp_switch_features
*osf
;
2200 unsigned int port_no
;
2201 struct ofport
*port
;
2203 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2204 osf
->datapath_id
= htonll(p
->datapath_id
);
2205 osf
->n_buffers
= htonl(pktbuf_capacity());
2207 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2208 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2209 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2210 (1u << OFPAT_SET_VLAN_VID
) |
2211 (1u << OFPAT_SET_VLAN_PCP
) |
2212 (1u << OFPAT_STRIP_VLAN
) |
2213 (1u << OFPAT_SET_DL_SRC
) |
2214 (1u << OFPAT_SET_DL_DST
) |
2215 (1u << OFPAT_SET_NW_SRC
) |
2216 (1u << OFPAT_SET_NW_DST
) |
2217 (1u << OFPAT_SET_NW_TOS
) |
2218 (1u << OFPAT_SET_TP_SRC
) |
2219 (1u << OFPAT_SET_TP_DST
) |
2220 (1u << OFPAT_ENQUEUE
));
2222 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
2223 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2226 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2231 handle_get_config_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2232 struct ofp_header
*oh
)
2235 struct ofp_switch_config
*osc
;
2239 /* Figure out flags. */
2240 dpif_get_drop_frags(p
->dpif
, &drop_frags
);
2241 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2244 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2245 osc
->flags
= htons(flags
);
2246 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2247 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2253 handle_set_config(struct ofproto
*p
, struct ofconn
*ofconn
,
2254 struct ofp_switch_config
*osc
)
2259 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2263 flags
= ntohs(osc
->flags
);
2265 if (ofconn
->type
== OFCONN_CONTROLLER
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2266 switch (flags
& OFPC_FRAG_MASK
) {
2267 case OFPC_FRAG_NORMAL
:
2268 dpif_set_drop_frags(p
->dpif
, false);
2270 case OFPC_FRAG_DROP
:
2271 dpif_set_drop_frags(p
->dpif
, true);
2274 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2280 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2286 add_output_group_action(struct odp_actions
*actions
, uint16_t group
,
2287 uint16_t *nf_output_iface
)
2289 odp_actions_add(actions
, ODPAT_OUTPUT_GROUP
)->output_group
.group
= group
;
2291 if (group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
) {
2292 *nf_output_iface
= NF_OUT_FLOOD
;
2297 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2299 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2300 a
->controller
.arg
= max_len
;
2303 struct action_xlate_ctx
{
2305 flow_t flow
; /* Flow to which these actions correspond. */
2306 int recurse
; /* Recursion level, via xlate_table_action. */
2307 struct ofproto
*ofproto
;
2308 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2309 * null pointer if we are revalidating
2310 * without a packet to refer to. */
2313 struct odp_actions
*out
; /* Datapath actions. */
2314 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
2315 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2316 * be reassessed for every packet. */
2317 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2320 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2321 struct action_xlate_ctx
*ctx
);
2324 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2326 const struct ofport
*ofport
= port_array_get(&ctx
->ofproto
->ports
, port
);
2329 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2330 /* Forwarding disabled on port. */
2335 * We don't have an ofport record for this port, but it doesn't hurt to
2336 * allow forwarding to it anyhow. Maybe such a port will appear later
2337 * and we're pre-populating the flow table.
2341 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2342 ctx
->nf_output_iface
= port
;
2345 static struct rule
*
2346 lookup_valid_rule(struct ofproto
*ofproto
, const flow_t
*flow
)
2349 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2351 /* The rule we found might not be valid, since we could be in need of
2352 * revalidation. If it is not valid, don't return it. */
2355 && ofproto
->need_revalidate
2356 && !revalidate_rule(ofproto
, rule
)) {
2357 COVERAGE_INC(ofproto_invalidated
);
2365 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2367 if (!ctx
->recurse
) {
2368 uint16_t old_in_port
;
2371 /* Look up a flow with 'in_port' as the input port. Then restore the
2372 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2373 * have surprising behavior). */
2374 old_in_port
= ctx
->flow
.in_port
;
2375 ctx
->flow
.in_port
= in_port
;
2376 rule
= lookup_valid_rule(ctx
->ofproto
, &ctx
->flow
);
2377 ctx
->flow
.in_port
= old_in_port
;
2385 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2392 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2393 uint16_t port
, uint16_t max_len
)
2396 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2398 ctx
->nf_output_iface
= NF_OUT_DROP
;
2402 add_output_action(ctx
, ctx
->flow
.in_port
);
2405 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2408 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2409 ctx
->out
, ctx
->tags
,
2410 &ctx
->nf_output_iface
,
2411 ctx
->ofproto
->aux
)) {
2412 COVERAGE_INC(ofproto_uninstallable
);
2413 ctx
->may_set_up_flow
= false;
2417 add_output_group_action(ctx
->out
, DP_GROUP_FLOOD
,
2418 &ctx
->nf_output_iface
);
2421 add_output_group_action(ctx
->out
, DP_GROUP_ALL
, &ctx
->nf_output_iface
);
2423 case OFPP_CONTROLLER
:
2424 add_controller_action(ctx
->out
, max_len
);
2427 add_output_action(ctx
, ODPP_LOCAL
);
2430 odp_port
= ofp_port_to_odp_port(port
);
2431 if (odp_port
!= ctx
->flow
.in_port
) {
2432 add_output_action(ctx
, odp_port
);
2437 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2438 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2439 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2440 ctx
->nf_output_iface
= prev_nf_output_iface
;
2441 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2442 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2443 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2448 xlate_output_action(struct action_xlate_ctx
*ctx
,
2449 const struct ofp_action_output
*oao
)
2451 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2454 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2455 * optimization, because we're going to add another action that sets the
2456 * priority immediately after, or because there are no actions following the
2459 remove_pop_action(struct action_xlate_ctx
*ctx
)
2461 size_t n
= ctx
->out
->n_actions
;
2462 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2463 ctx
->out
->n_actions
--;
2468 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2469 const struct ofp_action_enqueue
*oae
)
2471 uint16_t ofp_port
, odp_port
;
2475 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2478 /* Fall back to ordinary output action. */
2479 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2483 /* Figure out ODP output port. */
2484 ofp_port
= ntohs(oae
->port
);
2485 if (ofp_port
!= OFPP_IN_PORT
) {
2486 odp_port
= ofp_port_to_odp_port(ofp_port
);
2488 odp_port
= ctx
->flow
.in_port
;
2491 /* Add ODP actions. */
2492 remove_pop_action(ctx
);
2493 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2495 add_output_action(ctx
, odp_port
);
2496 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2498 /* Update NetFlow output port. */
2499 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2500 ctx
->nf_output_iface
= odp_port
;
2501 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2502 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2507 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2508 const struct nx_action_header
*nah
)
2510 const struct nx_action_resubmit
*nar
;
2511 const struct nx_action_set_tunnel
*nast
;
2512 union odp_action
*oa
;
2513 int subtype
= ntohs(nah
->subtype
);
2515 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2517 case NXAST_RESUBMIT
:
2518 nar
= (const struct nx_action_resubmit
*) nah
;
2519 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2522 case NXAST_SET_TUNNEL
:
2523 nast
= (const struct nx_action_set_tunnel
*) nah
;
2524 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2525 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2528 /* If you add a new action here that modifies flow data, don't forget to
2529 * update the flow key in ctx->flow at the same time. */
2532 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2538 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2539 struct action_xlate_ctx
*ctx
)
2541 struct actions_iterator iter
;
2542 const union ofp_action
*ia
;
2543 const struct ofport
*port
;
2545 port
= port_array_get(&ctx
->ofproto
->ports
, ctx
->flow
.in_port
);
2546 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2547 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, stp_eth_addr
)
2548 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2549 /* Drop this flow. */
2553 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2554 uint16_t type
= ntohs(ia
->type
);
2555 union odp_action
*oa
;
2559 xlate_output_action(ctx
, &ia
->output
);
2562 case OFPAT_SET_VLAN_VID
:
2563 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_VID
);
2564 ctx
->flow
.dl_vlan
= oa
->vlan_vid
.vlan_vid
= ia
->vlan_vid
.vlan_vid
;
2567 case OFPAT_SET_VLAN_PCP
:
2568 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_PCP
);
2569 ctx
->flow
.dl_vlan_pcp
= oa
->vlan_pcp
.vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2572 case OFPAT_STRIP_VLAN
:
2573 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2574 ctx
->flow
.dl_vlan
= htons(OFP_VLAN_NONE
);
2575 ctx
->flow
.dl_vlan_pcp
= 0;
2578 case OFPAT_SET_DL_SRC
:
2579 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2580 memcpy(oa
->dl_addr
.dl_addr
,
2581 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2582 memcpy(ctx
->flow
.dl_src
,
2583 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2586 case OFPAT_SET_DL_DST
:
2587 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2588 memcpy(oa
->dl_addr
.dl_addr
,
2589 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2590 memcpy(ctx
->flow
.dl_dst
,
2591 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2594 case OFPAT_SET_NW_SRC
:
2595 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2596 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2599 case OFPAT_SET_NW_DST
:
2600 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2601 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2604 case OFPAT_SET_NW_TOS
:
2605 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2606 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2609 case OFPAT_SET_TP_SRC
:
2610 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2611 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2614 case OFPAT_SET_TP_DST
:
2615 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2616 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2620 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2624 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2628 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2635 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2636 const flow_t
*flow
, struct ofproto
*ofproto
,
2637 const struct ofpbuf
*packet
,
2638 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2639 uint16_t *nf_output_iface
)
2641 tag_type no_tags
= 0;
2642 struct action_xlate_ctx ctx
;
2643 COVERAGE_INC(ofproto_ofp2odp
);
2644 odp_actions_init(out
);
2647 ctx
.ofproto
= ofproto
;
2648 ctx
.packet
= packet
;
2650 ctx
.tags
= tags
? tags
: &no_tags
;
2651 ctx
.may_set_up_flow
= true;
2652 ctx
.nf_output_iface
= NF_OUT_DROP
;
2653 do_xlate_actions(in
, n_in
, &ctx
);
2654 remove_pop_action(&ctx
);
2656 /* Check with in-band control to see if we're allowed to set up this
2658 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2659 ctx
.may_set_up_flow
= false;
2662 if (may_set_up_flow
) {
2663 *may_set_up_flow
= ctx
.may_set_up_flow
;
2665 if (nf_output_iface
) {
2666 *nf_output_iface
= ctx
.nf_output_iface
;
2668 if (odp_actions_overflow(out
)) {
2669 odp_actions_init(out
);
2670 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
2675 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2676 * error message code (composed with ofp_mkerr()) for the caller to propagate
2677 * upward. Otherwise, returns 0.
2679 * 'oh' is used to make log messages more informative. */
2681 reject_slave_controller(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2683 if (ofconn
->type
== OFCONN_CONTROLLER
&& ofconn
->role
== NX_ROLE_SLAVE
) {
2684 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2687 type_name
= ofp_message_type_to_string(oh
->type
);
2688 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
2692 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
2699 handle_packet_out(struct ofproto
*p
, struct ofconn
*ofconn
,
2700 struct ofp_header
*oh
)
2702 struct ofp_packet_out
*opo
;
2703 struct ofpbuf payload
, *buffer
;
2704 struct odp_actions actions
;
2710 error
= reject_slave_controller(ofconn
, oh
);
2715 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2719 opo
= (struct ofp_packet_out
*) oh
;
2721 COVERAGE_INC(ofproto_packet_out
);
2722 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2723 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2725 if (error
|| !buffer
) {
2733 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2734 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2735 &flow
, p
, &payload
, &actions
, NULL
, NULL
, NULL
);
2740 dpif_execute(p
->dpif
, flow
.in_port
, actions
.actions
, actions
.n_actions
,
2742 ofpbuf_delete(buffer
);
2748 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2749 uint32_t config
, uint32_t mask
)
2751 mask
&= config
^ port
->opp
.config
;
2752 if (mask
& OFPPC_PORT_DOWN
) {
2753 if (config
& OFPPC_PORT_DOWN
) {
2754 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2756 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2759 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2760 if (mask
& REVALIDATE_BITS
) {
2761 COVERAGE_INC(ofproto_costly_flags
);
2762 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2763 p
->need_revalidate
= true;
2765 #undef REVALIDATE_BITS
2766 if (mask
& OFPPC_NO_FLOOD
) {
2767 port
->opp
.config
^= OFPPC_NO_FLOOD
;
2768 refresh_port_groups(p
);
2770 if (mask
& OFPPC_NO_PACKET_IN
) {
2771 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2776 handle_port_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
2777 struct ofp_header
*oh
)
2779 const struct ofp_port_mod
*opm
;
2780 struct ofport
*port
;
2783 error
= reject_slave_controller(ofconn
, oh
);
2787 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2791 opm
= (struct ofp_port_mod
*) oh
;
2793 port
= port_array_get(&p
->ports
,
2794 ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2796 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2797 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2798 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2800 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2801 if (opm
->advertise
) {
2802 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2808 static struct ofpbuf
*
2809 make_stats_reply(uint32_t xid
, uint16_t type
, size_t body_len
)
2811 struct ofp_stats_reply
*osr
;
2814 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
2815 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
2817 osr
->flags
= htons(0);
2821 static struct ofpbuf
*
2822 start_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
2824 return make_stats_reply(request
->header
.xid
, request
->type
, body_len
);
2828 append_stats_reply(size_t nbytes
, struct ofconn
*ofconn
, struct ofpbuf
**msgp
)
2830 struct ofpbuf
*msg
= *msgp
;
2831 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
2832 if (nbytes
+ msg
->size
> UINT16_MAX
) {
2833 struct ofp_stats_reply
*reply
= msg
->data
;
2834 reply
->flags
= htons(OFPSF_REPLY_MORE
);
2835 *msgp
= make_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
2836 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2838 return ofpbuf_put_uninit(*msgp
, nbytes
);
2842 handle_desc_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2843 struct ofp_stats_request
*request
)
2845 struct ofp_desc_stats
*ods
;
2848 msg
= start_stats_reply(request
, sizeof *ods
);
2849 ods
= append_stats_reply(sizeof *ods
, ofconn
, &msg
);
2850 memset(ods
, 0, sizeof *ods
);
2851 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
2852 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
2853 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
2854 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
2855 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
2856 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2862 count_subrules(struct cls_rule
*cls_rule
, void *n_subrules_
)
2864 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
2865 int *n_subrules
= n_subrules_
;
2873 handle_table_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2874 struct ofp_stats_request
*request
)
2876 struct ofp_table_stats
*ots
;
2878 struct odp_stats dpstats
;
2879 int n_exact
, n_subrules
, n_wild
;
2881 msg
= start_stats_reply(request
, sizeof *ots
* 2);
2883 /* Count rules of various kinds. */
2885 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, count_subrules
, &n_subrules
);
2886 n_exact
= classifier_count_exact(&p
->cls
) - n_subrules
;
2887 n_wild
= classifier_count(&p
->cls
) - classifier_count_exact(&p
->cls
);
2890 dpif_get_dp_stats(p
->dpif
, &dpstats
);
2891 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2892 memset(ots
, 0, sizeof *ots
);
2893 ots
->table_id
= TABLEID_HASH
;
2894 strcpy(ots
->name
, "hash");
2895 ots
->wildcards
= htonl(0);
2896 ots
->max_entries
= htonl(dpstats
.max_capacity
);
2897 ots
->active_count
= htonl(n_exact
);
2898 ots
->lookup_count
= htonll(dpstats
.n_frags
+ dpstats
.n_hit
+
2900 ots
->matched_count
= htonll(dpstats
.n_hit
); /* XXX */
2902 /* Classifier table. */
2903 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2904 memset(ots
, 0, sizeof *ots
);
2905 ots
->table_id
= TABLEID_CLASSIFIER
;
2906 strcpy(ots
->name
, "classifier");
2907 ots
->wildcards
= p
->tun_id_from_cookie
? htonl(OVSFW_ALL
)
2909 ots
->max_entries
= htonl(65536);
2910 ots
->active_count
= htonl(n_wild
);
2911 ots
->lookup_count
= htonll(0); /* XXX */
2912 ots
->matched_count
= htonll(0); /* XXX */
2914 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2919 append_port_stat(struct ofport
*port
, uint16_t port_no
, struct ofconn
*ofconn
,
2920 struct ofpbuf
**msgp
)
2922 struct netdev_stats stats
;
2923 struct ofp_port_stats
*ops
;
2925 /* Intentionally ignore return value, since errors will set
2926 * 'stats' to all-1s, which is correct for OpenFlow, and
2927 * netdev_get_stats() will log errors. */
2928 netdev_get_stats(port
->netdev
, &stats
);
2930 ops
= append_stats_reply(sizeof *ops
, ofconn
, msgp
);
2931 ops
->port_no
= htons(odp_port_to_ofp_port(port_no
));
2932 memset(ops
->pad
, 0, sizeof ops
->pad
);
2933 ops
->rx_packets
= htonll(stats
.rx_packets
);
2934 ops
->tx_packets
= htonll(stats
.tx_packets
);
2935 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
2936 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
2937 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
2938 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
2939 ops
->rx_errors
= htonll(stats
.rx_errors
);
2940 ops
->tx_errors
= htonll(stats
.tx_errors
);
2941 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
2942 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
2943 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
2944 ops
->collisions
= htonll(stats
.collisions
);
2948 handle_port_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2949 struct ofp_stats_request
*osr
,
2952 struct ofp_port_stats_request
*psr
;
2953 struct ofp_port_stats
*ops
;
2955 struct ofport
*port
;
2956 unsigned int port_no
;
2958 if (arg_size
!= sizeof *psr
) {
2959 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
2961 psr
= (struct ofp_port_stats_request
*) osr
->body
;
2963 msg
= start_stats_reply(osr
, sizeof *ops
* 16);
2964 if (psr
->port_no
!= htons(OFPP_NONE
)) {
2965 port
= port_array_get(&p
->ports
,
2966 ofp_port_to_odp_port(ntohs(psr
->port_no
)));
2968 append_port_stat(port
, ntohs(psr
->port_no
), ofconn
, &msg
);
2971 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
2972 append_port_stat(port
, port_no
, ofconn
, &msg
);
2976 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2980 struct flow_stats_cbdata
{
2981 struct ofproto
*ofproto
;
2982 struct ofconn
*ofconn
;
2987 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2988 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2989 * returned statistic include statistics for all of 'rule''s subrules. */
2991 query_stats(struct ofproto
*p
, struct rule
*rule
,
2992 uint64_t *packet_countp
, uint64_t *byte_countp
)
2994 uint64_t packet_count
, byte_count
;
2995 struct rule
*subrule
;
2996 struct odp_flow
*odp_flows
;
2999 /* Start from historical data for 'rule' itself that are no longer tracked
3000 * by the datapath. This counts, for example, subrules that have
3002 packet_count
= rule
->packet_count
;
3003 byte_count
= rule
->byte_count
;
3005 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3006 * wildcarded then on all of its subrules.
3008 * Also, add any statistics that are not tracked by the datapath for each
3009 * subrule. This includes, for example, statistics for packets that were
3010 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3012 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
3013 odp_flows
= xzalloc(n_odp_flows
* sizeof *odp_flows
);
3014 if (rule
->cr
.wc
.wildcards
) {
3016 LIST_FOR_EACH (subrule
, struct rule
, list
, &rule
->list
) {
3017 odp_flows
[i
++].key
= subrule
->cr
.flow
;
3018 packet_count
+= subrule
->packet_count
;
3019 byte_count
+= subrule
->byte_count
;
3022 odp_flows
[0].key
= rule
->cr
.flow
;
3025 /* Fetch up-to-date statistics from the datapath and add them in. */
3026 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3028 for (i
= 0; i
< n_odp_flows
; i
++) {
3029 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3030 packet_count
+= odp_flow
->stats
.n_packets
;
3031 byte_count
+= odp_flow
->stats
.n_bytes
;
3036 /* Return the stats to the caller. */
3037 *packet_countp
= packet_count
;
3038 *byte_countp
= byte_count
;
3042 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3044 struct rule
*rule
= rule_from_cls_rule(rule_
);
3045 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3046 struct ofp_flow_stats
*ofs
;
3047 uint64_t packet_count
, byte_count
;
3048 size_t act_len
, len
;
3049 long long int tdiff
= time_msec() - rule
->created
;
3050 uint32_t sec
= tdiff
/ 1000;
3051 uint32_t msec
= tdiff
- (sec
* 1000);
3053 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3057 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3058 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3060 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3062 ofs
= append_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
3063 ofs
->length
= htons(len
);
3064 ofs
->table_id
= rule
->cr
.wc
.wildcards
? TABLEID_CLASSIFIER
: TABLEID_HASH
;
3066 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3067 cbdata
->ofproto
->tun_id_from_cookie
, &ofs
->match
);
3068 ofs
->duration_sec
= htonl(sec
);
3069 ofs
->duration_nsec
= htonl(msec
* 1000000);
3070 ofs
->cookie
= rule
->flow_cookie
;
3071 ofs
->priority
= htons(rule
->cr
.priority
);
3072 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3073 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3074 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3075 ofs
->packet_count
= htonll(packet_count
);
3076 ofs
->byte_count
= htonll(byte_count
);
3077 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3081 table_id_to_include(uint8_t table_id
)
3083 return (table_id
== TABLEID_HASH
? CLS_INC_EXACT
3084 : table_id
== TABLEID_CLASSIFIER
? CLS_INC_WILD
3085 : table_id
== 0xff ? CLS_INC_ALL
3090 handle_flow_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3091 const struct ofp_stats_request
*osr
,
3094 struct ofp_flow_stats_request
*fsr
;
3095 struct flow_stats_cbdata cbdata
;
3096 struct cls_rule target
;
3098 if (arg_size
!= sizeof *fsr
) {
3099 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3101 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3103 COVERAGE_INC(ofproto_flows_req
);
3105 cbdata
.ofconn
= ofconn
;
3106 cbdata
.out_port
= fsr
->out_port
;
3107 cbdata
.msg
= start_stats_reply(osr
, 1024);
3108 cls_rule_from_match(&fsr
->match
, 0, false, 0, &target
);
3109 classifier_for_each_match(&p
->cls
, &target
,
3110 table_id_to_include(fsr
->table_id
),
3111 flow_stats_cb
, &cbdata
);
3112 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3116 struct flow_stats_ds_cbdata
{
3117 struct ofproto
*ofproto
;
3122 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
3124 struct rule
*rule
= rule_from_cls_rule(rule_
);
3125 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
3126 struct ds
*results
= cbdata
->results
;
3127 struct ofp_match match
;
3128 uint64_t packet_count
, byte_count
;
3129 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3131 /* Don't report on subrules. */
3132 if (rule
->super
!= NULL
) {
3136 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3137 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3138 cbdata
->ofproto
->tun_id_from_cookie
, &match
);
3140 ds_put_format(results
, "duration=%llds, ",
3141 (time_msec() - rule
->created
) / 1000);
3142 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3143 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3144 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3145 ofp_print_match(results
, &match
, true);
3146 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3147 ds_put_cstr(results
, "\n");
3150 /* Adds a pretty-printed description of all flows to 'results', including
3151 * those marked hidden by secchan (e.g., by in-band control). */
3153 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3155 struct ofp_match match
;
3156 struct cls_rule target
;
3157 struct flow_stats_ds_cbdata cbdata
;
3159 memset(&match
, 0, sizeof match
);
3160 match
.wildcards
= htonl(OVSFW_ALL
);
3163 cbdata
.results
= results
;
3165 cls_rule_from_match(&match
, 0, false, 0, &target
);
3166 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3167 flow_stats_ds_cb
, &cbdata
);
3170 struct aggregate_stats_cbdata
{
3171 struct ofproto
*ofproto
;
3173 uint64_t packet_count
;
3174 uint64_t byte_count
;
3179 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3181 struct rule
*rule
= rule_from_cls_rule(rule_
);
3182 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
3183 uint64_t packet_count
, byte_count
;
3185 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3189 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3191 cbdata
->packet_count
+= packet_count
;
3192 cbdata
->byte_count
+= byte_count
;
3197 handle_aggregate_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3198 const struct ofp_stats_request
*osr
,
3201 struct ofp_aggregate_stats_request
*asr
;
3202 struct ofp_aggregate_stats_reply
*reply
;
3203 struct aggregate_stats_cbdata cbdata
;
3204 struct cls_rule target
;
3207 if (arg_size
!= sizeof *asr
) {
3208 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3210 asr
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3212 COVERAGE_INC(ofproto_agg_request
);
3214 cbdata
.out_port
= asr
->out_port
;
3215 cbdata
.packet_count
= 0;
3216 cbdata
.byte_count
= 0;
3218 cls_rule_from_match(&asr
->match
, 0, false, 0, &target
);
3219 classifier_for_each_match(&p
->cls
, &target
,
3220 table_id_to_include(asr
->table_id
),
3221 aggregate_stats_cb
, &cbdata
);
3223 msg
= start_stats_reply(osr
, sizeof *reply
);
3224 reply
= append_stats_reply(sizeof *reply
, ofconn
, &msg
);
3225 reply
->flow_count
= htonl(cbdata
.n_flows
);
3226 reply
->packet_count
= htonll(cbdata
.packet_count
);
3227 reply
->byte_count
= htonll(cbdata
.byte_count
);
3228 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3232 struct queue_stats_cbdata
{
3233 struct ofconn
*ofconn
;
3239 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3240 const struct netdev_queue_stats
*stats
)
3242 struct ofp_queue_stats
*reply
;
3244 reply
= append_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3245 reply
->port_no
= htons(cbdata
->port_no
);
3246 memset(reply
->pad
, 0, sizeof reply
->pad
);
3247 reply
->queue_id
= htonl(queue_id
);
3248 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3249 reply
->tx_packets
= htonll(stats
->tx_packets
);
3250 reply
->tx_errors
= htonll(stats
->tx_errors
);
3254 handle_queue_stats_dump_cb(uint32_t queue_id
,
3255 struct netdev_queue_stats
*stats
,
3258 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3260 put_queue_stats(cbdata
, queue_id
, stats
);
3264 handle_queue_stats_for_port(struct ofport
*port
, uint16_t port_no
,
3266 struct queue_stats_cbdata
*cbdata
)
3268 cbdata
->port_no
= port_no
;
3269 if (queue_id
== OFPQ_ALL
) {
3270 netdev_dump_queue_stats(port
->netdev
,
3271 handle_queue_stats_dump_cb
, cbdata
);
3273 struct netdev_queue_stats stats
;
3275 netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
);
3276 put_queue_stats(cbdata
, queue_id
, &stats
);
3281 handle_queue_stats_request(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3282 const struct ofp_stats_request
*osr
,
3285 struct ofp_queue_stats_request
*qsr
;
3286 struct queue_stats_cbdata cbdata
;
3287 struct ofport
*port
;
3288 unsigned int port_no
;
3291 if (arg_size
!= sizeof *qsr
) {
3292 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3294 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3296 COVERAGE_INC(ofproto_queue_req
);
3298 cbdata
.ofconn
= ofconn
;
3299 cbdata
.msg
= start_stats_reply(osr
, 128);
3301 port_no
= ntohs(qsr
->port_no
);
3302 queue_id
= ntohl(qsr
->queue_id
);
3303 if (port_no
== OFPP_ALL
) {
3304 PORT_ARRAY_FOR_EACH (port
, &ofproto
->ports
, port_no
) {
3305 handle_queue_stats_for_port(port
, port_no
, queue_id
, &cbdata
);
3307 } else if (port_no
< ofproto
->max_ports
) {
3308 port
= port_array_get(&ofproto
->ports
, port_no
);
3310 handle_queue_stats_for_port(port
, port_no
, queue_id
, &cbdata
);
3313 ofpbuf_delete(cbdata
.msg
);
3314 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3316 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3322 handle_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3323 struct ofp_header
*oh
)
3325 struct ofp_stats_request
*osr
;
3329 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3334 osr
= (struct ofp_stats_request
*) oh
;
3336 switch (ntohs(osr
->type
)) {
3338 return handle_desc_stats_request(p
, ofconn
, osr
);
3341 return handle_flow_stats_request(p
, ofconn
, osr
, arg_size
);
3343 case OFPST_AGGREGATE
:
3344 return handle_aggregate_stats_request(p
, ofconn
, osr
, arg_size
);
3347 return handle_table_stats_request(p
, ofconn
, osr
);
3350 return handle_port_stats_request(p
, ofconn
, osr
, arg_size
);
3353 return handle_queue_stats_request(p
, ofconn
, osr
, arg_size
);
3356 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3359 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3363 static long long int
3364 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3366 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3370 update_time(struct ofproto
*ofproto
, struct rule
*rule
,
3371 const struct odp_flow_stats
*stats
)
3373 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3374 if (used
> rule
->used
) {
3376 if (rule
->super
&& used
> rule
->super
->used
) {
3377 rule
->super
->used
= used
;
3379 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, used
);
3384 update_stats(struct ofproto
*ofproto
, struct rule
*rule
,
3385 const struct odp_flow_stats
*stats
)
3387 if (stats
->n_packets
) {
3388 update_time(ofproto
, rule
, stats
);
3389 rule
->packet_count
+= stats
->n_packets
;
3390 rule
->byte_count
+= stats
->n_bytes
;
3391 netflow_flow_update_flags(&rule
->nf_flow
, stats
->tcp_flags
);
3395 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3396 * in which no matching flow already exists in the flow table.
3398 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3399 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3400 * code as encoded by ofp_mkerr() on failure.
3402 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3405 add_flow(struct ofproto
*p
, struct ofconn
*ofconn
,
3406 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3408 struct ofpbuf
*packet
;
3413 if (ofm
->flags
& htons(OFPFF_CHECK_OVERLAP
)) {
3417 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3419 if (classifier_rule_overlaps(&p
->cls
, &flow
, wildcards
,
3420 ntohs(ofm
->priority
))) {
3421 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3425 rule
= rule_create(p
, NULL
, (const union ofp_action
*) ofm
->actions
,
3426 n_actions
, ntohs(ofm
->idle_timeout
),
3427 ntohs(ofm
->hard_timeout
), ofm
->cookie
,
3428 ofm
->flags
& htons(OFPFF_SEND_FLOW_REM
));
3429 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
),
3430 p
->tun_id_from_cookie
, ofm
->cookie
, &rule
->cr
);
3433 if (ofm
->buffer_id
!= htonl(UINT32_MAX
)) {
3434 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3438 in_port
= UINT16_MAX
;
3441 rule_insert(p
, rule
, packet
, in_port
);
3442 ofpbuf_delete(packet
);
3446 static struct rule
*
3447 find_flow_strict(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3452 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3454 return rule_from_cls_rule(classifier_find_rule_exactly(
3455 &p
->cls
, &flow
, wildcards
,
3456 ntohs(ofm
->priority
)));
3460 send_buffered_packet(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3461 struct rule
*rule
, const struct ofp_flow_mod
*ofm
)
3463 struct ofpbuf
*packet
;
3468 if (ofm
->buffer_id
== htonl(UINT32_MAX
)) {
3472 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3478 flow_extract(packet
, 0, in_port
, &flow
);
3479 rule_execute(ofproto
, rule
, packet
, &flow
);
3480 ofpbuf_delete(packet
);
3485 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3487 struct modify_flows_cbdata
{
3488 struct ofproto
*ofproto
;
3489 const struct ofp_flow_mod
*ofm
;
3494 static int modify_flow(struct ofproto
*, const struct ofp_flow_mod
*,
3495 size_t n_actions
, struct rule
*);
3496 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3498 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3499 * encoded by ofp_mkerr() on failure.
3501 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3504 modify_flows_loose(struct ofproto
*p
, struct ofconn
*ofconn
,
3505 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3507 struct modify_flows_cbdata cbdata
;
3508 struct cls_rule target
;
3512 cbdata
.n_actions
= n_actions
;
3513 cbdata
.match
= NULL
;
3515 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3518 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3519 modify_flows_cb
, &cbdata
);
3521 /* This credits the packet to whichever flow happened to happened to
3522 * match last. That's weird. Maybe we should do a lookup for the
3523 * flow that actually matches the packet? Who knows. */
3524 send_buffered_packet(p
, ofconn
, cbdata
.match
, ofm
);
3527 return add_flow(p
, ofconn
, ofm
, n_actions
);
3531 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3532 * code as encoded by ofp_mkerr() on failure.
3534 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3537 modify_flow_strict(struct ofproto
*p
, struct ofconn
*ofconn
,
3538 struct ofp_flow_mod
*ofm
, size_t n_actions
)
3540 struct rule
*rule
= find_flow_strict(p
, ofm
);
3541 if (rule
&& !rule_is_hidden(rule
)) {
3542 modify_flow(p
, ofm
, n_actions
, rule
);
3543 return send_buffered_packet(p
, ofconn
, rule
, ofm
);
3545 return add_flow(p
, ofconn
, ofm
, n_actions
);
3549 /* Callback for modify_flows_loose(). */
3551 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3553 struct rule
*rule
= rule_from_cls_rule(rule_
);
3554 struct modify_flows_cbdata
*cbdata
= cbdata_
;
3556 if (!rule_is_hidden(rule
)) {
3557 cbdata
->match
= rule
;
3558 modify_flow(cbdata
->ofproto
, cbdata
->ofm
, cbdata
->n_actions
, rule
);
3562 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3563 * been identified as a flow in 'p''s flow table to be modified, by changing
3564 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3565 * ofp_action[] structures). */
3567 modify_flow(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
3568 size_t n_actions
, struct rule
*rule
)
3570 size_t actions_len
= n_actions
* sizeof *rule
->actions
;
3572 rule
->flow_cookie
= ofm
->cookie
;
3574 /* If the actions are the same, do nothing. */
3575 if (n_actions
== rule
->n_actions
3576 && !memcmp(ofm
->actions
, rule
->actions
, actions_len
))
3581 /* Replace actions. */
3582 free(rule
->actions
);
3583 rule
->actions
= xmemdup(ofm
->actions
, actions_len
);
3584 rule
->n_actions
= n_actions
;
3586 /* Make sure that the datapath gets updated properly. */
3587 if (rule
->cr
.wc
.wildcards
) {
3588 COVERAGE_INC(ofproto_mod_wc_flow
);
3589 p
->need_revalidate
= true;
3591 rule_update_actions(p
, rule
);
3597 /* OFPFC_DELETE implementation. */
3599 struct delete_flows_cbdata
{
3600 struct ofproto
*ofproto
;
3604 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
3605 static void delete_flow(struct ofproto
*, struct rule
*, uint16_t out_port
);
3607 /* Implements OFPFC_DELETE. */
3609 delete_flows_loose(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3611 struct delete_flows_cbdata cbdata
;
3612 struct cls_rule target
;
3615 cbdata
.out_port
= ofm
->out_port
;
3617 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3620 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3621 delete_flows_cb
, &cbdata
);
3624 /* Implements OFPFC_DELETE_STRICT. */
3626 delete_flow_strict(struct ofproto
*p
, struct ofp_flow_mod
*ofm
)
3628 struct rule
*rule
= find_flow_strict(p
, ofm
);
3630 delete_flow(p
, rule
, ofm
->out_port
);
3634 /* Callback for delete_flows_loose(). */
3636 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3638 struct rule
*rule
= rule_from_cls_rule(rule_
);
3639 struct delete_flows_cbdata
*cbdata
= cbdata_
;
3641 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
3644 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3645 * been identified as a flow to delete from 'p''s flow table, by deleting the
3646 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3649 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3650 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3651 * specified 'out_port'. */
3653 delete_flow(struct ofproto
*p
, struct rule
*rule
, uint16_t out_port
)
3655 if (rule_is_hidden(rule
)) {
3659 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
3663 send_flow_removed(p
, rule
, time_msec(), OFPRR_DELETE
);
3664 rule_remove(p
, rule
);
3668 handle_flow_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
3669 struct ofp_flow_mod
*ofm
)
3671 struct ofp_match orig_match
;
3675 error
= reject_slave_controller(ofconn
, &ofm
->header
);
3679 error
= check_ofp_message_array(&ofm
->header
, OFPT_FLOW_MOD
, sizeof *ofm
,
3680 sizeof *ofm
->actions
, &n_actions
);
3685 /* We do not support the emergency flow cache. It will hopefully
3686 * get dropped from OpenFlow in the near future. */
3687 if (ofm
->flags
& htons(OFPFF_EMERG
)) {
3688 /* There isn't a good fit for an error code, so just state that the
3689 * flow table is full. */
3690 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
3693 /* Normalize ofp->match. If normalization actually changes anything, then
3694 * log the differences. */
3695 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
3696 orig_match
= ofm
->match
;
3697 normalize_match(&ofm
->match
);
3698 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
3699 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
3700 if (!VLOG_DROP_INFO(&normal_rl
)) {
3701 char *old
= ofp_match_to_literal_string(&orig_match
);
3702 char *new = ofp_match_to_literal_string(&ofm
->match
);
3703 VLOG_INFO("%s: normalization changed ofp_match, details:",
3704 rconn_get_name(ofconn
->rconn
));
3705 VLOG_INFO(" pre: %s", old
);
3706 VLOG_INFO("post: %s", new);
3712 if (!ofm
->match
.wildcards
) {
3713 ofm
->priority
= htons(UINT16_MAX
);
3716 error
= validate_actions((const union ofp_action
*) ofm
->actions
,
3717 n_actions
, p
->max_ports
);
3722 switch (ntohs(ofm
->command
)) {
3724 return add_flow(p
, ofconn
, ofm
, n_actions
);
3727 return modify_flows_loose(p
, ofconn
, ofm
, n_actions
);
3729 case OFPFC_MODIFY_STRICT
:
3730 return modify_flow_strict(p
, ofconn
, ofm
, n_actions
);
3733 delete_flows_loose(p
, ofm
);
3736 case OFPFC_DELETE_STRICT
:
3737 delete_flow_strict(p
, ofm
);
3741 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
3746 handle_tun_id_from_cookie(struct ofproto
*p
, struct nxt_tun_id_cookie
*msg
)
3750 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
3755 p
->tun_id_from_cookie
= !!msg
->set
;
3760 handle_role_request(struct ofproto
*ofproto
,
3761 struct ofconn
*ofconn
, struct nicira_header
*msg
)
3763 struct nx_role_request
*nrr
;
3764 struct nx_role_request
*reply
;
3768 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
3769 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
3770 ntohs(msg
->header
.length
), sizeof *nrr
);
3771 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3773 nrr
= (struct nx_role_request
*) msg
;
3775 if (ofconn
->type
!= OFCONN_CONTROLLER
) {
3776 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
3778 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3781 role
= ntohl(nrr
->role
);
3782 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3783 && role
!= NX_ROLE_SLAVE
) {
3784 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
3786 /* There's no good error code for this. */
3787 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
3790 if (role
== NX_ROLE_MASTER
) {
3791 struct ofconn
*other
;
3793 HMAP_FOR_EACH (other
, struct ofconn
, hmap_node
,
3794 &ofproto
->controllers
) {
3795 if (other
->role
== NX_ROLE_MASTER
) {
3796 other
->role
= NX_ROLE_SLAVE
;
3800 ofconn
->role
= role
;
3802 reply
= make_openflow_xid(sizeof *reply
, OFPT_VENDOR
, msg
->header
.xid
,
3804 reply
->nxh
.vendor
= htonl(NX_VENDOR_ID
);
3805 reply
->nxh
.subtype
= htonl(NXT_ROLE_REPLY
);
3806 reply
->role
= htonl(role
);
3807 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3813 handle_vendor(struct ofproto
*p
, struct ofconn
*ofconn
, void *msg
)
3815 struct ofp_vendor_header
*ovh
= msg
;
3816 struct nicira_header
*nh
;
3818 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
3819 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
3820 "(expected at least %zu)",
3821 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
3822 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3824 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
3825 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3827 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
3828 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
3829 "(expected at least %zu)",
3830 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
3831 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3835 switch (ntohl(nh
->subtype
)) {
3836 case NXT_STATUS_REQUEST
:
3837 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
3840 case NXT_TUN_ID_FROM_COOKIE
:
3841 return handle_tun_id_from_cookie(p
, msg
);
3843 case NXT_ROLE_REQUEST
:
3844 return handle_role_request(p
, ofconn
, msg
);
3847 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
3851 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3853 struct ofp_header
*ob
;
3856 /* Currently, everything executes synchronously, so we can just
3857 * immediately send the barrier reply. */
3858 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
3859 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3864 handle_openflow(struct ofconn
*ofconn
, struct ofproto
*p
,
3865 struct ofpbuf
*ofp_msg
)
3867 struct ofp_header
*oh
= ofp_msg
->data
;
3870 COVERAGE_INC(ofproto_recv_openflow
);
3872 case OFPT_ECHO_REQUEST
:
3873 error
= handle_echo_request(ofconn
, oh
);
3876 case OFPT_ECHO_REPLY
:
3880 case OFPT_FEATURES_REQUEST
:
3881 error
= handle_features_request(p
, ofconn
, oh
);
3884 case OFPT_GET_CONFIG_REQUEST
:
3885 error
= handle_get_config_request(p
, ofconn
, oh
);
3888 case OFPT_SET_CONFIG
:
3889 error
= handle_set_config(p
, ofconn
, ofp_msg
->data
);
3892 case OFPT_PACKET_OUT
:
3893 error
= handle_packet_out(p
, ofconn
, ofp_msg
->data
);
3897 error
= handle_port_mod(p
, ofconn
, oh
);
3901 error
= handle_flow_mod(p
, ofconn
, ofp_msg
->data
);
3904 case OFPT_STATS_REQUEST
:
3905 error
= handle_stats_request(p
, ofconn
, oh
);
3909 error
= handle_vendor(p
, ofconn
, ofp_msg
->data
);
3912 case OFPT_BARRIER_REQUEST
:
3913 error
= handle_barrier_request(ofconn
, oh
);
3917 if (VLOG_IS_WARN_ENABLED()) {
3918 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
3919 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
3922 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
3927 send_error_oh(ofconn
, ofp_msg
->data
, error
);
3932 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
3934 struct odp_msg
*msg
= packet
->data
;
3936 struct ofpbuf payload
;
3939 payload
.data
= msg
+ 1;
3940 payload
.size
= msg
->length
- sizeof *msg
;
3941 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
3943 /* Check with in-band control to see if this packet should be sent
3944 * to the local port regardless of the flow table. */
3945 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
3946 union odp_action action
;
3948 memset(&action
, 0, sizeof(action
));
3949 action
.output
.type
= ODPAT_OUTPUT
;
3950 action
.output
.port
= ODPP_LOCAL
;
3951 dpif_execute(p
->dpif
, flow
.in_port
, &action
, 1, &payload
);
3954 rule
= lookup_valid_rule(p
, &flow
);
3956 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3957 struct ofport
*port
= port_array_get(&p
->ports
, msg
->port
);
3959 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
3960 COVERAGE_INC(ofproto_no_packet_in
);
3961 /* XXX install 'drop' flow entry */
3962 ofpbuf_delete(packet
);
3966 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
3969 COVERAGE_INC(ofproto_packet_in
);
3970 send_packet_in(p
, packet
);
3974 if (rule
->cr
.wc
.wildcards
) {
3975 rule
= rule_create_subrule(p
, rule
, &flow
);
3976 rule_make_actions(p
, rule
, packet
);
3978 if (!rule
->may_install
) {
3979 /* The rule is not installable, that is, we need to process every
3980 * packet, so process the current packet and set its actions into
3982 rule_make_actions(p
, rule
, packet
);
3984 /* XXX revalidate rule if it needs it */
3988 rule_execute(p
, rule
, &payload
, &flow
);
3989 rule_reinstall(p
, rule
);
3991 if (rule
->super
&& rule
->super
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
3993 * Extra-special case for fail-open mode.
3995 * We are in fail-open mode and the packet matched the fail-open rule,
3996 * but we are connected to a controller too. We should send the packet
3997 * up to the controller in the hope that it will try to set up a flow
3998 * and thereby allow us to exit fail-open.
4000 * See the top-level comment in fail-open.c for more information.
4002 send_packet_in(p
, packet
);
4004 ofpbuf_delete(packet
);
4009 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4011 struct odp_msg
*msg
= packet
->data
;
4013 switch (msg
->type
) {
4014 case _ODPL_ACTION_NR
:
4015 COVERAGE_INC(ofproto_ctlr_action
);
4016 send_packet_in(p
, packet
);
4019 case _ODPL_SFLOW_NR
:
4021 ofproto_sflow_received(p
->sflow
, msg
);
4023 ofpbuf_delete(packet
);
4027 handle_odp_miss_msg(p
, packet
);
4031 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4038 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
4040 struct rule
*sub
= rule_from_cls_rule(sub_
);
4041 struct revalidate_cbdata
*cbdata
= cbdata_
;
4043 if (cbdata
->revalidate_all
4044 || (cbdata
->revalidate_subrules
&& sub
->super
)
4045 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
4046 revalidate_rule(cbdata
->ofproto
, sub
);
4051 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
4053 const flow_t
*flow
= &rule
->cr
.flow
;
4055 COVERAGE_INC(ofproto_revalidate_rule
);
4058 super
= rule_from_cls_rule(classifier_lookup_wild(&p
->cls
, flow
));
4060 rule_remove(p
, rule
);
4062 } else if (super
!= rule
->super
) {
4063 COVERAGE_INC(ofproto_revalidate_moved
);
4064 list_remove(&rule
->list
);
4065 list_push_back(&super
->list
, &rule
->list
);
4066 rule
->super
= super
;
4067 rule
->hard_timeout
= super
->hard_timeout
;
4068 rule
->idle_timeout
= super
->idle_timeout
;
4069 rule
->created
= super
->created
;
4074 rule_update_actions(p
, rule
);
4078 static struct ofpbuf
*
4079 compose_flow_removed(struct ofproto
*p
, const struct rule
*rule
,
4080 long long int now
, uint8_t reason
)
4082 struct ofp_flow_removed
*ofr
;
4084 long long int tdiff
= now
- rule
->created
;
4085 uint32_t sec
= tdiff
/ 1000;
4086 uint32_t msec
= tdiff
- (sec
* 1000);
4088 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4089 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, p
->tun_id_from_cookie
,
4091 ofr
->cookie
= rule
->flow_cookie
;
4092 ofr
->priority
= htons(rule
->cr
.priority
);
4093 ofr
->reason
= reason
;
4094 ofr
->duration_sec
= htonl(sec
);
4095 ofr
->duration_nsec
= htonl(msec
* 1000000);
4096 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4097 ofr
->packet_count
= htonll(rule
->packet_count
);
4098 ofr
->byte_count
= htonll(rule
->byte_count
);
4104 uninstall_idle_flow(struct ofproto
*ofproto
, struct rule
*rule
)
4106 assert(rule
->installed
);
4107 assert(!rule
->cr
.wc
.wildcards
);
4110 rule_remove(ofproto
, rule
);
4112 rule_uninstall(ofproto
, rule
);
4117 send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
4118 long long int now
, uint8_t reason
)
4120 struct ofconn
*ofconn
;
4121 struct ofconn
*prev
;
4122 struct ofpbuf
*buf
= NULL
;
4124 /* We limit the maximum number of queued flow expirations it by accounting
4125 * them under the counter for replies. That works because preventing
4126 * OpenFlow requests from being processed also prevents new flows from
4127 * being added (and expiring). (It also prevents processing OpenFlow
4128 * requests that would not add new flows, so it is imperfect.) */
4131 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
4132 if (rule
->send_flow_removed
&& rconn_is_connected(ofconn
->rconn
)
4133 && ofconn_receives_async_msgs(ofconn
)) {
4135 queue_tx(ofpbuf_clone(buf
), prev
, prev
->reply_counter
);
4137 buf
= compose_flow_removed(p
, rule
, now
, reason
);
4143 queue_tx(buf
, prev
, prev
->reply_counter
);
4149 expire_rule(struct cls_rule
*cls_rule
, void *p_
)
4151 struct ofproto
*p
= p_
;
4152 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
4153 long long int hard_expire
, idle_expire
, expire
, now
;
4155 hard_expire
= (rule
->hard_timeout
4156 ? rule
->created
+ rule
->hard_timeout
* 1000
4158 idle_expire
= (rule
->idle_timeout
4159 && (rule
->super
|| list_is_empty(&rule
->list
))
4160 ? rule
->used
+ rule
->idle_timeout
* 1000
4162 expire
= MIN(hard_expire
, idle_expire
);
4166 if (rule
->installed
&& now
>= rule
->used
+ 5000) {
4167 uninstall_idle_flow(p
, rule
);
4168 } else if (!rule
->cr
.wc
.wildcards
) {
4169 active_timeout(p
, rule
);
4175 COVERAGE_INC(ofproto_expired
);
4177 /* Update stats. This code will be a no-op if the rule expired
4178 * due to an idle timeout. */
4179 if (rule
->cr
.wc
.wildcards
) {
4180 struct rule
*subrule
, *next
;
4181 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
4182 rule_remove(p
, subrule
);
4185 rule_uninstall(p
, rule
);
4188 if (!rule_is_hidden(rule
)) {
4189 send_flow_removed(p
, rule
, now
,
4191 ? OFPRR_HARD_TIMEOUT
: OFPRR_IDLE_TIMEOUT
));
4193 rule_remove(p
, rule
);
4197 active_timeout(struct ofproto
*ofproto
, struct rule
*rule
)
4199 if (ofproto
->netflow
&& !is_controller_rule(rule
) &&
4200 netflow_active_timeout_expired(ofproto
->netflow
, &rule
->nf_flow
)) {
4201 struct ofexpired expired
;
4202 struct odp_flow odp_flow
;
4204 /* Get updated flow stats. */
4205 memset(&odp_flow
, 0, sizeof odp_flow
);
4206 if (rule
->installed
) {
4207 odp_flow
.key
= rule
->cr
.flow
;
4208 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4209 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4211 if (odp_flow
.stats
.n_packets
) {
4212 update_time(ofproto
, rule
, &odp_flow
.stats
);
4213 netflow_flow_update_flags(&rule
->nf_flow
,
4214 odp_flow
.stats
.tcp_flags
);
4218 expired
.flow
= rule
->cr
.flow
;
4219 expired
.packet_count
= rule
->packet_count
+
4220 odp_flow
.stats
.n_packets
;
4221 expired
.byte_count
= rule
->byte_count
+ odp_flow
.stats
.n_bytes
;
4222 expired
.used
= rule
->used
;
4224 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
4226 /* Schedule us to send the accumulated records once we have
4227 * collected all of them. */
4228 poll_immediate_wake();
4233 update_used(struct ofproto
*p
)
4235 struct odp_flow
*flows
;
4240 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4245 for (i
= 0; i
< n_flows
; i
++) {
4246 struct odp_flow
*f
= &flows
[i
];
4249 rule
= rule_from_cls_rule(
4250 classifier_find_rule_exactly(&p
->cls
, &f
->key
, 0, UINT16_MAX
));
4251 if (!rule
|| !rule
->installed
) {
4252 COVERAGE_INC(ofproto_unexpected_rule
);
4253 dpif_flow_del(p
->dpif
, f
);
4257 update_time(p
, rule
, &f
->stats
);
4258 rule_account(p
, rule
, f
->stats
.n_bytes
);
4263 /* pinsched callback for sending 'packet' on 'ofconn'. */
4265 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4267 struct ofconn
*ofconn
= ofconn_
;
4269 rconn_send_with_limit(ofconn
->rconn
, packet
,
4270 ofconn
->packet_in_counter
, 100);
4273 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4274 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4275 * packet scheduler for sending.
4277 * 'max_len' specifies the maximum number of bytes of the packet to send on
4278 * 'ofconn' (INT_MAX specifies no limit).
4280 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4281 * ownership is transferred to this function. */
4283 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4286 struct ofproto
*ofproto
= ofconn
->ofproto
;
4287 struct ofp_packet_in
*opi
= packet
->data
;
4288 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4289 int send_len
, trim_size
;
4293 if (opi
->reason
== OFPR_ACTION
) {
4294 buffer_id
= UINT32_MAX
;
4295 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4296 buffer_id
= pktbuf_get_null();
4297 } else if (!ofconn
->pktbuf
) {
4298 buffer_id
= UINT32_MAX
;
4300 struct ofpbuf payload
;
4301 payload
.data
= opi
->data
;
4302 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4303 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4306 /* Figure out how much of the packet to send. */
4307 send_len
= ntohs(opi
->total_len
);
4308 if (buffer_id
!= UINT32_MAX
) {
4309 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4311 send_len
= MIN(send_len
, max_len
);
4313 /* Adjust packet length and clone if necessary. */
4314 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4316 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4319 packet
->size
= trim_size
;
4322 /* Update packet headers. */
4323 opi
->buffer_id
= htonl(buffer_id
);
4324 update_openflow_length(packet
);
4326 /* Hand over to packet scheduler. It might immediately call into
4327 * do_send_packet_in() or it might buffer it for a while (until a later
4328 * call to pinsched_run()). */
4329 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4330 packet
, do_send_packet_in
, ofconn
);
4333 /* Replace struct odp_msg header in 'packet' by equivalent struct
4334 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4335 * returned by dpif_recv()).
4337 * The conversion is not complete: the caller still needs to trim any unneeded
4338 * payload off the end of the buffer, set the length in the OpenFlow header,
4339 * and set buffer_id. Those require us to know the controller settings and so
4340 * must be done on a per-controller basis.
4342 * Returns the maximum number of bytes of the packet that should be sent to
4343 * the controller (INT_MAX if no limit). */
4345 do_convert_to_packet_in(struct ofpbuf
*packet
)
4347 struct odp_msg
*msg
= packet
->data
;
4348 struct ofp_packet_in
*opi
;
4354 /* Extract relevant header fields */
4355 if (msg
->type
== _ODPL_ACTION_NR
) {
4356 reason
= OFPR_ACTION
;
4359 reason
= OFPR_NO_MATCH
;
4362 total_len
= msg
->length
- sizeof *msg
;
4363 in_port
= odp_port_to_ofp_port(msg
->port
);
4365 /* Repurpose packet buffer by overwriting header. */
4366 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4367 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4368 opi
->header
.version
= OFP_VERSION
;
4369 opi
->header
.type
= OFPT_PACKET_IN
;
4370 opi
->total_len
= htons(total_len
);
4371 opi
->in_port
= htons(in_port
);
4372 opi
->reason
= reason
;
4377 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4378 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4379 * as necessary according to their individual configurations.
4381 * 'packet' must have sufficient headroom to convert it into a struct
4382 * ofp_packet_in (e.g. as returned by dpif_recv()).
4384 * Takes ownership of 'packet'. */
4386 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4388 struct ofconn
*ofconn
, *prev
;
4391 max_len
= do_convert_to_packet_in(packet
);
4394 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &ofproto
->all_conns
) {
4395 if (ofconn_receives_async_msgs(ofconn
)) {
4397 schedule_packet_in(prev
, packet
, max_len
, true);
4403 schedule_packet_in(prev
, packet
, max_len
, false);
4405 ofpbuf_delete(packet
);
4410 pick_datapath_id(const struct ofproto
*ofproto
)
4412 const struct ofport
*port
;
4414 port
= port_array_get(&ofproto
->ports
, ODPP_LOCAL
);
4416 uint8_t ea
[ETH_ADDR_LEN
];
4419 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4421 return eth_addr_to_uint64(ea
);
4423 VLOG_WARN("could not get MAC address for %s (%s)",
4424 netdev_get_name(port
->netdev
), strerror(error
));
4426 return ofproto
->fallback_dpid
;
4430 pick_fallback_dpid(void)
4432 uint8_t ea
[ETH_ADDR_LEN
];
4433 eth_addr_nicira_random(ea
);
4434 return eth_addr_to_uint64(ea
);
4438 default_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
4439 struct odp_actions
*actions
, tag_type
*tags
,
4440 uint16_t *nf_output_iface
, void *ofproto_
)
4442 struct ofproto
*ofproto
= ofproto_
;
4445 /* Drop frames for reserved multicast addresses. */
4446 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4450 /* Learn source MAC (but don't try to learn from revalidation). */
4451 if (packet
!= NULL
) {
4452 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4454 GRAT_ARP_LOCK_NONE
);
4456 /* The log messages here could actually be useful in debugging,
4457 * so keep the rate limit relatively high. */
4458 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4459 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4460 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4461 ofproto_revalidate(ofproto
, rev_tag
);
4465 /* Determine output port. */
4466 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
4469 add_output_group_action(actions
, DP_GROUP_FLOOD
, nf_output_iface
);
4470 } else if (out_port
!= flow
->in_port
) {
4471 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4472 *nf_output_iface
= out_port
;
4480 static const struct ofhooks default_ofhooks
= {
4482 default_normal_ofhook_cb
,