2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
29 #include "classifier.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
38 #include "multipath.h"
44 #include "ofp-print.h"
46 #include "ofproto-sflow.h"
48 #include "openflow/nicira-ext.h"
49 #include "openflow/openflow.h"
50 #include "openvswitch/datapath-protocol.h"
54 #include "poll-loop.h"
57 #include "stream-ssl.h"
61 #include "unaligned.h"
66 VLOG_DEFINE_THIS_MODULE(ofproto
);
68 COVERAGE_DEFINE(facet_changed_rule
);
69 COVERAGE_DEFINE(facet_revalidate
);
70 COVERAGE_DEFINE(odp_overflow
);
71 COVERAGE_DEFINE(ofproto_agg_request
);
72 COVERAGE_DEFINE(ofproto_costly_flags
);
73 COVERAGE_DEFINE(ofproto_ctlr_action
);
74 COVERAGE_DEFINE(ofproto_del_rule
);
75 COVERAGE_DEFINE(ofproto_error
);
76 COVERAGE_DEFINE(ofproto_expiration
);
77 COVERAGE_DEFINE(ofproto_expired
);
78 COVERAGE_DEFINE(ofproto_flows_req
);
79 COVERAGE_DEFINE(ofproto_flush
);
80 COVERAGE_DEFINE(ofproto_invalidated
);
81 COVERAGE_DEFINE(ofproto_no_packet_in
);
82 COVERAGE_DEFINE(ofproto_ofconn_stuck
);
83 COVERAGE_DEFINE(ofproto_ofp2odp
);
84 COVERAGE_DEFINE(ofproto_packet_in
);
85 COVERAGE_DEFINE(ofproto_packet_out
);
86 COVERAGE_DEFINE(ofproto_queue_req
);
87 COVERAGE_DEFINE(ofproto_recv_openflow
);
88 COVERAGE_DEFINE(ofproto_reinit_ports
);
89 COVERAGE_DEFINE(ofproto_unexpected_rule
);
90 COVERAGE_DEFINE(ofproto_uninstallable
);
91 COVERAGE_DEFINE(ofproto_update_port
);
93 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
94 * flow translation. */
95 #define MAX_RESUBMIT_RECURSION 16
100 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
101 struct netdev
*netdev
;
102 struct ofp_phy_port opp
; /* In host byte order. */
104 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
107 static void ofport_free(struct ofport
*);
108 static void ofport_run(struct ofproto
*, struct ofport
*);
109 static void ofport_wait(struct ofport
*);
110 static void hton_ofp_phy_port(struct ofp_phy_port
*);
112 struct action_xlate_ctx
{
113 /* action_xlate_ctx_init() initializes these members. */
116 struct ofproto
*ofproto
;
118 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
119 * this flow when actions change header fields. */
122 /* The packet corresponding to 'flow', or a null pointer if we are
123 * revalidating without a packet to refer to. */
124 const struct ofpbuf
*packet
;
126 /* If nonnull, called just before executing a resubmit action.
128 * This is normally null so the client has to set it manually after
129 * calling action_xlate_ctx_init(). */
130 void (*resubmit_hook
)(struct action_xlate_ctx
*, struct rule
*);
132 /* If true, the speciality of 'flow' should be checked before executing
133 * its actions. If special_cb returns false on 'flow' rendered
134 * uninstallable and no actions will be executed. */
137 /* xlate_actions() initializes and uses these members. The client might want
138 * to look at them after it returns. */
140 struct ofpbuf
*odp_actions
; /* Datapath actions. */
141 tag_type tags
; /* Tags associated with OFPP_NORMAL actions. */
142 bool may_set_up_flow
; /* True ordinarily; false if the actions must
143 * be reassessed for every packet. */
144 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
146 /* xlate_actions() initializes and uses these members, but the client has no
147 * reason to look at them. */
149 int recurse
; /* Recursion level, via xlate_table_action. */
150 int last_pop_priority
; /* Offset in 'odp_actions' just past most
151 * recent ODP_ACTION_ATTR_SET_PRIORITY. */
154 static void action_xlate_ctx_init(struct action_xlate_ctx
*,
155 struct ofproto
*, const struct flow
*,
156 const struct ofpbuf
*);
157 static struct ofpbuf
*xlate_actions(struct action_xlate_ctx
*,
158 const union ofp_action
*in
, size_t n_in
);
160 /* An OpenFlow flow. */
162 long long int used
; /* Time last used; time created if not used. */
163 long long int created
; /* Creation time. */
167 * - Do include packets and bytes from facets that have been deleted or
168 * whose own statistics have been folded into the rule.
170 * - Do include packets and bytes sent "by hand" that were accounted to
171 * the rule without any facet being involved (this is a rare corner
172 * case in rule_execute()).
174 * - Do not include packet or bytes that can be obtained from any facet's
175 * packet_count or byte_count member or that can be obtained from the
176 * datapath by, e.g., dpif_flow_get() for any facet.
178 uint64_t packet_count
; /* Number of packets received. */
179 uint64_t byte_count
; /* Number of bytes received. */
181 ovs_be64 flow_cookie
; /* Controller-issued identifier. */
183 struct cls_rule cr
; /* In owning ofproto's classifier. */
184 uint16_t idle_timeout
; /* In seconds from time of last use. */
185 uint16_t hard_timeout
; /* In seconds from time of creation. */
186 bool send_flow_removed
; /* Send a flow removed message? */
187 int n_actions
; /* Number of elements in actions[]. */
188 union ofp_action
*actions
; /* OpenFlow actions. */
189 struct list facets
; /* List of "struct facet"s. */
192 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
193 static bool rule_is_hidden(const struct rule
*);
195 static struct rule
*rule_create(const struct cls_rule
*,
196 const union ofp_action
*, size_t n_actions
,
197 uint16_t idle_timeout
, uint16_t hard_timeout
,
198 ovs_be64 flow_cookie
, bool send_flow_removed
);
199 static void rule_destroy(struct ofproto
*, struct rule
*);
200 static void rule_free(struct rule
*);
202 static struct rule
*rule_lookup(struct ofproto
*, const struct flow
*);
203 static void rule_insert(struct ofproto
*, struct rule
*);
204 static void rule_remove(struct ofproto
*, struct rule
*);
206 static void rule_send_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
207 static void rule_get_stats(const struct rule
*, uint64_t *packets
,
210 /* An exact-match instantiation of an OpenFlow flow. */
212 long long int used
; /* Time last used; time created if not used. */
216 * - Do include packets and bytes sent "by hand", e.g. with
219 * - Do include packets and bytes that were obtained from the datapath
220 * when a flow was deleted (e.g. dpif_flow_del()) or when its
221 * statistics were reset (e.g. dpif_flow_put() with
222 * DPIF_FP_ZERO_STATS).
224 * - Do not include any packets or bytes that can currently be obtained
225 * from the datapath by, e.g., dpif_flow_get().
227 uint64_t packet_count
; /* Number of packets received. */
228 uint64_t byte_count
; /* Number of bytes received. */
230 uint64_t dp_packet_count
; /* Last known packet count in the datapath. */
231 uint64_t dp_byte_count
; /* Last known byte count in the datapath. */
233 uint64_t rs_packet_count
; /* Packets pushed to resubmit children. */
234 uint64_t rs_byte_count
; /* Bytes pushed to resubmit children. */
235 long long int rs_used
; /* Used time pushed to resubmit children. */
237 /* Number of bytes passed to account_cb. This may include bytes that can
238 * currently obtained from the datapath (thus, it can be greater than
240 uint64_t accounted_bytes
;
242 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
243 struct list list_node
; /* In owning rule's 'facets' list. */
244 struct rule
*rule
; /* Owning rule. */
245 struct flow flow
; /* Exact-match flow. */
246 bool installed
; /* Installed in datapath? */
247 bool may_install
; /* True ordinarily; false if actions must
248 * be reassessed for every packet. */
249 size_t actions_len
; /* Number of bytes in actions[]. */
250 struct nlattr
*actions
; /* Datapath actions. */
251 tag_type tags
; /* Tags (set only by hooks). */
252 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
255 static struct facet
*facet_create(struct ofproto
*, struct rule
*,
257 const struct ofpbuf
*packet
);
258 static void facet_remove(struct ofproto
*, struct facet
*);
259 static void facet_free(struct facet
*);
261 static struct facet
*facet_lookup_valid(struct ofproto
*, const struct flow
*);
262 static bool facet_revalidate(struct ofproto
*, struct facet
*);
264 static void facet_install(struct ofproto
*, struct facet
*, bool zero_stats
);
265 static void facet_uninstall(struct ofproto
*, struct facet
*);
266 static void facet_flush_stats(struct ofproto
*, struct facet
*);
268 static void facet_make_actions(struct ofproto
*, struct facet
*,
269 const struct ofpbuf
*packet
);
270 static void facet_update_stats(struct ofproto
*, struct facet
*,
271 const struct dpif_flow_stats
*);
272 static void facet_push_stats(struct ofproto
*, struct facet
*);
274 /* ofproto supports two kinds of OpenFlow connections:
276 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
277 * maintains persistent connections to these controllers and by default
278 * sends them asynchronous messages such as packet-ins.
280 * - "Service" connections, e.g. from ovs-ofctl. When these connections
281 * drop, it is the other side's responsibility to reconnect them if
282 * necessary. ofproto does not send them asynchronous messages by default.
284 * Currently, active (tcp, ssl, unix) connections are always "primary"
285 * connections and passive (ptcp, pssl, punix) connections are always "service"
286 * connections. There is no inherent reason for this, but it reflects the
290 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
291 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
294 /* A listener for incoming OpenFlow "service" connections. */
296 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
297 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
299 /* These are not used by ofservice directly. They are settings for
300 * accepted "struct ofconn"s from the pvconn. */
301 int probe_interval
; /* Max idle time before probing, in seconds. */
302 int rate_limit
; /* Max packet-in rate in packets per second. */
303 int burst_limit
; /* Limit on accumulating packet credits. */
306 static struct ofservice
*ofservice_lookup(struct ofproto
*,
308 static int ofservice_create(struct ofproto
*,
309 const struct ofproto_controller
*);
310 static void ofservice_reconfigure(struct ofservice
*,
311 const struct ofproto_controller
*);
312 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
314 /* An OpenFlow connection. */
316 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
317 struct list node
; /* In struct ofproto's "all_conns" list. */
318 struct rconn
*rconn
; /* OpenFlow connection. */
319 enum ofconn_type type
; /* Type. */
320 enum nx_flow_format flow_format
; /* Currently selected flow format. */
322 /* OFPT_PACKET_IN related data. */
323 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
324 #define N_SCHEDULERS 2
325 struct pinsched
*schedulers
[N_SCHEDULERS
];
326 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
327 int miss_send_len
; /* Bytes to send of buffered packets. */
329 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
330 * requests, and the maximum number before we stop reading OpenFlow
332 #define OFCONN_REPLY_MAX 100
333 struct rconn_packet_counter
*reply_counter
;
335 /* type == OFCONN_PRIMARY only. */
336 enum nx_role role
; /* Role. */
337 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
338 enum ofproto_band band
; /* In-band or out-of-band? */
342 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
344 static void ofconn_destroy(struct ofconn
*);
345 static void ofconn_run(struct ofconn
*);
346 static void ofconn_wait(struct ofconn
*);
347 static bool ofconn_receives_async_msgs(const struct ofconn
*);
348 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
349 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
351 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
352 struct rconn_packet_counter
*counter
);
354 static void send_packet_in(struct ofproto
*, struct dpif_upcall
*,
355 const struct flow
*, bool clone
);
356 static void do_send_packet_in(struct ofpbuf
*ofp_packet_in
, void *ofconn
);
360 uint64_t datapath_id
; /* Datapath ID. */
361 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
362 char *mfr_desc
; /* Manufacturer. */
363 char *hw_desc
; /* Hardware. */
364 char *sw_desc
; /* Software version. */
365 char *serial_desc
; /* Serial number. */
366 char *dp_desc
; /* Datapath description. */
370 struct netdev_monitor
*netdev_monitor
;
371 struct hmap ports
; /* Contains "struct ofport"s. */
372 struct shash port_by_name
;
376 struct fail_open
*fail_open
;
377 struct netflow
*netflow
;
378 struct ofproto_sflow
*sflow
;
380 /* In-band control. */
381 struct in_band
*in_band
;
382 long long int next_in_band_update
;
383 struct sockaddr_in
*extra_in_band_remotes
;
384 size_t n_extra_remotes
;
388 struct classifier cls
;
389 long long int next_expiration
;
393 bool need_revalidate
;
394 struct tag_set revalidate_set
;
396 /* OpenFlow connections. */
397 struct hmap controllers
; /* Controller "struct ofconn"s. */
398 struct list all_conns
; /* Contains "struct ofconn"s. */
399 enum ofproto_fail_mode fail_mode
;
401 /* OpenFlow listeners. */
402 struct hmap services
; /* Contains "struct ofservice"s. */
403 struct pvconn
**snoops
;
406 /* Hooks for ovs-vswitchd. */
407 const struct ofhooks
*ofhooks
;
410 /* Used by default ofhooks. */
411 struct mac_learning
*ml
;
414 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
415 static struct shash all_ofprotos
= SHASH_INITIALIZER(&all_ofprotos
);
417 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
419 static const struct ofhooks default_ofhooks
;
421 static uint64_t pick_datapath_id(const struct ofproto
*);
422 static uint64_t pick_fallback_dpid(void);
424 static int ofproto_expire(struct ofproto
*);
425 static void flow_push_stats(struct ofproto
*, const struct rule
*,
426 struct flow
*, uint64_t packets
, uint64_t bytes
,
429 static void handle_upcall(struct ofproto
*, struct dpif_upcall
*);
431 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
433 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
434 static void update_port(struct ofproto
*, const char *devname
);
435 static int init_ports(struct ofproto
*);
436 static void reinit_ports(struct ofproto
*);
438 static void ofproto_unixctl_init(void);
441 ofproto_create(const char *datapath
, const char *datapath_type
,
442 const struct ofhooks
*ofhooks
, void *aux
,
443 struct ofproto
**ofprotop
)
451 ofproto_unixctl_init();
453 /* Connect to datapath and start listening for messages. */
454 error
= dpif_open(datapath
, datapath_type
, &dpif
);
456 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
459 error
= dpif_recv_set_mask(dpif
,
460 ((1u << DPIF_UC_MISS
) |
461 (1u << DPIF_UC_ACTION
) |
462 (1u << DPIF_UC_SAMPLE
)));
464 VLOG_ERR("failed to listen on datapath %s: %s",
465 datapath
, strerror(error
));
469 dpif_flow_flush(dpif
);
470 dpif_recv_purge(dpif
);
472 /* Initialize settings. */
473 p
= xzalloc(sizeof *p
);
474 p
->fallback_dpid
= pick_fallback_dpid();
475 p
->datapath_id
= p
->fallback_dpid
;
476 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
477 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
478 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
479 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
480 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
482 /* Initialize datapath. */
484 p
->netdev_monitor
= netdev_monitor_create();
485 hmap_init(&p
->ports
);
486 shash_init(&p
->port_by_name
);
487 p
->max_ports
= dpif_get_max_ports(dpif
);
489 /* Initialize submodules. */
494 /* Initialize in-band control. */
496 p
->in_band_queue
= -1;
498 /* Initialize flow table. */
499 classifier_init(&p
->cls
);
500 p
->next_expiration
= time_msec() + 1000;
502 /* Initialize facet table. */
503 hmap_init(&p
->facets
);
504 p
->need_revalidate
= false;
505 tag_set_init(&p
->revalidate_set
);
507 /* Initialize OpenFlow connections. */
508 list_init(&p
->all_conns
);
509 hmap_init(&p
->controllers
);
510 hmap_init(&p
->services
);
514 /* Initialize hooks. */
516 p
->ofhooks
= ofhooks
;
520 p
->ofhooks
= &default_ofhooks
;
522 p
->ml
= mac_learning_create();
525 /* Pick final datapath ID. */
526 p
->datapath_id
= pick_datapath_id(p
);
527 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
529 shash_add_once(&all_ofprotos
, dpif_name(p
->dpif
), p
);
536 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
538 uint64_t old_dpid
= p
->datapath_id
;
539 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
540 if (p
->datapath_id
!= old_dpid
) {
541 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
543 /* Force all active connections to reconnect, since there is no way to
544 * notify a controller that the datapath ID has changed. */
545 ofproto_reconnect_controllers(p
);
549 /* Creates a new controller in 'ofproto'. Some of the settings are initially
550 * drawn from 'c', but update_controller() needs to be called later to finish
551 * the new ofconn's configuration. */
553 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
555 char *name
= ofconn_make_name(ofproto
, c
->target
);
556 struct ofconn
*ofconn
;
558 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
559 ofconn
->pktbuf
= pktbuf_create();
560 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
561 rconn_connect(ofconn
->rconn
, c
->target
, name
);
562 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
563 hash_string(c
->target
, 0));
568 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
569 * target (this is done by creating new ofconns and deleting old ones), but it
570 * can update the rest of an ofconn's settings. */
572 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
576 ofconn
->band
= c
->band
;
578 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
580 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
581 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
583 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
587 ofconn_get_target(const struct ofconn
*ofconn
)
589 return rconn_get_target(ofconn
->rconn
);
592 static struct ofconn
*
593 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
595 struct ofconn
*ofconn
;
597 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
598 hash_string(target
, 0), &ofproto
->controllers
) {
599 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
607 update_in_band_remotes(struct ofproto
*ofproto
)
609 const struct ofconn
*ofconn
;
610 struct sockaddr_in
*addrs
;
611 size_t max_addrs
, n_addrs
;
614 /* Allocate enough memory for as many remotes as we could possibly have. */
615 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
616 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
619 /* Add all the remotes. */
620 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
621 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
623 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
627 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
628 if (sin
->sin_addr
.s_addr
) {
629 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
633 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
634 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
637 /* Create or update or destroy in-band. */
639 if (!ofproto
->in_band
) {
640 in_band_create(ofproto
, ofproto
->dpif
, &ofproto
->in_band
);
642 if (ofproto
->in_band
) {
643 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
645 in_band_set_queue(ofproto
->in_band
, ofproto
->in_band_queue
);
646 ofproto
->next_in_band_update
= time_msec() + 1000;
648 in_band_destroy(ofproto
->in_band
);
649 ofproto
->in_band
= NULL
;
657 update_fail_open(struct ofproto
*p
)
659 struct ofconn
*ofconn
;
661 if (!hmap_is_empty(&p
->controllers
)
662 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
663 struct rconn
**rconns
;
667 p
->fail_open
= fail_open_create(p
);
671 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
672 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
673 rconns
[n
++] = ofconn
->rconn
;
676 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
677 /* p->fail_open takes ownership of 'rconns'. */
679 fail_open_destroy(p
->fail_open
);
685 ofproto_set_controllers(struct ofproto
*p
,
686 const struct ofproto_controller
*controllers
,
687 size_t n_controllers
)
689 struct shash new_controllers
;
690 struct ofconn
*ofconn
, *next_ofconn
;
691 struct ofservice
*ofservice
, *next_ofservice
;
694 /* Create newly configured controllers and services.
695 * Create a name to ofproto_controller mapping in 'new_controllers'. */
696 shash_init(&new_controllers
);
697 for (i
= 0; i
< n_controllers
; i
++) {
698 const struct ofproto_controller
*c
= &controllers
[i
];
700 if (!vconn_verify_name(c
->target
)) {
701 if (!find_controller_by_target(p
, c
->target
)) {
702 add_controller(p
, c
);
704 } else if (!pvconn_verify_name(c
->target
)) {
705 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
709 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
710 dpif_name(p
->dpif
), c
->target
);
714 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
717 /* Delete controllers that are no longer configured.
718 * Update configuration of all now-existing controllers. */
719 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
720 struct ofproto_controller
*c
;
722 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
724 ofconn_destroy(ofconn
);
726 update_controller(ofconn
, c
);
730 /* Delete services that are no longer configured.
731 * Update configuration of all now-existing services. */
732 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
733 struct ofproto_controller
*c
;
735 c
= shash_find_data(&new_controllers
,
736 pvconn_get_name(ofservice
->pvconn
));
738 ofservice_destroy(p
, ofservice
);
740 ofservice_reconfigure(ofservice
, c
);
744 shash_destroy(&new_controllers
);
746 update_in_band_remotes(p
);
751 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
753 p
->fail_mode
= fail_mode
;
757 /* Drops the connections between 'ofproto' and all of its controllers, forcing
758 * them to reconnect. */
760 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
762 struct ofconn
*ofconn
;
764 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
765 rconn_reconnect(ofconn
->rconn
);
770 any_extras_changed(const struct ofproto
*ofproto
,
771 const struct sockaddr_in
*extras
, size_t n
)
775 if (n
!= ofproto
->n_extra_remotes
) {
779 for (i
= 0; i
< n
; i
++) {
780 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
781 const struct sockaddr_in
*new = &extras
[i
];
783 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
784 old
->sin_port
!= new->sin_port
) {
792 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
793 * in-band control should guarantee access, in the same way that in-band
794 * control guarantees access to OpenFlow controllers. */
796 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
797 const struct sockaddr_in
*extras
, size_t n
)
799 if (!any_extras_changed(ofproto
, extras
, n
)) {
803 free(ofproto
->extra_in_band_remotes
);
804 ofproto
->n_extra_remotes
= n
;
805 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
807 update_in_band_remotes(ofproto
);
810 /* Sets the OpenFlow queue used by flows set up by in-band control on
811 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
812 * flows will use the default queue. */
814 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
816 if (queue_id
!= ofproto
->in_band_queue
) {
817 ofproto
->in_band_queue
= queue_id
;
818 update_in_band_remotes(ofproto
);
823 ofproto_set_desc(struct ofproto
*p
,
824 const char *mfr_desc
, const char *hw_desc
,
825 const char *sw_desc
, const char *serial_desc
,
828 struct ofp_desc_stats
*ods
;
831 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
832 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
833 sizeof ods
->mfr_desc
);
836 p
->mfr_desc
= xstrdup(mfr_desc
);
839 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
840 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
841 sizeof ods
->hw_desc
);
844 p
->hw_desc
= xstrdup(hw_desc
);
847 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
848 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
849 sizeof ods
->sw_desc
);
852 p
->sw_desc
= xstrdup(sw_desc
);
855 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
856 VLOG_WARN("truncating serial_desc, must be less than %zu "
858 sizeof ods
->serial_num
);
860 free(p
->serial_desc
);
861 p
->serial_desc
= xstrdup(serial_desc
);
864 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
865 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
866 sizeof ods
->dp_desc
);
869 p
->dp_desc
= xstrdup(dp_desc
);
874 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
875 const struct svec
*svec
)
877 struct pvconn
**pvconns
= *pvconnsp
;
878 size_t n_pvconns
= *n_pvconnsp
;
882 for (i
= 0; i
< n_pvconns
; i
++) {
883 pvconn_close(pvconns
[i
]);
887 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
889 for (i
= 0; i
< svec
->n
; i
++) {
890 const char *name
= svec
->names
[i
];
891 struct pvconn
*pvconn
;
894 error
= pvconn_open(name
, &pvconn
);
896 pvconns
[n_pvconns
++] = pvconn
;
898 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
906 *n_pvconnsp
= n_pvconns
;
912 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
914 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
918 ofproto_set_netflow(struct ofproto
*ofproto
,
919 const struct netflow_options
*nf_options
)
921 if (nf_options
&& nf_options
->collectors
.n
) {
922 if (!ofproto
->netflow
) {
923 ofproto
->netflow
= netflow_create();
925 return netflow_set_options(ofproto
->netflow
, nf_options
);
927 netflow_destroy(ofproto
->netflow
);
928 ofproto
->netflow
= NULL
;
934 ofproto_set_sflow(struct ofproto
*ofproto
,
935 const struct ofproto_sflow_options
*oso
)
937 struct ofproto_sflow
*os
= ofproto
->sflow
;
940 struct ofport
*ofport
;
942 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
943 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
944 ofproto_sflow_add_port(os
, ofport
->odp_port
,
945 netdev_get_name(ofport
->netdev
));
948 ofproto_sflow_set_options(os
, oso
);
950 ofproto_sflow_destroy(os
);
951 ofproto
->sflow
= NULL
;
955 /* Connectivity Fault Management configuration. */
957 /* Clears the CFM configuration from 'port_no' on 'ofproto'. */
959 ofproto_iface_clear_cfm(struct ofproto
*ofproto
, uint32_t port_no
)
961 struct ofport
*ofport
= get_port(ofproto
, port_no
);
962 if (ofport
&& ofport
->cfm
){
963 cfm_destroy(ofport
->cfm
);
968 /* Configures connectivity fault management on 'port_no' in 'ofproto'. Takes
969 * basic configuration from the configuration members in 'cfm', and the set of
970 * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'.
971 * Ignores the statistics members of 'cfm'.
973 * This function has no effect if 'ofproto' does not have a port 'port_no'. */
975 ofproto_iface_set_cfm(struct ofproto
*ofproto
, uint32_t port_no
,
976 const struct cfm
*cfm
,
977 const uint16_t *remote_mps
, size_t n_remote_mps
)
979 struct ofport
*ofport
;
981 ofport
= get_port(ofproto
, port_no
);
983 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32
,
984 dpif_name(ofproto
->dpif
), port_no
);
989 ofport
->cfm
= cfm_create();
992 ofport
->cfm
->mpid
= cfm
->mpid
;
993 ofport
->cfm
->interval
= cfm
->interval
;
994 memcpy(ofport
->cfm
->maid
, cfm
->maid
, CCM_MAID_LEN
);
996 cfm_update_remote_mps(ofport
->cfm
, remote_mps
, n_remote_mps
);
998 if (!cfm_configure(ofport
->cfm
)) {
999 VLOG_WARN("%s: CFM configuration on port %"PRIu32
" (%s) failed",
1000 dpif_name(ofproto
->dpif
), port_no
,
1001 netdev_get_name(ofport
->netdev
));
1002 cfm_destroy(ofport
->cfm
);
1007 /* Returns the connectivity fault management object associated with 'port_no'
1008 * within 'ofproto', or a null pointer if 'ofproto' does not have a port
1009 * 'port_no' or if that port does not have CFM configured. The caller must not
1010 * modify or destroy the returned object. */
1012 ofproto_iface_get_cfm(struct ofproto
*ofproto
, uint32_t port_no
)
1014 struct ofport
*ofport
= get_port(ofproto
, port_no
);
1015 return ofport
? ofport
->cfm
: NULL
;
1019 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
1021 return ofproto
->datapath_id
;
1025 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
1027 return !hmap_is_empty(&ofproto
->controllers
);
1030 enum ofproto_fail_mode
1031 ofproto_get_fail_mode(const struct ofproto
*p
)
1033 return p
->fail_mode
;
1037 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
1041 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
1042 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
1047 ofproto_destroy(struct ofproto
*p
)
1049 struct ofservice
*ofservice
, *next_ofservice
;
1050 struct ofconn
*ofconn
, *next_ofconn
;
1051 struct ofport
*ofport
, *next_ofport
;
1058 shash_find_and_delete(&all_ofprotos
, dpif_name(p
->dpif
));
1060 /* Destroy fail-open and in-band early, since they touch the classifier. */
1061 fail_open_destroy(p
->fail_open
);
1062 p
->fail_open
= NULL
;
1064 in_band_destroy(p
->in_band
);
1066 free(p
->extra_in_band_remotes
);
1068 ofproto_flush_flows(p
);
1069 classifier_destroy(&p
->cls
);
1070 hmap_destroy(&p
->facets
);
1072 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1073 ofconn_destroy(ofconn
);
1075 hmap_destroy(&p
->controllers
);
1077 dpif_close(p
->dpif
);
1078 netdev_monitor_destroy(p
->netdev_monitor
);
1079 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1080 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1081 ofport_free(ofport
);
1083 shash_destroy(&p
->port_by_name
);
1085 netflow_destroy(p
->netflow
);
1086 ofproto_sflow_destroy(p
->sflow
);
1088 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
1089 ofservice_destroy(p
, ofservice
);
1091 hmap_destroy(&p
->services
);
1093 for (i
= 0; i
< p
->n_snoops
; i
++) {
1094 pvconn_close(p
->snoops
[i
]);
1098 mac_learning_destroy(p
->ml
);
1103 free(p
->serial_desc
);
1106 hmap_destroy(&p
->ports
);
1112 ofproto_run(struct ofproto
*p
)
1114 int error
= ofproto_run1(p
);
1116 error
= ofproto_run2(p
, false);
1122 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1124 if (error
== ENOBUFS
) {
1125 reinit_ports(ofproto
);
1126 } else if (!error
) {
1127 update_port(ofproto
, devname
);
1132 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1133 * means that 'ofconn' is more interesting for monitoring than a lower return
1136 snoop_preference(const struct ofconn
*ofconn
)
1138 switch (ofconn
->role
) {
1139 case NX_ROLE_MASTER
:
1146 /* Shouldn't happen. */
1151 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1152 * Connects this vconn to a controller. */
1154 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1156 struct ofconn
*ofconn
, *best
;
1158 /* Pick a controller for monitoring. */
1160 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1161 if (ofconn
->type
== OFCONN_PRIMARY
1162 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1168 rconn_add_monitor(best
->rconn
, vconn
);
1170 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1176 ofproto_run1(struct ofproto
*p
)
1178 struct ofconn
*ofconn
, *next_ofconn
;
1179 struct ofservice
*ofservice
;
1180 struct ofport
*ofport
;
1185 if (shash_is_empty(&p
->port_by_name
)) {
1189 for (i
= 0; i
< 50; i
++) {
1190 struct dpif_upcall packet
;
1192 error
= dpif_recv(p
->dpif
, &packet
);
1194 if (error
== ENODEV
) {
1195 /* Someone destroyed the datapath behind our back. The caller
1196 * better destroy us and give up, because we're just going to
1197 * spin from here on out. */
1198 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1199 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1200 dpif_name(p
->dpif
));
1206 handle_upcall(p
, &packet
);
1209 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1210 process_port_change(p
, error
, devname
);
1212 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1213 &devname
)) != EAGAIN
) {
1214 process_port_change(p
, error
, devname
);
1217 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1218 ofport_run(p
, ofport
);
1222 if (time_msec() >= p
->next_in_band_update
) {
1223 update_in_band_remotes(p
);
1225 in_band_run(p
->in_band
);
1228 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1232 /* Fail-open maintenance. Do this after processing the ofconns since
1233 * fail-open checks the status of the controller rconn. */
1235 fail_open_run(p
->fail_open
);
1238 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1239 struct vconn
*vconn
;
1242 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1244 struct rconn
*rconn
;
1247 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1248 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1249 rconn_connect_unreliably(rconn
, vconn
, name
);
1252 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1253 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1254 ofservice
->burst_limit
);
1255 } else if (retval
!= EAGAIN
) {
1256 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1260 for (i
= 0; i
< p
->n_snoops
; i
++) {
1261 struct vconn
*vconn
;
1264 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1266 add_snooper(p
, vconn
);
1267 } else if (retval
!= EAGAIN
) {
1268 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1272 if (time_msec() >= p
->next_expiration
) {
1273 int delay
= ofproto_expire(p
);
1274 p
->next_expiration
= time_msec() + delay
;
1275 COVERAGE_INC(ofproto_expiration
);
1279 netflow_run(p
->netflow
);
1282 ofproto_sflow_run(p
->sflow
);
1289 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1291 /* Figure out what we need to revalidate now, if anything. */
1292 struct tag_set revalidate_set
= p
->revalidate_set
;
1293 if (p
->need_revalidate
) {
1294 revalidate_all
= true;
1297 /* Clear the revalidation flags. */
1298 tag_set_init(&p
->revalidate_set
);
1299 p
->need_revalidate
= false;
1301 /* Now revalidate if there's anything to do. */
1302 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1303 struct facet
*facet
, *next
;
1305 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1307 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1308 facet_revalidate(p
, facet
);
1317 ofproto_wait(struct ofproto
*p
)
1319 struct ofservice
*ofservice
;
1320 struct ofconn
*ofconn
;
1321 struct ofport
*ofport
;
1324 dpif_recv_wait(p
->dpif
);
1325 dpif_port_poll_wait(p
->dpif
);
1326 netdev_monitor_poll_wait(p
->netdev_monitor
);
1327 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1328 ofport_wait(ofport
);
1330 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1331 ofconn_wait(ofconn
);
1334 poll_timer_wait_until(p
->next_in_band_update
);
1335 in_band_wait(p
->in_band
);
1338 fail_open_wait(p
->fail_open
);
1341 ofproto_sflow_wait(p
->sflow
);
1343 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1344 poll_immediate_wake();
1346 if (p
->need_revalidate
) {
1347 /* Shouldn't happen, but if it does just go around again. */
1348 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1349 poll_immediate_wake();
1350 } else if (p
->next_expiration
!= LLONG_MAX
) {
1351 poll_timer_wait_until(p
->next_expiration
);
1353 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1354 pvconn_wait(ofservice
->pvconn
);
1356 for (i
= 0; i
< p
->n_snoops
; i
++) {
1357 pvconn_wait(p
->snoops
[i
]);
1362 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1364 tag_set_add(&ofproto
->revalidate_set
, tag
);
1368 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1370 return &ofproto
->revalidate_set
;
1374 ofproto_is_alive(const struct ofproto
*p
)
1376 return !hmap_is_empty(&p
->controllers
);
1380 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1383 const struct ofconn
*ofconn
;
1387 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
1388 const struct rconn
*rconn
= ofconn
->rconn
;
1389 time_t now
= time_now();
1390 time_t last_connection
= rconn_get_last_connection(rconn
);
1391 time_t last_disconnect
= rconn_get_last_disconnect(rconn
);
1392 const int last_error
= rconn_get_last_error(rconn
);
1393 struct ofproto_controller_info
*cinfo
= xmalloc(sizeof *cinfo
);
1395 shash_add(info
, rconn_get_target(rconn
), cinfo
);
1397 cinfo
->is_connected
= rconn_is_connected(rconn
);
1398 cinfo
->role
= ofconn
->role
;
1403 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "last_error";
1404 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1405 xstrdup(ovs_retval_to_string(last_error
));
1408 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "state";
1409 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1410 xstrdup(rconn_get_state(rconn
));
1412 if (last_connection
!= TIME_MIN
) {
1413 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "sec_since_connect";
1414 cinfo
->pairs
.values
[cinfo
->pairs
.n
++]
1415 = xasprintf("%ld", (long int) (now
- last_connection
));
1418 if (last_disconnect
!= TIME_MIN
) {
1419 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "sec_since_disconnect";
1420 cinfo
->pairs
.values
[cinfo
->pairs
.n
++]
1421 = xasprintf("%ld", (long int) (now
- last_disconnect
));
1427 ofproto_free_ofproto_controller_info(struct shash
*info
)
1429 struct shash_node
*node
;
1431 SHASH_FOR_EACH (node
, info
) {
1432 struct ofproto_controller_info
*cinfo
= node
->data
;
1433 while (cinfo
->pairs
.n
) {
1434 free((char *) cinfo
->pairs
.values
[--cinfo
->pairs
.n
]);
1438 shash_destroy(info
);
1441 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1443 * This is almost the same as calling dpif_port_del() directly on the
1444 * datapath, but it also makes 'ofproto' close its open netdev for the port
1445 * (if any). This makes it possible to create a new netdev of a different
1446 * type under the same name, which otherwise the netdev library would refuse
1447 * to do because of the conflict. (The netdev would eventually get closed on
1448 * the next trip through ofproto_run(), but this interface is more direct.)
1450 * Returns 0 if successful, otherwise a positive errno. */
1452 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1454 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1455 const char *name
= ofport
? ofport
->opp
.name
: "<unknown>";
1458 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1460 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1461 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1462 } else if (ofport
) {
1463 /* 'name' is ofport->opp.name and update_port() is going to destroy
1464 * 'ofport'. Just in case update_port() refers to 'name' after it
1465 * destroys 'ofport', make a copy of it around the update_port()
1467 char *devname
= xstrdup(name
);
1468 update_port(ofproto
, devname
);
1474 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1475 * true if 'odp_port' exists and should be included, false otherwise. */
1477 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1479 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1480 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1483 /* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the
1484 * packet will not have any 802.1Q hader; if it is nonzero, then the packet
1485 * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
1487 * Returns 0 if successful, otherwise a positive errno value. */
1489 ofproto_send_packet(struct ofproto
*ofproto
,
1490 uint32_t port_no
, uint16_t vlan_tci
,
1491 const struct ofpbuf
*packet
)
1493 struct ofpbuf odp_actions
;
1496 ofpbuf_init(&odp_actions
, 32);
1497 if (vlan_tci
!= 0) {
1498 nl_msg_put_u32(&odp_actions
, ODP_ACTION_ATTR_SET_DL_TCI
,
1499 ntohs(vlan_tci
& ~VLAN_CFI
));
1501 nl_msg_put_u32(&odp_actions
, ODP_ACTION_ATTR_OUTPUT
, port_no
);
1502 error
= dpif_execute(ofproto
->dpif
, odp_actions
.data
, odp_actions
.size
,
1504 ofpbuf_uninit(&odp_actions
);
1507 VLOG_WARN_RL(&rl
, "%s: failed to send packet on port %"PRIu32
" (%s)",
1508 dpif_name(ofproto
->dpif
), port_no
, strerror(error
));
1513 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1514 * performs the 'n_actions' actions in 'actions'. The new flow will not
1517 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1518 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1519 * controllers; otherwise, it will be hidden.
1521 * The caller retains ownership of 'cls_rule' and 'actions'. */
1523 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1524 const union ofp_action
*actions
, size_t n_actions
)
1527 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1528 rule_insert(p
, rule
);
1532 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1536 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1539 rule_remove(ofproto
, rule
);
1544 ofproto_flush_flows(struct ofproto
*ofproto
)
1546 struct facet
*facet
, *next_facet
;
1547 struct rule
*rule
, *next_rule
;
1548 struct cls_cursor cursor
;
1550 COVERAGE_INC(ofproto_flush
);
1552 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1553 /* Mark the facet as not installed so that facet_remove() doesn't
1554 * bother trying to uninstall it. There is no point in uninstalling it
1555 * individually since we are about to blow away all the facets with
1556 * dpif_flow_flush(). */
1557 facet
->installed
= false;
1558 facet
->dp_packet_count
= 0;
1559 facet
->dp_byte_count
= 0;
1560 facet_remove(ofproto
, facet
);
1563 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1564 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1565 rule_remove(ofproto
, rule
);
1568 dpif_flow_flush(ofproto
->dpif
);
1569 if (ofproto
->in_band
) {
1570 in_band_flushed(ofproto
->in_band
);
1572 if (ofproto
->fail_open
) {
1573 fail_open_flushed(ofproto
->fail_open
);
1578 reinit_ports(struct ofproto
*p
)
1580 struct dpif_port_dump dump
;
1581 struct shash_node
*node
;
1582 struct shash devnames
;
1583 struct ofport
*ofport
;
1584 struct dpif_port dpif_port
;
1586 COVERAGE_INC(ofproto_reinit_ports
);
1588 shash_init(&devnames
);
1589 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1590 shash_add_once (&devnames
, ofport
->opp
.name
, NULL
);
1592 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1593 shash_add_once (&devnames
, dpif_port
.name
, NULL
);
1596 SHASH_FOR_EACH (node
, &devnames
) {
1597 update_port(p
, node
->name
);
1599 shash_destroy(&devnames
);
1602 static struct ofport
*
1603 make_ofport(const struct dpif_port
*dpif_port
)
1605 struct netdev_options netdev_options
;
1606 enum netdev_flags flags
;
1607 struct ofport
*ofport
;
1608 struct netdev
*netdev
;
1611 memset(&netdev_options
, 0, sizeof netdev_options
);
1612 netdev_options
.name
= dpif_port
->name
;
1613 netdev_options
.type
= dpif_port
->type
;
1614 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1616 error
= netdev_open(&netdev_options
, &netdev
);
1618 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1619 "cannot be opened (%s)",
1620 dpif_port
->name
, dpif_port
->port_no
,
1621 dpif_port
->name
, strerror(error
));
1625 ofport
= xzalloc(sizeof *ofport
);
1626 ofport
->netdev
= netdev
;
1627 ofport
->odp_port
= dpif_port
->port_no
;
1628 ofport
->opp
.port_no
= odp_port_to_ofp_port(dpif_port
->port_no
);
1629 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1630 ovs_strlcpy(ofport
->opp
.name
, dpif_port
->name
, sizeof ofport
->opp
.name
);
1632 netdev_get_flags(netdev
, &flags
);
1633 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1635 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1637 netdev_get_features(netdev
,
1638 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1639 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1644 ofport_conflicts(const struct ofproto
*p
, const struct dpif_port
*dpif_port
)
1646 if (get_port(p
, dpif_port
->port_no
)) {
1647 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1648 dpif_port
->port_no
);
1650 } else if (shash_find(&p
->port_by_name
, dpif_port
->name
)) {
1651 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1660 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1662 const struct ofp_phy_port
*a
= &a_
->opp
;
1663 const struct ofp_phy_port
*b
= &b_
->opp
;
1665 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1666 return (a
->port_no
== b
->port_no
1667 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1668 && !strcmp(a
->name
, b
->name
)
1669 && a
->state
== b
->state
1670 && a
->config
== b
->config
1671 && a
->curr
== b
->curr
1672 && a
->advertised
== b
->advertised
1673 && a
->supported
== b
->supported
1674 && a
->peer
== b
->peer
);
1678 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1681 /* XXX Should limit the number of queued port status change messages. */
1682 struct ofconn
*ofconn
;
1683 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1684 struct ofp_port_status
*ops
;
1687 /* Primary controllers, even slaves, should always get port status
1688 updates. Otherwise obey ofconn_receives_async_msgs(). */
1689 if (ofconn
->type
!= OFCONN_PRIMARY
1690 && !ofconn_receives_async_msgs(ofconn
)) {
1694 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1695 ops
->reason
= reason
;
1696 ops
->desc
= ofport
->opp
;
1697 hton_ofp_phy_port(&ops
->desc
);
1698 queue_tx(b
, ofconn
, NULL
);
1703 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1705 const char *netdev_name
= ofport
->opp
.name
;
1707 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1708 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1709 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1711 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1716 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1718 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1719 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1720 shash_delete(&p
->port_by_name
,
1721 shash_find(&p
->port_by_name
, ofport
->opp
.name
));
1723 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1728 ofport_run(struct ofproto
*ofproto
, struct ofport
*ofport
)
1731 cfm_run(ofport
->cfm
);
1733 if (cfm_should_send_ccm(ofport
->cfm
)) {
1734 struct ofpbuf packet
;
1737 ofpbuf_init(&packet
, 0);
1738 ccm
= compose_packet(&packet
, eth_addr_ccm
, ofport
->opp
.hw_addr
,
1739 ETH_TYPE_CFM
, sizeof *ccm
);
1740 cfm_compose_ccm(ofport
->cfm
, ccm
);
1741 ofproto_send_packet(ofproto
, ofport
->odp_port
, 0, &packet
);
1742 ofpbuf_uninit(&packet
);
1748 ofport_wait(struct ofport
*ofport
)
1751 cfm_wait(ofport
->cfm
);
1756 ofport_free(struct ofport
*ofport
)
1759 cfm_destroy(ofport
->cfm
);
1760 netdev_close(ofport
->netdev
);
1765 static struct ofport
*
1766 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1768 struct ofport
*port
;
1770 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1771 hash_int(odp_port
, 0), &ofproto
->ports
) {
1772 if (port
->odp_port
== odp_port
) {
1780 update_port(struct ofproto
*p
, const char *devname
)
1782 struct dpif_port dpif_port
;
1783 struct ofport
*old_ofport
;
1784 struct ofport
*new_ofport
;
1787 COVERAGE_INC(ofproto_update_port
);
1789 /* Query the datapath for port information. */
1790 error
= dpif_port_query_by_name(p
->dpif
, devname
, &dpif_port
);
1792 /* Find the old ofport. */
1793 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1796 /* There's no port named 'devname' but there might be a port with
1797 * the same port number. This could happen if a port is deleted
1798 * and then a new one added in its place very quickly, or if a port
1799 * is renamed. In the former case we want to send an OFPPR_DELETE
1800 * and an OFPPR_ADD, and in the latter case we want to send a
1801 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1802 * the old port's ifindex against the new port, or perhaps less
1803 * reliably but more portably by comparing the old port's MAC
1804 * against the new port's MAC. However, this code isn't that smart
1805 * and always sends an OFPPR_MODIFY (XXX). */
1806 old_ofport
= get_port(p
, dpif_port
.port_no
);
1808 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1809 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1810 "%s", strerror(error
));
1814 /* Create a new ofport. */
1815 new_ofport
= !error
? make_ofport(&dpif_port
) : NULL
;
1817 /* Eliminate a few pathological cases. */
1818 if (!old_ofport
&& !new_ofport
) {
1820 } else if (old_ofport
&& new_ofport
) {
1821 /* Most of the 'config' bits are OpenFlow soft state, but
1822 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1823 * OpenFlow bits from old_ofport. (make_ofport() only sets
1824 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1825 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1827 if (ofport_equal(old_ofport
, new_ofport
)) {
1828 /* False alarm--no change. */
1829 ofport_free(new_ofport
);
1834 /* Now deal with the normal cases. */
1836 ofport_remove(p
, old_ofport
);
1839 ofport_install(p
, new_ofport
);
1841 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1842 (!old_ofport
? OFPPR_ADD
1843 : !new_ofport
? OFPPR_DELETE
1845 ofport_free(old_ofport
);
1848 dpif_port_destroy(&dpif_port
);
1852 init_ports(struct ofproto
*p
)
1854 struct dpif_port_dump dump
;
1855 struct dpif_port dpif_port
;
1857 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1858 if (!ofport_conflicts(p
, &dpif_port
)) {
1859 struct ofport
*ofport
= make_ofport(&dpif_port
);
1861 ofport_install(p
, ofport
);
1869 static struct ofconn
*
1870 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1872 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1873 ofconn
->ofproto
= p
;
1874 list_push_back(&p
->all_conns
, &ofconn
->node
);
1875 ofconn
->rconn
= rconn
;
1876 ofconn
->type
= type
;
1877 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1878 ofconn
->role
= NX_ROLE_OTHER
;
1879 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1880 ofconn
->pktbuf
= NULL
;
1881 ofconn
->miss_send_len
= 0;
1882 ofconn
->reply_counter
= rconn_packet_counter_create ();
1887 ofconn_destroy(struct ofconn
*ofconn
)
1889 if (ofconn
->type
== OFCONN_PRIMARY
) {
1890 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1893 list_remove(&ofconn
->node
);
1894 rconn_destroy(ofconn
->rconn
);
1895 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1896 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1897 pktbuf_destroy(ofconn
->pktbuf
);
1902 ofconn_run(struct ofconn
*ofconn
)
1904 struct ofproto
*p
= ofconn
->ofproto
;
1908 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1909 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1912 rconn_run(ofconn
->rconn
);
1914 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1915 /* Limit the number of iterations to prevent other tasks from
1917 for (iteration
= 0; iteration
< 50; iteration
++) {
1918 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1923 fail_open_maybe_recover(p
->fail_open
);
1925 handle_openflow(ofconn
, of_msg
);
1926 ofpbuf_delete(of_msg
);
1930 if (!rconn_is_alive(ofconn
->rconn
)) {
1931 ofconn_destroy(ofconn
);
1936 ofconn_wait(struct ofconn
*ofconn
)
1940 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1941 pinsched_wait(ofconn
->schedulers
[i
]);
1943 rconn_run_wait(ofconn
->rconn
);
1944 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1945 rconn_recv_wait(ofconn
->rconn
);
1947 COVERAGE_INC(ofproto_ofconn_stuck
);
1951 /* Returns true if 'ofconn' should receive asynchronous messages. */
1953 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1955 if (ofconn
->type
== OFCONN_PRIMARY
) {
1956 /* Primary controllers always get asynchronous messages unless they
1957 * have configured themselves as "slaves". */
1958 return ofconn
->role
!= NX_ROLE_SLAVE
;
1960 /* Service connections don't get asynchronous messages unless they have
1961 * explicitly asked for them by setting a nonzero miss send length. */
1962 return ofconn
->miss_send_len
> 0;
1966 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1967 * and 'target', suitable for use in log messages for identifying the
1970 * The name is dynamically allocated. The caller should free it (with free())
1971 * when it is no longer needed. */
1973 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1975 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1979 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1983 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1984 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1988 *s
= pinsched_create(rate
, burst
);
1990 pinsched_set_limits(*s
, rate
, burst
);
1993 pinsched_destroy(*s
);
2000 ofservice_reconfigure(struct ofservice
*ofservice
,
2001 const struct ofproto_controller
*c
)
2003 ofservice
->probe_interval
= c
->probe_interval
;
2004 ofservice
->rate_limit
= c
->rate_limit
;
2005 ofservice
->burst_limit
= c
->burst_limit
;
2008 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
2009 * positive errno value. */
2011 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
2013 struct ofservice
*ofservice
;
2014 struct pvconn
*pvconn
;
2017 error
= pvconn_open(c
->target
, &pvconn
);
2022 ofservice
= xzalloc(sizeof *ofservice
);
2023 hmap_insert(&ofproto
->services
, &ofservice
->node
,
2024 hash_string(c
->target
, 0));
2025 ofservice
->pvconn
= pvconn
;
2027 ofservice_reconfigure(ofservice
, c
);
2033 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
2035 hmap_remove(&ofproto
->services
, &ofservice
->node
);
2036 pvconn_close(ofservice
->pvconn
);
2040 /* Finds and returns the ofservice within 'ofproto' that has the given
2041 * 'target', or a null pointer if none exists. */
2042 static struct ofservice
*
2043 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
2045 struct ofservice
*ofservice
;
2047 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
2048 &ofproto
->services
) {
2049 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
2056 /* Returns true if 'rule' should be hidden from the controller.
2058 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2059 * (e.g. by in-band control) and are intentionally hidden from the
2062 rule_is_hidden(const struct rule
*rule
)
2064 return rule
->cr
.priority
> UINT16_MAX
;
2067 /* Creates and returns a new rule initialized as specified.
2069 * The caller is responsible for inserting the rule into the classifier (with
2070 * rule_insert()). */
2071 static struct rule
*
2072 rule_create(const struct cls_rule
*cls_rule
,
2073 const union ofp_action
*actions
, size_t n_actions
,
2074 uint16_t idle_timeout
, uint16_t hard_timeout
,
2075 ovs_be64 flow_cookie
, bool send_flow_removed
)
2077 struct rule
*rule
= xzalloc(sizeof *rule
);
2078 rule
->cr
= *cls_rule
;
2079 rule
->idle_timeout
= idle_timeout
;
2080 rule
->hard_timeout
= hard_timeout
;
2081 rule
->flow_cookie
= flow_cookie
;
2082 rule
->used
= rule
->created
= time_msec();
2083 rule
->send_flow_removed
= send_flow_removed
;
2084 list_init(&rule
->facets
);
2085 if (n_actions
> 0) {
2086 rule
->n_actions
= n_actions
;
2087 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
2093 static struct rule
*
2094 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
2096 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
2100 rule_free(struct rule
*rule
)
2102 free(rule
->actions
);
2106 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2107 * destroying any that no longer has a rule (which is probably all of them).
2109 * The caller must have already removed 'rule' from the classifier. */
2111 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
2113 struct facet
*facet
, *next_facet
;
2114 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
2115 facet_revalidate(ofproto
, facet
);
2120 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2121 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2124 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
2126 const union ofp_action
*oa
;
2127 struct actions_iterator i
;
2129 if (out_port
== htons(OFPP_NONE
)) {
2132 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
2133 oa
= actions_next(&i
)) {
2134 if (action_outputs_to_port(oa
, out_port
)) {
2141 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2142 * 'packet', which arrived on 'in_port'.
2144 * Takes ownership of 'packet'. */
2146 execute_odp_actions(struct ofproto
*ofproto
, const struct flow
*flow
,
2147 const struct nlattr
*odp_actions
, size_t actions_len
,
2148 struct ofpbuf
*packet
)
2150 if (actions_len
== NLA_ALIGN(NLA_HDRLEN
+ sizeof(uint64_t))
2151 && odp_actions
->nla_type
== ODP_ACTION_ATTR_CONTROLLER
) {
2152 /* As an optimization, avoid a round-trip from userspace to kernel to
2153 * userspace. This also avoids possibly filling up kernel packet
2154 * buffers along the way. */
2155 struct dpif_upcall upcall
;
2157 upcall
.type
= DPIF_UC_ACTION
;
2158 upcall
.packet
= packet
;
2161 upcall
.userdata
= nl_attr_get_u64(odp_actions
);
2162 upcall
.sample_pool
= 0;
2163 upcall
.actions
= NULL
;
2164 upcall
.actions_len
= 0;
2166 send_packet_in(ofproto
, &upcall
, flow
, false);
2172 error
= dpif_execute(ofproto
->dpif
, odp_actions
, actions_len
, packet
);
2173 ofpbuf_delete(packet
);
2178 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2179 * statistics appropriately. 'packet' must have at least sizeof(struct
2180 * ofp_packet_in) bytes of headroom.
2182 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2183 * applying flow_extract() to 'packet' would yield the same flow as
2186 * 'facet' must have accurately composed ODP actions; that is, it must not be
2187 * in need of revalidation.
2189 * Takes ownership of 'packet'. */
2191 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2192 struct ofpbuf
*packet
)
2194 struct dpif_flow_stats stats
;
2196 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2198 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2199 stats
.used
= time_msec();
2200 if (execute_odp_actions(ofproto
, &facet
->flow
,
2201 facet
->actions
, facet
->actions_len
, packet
)) {
2202 facet_update_stats(ofproto
, facet
, &stats
);
2206 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2207 * statistics (or the statistics for one of its facets) appropriately.
2208 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2210 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2211 * with statistics for 'packet' either way.
2213 * Takes ownership of 'packet'. */
2215 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2216 struct ofpbuf
*packet
)
2218 struct action_xlate_ctx ctx
;
2219 struct ofpbuf
*odp_actions
;
2220 struct facet
*facet
;
2224 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2226 flow_extract(packet
, 0, in_port
, &flow
);
2228 /* First look for a related facet. If we find one, account it to that. */
2229 facet
= facet_lookup_valid(ofproto
, &flow
);
2230 if (facet
&& facet
->rule
== rule
) {
2231 facet_execute(ofproto
, facet
, packet
);
2235 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2236 * create a new facet for it and use that. */
2237 if (rule_lookup(ofproto
, &flow
) == rule
) {
2238 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2239 facet_execute(ofproto
, facet
, packet
);
2240 facet_install(ofproto
, facet
, true);
2244 /* We can't account anything to a facet. If we were to try, then that
2245 * facet would have a non-matching rule, busting our invariants. */
2246 action_xlate_ctx_init(&ctx
, ofproto
, &flow
, packet
);
2247 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2248 size
= packet
->size
;
2249 if (execute_odp_actions(ofproto
, &flow
, odp_actions
->data
,
2250 odp_actions
->size
, packet
)) {
2251 rule
->used
= time_msec();
2252 rule
->packet_count
++;
2253 rule
->byte_count
+= size
;
2254 flow_push_stats(ofproto
, rule
, &flow
, 1, size
, rule
->used
);
2256 ofpbuf_delete(odp_actions
);
2259 /* Inserts 'rule' into 'p''s flow table. */
2261 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2263 struct rule
*displaced_rule
;
2265 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2266 if (displaced_rule
) {
2267 rule_destroy(p
, displaced_rule
);
2269 p
->need_revalidate
= true;
2272 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2273 * 'flow' and an example 'packet' within that flow.
2275 * The caller must already have determined that no facet with an identical
2276 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2277 * 'ofproto''s classifier table. */
2278 static struct facet
*
2279 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2280 const struct flow
*flow
, const struct ofpbuf
*packet
)
2282 struct facet
*facet
;
2284 facet
= xzalloc(sizeof *facet
);
2285 facet
->used
= time_msec();
2286 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2287 list_push_back(&rule
->facets
, &facet
->list_node
);
2289 facet
->flow
= *flow
;
2290 netflow_flow_init(&facet
->nf_flow
);
2291 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2293 facet_make_actions(ofproto
, facet
, packet
);
2299 facet_free(struct facet
*facet
)
2301 free(facet
->actions
);
2305 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2307 * - Removes 'rule' from the classifier.
2309 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2310 * destroys them), via rule_destroy().
2313 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2315 COVERAGE_INC(ofproto_del_rule
);
2316 ofproto
->need_revalidate
= true;
2317 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2318 rule_destroy(ofproto
, rule
);
2321 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2323 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2324 * rule's statistics, via facet_uninstall().
2326 * - Removes 'facet' from its rule and from ofproto->facets.
2329 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2331 facet_uninstall(ofproto
, facet
);
2332 facet_flush_stats(ofproto
, facet
);
2333 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2334 list_remove(&facet
->list_node
);
2338 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2340 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2341 const struct ofpbuf
*packet
)
2343 const struct rule
*rule
= facet
->rule
;
2344 struct ofpbuf
*odp_actions
;
2345 struct action_xlate_ctx ctx
;
2347 action_xlate_ctx_init(&ctx
, p
, &facet
->flow
, packet
);
2348 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2349 facet
->tags
= ctx
.tags
;
2350 facet
->may_install
= ctx
.may_set_up_flow
;
2351 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2353 if (facet
->actions_len
!= odp_actions
->size
2354 || memcmp(facet
->actions
, odp_actions
->data
, odp_actions
->size
)) {
2355 free(facet
->actions
);
2356 facet
->actions_len
= odp_actions
->size
;
2357 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2360 ofpbuf_delete(odp_actions
);
2364 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
,
2365 const struct nlattr
*actions
, size_t actions_len
,
2366 struct dpif_flow_stats
*stats
)
2368 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2369 enum dpif_flow_put_flags flags
;
2372 flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
2374 flags
|= DPIF_FP_ZERO_STATS
;
2375 facet
->dp_packet_count
= 0;
2376 facet
->dp_byte_count
= 0;
2379 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2380 odp_flow_key_from_flow(&key
, &facet
->flow
);
2381 assert(key
.base
== keybuf
);
2383 return dpif_flow_put(ofproto
->dpif
, flags
, key
.data
, key
.size
,
2384 actions
, actions_len
, stats
);
2387 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2388 * 'zero_stats' is true, clears any existing statistics from the datapath for
2391 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2393 struct dpif_flow_stats stats
;
2395 if (facet
->may_install
2396 && !facet_put__(p
, facet
, facet
->actions
, facet
->actions_len
,
2397 zero_stats
? &stats
: NULL
)) {
2398 facet
->installed
= true;
2402 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2403 * to the accounting hook function in the ofhooks structure. */
2405 facet_account(struct ofproto
*ofproto
,
2406 struct facet
*facet
, uint64_t extra_bytes
)
2408 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2410 if (ofproto
->ofhooks
->account_flow_cb
2411 && total_bytes
> facet
->accounted_bytes
)
2413 ofproto
->ofhooks
->account_flow_cb(
2414 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->actions_len
,
2415 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2416 facet
->accounted_bytes
= total_bytes
;
2420 /* If 'rule' is installed in the datapath, uninstalls it. */
2422 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2424 if (facet
->installed
) {
2425 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2426 struct dpif_flow_stats stats
;
2429 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2430 odp_flow_key_from_flow(&key
, &facet
->flow
);
2431 assert(key
.base
== keybuf
);
2433 if (!dpif_flow_del(p
->dpif
, key
.data
, key
.size
, &stats
)) {
2434 facet_update_stats(p
, facet
, &stats
);
2436 facet
->installed
= false;
2437 facet
->dp_packet_count
= 0;
2438 facet
->dp_byte_count
= 0;
2440 assert(facet
->dp_packet_count
== 0);
2441 assert(facet
->dp_byte_count
== 0);
2445 /* Returns true if the only action for 'facet' is to send to the controller.
2446 * (We don't report NetFlow expiration messages for such facets because they
2447 * are just part of the control logic for the network, not real traffic). */
2449 facet_is_controller_flow(struct facet
*facet
)
2452 && facet
->rule
->n_actions
== 1
2453 && action_outputs_to_port(&facet
->rule
->actions
[0],
2454 htons(OFPP_CONTROLLER
)));
2457 /* Folds all of 'facet''s statistics into its rule. Also updates the
2458 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2459 * 'facet''s statistics in the datapath should have been zeroed and folded into
2460 * its packet and byte counts before this function is called. */
2462 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2464 assert(!facet
->dp_byte_count
);
2465 assert(!facet
->dp_packet_count
);
2467 facet_push_stats(ofproto
, facet
);
2468 facet_account(ofproto
, facet
, 0);
2470 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2471 struct ofexpired expired
;
2472 expired
.flow
= facet
->flow
;
2473 expired
.packet_count
= facet
->packet_count
;
2474 expired
.byte_count
= facet
->byte_count
;
2475 expired
.used
= facet
->used
;
2476 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2479 facet
->rule
->packet_count
+= facet
->packet_count
;
2480 facet
->rule
->byte_count
+= facet
->byte_count
;
2482 /* Reset counters to prevent double counting if 'facet' ever gets
2484 facet
->packet_count
= 0;
2485 facet
->byte_count
= 0;
2486 facet
->rs_packet_count
= 0;
2487 facet
->rs_byte_count
= 0;
2488 facet
->accounted_bytes
= 0;
2490 netflow_flow_clear(&facet
->nf_flow
);
2493 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2494 * Returns it if found, otherwise a null pointer.
2496 * The returned facet might need revalidation; use facet_lookup_valid()
2497 * instead if that is important. */
2498 static struct facet
*
2499 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2501 struct facet
*facet
;
2503 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2505 if (flow_equal(flow
, &facet
->flow
)) {
2513 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2514 * Returns it if found, otherwise a null pointer.
2516 * The returned facet is guaranteed to be valid. */
2517 static struct facet
*
2518 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2520 struct facet
*facet
= facet_find(ofproto
, flow
);
2522 /* The facet we found might not be valid, since we could be in need of
2523 * revalidation. If it is not valid, don't return it. */
2525 && ofproto
->need_revalidate
2526 && !facet_revalidate(ofproto
, facet
)) {
2527 COVERAGE_INC(ofproto_invalidated
);
2534 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2536 * - If the rule found is different from 'facet''s current rule, moves
2537 * 'facet' to the new rule and recompiles its actions.
2539 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2540 * where it is and recompiles its actions anyway.
2542 * - If there is none, destroys 'facet'.
2544 * Returns true if 'facet' still exists, false if it has been destroyed. */
2546 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2548 struct action_xlate_ctx ctx
;
2549 struct ofpbuf
*odp_actions
;
2550 struct rule
*new_rule
;
2551 bool actions_changed
;
2553 COVERAGE_INC(facet_revalidate
);
2555 /* Determine the new rule. */
2556 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2558 /* No new rule, so delete the facet. */
2559 facet_remove(ofproto
, facet
);
2563 /* Calculate new ODP actions.
2565 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2566 * emit a NetFlow expiration and, if so, we need to have the old state
2567 * around to properly compose it. */
2568 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
, NULL
);
2569 odp_actions
= xlate_actions(&ctx
, new_rule
->actions
, new_rule
->n_actions
);
2570 actions_changed
= (facet
->actions_len
!= odp_actions
->size
2571 || memcmp(facet
->actions
, odp_actions
->data
,
2572 facet
->actions_len
));
2574 /* If the ODP actions changed or the installability changed, then we need
2575 * to talk to the datapath. */
2576 if (actions_changed
|| ctx
.may_set_up_flow
!= facet
->installed
) {
2577 if (ctx
.may_set_up_flow
) {
2578 struct dpif_flow_stats stats
;
2580 facet_put__(ofproto
, facet
,
2581 odp_actions
->data
, odp_actions
->size
, &stats
);
2582 facet_update_stats(ofproto
, facet
, &stats
);
2584 facet_uninstall(ofproto
, facet
);
2587 /* The datapath flow is gone or has zeroed stats, so push stats out of
2588 * 'facet' into 'rule'. */
2589 facet_flush_stats(ofproto
, facet
);
2592 /* Update 'facet' now that we've taken care of all the old state. */
2593 facet
->tags
= ctx
.tags
;
2594 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2595 facet
->may_install
= ctx
.may_set_up_flow
;
2596 if (actions_changed
) {
2597 free(facet
->actions
);
2598 facet
->actions_len
= odp_actions
->size
;
2599 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2601 if (facet
->rule
!= new_rule
) {
2602 COVERAGE_INC(facet_changed_rule
);
2603 list_remove(&facet
->list_node
);
2604 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2605 facet
->rule
= new_rule
;
2606 facet
->used
= new_rule
->created
;
2607 facet
->rs_used
= facet
->used
;
2610 ofpbuf_delete(odp_actions
);
2616 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2617 struct rconn_packet_counter
*counter
)
2619 update_openflow_length(msg
);
2620 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2626 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2629 struct ofpbuf
*buf
= ofputil_encode_error_msg(error
, oh
);
2631 COVERAGE_INC(ofproto_error
);
2632 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2637 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2639 opp
->port_no
= htons(opp
->port_no
);
2640 opp
->config
= htonl(opp
->config
);
2641 opp
->state
= htonl(opp
->state
);
2642 opp
->curr
= htonl(opp
->curr
);
2643 opp
->advertised
= htonl(opp
->advertised
);
2644 opp
->supported
= htonl(opp
->supported
);
2645 opp
->peer
= htonl(opp
->peer
);
2649 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2651 queue_tx(make_echo_reply(oh
), ofconn
, ofconn
->reply_counter
);
2656 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2658 struct ofp_switch_features
*osf
;
2660 struct ofport
*port
;
2662 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2663 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2664 osf
->n_buffers
= htonl(pktbuf_capacity());
2666 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2667 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2668 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2669 (1u << OFPAT_SET_VLAN_VID
) |
2670 (1u << OFPAT_SET_VLAN_PCP
) |
2671 (1u << OFPAT_STRIP_VLAN
) |
2672 (1u << OFPAT_SET_DL_SRC
) |
2673 (1u << OFPAT_SET_DL_DST
) |
2674 (1u << OFPAT_SET_NW_SRC
) |
2675 (1u << OFPAT_SET_NW_DST
) |
2676 (1u << OFPAT_SET_NW_TOS
) |
2677 (1u << OFPAT_SET_TP_SRC
) |
2678 (1u << OFPAT_SET_TP_DST
) |
2679 (1u << OFPAT_ENQUEUE
));
2681 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2682 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2685 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2690 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2693 struct ofp_switch_config
*osc
;
2697 /* Figure out flags. */
2698 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2699 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2702 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2703 osc
->flags
= htons(flags
);
2704 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2705 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2711 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
2713 uint16_t flags
= ntohs(osc
->flags
);
2715 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2716 switch (flags
& OFPC_FRAG_MASK
) {
2717 case OFPC_FRAG_NORMAL
:
2718 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2720 case OFPC_FRAG_DROP
:
2721 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2724 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2730 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2735 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2736 struct action_xlate_ctx
*ctx
);
2739 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2741 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2744 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2745 /* Forwarding disabled on port. */
2750 * We don't have an ofport record for this port, but it doesn't hurt to
2751 * allow forwarding to it anyhow. Maybe such a port will appear later
2752 * and we're pre-populating the flow table.
2756 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_OUTPUT
, port
);
2757 ctx
->nf_output_iface
= port
;
2760 static struct rule
*
2761 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2763 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2767 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2769 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2770 uint16_t old_in_port
;
2773 /* Look up a flow with 'in_port' as the input port. Then restore the
2774 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2775 * have surprising behavior). */
2776 old_in_port
= ctx
->flow
.in_port
;
2777 ctx
->flow
.in_port
= in_port
;
2778 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2779 ctx
->flow
.in_port
= old_in_port
;
2781 if (ctx
->resubmit_hook
) {
2782 ctx
->resubmit_hook(ctx
, rule
);
2787 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2791 static struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2793 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2794 MAX_RESUBMIT_RECURSION
);
2799 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2800 uint16_t *nf_output_iface
, struct ofpbuf
*odp_actions
)
2802 struct ofport
*ofport
;
2804 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2805 uint16_t odp_port
= ofport
->odp_port
;
2806 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2807 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, odp_port
);
2810 *nf_output_iface
= NF_OUT_FLOOD
;
2814 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2815 uint16_t port
, uint16_t max_len
)
2818 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2820 ctx
->nf_output_iface
= NF_OUT_DROP
;
2824 add_output_action(ctx
, ctx
->flow
.in_port
);
2827 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2830 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2831 ctx
->odp_actions
, &ctx
->tags
,
2832 &ctx
->nf_output_iface
,
2833 ctx
->ofproto
->aux
)) {
2834 COVERAGE_INC(ofproto_uninstallable
);
2835 ctx
->may_set_up_flow
= false;
2839 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2840 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2843 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2844 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2846 case OFPP_CONTROLLER
:
2847 nl_msg_put_u64(ctx
->odp_actions
, ODP_ACTION_ATTR_CONTROLLER
, max_len
);
2850 add_output_action(ctx
, ODPP_LOCAL
);
2853 odp_port
= ofp_port_to_odp_port(port
);
2854 if (odp_port
!= ctx
->flow
.in_port
) {
2855 add_output_action(ctx
, odp_port
);
2860 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2861 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2862 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2863 ctx
->nf_output_iface
= prev_nf_output_iface
;
2864 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2865 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2866 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2871 xlate_output_action(struct action_xlate_ctx
*ctx
,
2872 const struct ofp_action_output
*oao
)
2874 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2877 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2878 * optimization, because we're going to add another action that sets the
2879 * priority immediately after, or because there are no actions following the
2882 remove_pop_action(struct action_xlate_ctx
*ctx
)
2884 if (ctx
->odp_actions
->size
== ctx
->last_pop_priority
) {
2885 ctx
->odp_actions
->size
-= NLA_ALIGN(NLA_HDRLEN
);
2886 ctx
->last_pop_priority
= -1;
2891 add_pop_action(struct action_xlate_ctx
*ctx
)
2893 if (ctx
->odp_actions
->size
!= ctx
->last_pop_priority
) {
2894 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_POP_PRIORITY
);
2895 ctx
->last_pop_priority
= ctx
->odp_actions
->size
;
2900 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2901 const struct ofp_action_enqueue
*oae
)
2903 uint16_t ofp_port
, odp_port
;
2907 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2910 /* Fall back to ordinary output action. */
2911 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2915 /* Figure out ODP output port. */
2916 ofp_port
= ntohs(oae
->port
);
2917 if (ofp_port
!= OFPP_IN_PORT
) {
2918 odp_port
= ofp_port_to_odp_port(ofp_port
);
2920 odp_port
= ctx
->flow
.in_port
;
2923 /* Add ODP actions. */
2924 remove_pop_action(ctx
);
2925 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2926 add_output_action(ctx
, odp_port
);
2927 add_pop_action(ctx
);
2929 /* Update NetFlow output port. */
2930 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2931 ctx
->nf_output_iface
= odp_port
;
2932 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2933 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2938 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2939 const struct nx_action_set_queue
*nasq
)
2944 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2947 /* Couldn't translate queue to a priority, so ignore. A warning
2948 * has already been logged. */
2952 remove_pop_action(ctx
);
2953 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2957 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2959 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2960 if (!(tci
& htons(VLAN_CFI
))) {
2961 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_STRIP_VLAN
);
2963 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_TCI
,
2964 tci
& ~htons(VLAN_CFI
));
2968 struct xlate_reg_state
{
2974 save_reg_state(const struct action_xlate_ctx
*ctx
,
2975 struct xlate_reg_state
*state
)
2977 state
->vlan_tci
= ctx
->flow
.vlan_tci
;
2978 state
->tun_id
= ctx
->flow
.tun_id
;
2982 update_reg_state(struct action_xlate_ctx
*ctx
,
2983 const struct xlate_reg_state
*state
)
2985 if (ctx
->flow
.vlan_tci
!= state
->vlan_tci
) {
2986 xlate_set_dl_tci(ctx
);
2988 if (ctx
->flow
.tun_id
!= state
->tun_id
) {
2989 nl_msg_put_be64(ctx
->odp_actions
,
2990 ODP_ACTION_ATTR_SET_TUNNEL
, ctx
->flow
.tun_id
);
2995 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2996 const struct nx_action_header
*nah
)
2998 const struct nx_action_resubmit
*nar
;
2999 const struct nx_action_set_tunnel
*nast
;
3000 const struct nx_action_set_queue
*nasq
;
3001 const struct nx_action_multipath
*nam
;
3002 enum nx_action_subtype subtype
= ntohs(nah
->subtype
);
3003 struct xlate_reg_state state
;
3006 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
3008 case NXAST_RESUBMIT
:
3009 nar
= (const struct nx_action_resubmit
*) nah
;
3010 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
3013 case NXAST_SET_TUNNEL
:
3014 nast
= (const struct nx_action_set_tunnel
*) nah
;
3015 tun_id
= htonll(ntohl(nast
->tun_id
));
3016 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
3017 ctx
->flow
.tun_id
= tun_id
;
3020 case NXAST_DROP_SPOOFED_ARP
:
3021 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
3022 nl_msg_put_flag(ctx
->odp_actions
,
3023 ODP_ACTION_ATTR_DROP_SPOOFED_ARP
);
3027 case NXAST_SET_QUEUE
:
3028 nasq
= (const struct nx_action_set_queue
*) nah
;
3029 xlate_set_queue_action(ctx
, nasq
);
3032 case NXAST_POP_QUEUE
:
3033 add_pop_action(ctx
);
3036 case NXAST_REG_MOVE
:
3037 save_reg_state(ctx
, &state
);
3038 nxm_execute_reg_move((const struct nx_action_reg_move
*) nah
,
3040 update_reg_state(ctx
, &state
);
3043 case NXAST_REG_LOAD
:
3044 save_reg_state(ctx
, &state
);
3045 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
3047 update_reg_state(ctx
, &state
);
3051 /* Nothing to do. */
3054 case NXAST_SET_TUNNEL64
:
3055 tun_id
= ((const struct nx_action_set_tunnel64
*) nah
)->tun_id
;
3056 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
3057 ctx
->flow
.tun_id
= tun_id
;
3060 case NXAST_MULTIPATH
:
3061 nam
= (const struct nx_action_multipath
*) nah
;
3062 multipath_execute(nam
, &ctx
->flow
);
3065 /* If you add a new action here that modifies flow data, don't forget to
3066 * update the flow key in ctx->flow at the same time. */
3068 case NXAST_SNAT__OBSOLETE
:
3070 VLOG_DBG_RL(&rl
, "unknown Nicira action type %d", (int) subtype
);
3076 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
3077 struct action_xlate_ctx
*ctx
)
3079 struct actions_iterator iter
;
3080 const union ofp_action
*ia
;
3081 const struct ofport
*port
;
3083 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
3084 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
3085 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
3086 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
3087 /* Drop this flow. */
3091 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
3092 enum ofp_action_type type
= ntohs(ia
->type
);
3093 const struct ofp_action_dl_addr
*oada
;
3097 xlate_output_action(ctx
, &ia
->output
);
3100 case OFPAT_SET_VLAN_VID
:
3101 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
3102 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
3103 xlate_set_dl_tci(ctx
);
3106 case OFPAT_SET_VLAN_PCP
:
3107 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
3108 ctx
->flow
.vlan_tci
|= htons(
3109 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
3110 xlate_set_dl_tci(ctx
);
3113 case OFPAT_STRIP_VLAN
:
3114 ctx
->flow
.vlan_tci
= htons(0);
3115 xlate_set_dl_tci(ctx
);
3118 case OFPAT_SET_DL_SRC
:
3119 oada
= ((struct ofp_action_dl_addr
*) ia
);
3120 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_SRC
,
3121 oada
->dl_addr
, ETH_ADDR_LEN
);
3122 memcpy(ctx
->flow
.dl_src
, oada
->dl_addr
, ETH_ADDR_LEN
);
3125 case OFPAT_SET_DL_DST
:
3126 oada
= ((struct ofp_action_dl_addr
*) ia
);
3127 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_DST
,
3128 oada
->dl_addr
, ETH_ADDR_LEN
);
3129 memcpy(ctx
->flow
.dl_dst
, oada
->dl_addr
, ETH_ADDR_LEN
);
3132 case OFPAT_SET_NW_SRC
:
3133 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_SRC
,
3134 ia
->nw_addr
.nw_addr
);
3135 ctx
->flow
.nw_src
= ia
->nw_addr
.nw_addr
;
3138 case OFPAT_SET_NW_DST
:
3139 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_DST
,
3140 ia
->nw_addr
.nw_addr
);
3141 ctx
->flow
.nw_dst
= ia
->nw_addr
.nw_addr
;
3144 case OFPAT_SET_NW_TOS
:
3145 nl_msg_put_u8(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_TOS
,
3147 ctx
->flow
.nw_tos
= ia
->nw_tos
.nw_tos
;
3150 case OFPAT_SET_TP_SRC
:
3151 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_SRC
,
3152 ia
->tp_port
.tp_port
);
3153 ctx
->flow
.tp_src
= ia
->tp_port
.tp_port
;
3156 case OFPAT_SET_TP_DST
:
3157 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_DST
,
3158 ia
->tp_port
.tp_port
);
3159 ctx
->flow
.tp_dst
= ia
->tp_port
.tp_port
;
3163 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
3167 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
3171 VLOG_DBG_RL(&rl
, "unknown action type %d", (int) type
);
3178 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
3179 struct ofproto
*ofproto
, const struct flow
*flow
,
3180 const struct ofpbuf
*packet
)
3182 ctx
->ofproto
= ofproto
;
3184 ctx
->packet
= packet
;
3185 ctx
->resubmit_hook
= NULL
;
3186 ctx
->check_special
= true;
3190 ofproto_process_cfm(struct ofproto
*ofproto
, const struct flow
*flow
,
3191 const struct ofpbuf
*packet
)
3193 struct ofport
*ofport
;
3195 ofport
= get_port(ofproto
, flow
->in_port
);
3196 if (ofport
&& ofport
->cfm
) {
3197 cfm_process_heartbeat(ofport
->cfm
, packet
);
3201 static struct ofpbuf
*
3202 xlate_actions(struct action_xlate_ctx
*ctx
,
3203 const union ofp_action
*in
, size_t n_in
)
3205 COVERAGE_INC(ofproto_ofp2odp
);
3207 ctx
->odp_actions
= ofpbuf_new(512);
3209 ctx
->may_set_up_flow
= true;
3210 ctx
->nf_output_iface
= NF_OUT_DROP
;
3212 ctx
->last_pop_priority
= -1;
3214 if (ctx
->check_special
&& cfm_should_process_flow(&ctx
->flow
)) {
3216 ofproto_process_cfm(ctx
->ofproto
, &ctx
->flow
, ctx
->packet
);
3218 ctx
->may_set_up_flow
= false;
3219 } else if (ctx
->check_special
3220 && ctx
->ofproto
->ofhooks
->special_cb
3221 && !ctx
->ofproto
->ofhooks
->special_cb(&ctx
->flow
, ctx
->packet
,
3222 ctx
->ofproto
->aux
)) {
3223 ctx
->may_set_up_flow
= false;
3225 do_xlate_actions(in
, n_in
, ctx
);
3228 remove_pop_action(ctx
);
3230 /* Check with in-band control to see if we're allowed to set up this
3232 if (!in_band_rule_check(ctx
->ofproto
->in_band
, &ctx
->flow
,
3233 ctx
->odp_actions
->data
, ctx
->odp_actions
->size
)) {
3234 ctx
->may_set_up_flow
= false;
3237 return ctx
->odp_actions
;
3240 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3241 * error message code (composed with ofp_mkerr()) for the caller to propagate
3242 * upward. Otherwise, returns 0.
3244 * The log message mentions 'msg_type'. */
3246 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3248 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3249 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3250 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3253 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3260 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3262 struct ofproto
*p
= ofconn
->ofproto
;
3263 struct ofp_packet_out
*opo
;
3264 struct ofpbuf payload
, *buffer
;
3265 union ofp_action
*ofp_actions
;
3266 struct action_xlate_ctx ctx
;
3267 struct ofpbuf
*odp_actions
;
3268 struct ofpbuf request
;
3270 size_t n_ofp_actions
;
3274 COVERAGE_INC(ofproto_packet_out
);
3276 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3281 /* Get ofp_packet_out. */
3282 ofpbuf_use_const(&request
, oh
, ntohs(oh
->length
));
3283 opo
= ofpbuf_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3286 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3287 &ofp_actions
, &n_ofp_actions
);
3293 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3294 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3296 if (error
|| !buffer
) {
3305 /* Extract flow, check actions. */
3306 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3308 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3314 action_xlate_ctx_init(&ctx
, p
, &flow
, &payload
);
3315 odp_actions
= xlate_actions(&ctx
, ofp_actions
, n_ofp_actions
);
3316 dpif_execute(p
->dpif
, odp_actions
->data
, odp_actions
->size
, &payload
);
3317 ofpbuf_delete(odp_actions
);
3320 ofpbuf_delete(buffer
);
3325 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3326 uint32_t config
, uint32_t mask
)
3328 mask
&= config
^ port
->opp
.config
;
3329 if (mask
& OFPPC_PORT_DOWN
) {
3330 if (config
& OFPPC_PORT_DOWN
) {
3331 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3333 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3336 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3337 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3338 if (mask
& REVALIDATE_BITS
) {
3339 COVERAGE_INC(ofproto_costly_flags
);
3340 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3341 p
->need_revalidate
= true;
3343 #undef REVALIDATE_BITS
3344 if (mask
& OFPPC_NO_PACKET_IN
) {
3345 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3350 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3352 struct ofproto
*p
= ofconn
->ofproto
;
3353 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
3354 struct ofport
*port
;
3357 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3362 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3364 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3365 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3366 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3368 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3369 if (opm
->advertise
) {
3370 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3376 static struct ofpbuf
*
3377 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3379 struct ofp_stats_reply
*osr
;
3382 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3383 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3385 osr
->flags
= htons(0);
3389 static struct ofpbuf
*
3390 start_ofp_stats_reply(const struct ofp_header
*request
, size_t body_len
)
3392 const struct ofp_stats_request
*osr
3393 = (const struct ofp_stats_request
*) request
;
3394 return make_ofp_stats_reply(osr
->header
.xid
, osr
->type
, body_len
);
3398 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3399 struct ofpbuf
**msgp
)
3401 struct ofpbuf
*msg
= *msgp
;
3402 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3403 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3404 struct ofp_stats_reply
*reply
= msg
->data
;
3405 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3406 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3407 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3409 return ofpbuf_put_uninit(*msgp
, nbytes
);
3412 static struct ofpbuf
*
3413 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3415 struct nicira_stats_msg
*nsm
;
3418 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3419 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3420 nsm
->type
= htons(OFPST_VENDOR
);
3421 nsm
->flags
= htons(0);
3422 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3423 nsm
->subtype
= subtype
;
3427 static struct ofpbuf
*
3428 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3430 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3434 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3435 struct ofpbuf
**msgp
)
3437 struct ofpbuf
*msg
= *msgp
;
3438 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3439 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3440 struct nicira_stats_msg
*reply
= msg
->data
;
3441 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3442 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3443 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3445 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3449 handle_desc_stats_request(struct ofconn
*ofconn
,
3450 const struct ofp_header
*request
)
3452 struct ofproto
*p
= ofconn
->ofproto
;
3453 struct ofp_desc_stats
*ods
;
3456 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3457 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3458 memset(ods
, 0, sizeof *ods
);
3459 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3460 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3461 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3462 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3463 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3464 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3470 handle_table_stats_request(struct ofconn
*ofconn
,
3471 const struct ofp_header
*request
)
3473 struct ofproto
*p
= ofconn
->ofproto
;
3474 struct ofp_table_stats
*ots
;
3477 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3479 /* Classifier table. */
3480 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3481 memset(ots
, 0, sizeof *ots
);
3482 strcpy(ots
->name
, "classifier");
3483 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3484 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3485 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3486 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3487 put_32aligned_be64(&ots
->lookup_count
, htonll(0)); /* XXX */
3488 put_32aligned_be64(&ots
->matched_count
, htonll(0)); /* XXX */
3490 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3495 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3496 struct ofpbuf
**msgp
)
3498 struct netdev_stats stats
;
3499 struct ofp_port_stats
*ops
;
3501 /* Intentionally ignore return value, since errors will set
3502 * 'stats' to all-1s, which is correct for OpenFlow, and
3503 * netdev_get_stats() will log errors. */
3504 netdev_get_stats(port
->netdev
, &stats
);
3506 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3507 ops
->port_no
= htons(port
->opp
.port_no
);
3508 memset(ops
->pad
, 0, sizeof ops
->pad
);
3509 put_32aligned_be64(&ops
->rx_packets
, htonll(stats
.rx_packets
));
3510 put_32aligned_be64(&ops
->tx_packets
, htonll(stats
.tx_packets
));
3511 put_32aligned_be64(&ops
->rx_bytes
, htonll(stats
.rx_bytes
));
3512 put_32aligned_be64(&ops
->tx_bytes
, htonll(stats
.tx_bytes
));
3513 put_32aligned_be64(&ops
->rx_dropped
, htonll(stats
.rx_dropped
));
3514 put_32aligned_be64(&ops
->tx_dropped
, htonll(stats
.tx_dropped
));
3515 put_32aligned_be64(&ops
->rx_errors
, htonll(stats
.rx_errors
));
3516 put_32aligned_be64(&ops
->tx_errors
, htonll(stats
.tx_errors
));
3517 put_32aligned_be64(&ops
->rx_frame_err
, htonll(stats
.rx_frame_errors
));
3518 put_32aligned_be64(&ops
->rx_over_err
, htonll(stats
.rx_over_errors
));
3519 put_32aligned_be64(&ops
->rx_crc_err
, htonll(stats
.rx_crc_errors
));
3520 put_32aligned_be64(&ops
->collisions
, htonll(stats
.collisions
));
3524 handle_port_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3526 struct ofproto
*p
= ofconn
->ofproto
;
3527 const struct ofp_port_stats_request
*psr
= ofputil_stats_body(oh
);
3528 struct ofp_port_stats
*ops
;
3530 struct ofport
*port
;
3532 msg
= start_ofp_stats_reply(oh
, sizeof *ops
* 16);
3533 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3534 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3536 append_port_stat(port
, ofconn
, &msg
);
3539 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3540 append_port_stat(port
, ofconn
, &msg
);
3544 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3549 calc_flow_duration__(long long int start
, uint32_t *sec
, uint32_t *nsec
)
3551 long long int msecs
= time_msec() - start
;
3552 *sec
= msecs
/ 1000;
3553 *nsec
= (msecs
% 1000) * (1000 * 1000);
3557 calc_flow_duration(long long int start
, ovs_be32
*sec_be
, ovs_be32
*nsec_be
)
3561 calc_flow_duration__(start
, &sec
, &nsec
);
3562 *sec_be
= htonl(sec
);
3563 *nsec_be
= htonl(nsec
);
3567 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3568 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3570 struct ofp_flow_stats
*ofs
;
3571 uint64_t packet_count
, byte_count
;
3573 size_t act_len
, len
;
3575 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3579 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3580 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3582 rule_get_stats(rule
, &packet_count
, &byte_count
);
3584 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3585 ofs
->length
= htons(len
);
3588 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
,
3589 rule
->flow_cookie
, &cookie
);
3590 put_32aligned_be64(&ofs
->cookie
, cookie
);
3591 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3592 ofs
->priority
= htons(rule
->cr
.priority
);
3593 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3594 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3595 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3596 put_32aligned_be64(&ofs
->packet_count
, htonll(packet_count
));
3597 put_32aligned_be64(&ofs
->byte_count
, htonll(byte_count
));
3598 if (rule
->n_actions
> 0) {
3599 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3604 is_valid_table(uint8_t table_id
)
3606 if (table_id
== 0 || table_id
== 0xff) {
3609 /* It would probably be better to reply with an error but there doesn't
3610 * seem to be any appropriate value, so that might just be
3612 VLOG_WARN_RL(&rl
, "controller asked for invalid table %"PRIu8
,
3619 handle_flow_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3621 const struct ofp_flow_stats_request
*fsr
= ofputil_stats_body(oh
);
3622 struct ofpbuf
*reply
;
3624 COVERAGE_INC(ofproto_flows_req
);
3625 reply
= start_ofp_stats_reply(oh
, 1024);
3626 if (is_valid_table(fsr
->table_id
)) {
3627 struct cls_cursor cursor
;
3628 struct cls_rule target
;
3631 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3633 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3634 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3635 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3638 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3644 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3645 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3647 struct nx_flow_stats
*nfs
;
3648 uint64_t packet_count
, byte_count
;
3649 size_t act_len
, start_len
;
3650 struct ofpbuf
*reply
;
3652 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3656 rule_get_stats(rule
, &packet_count
, &byte_count
);
3658 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3660 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3661 start_len
= (*replyp
)->size
;
3664 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3667 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3668 nfs
->cookie
= rule
->flow_cookie
;
3669 nfs
->priority
= htons(rule
->cr
.priority
);
3670 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3671 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3672 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3673 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3674 nfs
->packet_count
= htonll(packet_count
);
3675 nfs
->byte_count
= htonll(byte_count
);
3676 if (rule
->n_actions
> 0) {
3677 ofpbuf_put(reply
, rule
->actions
, act_len
);
3679 nfs
->length
= htons(reply
->size
- start_len
);
3683 handle_nxst_flow(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3685 struct nx_flow_stats_request
*nfsr
;
3686 struct cls_rule target
;
3687 struct ofpbuf
*reply
;
3691 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3693 /* Dissect the message. */
3694 nfsr
= ofpbuf_pull(&b
, sizeof *nfsr
);
3695 error
= nx_pull_match(&b
, ntohs(nfsr
->match_len
), 0, &target
);
3700 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3703 COVERAGE_INC(ofproto_flows_req
);
3704 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3705 if (is_valid_table(nfsr
->table_id
)) {
3706 struct cls_cursor cursor
;
3709 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3710 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3711 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3714 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3720 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
3722 uint64_t packet_count
, byte_count
;
3723 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3725 rule_get_stats(rule
, &packet_count
, &byte_count
);
3727 ds_put_format(results
, "duration=%llds, ",
3728 (time_msec() - rule
->created
) / 1000);
3729 ds_put_format(results
, "idle=%.3fs, ", (time_msec() - rule
->used
) / 1000.0);
3730 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3731 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3732 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3733 cls_rule_format(&rule
->cr
, results
);
3734 ds_put_char(results
, ',');
3736 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3738 ds_put_cstr(results
, "drop");
3740 ds_put_cstr(results
, "\n");
3743 /* Adds a pretty-printed description of all flows to 'results', including
3744 * hidden flows (e.g., set up by in-band control). */
3746 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3748 struct cls_cursor cursor
;
3751 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3752 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3753 flow_stats_ds(rule
, results
);
3758 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3759 ovs_be16 out_port
, uint8_t table_id
,
3760 struct ofp_aggregate_stats_reply
*oasr
)
3762 uint64_t total_packets
= 0;
3763 uint64_t total_bytes
= 0;
3766 COVERAGE_INC(ofproto_agg_request
);
3768 if (is_valid_table(table_id
)) {
3769 struct cls_cursor cursor
;
3772 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3773 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3774 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3775 uint64_t packet_count
;
3776 uint64_t byte_count
;
3778 rule_get_stats(rule
, &packet_count
, &byte_count
);
3780 total_packets
+= packet_count
;
3781 total_bytes
+= byte_count
;
3787 oasr
->flow_count
= htonl(n_flows
);
3788 put_32aligned_be64(&oasr
->packet_count
, htonll(total_packets
));
3789 put_32aligned_be64(&oasr
->byte_count
, htonll(total_bytes
));
3790 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3794 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3795 const struct ofp_header
*oh
)
3797 const struct ofp_aggregate_stats_request
*request
= ofputil_stats_body(oh
);
3798 struct ofp_aggregate_stats_reply
*reply
;
3799 struct cls_rule target
;
3802 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3805 msg
= start_ofp_stats_reply(oh
, sizeof *reply
);
3806 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3807 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3808 request
->table_id
, reply
);
3809 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3814 handle_nxst_aggregate(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3816 struct nx_aggregate_stats_request
*request
;
3817 struct ofp_aggregate_stats_reply
*reply
;
3818 struct cls_rule target
;
3823 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3825 /* Dissect the message. */
3826 request
= ofpbuf_pull(&b
, sizeof *request
);
3827 error
= nx_pull_match(&b
, ntohs(request
->match_len
), 0, &target
);
3832 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3836 COVERAGE_INC(ofproto_flows_req
);
3837 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3838 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3839 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3840 request
->table_id
, reply
);
3841 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3846 struct queue_stats_cbdata
{
3847 struct ofconn
*ofconn
;
3848 struct ofport
*ofport
;
3853 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3854 const struct netdev_queue_stats
*stats
)
3856 struct ofp_queue_stats
*reply
;
3858 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3859 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3860 memset(reply
->pad
, 0, sizeof reply
->pad
);
3861 reply
->queue_id
= htonl(queue_id
);
3862 put_32aligned_be64(&reply
->tx_bytes
, htonll(stats
->tx_bytes
));
3863 put_32aligned_be64(&reply
->tx_packets
, htonll(stats
->tx_packets
));
3864 put_32aligned_be64(&reply
->tx_errors
, htonll(stats
->tx_errors
));
3868 handle_queue_stats_dump_cb(uint32_t queue_id
,
3869 struct netdev_queue_stats
*stats
,
3872 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3874 put_queue_stats(cbdata
, queue_id
, stats
);
3878 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3879 struct queue_stats_cbdata
*cbdata
)
3881 cbdata
->ofport
= port
;
3882 if (queue_id
== OFPQ_ALL
) {
3883 netdev_dump_queue_stats(port
->netdev
,
3884 handle_queue_stats_dump_cb
, cbdata
);
3886 struct netdev_queue_stats stats
;
3888 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3889 put_queue_stats(cbdata
, queue_id
, &stats
);
3895 handle_queue_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3897 struct ofproto
*ofproto
= ofconn
->ofproto
;
3898 const struct ofp_queue_stats_request
*qsr
;
3899 struct queue_stats_cbdata cbdata
;
3900 struct ofport
*port
;
3901 unsigned int port_no
;
3904 qsr
= ofputil_stats_body(oh
);
3906 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3909 COVERAGE_INC(ofproto_queue_req
);
3911 cbdata
.ofconn
= ofconn
;
3912 cbdata
.msg
= start_ofp_stats_reply(oh
, 128);
3914 port_no
= ntohs(qsr
->port_no
);
3915 queue_id
= ntohl(qsr
->queue_id
);
3916 if (port_no
== OFPP_ALL
) {
3917 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3918 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3920 } else if (port_no
< ofproto
->max_ports
) {
3921 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3923 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3926 ofpbuf_delete(cbdata
.msg
);
3927 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3929 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3934 /* Updates 'facet''s used time. Caller is responsible for calling
3935 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3937 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3940 if (used
> facet
->used
) {
3942 if (used
> facet
->rule
->used
) {
3943 facet
->rule
->used
= used
;
3945 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3949 /* Folds the statistics from 'stats' into the counters in 'facet'.
3951 * Because of the meaning of a facet's counters, it only makes sense to do this
3952 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3953 * packet that was sent by hand or if it represents statistics that have been
3954 * cleared out of the datapath. */
3956 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3957 const struct dpif_flow_stats
*stats
)
3959 if (stats
->n_packets
|| stats
->used
> facet
->used
) {
3960 facet_update_time(ofproto
, facet
, stats
->used
);
3961 facet
->packet_count
+= stats
->n_packets
;
3962 facet
->byte_count
+= stats
->n_bytes
;
3963 facet_push_stats(ofproto
, facet
);
3964 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3969 facet_push_stats(struct ofproto
*ofproto
, struct facet
*facet
)
3971 uint64_t rs_packets
, rs_bytes
;
3973 assert(facet
->packet_count
>= facet
->rs_packet_count
);
3974 assert(facet
->byte_count
>= facet
->rs_byte_count
);
3975 assert(facet
->used
>= facet
->rs_used
);
3977 rs_packets
= facet
->packet_count
- facet
->rs_packet_count
;
3978 rs_bytes
= facet
->byte_count
- facet
->rs_byte_count
;
3980 if (rs_packets
|| rs_bytes
|| facet
->used
> facet
->rs_used
) {
3981 facet
->rs_packet_count
= facet
->packet_count
;
3982 facet
->rs_byte_count
= facet
->byte_count
;
3983 facet
->rs_used
= facet
->used
;
3985 flow_push_stats(ofproto
, facet
->rule
, &facet
->flow
,
3986 rs_packets
, rs_bytes
, facet
->used
);
3990 struct ofproto_push
{
3991 struct action_xlate_ctx ctx
;
3998 push_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
4000 struct ofproto_push
*push
= CONTAINER_OF(ctx
, struct ofproto_push
, ctx
);
4003 rule
->packet_count
+= push
->packets
;
4004 rule
->byte_count
+= push
->bytes
;
4005 rule
->used
= MAX(push
->used
, rule
->used
);
4009 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4010 * 'rule''s actions. */
4012 flow_push_stats(struct ofproto
*ofproto
, const struct rule
*rule
,
4013 struct flow
*flow
, uint64_t packets
, uint64_t bytes
,
4016 struct ofproto_push push
;
4018 push
.packets
= packets
;
4022 action_xlate_ctx_init(&push
.ctx
, ofproto
, flow
, NULL
);
4023 push
.ctx
.resubmit_hook
= push_resubmit
;
4024 ofpbuf_delete(xlate_actions(&push
.ctx
, rule
->actions
, rule
->n_actions
));
4027 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4028 * in which no matching flow already exists in the flow table.
4030 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
4031 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
4032 * OpenFlow error code as encoded by ofp_mkerr() on failure.
4034 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4037 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4039 struct ofproto
*p
= ofconn
->ofproto
;
4040 struct ofpbuf
*packet
;
4045 if (fm
->flags
& OFPFF_CHECK_OVERLAP
4046 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
4047 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
4051 if (fm
->buffer_id
!= UINT32_MAX
) {
4052 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
4056 in_port
= UINT16_MAX
;
4059 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
4060 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
4061 fm
->flags
& OFPFF_SEND_FLOW_REM
);
4062 rule_insert(p
, rule
);
4064 rule_execute(p
, rule
, in_port
, packet
);
4069 static struct rule
*
4070 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
4072 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
4076 send_buffered_packet(struct ofconn
*ofconn
,
4077 struct rule
*rule
, uint32_t buffer_id
)
4079 struct ofpbuf
*packet
;
4083 if (buffer_id
== UINT32_MAX
) {
4087 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
4092 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
4097 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4099 struct modify_flows_cbdata
{
4100 struct ofproto
*ofproto
;
4101 const struct flow_mod
*fm
;
4105 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
4108 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4109 * encoded by ofp_mkerr() on failure.
4111 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4114 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4116 struct ofproto
*p
= ofconn
->ofproto
;
4117 struct rule
*match
= NULL
;
4118 struct cls_cursor cursor
;
4121 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4122 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4123 if (!rule_is_hidden(rule
)) {
4125 modify_flow(p
, fm
, rule
);
4130 /* This credits the packet to whichever flow happened to match last.
4131 * That's weird. Maybe we should do a lookup for the flow that
4132 * actually matches the packet? Who knows. */
4133 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
4136 return add_flow(ofconn
, fm
);
4140 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4141 * code as encoded by ofp_mkerr() on failure.
4143 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4146 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4148 struct ofproto
*p
= ofconn
->ofproto
;
4149 struct rule
*rule
= find_flow_strict(p
, fm
);
4150 if (rule
&& !rule_is_hidden(rule
)) {
4151 modify_flow(p
, fm
, rule
);
4152 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
4154 return add_flow(ofconn
, fm
);
4158 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4159 * been identified as a flow in 'p''s flow table to be modified, by changing
4160 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4161 * ofp_action[] structures). */
4163 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
4165 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
4167 rule
->flow_cookie
= fm
->cookie
;
4169 /* If the actions are the same, do nothing. */
4170 if (fm
->n_actions
== rule
->n_actions
4172 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
4176 /* Replace actions. */
4177 free(rule
->actions
);
4178 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
4179 rule
->n_actions
= fm
->n_actions
;
4181 p
->need_revalidate
= true;
4186 /* OFPFC_DELETE implementation. */
4188 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
4190 /* Implements OFPFC_DELETE. */
4192 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
4194 struct rule
*rule
, *next_rule
;
4195 struct cls_cursor cursor
;
4197 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4198 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4199 delete_flow(p
, rule
, htons(fm
->out_port
));
4203 /* Implements OFPFC_DELETE_STRICT. */
4205 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4207 struct rule
*rule
= find_flow_strict(p
, fm
);
4209 delete_flow(p
, rule
, htons(fm
->out_port
));
4213 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4214 * been identified as a flow to delete from 'p''s flow table, by deleting the
4215 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4218 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4219 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4220 * specified 'out_port'. */
4222 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4224 if (rule_is_hidden(rule
)) {
4228 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4232 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4233 rule_remove(p
, rule
);
4237 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4239 struct ofproto
*p
= ofconn
->ofproto
;
4243 error
= reject_slave_controller(ofconn
, "flow_mod");
4248 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn
->flow_format
);
4253 /* We do not support the emergency flow cache. It will hopefully get
4254 * dropped from OpenFlow in the near future. */
4255 if (fm
.flags
& OFPFF_EMERG
) {
4256 /* There isn't a good fit for an error code, so just state that the
4257 * flow table is full. */
4258 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4261 error
= validate_actions(fm
.actions
, fm
.n_actions
,
4262 &fm
.cr
.flow
, p
->max_ports
);
4267 switch (fm
.command
) {
4269 return add_flow(ofconn
, &fm
);
4272 return modify_flows_loose(ofconn
, &fm
);
4274 case OFPFC_MODIFY_STRICT
:
4275 return modify_flow_strict(ofconn
, &fm
);
4278 delete_flows_loose(p
, &fm
);
4281 case OFPFC_DELETE_STRICT
:
4282 delete_flow_strict(p
, &fm
);
4286 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4291 handle_tun_id_from_cookie(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4293 const struct nxt_tun_id_cookie
*msg
4294 = (const struct nxt_tun_id_cookie
*) oh
;
4296 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4301 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4303 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
4304 struct nx_role_request
*reply
;
4308 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4309 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4311 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4314 role
= ntohl(nrr
->role
);
4315 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4316 && role
!= NX_ROLE_SLAVE
) {
4317 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4319 /* There's no good error code for this. */
4320 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4323 if (role
== NX_ROLE_MASTER
) {
4324 struct ofconn
*other
;
4326 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4327 if (other
->role
== NX_ROLE_MASTER
) {
4328 other
->role
= NX_ROLE_SLAVE
;
4332 ofconn
->role
= role
;
4334 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
4335 reply
->role
= htonl(role
);
4336 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4342 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4344 const struct nxt_set_flow_format
*msg
4345 = (const struct nxt_set_flow_format
*) oh
;
4348 format
= ntohl(msg
->format
);
4349 if (format
== NXFF_OPENFLOW10
4350 || format
== NXFF_TUN_ID_FROM_COOKIE
4351 || format
== NXFF_NXM
) {
4352 ofconn
->flow_format
= format
;
4355 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4360 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4362 struct ofp_header
*ob
;
4365 /* Currently, everything executes synchronously, so we can just
4366 * immediately send the barrier reply. */
4367 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4368 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4373 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4375 const struct ofp_header
*oh
= msg
->data
;
4376 const struct ofputil_msg_type
*type
;
4379 error
= ofputil_decode_msg_type(oh
, &type
);
4384 switch (ofputil_msg_type_code(type
)) {
4385 /* OpenFlow requests. */
4386 case OFPUTIL_OFPT_ECHO_REQUEST
:
4387 return handle_echo_request(ofconn
, oh
);
4389 case OFPUTIL_OFPT_FEATURES_REQUEST
:
4390 return handle_features_request(ofconn
, oh
);
4392 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
4393 return handle_get_config_request(ofconn
, oh
);
4395 case OFPUTIL_OFPT_SET_CONFIG
:
4396 return handle_set_config(ofconn
, msg
->data
);
4398 case OFPUTIL_OFPT_PACKET_OUT
:
4399 return handle_packet_out(ofconn
, oh
);
4401 case OFPUTIL_OFPT_PORT_MOD
:
4402 return handle_port_mod(ofconn
, oh
);
4404 case OFPUTIL_OFPT_FLOW_MOD
:
4405 return handle_flow_mod(ofconn
, oh
);
4407 case OFPUTIL_OFPT_BARRIER_REQUEST
:
4408 return handle_barrier_request(ofconn
, oh
);
4410 /* OpenFlow replies. */
4411 case OFPUTIL_OFPT_ECHO_REPLY
:
4414 /* Nicira extension requests. */
4415 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE
:
4416 return handle_tun_id_from_cookie(ofconn
, oh
);
4418 case OFPUTIL_NXT_ROLE_REQUEST
:
4419 return handle_role_request(ofconn
, oh
);
4421 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
4422 return handle_nxt_set_flow_format(ofconn
, oh
);
4424 case OFPUTIL_NXT_FLOW_MOD
:
4425 return handle_flow_mod(ofconn
, oh
);
4427 /* OpenFlow statistics requests. */
4428 case OFPUTIL_OFPST_DESC_REQUEST
:
4429 return handle_desc_stats_request(ofconn
, oh
);
4431 case OFPUTIL_OFPST_FLOW_REQUEST
:
4432 return handle_flow_stats_request(ofconn
, oh
);
4434 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
4435 return handle_aggregate_stats_request(ofconn
, oh
);
4437 case OFPUTIL_OFPST_TABLE_REQUEST
:
4438 return handle_table_stats_request(ofconn
, oh
);
4440 case OFPUTIL_OFPST_PORT_REQUEST
:
4441 return handle_port_stats_request(ofconn
, oh
);
4443 case OFPUTIL_OFPST_QUEUE_REQUEST
:
4444 return handle_queue_stats_request(ofconn
, oh
);
4446 /* Nicira extension statistics requests. */
4447 case OFPUTIL_NXST_FLOW_REQUEST
:
4448 return handle_nxst_flow(ofconn
, oh
);
4450 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
4451 return handle_nxst_aggregate(ofconn
, oh
);
4453 case OFPUTIL_INVALID
:
4454 case OFPUTIL_OFPT_HELLO
:
4455 case OFPUTIL_OFPT_ERROR
:
4456 case OFPUTIL_OFPT_FEATURES_REPLY
:
4457 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
4458 case OFPUTIL_OFPT_PACKET_IN
:
4459 case OFPUTIL_OFPT_FLOW_REMOVED
:
4460 case OFPUTIL_OFPT_PORT_STATUS
:
4461 case OFPUTIL_OFPT_BARRIER_REPLY
:
4462 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
4463 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
4464 case OFPUTIL_OFPST_DESC_REPLY
:
4465 case OFPUTIL_OFPST_FLOW_REPLY
:
4466 case OFPUTIL_OFPST_QUEUE_REPLY
:
4467 case OFPUTIL_OFPST_PORT_REPLY
:
4468 case OFPUTIL_OFPST_TABLE_REPLY
:
4469 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
4470 case OFPUTIL_NXT_ROLE_REPLY
:
4471 case OFPUTIL_NXT_FLOW_REMOVED
:
4472 case OFPUTIL_NXST_FLOW_REPLY
:
4473 case OFPUTIL_NXST_AGGREGATE_REPLY
:
4475 if (VLOG_IS_WARN_ENABLED()) {
4476 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4477 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4480 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
4481 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
4483 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4489 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4491 int error
= handle_openflow__(ofconn
, ofp_msg
);
4493 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4495 COVERAGE_INC(ofproto_recv_openflow
);
4499 handle_miss_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4501 struct facet
*facet
;
4504 /* Obtain in_port and tun_id, at least. */
4505 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4507 /* Set header pointers in 'flow'. */
4508 flow_extract(upcall
->packet
, flow
.tun_id
, flow
.in_port
, &flow
);
4510 if (cfm_should_process_flow(&flow
)) {
4511 ofproto_process_cfm(p
, &flow
, upcall
->packet
);
4512 ofpbuf_delete(upcall
->packet
);
4514 } else if (p
->ofhooks
->special_cb
4515 && !p
->ofhooks
->special_cb(&flow
, upcall
->packet
, p
->aux
)) {
4516 ofpbuf_delete(upcall
->packet
);
4520 /* Check with in-band control to see if this packet should be sent
4521 * to the local port regardless of the flow table. */
4522 if (in_band_msg_in_hook(p
->in_band
, &flow
, upcall
->packet
)) {
4523 ofproto_send_packet(p
, ODPP_LOCAL
, 0, upcall
->packet
);
4526 facet
= facet_lookup_valid(p
, &flow
);
4528 struct rule
*rule
= rule_lookup(p
, &flow
);
4530 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4531 struct ofport
*port
= get_port(p
, flow
.in_port
);
4533 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4534 COVERAGE_INC(ofproto_no_packet_in
);
4535 /* XXX install 'drop' flow entry */
4536 ofpbuf_delete(upcall
->packet
);
4540 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4544 COVERAGE_INC(ofproto_packet_in
);
4545 send_packet_in(p
, upcall
, &flow
, false);
4549 facet
= facet_create(p
, rule
, &flow
, upcall
->packet
);
4550 } else if (!facet
->may_install
) {
4551 /* The facet is not installable, that is, we need to process every
4552 * packet, so process the current packet's actions into 'facet'. */
4553 facet_make_actions(p
, facet
, upcall
->packet
);
4556 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4558 * Extra-special case for fail-open mode.
4560 * We are in fail-open mode and the packet matched the fail-open rule,
4561 * but we are connected to a controller too. We should send the packet
4562 * up to the controller in the hope that it will try to set up a flow
4563 * and thereby allow us to exit fail-open.
4565 * See the top-level comment in fail-open.c for more information.
4567 send_packet_in(p
, upcall
, &flow
, true);
4570 facet_execute(p
, facet
, upcall
->packet
);
4571 facet_install(p
, facet
, false);
4575 handle_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4579 switch (upcall
->type
) {
4580 case DPIF_UC_ACTION
:
4581 COVERAGE_INC(ofproto_ctlr_action
);
4582 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4583 send_packet_in(p
, upcall
, &flow
, false);
4586 case DPIF_UC_SAMPLE
:
4588 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4589 ofproto_sflow_received(p
->sflow
, upcall
, &flow
);
4591 ofpbuf_delete(upcall
->packet
);
4595 handle_miss_upcall(p
, upcall
);
4598 case DPIF_N_UC_TYPES
:
4600 VLOG_WARN_RL(&rl
, "upcall has unexpected type %"PRIu32
, upcall
->type
);
4605 /* Flow expiration. */
4607 static int ofproto_dp_max_idle(const struct ofproto
*);
4608 static void ofproto_update_stats(struct ofproto
*);
4609 static void rule_expire(struct ofproto
*, struct rule
*);
4610 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4612 /* This function is called periodically by ofproto_run(). Its job is to
4613 * collect updates for the flows that have been installed into the datapath,
4614 * most importantly when they last were used, and then use that information to
4615 * expire flows that have not been used recently.
4617 * Returns the number of milliseconds after which it should be called again. */
4619 ofproto_expire(struct ofproto
*ofproto
)
4621 struct rule
*rule
, *next_rule
;
4622 struct cls_cursor cursor
;
4625 /* Update stats for each flow in the datapath. */
4626 ofproto_update_stats(ofproto
);
4628 /* Expire facets that have been idle too long. */
4629 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4630 ofproto_expire_facets(ofproto
, dp_max_idle
);
4632 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4633 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4634 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4635 rule_expire(ofproto
, rule
);
4638 /* Let the hook know that we're at a stable point: all outstanding data
4639 * in existing flows has been accounted to the account_cb. Thus, the
4640 * hook can now reasonably do operations that depend on having accurate
4641 * flow volume accounting (currently, that's just bond rebalancing). */
4642 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4643 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4646 return MIN(dp_max_idle
, 1000);
4649 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4651 * This function also pushes statistics updates to rules which each facet
4652 * resubmits into. Generally these statistics will be accurate. However, if a
4653 * facet changes the rule it resubmits into at some time in between
4654 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4655 * old rule will be incorrectly attributed to the new rule. This could be
4656 * avoided by calling ofproto_update_stats() whenever rules are created or
4657 * deleted. However, the performance impact of making so many calls to the
4658 * datapath do not justify the benefit of having perfectly accurate statistics.
4661 ofproto_update_stats(struct ofproto
*p
)
4663 const struct dpif_flow_stats
*stats
;
4664 struct dpif_flow_dump dump
;
4665 const struct nlattr
*key
;
4668 dpif_flow_dump_start(&dump
, p
->dpif
);
4669 while (dpif_flow_dump_next(&dump
, &key
, &key_len
, NULL
, NULL
, &stats
)) {
4670 struct facet
*facet
;
4673 if (odp_flow_key_to_flow(key
, key_len
, &flow
)) {
4677 odp_flow_key_format(key
, key_len
, &s
);
4678 VLOG_WARN_RL(&rl
, "failed to convert ODP flow key to flow: %s",
4684 facet
= facet_find(p
, &flow
);
4686 if (facet
&& facet
->installed
) {
4688 if (stats
->n_packets
>= facet
->dp_packet_count
) {
4689 facet
->packet_count
+= stats
->n_packets
- facet
->dp_packet_count
;
4691 VLOG_WARN_RL(&rl
, "unexpected packet count from the datapath");
4694 if (stats
->n_bytes
>= facet
->dp_byte_count
) {
4695 facet
->byte_count
+= stats
->n_bytes
- facet
->dp_byte_count
;
4697 VLOG_WARN_RL(&rl
, "unexpected byte count from datapath");
4700 facet
->dp_packet_count
= stats
->n_packets
;
4701 facet
->dp_byte_count
= stats
->n_bytes
;
4703 facet_update_time(p
, facet
, stats
->used
);
4704 facet_account(p
, facet
, stats
->n_bytes
);
4705 facet_push_stats(p
, facet
);
4707 /* There's a flow in the datapath that we know nothing about.
4709 COVERAGE_INC(ofproto_unexpected_rule
);
4710 dpif_flow_del(p
->dpif
, key
, key_len
, NULL
);
4713 dpif_flow_dump_done(&dump
);
4716 /* Calculates and returns the number of milliseconds of idle time after which
4717 * facets should expire from the datapath and we should fold their statistics
4718 * into their parent rules in userspace. */
4720 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4723 * Idle time histogram.
4725 * Most of the time a switch has a relatively small number of facets. When
4726 * this is the case we might as well keep statistics for all of them in
4727 * userspace and to cache them in the kernel datapath for performance as
4730 * As the number of facets increases, the memory required to maintain
4731 * statistics about them in userspace and in the kernel becomes
4732 * significant. However, with a large number of facets it is likely that
4733 * only a few of them are "heavy hitters" that consume a large amount of
4734 * bandwidth. At this point, only heavy hitters are worth caching in the
4735 * kernel and maintaining in userspaces; other facets we can discard.
4737 * The technique used to compute the idle time is to build a histogram with
4738 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4739 * that is installed in the kernel gets dropped in the appropriate bucket.
4740 * After the histogram has been built, we compute the cutoff so that only
4741 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4742 * cached. At least the most-recently-used bucket of facets is kept, so
4743 * actually an arbitrary number of facets can be kept in any given
4744 * expiration run (though the next run will delete most of those unless
4745 * they receive additional data).
4747 * This requires a second pass through the facets, in addition to the pass
4748 * made by ofproto_update_stats(), because the former function never looks
4749 * at uninstallable facets.
4751 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4752 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4753 int buckets
[N_BUCKETS
] = { 0 };
4754 struct facet
*facet
;
4759 total
= hmap_count(&ofproto
->facets
);
4760 if (total
<= 1000) {
4761 return N_BUCKETS
* BUCKET_WIDTH
;
4764 /* Build histogram. */
4766 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4767 long long int idle
= now
- facet
->used
;
4768 int bucket
= (idle
<= 0 ? 0
4769 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4770 : (unsigned int) idle
/ BUCKET_WIDTH
);
4774 /* Find the first bucket whose flows should be expired. */
4775 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4776 if (buckets
[bucket
]) {
4779 subtotal
+= buckets
[bucket
++];
4780 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4785 if (VLOG_IS_DBG_ENABLED()) {
4789 ds_put_cstr(&s
, "keep");
4790 for (i
= 0; i
< N_BUCKETS
; i
++) {
4792 ds_put_cstr(&s
, ", drop");
4795 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4798 VLOG_INFO("%s: %s (msec:count)",
4799 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4803 return bucket
* BUCKET_WIDTH
;
4807 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4809 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4810 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4811 struct ofexpired expired
;
4813 if (facet
->installed
) {
4814 struct dpif_flow_stats stats
;
4816 facet_put__(ofproto
, facet
, facet
->actions
, facet
->actions_len
,
4818 facet_update_stats(ofproto
, facet
, &stats
);
4821 expired
.flow
= facet
->flow
;
4822 expired
.packet_count
= facet
->packet_count
;
4823 expired
.byte_count
= facet
->byte_count
;
4824 expired
.used
= facet
->used
;
4825 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4830 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4832 long long int cutoff
= time_msec() - dp_max_idle
;
4833 struct facet
*facet
, *next_facet
;
4835 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4836 facet_active_timeout(ofproto
, facet
);
4837 if (facet
->used
< cutoff
) {
4838 facet_remove(ofproto
, facet
);
4843 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4844 * then delete it entirely. */
4846 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4848 struct facet
*facet
, *next_facet
;
4852 /* Has 'rule' expired? */
4854 if (rule
->hard_timeout
4855 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4856 reason
= OFPRR_HARD_TIMEOUT
;
4857 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4858 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4859 reason
= OFPRR_IDLE_TIMEOUT
;
4864 COVERAGE_INC(ofproto_expired
);
4866 /* Update stats. (This is a no-op if the rule expired due to an idle
4867 * timeout, because that only happens when the rule has no facets left.) */
4868 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4869 facet_remove(ofproto
, facet
);
4872 /* Get rid of the rule. */
4873 if (!rule_is_hidden(rule
)) {
4874 rule_send_removed(ofproto
, rule
, reason
);
4876 rule_remove(ofproto
, rule
);
4880 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4882 struct ofputil_flow_removed fr
;
4883 struct ofconn
*ofconn
;
4885 if (!rule
->send_flow_removed
) {
4890 fr
.cookie
= rule
->flow_cookie
;
4892 calc_flow_duration__(rule
->created
, &fr
.duration_sec
, &fr
.duration_nsec
);
4893 fr
.idle_timeout
= rule
->idle_timeout
;
4894 fr
.packet_count
= rule
->packet_count
;
4895 fr
.byte_count
= rule
->byte_count
;
4897 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4898 if (!rconn_is_connected(ofconn
->rconn
)
4899 || !ofconn_receives_async_msgs(ofconn
)) {
4903 /* Account flow expirations under ofconn->reply_counter, the counter
4904 * for replies to OpenFlow requests. That works because preventing
4905 * OpenFlow requests from being processed also prevents new flows from
4906 * being added (and expiring). (It also prevents processing OpenFlow
4907 * requests that would not add new flows, so it is imperfect.) */
4908 queue_tx(ofputil_encode_flow_removed(&fr
, ofconn
->flow_format
),
4909 ofconn
, ofconn
->reply_counter
);
4913 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4914 * The returned statistics include statistics for all of 'rule''s facets. */
4916 rule_get_stats(const struct rule
*rule
, uint64_t *packets
, uint64_t *bytes
)
4919 struct facet
*facet
;
4921 /* Start from historical data for 'rule' itself that are no longer tracked
4922 * in facets. This counts, for example, facets that have expired. */
4923 p
= rule
->packet_count
;
4924 b
= rule
->byte_count
;
4926 /* Add any statistics that are tracked by facets. This includes
4927 * statistical data recently updated by ofproto_update_stats() as well as
4928 * stats for packets that were executed "by hand" via dpif_execute(). */
4929 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
4930 p
+= facet
->packet_count
;
4931 b
+= facet
->byte_count
;
4938 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4940 do_send_packet_in(struct ofpbuf
*ofp_packet_in
, void *ofconn_
)
4942 struct ofconn
*ofconn
= ofconn_
;
4944 rconn_send_with_limit(ofconn
->rconn
, ofp_packet_in
,
4945 ofconn
->packet_in_counter
, 100);
4948 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4949 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4950 * scheduler for sending.
4952 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4953 * Otherwise, ownership is transferred to this function. */
4955 schedule_packet_in(struct ofconn
*ofconn
, struct dpif_upcall
*upcall
,
4956 const struct flow
*flow
, bool clone
)
4958 enum { OPI_SIZE
= offsetof(struct ofp_packet_in
, data
) };
4959 struct ofproto
*ofproto
= ofconn
->ofproto
;
4960 struct ofp_packet_in
*opi
;
4961 int total_len
, send_len
;
4962 struct ofpbuf
*packet
;
4966 /* Get OpenFlow buffer_id. */
4967 if (upcall
->type
== DPIF_UC_ACTION
) {
4968 buffer_id
= UINT32_MAX
;
4969 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4970 buffer_id
= pktbuf_get_null();
4971 } else if (!ofconn
->pktbuf
) {
4972 buffer_id
= UINT32_MAX
;
4974 buffer_id
= pktbuf_save(ofconn
->pktbuf
, upcall
->packet
, flow
->in_port
);
4977 /* Figure out how much of the packet to send. */
4978 total_len
= send_len
= upcall
->packet
->size
;
4979 if (buffer_id
!= UINT32_MAX
) {
4980 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4982 if (upcall
->type
== DPIF_UC_ACTION
) {
4983 send_len
= MIN(send_len
, upcall
->userdata
);
4986 /* Copy or steal buffer for OFPT_PACKET_IN. */
4988 packet
= ofpbuf_clone_data_with_headroom(upcall
->packet
->data
,
4989 send_len
, OPI_SIZE
);
4991 packet
= upcall
->packet
;
4992 packet
->size
= send_len
;
4995 /* Add OFPT_PACKET_IN. */
4996 opi
= ofpbuf_push_zeros(packet
, OPI_SIZE
);
4997 opi
->header
.version
= OFP_VERSION
;
4998 opi
->header
.type
= OFPT_PACKET_IN
;
4999 opi
->total_len
= htons(total_len
);
5000 opi
->in_port
= htons(odp_port_to_ofp_port(flow
->in_port
));
5001 opi
->reason
= upcall
->type
== DPIF_UC_MISS
? OFPR_NO_MATCH
: OFPR_ACTION
;
5002 opi
->buffer_id
= htonl(buffer_id
);
5003 update_openflow_length(packet
);
5005 /* Hand over to packet scheduler. It might immediately call into
5006 * do_send_packet_in() or it might buffer it for a while (until a later
5007 * call to pinsched_run()). */
5008 idx
= upcall
->type
== DPIF_UC_MISS
? 0 : 1;
5009 pinsched_send(ofconn
->schedulers
[idx
], flow
->in_port
,
5010 packet
, do_send_packet_in
, ofconn
);
5013 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5014 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5015 * their individual configurations.
5017 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
5018 * Otherwise, ownership is transferred to this function. */
5020 send_packet_in(struct ofproto
*ofproto
, struct dpif_upcall
*upcall
,
5021 const struct flow
*flow
, bool clone
)
5023 struct ofconn
*ofconn
, *prev
;
5026 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
5027 if (ofconn_receives_async_msgs(ofconn
)) {
5029 schedule_packet_in(prev
, upcall
, flow
, true);
5035 schedule_packet_in(prev
, upcall
, flow
, clone
);
5036 } else if (!clone
) {
5037 ofpbuf_delete(upcall
->packet
);
5042 pick_datapath_id(const struct ofproto
*ofproto
)
5044 const struct ofport
*port
;
5046 port
= get_port(ofproto
, ODPP_LOCAL
);
5048 uint8_t ea
[ETH_ADDR_LEN
];
5051 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5053 return eth_addr_to_uint64(ea
);
5055 VLOG_WARN("could not get MAC address for %s (%s)",
5056 netdev_get_name(port
->netdev
), strerror(error
));
5058 return ofproto
->fallback_dpid
;
5062 pick_fallback_dpid(void)
5064 uint8_t ea
[ETH_ADDR_LEN
];
5065 eth_addr_nicira_random(ea
);
5066 return eth_addr_to_uint64(ea
);
5070 ofproto_unixctl_list(struct unixctl_conn
*conn
, const char *arg OVS_UNUSED
,
5071 void *aux OVS_UNUSED
)
5073 const struct shash_node
*node
;
5077 SHASH_FOR_EACH (node
, &all_ofprotos
) {
5078 ds_put_format(&results
, "%s\n", node
->name
);
5080 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
5081 ds_destroy(&results
);
5084 struct ofproto_trace
{
5085 struct action_xlate_ctx ctx
;
5091 trace_format_rule(struct ds
*result
, int level
, const struct rule
*rule
)
5093 ds_put_char_multiple(result
, '\t', level
);
5095 ds_put_cstr(result
, "No match\n");
5099 ds_put_format(result
, "Rule: cookie=%#"PRIx64
" ",
5100 ntohll(rule
->flow_cookie
));
5101 cls_rule_format(&rule
->cr
, result
);
5102 ds_put_char(result
, '\n');
5104 ds_put_char_multiple(result
, '\t', level
);
5105 ds_put_cstr(result
, "OpenFlow ");
5106 ofp_print_actions(result
, (const struct ofp_action_header
*) rule
->actions
,
5107 rule
->n_actions
* sizeof *rule
->actions
);
5108 ds_put_char(result
, '\n');
5112 trace_format_flow(struct ds
*result
, int level
, const char *title
,
5113 struct ofproto_trace
*trace
)
5115 ds_put_char_multiple(result
, '\t', level
);
5116 ds_put_format(result
, "%s: ", title
);
5117 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
5118 ds_put_cstr(result
, "unchanged");
5120 flow_format(result
, &trace
->ctx
.flow
);
5121 trace
->flow
= trace
->ctx
.flow
;
5123 ds_put_char(result
, '\n');
5127 trace_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
5129 struct ofproto_trace
*trace
= CONTAINER_OF(ctx
, struct ofproto_trace
, ctx
);
5130 struct ds
*result
= trace
->result
;
5132 ds_put_char(result
, '\n');
5133 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
5134 trace_format_rule(result
, ctx
->recurse
+ 1, rule
);
5138 ofproto_unixctl_trace(struct unixctl_conn
*conn
, const char *args_
,
5139 void *aux OVS_UNUSED
)
5141 char *dpname
, *in_port_s
, *tun_id_s
, *packet_s
;
5142 char *args
= xstrdup(args_
);
5143 char *save_ptr
= NULL
;
5144 struct ofproto
*ofproto
;
5145 struct ofpbuf packet
;
5153 ofpbuf_init(&packet
, strlen(args
) / 2);
5156 dpname
= strtok_r(args
, " ", &save_ptr
);
5157 tun_id_s
= strtok_r(NULL
, " ", &save_ptr
);
5158 in_port_s
= strtok_r(NULL
, " ", &save_ptr
);
5159 packet_s
= strtok_r(NULL
, "", &save_ptr
); /* Get entire rest of line. */
5160 if (!dpname
|| !in_port_s
|| !packet_s
) {
5161 unixctl_command_reply(conn
, 501, "Bad command syntax");
5165 ofproto
= shash_find_data(&all_ofprotos
, dpname
);
5167 unixctl_command_reply(conn
, 501, "Unknown ofproto (use ofproto/list "
5172 tun_id
= htonll(strtoull(tun_id_s
, NULL
, 0));
5173 in_port
= ofp_port_to_odp_port(atoi(in_port_s
));
5175 packet_s
= ofpbuf_put_hex(&packet
, packet_s
, NULL
);
5176 packet_s
+= strspn(packet_s
, " ");
5177 if (*packet_s
!= '\0') {
5178 unixctl_command_reply(conn
, 501, "Trailing garbage in command");
5181 if (packet
.size
< ETH_HEADER_LEN
) {
5182 unixctl_command_reply(conn
, 501, "Packet data too short for Ethernet");
5186 ds_put_cstr(&result
, "Packet: ");
5187 s
= ofp_packet_to_string(packet
.data
, packet
.size
, packet
.size
);
5188 ds_put_cstr(&result
, s
);
5191 flow_extract(&packet
, tun_id
, in_port
, &flow
);
5192 ds_put_cstr(&result
, "Flow: ");
5193 flow_format(&result
, &flow
);
5194 ds_put_char(&result
, '\n');
5196 rule
= rule_lookup(ofproto
, &flow
);
5197 trace_format_rule(&result
, 0, rule
);
5199 struct ofproto_trace trace
;
5200 struct ofpbuf
*odp_actions
;
5202 trace
.result
= &result
;
5204 action_xlate_ctx_init(&trace
.ctx
, ofproto
, &flow
, &packet
);
5205 trace
.ctx
.resubmit_hook
= trace_resubmit
;
5206 odp_actions
= xlate_actions(&trace
.ctx
,
5207 rule
->actions
, rule
->n_actions
);
5209 ds_put_char(&result
, '\n');
5210 trace_format_flow(&result
, 0, "Final flow", &trace
);
5211 ds_put_cstr(&result
, "Datapath actions: ");
5212 format_odp_actions(&result
, odp_actions
->data
, odp_actions
->size
);
5213 ofpbuf_delete(odp_actions
);
5216 unixctl_command_reply(conn
, 200, ds_cstr(&result
));
5219 ds_destroy(&result
);
5220 ofpbuf_uninit(&packet
);
5225 ofproto_unixctl_init(void)
5227 static bool registered
;
5233 unixctl_command_register("ofproto/list", ofproto_unixctl_list
, NULL
);
5234 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace
, NULL
);
5238 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5239 struct ofpbuf
*odp_actions
, tag_type
*tags
,
5240 uint16_t *nf_output_iface
, void *ofproto_
)
5242 struct ofproto
*ofproto
= ofproto_
;
5243 struct mac_entry
*dst_mac
;
5245 /* Drop frames for reserved multicast addresses. */
5246 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5250 /* Learn source MAC (but don't try to learn from revalidation). */
5252 && mac_learning_may_learn(ofproto
->ml
, flow
->dl_src
, 0)) {
5253 struct mac_entry
*src_mac
;
5255 src_mac
= mac_learning_insert(ofproto
->ml
, flow
->dl_src
, 0);
5256 if (mac_entry_is_new(src_mac
) || src_mac
->port
.i
!= flow
->in_port
) {
5257 /* The log messages here could actually be useful in debugging,
5258 * so keep the rate limit relatively high. */
5259 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5260 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5261 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5263 ofproto_revalidate(ofproto
,
5264 mac_learning_changed(ofproto
->ml
, src_mac
));
5265 src_mac
->port
.i
= flow
->in_port
;
5269 /* Determine output port. */
5270 dst_mac
= mac_learning_lookup(ofproto
->ml
, flow
->dl_dst
, 0, tags
);
5272 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5273 nf_output_iface
, odp_actions
);
5275 int out_port
= dst_mac
->port
.i
;
5276 if (out_port
!= flow
->in_port
) {
5277 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, out_port
);
5278 *nf_output_iface
= out_port
;
5287 static const struct ofhooks default_ofhooks
= {
5288 default_normal_ofhook_cb
,