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"
28 #include "classifier.h"
30 #include "discovery.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"
58 #include "stream-ssl.h"
62 #include "unaligned.h"
67 VLOG_DEFINE_THIS_MODULE(ofproto
);
69 COVERAGE_DEFINE(facet_changed_rule
);
70 COVERAGE_DEFINE(facet_revalidate
);
71 COVERAGE_DEFINE(odp_overflow
);
72 COVERAGE_DEFINE(ofproto_agg_request
);
73 COVERAGE_DEFINE(ofproto_costly_flags
);
74 COVERAGE_DEFINE(ofproto_ctlr_action
);
75 COVERAGE_DEFINE(ofproto_del_rule
);
76 COVERAGE_DEFINE(ofproto_error
);
77 COVERAGE_DEFINE(ofproto_expiration
);
78 COVERAGE_DEFINE(ofproto_expired
);
79 COVERAGE_DEFINE(ofproto_flows_req
);
80 COVERAGE_DEFINE(ofproto_flush
);
81 COVERAGE_DEFINE(ofproto_invalidated
);
82 COVERAGE_DEFINE(ofproto_no_packet_in
);
83 COVERAGE_DEFINE(ofproto_ofconn_stuck
);
84 COVERAGE_DEFINE(ofproto_ofp2odp
);
85 COVERAGE_DEFINE(ofproto_packet_in
);
86 COVERAGE_DEFINE(ofproto_packet_out
);
87 COVERAGE_DEFINE(ofproto_queue_req
);
88 COVERAGE_DEFINE(ofproto_recv_openflow
);
89 COVERAGE_DEFINE(ofproto_reinit_ports
);
90 COVERAGE_DEFINE(ofproto_unexpected_rule
);
91 COVERAGE_DEFINE(ofproto_uninstallable
);
92 COVERAGE_DEFINE(ofproto_update_port
);
94 #include "sflow_api.h"
96 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
97 * flow translation. */
98 #define MAX_RESUBMIT_RECURSION 16
103 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
104 struct netdev
*netdev
;
105 struct ofp_phy_port opp
; /* In host byte order. */
109 static void ofport_free(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 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
339 struct status_category
*ss
; /* Switch status category. */
340 enum ofproto_band band
; /* In-band or out-of-band? */
344 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
346 static void ofconn_destroy(struct ofconn
*);
347 static void ofconn_run(struct ofconn
*);
348 static void ofconn_wait(struct ofconn
*);
349 static bool ofconn_receives_async_msgs(const struct ofconn
*);
350 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
351 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
353 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
354 struct rconn_packet_counter
*counter
);
356 static void send_packet_in(struct ofproto
*, struct dpif_upcall
*,
357 const struct flow
*, bool clone
);
358 static void do_send_packet_in(struct ofpbuf
*ofp_packet_in
, void *ofconn
);
362 uint64_t datapath_id
; /* Datapath ID. */
363 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
364 char *mfr_desc
; /* Manufacturer. */
365 char *hw_desc
; /* Hardware. */
366 char *sw_desc
; /* Software version. */
367 char *serial_desc
; /* Serial number. */
368 char *dp_desc
; /* Datapath description. */
372 struct netdev_monitor
*netdev_monitor
;
373 struct hmap ports
; /* Contains "struct ofport"s. */
374 struct shash port_by_name
;
378 struct switch_status
*switch_status
;
379 struct fail_open
*fail_open
;
380 struct netflow
*netflow
;
381 struct ofproto_sflow
*sflow
;
383 /* In-band control. */
384 struct in_band
*in_band
;
385 long long int next_in_band_update
;
386 struct sockaddr_in
*extra_in_band_remotes
;
387 size_t n_extra_remotes
;
391 struct classifier cls
;
392 long long int next_expiration
;
396 bool need_revalidate
;
397 struct tag_set revalidate_set
;
399 /* OpenFlow connections. */
400 struct hmap controllers
; /* Controller "struct ofconn"s. */
401 struct list all_conns
; /* Contains "struct ofconn"s. */
402 enum ofproto_fail_mode fail_mode
;
404 /* OpenFlow listeners. */
405 struct hmap services
; /* Contains "struct ofservice"s. */
406 struct pvconn
**snoops
;
409 /* Hooks for ovs-vswitchd. */
410 const struct ofhooks
*ofhooks
;
413 /* Used by default ofhooks. */
414 struct mac_learning
*ml
;
417 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
418 static struct shash all_ofprotos
= SHASH_INITIALIZER(&all_ofprotos
);
420 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
422 static const struct ofhooks default_ofhooks
;
424 static uint64_t pick_datapath_id(const struct ofproto
*);
425 static uint64_t pick_fallback_dpid(void);
427 static int ofproto_expire(struct ofproto
*);
428 static void flow_push_stats(struct ofproto
*, const struct rule
*,
429 struct flow
*, uint64_t packets
, uint64_t bytes
,
432 static void handle_upcall(struct ofproto
*, struct dpif_upcall
*);
434 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
436 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
437 static void update_port(struct ofproto
*, const char *devname
);
438 static int init_ports(struct ofproto
*);
439 static void reinit_ports(struct ofproto
*);
441 static void ofproto_unixctl_init(void);
444 ofproto_create(const char *datapath
, const char *datapath_type
,
445 const struct ofhooks
*ofhooks
, void *aux
,
446 struct ofproto
**ofprotop
)
454 ofproto_unixctl_init();
456 /* Connect to datapath and start listening for messages. */
457 error
= dpif_open(datapath
, datapath_type
, &dpif
);
459 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
462 error
= dpif_recv_set_mask(dpif
,
463 ((1u << DPIF_UC_MISS
) |
464 (1u << DPIF_UC_ACTION
) |
465 (1u << DPIF_UC_SAMPLE
)));
467 VLOG_ERR("failed to listen on datapath %s: %s",
468 datapath
, strerror(error
));
472 dpif_flow_flush(dpif
);
473 dpif_recv_purge(dpif
);
475 /* Initialize settings. */
476 p
= xzalloc(sizeof *p
);
477 p
->fallback_dpid
= pick_fallback_dpid();
478 p
->datapath_id
= p
->fallback_dpid
;
479 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
480 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
481 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
482 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
483 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
485 /* Initialize datapath. */
487 p
->netdev_monitor
= netdev_monitor_create();
488 hmap_init(&p
->ports
);
489 shash_init(&p
->port_by_name
);
490 p
->max_ports
= dpif_get_max_ports(dpif
);
492 /* Initialize submodules. */
493 p
->switch_status
= switch_status_create(p
);
498 /* Initialize in-band control. */
500 p
->in_band_queue
= -1;
502 /* Initialize flow table. */
503 classifier_init(&p
->cls
);
504 p
->next_expiration
= time_msec() + 1000;
506 /* Initialize facet table. */
507 hmap_init(&p
->facets
);
508 p
->need_revalidate
= false;
509 tag_set_init(&p
->revalidate_set
);
511 /* Initialize OpenFlow connections. */
512 list_init(&p
->all_conns
);
513 hmap_init(&p
->controllers
);
514 hmap_init(&p
->services
);
518 /* Initialize hooks. */
520 p
->ofhooks
= ofhooks
;
524 p
->ofhooks
= &default_ofhooks
;
526 p
->ml
= mac_learning_create();
529 /* Pick final datapath ID. */
530 p
->datapath_id
= pick_datapath_id(p
);
531 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
533 shash_add_once(&all_ofprotos
, dpif_name(p
->dpif
), p
);
540 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
542 uint64_t old_dpid
= p
->datapath_id
;
543 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
544 if (p
->datapath_id
!= old_dpid
) {
545 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
547 /* Force all active connections to reconnect, since there is no way to
548 * notify a controller that the datapath ID has changed. */
549 ofproto_reconnect_controllers(p
);
554 is_discovery_controller(const struct ofproto_controller
*c
)
556 return !strcmp(c
->target
, "discover");
560 is_in_band_controller(const struct ofproto_controller
*c
)
562 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
565 /* Creates a new controller in 'ofproto'. Some of the settings are initially
566 * drawn from 'c', but update_controller() needs to be called later to finish
567 * the new ofconn's configuration. */
569 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
571 struct discovery
*discovery
;
572 struct ofconn
*ofconn
;
574 if (is_discovery_controller(c
)) {
575 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
576 ofproto
->dpif
, ofproto
->switch_status
,
585 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
586 ofconn
->pktbuf
= pktbuf_create();
587 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
589 ofconn
->discovery
= discovery
;
591 char *name
= ofconn_make_name(ofproto
, c
->target
);
592 rconn_connect(ofconn
->rconn
, c
->target
, name
);
595 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
596 hash_string(c
->target
, 0));
599 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
600 * target or turn discovery on or off (these are done by creating new ofconns
601 * and deleting old ones), but it can update the rest of an ofconn's
604 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
608 ofconn
->band
= (is_in_band_controller(c
)
609 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
611 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
613 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
614 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
616 if (ofconn
->discovery
) {
617 discovery_set_update_resolv_conf(ofconn
->discovery
,
618 c
->update_resolv_conf
);
619 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
622 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
626 ofconn_get_target(const struct ofconn
*ofconn
)
628 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
631 static struct ofconn
*
632 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
634 struct ofconn
*ofconn
;
636 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
637 hash_string(target
, 0), &ofproto
->controllers
) {
638 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
646 update_in_band_remotes(struct ofproto
*ofproto
)
648 const struct ofconn
*ofconn
;
649 struct sockaddr_in
*addrs
;
650 size_t max_addrs
, n_addrs
;
654 /* Allocate enough memory for as many remotes as we could possibly have. */
655 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
656 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
659 /* Add all the remotes. */
661 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
662 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
664 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
668 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
669 if (sin
->sin_addr
.s_addr
) {
670 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
673 if (ofconn
->discovery
) {
677 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
678 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
681 /* Create or update or destroy in-band.
683 * Ordinarily we only enable in-band if there's at least one remote
684 * address, but discovery needs the in-band rules for DHCP to be installed
685 * even before we know any remote addresses. */
686 if (n_addrs
|| discovery
) {
687 if (!ofproto
->in_band
) {
688 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
691 if (ofproto
->in_band
) {
692 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
694 in_band_set_queue(ofproto
->in_band
, ofproto
->in_band_queue
);
695 ofproto
->next_in_band_update
= time_msec() + 1000;
697 in_band_destroy(ofproto
->in_band
);
698 ofproto
->in_band
= NULL
;
706 update_fail_open(struct ofproto
*p
)
708 struct ofconn
*ofconn
;
710 if (!hmap_is_empty(&p
->controllers
)
711 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
712 struct rconn
**rconns
;
716 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
720 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
721 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
722 rconns
[n
++] = ofconn
->rconn
;
725 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
726 /* p->fail_open takes ownership of 'rconns'. */
728 fail_open_destroy(p
->fail_open
);
734 ofproto_set_controllers(struct ofproto
*p
,
735 const struct ofproto_controller
*controllers
,
736 size_t n_controllers
)
738 struct shash new_controllers
;
739 struct ofconn
*ofconn
, *next_ofconn
;
740 struct ofservice
*ofservice
, *next_ofservice
;
744 /* Create newly configured controllers and services.
745 * Create a name to ofproto_controller mapping in 'new_controllers'. */
746 shash_init(&new_controllers
);
747 for (i
= 0; i
< n_controllers
; i
++) {
748 const struct ofproto_controller
*c
= &controllers
[i
];
750 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
751 if (!find_controller_by_target(p
, c
->target
)) {
752 add_controller(p
, c
);
754 } else if (!pvconn_verify_name(c
->target
)) {
755 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
759 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
760 dpif_name(p
->dpif
), c
->target
);
764 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
767 /* Delete controllers that are no longer configured.
768 * Update configuration of all now-existing controllers. */
770 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
771 struct ofproto_controller
*c
;
773 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
775 ofconn_destroy(ofconn
);
777 update_controller(ofconn
, c
);
784 /* Delete services that are no longer configured.
785 * Update configuration of all now-existing services. */
786 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
787 struct ofproto_controller
*c
;
789 c
= shash_find_data(&new_controllers
,
790 pvconn_get_name(ofservice
->pvconn
));
792 ofservice_destroy(p
, ofservice
);
794 ofservice_reconfigure(ofservice
, c
);
798 shash_destroy(&new_controllers
);
800 update_in_band_remotes(p
);
803 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
804 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
805 struct ofconn
, hmap_node
);
806 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
807 rconn_status_cb
, ofconn
->rconn
);
812 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
814 p
->fail_mode
= fail_mode
;
818 /* Drops the connections between 'ofproto' and all of its controllers, forcing
819 * them to reconnect. */
821 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
823 struct ofconn
*ofconn
;
825 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
826 rconn_reconnect(ofconn
->rconn
);
831 any_extras_changed(const struct ofproto
*ofproto
,
832 const struct sockaddr_in
*extras
, size_t n
)
836 if (n
!= ofproto
->n_extra_remotes
) {
840 for (i
= 0; i
< n
; i
++) {
841 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
842 const struct sockaddr_in
*new = &extras
[i
];
844 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
845 old
->sin_port
!= new->sin_port
) {
853 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
854 * in-band control should guarantee access, in the same way that in-band
855 * control guarantees access to OpenFlow controllers. */
857 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
858 const struct sockaddr_in
*extras
, size_t n
)
860 if (!any_extras_changed(ofproto
, extras
, n
)) {
864 free(ofproto
->extra_in_band_remotes
);
865 ofproto
->n_extra_remotes
= n
;
866 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
868 update_in_band_remotes(ofproto
);
871 /* Sets the OpenFlow queue used by flows set up by in-band control on
872 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
873 * flows will use the default queue. */
875 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
877 if (queue_id
!= ofproto
->in_band_queue
) {
878 ofproto
->in_band_queue
= queue_id
;
879 update_in_band_remotes(ofproto
);
884 ofproto_set_desc(struct ofproto
*p
,
885 const char *mfr_desc
, const char *hw_desc
,
886 const char *sw_desc
, const char *serial_desc
,
889 struct ofp_desc_stats
*ods
;
892 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
893 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
894 sizeof ods
->mfr_desc
);
897 p
->mfr_desc
= xstrdup(mfr_desc
);
900 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
901 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
902 sizeof ods
->hw_desc
);
905 p
->hw_desc
= xstrdup(hw_desc
);
908 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
909 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
910 sizeof ods
->sw_desc
);
913 p
->sw_desc
= xstrdup(sw_desc
);
916 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
917 VLOG_WARN("truncating serial_desc, must be less than %zu "
919 sizeof ods
->serial_num
);
921 free(p
->serial_desc
);
922 p
->serial_desc
= xstrdup(serial_desc
);
925 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
926 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
927 sizeof ods
->dp_desc
);
930 p
->dp_desc
= xstrdup(dp_desc
);
935 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
936 const struct svec
*svec
)
938 struct pvconn
**pvconns
= *pvconnsp
;
939 size_t n_pvconns
= *n_pvconnsp
;
943 for (i
= 0; i
< n_pvconns
; i
++) {
944 pvconn_close(pvconns
[i
]);
948 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
950 for (i
= 0; i
< svec
->n
; i
++) {
951 const char *name
= svec
->names
[i
];
952 struct pvconn
*pvconn
;
955 error
= pvconn_open(name
, &pvconn
);
957 pvconns
[n_pvconns
++] = pvconn
;
959 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
967 *n_pvconnsp
= n_pvconns
;
973 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
975 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
979 ofproto_set_netflow(struct ofproto
*ofproto
,
980 const struct netflow_options
*nf_options
)
982 if (nf_options
&& nf_options
->collectors
.n
) {
983 if (!ofproto
->netflow
) {
984 ofproto
->netflow
= netflow_create();
986 return netflow_set_options(ofproto
->netflow
, nf_options
);
988 netflow_destroy(ofproto
->netflow
);
989 ofproto
->netflow
= NULL
;
995 ofproto_set_sflow(struct ofproto
*ofproto
,
996 const struct ofproto_sflow_options
*oso
)
998 struct ofproto_sflow
*os
= ofproto
->sflow
;
1001 struct ofport
*ofport
;
1003 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
1004 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
1005 ofproto_sflow_add_port(os
, ofport
->odp_port
,
1006 netdev_get_name(ofport
->netdev
));
1009 ofproto_sflow_set_options(os
, oso
);
1011 ofproto_sflow_destroy(os
);
1012 ofproto
->sflow
= NULL
;
1017 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
1019 return ofproto
->datapath_id
;
1023 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
1025 return !hmap_is_empty(&ofproto
->controllers
);
1028 enum ofproto_fail_mode
1029 ofproto_get_fail_mode(const struct ofproto
*p
)
1031 return p
->fail_mode
;
1035 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
1039 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
1040 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
1045 ofproto_destroy(struct ofproto
*p
)
1047 struct ofservice
*ofservice
, *next_ofservice
;
1048 struct ofconn
*ofconn
, *next_ofconn
;
1049 struct ofport
*ofport
, *next_ofport
;
1056 shash_find_and_delete(&all_ofprotos
, dpif_name(p
->dpif
));
1058 /* Destroy fail-open and in-band early, since they touch the classifier. */
1059 fail_open_destroy(p
->fail_open
);
1060 p
->fail_open
= NULL
;
1062 in_band_destroy(p
->in_band
);
1064 free(p
->extra_in_band_remotes
);
1066 ofproto_flush_flows(p
);
1067 classifier_destroy(&p
->cls
);
1068 hmap_destroy(&p
->facets
);
1070 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1071 ofconn_destroy(ofconn
);
1073 hmap_destroy(&p
->controllers
);
1075 dpif_close(p
->dpif
);
1076 netdev_monitor_destroy(p
->netdev_monitor
);
1077 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1078 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1079 ofport_free(ofport
);
1081 shash_destroy(&p
->port_by_name
);
1083 switch_status_destroy(p
->switch_status
);
1084 netflow_destroy(p
->netflow
);
1085 ofproto_sflow_destroy(p
->sflow
);
1087 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
1088 ofservice_destroy(p
, ofservice
);
1090 hmap_destroy(&p
->services
);
1092 for (i
= 0; i
< p
->n_snoops
; i
++) {
1093 pvconn_close(p
->snoops
[i
]);
1097 mac_learning_destroy(p
->ml
);
1102 free(p
->serial_desc
);
1105 hmap_destroy(&p
->ports
);
1111 ofproto_run(struct ofproto
*p
)
1113 int error
= ofproto_run1(p
);
1115 error
= ofproto_run2(p
, false);
1121 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1123 if (error
== ENOBUFS
) {
1124 reinit_ports(ofproto
);
1125 } else if (!error
) {
1126 update_port(ofproto
, devname
);
1131 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1132 * means that 'ofconn' is more interesting for monitoring than a lower return
1135 snoop_preference(const struct ofconn
*ofconn
)
1137 switch (ofconn
->role
) {
1138 case NX_ROLE_MASTER
:
1145 /* Shouldn't happen. */
1150 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1151 * Connects this vconn to a controller. */
1153 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1155 struct ofconn
*ofconn
, *best
;
1157 /* Pick a controller for monitoring. */
1159 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1160 if (ofconn
->type
== OFCONN_PRIMARY
1161 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1167 rconn_add_monitor(best
->rconn
, vconn
);
1169 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1175 ofproto_run1(struct ofproto
*p
)
1177 struct ofconn
*ofconn
, *next_ofconn
;
1178 struct ofservice
*ofservice
;
1183 if (shash_is_empty(&p
->port_by_name
)) {
1187 for (i
= 0; i
< 50; i
++) {
1188 struct dpif_upcall packet
;
1190 error
= dpif_recv(p
->dpif
, &packet
);
1192 if (error
== ENODEV
) {
1193 /* Someone destroyed the datapath behind our back. The caller
1194 * better destroy us and give up, because we're just going to
1195 * spin from here on out. */
1196 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1197 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1198 dpif_name(p
->dpif
));
1204 handle_upcall(p
, &packet
);
1207 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1208 process_port_change(p
, error
, devname
);
1210 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1211 &devname
)) != EAGAIN
) {
1212 process_port_change(p
, error
, devname
);
1216 if (time_msec() >= p
->next_in_band_update
) {
1217 update_in_band_remotes(p
);
1219 in_band_run(p
->in_band
);
1222 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1226 /* Fail-open maintenance. Do this after processing the ofconns since
1227 * fail-open checks the status of the controller rconn. */
1229 fail_open_run(p
->fail_open
);
1232 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1233 struct vconn
*vconn
;
1236 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1238 struct rconn
*rconn
;
1241 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1242 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1243 rconn_connect_unreliably(rconn
, vconn
, name
);
1246 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1247 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1248 ofservice
->burst_limit
);
1249 } else if (retval
!= EAGAIN
) {
1250 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1254 for (i
= 0; i
< p
->n_snoops
; i
++) {
1255 struct vconn
*vconn
;
1258 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1260 add_snooper(p
, vconn
);
1261 } else if (retval
!= EAGAIN
) {
1262 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1266 if (time_msec() >= p
->next_expiration
) {
1267 int delay
= ofproto_expire(p
);
1268 p
->next_expiration
= time_msec() + delay
;
1269 COVERAGE_INC(ofproto_expiration
);
1273 netflow_run(p
->netflow
);
1276 ofproto_sflow_run(p
->sflow
);
1283 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1285 /* Figure out what we need to revalidate now, if anything. */
1286 struct tag_set revalidate_set
= p
->revalidate_set
;
1287 if (p
->need_revalidate
) {
1288 revalidate_all
= true;
1291 /* Clear the revalidation flags. */
1292 tag_set_init(&p
->revalidate_set
);
1293 p
->need_revalidate
= false;
1295 /* Now revalidate if there's anything to do. */
1296 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1297 struct facet
*facet
, *next
;
1299 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1301 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1302 facet_revalidate(p
, facet
);
1311 ofproto_wait(struct ofproto
*p
)
1313 struct ofservice
*ofservice
;
1314 struct ofconn
*ofconn
;
1317 dpif_recv_wait(p
->dpif
);
1318 dpif_port_poll_wait(p
->dpif
);
1319 netdev_monitor_poll_wait(p
->netdev_monitor
);
1320 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1321 ofconn_wait(ofconn
);
1324 poll_timer_wait_until(p
->next_in_band_update
);
1325 in_band_wait(p
->in_band
);
1328 fail_open_wait(p
->fail_open
);
1331 ofproto_sflow_wait(p
->sflow
);
1333 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1334 poll_immediate_wake();
1336 if (p
->need_revalidate
) {
1337 /* Shouldn't happen, but if it does just go around again. */
1338 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1339 poll_immediate_wake();
1340 } else if (p
->next_expiration
!= LLONG_MAX
) {
1341 poll_timer_wait_until(p
->next_expiration
);
1343 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1344 pvconn_wait(ofservice
->pvconn
);
1346 for (i
= 0; i
< p
->n_snoops
; i
++) {
1347 pvconn_wait(p
->snoops
[i
]);
1352 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1354 tag_set_add(&ofproto
->revalidate_set
, tag
);
1358 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1360 return &ofproto
->revalidate_set
;
1364 ofproto_is_alive(const struct ofproto
*p
)
1366 return !hmap_is_empty(&p
->controllers
);
1370 ofproto_get_ofproto_controller_info(const struct ofproto
*ofproto
,
1373 const struct ofconn
*ofconn
;
1377 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
1378 const struct rconn
*rconn
= ofconn
->rconn
;
1379 time_t now
= time_now();
1380 time_t last_connection
= rconn_get_last_connection(rconn
);
1381 time_t last_disconnect
= rconn_get_last_disconnect(rconn
);
1382 const int last_error
= rconn_get_last_error(rconn
);
1383 struct ofproto_controller_info
*cinfo
= xmalloc(sizeof *cinfo
);
1385 shash_add(info
, rconn_get_target(rconn
), cinfo
);
1387 cinfo
->is_connected
= rconn_is_connected(rconn
);
1388 cinfo
->role
= ofconn
->role
;
1393 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "last_error";
1394 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1395 xstrdup(ovs_retval_to_string(last_error
));
1398 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "state";
1399 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1400 xstrdup(rconn_get_state(rconn
));
1402 if (last_connection
!= TIME_MIN
) {
1403 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "sec_since_connect";
1404 cinfo
->pairs
.values
[cinfo
->pairs
.n
++]
1405 = xasprintf("%ld", (long int) (now
- last_connection
));
1408 if (last_disconnect
!= TIME_MIN
) {
1409 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "sec_since_disconnect";
1410 cinfo
->pairs
.values
[cinfo
->pairs
.n
++]
1411 = xasprintf("%ld", (long int) (now
- last_disconnect
));
1417 ofproto_free_ofproto_controller_info(struct shash
*info
)
1419 struct shash_node
*node
;
1421 SHASH_FOR_EACH (node
, info
) {
1422 struct ofproto_controller_info
*cinfo
= node
->data
;
1423 while (cinfo
->pairs
.n
) {
1424 free((char *) cinfo
->pairs
.values
[--cinfo
->pairs
.n
]);
1428 shash_destroy(info
);
1431 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1433 * This is almost the same as calling dpif_port_del() directly on the
1434 * datapath, but it also makes 'ofproto' close its open netdev for the port
1435 * (if any). This makes it possible to create a new netdev of a different
1436 * type under the same name, which otherwise the netdev library would refuse
1437 * to do because of the conflict. (The netdev would eventually get closed on
1438 * the next trip through ofproto_run(), but this interface is more direct.)
1440 * Returns 0 if successful, otherwise a positive errno. */
1442 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1444 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1445 const char *name
= ofport
? ofport
->opp
.name
: "<unknown>";
1448 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1450 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1451 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1452 } else if (ofport
) {
1453 /* 'name' is ofport->opp.name and update_port() is going to destroy
1454 * 'ofport'. Just in case update_port() refers to 'name' after it
1455 * destroys 'ofport', make a copy of it around the update_port()
1457 char *devname
= xstrdup(name
);
1458 update_port(ofproto
, devname
);
1464 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1465 * true if 'odp_port' exists and should be included, false otherwise. */
1467 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1469 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1470 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1474 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1475 const union ofp_action
*actions
, size_t n_actions
,
1476 const struct ofpbuf
*packet
)
1478 struct action_xlate_ctx ctx
;
1479 struct ofpbuf
*odp_actions
;
1481 action_xlate_ctx_init(&ctx
, p
, flow
, packet
);
1482 /* Always xlate packets originated in this function. */
1483 ctx
.check_special
= false;
1484 odp_actions
= xlate_actions(&ctx
, actions
, n_actions
);
1486 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1488 dpif_execute(p
->dpif
, odp_actions
->data
, odp_actions
->size
, packet
);
1490 ofpbuf_delete(odp_actions
);
1495 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1496 * performs the 'n_actions' actions in 'actions'. The new flow will not
1499 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1500 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1501 * controllers; otherwise, it will be hidden.
1503 * The caller retains ownership of 'cls_rule' and 'actions'. */
1505 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1506 const union ofp_action
*actions
, size_t n_actions
)
1509 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1510 rule_insert(p
, rule
);
1514 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1518 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1521 rule_remove(ofproto
, rule
);
1526 ofproto_flush_flows(struct ofproto
*ofproto
)
1528 struct facet
*facet
, *next_facet
;
1529 struct rule
*rule
, *next_rule
;
1530 struct cls_cursor cursor
;
1532 COVERAGE_INC(ofproto_flush
);
1534 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1535 /* Mark the facet as not installed so that facet_remove() doesn't
1536 * bother trying to uninstall it. There is no point in uninstalling it
1537 * individually since we are about to blow away all the facets with
1538 * dpif_flow_flush(). */
1539 facet
->installed
= false;
1540 facet
->dp_packet_count
= 0;
1541 facet
->dp_byte_count
= 0;
1542 facet_remove(ofproto
, facet
);
1545 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1546 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1547 rule_remove(ofproto
, rule
);
1550 dpif_flow_flush(ofproto
->dpif
);
1551 if (ofproto
->in_band
) {
1552 in_band_flushed(ofproto
->in_band
);
1554 if (ofproto
->fail_open
) {
1555 fail_open_flushed(ofproto
->fail_open
);
1560 reinit_ports(struct ofproto
*p
)
1562 struct dpif_port_dump dump
;
1563 struct shash_node
*node
;
1564 struct shash devnames
;
1565 struct ofport
*ofport
;
1566 struct dpif_port dpif_port
;
1568 COVERAGE_INC(ofproto_reinit_ports
);
1570 shash_init(&devnames
);
1571 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1572 shash_add_once (&devnames
, ofport
->opp
.name
, NULL
);
1574 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1575 shash_add_once (&devnames
, dpif_port
.name
, NULL
);
1578 SHASH_FOR_EACH (node
, &devnames
) {
1579 update_port(p
, node
->name
);
1581 shash_destroy(&devnames
);
1584 static struct ofport
*
1585 make_ofport(const struct dpif_port
*dpif_port
)
1587 struct netdev_options netdev_options
;
1588 enum netdev_flags flags
;
1589 struct ofport
*ofport
;
1590 struct netdev
*netdev
;
1593 memset(&netdev_options
, 0, sizeof netdev_options
);
1594 netdev_options
.name
= dpif_port
->name
;
1595 netdev_options
.type
= dpif_port
->type
;
1596 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1598 error
= netdev_open(&netdev_options
, &netdev
);
1600 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1601 "cannot be opened (%s)",
1602 dpif_port
->name
, dpif_port
->port_no
,
1603 dpif_port
->name
, strerror(error
));
1607 ofport
= xzalloc(sizeof *ofport
);
1608 ofport
->netdev
= netdev
;
1609 ofport
->odp_port
= dpif_port
->port_no
;
1610 ofport
->opp
.port_no
= odp_port_to_ofp_port(dpif_port
->port_no
);
1611 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1612 ovs_strlcpy(ofport
->opp
.name
, dpif_port
->name
, sizeof ofport
->opp
.name
);
1614 netdev_get_flags(netdev
, &flags
);
1615 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1617 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1619 netdev_get_features(netdev
,
1620 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1621 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1626 ofport_conflicts(const struct ofproto
*p
, const struct dpif_port
*dpif_port
)
1628 if (get_port(p
, dpif_port
->port_no
)) {
1629 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1630 dpif_port
->port_no
);
1632 } else if (shash_find(&p
->port_by_name
, dpif_port
->name
)) {
1633 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1642 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1644 const struct ofp_phy_port
*a
= &a_
->opp
;
1645 const struct ofp_phy_port
*b
= &b_
->opp
;
1647 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1648 return (a
->port_no
== b
->port_no
1649 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1650 && !strcmp(a
->name
, b
->name
)
1651 && a
->state
== b
->state
1652 && a
->config
== b
->config
1653 && a
->curr
== b
->curr
1654 && a
->advertised
== b
->advertised
1655 && a
->supported
== b
->supported
1656 && a
->peer
== b
->peer
);
1660 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1663 /* XXX Should limit the number of queued port status change messages. */
1664 struct ofconn
*ofconn
;
1665 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1666 struct ofp_port_status
*ops
;
1669 /* Primary controllers, even slaves, should always get port status
1670 updates. Otherwise obey ofconn_receives_async_msgs(). */
1671 if (ofconn
->type
!= OFCONN_PRIMARY
1672 && !ofconn_receives_async_msgs(ofconn
)) {
1676 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1677 ops
->reason
= reason
;
1678 ops
->desc
= ofport
->opp
;
1679 hton_ofp_phy_port(&ops
->desc
);
1680 queue_tx(b
, ofconn
, NULL
);
1685 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1687 const char *netdev_name
= ofport
->opp
.name
;
1689 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1690 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1691 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1693 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1698 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1700 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1701 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1702 shash_delete(&p
->port_by_name
,
1703 shash_find(&p
->port_by_name
, ofport
->opp
.name
));
1705 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1710 ofport_free(struct ofport
*ofport
)
1713 netdev_close(ofport
->netdev
);
1718 static struct ofport
*
1719 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1721 struct ofport
*port
;
1723 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1724 hash_int(odp_port
, 0), &ofproto
->ports
) {
1725 if (port
->odp_port
== odp_port
) {
1733 update_port(struct ofproto
*p
, const char *devname
)
1735 struct dpif_port dpif_port
;
1736 struct ofport
*old_ofport
;
1737 struct ofport
*new_ofport
;
1740 COVERAGE_INC(ofproto_update_port
);
1742 /* Query the datapath for port information. */
1743 error
= dpif_port_query_by_name(p
->dpif
, devname
, &dpif_port
);
1745 /* Find the old ofport. */
1746 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1749 /* There's no port named 'devname' but there might be a port with
1750 * the same port number. This could happen if a port is deleted
1751 * and then a new one added in its place very quickly, or if a port
1752 * is renamed. In the former case we want to send an OFPPR_DELETE
1753 * and an OFPPR_ADD, and in the latter case we want to send a
1754 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1755 * the old port's ifindex against the new port, or perhaps less
1756 * reliably but more portably by comparing the old port's MAC
1757 * against the new port's MAC. However, this code isn't that smart
1758 * and always sends an OFPPR_MODIFY (XXX). */
1759 old_ofport
= get_port(p
, dpif_port
.port_no
);
1761 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1762 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1763 "%s", strerror(error
));
1767 /* Create a new ofport. */
1768 new_ofport
= !error
? make_ofport(&dpif_port
) : NULL
;
1770 /* Eliminate a few pathological cases. */
1771 if (!old_ofport
&& !new_ofport
) {
1773 } else if (old_ofport
&& new_ofport
) {
1774 /* Most of the 'config' bits are OpenFlow soft state, but
1775 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1776 * OpenFlow bits from old_ofport. (make_ofport() only sets
1777 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1778 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1780 if (ofport_equal(old_ofport
, new_ofport
)) {
1781 /* False alarm--no change. */
1782 ofport_free(new_ofport
);
1787 /* Now deal with the normal cases. */
1789 ofport_remove(p
, old_ofport
);
1792 ofport_install(p
, new_ofport
);
1794 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1795 (!old_ofport
? OFPPR_ADD
1796 : !new_ofport
? OFPPR_DELETE
1798 ofport_free(old_ofport
);
1801 dpif_port_destroy(&dpif_port
);
1805 init_ports(struct ofproto
*p
)
1807 struct dpif_port_dump dump
;
1808 struct dpif_port dpif_port
;
1810 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1811 if (!ofport_conflicts(p
, &dpif_port
)) {
1812 struct ofport
*ofport
= make_ofport(&dpif_port
);
1814 ofport_install(p
, ofport
);
1822 static struct ofconn
*
1823 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1825 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1826 ofconn
->ofproto
= p
;
1827 list_push_back(&p
->all_conns
, &ofconn
->node
);
1828 ofconn
->rconn
= rconn
;
1829 ofconn
->type
= type
;
1830 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1831 ofconn
->role
= NX_ROLE_OTHER
;
1832 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1833 ofconn
->pktbuf
= NULL
;
1834 ofconn
->miss_send_len
= 0;
1835 ofconn
->reply_counter
= rconn_packet_counter_create ();
1840 ofconn_destroy(struct ofconn
*ofconn
)
1842 if (ofconn
->type
== OFCONN_PRIMARY
) {
1843 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1845 discovery_destroy(ofconn
->discovery
);
1847 list_remove(&ofconn
->node
);
1848 switch_status_unregister(ofconn
->ss
);
1849 rconn_destroy(ofconn
->rconn
);
1850 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1851 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1852 pktbuf_destroy(ofconn
->pktbuf
);
1857 ofconn_run(struct ofconn
*ofconn
)
1859 struct ofproto
*p
= ofconn
->ofproto
;
1863 if (ofconn
->discovery
) {
1864 char *controller_name
;
1865 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1866 discovery_question_connectivity(ofconn
->discovery
);
1868 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1869 if (controller_name
) {
1870 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1871 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1873 free(controller_name
);
1875 rconn_disconnect(ofconn
->rconn
);
1880 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1881 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1884 rconn_run(ofconn
->rconn
);
1886 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1887 /* Limit the number of iterations to prevent other tasks from
1889 for (iteration
= 0; iteration
< 50; iteration
++) {
1890 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1895 fail_open_maybe_recover(p
->fail_open
);
1897 handle_openflow(ofconn
, of_msg
);
1898 ofpbuf_delete(of_msg
);
1902 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1903 ofconn_destroy(ofconn
);
1908 ofconn_wait(struct ofconn
*ofconn
)
1912 if (ofconn
->discovery
) {
1913 discovery_wait(ofconn
->discovery
);
1915 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1916 pinsched_wait(ofconn
->schedulers
[i
]);
1918 rconn_run_wait(ofconn
->rconn
);
1919 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1920 rconn_recv_wait(ofconn
->rconn
);
1922 COVERAGE_INC(ofproto_ofconn_stuck
);
1926 /* Returns true if 'ofconn' should receive asynchronous messages. */
1928 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1930 if (ofconn
->type
== OFCONN_PRIMARY
) {
1931 /* Primary controllers always get asynchronous messages unless they
1932 * have configured themselves as "slaves". */
1933 return ofconn
->role
!= NX_ROLE_SLAVE
;
1935 /* Service connections don't get asynchronous messages unless they have
1936 * explicitly asked for them by setting a nonzero miss send length. */
1937 return ofconn
->miss_send_len
> 0;
1941 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1942 * and 'target', suitable for use in log messages for identifying the
1945 * The name is dynamically allocated. The caller should free it (with free())
1946 * when it is no longer needed. */
1948 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1950 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1954 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1958 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1959 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1963 *s
= pinsched_create(rate
, burst
,
1964 ofconn
->ofproto
->switch_status
);
1966 pinsched_set_limits(*s
, rate
, burst
);
1969 pinsched_destroy(*s
);
1976 ofservice_reconfigure(struct ofservice
*ofservice
,
1977 const struct ofproto_controller
*c
)
1979 ofservice
->probe_interval
= c
->probe_interval
;
1980 ofservice
->rate_limit
= c
->rate_limit
;
1981 ofservice
->burst_limit
= c
->burst_limit
;
1984 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1985 * positive errno value. */
1987 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1989 struct ofservice
*ofservice
;
1990 struct pvconn
*pvconn
;
1993 error
= pvconn_open(c
->target
, &pvconn
);
1998 ofservice
= xzalloc(sizeof *ofservice
);
1999 hmap_insert(&ofproto
->services
, &ofservice
->node
,
2000 hash_string(c
->target
, 0));
2001 ofservice
->pvconn
= pvconn
;
2003 ofservice_reconfigure(ofservice
, c
);
2009 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
2011 hmap_remove(&ofproto
->services
, &ofservice
->node
);
2012 pvconn_close(ofservice
->pvconn
);
2016 /* Finds and returns the ofservice within 'ofproto' that has the given
2017 * 'target', or a null pointer if none exists. */
2018 static struct ofservice
*
2019 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
2021 struct ofservice
*ofservice
;
2023 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
2024 &ofproto
->services
) {
2025 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
2032 /* Returns true if 'rule' should be hidden from the controller.
2034 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2035 * (e.g. by in-band control) and are intentionally hidden from the
2038 rule_is_hidden(const struct rule
*rule
)
2040 return rule
->cr
.priority
> UINT16_MAX
;
2043 /* Creates and returns a new rule initialized as specified.
2045 * The caller is responsible for inserting the rule into the classifier (with
2046 * rule_insert()). */
2047 static struct rule
*
2048 rule_create(const struct cls_rule
*cls_rule
,
2049 const union ofp_action
*actions
, size_t n_actions
,
2050 uint16_t idle_timeout
, uint16_t hard_timeout
,
2051 ovs_be64 flow_cookie
, bool send_flow_removed
)
2053 struct rule
*rule
= xzalloc(sizeof *rule
);
2054 rule
->cr
= *cls_rule
;
2055 rule
->idle_timeout
= idle_timeout
;
2056 rule
->hard_timeout
= hard_timeout
;
2057 rule
->flow_cookie
= flow_cookie
;
2058 rule
->used
= rule
->created
= time_msec();
2059 rule
->send_flow_removed
= send_flow_removed
;
2060 list_init(&rule
->facets
);
2061 if (n_actions
> 0) {
2062 rule
->n_actions
= n_actions
;
2063 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
2069 static struct rule
*
2070 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
2072 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
2076 rule_free(struct rule
*rule
)
2078 free(rule
->actions
);
2082 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2083 * destroying any that no longer has a rule (which is probably all of them).
2085 * The caller must have already removed 'rule' from the classifier. */
2087 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
2089 struct facet
*facet
, *next_facet
;
2090 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
2091 facet_revalidate(ofproto
, facet
);
2096 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2097 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2100 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
2102 const union ofp_action
*oa
;
2103 struct actions_iterator i
;
2105 if (out_port
== htons(OFPP_NONE
)) {
2108 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
2109 oa
= actions_next(&i
)) {
2110 if (action_outputs_to_port(oa
, out_port
)) {
2117 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2118 * 'packet', which arrived on 'in_port'.
2120 * Takes ownership of 'packet'. */
2122 execute_odp_actions(struct ofproto
*ofproto
, const struct flow
*flow
,
2123 const struct nlattr
*odp_actions
, size_t actions_len
,
2124 struct ofpbuf
*packet
)
2126 if (actions_len
== NLA_ALIGN(NLA_HDRLEN
+ sizeof(uint64_t))
2127 && odp_actions
->nla_type
== ODP_ACTION_ATTR_CONTROLLER
) {
2128 /* As an optimization, avoid a round-trip from userspace to kernel to
2129 * userspace. This also avoids possibly filling up kernel packet
2130 * buffers along the way. */
2131 struct dpif_upcall upcall
;
2133 upcall
.type
= DPIF_UC_ACTION
;
2134 upcall
.packet
= packet
;
2137 upcall
.userdata
= nl_attr_get_u64(odp_actions
);
2138 upcall
.sample_pool
= 0;
2139 upcall
.actions
= NULL
;
2140 upcall
.actions_len
= 0;
2142 send_packet_in(ofproto
, &upcall
, flow
, false);
2148 error
= dpif_execute(ofproto
->dpif
, odp_actions
, actions_len
, packet
);
2149 ofpbuf_delete(packet
);
2154 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2155 * statistics appropriately. 'packet' must have at least sizeof(struct
2156 * ofp_packet_in) bytes of headroom.
2158 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2159 * applying flow_extract() to 'packet' would yield the same flow as
2162 * 'facet' must have accurately composed ODP actions; that is, it must not be
2163 * in need of revalidation.
2165 * Takes ownership of 'packet'. */
2167 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2168 struct ofpbuf
*packet
)
2170 struct dpif_flow_stats stats
;
2172 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2174 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2175 stats
.used
= time_msec();
2176 if (execute_odp_actions(ofproto
, &facet
->flow
,
2177 facet
->actions
, facet
->actions_len
, packet
)) {
2178 facet_update_stats(ofproto
, facet
, &stats
);
2182 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2183 * statistics (or the statistics for one of its facets) appropriately.
2184 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2186 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2187 * with statistics for 'packet' either way.
2189 * Takes ownership of 'packet'. */
2191 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2192 struct ofpbuf
*packet
)
2194 struct action_xlate_ctx ctx
;
2195 struct ofpbuf
*odp_actions
;
2196 struct facet
*facet
;
2200 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2202 flow_extract(packet
, 0, in_port
, &flow
);
2204 /* First look for a related facet. If we find one, account it to that. */
2205 facet
= facet_lookup_valid(ofproto
, &flow
);
2206 if (facet
&& facet
->rule
== rule
) {
2207 facet_execute(ofproto
, facet
, packet
);
2211 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2212 * create a new facet for it and use that. */
2213 if (rule_lookup(ofproto
, &flow
) == rule
) {
2214 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2215 facet_execute(ofproto
, facet
, packet
);
2216 facet_install(ofproto
, facet
, true);
2220 /* We can't account anything to a facet. If we were to try, then that
2221 * facet would have a non-matching rule, busting our invariants. */
2222 action_xlate_ctx_init(&ctx
, ofproto
, &flow
, packet
);
2223 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2224 size
= packet
->size
;
2225 if (execute_odp_actions(ofproto
, &flow
, odp_actions
->data
,
2226 odp_actions
->size
, packet
)) {
2227 rule
->used
= time_msec();
2228 rule
->packet_count
++;
2229 rule
->byte_count
+= size
;
2230 flow_push_stats(ofproto
, rule
, &flow
, 1, size
, rule
->used
);
2232 ofpbuf_delete(odp_actions
);
2235 /* Inserts 'rule' into 'p''s flow table. */
2237 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2239 struct rule
*displaced_rule
;
2241 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2242 if (displaced_rule
) {
2243 rule_destroy(p
, displaced_rule
);
2245 p
->need_revalidate
= true;
2248 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2249 * 'flow' and an example 'packet' within that flow.
2251 * The caller must already have determined that no facet with an identical
2252 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2253 * 'ofproto''s classifier table. */
2254 static struct facet
*
2255 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2256 const struct flow
*flow
, const struct ofpbuf
*packet
)
2258 struct facet
*facet
;
2260 facet
= xzalloc(sizeof *facet
);
2261 facet
->used
= time_msec();
2262 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2263 list_push_back(&rule
->facets
, &facet
->list_node
);
2265 facet
->flow
= *flow
;
2266 netflow_flow_init(&facet
->nf_flow
);
2267 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2269 facet_make_actions(ofproto
, facet
, packet
);
2275 facet_free(struct facet
*facet
)
2277 free(facet
->actions
);
2281 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2283 * - Removes 'rule' from the classifier.
2285 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2286 * destroys them), via rule_destroy().
2289 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2291 COVERAGE_INC(ofproto_del_rule
);
2292 ofproto
->need_revalidate
= true;
2293 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2294 rule_destroy(ofproto
, rule
);
2297 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2299 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2300 * rule's statistics, via facet_uninstall().
2302 * - Removes 'facet' from its rule and from ofproto->facets.
2305 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2307 facet_uninstall(ofproto
, facet
);
2308 facet_flush_stats(ofproto
, facet
);
2309 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2310 list_remove(&facet
->list_node
);
2314 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2316 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2317 const struct ofpbuf
*packet
)
2319 const struct rule
*rule
= facet
->rule
;
2320 struct ofpbuf
*odp_actions
;
2321 struct action_xlate_ctx ctx
;
2323 action_xlate_ctx_init(&ctx
, p
, &facet
->flow
, packet
);
2324 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2325 facet
->tags
= ctx
.tags
;
2326 facet
->may_install
= ctx
.may_set_up_flow
;
2327 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2329 if (facet
->actions_len
!= odp_actions
->size
2330 || memcmp(facet
->actions
, odp_actions
->data
, odp_actions
->size
)) {
2331 free(facet
->actions
);
2332 facet
->actions_len
= odp_actions
->size
;
2333 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2336 ofpbuf_delete(odp_actions
);
2340 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
,
2341 const struct nlattr
*actions
, size_t actions_len
,
2342 struct dpif_flow_stats
*stats
)
2344 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2345 enum dpif_flow_put_flags flags
;
2348 flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
2350 flags
|= DPIF_FP_ZERO_STATS
;
2351 facet
->dp_packet_count
= 0;
2352 facet
->dp_byte_count
= 0;
2355 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2356 odp_flow_key_from_flow(&key
, &facet
->flow
);
2357 assert(key
.base
== keybuf
);
2359 return dpif_flow_put(ofproto
->dpif
, flags
, key
.data
, key
.size
,
2360 actions
, actions_len
, stats
);
2363 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2364 * 'zero_stats' is true, clears any existing statistics from the datapath for
2367 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2369 struct dpif_flow_stats stats
;
2371 if (facet
->may_install
2372 && !facet_put__(p
, facet
, facet
->actions
, facet
->actions_len
,
2373 zero_stats
? &stats
: NULL
)) {
2374 facet
->installed
= true;
2378 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2379 * to the accounting hook function in the ofhooks structure. */
2381 facet_account(struct ofproto
*ofproto
,
2382 struct facet
*facet
, uint64_t extra_bytes
)
2384 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2386 if (ofproto
->ofhooks
->account_flow_cb
2387 && total_bytes
> facet
->accounted_bytes
)
2389 ofproto
->ofhooks
->account_flow_cb(
2390 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->actions_len
,
2391 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2392 facet
->accounted_bytes
= total_bytes
;
2396 /* If 'rule' is installed in the datapath, uninstalls it. */
2398 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2400 if (facet
->installed
) {
2401 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2402 struct dpif_flow_stats stats
;
2405 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2406 odp_flow_key_from_flow(&key
, &facet
->flow
);
2407 assert(key
.base
== keybuf
);
2409 if (!dpif_flow_del(p
->dpif
, key
.data
, key
.size
, &stats
)) {
2410 facet_update_stats(p
, facet
, &stats
);
2412 facet
->installed
= false;
2413 facet
->dp_packet_count
= 0;
2414 facet
->dp_byte_count
= 0;
2416 assert(facet
->dp_packet_count
== 0);
2417 assert(facet
->dp_byte_count
== 0);
2421 /* Returns true if the only action for 'facet' is to send to the controller.
2422 * (We don't report NetFlow expiration messages for such facets because they
2423 * are just part of the control logic for the network, not real traffic). */
2425 facet_is_controller_flow(struct facet
*facet
)
2428 && facet
->rule
->n_actions
== 1
2429 && action_outputs_to_port(&facet
->rule
->actions
[0],
2430 htons(OFPP_CONTROLLER
)));
2433 /* Folds all of 'facet''s statistics into its rule. Also updates the
2434 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2435 * 'facet''s statistics in the datapath should have been zeroed and folded into
2436 * its packet and byte counts before this function is called. */
2438 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2440 assert(!facet
->dp_byte_count
);
2441 assert(!facet
->dp_packet_count
);
2443 facet_push_stats(ofproto
, facet
);
2444 facet_account(ofproto
, facet
, 0);
2446 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2447 struct ofexpired expired
;
2448 expired
.flow
= facet
->flow
;
2449 expired
.packet_count
= facet
->packet_count
;
2450 expired
.byte_count
= facet
->byte_count
;
2451 expired
.used
= facet
->used
;
2452 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2455 facet
->rule
->packet_count
+= facet
->packet_count
;
2456 facet
->rule
->byte_count
+= facet
->byte_count
;
2458 /* Reset counters to prevent double counting if 'facet' ever gets
2460 facet
->packet_count
= 0;
2461 facet
->byte_count
= 0;
2462 facet
->rs_packet_count
= 0;
2463 facet
->rs_byte_count
= 0;
2464 facet
->accounted_bytes
= 0;
2466 netflow_flow_clear(&facet
->nf_flow
);
2469 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2470 * Returns it if found, otherwise a null pointer.
2472 * The returned facet might need revalidation; use facet_lookup_valid()
2473 * instead if that is important. */
2474 static struct facet
*
2475 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2477 struct facet
*facet
;
2479 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2481 if (flow_equal(flow
, &facet
->flow
)) {
2489 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2490 * Returns it if found, otherwise a null pointer.
2492 * The returned facet is guaranteed to be valid. */
2493 static struct facet
*
2494 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2496 struct facet
*facet
= facet_find(ofproto
, flow
);
2498 /* The facet we found might not be valid, since we could be in need of
2499 * revalidation. If it is not valid, don't return it. */
2501 && ofproto
->need_revalidate
2502 && !facet_revalidate(ofproto
, facet
)) {
2503 COVERAGE_INC(ofproto_invalidated
);
2510 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2512 * - If the rule found is different from 'facet''s current rule, moves
2513 * 'facet' to the new rule and recompiles its actions.
2515 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2516 * where it is and recompiles its actions anyway.
2518 * - If there is none, destroys 'facet'.
2520 * Returns true if 'facet' still exists, false if it has been destroyed. */
2522 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2524 struct action_xlate_ctx ctx
;
2525 struct ofpbuf
*odp_actions
;
2526 struct rule
*new_rule
;
2527 bool actions_changed
;
2529 COVERAGE_INC(facet_revalidate
);
2531 /* Determine the new rule. */
2532 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2534 /* No new rule, so delete the facet. */
2535 facet_remove(ofproto
, facet
);
2539 /* Calculate new ODP actions.
2541 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2542 * emit a NetFlow expiration and, if so, we need to have the old state
2543 * around to properly compose it. */
2544 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
, NULL
);
2545 odp_actions
= xlate_actions(&ctx
, new_rule
->actions
, new_rule
->n_actions
);
2546 actions_changed
= (facet
->actions_len
!= odp_actions
->size
2547 || memcmp(facet
->actions
, odp_actions
->data
,
2548 facet
->actions_len
));
2550 /* If the ODP actions changed or the installability changed, then we need
2551 * to talk to the datapath. */
2552 if (actions_changed
|| ctx
.may_set_up_flow
!= facet
->installed
) {
2553 if (ctx
.may_set_up_flow
) {
2554 struct dpif_flow_stats stats
;
2556 facet_put__(ofproto
, facet
,
2557 odp_actions
->data
, odp_actions
->size
, &stats
);
2558 facet_update_stats(ofproto
, facet
, &stats
);
2560 facet_uninstall(ofproto
, facet
);
2563 /* The datapath flow is gone or has zeroed stats, so push stats out of
2564 * 'facet' into 'rule'. */
2565 facet_flush_stats(ofproto
, facet
);
2568 /* Update 'facet' now that we've taken care of all the old state. */
2569 facet
->tags
= ctx
.tags
;
2570 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2571 facet
->may_install
= ctx
.may_set_up_flow
;
2572 if (actions_changed
) {
2573 free(facet
->actions
);
2574 facet
->actions_len
= odp_actions
->size
;
2575 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2577 if (facet
->rule
!= new_rule
) {
2578 COVERAGE_INC(facet_changed_rule
);
2579 list_remove(&facet
->list_node
);
2580 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2581 facet
->rule
= new_rule
;
2582 facet
->used
= new_rule
->created
;
2583 facet
->rs_used
= facet
->used
;
2586 ofpbuf_delete(odp_actions
);
2592 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2593 struct rconn_packet_counter
*counter
)
2595 update_openflow_length(msg
);
2596 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2602 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2605 struct ofpbuf
*buf
= ofputil_encode_error_msg(error
, oh
);
2607 COVERAGE_INC(ofproto_error
);
2608 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2613 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2615 opp
->port_no
= htons(opp
->port_no
);
2616 opp
->config
= htonl(opp
->config
);
2617 opp
->state
= htonl(opp
->state
);
2618 opp
->curr
= htonl(opp
->curr
);
2619 opp
->advertised
= htonl(opp
->advertised
);
2620 opp
->supported
= htonl(opp
->supported
);
2621 opp
->peer
= htonl(opp
->peer
);
2625 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2627 queue_tx(make_echo_reply(oh
), ofconn
, ofconn
->reply_counter
);
2632 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2634 struct ofp_switch_features
*osf
;
2636 struct ofport
*port
;
2638 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2639 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2640 osf
->n_buffers
= htonl(pktbuf_capacity());
2642 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2643 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2644 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2645 (1u << OFPAT_SET_VLAN_VID
) |
2646 (1u << OFPAT_SET_VLAN_PCP
) |
2647 (1u << OFPAT_STRIP_VLAN
) |
2648 (1u << OFPAT_SET_DL_SRC
) |
2649 (1u << OFPAT_SET_DL_DST
) |
2650 (1u << OFPAT_SET_NW_SRC
) |
2651 (1u << OFPAT_SET_NW_DST
) |
2652 (1u << OFPAT_SET_NW_TOS
) |
2653 (1u << OFPAT_SET_TP_SRC
) |
2654 (1u << OFPAT_SET_TP_DST
) |
2655 (1u << OFPAT_ENQUEUE
));
2657 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2658 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2661 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2666 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2669 struct ofp_switch_config
*osc
;
2673 /* Figure out flags. */
2674 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2675 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2678 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2679 osc
->flags
= htons(flags
);
2680 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2681 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2687 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
2689 uint16_t flags
= ntohs(osc
->flags
);
2691 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2692 switch (flags
& OFPC_FRAG_MASK
) {
2693 case OFPC_FRAG_NORMAL
:
2694 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2696 case OFPC_FRAG_DROP
:
2697 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2700 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2706 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2711 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2712 struct action_xlate_ctx
*ctx
);
2715 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2717 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2720 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2721 /* Forwarding disabled on port. */
2726 * We don't have an ofport record for this port, but it doesn't hurt to
2727 * allow forwarding to it anyhow. Maybe such a port will appear later
2728 * and we're pre-populating the flow table.
2732 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_OUTPUT
, port
);
2733 ctx
->nf_output_iface
= port
;
2736 static struct rule
*
2737 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2739 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2743 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2745 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2746 uint16_t old_in_port
;
2749 /* Look up a flow with 'in_port' as the input port. Then restore the
2750 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2751 * have surprising behavior). */
2752 old_in_port
= ctx
->flow
.in_port
;
2753 ctx
->flow
.in_port
= in_port
;
2754 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2755 ctx
->flow
.in_port
= old_in_port
;
2757 if (ctx
->resubmit_hook
) {
2758 ctx
->resubmit_hook(ctx
, rule
);
2763 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2767 static struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2769 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2770 MAX_RESUBMIT_RECURSION
);
2775 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2776 uint16_t *nf_output_iface
, struct ofpbuf
*odp_actions
)
2778 struct ofport
*ofport
;
2780 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2781 uint16_t odp_port
= ofport
->odp_port
;
2782 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2783 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, odp_port
);
2786 *nf_output_iface
= NF_OUT_FLOOD
;
2790 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2791 uint16_t port
, uint16_t max_len
)
2794 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2796 ctx
->nf_output_iface
= NF_OUT_DROP
;
2800 add_output_action(ctx
, ctx
->flow
.in_port
);
2803 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2806 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2807 ctx
->odp_actions
, &ctx
->tags
,
2808 &ctx
->nf_output_iface
,
2809 ctx
->ofproto
->aux
)) {
2810 COVERAGE_INC(ofproto_uninstallable
);
2811 ctx
->may_set_up_flow
= false;
2815 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2816 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2819 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2820 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2822 case OFPP_CONTROLLER
:
2823 nl_msg_put_u64(ctx
->odp_actions
, ODP_ACTION_ATTR_CONTROLLER
, max_len
);
2826 add_output_action(ctx
, ODPP_LOCAL
);
2829 odp_port
= ofp_port_to_odp_port(port
);
2830 if (odp_port
!= ctx
->flow
.in_port
) {
2831 add_output_action(ctx
, odp_port
);
2836 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2837 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2838 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2839 ctx
->nf_output_iface
= prev_nf_output_iface
;
2840 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2841 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2842 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2847 xlate_output_action(struct action_xlate_ctx
*ctx
,
2848 const struct ofp_action_output
*oao
)
2850 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2853 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2854 * optimization, because we're going to add another action that sets the
2855 * priority immediately after, or because there are no actions following the
2858 remove_pop_action(struct action_xlate_ctx
*ctx
)
2860 if (ctx
->odp_actions
->size
== ctx
->last_pop_priority
) {
2861 ctx
->odp_actions
->size
-= NLA_ALIGN(NLA_HDRLEN
);
2862 ctx
->last_pop_priority
= -1;
2867 add_pop_action(struct action_xlate_ctx
*ctx
)
2869 if (ctx
->odp_actions
->size
!= ctx
->last_pop_priority
) {
2870 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_POP_PRIORITY
);
2871 ctx
->last_pop_priority
= ctx
->odp_actions
->size
;
2876 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2877 const struct ofp_action_enqueue
*oae
)
2879 uint16_t ofp_port
, odp_port
;
2883 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2886 /* Fall back to ordinary output action. */
2887 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2891 /* Figure out ODP output port. */
2892 ofp_port
= ntohs(oae
->port
);
2893 if (ofp_port
!= OFPP_IN_PORT
) {
2894 odp_port
= ofp_port_to_odp_port(ofp_port
);
2896 odp_port
= ctx
->flow
.in_port
;
2899 /* Add ODP actions. */
2900 remove_pop_action(ctx
);
2901 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2902 add_output_action(ctx
, odp_port
);
2903 add_pop_action(ctx
);
2905 /* Update NetFlow output port. */
2906 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2907 ctx
->nf_output_iface
= odp_port
;
2908 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2909 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2914 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2915 const struct nx_action_set_queue
*nasq
)
2920 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2923 /* Couldn't translate queue to a priority, so ignore. A warning
2924 * has already been logged. */
2928 remove_pop_action(ctx
);
2929 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2933 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2935 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2936 if (!(tci
& htons(VLAN_CFI
))) {
2937 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_STRIP_VLAN
);
2939 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_TCI
,
2940 tci
& ~htons(VLAN_CFI
));
2944 struct xlate_reg_state
{
2950 save_reg_state(const struct action_xlate_ctx
*ctx
,
2951 struct xlate_reg_state
*state
)
2953 state
->vlan_tci
= ctx
->flow
.vlan_tci
;
2954 state
->tun_id
= ctx
->flow
.tun_id
;
2958 update_reg_state(struct action_xlate_ctx
*ctx
,
2959 const struct xlate_reg_state
*state
)
2961 if (ctx
->flow
.vlan_tci
!= state
->vlan_tci
) {
2962 xlate_set_dl_tci(ctx
);
2964 if (ctx
->flow
.tun_id
!= state
->tun_id
) {
2965 nl_msg_put_be64(ctx
->odp_actions
,
2966 ODP_ACTION_ATTR_SET_TUNNEL
, ctx
->flow
.tun_id
);
2971 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2972 const struct nx_action_header
*nah
)
2974 const struct nx_action_resubmit
*nar
;
2975 const struct nx_action_set_tunnel
*nast
;
2976 const struct nx_action_set_queue
*nasq
;
2977 const struct nx_action_multipath
*nam
;
2978 enum nx_action_subtype subtype
= ntohs(nah
->subtype
);
2979 struct xlate_reg_state state
;
2982 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2984 case NXAST_RESUBMIT
:
2985 nar
= (const struct nx_action_resubmit
*) nah
;
2986 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2989 case NXAST_SET_TUNNEL
:
2990 nast
= (const struct nx_action_set_tunnel
*) nah
;
2991 tun_id
= htonll(ntohl(nast
->tun_id
));
2992 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
2993 ctx
->flow
.tun_id
= tun_id
;
2996 case NXAST_DROP_SPOOFED_ARP
:
2997 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2998 nl_msg_put_flag(ctx
->odp_actions
,
2999 ODP_ACTION_ATTR_DROP_SPOOFED_ARP
);
3003 case NXAST_SET_QUEUE
:
3004 nasq
= (const struct nx_action_set_queue
*) nah
;
3005 xlate_set_queue_action(ctx
, nasq
);
3008 case NXAST_POP_QUEUE
:
3009 add_pop_action(ctx
);
3012 case NXAST_REG_MOVE
:
3013 save_reg_state(ctx
, &state
);
3014 nxm_execute_reg_move((const struct nx_action_reg_move
*) nah
,
3016 update_reg_state(ctx
, &state
);
3019 case NXAST_REG_LOAD
:
3020 save_reg_state(ctx
, &state
);
3021 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
3023 update_reg_state(ctx
, &state
);
3027 /* Nothing to do. */
3030 case NXAST_SET_TUNNEL64
:
3031 tun_id
= ((const struct nx_action_set_tunnel64
*) nah
)->tun_id
;
3032 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
3033 ctx
->flow
.tun_id
= tun_id
;
3036 case NXAST_MULTIPATH
:
3037 nam
= (const struct nx_action_multipath
*) nah
;
3038 multipath_execute(nam
, &ctx
->flow
);
3041 /* If you add a new action here that modifies flow data, don't forget to
3042 * update the flow key in ctx->flow at the same time. */
3044 case NXAST_SNAT__OBSOLETE
:
3046 VLOG_DBG_RL(&rl
, "unknown Nicira action type %d", (int) subtype
);
3052 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
3053 struct action_xlate_ctx
*ctx
)
3055 struct actions_iterator iter
;
3056 const union ofp_action
*ia
;
3057 const struct ofport
*port
;
3059 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
3060 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
3061 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
3062 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
3063 /* Drop this flow. */
3067 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
3068 enum ofp_action_type type
= ntohs(ia
->type
);
3069 const struct ofp_action_dl_addr
*oada
;
3073 xlate_output_action(ctx
, &ia
->output
);
3076 case OFPAT_SET_VLAN_VID
:
3077 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
3078 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
3079 xlate_set_dl_tci(ctx
);
3082 case OFPAT_SET_VLAN_PCP
:
3083 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
3084 ctx
->flow
.vlan_tci
|= htons(
3085 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
3086 xlate_set_dl_tci(ctx
);
3089 case OFPAT_STRIP_VLAN
:
3090 ctx
->flow
.vlan_tci
= htons(0);
3091 xlate_set_dl_tci(ctx
);
3094 case OFPAT_SET_DL_SRC
:
3095 oada
= ((struct ofp_action_dl_addr
*) ia
);
3096 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_SRC
,
3097 oada
->dl_addr
, ETH_ADDR_LEN
);
3098 memcpy(ctx
->flow
.dl_src
, oada
->dl_addr
, ETH_ADDR_LEN
);
3101 case OFPAT_SET_DL_DST
:
3102 oada
= ((struct ofp_action_dl_addr
*) ia
);
3103 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_DST
,
3104 oada
->dl_addr
, ETH_ADDR_LEN
);
3105 memcpy(ctx
->flow
.dl_dst
, oada
->dl_addr
, ETH_ADDR_LEN
);
3108 case OFPAT_SET_NW_SRC
:
3109 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_SRC
,
3110 ia
->nw_addr
.nw_addr
);
3111 ctx
->flow
.nw_src
= ia
->nw_addr
.nw_addr
;
3114 case OFPAT_SET_NW_DST
:
3115 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_DST
,
3116 ia
->nw_addr
.nw_addr
);
3117 ctx
->flow
.nw_dst
= ia
->nw_addr
.nw_addr
;
3120 case OFPAT_SET_NW_TOS
:
3121 nl_msg_put_u8(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_TOS
,
3123 ctx
->flow
.nw_tos
= ia
->nw_tos
.nw_tos
;
3126 case OFPAT_SET_TP_SRC
:
3127 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_SRC
,
3128 ia
->tp_port
.tp_port
);
3129 ctx
->flow
.tp_src
= ia
->tp_port
.tp_port
;
3132 case OFPAT_SET_TP_DST
:
3133 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_DST
,
3134 ia
->tp_port
.tp_port
);
3135 ctx
->flow
.tp_dst
= ia
->tp_port
.tp_port
;
3139 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
3143 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
3147 VLOG_DBG_RL(&rl
, "unknown action type %d", (int) type
);
3154 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
3155 struct ofproto
*ofproto
, const struct flow
*flow
,
3156 const struct ofpbuf
*packet
)
3158 ctx
->ofproto
= ofproto
;
3160 ctx
->packet
= packet
;
3161 ctx
->resubmit_hook
= NULL
;
3162 ctx
->check_special
= true;
3165 static struct ofpbuf
*
3166 xlate_actions(struct action_xlate_ctx
*ctx
,
3167 const union ofp_action
*in
, size_t n_in
)
3169 COVERAGE_INC(ofproto_ofp2odp
);
3171 ctx
->odp_actions
= ofpbuf_new(512);
3173 ctx
->may_set_up_flow
= true;
3174 ctx
->nf_output_iface
= NF_OUT_DROP
;
3176 ctx
->last_pop_priority
= -1;
3178 if (!ctx
->check_special
3179 || !ctx
->ofproto
->ofhooks
->special_cb
3180 || ctx
->ofproto
->ofhooks
->special_cb(&ctx
->flow
, ctx
->packet
,
3181 ctx
->ofproto
->aux
)) {
3182 do_xlate_actions(in
, n_in
, ctx
);
3184 ctx
->may_set_up_flow
= false;
3187 remove_pop_action(ctx
);
3189 /* Check with in-band control to see if we're allowed to set up this
3191 if (!in_band_rule_check(ctx
->ofproto
->in_band
, &ctx
->flow
,
3192 ctx
->odp_actions
->data
, ctx
->odp_actions
->size
)) {
3193 ctx
->may_set_up_flow
= false;
3196 return ctx
->odp_actions
;
3199 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3200 * error message code (composed with ofp_mkerr()) for the caller to propagate
3201 * upward. Otherwise, returns 0.
3203 * The log message mentions 'msg_type'. */
3205 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3207 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3208 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3209 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3212 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3219 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3221 struct ofproto
*p
= ofconn
->ofproto
;
3222 struct ofp_packet_out
*opo
;
3223 struct ofpbuf payload
, *buffer
;
3224 union ofp_action
*ofp_actions
;
3225 struct action_xlate_ctx ctx
;
3226 struct ofpbuf
*odp_actions
;
3227 struct ofpbuf request
;
3229 size_t n_ofp_actions
;
3233 COVERAGE_INC(ofproto_packet_out
);
3235 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3240 /* Get ofp_packet_out. */
3241 ofpbuf_use_const(&request
, oh
, ntohs(oh
->length
));
3242 opo
= ofpbuf_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3245 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3246 &ofp_actions
, &n_ofp_actions
);
3252 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3253 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3255 if (error
|| !buffer
) {
3264 /* Extract flow, check actions. */
3265 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3267 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3273 action_xlate_ctx_init(&ctx
, p
, &flow
, &payload
);
3274 odp_actions
= xlate_actions(&ctx
, ofp_actions
, n_ofp_actions
);
3275 dpif_execute(p
->dpif
, odp_actions
->data
, odp_actions
->size
, &payload
);
3276 ofpbuf_delete(odp_actions
);
3279 ofpbuf_delete(buffer
);
3284 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3285 uint32_t config
, uint32_t mask
)
3287 mask
&= config
^ port
->opp
.config
;
3288 if (mask
& OFPPC_PORT_DOWN
) {
3289 if (config
& OFPPC_PORT_DOWN
) {
3290 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3292 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3295 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3296 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3297 if (mask
& REVALIDATE_BITS
) {
3298 COVERAGE_INC(ofproto_costly_flags
);
3299 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3300 p
->need_revalidate
= true;
3302 #undef REVALIDATE_BITS
3303 if (mask
& OFPPC_NO_PACKET_IN
) {
3304 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3309 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3311 struct ofproto
*p
= ofconn
->ofproto
;
3312 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
3313 struct ofport
*port
;
3316 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3321 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3323 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3324 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3325 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3327 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3328 if (opm
->advertise
) {
3329 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3335 static struct ofpbuf
*
3336 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3338 struct ofp_stats_reply
*osr
;
3341 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3342 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3344 osr
->flags
= htons(0);
3348 static struct ofpbuf
*
3349 start_ofp_stats_reply(const struct ofp_header
*request
, size_t body_len
)
3351 const struct ofp_stats_request
*osr
3352 = (const struct ofp_stats_request
*) request
;
3353 return make_ofp_stats_reply(osr
->header
.xid
, osr
->type
, body_len
);
3357 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3358 struct ofpbuf
**msgp
)
3360 struct ofpbuf
*msg
= *msgp
;
3361 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3362 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3363 struct ofp_stats_reply
*reply
= msg
->data
;
3364 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3365 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3366 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3368 return ofpbuf_put_uninit(*msgp
, nbytes
);
3371 static struct ofpbuf
*
3372 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3374 struct nicira_stats_msg
*nsm
;
3377 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3378 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3379 nsm
->type
= htons(OFPST_VENDOR
);
3380 nsm
->flags
= htons(0);
3381 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3382 nsm
->subtype
= subtype
;
3386 static struct ofpbuf
*
3387 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3389 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3393 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3394 struct ofpbuf
**msgp
)
3396 struct ofpbuf
*msg
= *msgp
;
3397 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3398 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3399 struct nicira_stats_msg
*reply
= msg
->data
;
3400 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3401 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3402 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3404 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3408 handle_desc_stats_request(struct ofconn
*ofconn
,
3409 const struct ofp_header
*request
)
3411 struct ofproto
*p
= ofconn
->ofproto
;
3412 struct ofp_desc_stats
*ods
;
3415 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3416 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3417 memset(ods
, 0, sizeof *ods
);
3418 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3419 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3420 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3421 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3422 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3423 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3429 handle_table_stats_request(struct ofconn
*ofconn
,
3430 const struct ofp_header
*request
)
3432 struct ofproto
*p
= ofconn
->ofproto
;
3433 struct ofp_table_stats
*ots
;
3436 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3438 /* Classifier table. */
3439 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3440 memset(ots
, 0, sizeof *ots
);
3441 strcpy(ots
->name
, "classifier");
3442 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3443 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3444 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3445 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3446 put_32aligned_be64(&ots
->lookup_count
, htonll(0)); /* XXX */
3447 put_32aligned_be64(&ots
->matched_count
, htonll(0)); /* XXX */
3449 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3454 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3455 struct ofpbuf
**msgp
)
3457 struct netdev_stats stats
;
3458 struct ofp_port_stats
*ops
;
3460 /* Intentionally ignore return value, since errors will set
3461 * 'stats' to all-1s, which is correct for OpenFlow, and
3462 * netdev_get_stats() will log errors. */
3463 netdev_get_stats(port
->netdev
, &stats
);
3465 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3466 ops
->port_no
= htons(port
->opp
.port_no
);
3467 memset(ops
->pad
, 0, sizeof ops
->pad
);
3468 put_32aligned_be64(&ops
->rx_packets
, htonll(stats
.rx_packets
));
3469 put_32aligned_be64(&ops
->tx_packets
, htonll(stats
.tx_packets
));
3470 put_32aligned_be64(&ops
->rx_bytes
, htonll(stats
.rx_bytes
));
3471 put_32aligned_be64(&ops
->tx_bytes
, htonll(stats
.tx_bytes
));
3472 put_32aligned_be64(&ops
->rx_dropped
, htonll(stats
.rx_dropped
));
3473 put_32aligned_be64(&ops
->tx_dropped
, htonll(stats
.tx_dropped
));
3474 put_32aligned_be64(&ops
->rx_errors
, htonll(stats
.rx_errors
));
3475 put_32aligned_be64(&ops
->tx_errors
, htonll(stats
.tx_errors
));
3476 put_32aligned_be64(&ops
->rx_frame_err
, htonll(stats
.rx_frame_errors
));
3477 put_32aligned_be64(&ops
->rx_over_err
, htonll(stats
.rx_over_errors
));
3478 put_32aligned_be64(&ops
->rx_crc_err
, htonll(stats
.rx_crc_errors
));
3479 put_32aligned_be64(&ops
->collisions
, htonll(stats
.collisions
));
3483 handle_port_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3485 struct ofproto
*p
= ofconn
->ofproto
;
3486 const struct ofp_port_stats_request
*psr
= ofputil_stats_body(oh
);
3487 struct ofp_port_stats
*ops
;
3489 struct ofport
*port
;
3491 msg
= start_ofp_stats_reply(oh
, sizeof *ops
* 16);
3492 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3493 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3495 append_port_stat(port
, ofconn
, &msg
);
3498 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3499 append_port_stat(port
, ofconn
, &msg
);
3503 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3508 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3510 long long int msecs
= time_msec() - start
;
3511 *sec
= htonl(msecs
/ 1000);
3512 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3516 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3517 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3519 struct ofp_flow_stats
*ofs
;
3520 uint64_t packet_count
, byte_count
;
3522 size_t act_len
, len
;
3524 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3528 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3529 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3531 rule_get_stats(rule
, &packet_count
, &byte_count
);
3533 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3534 ofs
->length
= htons(len
);
3537 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
,
3538 rule
->flow_cookie
, &cookie
);
3539 put_32aligned_be64(&ofs
->cookie
, cookie
);
3540 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3541 ofs
->priority
= htons(rule
->cr
.priority
);
3542 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3543 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3544 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3545 put_32aligned_be64(&ofs
->packet_count
, htonll(packet_count
));
3546 put_32aligned_be64(&ofs
->byte_count
, htonll(byte_count
));
3547 if (rule
->n_actions
> 0) {
3548 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3553 is_valid_table(uint8_t table_id
)
3555 if (table_id
== 0 || table_id
== 0xff) {
3558 /* It would probably be better to reply with an error but there doesn't
3559 * seem to be any appropriate value, so that might just be
3561 VLOG_WARN_RL(&rl
, "controller asked for invalid table %"PRIu8
,
3568 handle_flow_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3570 const struct ofp_flow_stats_request
*fsr
= ofputil_stats_body(oh
);
3571 struct ofpbuf
*reply
;
3573 COVERAGE_INC(ofproto_flows_req
);
3574 reply
= start_ofp_stats_reply(oh
, 1024);
3575 if (is_valid_table(fsr
->table_id
)) {
3576 struct cls_cursor cursor
;
3577 struct cls_rule target
;
3580 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3582 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3583 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3584 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3587 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3593 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3594 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3596 struct nx_flow_stats
*nfs
;
3597 uint64_t packet_count
, byte_count
;
3598 size_t act_len
, start_len
;
3599 struct ofpbuf
*reply
;
3601 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3605 rule_get_stats(rule
, &packet_count
, &byte_count
);
3607 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3609 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3610 start_len
= (*replyp
)->size
;
3613 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3616 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3617 nfs
->cookie
= rule
->flow_cookie
;
3618 nfs
->priority
= htons(rule
->cr
.priority
);
3619 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3620 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3621 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3622 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3623 nfs
->packet_count
= htonll(packet_count
);
3624 nfs
->byte_count
= htonll(byte_count
);
3625 if (rule
->n_actions
> 0) {
3626 ofpbuf_put(reply
, rule
->actions
, act_len
);
3628 nfs
->length
= htons(reply
->size
- start_len
);
3632 handle_nxst_flow(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3634 struct nx_flow_stats_request
*nfsr
;
3635 struct cls_rule target
;
3636 struct ofpbuf
*reply
;
3640 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3642 /* Dissect the message. */
3643 nfsr
= ofpbuf_pull(&b
, sizeof *nfsr
);
3644 error
= nx_pull_match(&b
, ntohs(nfsr
->match_len
), 0, &target
);
3649 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3652 COVERAGE_INC(ofproto_flows_req
);
3653 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3654 if (is_valid_table(nfsr
->table_id
)) {
3655 struct cls_cursor cursor
;
3658 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3659 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3660 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3663 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3669 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
3671 uint64_t packet_count
, byte_count
;
3672 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3674 rule_get_stats(rule
, &packet_count
, &byte_count
);
3676 ds_put_format(results
, "duration=%llds, ",
3677 (time_msec() - rule
->created
) / 1000);
3678 ds_put_format(results
, "idle=%.3fs, ", (time_msec() - rule
->used
) / 1000.0);
3679 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3680 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3681 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3682 cls_rule_format(&rule
->cr
, results
);
3683 ds_put_char(results
, ',');
3685 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3687 ds_put_cstr(results
, "drop");
3689 ds_put_cstr(results
, "\n");
3692 /* Adds a pretty-printed description of all flows to 'results', including
3693 * those marked hidden by secchan (e.g., by in-band control). */
3695 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3697 struct cls_cursor cursor
;
3700 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3701 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3702 flow_stats_ds(rule
, results
);
3707 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3708 ovs_be16 out_port
, uint8_t table_id
,
3709 struct ofp_aggregate_stats_reply
*oasr
)
3711 uint64_t total_packets
= 0;
3712 uint64_t total_bytes
= 0;
3715 COVERAGE_INC(ofproto_agg_request
);
3717 if (is_valid_table(table_id
)) {
3718 struct cls_cursor cursor
;
3721 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3722 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3723 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3724 uint64_t packet_count
;
3725 uint64_t byte_count
;
3727 rule_get_stats(rule
, &packet_count
, &byte_count
);
3729 total_packets
+= packet_count
;
3730 total_bytes
+= byte_count
;
3736 oasr
->flow_count
= htonl(n_flows
);
3737 put_32aligned_be64(&oasr
->packet_count
, htonll(total_packets
));
3738 put_32aligned_be64(&oasr
->byte_count
, htonll(total_bytes
));
3739 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3743 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3744 const struct ofp_header
*oh
)
3746 const struct ofp_aggregate_stats_request
*request
= ofputil_stats_body(oh
);
3747 struct ofp_aggregate_stats_reply
*reply
;
3748 struct cls_rule target
;
3751 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3754 msg
= start_ofp_stats_reply(oh
, sizeof *reply
);
3755 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3756 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3757 request
->table_id
, reply
);
3758 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3763 handle_nxst_aggregate(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3765 struct nx_aggregate_stats_request
*request
;
3766 struct ofp_aggregate_stats_reply
*reply
;
3767 struct cls_rule target
;
3772 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3774 /* Dissect the message. */
3775 request
= ofpbuf_pull(&b
, sizeof *request
);
3776 error
= nx_pull_match(&b
, ntohs(request
->match_len
), 0, &target
);
3781 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3785 COVERAGE_INC(ofproto_flows_req
);
3786 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3787 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3788 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3789 request
->table_id
, reply
);
3790 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3795 struct queue_stats_cbdata
{
3796 struct ofconn
*ofconn
;
3797 struct ofport
*ofport
;
3802 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3803 const struct netdev_queue_stats
*stats
)
3805 struct ofp_queue_stats
*reply
;
3807 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3808 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3809 memset(reply
->pad
, 0, sizeof reply
->pad
);
3810 reply
->queue_id
= htonl(queue_id
);
3811 put_32aligned_be64(&reply
->tx_bytes
, htonll(stats
->tx_bytes
));
3812 put_32aligned_be64(&reply
->tx_packets
, htonll(stats
->tx_packets
));
3813 put_32aligned_be64(&reply
->tx_errors
, htonll(stats
->tx_errors
));
3817 handle_queue_stats_dump_cb(uint32_t queue_id
,
3818 struct netdev_queue_stats
*stats
,
3821 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3823 put_queue_stats(cbdata
, queue_id
, stats
);
3827 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3828 struct queue_stats_cbdata
*cbdata
)
3830 cbdata
->ofport
= port
;
3831 if (queue_id
== OFPQ_ALL
) {
3832 netdev_dump_queue_stats(port
->netdev
,
3833 handle_queue_stats_dump_cb
, cbdata
);
3835 struct netdev_queue_stats stats
;
3837 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3838 put_queue_stats(cbdata
, queue_id
, &stats
);
3844 handle_queue_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3846 struct ofproto
*ofproto
= ofconn
->ofproto
;
3847 const struct ofp_queue_stats_request
*qsr
;
3848 struct queue_stats_cbdata cbdata
;
3849 struct ofport
*port
;
3850 unsigned int port_no
;
3853 qsr
= ofputil_stats_body(oh
);
3855 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3858 COVERAGE_INC(ofproto_queue_req
);
3860 cbdata
.ofconn
= ofconn
;
3861 cbdata
.msg
= start_ofp_stats_reply(oh
, 128);
3863 port_no
= ntohs(qsr
->port_no
);
3864 queue_id
= ntohl(qsr
->queue_id
);
3865 if (port_no
== OFPP_ALL
) {
3866 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3867 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3869 } else if (port_no
< ofproto
->max_ports
) {
3870 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3872 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3875 ofpbuf_delete(cbdata
.msg
);
3876 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3878 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3883 /* Updates 'facet''s used time. Caller is responsible for calling
3884 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3886 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3889 if (used
> facet
->used
) {
3891 if (used
> facet
->rule
->used
) {
3892 facet
->rule
->used
= used
;
3894 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3898 /* Folds the statistics from 'stats' into the counters in 'facet'.
3900 * Because of the meaning of a facet's counters, it only makes sense to do this
3901 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3902 * packet that was sent by hand or if it represents statistics that have been
3903 * cleared out of the datapath. */
3905 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3906 const struct dpif_flow_stats
*stats
)
3908 if (stats
->n_packets
|| stats
->used
> facet
->used
) {
3909 facet_update_time(ofproto
, facet
, stats
->used
);
3910 facet
->packet_count
+= stats
->n_packets
;
3911 facet
->byte_count
+= stats
->n_bytes
;
3912 facet_push_stats(ofproto
, facet
);
3913 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3918 facet_push_stats(struct ofproto
*ofproto
, struct facet
*facet
)
3920 uint64_t rs_packets
, rs_bytes
;
3922 assert(facet
->packet_count
>= facet
->rs_packet_count
);
3923 assert(facet
->byte_count
>= facet
->rs_byte_count
);
3924 assert(facet
->used
>= facet
->rs_used
);
3926 rs_packets
= facet
->packet_count
- facet
->rs_packet_count
;
3927 rs_bytes
= facet
->byte_count
- facet
->rs_byte_count
;
3929 if (rs_packets
|| rs_bytes
|| facet
->used
> facet
->rs_used
) {
3930 facet
->rs_packet_count
= facet
->packet_count
;
3931 facet
->rs_byte_count
= facet
->byte_count
;
3932 facet
->rs_used
= facet
->used
;
3934 flow_push_stats(ofproto
, facet
->rule
, &facet
->flow
,
3935 rs_packets
, rs_bytes
, facet
->used
);
3939 struct ofproto_push
{
3940 struct action_xlate_ctx ctx
;
3947 push_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
3949 struct ofproto_push
*push
= CONTAINER_OF(ctx
, struct ofproto_push
, ctx
);
3952 rule
->packet_count
+= push
->packets
;
3953 rule
->byte_count
+= push
->bytes
;
3954 rule
->used
= MAX(push
->used
, rule
->used
);
3958 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3959 * 'rule''s actions. */
3961 flow_push_stats(struct ofproto
*ofproto
, const struct rule
*rule
,
3962 struct flow
*flow
, uint64_t packets
, uint64_t bytes
,
3965 struct ofproto_push push
;
3967 push
.packets
= packets
;
3971 action_xlate_ctx_init(&push
.ctx
, ofproto
, flow
, NULL
);
3972 push
.ctx
.resubmit_hook
= push_resubmit
;
3973 ofpbuf_delete(xlate_actions(&push
.ctx
, rule
->actions
, rule
->n_actions
));
3976 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3977 * in which no matching flow already exists in the flow table.
3979 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3980 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3981 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3983 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3986 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3988 struct ofproto
*p
= ofconn
->ofproto
;
3989 struct ofpbuf
*packet
;
3994 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3995 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3996 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
4000 if (fm
->buffer_id
!= UINT32_MAX
) {
4001 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
4005 in_port
= UINT16_MAX
;
4008 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
4009 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
4010 fm
->flags
& OFPFF_SEND_FLOW_REM
);
4011 rule_insert(p
, rule
);
4013 rule_execute(p
, rule
, in_port
, packet
);
4018 static struct rule
*
4019 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
4021 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
4025 send_buffered_packet(struct ofconn
*ofconn
,
4026 struct rule
*rule
, uint32_t buffer_id
)
4028 struct ofpbuf
*packet
;
4032 if (buffer_id
== UINT32_MAX
) {
4036 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
4041 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
4046 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4048 struct modify_flows_cbdata
{
4049 struct ofproto
*ofproto
;
4050 const struct flow_mod
*fm
;
4054 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
4057 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4058 * encoded by ofp_mkerr() on failure.
4060 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4063 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4065 struct ofproto
*p
= ofconn
->ofproto
;
4066 struct rule
*match
= NULL
;
4067 struct cls_cursor cursor
;
4070 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4071 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4072 if (!rule_is_hidden(rule
)) {
4074 modify_flow(p
, fm
, rule
);
4079 /* This credits the packet to whichever flow happened to match last.
4080 * That's weird. Maybe we should do a lookup for the flow that
4081 * actually matches the packet? Who knows. */
4082 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
4085 return add_flow(ofconn
, fm
);
4089 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4090 * code as encoded by ofp_mkerr() on failure.
4092 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4095 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4097 struct ofproto
*p
= ofconn
->ofproto
;
4098 struct rule
*rule
= find_flow_strict(p
, fm
);
4099 if (rule
&& !rule_is_hidden(rule
)) {
4100 modify_flow(p
, fm
, rule
);
4101 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
4103 return add_flow(ofconn
, fm
);
4107 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4108 * been identified as a flow in 'p''s flow table to be modified, by changing
4109 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4110 * ofp_action[] structures). */
4112 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
4114 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
4116 rule
->flow_cookie
= fm
->cookie
;
4118 /* If the actions are the same, do nothing. */
4119 if (fm
->n_actions
== rule
->n_actions
4121 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
4125 /* Replace actions. */
4126 free(rule
->actions
);
4127 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
4128 rule
->n_actions
= fm
->n_actions
;
4130 p
->need_revalidate
= true;
4135 /* OFPFC_DELETE implementation. */
4137 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
4139 /* Implements OFPFC_DELETE. */
4141 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
4143 struct rule
*rule
, *next_rule
;
4144 struct cls_cursor cursor
;
4146 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4147 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4148 delete_flow(p
, rule
, htons(fm
->out_port
));
4152 /* Implements OFPFC_DELETE_STRICT. */
4154 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4156 struct rule
*rule
= find_flow_strict(p
, fm
);
4158 delete_flow(p
, rule
, htons(fm
->out_port
));
4162 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4163 * been identified as a flow to delete from 'p''s flow table, by deleting the
4164 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4167 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4168 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4169 * specified 'out_port'. */
4171 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4173 if (rule_is_hidden(rule
)) {
4177 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4181 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4182 rule_remove(p
, rule
);
4186 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4188 struct ofproto
*p
= ofconn
->ofproto
;
4192 error
= reject_slave_controller(ofconn
, "flow_mod");
4197 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn
->flow_format
);
4202 /* We do not support the emergency flow cache. It will hopefully get
4203 * dropped from OpenFlow in the near future. */
4204 if (fm
.flags
& OFPFF_EMERG
) {
4205 /* There isn't a good fit for an error code, so just state that the
4206 * flow table is full. */
4207 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4210 error
= validate_actions(fm
.actions
, fm
.n_actions
,
4211 &fm
.cr
.flow
, p
->max_ports
);
4216 switch (fm
.command
) {
4218 return add_flow(ofconn
, &fm
);
4221 return modify_flows_loose(ofconn
, &fm
);
4223 case OFPFC_MODIFY_STRICT
:
4224 return modify_flow_strict(ofconn
, &fm
);
4227 delete_flows_loose(p
, &fm
);
4230 case OFPFC_DELETE_STRICT
:
4231 delete_flow_strict(p
, &fm
);
4235 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4240 handle_tun_id_from_cookie(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4242 const struct nxt_tun_id_cookie
*msg
4243 = (const struct nxt_tun_id_cookie
*) oh
;
4245 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4250 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4252 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
4253 struct nx_role_request
*reply
;
4257 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4258 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4260 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4263 role
= ntohl(nrr
->role
);
4264 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4265 && role
!= NX_ROLE_SLAVE
) {
4266 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4268 /* There's no good error code for this. */
4269 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4272 if (role
== NX_ROLE_MASTER
) {
4273 struct ofconn
*other
;
4275 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4276 if (other
->role
== NX_ROLE_MASTER
) {
4277 other
->role
= NX_ROLE_SLAVE
;
4281 ofconn
->role
= role
;
4283 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
4284 reply
->role
= htonl(role
);
4285 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4291 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4293 const struct nxt_set_flow_format
*msg
4294 = (const struct nxt_set_flow_format
*) oh
;
4297 format
= ntohl(msg
->format
);
4298 if (format
== NXFF_OPENFLOW10
4299 || format
== NXFF_TUN_ID_FROM_COOKIE
4300 || format
== NXFF_NXM
) {
4301 ofconn
->flow_format
= format
;
4304 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4309 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4311 struct ofp_header
*ob
;
4314 /* Currently, everything executes synchronously, so we can just
4315 * immediately send the barrier reply. */
4316 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4317 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4322 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4324 const struct ofp_header
*oh
= msg
->data
;
4325 const struct ofputil_msg_type
*type
;
4328 error
= ofputil_decode_msg_type(oh
, &type
);
4333 switch (ofputil_msg_type_code(type
)) {
4334 /* OpenFlow requests. */
4335 case OFPUTIL_OFPT_ECHO_REQUEST
:
4336 return handle_echo_request(ofconn
, oh
);
4338 case OFPUTIL_OFPT_FEATURES_REQUEST
:
4339 return handle_features_request(ofconn
, oh
);
4341 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
4342 return handle_get_config_request(ofconn
, oh
);
4344 case OFPUTIL_OFPT_SET_CONFIG
:
4345 return handle_set_config(ofconn
, msg
->data
);
4347 case OFPUTIL_OFPT_PACKET_OUT
:
4348 return handle_packet_out(ofconn
, oh
);
4350 case OFPUTIL_OFPT_PORT_MOD
:
4351 return handle_port_mod(ofconn
, oh
);
4353 case OFPUTIL_OFPT_FLOW_MOD
:
4354 return handle_flow_mod(ofconn
, oh
);
4356 case OFPUTIL_OFPT_BARRIER_REQUEST
:
4357 return handle_barrier_request(ofconn
, oh
);
4359 /* OpenFlow replies. */
4360 case OFPUTIL_OFPT_ECHO_REPLY
:
4363 /* Nicira extension requests. */
4364 case OFPUTIL_NXT_STATUS_REQUEST
:
4365 return switch_status_handle_request(
4366 ofconn
->ofproto
->switch_status
, ofconn
->rconn
, oh
);
4368 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE
:
4369 return handle_tun_id_from_cookie(ofconn
, oh
);
4371 case OFPUTIL_NXT_ROLE_REQUEST
:
4372 return handle_role_request(ofconn
, oh
);
4374 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
4375 return handle_nxt_set_flow_format(ofconn
, oh
);
4377 case OFPUTIL_NXT_FLOW_MOD
:
4378 return handle_flow_mod(ofconn
, oh
);
4380 /* OpenFlow statistics requests. */
4381 case OFPUTIL_OFPST_DESC_REQUEST
:
4382 return handle_desc_stats_request(ofconn
, oh
);
4384 case OFPUTIL_OFPST_FLOW_REQUEST
:
4385 return handle_flow_stats_request(ofconn
, oh
);
4387 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
4388 return handle_aggregate_stats_request(ofconn
, oh
);
4390 case OFPUTIL_OFPST_TABLE_REQUEST
:
4391 return handle_table_stats_request(ofconn
, oh
);
4393 case OFPUTIL_OFPST_PORT_REQUEST
:
4394 return handle_port_stats_request(ofconn
, oh
);
4396 case OFPUTIL_OFPST_QUEUE_REQUEST
:
4397 return handle_queue_stats_request(ofconn
, oh
);
4399 /* Nicira extension statistics requests. */
4400 case OFPUTIL_NXST_FLOW_REQUEST
:
4401 return handle_nxst_flow(ofconn
, oh
);
4403 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
4404 return handle_nxst_aggregate(ofconn
, oh
);
4406 case OFPUTIL_INVALID
:
4407 case OFPUTIL_OFPT_HELLO
:
4408 case OFPUTIL_OFPT_ERROR
:
4409 case OFPUTIL_OFPT_FEATURES_REPLY
:
4410 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
4411 case OFPUTIL_OFPT_PACKET_IN
:
4412 case OFPUTIL_OFPT_FLOW_REMOVED
:
4413 case OFPUTIL_OFPT_PORT_STATUS
:
4414 case OFPUTIL_OFPT_BARRIER_REPLY
:
4415 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
4416 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
4417 case OFPUTIL_OFPST_DESC_REPLY
:
4418 case OFPUTIL_OFPST_FLOW_REPLY
:
4419 case OFPUTIL_OFPST_QUEUE_REPLY
:
4420 case OFPUTIL_OFPST_PORT_REPLY
:
4421 case OFPUTIL_OFPST_TABLE_REPLY
:
4422 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
4423 case OFPUTIL_NXT_STATUS_REPLY
:
4424 case OFPUTIL_NXT_ROLE_REPLY
:
4425 case OFPUTIL_NXT_FLOW_REMOVED
:
4426 case OFPUTIL_NXST_FLOW_REPLY
:
4427 case OFPUTIL_NXST_AGGREGATE_REPLY
:
4429 if (VLOG_IS_WARN_ENABLED()) {
4430 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4431 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4434 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
4435 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
4437 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4443 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4445 int error
= handle_openflow__(ofconn
, ofp_msg
);
4447 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4449 COVERAGE_INC(ofproto_recv_openflow
);
4453 handle_miss_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4455 struct facet
*facet
;
4458 /* Obtain in_port and tun_id, at least. */
4459 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4461 /* Set header pointers in 'flow'. */
4462 flow_extract(upcall
->packet
, flow
.tun_id
, flow
.in_port
, &flow
);
4464 if (p
->ofhooks
->special_cb
4465 && !p
->ofhooks
->special_cb(&flow
, upcall
->packet
, p
->aux
)) {
4466 ofpbuf_delete(upcall
->packet
);
4470 /* Check with in-band control to see if this packet should be sent
4471 * to the local port regardless of the flow table. */
4472 if (in_band_msg_in_hook(p
->in_band
, &flow
, upcall
->packet
)) {
4473 struct ofpbuf odp_actions
;
4475 ofpbuf_init(&odp_actions
, 32);
4476 nl_msg_put_u32(&odp_actions
, ODP_ACTION_ATTR_OUTPUT
, ODPP_LOCAL
);
4477 dpif_execute(p
->dpif
, odp_actions
.data
, odp_actions
.size
,
4479 ofpbuf_uninit(&odp_actions
);
4482 facet
= facet_lookup_valid(p
, &flow
);
4484 struct rule
*rule
= rule_lookup(p
, &flow
);
4486 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4487 struct ofport
*port
= get_port(p
, flow
.in_port
);
4489 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4490 COVERAGE_INC(ofproto_no_packet_in
);
4491 /* XXX install 'drop' flow entry */
4492 ofpbuf_delete(upcall
->packet
);
4496 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4500 COVERAGE_INC(ofproto_packet_in
);
4501 send_packet_in(p
, upcall
, &flow
, false);
4505 facet
= facet_create(p
, rule
, &flow
, upcall
->packet
);
4506 } else if (!facet
->may_install
) {
4507 /* The facet is not installable, that is, we need to process every
4508 * packet, so process the current packet's actions into 'facet'. */
4509 facet_make_actions(p
, facet
, upcall
->packet
);
4512 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4514 * Extra-special case for fail-open mode.
4516 * We are in fail-open mode and the packet matched the fail-open rule,
4517 * but we are connected to a controller too. We should send the packet
4518 * up to the controller in the hope that it will try to set up a flow
4519 * and thereby allow us to exit fail-open.
4521 * See the top-level comment in fail-open.c for more information.
4523 send_packet_in(p
, upcall
, &flow
, true);
4526 facet_execute(p
, facet
, upcall
->packet
);
4527 facet_install(p
, facet
, false);
4531 handle_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4535 switch (upcall
->type
) {
4536 case DPIF_UC_ACTION
:
4537 COVERAGE_INC(ofproto_ctlr_action
);
4538 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4539 send_packet_in(p
, upcall
, &flow
, false);
4542 case DPIF_UC_SAMPLE
:
4544 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4545 ofproto_sflow_received(p
->sflow
, upcall
, &flow
);
4547 ofpbuf_delete(upcall
->packet
);
4551 handle_miss_upcall(p
, upcall
);
4554 case DPIF_N_UC_TYPES
:
4556 VLOG_WARN_RL(&rl
, "upcall has unexpected type %"PRIu32
, upcall
->type
);
4561 /* Flow expiration. */
4563 static int ofproto_dp_max_idle(const struct ofproto
*);
4564 static void ofproto_update_stats(struct ofproto
*);
4565 static void rule_expire(struct ofproto
*, struct rule
*);
4566 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4568 /* This function is called periodically by ofproto_run(). Its job is to
4569 * collect updates for the flows that have been installed into the datapath,
4570 * most importantly when they last were used, and then use that information to
4571 * expire flows that have not been used recently.
4573 * Returns the number of milliseconds after which it should be called again. */
4575 ofproto_expire(struct ofproto
*ofproto
)
4577 struct rule
*rule
, *next_rule
;
4578 struct cls_cursor cursor
;
4581 /* Update stats for each flow in the datapath. */
4582 ofproto_update_stats(ofproto
);
4584 /* Expire facets that have been idle too long. */
4585 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4586 ofproto_expire_facets(ofproto
, dp_max_idle
);
4588 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4589 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4590 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4591 rule_expire(ofproto
, rule
);
4594 /* Let the hook know that we're at a stable point: all outstanding data
4595 * in existing flows has been accounted to the account_cb. Thus, the
4596 * hook can now reasonably do operations that depend on having accurate
4597 * flow volume accounting (currently, that's just bond rebalancing). */
4598 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4599 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4602 return MIN(dp_max_idle
, 1000);
4605 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4607 * This function also pushes statistics updates to rules which each facet
4608 * resubmits into. Generally these statistics will be accurate. However, if a
4609 * facet changes the rule it resubmits into at some time in between
4610 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4611 * old rule will be incorrectly attributed to the new rule. This could be
4612 * avoided by calling ofproto_update_stats() whenever rules are created or
4613 * deleted. However, the performance impact of making so many calls to the
4614 * datapath do not justify the benefit of having perfectly accurate statistics.
4617 ofproto_update_stats(struct ofproto
*p
)
4619 const struct dpif_flow_stats
*stats
;
4620 struct dpif_flow_dump dump
;
4621 const struct nlattr
*key
;
4624 dpif_flow_dump_start(&dump
, p
->dpif
);
4625 while (dpif_flow_dump_next(&dump
, &key
, &key_len
, NULL
, NULL
, &stats
)) {
4626 struct facet
*facet
;
4629 if (odp_flow_key_to_flow(key
, key_len
, &flow
)) {
4633 odp_flow_key_format(key
, key_len
, &s
);
4634 VLOG_WARN_RL(&rl
, "failed to convert ODP flow key to flow: %s",
4640 facet
= facet_find(p
, &flow
);
4642 if (facet
&& facet
->installed
) {
4644 if (stats
->n_packets
>= facet
->dp_packet_count
) {
4645 facet
->packet_count
+= stats
->n_packets
- facet
->dp_packet_count
;
4647 VLOG_WARN_RL(&rl
, "unexpected packet count from the datapath");
4650 if (stats
->n_bytes
>= facet
->dp_byte_count
) {
4651 facet
->byte_count
+= stats
->n_bytes
- facet
->dp_byte_count
;
4653 VLOG_WARN_RL(&rl
, "unexpected byte count from datapath");
4656 facet
->dp_packet_count
= stats
->n_packets
;
4657 facet
->dp_byte_count
= stats
->n_bytes
;
4659 facet_update_time(p
, facet
, stats
->used
);
4660 facet_account(p
, facet
, stats
->n_bytes
);
4661 facet_push_stats(p
, facet
);
4663 /* There's a flow in the datapath that we know nothing about.
4665 COVERAGE_INC(ofproto_unexpected_rule
);
4666 dpif_flow_del(p
->dpif
, key
, key_len
, NULL
);
4669 dpif_flow_dump_done(&dump
);
4672 /* Calculates and returns the number of milliseconds of idle time after which
4673 * facets should expire from the datapath and we should fold their statistics
4674 * into their parent rules in userspace. */
4676 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4679 * Idle time histogram.
4681 * Most of the time a switch has a relatively small number of facets. When
4682 * this is the case we might as well keep statistics for all of them in
4683 * userspace and to cache them in the kernel datapath for performance as
4686 * As the number of facets increases, the memory required to maintain
4687 * statistics about them in userspace and in the kernel becomes
4688 * significant. However, with a large number of facets it is likely that
4689 * only a few of them are "heavy hitters" that consume a large amount of
4690 * bandwidth. At this point, only heavy hitters are worth caching in the
4691 * kernel and maintaining in userspaces; other facets we can discard.
4693 * The technique used to compute the idle time is to build a histogram with
4694 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4695 * that is installed in the kernel gets dropped in the appropriate bucket.
4696 * After the histogram has been built, we compute the cutoff so that only
4697 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4698 * cached. At least the most-recently-used bucket of facets is kept, so
4699 * actually an arbitrary number of facets can be kept in any given
4700 * expiration run (though the next run will delete most of those unless
4701 * they receive additional data).
4703 * This requires a second pass through the facets, in addition to the pass
4704 * made by ofproto_update_stats(), because the former function never looks
4705 * at uninstallable facets.
4707 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4708 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4709 int buckets
[N_BUCKETS
] = { 0 };
4710 struct facet
*facet
;
4715 total
= hmap_count(&ofproto
->facets
);
4716 if (total
<= 1000) {
4717 return N_BUCKETS
* BUCKET_WIDTH
;
4720 /* Build histogram. */
4722 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4723 long long int idle
= now
- facet
->used
;
4724 int bucket
= (idle
<= 0 ? 0
4725 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4726 : (unsigned int) idle
/ BUCKET_WIDTH
);
4730 /* Find the first bucket whose flows should be expired. */
4731 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4732 if (buckets
[bucket
]) {
4735 subtotal
+= buckets
[bucket
++];
4736 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4741 if (VLOG_IS_DBG_ENABLED()) {
4745 ds_put_cstr(&s
, "keep");
4746 for (i
= 0; i
< N_BUCKETS
; i
++) {
4748 ds_put_cstr(&s
, ", drop");
4751 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4754 VLOG_INFO("%s: %s (msec:count)",
4755 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4759 return bucket
* BUCKET_WIDTH
;
4763 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4765 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4766 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4767 struct ofexpired expired
;
4769 if (facet
->installed
) {
4770 struct dpif_flow_stats stats
;
4772 facet_put__(ofproto
, facet
, facet
->actions
, facet
->actions_len
,
4774 facet_update_stats(ofproto
, facet
, &stats
);
4777 expired
.flow
= facet
->flow
;
4778 expired
.packet_count
= facet
->packet_count
;
4779 expired
.byte_count
= facet
->byte_count
;
4780 expired
.used
= facet
->used
;
4781 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4786 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4788 long long int cutoff
= time_msec() - dp_max_idle
;
4789 struct facet
*facet
, *next_facet
;
4791 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4792 facet_active_timeout(ofproto
, facet
);
4793 if (facet
->used
< cutoff
) {
4794 facet_remove(ofproto
, facet
);
4799 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4800 * then delete it entirely. */
4802 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4804 struct facet
*facet
, *next_facet
;
4808 /* Has 'rule' expired? */
4810 if (rule
->hard_timeout
4811 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4812 reason
= OFPRR_HARD_TIMEOUT
;
4813 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4814 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4815 reason
= OFPRR_IDLE_TIMEOUT
;
4820 COVERAGE_INC(ofproto_expired
);
4822 /* Update stats. (This is a no-op if the rule expired due to an idle
4823 * timeout, because that only happens when the rule has no facets left.) */
4824 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4825 facet_remove(ofproto
, facet
);
4828 /* Get rid of the rule. */
4829 if (!rule_is_hidden(rule
)) {
4830 rule_send_removed(ofproto
, rule
, reason
);
4832 rule_remove(ofproto
, rule
);
4835 static struct ofpbuf
*
4836 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4839 struct ofp_flow_removed
*ofr
;
4842 ofr
= make_openflow_xid(sizeof *ofr
, OFPT_FLOW_REMOVED
, htonl(0), &buf
);
4843 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofr
->match
,
4844 rule
->flow_cookie
, &ofr
->cookie
);
4845 ofr
->priority
= htons(rule
->cr
.priority
);
4846 ofr
->reason
= reason
;
4847 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4848 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4849 ofr
->packet_count
= htonll(rule
->packet_count
);
4850 ofr
->byte_count
= htonll(rule
->byte_count
);
4855 static struct ofpbuf
*
4856 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4858 struct nx_flow_removed
*nfr
;
4862 make_nxmsg_xid(sizeof *nfr
, NXT_FLOW_REMOVED
, htonl(0), &buf
);
4863 match_len
= nx_put_match(buf
, &rule
->cr
);
4866 nfr
->cookie
= rule
->flow_cookie
;
4867 nfr
->priority
= htons(rule
->cr
.priority
);
4868 nfr
->reason
= reason
;
4869 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4870 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4871 nfr
->match_len
= htons(match_len
);
4872 nfr
->packet_count
= htonll(rule
->packet_count
);
4873 nfr
->byte_count
= htonll(rule
->byte_count
);
4879 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4881 struct ofconn
*ofconn
;
4883 if (!rule
->send_flow_removed
) {
4887 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4890 if (!rconn_is_connected(ofconn
->rconn
)
4891 || !ofconn_receives_async_msgs(ofconn
)) {
4895 msg
= (ofconn
->flow_format
== NXFF_NXM
4896 ? compose_nx_flow_removed(rule
, reason
)
4897 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4899 /* Account flow expirations under ofconn->reply_counter, the counter
4900 * for replies to OpenFlow requests. That works because preventing
4901 * OpenFlow requests from being processed also prevents new flows from
4902 * being added (and expiring). (It also prevents processing OpenFlow
4903 * requests that would not add new flows, so it is imperfect.) */
4904 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4908 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4909 * The returned statistics include statistics for all of 'rule''s facets. */
4911 rule_get_stats(const struct rule
*rule
, uint64_t *packets
, uint64_t *bytes
)
4914 struct facet
*facet
;
4916 /* Start from historical data for 'rule' itself that are no longer tracked
4917 * in facets. This counts, for example, facets that have expired. */
4918 p
= rule
->packet_count
;
4919 b
= rule
->byte_count
;
4921 /* Add any statistics that are tracked by facets. This includes
4922 * statistical data recently updated by ofproto_update_stats() as well as
4923 * stats for packets that were executed "by hand" via dpif_execute(). */
4924 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
4925 p
+= facet
->packet_count
;
4926 b
+= facet
->byte_count
;
4933 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4935 do_send_packet_in(struct ofpbuf
*ofp_packet_in
, void *ofconn_
)
4937 struct ofconn
*ofconn
= ofconn_
;
4939 rconn_send_with_limit(ofconn
->rconn
, ofp_packet_in
,
4940 ofconn
->packet_in_counter
, 100);
4943 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4944 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4945 * scheduler for sending.
4947 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4948 * Otherwise, ownership is transferred to this function. */
4950 schedule_packet_in(struct ofconn
*ofconn
, struct dpif_upcall
*upcall
,
4951 const struct flow
*flow
, bool clone
)
4953 enum { OPI_SIZE
= offsetof(struct ofp_packet_in
, data
) };
4954 struct ofproto
*ofproto
= ofconn
->ofproto
;
4955 struct ofp_packet_in
*opi
;
4956 int total_len
, send_len
;
4957 struct ofpbuf
*packet
;
4961 /* Get OpenFlow buffer_id. */
4962 if (upcall
->type
== DPIF_UC_ACTION
) {
4963 buffer_id
= UINT32_MAX
;
4964 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4965 buffer_id
= pktbuf_get_null();
4966 } else if (!ofconn
->pktbuf
) {
4967 buffer_id
= UINT32_MAX
;
4969 buffer_id
= pktbuf_save(ofconn
->pktbuf
, upcall
->packet
, flow
->in_port
);
4972 /* Figure out how much of the packet to send. */
4973 total_len
= send_len
= upcall
->packet
->size
;
4974 if (buffer_id
!= UINT32_MAX
) {
4975 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4977 if (upcall
->type
== DPIF_UC_ACTION
) {
4978 send_len
= MIN(send_len
, upcall
->userdata
);
4981 /* Copy or steal buffer for OFPT_PACKET_IN. */
4983 packet
= ofpbuf_clone_data_with_headroom(upcall
->packet
->data
,
4984 send_len
, OPI_SIZE
);
4986 packet
= upcall
->packet
;
4987 packet
->size
= send_len
;
4990 /* Add OFPT_PACKET_IN. */
4991 opi
= ofpbuf_push_zeros(packet
, OPI_SIZE
);
4992 opi
->header
.version
= OFP_VERSION
;
4993 opi
->header
.type
= OFPT_PACKET_IN
;
4994 opi
->total_len
= htons(total_len
);
4995 opi
->in_port
= htons(odp_port_to_ofp_port(flow
->in_port
));
4996 opi
->reason
= upcall
->type
== DPIF_UC_MISS
? OFPR_NO_MATCH
: OFPR_ACTION
;
4997 opi
->buffer_id
= htonl(buffer_id
);
4998 update_openflow_length(packet
);
5000 /* Hand over to packet scheduler. It might immediately call into
5001 * do_send_packet_in() or it might buffer it for a while (until a later
5002 * call to pinsched_run()). */
5003 idx
= upcall
->type
== DPIF_UC_MISS
? 0 : 1;
5004 pinsched_send(ofconn
->schedulers
[idx
], flow
->in_port
,
5005 packet
, do_send_packet_in
, ofconn
);
5008 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
5009 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
5010 * their individual configurations.
5012 * Takes ownership of 'packet'. */
5014 send_packet_in(struct ofproto
*ofproto
, struct dpif_upcall
*upcall
,
5015 const struct flow
*flow
, bool clone
)
5017 struct ofconn
*ofconn
, *prev
;
5020 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
5021 if (ofconn_receives_async_msgs(ofconn
)) {
5023 schedule_packet_in(prev
, upcall
, flow
, true);
5029 schedule_packet_in(prev
, upcall
, flow
, clone
);
5030 } else if (!clone
) {
5031 ofpbuf_delete(upcall
->packet
);
5036 pick_datapath_id(const struct ofproto
*ofproto
)
5038 const struct ofport
*port
;
5040 port
= get_port(ofproto
, ODPP_LOCAL
);
5042 uint8_t ea
[ETH_ADDR_LEN
];
5045 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5047 return eth_addr_to_uint64(ea
);
5049 VLOG_WARN("could not get MAC address for %s (%s)",
5050 netdev_get_name(port
->netdev
), strerror(error
));
5052 return ofproto
->fallback_dpid
;
5056 pick_fallback_dpid(void)
5058 uint8_t ea
[ETH_ADDR_LEN
];
5059 eth_addr_nicira_random(ea
);
5060 return eth_addr_to_uint64(ea
);
5064 ofproto_unixctl_list(struct unixctl_conn
*conn
, const char *arg OVS_UNUSED
,
5065 void *aux OVS_UNUSED
)
5067 const struct shash_node
*node
;
5071 SHASH_FOR_EACH (node
, &all_ofprotos
) {
5072 ds_put_format(&results
, "%s\n", node
->name
);
5074 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
5075 ds_destroy(&results
);
5078 struct ofproto_trace
{
5079 struct action_xlate_ctx ctx
;
5085 trace_format_rule(struct ds
*result
, int level
, const struct rule
*rule
)
5087 ds_put_char_multiple(result
, '\t', level
);
5089 ds_put_cstr(result
, "No match\n");
5093 ds_put_format(result
, "Rule: cookie=%#"PRIx64
" ",
5094 ntohll(rule
->flow_cookie
));
5095 cls_rule_format(&rule
->cr
, result
);
5096 ds_put_char(result
, '\n');
5098 ds_put_char_multiple(result
, '\t', level
);
5099 ds_put_cstr(result
, "OpenFlow ");
5100 ofp_print_actions(result
, (const struct ofp_action_header
*) rule
->actions
,
5101 rule
->n_actions
* sizeof *rule
->actions
);
5102 ds_put_char(result
, '\n');
5106 trace_format_flow(struct ds
*result
, int level
, const char *title
,
5107 struct ofproto_trace
*trace
)
5109 ds_put_char_multiple(result
, '\t', level
);
5110 ds_put_format(result
, "%s: ", title
);
5111 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
5112 ds_put_cstr(result
, "unchanged");
5114 flow_format(result
, &trace
->ctx
.flow
);
5115 trace
->flow
= trace
->ctx
.flow
;
5117 ds_put_char(result
, '\n');
5121 trace_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
5123 struct ofproto_trace
*trace
= CONTAINER_OF(ctx
, struct ofproto_trace
, ctx
);
5124 struct ds
*result
= trace
->result
;
5126 ds_put_char(result
, '\n');
5127 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
5128 trace_format_rule(result
, ctx
->recurse
+ 1, rule
);
5132 ofproto_unixctl_trace(struct unixctl_conn
*conn
, const char *args_
,
5133 void *aux OVS_UNUSED
)
5135 char *dpname
, *in_port_s
, *tun_id_s
, *packet_s
;
5136 char *args
= xstrdup(args_
);
5137 char *save_ptr
= NULL
;
5138 struct ofproto
*ofproto
;
5139 struct ofpbuf packet
;
5147 ofpbuf_init(&packet
, strlen(args
) / 2);
5150 dpname
= strtok_r(args
, " ", &save_ptr
);
5151 tun_id_s
= strtok_r(NULL
, " ", &save_ptr
);
5152 in_port_s
= strtok_r(NULL
, " ", &save_ptr
);
5153 packet_s
= strtok_r(NULL
, "", &save_ptr
); /* Get entire rest of line. */
5154 if (!dpname
|| !in_port_s
|| !packet_s
) {
5155 unixctl_command_reply(conn
, 501, "Bad command syntax");
5159 ofproto
= shash_find_data(&all_ofprotos
, dpname
);
5161 unixctl_command_reply(conn
, 501, "Unknown ofproto (use ofproto/list "
5166 tun_id
= htonll(strtoull(tun_id_s
, NULL
, 10));
5167 in_port
= ofp_port_to_odp_port(atoi(in_port_s
));
5169 packet_s
= ofpbuf_put_hex(&packet
, packet_s
, NULL
);
5170 packet_s
+= strspn(packet_s
, " ");
5171 if (*packet_s
!= '\0') {
5172 unixctl_command_reply(conn
, 501, "Trailing garbage in command");
5175 if (packet
.size
< ETH_HEADER_LEN
) {
5176 unixctl_command_reply(conn
, 501, "Packet data too short for Ethernet");
5180 ds_put_cstr(&result
, "Packet: ");
5181 s
= ofp_packet_to_string(packet
.data
, packet
.size
, packet
.size
);
5182 ds_put_cstr(&result
, s
);
5185 flow_extract(&packet
, tun_id
, in_port
, &flow
);
5186 ds_put_cstr(&result
, "Flow: ");
5187 flow_format(&result
, &flow
);
5188 ds_put_char(&result
, '\n');
5190 rule
= rule_lookup(ofproto
, &flow
);
5191 trace_format_rule(&result
, 0, rule
);
5193 struct ofproto_trace trace
;
5194 struct ofpbuf
*odp_actions
;
5196 trace
.result
= &result
;
5198 action_xlate_ctx_init(&trace
.ctx
, ofproto
, &flow
, &packet
);
5199 trace
.ctx
.resubmit_hook
= trace_resubmit
;
5200 odp_actions
= xlate_actions(&trace
.ctx
,
5201 rule
->actions
, rule
->n_actions
);
5203 ds_put_char(&result
, '\n');
5204 trace_format_flow(&result
, 0, "Final flow", &trace
);
5205 ds_put_cstr(&result
, "Datapath actions: ");
5206 format_odp_actions(&result
, odp_actions
->data
, odp_actions
->size
);
5207 ofpbuf_delete(odp_actions
);
5210 unixctl_command_reply(conn
, 200, ds_cstr(&result
));
5213 ds_destroy(&result
);
5214 ofpbuf_uninit(&packet
);
5219 ofproto_unixctl_init(void)
5221 static bool registered
;
5227 unixctl_command_register("ofproto/list", ofproto_unixctl_list
, NULL
);
5228 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace
, NULL
);
5232 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5233 struct ofpbuf
*odp_actions
, tag_type
*tags
,
5234 uint16_t *nf_output_iface
, void *ofproto_
)
5236 struct ofproto
*ofproto
= ofproto_
;
5239 /* Drop frames for reserved multicast addresses. */
5240 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5244 /* Learn source MAC (but don't try to learn from revalidation). */
5245 if (packet
!= NULL
) {
5246 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
5248 GRAT_ARP_LOCK_NONE
);
5250 /* The log messages here could actually be useful in debugging,
5251 * so keep the rate limit relatively high. */
5252 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5253 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5254 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5255 ofproto_revalidate(ofproto
, rev_tag
);
5259 /* Determine output port. */
5260 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
5263 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5264 nf_output_iface
, odp_actions
);
5265 } else if (out_port
!= flow
->in_port
) {
5266 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, out_port
);
5267 *nf_output_iface
= out_port
;
5275 static const struct ofhooks default_ofhooks
= {
5276 default_normal_ofhook_cb
,