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 const int last_error
= rconn_get_last_error(rconn
);
1380 struct ofproto_controller_info
*cinfo
= xmalloc(sizeof *cinfo
);
1382 shash_add(info
, rconn_get_target(rconn
), cinfo
);
1384 cinfo
->is_connected
= rconn_is_connected(rconn
);
1385 cinfo
->role
= ofconn
->role
;
1390 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "last_error";
1391 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1392 xstrdup(ovs_retval_to_string(last_error
));
1395 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "state";
1396 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1397 xstrdup(rconn_get_state(rconn
));
1399 cinfo
->pairs
.keys
[cinfo
->pairs
.n
] = "time_in_state";
1400 cinfo
->pairs
.values
[cinfo
->pairs
.n
++] =
1401 xasprintf("%u", rconn_get_state_elapsed(rconn
));
1406 ofproto_free_ofproto_controller_info(struct shash
*info
)
1408 struct shash_node
*node
;
1410 SHASH_FOR_EACH (node
, info
) {
1411 struct ofproto_controller_info
*cinfo
= node
->data
;
1412 while (cinfo
->pairs
.n
) {
1413 free((char *) cinfo
->pairs
.values
[--cinfo
->pairs
.n
]);
1417 shash_destroy(info
);
1420 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1422 * This is almost the same as calling dpif_port_del() directly on the
1423 * datapath, but it also makes 'ofproto' close its open netdev for the port
1424 * (if any). This makes it possible to create a new netdev of a different
1425 * type under the same name, which otherwise the netdev library would refuse
1426 * to do because of the conflict. (The netdev would eventually get closed on
1427 * the next trip through ofproto_run(), but this interface is more direct.)
1429 * Returns 0 if successful, otherwise a positive errno. */
1431 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1433 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1434 const char *name
= ofport
? ofport
->opp
.name
: "<unknown>";
1437 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1439 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1440 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1441 } else if (ofport
) {
1442 /* 'name' is ofport->opp.name and update_port() is going to destroy
1443 * 'ofport'. Just in case update_port() refers to 'name' after it
1444 * destroys 'ofport', make a copy of it around the update_port()
1446 char *devname
= xstrdup(name
);
1447 update_port(ofproto
, devname
);
1453 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1454 * true if 'odp_port' exists and should be included, false otherwise. */
1456 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1458 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1459 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1463 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1464 const union ofp_action
*actions
, size_t n_actions
,
1465 const struct ofpbuf
*packet
)
1467 struct action_xlate_ctx ctx
;
1468 struct ofpbuf
*odp_actions
;
1470 action_xlate_ctx_init(&ctx
, p
, flow
, packet
);
1471 /* Always xlate packets originated in this function. */
1472 ctx
.check_special
= false;
1473 odp_actions
= xlate_actions(&ctx
, actions
, n_actions
);
1475 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1477 dpif_execute(p
->dpif
, odp_actions
->data
, odp_actions
->size
, packet
);
1479 ofpbuf_delete(odp_actions
);
1484 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1485 * performs the 'n_actions' actions in 'actions'. The new flow will not
1488 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1489 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1490 * controllers; otherwise, it will be hidden.
1492 * The caller retains ownership of 'cls_rule' and 'actions'. */
1494 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1495 const union ofp_action
*actions
, size_t n_actions
)
1498 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1499 rule_insert(p
, rule
);
1503 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1507 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1510 rule_remove(ofproto
, rule
);
1515 ofproto_flush_flows(struct ofproto
*ofproto
)
1517 struct facet
*facet
, *next_facet
;
1518 struct rule
*rule
, *next_rule
;
1519 struct cls_cursor cursor
;
1521 COVERAGE_INC(ofproto_flush
);
1523 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1524 /* Mark the facet as not installed so that facet_remove() doesn't
1525 * bother trying to uninstall it. There is no point in uninstalling it
1526 * individually since we are about to blow away all the facets with
1527 * dpif_flow_flush(). */
1528 facet
->installed
= false;
1529 facet_remove(ofproto
, facet
);
1532 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1533 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1534 rule_remove(ofproto
, rule
);
1537 dpif_flow_flush(ofproto
->dpif
);
1538 if (ofproto
->in_band
) {
1539 in_band_flushed(ofproto
->in_band
);
1541 if (ofproto
->fail_open
) {
1542 fail_open_flushed(ofproto
->fail_open
);
1547 reinit_ports(struct ofproto
*p
)
1549 struct dpif_port_dump dump
;
1550 struct shash_node
*node
;
1551 struct shash devnames
;
1552 struct ofport
*ofport
;
1553 struct dpif_port dpif_port
;
1555 COVERAGE_INC(ofproto_reinit_ports
);
1557 shash_init(&devnames
);
1558 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1559 shash_add_once (&devnames
, ofport
->opp
.name
, NULL
);
1561 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1562 shash_add_once (&devnames
, dpif_port
.name
, NULL
);
1565 SHASH_FOR_EACH (node
, &devnames
) {
1566 update_port(p
, node
->name
);
1568 shash_destroy(&devnames
);
1571 static struct ofport
*
1572 make_ofport(const struct dpif_port
*dpif_port
)
1574 struct netdev_options netdev_options
;
1575 enum netdev_flags flags
;
1576 struct ofport
*ofport
;
1577 struct netdev
*netdev
;
1580 memset(&netdev_options
, 0, sizeof netdev_options
);
1581 netdev_options
.name
= dpif_port
->name
;
1582 netdev_options
.type
= dpif_port
->type
;
1583 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1585 error
= netdev_open(&netdev_options
, &netdev
);
1587 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1588 "cannot be opened (%s)",
1589 dpif_port
->name
, dpif_port
->port_no
,
1590 dpif_port
->name
, strerror(error
));
1594 ofport
= xzalloc(sizeof *ofport
);
1595 ofport
->netdev
= netdev
;
1596 ofport
->odp_port
= dpif_port
->port_no
;
1597 ofport
->opp
.port_no
= odp_port_to_ofp_port(dpif_port
->port_no
);
1598 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1599 ovs_strlcpy(ofport
->opp
.name
, dpif_port
->name
, sizeof ofport
->opp
.name
);
1601 netdev_get_flags(netdev
, &flags
);
1602 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1604 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1606 netdev_get_features(netdev
,
1607 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1608 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1613 ofport_conflicts(const struct ofproto
*p
, const struct dpif_port
*dpif_port
)
1615 if (get_port(p
, dpif_port
->port_no
)) {
1616 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1617 dpif_port
->port_no
);
1619 } else if (shash_find(&p
->port_by_name
, dpif_port
->name
)) {
1620 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1629 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1631 const struct ofp_phy_port
*a
= &a_
->opp
;
1632 const struct ofp_phy_port
*b
= &b_
->opp
;
1634 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1635 return (a
->port_no
== b
->port_no
1636 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1637 && !strcmp(a
->name
, b
->name
)
1638 && a
->state
== b
->state
1639 && a
->config
== b
->config
1640 && a
->curr
== b
->curr
1641 && a
->advertised
== b
->advertised
1642 && a
->supported
== b
->supported
1643 && a
->peer
== b
->peer
);
1647 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1650 /* XXX Should limit the number of queued port status change messages. */
1651 struct ofconn
*ofconn
;
1652 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1653 struct ofp_port_status
*ops
;
1656 /* Primary controllers, even slaves, should always get port status
1657 updates. Otherwise obey ofconn_receives_async_msgs(). */
1658 if (ofconn
->type
!= OFCONN_PRIMARY
1659 && !ofconn_receives_async_msgs(ofconn
)) {
1663 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1664 ops
->reason
= reason
;
1665 ops
->desc
= ofport
->opp
;
1666 hton_ofp_phy_port(&ops
->desc
);
1667 queue_tx(b
, ofconn
, NULL
);
1672 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1674 const char *netdev_name
= ofport
->opp
.name
;
1676 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1677 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1678 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1680 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1685 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1687 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1688 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1689 shash_delete(&p
->port_by_name
,
1690 shash_find(&p
->port_by_name
, ofport
->opp
.name
));
1692 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1697 ofport_free(struct ofport
*ofport
)
1700 netdev_close(ofport
->netdev
);
1705 static struct ofport
*
1706 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1708 struct ofport
*port
;
1710 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1711 hash_int(odp_port
, 0), &ofproto
->ports
) {
1712 if (port
->odp_port
== odp_port
) {
1720 update_port(struct ofproto
*p
, const char *devname
)
1722 struct dpif_port dpif_port
;
1723 struct ofport
*old_ofport
;
1724 struct ofport
*new_ofport
;
1727 COVERAGE_INC(ofproto_update_port
);
1729 /* Query the datapath for port information. */
1730 error
= dpif_port_query_by_name(p
->dpif
, devname
, &dpif_port
);
1732 /* Find the old ofport. */
1733 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1736 /* There's no port named 'devname' but there might be a port with
1737 * the same port number. This could happen if a port is deleted
1738 * and then a new one added in its place very quickly, or if a port
1739 * is renamed. In the former case we want to send an OFPPR_DELETE
1740 * and an OFPPR_ADD, and in the latter case we want to send a
1741 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1742 * the old port's ifindex against the new port, or perhaps less
1743 * reliably but more portably by comparing the old port's MAC
1744 * against the new port's MAC. However, this code isn't that smart
1745 * and always sends an OFPPR_MODIFY (XXX). */
1746 old_ofport
= get_port(p
, dpif_port
.port_no
);
1748 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1749 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1750 "%s", strerror(error
));
1754 /* Create a new ofport. */
1755 new_ofport
= !error
? make_ofport(&dpif_port
) : NULL
;
1757 /* Eliminate a few pathological cases. */
1758 if (!old_ofport
&& !new_ofport
) {
1760 } else if (old_ofport
&& new_ofport
) {
1761 /* Most of the 'config' bits are OpenFlow soft state, but
1762 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1763 * OpenFlow bits from old_ofport. (make_ofport() only sets
1764 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1765 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1767 if (ofport_equal(old_ofport
, new_ofport
)) {
1768 /* False alarm--no change. */
1769 ofport_free(new_ofport
);
1774 /* Now deal with the normal cases. */
1776 ofport_remove(p
, old_ofport
);
1779 ofport_install(p
, new_ofport
);
1781 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1782 (!old_ofport
? OFPPR_ADD
1783 : !new_ofport
? OFPPR_DELETE
1785 ofport_free(old_ofport
);
1788 dpif_port_destroy(&dpif_port
);
1792 init_ports(struct ofproto
*p
)
1794 struct dpif_port_dump dump
;
1795 struct dpif_port dpif_port
;
1797 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, p
->dpif
) {
1798 if (!ofport_conflicts(p
, &dpif_port
)) {
1799 struct ofport
*ofport
= make_ofport(&dpif_port
);
1801 ofport_install(p
, ofport
);
1809 static struct ofconn
*
1810 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1812 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1813 ofconn
->ofproto
= p
;
1814 list_push_back(&p
->all_conns
, &ofconn
->node
);
1815 ofconn
->rconn
= rconn
;
1816 ofconn
->type
= type
;
1817 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1818 ofconn
->role
= NX_ROLE_OTHER
;
1819 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1820 ofconn
->pktbuf
= NULL
;
1821 ofconn
->miss_send_len
= 0;
1822 ofconn
->reply_counter
= rconn_packet_counter_create ();
1827 ofconn_destroy(struct ofconn
*ofconn
)
1829 if (ofconn
->type
== OFCONN_PRIMARY
) {
1830 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1832 discovery_destroy(ofconn
->discovery
);
1834 list_remove(&ofconn
->node
);
1835 switch_status_unregister(ofconn
->ss
);
1836 rconn_destroy(ofconn
->rconn
);
1837 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1838 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1839 pktbuf_destroy(ofconn
->pktbuf
);
1844 ofconn_run(struct ofconn
*ofconn
)
1846 struct ofproto
*p
= ofconn
->ofproto
;
1850 if (ofconn
->discovery
) {
1851 char *controller_name
;
1852 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1853 discovery_question_connectivity(ofconn
->discovery
);
1855 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1856 if (controller_name
) {
1857 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1858 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1860 free(controller_name
);
1862 rconn_disconnect(ofconn
->rconn
);
1867 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1868 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1871 rconn_run(ofconn
->rconn
);
1873 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1874 /* Limit the number of iterations to prevent other tasks from
1876 for (iteration
= 0; iteration
< 50; iteration
++) {
1877 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1882 fail_open_maybe_recover(p
->fail_open
);
1884 handle_openflow(ofconn
, of_msg
);
1885 ofpbuf_delete(of_msg
);
1889 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1890 ofconn_destroy(ofconn
);
1895 ofconn_wait(struct ofconn
*ofconn
)
1899 if (ofconn
->discovery
) {
1900 discovery_wait(ofconn
->discovery
);
1902 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1903 pinsched_wait(ofconn
->schedulers
[i
]);
1905 rconn_run_wait(ofconn
->rconn
);
1906 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1907 rconn_recv_wait(ofconn
->rconn
);
1909 COVERAGE_INC(ofproto_ofconn_stuck
);
1913 /* Returns true if 'ofconn' should receive asynchronous messages. */
1915 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1917 if (ofconn
->type
== OFCONN_PRIMARY
) {
1918 /* Primary controllers always get asynchronous messages unless they
1919 * have configured themselves as "slaves". */
1920 return ofconn
->role
!= NX_ROLE_SLAVE
;
1922 /* Service connections don't get asynchronous messages unless they have
1923 * explicitly asked for them by setting a nonzero miss send length. */
1924 return ofconn
->miss_send_len
> 0;
1928 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1929 * and 'target', suitable for use in log messages for identifying the
1932 * The name is dynamically allocated. The caller should free it (with free())
1933 * when it is no longer needed. */
1935 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1937 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1941 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1945 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1946 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1950 *s
= pinsched_create(rate
, burst
,
1951 ofconn
->ofproto
->switch_status
);
1953 pinsched_set_limits(*s
, rate
, burst
);
1956 pinsched_destroy(*s
);
1963 ofservice_reconfigure(struct ofservice
*ofservice
,
1964 const struct ofproto_controller
*c
)
1966 ofservice
->probe_interval
= c
->probe_interval
;
1967 ofservice
->rate_limit
= c
->rate_limit
;
1968 ofservice
->burst_limit
= c
->burst_limit
;
1971 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1972 * positive errno value. */
1974 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1976 struct ofservice
*ofservice
;
1977 struct pvconn
*pvconn
;
1980 error
= pvconn_open(c
->target
, &pvconn
);
1985 ofservice
= xzalloc(sizeof *ofservice
);
1986 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1987 hash_string(c
->target
, 0));
1988 ofservice
->pvconn
= pvconn
;
1990 ofservice_reconfigure(ofservice
, c
);
1996 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1998 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1999 pvconn_close(ofservice
->pvconn
);
2003 /* Finds and returns the ofservice within 'ofproto' that has the given
2004 * 'target', or a null pointer if none exists. */
2005 static struct ofservice
*
2006 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
2008 struct ofservice
*ofservice
;
2010 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
2011 &ofproto
->services
) {
2012 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
2019 /* Returns true if 'rule' should be hidden from the controller.
2021 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2022 * (e.g. by in-band control) and are intentionally hidden from the
2025 rule_is_hidden(const struct rule
*rule
)
2027 return rule
->cr
.priority
> UINT16_MAX
;
2030 /* Creates and returns a new rule initialized as specified.
2032 * The caller is responsible for inserting the rule into the classifier (with
2033 * rule_insert()). */
2034 static struct rule
*
2035 rule_create(const struct cls_rule
*cls_rule
,
2036 const union ofp_action
*actions
, size_t n_actions
,
2037 uint16_t idle_timeout
, uint16_t hard_timeout
,
2038 ovs_be64 flow_cookie
, bool send_flow_removed
)
2040 struct rule
*rule
= xzalloc(sizeof *rule
);
2041 rule
->cr
= *cls_rule
;
2042 rule
->idle_timeout
= idle_timeout
;
2043 rule
->hard_timeout
= hard_timeout
;
2044 rule
->flow_cookie
= flow_cookie
;
2045 rule
->used
= rule
->created
= time_msec();
2046 rule
->send_flow_removed
= send_flow_removed
;
2047 list_init(&rule
->facets
);
2048 if (n_actions
> 0) {
2049 rule
->n_actions
= n_actions
;
2050 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
2056 static struct rule
*
2057 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
2059 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
2063 rule_free(struct rule
*rule
)
2065 free(rule
->actions
);
2069 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2070 * destroying any that no longer has a rule (which is probably all of them).
2072 * The caller must have already removed 'rule' from the classifier. */
2074 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
2076 struct facet
*facet
, *next_facet
;
2077 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
2078 facet_revalidate(ofproto
, facet
);
2083 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2084 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2087 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
2089 const union ofp_action
*oa
;
2090 struct actions_iterator i
;
2092 if (out_port
== htons(OFPP_NONE
)) {
2095 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
2096 oa
= actions_next(&i
)) {
2097 if (action_outputs_to_port(oa
, out_port
)) {
2104 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2105 * 'packet', which arrived on 'in_port'.
2107 * Takes ownership of 'packet'. */
2109 execute_odp_actions(struct ofproto
*ofproto
, const struct flow
*flow
,
2110 const struct nlattr
*odp_actions
, size_t actions_len
,
2111 struct ofpbuf
*packet
)
2113 if (actions_len
== NLA_ALIGN(NLA_HDRLEN
+ sizeof(uint64_t))
2114 && odp_actions
->nla_type
== ODP_ACTION_ATTR_CONTROLLER
) {
2115 /* As an optimization, avoid a round-trip from userspace to kernel to
2116 * userspace. This also avoids possibly filling up kernel packet
2117 * buffers along the way. */
2118 struct dpif_upcall upcall
;
2120 upcall
.type
= DPIF_UC_ACTION
;
2121 upcall
.packet
= packet
;
2124 upcall
.userdata
= nl_attr_get_u64(odp_actions
);
2125 upcall
.sample_pool
= 0;
2126 upcall
.actions
= NULL
;
2127 upcall
.actions_len
= 0;
2129 send_packet_in(ofproto
, &upcall
, flow
, false);
2135 error
= dpif_execute(ofproto
->dpif
, odp_actions
, actions_len
, packet
);
2136 ofpbuf_delete(packet
);
2141 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2142 * statistics appropriately. 'packet' must have at least sizeof(struct
2143 * ofp_packet_in) bytes of headroom.
2145 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2146 * applying flow_extract() to 'packet' would yield the same flow as
2149 * 'facet' must have accurately composed ODP actions; that is, it must not be
2150 * in need of revalidation.
2152 * Takes ownership of 'packet'. */
2154 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2155 struct ofpbuf
*packet
)
2157 struct dpif_flow_stats stats
;
2159 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2161 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2162 if (execute_odp_actions(ofproto
, &facet
->flow
,
2163 facet
->actions
, facet
->actions_len
, packet
)) {
2164 facet
->used
= time_msec();
2165 facet_update_stats(ofproto
, facet
, &stats
);
2166 netflow_flow_update_time(ofproto
->netflow
,
2167 &facet
->nf_flow
, facet
->used
);
2171 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2172 * statistics (or the statistics for one of its facets) appropriately.
2173 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2175 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2176 * with statistics for 'packet' either way.
2178 * Takes ownership of 'packet'. */
2180 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2181 struct ofpbuf
*packet
)
2183 struct action_xlate_ctx ctx
;
2184 struct ofpbuf
*odp_actions
;
2185 struct facet
*facet
;
2189 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2191 flow_extract(packet
, 0, in_port
, &flow
);
2193 /* First look for a related facet. If we find one, account it to that. */
2194 facet
= facet_lookup_valid(ofproto
, &flow
);
2195 if (facet
&& facet
->rule
== rule
) {
2196 facet_execute(ofproto
, facet
, packet
);
2200 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2201 * create a new facet for it and use that. */
2202 if (rule_lookup(ofproto
, &flow
) == rule
) {
2203 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2204 facet_execute(ofproto
, facet
, packet
);
2205 facet_install(ofproto
, facet
, true);
2209 /* We can't account anything to a facet. If we were to try, then that
2210 * facet would have a non-matching rule, busting our invariants. */
2211 action_xlate_ctx_init(&ctx
, ofproto
, &flow
, packet
);
2212 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2213 size
= packet
->size
;
2214 if (execute_odp_actions(ofproto
, &flow
, odp_actions
->data
,
2215 odp_actions
->size
, packet
)) {
2216 rule
->used
= time_msec();
2217 rule
->packet_count
++;
2218 rule
->byte_count
+= size
;
2219 flow_push_stats(ofproto
, rule
, &flow
, 1, size
, rule
->used
);
2221 ofpbuf_delete(odp_actions
);
2224 /* Inserts 'rule' into 'p''s flow table. */
2226 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2228 struct rule
*displaced_rule
;
2230 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2231 if (displaced_rule
) {
2232 rule_destroy(p
, displaced_rule
);
2234 p
->need_revalidate
= true;
2237 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2238 * 'flow' and an example 'packet' within that flow.
2240 * The caller must already have determined that no facet with an identical
2241 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2242 * 'ofproto''s classifier table. */
2243 static struct facet
*
2244 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2245 const struct flow
*flow
, const struct ofpbuf
*packet
)
2247 struct facet
*facet
;
2249 facet
= xzalloc(sizeof *facet
);
2250 facet
->used
= time_msec();
2251 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2252 list_push_back(&rule
->facets
, &facet
->list_node
);
2254 facet
->flow
= *flow
;
2255 netflow_flow_init(&facet
->nf_flow
);
2256 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2258 facet_make_actions(ofproto
, facet
, packet
);
2264 facet_free(struct facet
*facet
)
2266 free(facet
->actions
);
2270 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2272 * - Removes 'rule' from the classifier.
2274 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2275 * destroys them), via rule_destroy().
2278 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2280 COVERAGE_INC(ofproto_del_rule
);
2281 ofproto
->need_revalidate
= true;
2282 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2283 rule_destroy(ofproto
, rule
);
2286 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2288 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2289 * rule's statistics, via facet_uninstall().
2291 * - Removes 'facet' from its rule and from ofproto->facets.
2294 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2296 facet_uninstall(ofproto
, facet
);
2297 facet_flush_stats(ofproto
, facet
);
2298 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2299 list_remove(&facet
->list_node
);
2303 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2305 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2306 const struct ofpbuf
*packet
)
2308 const struct rule
*rule
= facet
->rule
;
2309 struct ofpbuf
*odp_actions
;
2310 struct action_xlate_ctx ctx
;
2312 action_xlate_ctx_init(&ctx
, p
, &facet
->flow
, packet
);
2313 odp_actions
= xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2314 facet
->tags
= ctx
.tags
;
2315 facet
->may_install
= ctx
.may_set_up_flow
;
2316 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2318 if (facet
->actions_len
!= odp_actions
->size
2319 || memcmp(facet
->actions
, odp_actions
->data
, odp_actions
->size
)) {
2320 free(facet
->actions
);
2321 facet
->actions_len
= odp_actions
->size
;
2322 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2325 ofpbuf_delete(odp_actions
);
2329 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
,
2330 const struct nlattr
*actions
, size_t actions_len
,
2331 struct dpif_flow_stats
*stats
)
2333 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2334 enum dpif_flow_put_flags flags
;
2337 flags
= DPIF_FP_CREATE
| DPIF_FP_MODIFY
;
2339 flags
|= DPIF_FP_ZERO_STATS
;
2340 facet
->dp_packet_count
= 0;
2341 facet
->dp_byte_count
= 0;
2344 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2345 odp_flow_key_from_flow(&key
, &facet
->flow
);
2346 assert(key
.base
== keybuf
);
2348 return dpif_flow_put(ofproto
->dpif
, flags
, key
.data
, key
.size
,
2349 actions
, actions_len
, stats
);
2352 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2353 * 'zero_stats' is true, clears any existing statistics from the datapath for
2356 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2358 struct dpif_flow_stats stats
;
2360 if (facet
->may_install
2361 && !facet_put__(p
, facet
, facet
->actions
, facet
->actions_len
,
2362 zero_stats
? &stats
: NULL
)) {
2363 facet
->installed
= true;
2367 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2368 * to the accounting hook function in the ofhooks structure. */
2370 facet_account(struct ofproto
*ofproto
,
2371 struct facet
*facet
, uint64_t extra_bytes
)
2373 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2375 if (ofproto
->ofhooks
->account_flow_cb
2376 && total_bytes
> facet
->accounted_bytes
)
2378 ofproto
->ofhooks
->account_flow_cb(
2379 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->actions_len
,
2380 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2381 facet
->accounted_bytes
= total_bytes
;
2385 /* If 'rule' is installed in the datapath, uninstalls it. */
2387 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2389 if (facet
->installed
) {
2390 uint32_t keybuf
[ODPUTIL_FLOW_KEY_U32S
];
2391 struct dpif_flow_stats stats
;
2394 ofpbuf_use_stack(&key
, keybuf
, sizeof keybuf
);
2395 odp_flow_key_from_flow(&key
, &facet
->flow
);
2396 assert(key
.base
== keybuf
);
2398 if (!dpif_flow_del(p
->dpif
, key
.data
, key
.size
, &stats
)) {
2399 facet_update_stats(p
, facet
, &stats
);
2401 facet
->installed
= false;
2402 facet
->dp_packet_count
= 0;
2403 facet
->dp_byte_count
= 0;
2407 /* Returns true if the only action for 'facet' is to send to the controller.
2408 * (We don't report NetFlow expiration messages for such facets because they
2409 * are just part of the control logic for the network, not real traffic). */
2411 facet_is_controller_flow(struct facet
*facet
)
2414 && facet
->rule
->n_actions
== 1
2415 && action_outputs_to_port(&facet
->rule
->actions
[0],
2416 htons(OFPP_CONTROLLER
)));
2419 /* Folds all of 'facet''s statistics into its rule. Also updates the
2420 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
2421 * 'facet''s statistics in the datapath should have been zeroed and folded into
2422 * its packet and byte counts before this function is called. */
2424 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2426 assert(!facet
->dp_byte_count
);
2427 assert(!facet
->dp_packet_count
);
2429 facet_push_stats(ofproto
, facet
);
2430 facet_account(ofproto
, facet
, 0);
2432 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2433 struct ofexpired expired
;
2434 expired
.flow
= facet
->flow
;
2435 expired
.packet_count
= facet
->packet_count
;
2436 expired
.byte_count
= facet
->byte_count
;
2437 expired
.used
= facet
->used
;
2438 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2441 facet
->rule
->packet_count
+= facet
->packet_count
;
2442 facet
->rule
->byte_count
+= facet
->byte_count
;
2444 /* Reset counters to prevent double counting if 'facet' ever gets
2446 facet
->packet_count
= 0;
2447 facet
->byte_count
= 0;
2448 facet
->rs_packet_count
= 0;
2449 facet
->rs_byte_count
= 0;
2450 facet
->accounted_bytes
= 0;
2452 netflow_flow_clear(&facet
->nf_flow
);
2455 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2456 * Returns it if found, otherwise a null pointer.
2458 * The returned facet might need revalidation; use facet_lookup_valid()
2459 * instead if that is important. */
2460 static struct facet
*
2461 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2463 struct facet
*facet
;
2465 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2467 if (flow_equal(flow
, &facet
->flow
)) {
2475 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2476 * Returns it if found, otherwise a null pointer.
2478 * The returned facet is guaranteed to be valid. */
2479 static struct facet
*
2480 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2482 struct facet
*facet
= facet_find(ofproto
, flow
);
2484 /* The facet we found might not be valid, since we could be in need of
2485 * revalidation. If it is not valid, don't return it. */
2487 && ofproto
->need_revalidate
2488 && !facet_revalidate(ofproto
, facet
)) {
2489 COVERAGE_INC(ofproto_invalidated
);
2496 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2498 * - If the rule found is different from 'facet''s current rule, moves
2499 * 'facet' to the new rule and recompiles its actions.
2501 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2502 * where it is and recompiles its actions anyway.
2504 * - If there is none, destroys 'facet'.
2506 * Returns true if 'facet' still exists, false if it has been destroyed. */
2508 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2510 struct action_xlate_ctx ctx
;
2511 struct ofpbuf
*odp_actions
;
2512 struct rule
*new_rule
;
2513 bool actions_changed
;
2515 COVERAGE_INC(facet_revalidate
);
2517 /* Determine the new rule. */
2518 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2520 /* No new rule, so delete the facet. */
2521 facet_remove(ofproto
, facet
);
2525 /* Calculate new ODP actions.
2527 * We do not modify any 'facet' state yet, because we might need to, e.g.,
2528 * emit a NetFlow expiration and, if so, we need to have the old state
2529 * around to properly compose it. */
2530 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
, NULL
);
2531 odp_actions
= xlate_actions(&ctx
, new_rule
->actions
, new_rule
->n_actions
);
2532 actions_changed
= (facet
->actions_len
!= odp_actions
->size
2533 || memcmp(facet
->actions
, odp_actions
->data
,
2534 facet
->actions_len
));
2536 /* If the ODP actions changed or the installability changed, then we need
2537 * to talk to the datapath. */
2538 if (actions_changed
|| ctx
.may_set_up_flow
!= facet
->installed
) {
2539 if (ctx
.may_set_up_flow
) {
2540 struct dpif_flow_stats stats
;
2542 facet_put__(ofproto
, facet
,
2543 odp_actions
->data
, odp_actions
->size
, &stats
);
2544 facet_update_stats(ofproto
, facet
, &stats
);
2546 facet_uninstall(ofproto
, facet
);
2549 /* The datapath flow is gone or has zeroed stats, so push stats out of
2550 * 'facet' into 'rule'. */
2551 facet_flush_stats(ofproto
, facet
);
2554 /* Update 'facet' now that we've taken care of all the old state. */
2555 facet
->tags
= ctx
.tags
;
2556 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2557 facet
->may_install
= ctx
.may_set_up_flow
;
2558 if (actions_changed
) {
2559 free(facet
->actions
);
2560 facet
->actions_len
= odp_actions
->size
;
2561 facet
->actions
= xmemdup(odp_actions
->data
, odp_actions
->size
);
2563 if (facet
->rule
!= new_rule
) {
2564 COVERAGE_INC(facet_changed_rule
);
2565 list_remove(&facet
->list_node
);
2566 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2567 facet
->rule
= new_rule
;
2568 facet
->used
= new_rule
->created
;
2571 ofpbuf_delete(odp_actions
);
2577 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2578 struct rconn_packet_counter
*counter
)
2580 update_openflow_length(msg
);
2581 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2587 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2590 struct ofpbuf
*buf
= ofputil_encode_error_msg(error
, oh
);
2592 COVERAGE_INC(ofproto_error
);
2593 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2598 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2600 opp
->port_no
= htons(opp
->port_no
);
2601 opp
->config
= htonl(opp
->config
);
2602 opp
->state
= htonl(opp
->state
);
2603 opp
->curr
= htonl(opp
->curr
);
2604 opp
->advertised
= htonl(opp
->advertised
);
2605 opp
->supported
= htonl(opp
->supported
);
2606 opp
->peer
= htonl(opp
->peer
);
2610 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2612 queue_tx(make_echo_reply(oh
), ofconn
, ofconn
->reply_counter
);
2617 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2619 struct ofp_switch_features
*osf
;
2621 struct ofport
*port
;
2623 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2624 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2625 osf
->n_buffers
= htonl(pktbuf_capacity());
2627 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2628 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2629 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2630 (1u << OFPAT_SET_VLAN_VID
) |
2631 (1u << OFPAT_SET_VLAN_PCP
) |
2632 (1u << OFPAT_STRIP_VLAN
) |
2633 (1u << OFPAT_SET_DL_SRC
) |
2634 (1u << OFPAT_SET_DL_DST
) |
2635 (1u << OFPAT_SET_NW_SRC
) |
2636 (1u << OFPAT_SET_NW_DST
) |
2637 (1u << OFPAT_SET_NW_TOS
) |
2638 (1u << OFPAT_SET_TP_SRC
) |
2639 (1u << OFPAT_SET_TP_DST
) |
2640 (1u << OFPAT_ENQUEUE
));
2642 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2643 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2646 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2651 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2654 struct ofp_switch_config
*osc
;
2658 /* Figure out flags. */
2659 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2660 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2663 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2664 osc
->flags
= htons(flags
);
2665 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2666 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2672 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
2674 uint16_t flags
= ntohs(osc
->flags
);
2676 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2677 switch (flags
& OFPC_FRAG_MASK
) {
2678 case OFPC_FRAG_NORMAL
:
2679 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2681 case OFPC_FRAG_DROP
:
2682 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2685 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2691 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2696 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2697 struct action_xlate_ctx
*ctx
);
2700 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2702 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2705 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2706 /* Forwarding disabled on port. */
2711 * We don't have an ofport record for this port, but it doesn't hurt to
2712 * allow forwarding to it anyhow. Maybe such a port will appear later
2713 * and we're pre-populating the flow table.
2717 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_OUTPUT
, port
);
2718 ctx
->nf_output_iface
= port
;
2721 static struct rule
*
2722 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2724 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2728 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2730 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2731 uint16_t old_in_port
;
2734 /* Look up a flow with 'in_port' as the input port. Then restore the
2735 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2736 * have surprising behavior). */
2737 old_in_port
= ctx
->flow
.in_port
;
2738 ctx
->flow
.in_port
= in_port
;
2739 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2740 ctx
->flow
.in_port
= old_in_port
;
2742 if (ctx
->resubmit_hook
) {
2743 ctx
->resubmit_hook(ctx
, rule
);
2748 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2752 static struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2754 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2755 MAX_RESUBMIT_RECURSION
);
2760 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2761 uint16_t *nf_output_iface
, struct ofpbuf
*odp_actions
)
2763 struct ofport
*ofport
;
2765 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2766 uint16_t odp_port
= ofport
->odp_port
;
2767 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2768 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, odp_port
);
2771 *nf_output_iface
= NF_OUT_FLOOD
;
2775 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2776 uint16_t port
, uint16_t max_len
)
2779 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2781 ctx
->nf_output_iface
= NF_OUT_DROP
;
2785 add_output_action(ctx
, ctx
->flow
.in_port
);
2788 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2791 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2792 ctx
->odp_actions
, &ctx
->tags
,
2793 &ctx
->nf_output_iface
,
2794 ctx
->ofproto
->aux
)) {
2795 COVERAGE_INC(ofproto_uninstallable
);
2796 ctx
->may_set_up_flow
= false;
2800 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2801 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2804 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2805 &ctx
->nf_output_iface
, ctx
->odp_actions
);
2807 case OFPP_CONTROLLER
:
2808 nl_msg_put_u64(ctx
->odp_actions
, ODP_ACTION_ATTR_CONTROLLER
, max_len
);
2811 add_output_action(ctx
, ODPP_LOCAL
);
2814 odp_port
= ofp_port_to_odp_port(port
);
2815 if (odp_port
!= ctx
->flow
.in_port
) {
2816 add_output_action(ctx
, odp_port
);
2821 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2822 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2823 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2824 ctx
->nf_output_iface
= prev_nf_output_iface
;
2825 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2826 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2827 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2832 xlate_output_action(struct action_xlate_ctx
*ctx
,
2833 const struct ofp_action_output
*oao
)
2835 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2838 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2839 * optimization, because we're going to add another action that sets the
2840 * priority immediately after, or because there are no actions following the
2843 remove_pop_action(struct action_xlate_ctx
*ctx
)
2845 if (ctx
->odp_actions
->size
== ctx
->last_pop_priority
) {
2846 ctx
->odp_actions
->size
-= NLA_ALIGN(NLA_HDRLEN
);
2847 ctx
->last_pop_priority
= -1;
2852 add_pop_action(struct action_xlate_ctx
*ctx
)
2854 if (ctx
->odp_actions
->size
!= ctx
->last_pop_priority
) {
2855 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_POP_PRIORITY
);
2856 ctx
->last_pop_priority
= ctx
->odp_actions
->size
;
2861 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2862 const struct ofp_action_enqueue
*oae
)
2864 uint16_t ofp_port
, odp_port
;
2868 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2871 /* Fall back to ordinary output action. */
2872 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2876 /* Figure out ODP output port. */
2877 ofp_port
= ntohs(oae
->port
);
2878 if (ofp_port
!= OFPP_IN_PORT
) {
2879 odp_port
= ofp_port_to_odp_port(ofp_port
);
2881 odp_port
= ctx
->flow
.in_port
;
2884 /* Add ODP actions. */
2885 remove_pop_action(ctx
);
2886 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2887 add_output_action(ctx
, odp_port
);
2888 add_pop_action(ctx
);
2890 /* Update NetFlow output port. */
2891 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2892 ctx
->nf_output_iface
= odp_port
;
2893 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2894 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2899 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2900 const struct nx_action_set_queue
*nasq
)
2905 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2908 /* Couldn't translate queue to a priority, so ignore. A warning
2909 * has already been logged. */
2913 remove_pop_action(ctx
);
2914 nl_msg_put_u32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_PRIORITY
, priority
);
2918 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2920 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2921 if (!(tci
& htons(VLAN_CFI
))) {
2922 nl_msg_put_flag(ctx
->odp_actions
, ODP_ACTION_ATTR_STRIP_VLAN
);
2924 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_TCI
,
2925 tci
& ~htons(VLAN_CFI
));
2929 struct xlate_reg_state
{
2935 save_reg_state(const struct action_xlate_ctx
*ctx
,
2936 struct xlate_reg_state
*state
)
2938 state
->vlan_tci
= ctx
->flow
.vlan_tci
;
2939 state
->tun_id
= ctx
->flow
.tun_id
;
2943 update_reg_state(struct action_xlate_ctx
*ctx
,
2944 const struct xlate_reg_state
*state
)
2946 if (ctx
->flow
.vlan_tci
!= state
->vlan_tci
) {
2947 xlate_set_dl_tci(ctx
);
2949 if (ctx
->flow
.tun_id
!= state
->tun_id
) {
2950 nl_msg_put_be64(ctx
->odp_actions
,
2951 ODP_ACTION_ATTR_SET_TUNNEL
, ctx
->flow
.tun_id
);
2956 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2957 const struct nx_action_header
*nah
)
2959 const struct nx_action_resubmit
*nar
;
2960 const struct nx_action_set_tunnel
*nast
;
2961 const struct nx_action_set_queue
*nasq
;
2962 const struct nx_action_multipath
*nam
;
2963 enum nx_action_subtype subtype
= ntohs(nah
->subtype
);
2964 struct xlate_reg_state state
;
2967 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2969 case NXAST_RESUBMIT
:
2970 nar
= (const struct nx_action_resubmit
*) nah
;
2971 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2974 case NXAST_SET_TUNNEL
:
2975 nast
= (const struct nx_action_set_tunnel
*) nah
;
2976 tun_id
= htonll(ntohl(nast
->tun_id
));
2977 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
2978 ctx
->flow
.tun_id
= tun_id
;
2981 case NXAST_DROP_SPOOFED_ARP
:
2982 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2983 nl_msg_put_flag(ctx
->odp_actions
,
2984 ODP_ACTION_ATTR_DROP_SPOOFED_ARP
);
2988 case NXAST_SET_QUEUE
:
2989 nasq
= (const struct nx_action_set_queue
*) nah
;
2990 xlate_set_queue_action(ctx
, nasq
);
2993 case NXAST_POP_QUEUE
:
2994 add_pop_action(ctx
);
2997 case NXAST_REG_MOVE
:
2998 save_reg_state(ctx
, &state
);
2999 nxm_execute_reg_move((const struct nx_action_reg_move
*) nah
,
3001 update_reg_state(ctx
, &state
);
3004 case NXAST_REG_LOAD
:
3005 save_reg_state(ctx
, &state
);
3006 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
3008 update_reg_state(ctx
, &state
);
3012 /* Nothing to do. */
3015 case NXAST_SET_TUNNEL64
:
3016 tun_id
= ((const struct nx_action_set_tunnel64
*) nah
)->tun_id
;
3017 nl_msg_put_be64(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TUNNEL
, tun_id
);
3018 ctx
->flow
.tun_id
= tun_id
;
3021 case NXAST_MULTIPATH
:
3022 nam
= (const struct nx_action_multipath
*) nah
;
3023 multipath_execute(nam
, &ctx
->flow
);
3026 /* If you add a new action here that modifies flow data, don't forget to
3027 * update the flow key in ctx->flow at the same time. */
3029 case NXAST_SNAT__OBSOLETE
:
3031 VLOG_DBG_RL(&rl
, "unknown Nicira action type %d", (int) subtype
);
3037 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
3038 struct action_xlate_ctx
*ctx
)
3040 struct actions_iterator iter
;
3041 const union ofp_action
*ia
;
3042 const struct ofport
*port
;
3044 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
3045 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
3046 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
3047 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
3048 /* Drop this flow. */
3052 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
3053 enum ofp_action_type type
= ntohs(ia
->type
);
3054 const struct ofp_action_dl_addr
*oada
;
3058 xlate_output_action(ctx
, &ia
->output
);
3061 case OFPAT_SET_VLAN_VID
:
3062 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
3063 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
3064 xlate_set_dl_tci(ctx
);
3067 case OFPAT_SET_VLAN_PCP
:
3068 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
3069 ctx
->flow
.vlan_tci
|= htons(
3070 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
3071 xlate_set_dl_tci(ctx
);
3074 case OFPAT_STRIP_VLAN
:
3075 ctx
->flow
.vlan_tci
= htons(0);
3076 xlate_set_dl_tci(ctx
);
3079 case OFPAT_SET_DL_SRC
:
3080 oada
= ((struct ofp_action_dl_addr
*) ia
);
3081 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_SRC
,
3082 oada
->dl_addr
, ETH_ADDR_LEN
);
3083 memcpy(ctx
->flow
.dl_src
, oada
->dl_addr
, ETH_ADDR_LEN
);
3086 case OFPAT_SET_DL_DST
:
3087 oada
= ((struct ofp_action_dl_addr
*) ia
);
3088 nl_msg_put_unspec(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_DL_DST
,
3089 oada
->dl_addr
, ETH_ADDR_LEN
);
3090 memcpy(ctx
->flow
.dl_dst
, oada
->dl_addr
, ETH_ADDR_LEN
);
3093 case OFPAT_SET_NW_SRC
:
3094 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_SRC
,
3095 ia
->nw_addr
.nw_addr
);
3096 ctx
->flow
.nw_src
= ia
->nw_addr
.nw_addr
;
3099 case OFPAT_SET_NW_DST
:
3100 nl_msg_put_be32(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_DST
,
3101 ia
->nw_addr
.nw_addr
);
3102 ctx
->flow
.nw_dst
= ia
->nw_addr
.nw_addr
;
3105 case OFPAT_SET_NW_TOS
:
3106 nl_msg_put_u8(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_NW_TOS
,
3108 ctx
->flow
.nw_tos
= ia
->nw_tos
.nw_tos
;
3111 case OFPAT_SET_TP_SRC
:
3112 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_SRC
,
3113 ia
->tp_port
.tp_port
);
3114 ctx
->flow
.tp_src
= ia
->tp_port
.tp_port
;
3117 case OFPAT_SET_TP_DST
:
3118 nl_msg_put_be16(ctx
->odp_actions
, ODP_ACTION_ATTR_SET_TP_DST
,
3119 ia
->tp_port
.tp_port
);
3120 ctx
->flow
.tp_dst
= ia
->tp_port
.tp_port
;
3124 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
3128 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
3132 VLOG_DBG_RL(&rl
, "unknown action type %d", (int) type
);
3139 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
3140 struct ofproto
*ofproto
, const struct flow
*flow
,
3141 const struct ofpbuf
*packet
)
3143 ctx
->ofproto
= ofproto
;
3145 ctx
->packet
= packet
;
3146 ctx
->resubmit_hook
= NULL
;
3147 ctx
->check_special
= true;
3150 static struct ofpbuf
*
3151 xlate_actions(struct action_xlate_ctx
*ctx
,
3152 const union ofp_action
*in
, size_t n_in
)
3154 COVERAGE_INC(ofproto_ofp2odp
);
3156 ctx
->odp_actions
= ofpbuf_new(512);
3158 ctx
->may_set_up_flow
= true;
3159 ctx
->nf_output_iface
= NF_OUT_DROP
;
3161 ctx
->last_pop_priority
= -1;
3163 if (!ctx
->check_special
3164 || !ctx
->ofproto
->ofhooks
->special_cb
3165 || ctx
->ofproto
->ofhooks
->special_cb(&ctx
->flow
, ctx
->packet
,
3166 ctx
->ofproto
->aux
)) {
3167 do_xlate_actions(in
, n_in
, ctx
);
3169 ctx
->may_set_up_flow
= false;
3172 remove_pop_action(ctx
);
3174 /* Check with in-band control to see if we're allowed to set up this
3176 if (!in_band_rule_check(ctx
->ofproto
->in_band
, &ctx
->flow
,
3177 ctx
->odp_actions
->data
, ctx
->odp_actions
->size
)) {
3178 ctx
->may_set_up_flow
= false;
3181 return ctx
->odp_actions
;
3184 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3185 * error message code (composed with ofp_mkerr()) for the caller to propagate
3186 * upward. Otherwise, returns 0.
3188 * The log message mentions 'msg_type'. */
3190 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3192 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3193 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3194 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3197 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3204 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3206 struct ofproto
*p
= ofconn
->ofproto
;
3207 struct ofp_packet_out
*opo
;
3208 struct ofpbuf payload
, *buffer
;
3209 union ofp_action
*ofp_actions
;
3210 struct action_xlate_ctx ctx
;
3211 struct ofpbuf
*odp_actions
;
3212 struct ofpbuf request
;
3214 size_t n_ofp_actions
;
3218 COVERAGE_INC(ofproto_packet_out
);
3220 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3225 /* Get ofp_packet_out. */
3226 ofpbuf_use_const(&request
, oh
, ntohs(oh
->length
));
3227 opo
= ofpbuf_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3230 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3231 &ofp_actions
, &n_ofp_actions
);
3237 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3238 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3240 if (error
|| !buffer
) {
3249 /* Extract flow, check actions. */
3250 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3252 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3258 action_xlate_ctx_init(&ctx
, p
, &flow
, &payload
);
3259 odp_actions
= xlate_actions(&ctx
, ofp_actions
, n_ofp_actions
);
3260 dpif_execute(p
->dpif
, odp_actions
->data
, odp_actions
->size
, &payload
);
3261 ofpbuf_delete(odp_actions
);
3264 ofpbuf_delete(buffer
);
3269 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3270 uint32_t config
, uint32_t mask
)
3272 mask
&= config
^ port
->opp
.config
;
3273 if (mask
& OFPPC_PORT_DOWN
) {
3274 if (config
& OFPPC_PORT_DOWN
) {
3275 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3277 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3280 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3281 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3282 if (mask
& REVALIDATE_BITS
) {
3283 COVERAGE_INC(ofproto_costly_flags
);
3284 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3285 p
->need_revalidate
= true;
3287 #undef REVALIDATE_BITS
3288 if (mask
& OFPPC_NO_PACKET_IN
) {
3289 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3294 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3296 struct ofproto
*p
= ofconn
->ofproto
;
3297 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
3298 struct ofport
*port
;
3301 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3306 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3308 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3309 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3310 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3312 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3313 if (opm
->advertise
) {
3314 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3320 static struct ofpbuf
*
3321 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3323 struct ofp_stats_reply
*osr
;
3326 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3327 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3329 osr
->flags
= htons(0);
3333 static struct ofpbuf
*
3334 start_ofp_stats_reply(const struct ofp_header
*request
, size_t body_len
)
3336 const struct ofp_stats_request
*osr
3337 = (const struct ofp_stats_request
*) request
;
3338 return make_ofp_stats_reply(osr
->header
.xid
, osr
->type
, body_len
);
3342 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3343 struct ofpbuf
**msgp
)
3345 struct ofpbuf
*msg
= *msgp
;
3346 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3347 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3348 struct ofp_stats_reply
*reply
= msg
->data
;
3349 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3350 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3351 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3353 return ofpbuf_put_uninit(*msgp
, nbytes
);
3356 static struct ofpbuf
*
3357 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3359 struct nicira_stats_msg
*nsm
;
3362 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3363 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3364 nsm
->type
= htons(OFPST_VENDOR
);
3365 nsm
->flags
= htons(0);
3366 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3367 nsm
->subtype
= subtype
;
3371 static struct ofpbuf
*
3372 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3374 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3378 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3379 struct ofpbuf
**msgp
)
3381 struct ofpbuf
*msg
= *msgp
;
3382 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3383 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3384 struct nicira_stats_msg
*reply
= msg
->data
;
3385 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3386 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3387 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3389 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3393 handle_desc_stats_request(struct ofconn
*ofconn
,
3394 const struct ofp_header
*request
)
3396 struct ofproto
*p
= ofconn
->ofproto
;
3397 struct ofp_desc_stats
*ods
;
3400 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3401 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3402 memset(ods
, 0, sizeof *ods
);
3403 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3404 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3405 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3406 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3407 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3408 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3414 handle_table_stats_request(struct ofconn
*ofconn
,
3415 const struct ofp_header
*request
)
3417 struct ofproto
*p
= ofconn
->ofproto
;
3418 struct ofp_table_stats
*ots
;
3421 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3423 /* Classifier table. */
3424 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3425 memset(ots
, 0, sizeof *ots
);
3426 strcpy(ots
->name
, "classifier");
3427 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3428 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3429 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3430 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3431 put_32aligned_be64(&ots
->lookup_count
, htonll(0)); /* XXX */
3432 put_32aligned_be64(&ots
->matched_count
, htonll(0)); /* XXX */
3434 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3439 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3440 struct ofpbuf
**msgp
)
3442 struct netdev_stats stats
;
3443 struct ofp_port_stats
*ops
;
3445 /* Intentionally ignore return value, since errors will set
3446 * 'stats' to all-1s, which is correct for OpenFlow, and
3447 * netdev_get_stats() will log errors. */
3448 netdev_get_stats(port
->netdev
, &stats
);
3450 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3451 ops
->port_no
= htons(port
->opp
.port_no
);
3452 memset(ops
->pad
, 0, sizeof ops
->pad
);
3453 put_32aligned_be64(&ops
->rx_packets
, htonll(stats
.rx_packets
));
3454 put_32aligned_be64(&ops
->tx_packets
, htonll(stats
.tx_packets
));
3455 put_32aligned_be64(&ops
->rx_bytes
, htonll(stats
.rx_bytes
));
3456 put_32aligned_be64(&ops
->tx_bytes
, htonll(stats
.tx_bytes
));
3457 put_32aligned_be64(&ops
->rx_dropped
, htonll(stats
.rx_dropped
));
3458 put_32aligned_be64(&ops
->tx_dropped
, htonll(stats
.tx_dropped
));
3459 put_32aligned_be64(&ops
->rx_errors
, htonll(stats
.rx_errors
));
3460 put_32aligned_be64(&ops
->tx_errors
, htonll(stats
.tx_errors
));
3461 put_32aligned_be64(&ops
->rx_frame_err
, htonll(stats
.rx_frame_errors
));
3462 put_32aligned_be64(&ops
->rx_over_err
, htonll(stats
.rx_over_errors
));
3463 put_32aligned_be64(&ops
->rx_crc_err
, htonll(stats
.rx_crc_errors
));
3464 put_32aligned_be64(&ops
->collisions
, htonll(stats
.collisions
));
3468 handle_port_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3470 struct ofproto
*p
= ofconn
->ofproto
;
3471 const struct ofp_port_stats_request
*psr
= ofputil_stats_body(oh
);
3472 struct ofp_port_stats
*ops
;
3474 struct ofport
*port
;
3476 msg
= start_ofp_stats_reply(oh
, sizeof *ops
* 16);
3477 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3478 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3480 append_port_stat(port
, ofconn
, &msg
);
3483 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3484 append_port_stat(port
, ofconn
, &msg
);
3488 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3493 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3495 long long int msecs
= time_msec() - start
;
3496 *sec
= htonl(msecs
/ 1000);
3497 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3501 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3502 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3504 struct ofp_flow_stats
*ofs
;
3505 uint64_t packet_count
, byte_count
;
3507 size_t act_len
, len
;
3509 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3513 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3514 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3516 rule_get_stats(rule
, &packet_count
, &byte_count
);
3518 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3519 ofs
->length
= htons(len
);
3522 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
,
3523 rule
->flow_cookie
, &cookie
);
3524 put_32aligned_be64(&ofs
->cookie
, cookie
);
3525 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3526 ofs
->priority
= htons(rule
->cr
.priority
);
3527 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3528 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3529 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3530 put_32aligned_be64(&ofs
->packet_count
, htonll(packet_count
));
3531 put_32aligned_be64(&ofs
->byte_count
, htonll(byte_count
));
3532 if (rule
->n_actions
> 0) {
3533 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3538 is_valid_table(uint8_t table_id
)
3540 return table_id
== 0 || table_id
== 0xff;
3544 handle_flow_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3546 const struct ofp_flow_stats_request
*fsr
= ofputil_stats_body(oh
);
3547 struct ofpbuf
*reply
;
3549 COVERAGE_INC(ofproto_flows_req
);
3550 reply
= start_ofp_stats_reply(oh
, 1024);
3551 if (is_valid_table(fsr
->table_id
)) {
3552 struct cls_cursor cursor
;
3553 struct cls_rule target
;
3556 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3558 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3559 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3560 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3563 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3569 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3570 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3572 struct nx_flow_stats
*nfs
;
3573 uint64_t packet_count
, byte_count
;
3574 size_t act_len
, start_len
;
3575 struct ofpbuf
*reply
;
3577 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3581 rule_get_stats(rule
, &packet_count
, &byte_count
);
3583 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3585 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3586 start_len
= (*replyp
)->size
;
3589 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3592 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3593 nfs
->cookie
= rule
->flow_cookie
;
3594 nfs
->priority
= htons(rule
->cr
.priority
);
3595 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3596 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3597 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3598 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3599 nfs
->packet_count
= htonll(packet_count
);
3600 nfs
->byte_count
= htonll(byte_count
);
3601 if (rule
->n_actions
> 0) {
3602 ofpbuf_put(reply
, rule
->actions
, act_len
);
3604 nfs
->length
= htons(reply
->size
- start_len
);
3608 handle_nxst_flow(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3610 struct nx_flow_stats_request
*nfsr
;
3611 struct cls_rule target
;
3612 struct ofpbuf
*reply
;
3616 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3618 /* Dissect the message. */
3619 nfsr
= ofpbuf_pull(&b
, sizeof *nfsr
);
3620 error
= nx_pull_match(&b
, ntohs(nfsr
->match_len
), 0, &target
);
3625 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3628 COVERAGE_INC(ofproto_flows_req
);
3629 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3630 if (is_valid_table(nfsr
->table_id
)) {
3631 struct cls_cursor cursor
;
3634 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3635 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3636 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3639 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3645 flow_stats_ds(struct rule
*rule
, struct ds
*results
)
3647 uint64_t packet_count
, byte_count
;
3648 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3650 rule_get_stats(rule
, &packet_count
, &byte_count
);
3652 ds_put_format(results
, "duration=%llds, ",
3653 (time_msec() - rule
->created
) / 1000);
3654 ds_put_format(results
, "idle=%.3fs, ", (time_msec() - rule
->used
) / 1000.0);
3655 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3656 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3657 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3658 cls_rule_format(&rule
->cr
, results
);
3659 ds_put_char(results
, ',');
3661 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3663 ds_put_cstr(results
, "drop");
3665 ds_put_cstr(results
, "\n");
3668 /* Adds a pretty-printed description of all flows to 'results', including
3669 * those marked hidden by secchan (e.g., by in-band control). */
3671 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3673 struct cls_cursor cursor
;
3676 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3677 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3678 flow_stats_ds(rule
, results
);
3683 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3684 ovs_be16 out_port
, uint8_t table_id
,
3685 struct ofp_aggregate_stats_reply
*oasr
)
3687 uint64_t total_packets
= 0;
3688 uint64_t total_bytes
= 0;
3691 COVERAGE_INC(ofproto_agg_request
);
3693 if (is_valid_table(table_id
)) {
3694 struct cls_cursor cursor
;
3697 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3698 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3699 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3700 uint64_t packet_count
;
3701 uint64_t byte_count
;
3703 rule_get_stats(rule
, &packet_count
, &byte_count
);
3705 total_packets
+= packet_count
;
3706 total_bytes
+= byte_count
;
3712 oasr
->flow_count
= htonl(n_flows
);
3713 put_32aligned_be64(&oasr
->packet_count
, htonll(total_packets
));
3714 put_32aligned_be64(&oasr
->byte_count
, htonll(total_bytes
));
3715 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3719 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3720 const struct ofp_header
*oh
)
3722 const struct ofp_aggregate_stats_request
*request
= ofputil_stats_body(oh
);
3723 struct ofp_aggregate_stats_reply
*reply
;
3724 struct cls_rule target
;
3727 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3730 msg
= start_ofp_stats_reply(oh
, sizeof *reply
);
3731 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3732 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3733 request
->table_id
, reply
);
3734 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3739 handle_nxst_aggregate(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3741 struct nx_aggregate_stats_request
*request
;
3742 struct ofp_aggregate_stats_reply
*reply
;
3743 struct cls_rule target
;
3748 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3750 /* Dissect the message. */
3751 request
= ofpbuf_pull(&b
, sizeof *request
);
3752 error
= nx_pull_match(&b
, ntohs(request
->match_len
), 0, &target
);
3757 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3761 COVERAGE_INC(ofproto_flows_req
);
3762 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3763 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3764 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3765 request
->table_id
, reply
);
3766 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3771 struct queue_stats_cbdata
{
3772 struct ofconn
*ofconn
;
3773 struct ofport
*ofport
;
3778 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3779 const struct netdev_queue_stats
*stats
)
3781 struct ofp_queue_stats
*reply
;
3783 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3784 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3785 memset(reply
->pad
, 0, sizeof reply
->pad
);
3786 reply
->queue_id
= htonl(queue_id
);
3787 put_32aligned_be64(&reply
->tx_bytes
, htonll(stats
->tx_bytes
));
3788 put_32aligned_be64(&reply
->tx_packets
, htonll(stats
->tx_packets
));
3789 put_32aligned_be64(&reply
->tx_errors
, htonll(stats
->tx_errors
));
3793 handle_queue_stats_dump_cb(uint32_t queue_id
,
3794 struct netdev_queue_stats
*stats
,
3797 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3799 put_queue_stats(cbdata
, queue_id
, stats
);
3803 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3804 struct queue_stats_cbdata
*cbdata
)
3806 cbdata
->ofport
= port
;
3807 if (queue_id
== OFPQ_ALL
) {
3808 netdev_dump_queue_stats(port
->netdev
,
3809 handle_queue_stats_dump_cb
, cbdata
);
3811 struct netdev_queue_stats stats
;
3813 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3814 put_queue_stats(cbdata
, queue_id
, &stats
);
3820 handle_queue_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3822 struct ofproto
*ofproto
= ofconn
->ofproto
;
3823 const struct ofp_queue_stats_request
*qsr
;
3824 struct queue_stats_cbdata cbdata
;
3825 struct ofport
*port
;
3826 unsigned int port_no
;
3829 qsr
= ofputil_stats_body(oh
);
3831 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3834 COVERAGE_INC(ofproto_queue_req
);
3836 cbdata
.ofconn
= ofconn
;
3837 cbdata
.msg
= start_ofp_stats_reply(oh
, 128);
3839 port_no
= ntohs(qsr
->port_no
);
3840 queue_id
= ntohl(qsr
->queue_id
);
3841 if (port_no
== OFPP_ALL
) {
3842 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3843 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3845 } else if (port_no
< ofproto
->max_ports
) {
3846 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3848 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3851 ofpbuf_delete(cbdata
.msg
);
3852 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3854 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3859 /* Updates 'facet''s used time. Caller is responsible for calling
3860 * facet_push_stats() to update the flows which 'facet' resubmits into. */
3862 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3865 if (used
> facet
->used
) {
3867 if (used
> facet
->rule
->used
) {
3868 facet
->rule
->used
= used
;
3870 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3874 /* Folds the statistics from 'stats' into the counters in 'facet'.
3876 * Because of the meaning of a facet's counters, it only makes sense to do this
3877 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3878 * packet that was sent by hand or if it represents statistics that have been
3879 * cleared out of the datapath. */
3881 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3882 const struct dpif_flow_stats
*stats
)
3884 if (stats
->n_packets
) {
3885 facet_update_time(ofproto
, facet
, stats
->used
);
3886 facet
->packet_count
+= stats
->n_packets
;
3887 facet
->byte_count
+= stats
->n_bytes
;
3888 facet_push_stats(ofproto
, facet
);
3889 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3894 facet_push_stats(struct ofproto
*ofproto
, struct facet
*facet
)
3896 uint64_t rs_packets
, rs_bytes
;
3898 assert(facet
->packet_count
>= facet
->rs_packet_count
);
3899 assert(facet
->byte_count
>= facet
->rs_byte_count
);
3900 assert(facet
->used
>= facet
->rs_used
);
3902 rs_packets
= facet
->packet_count
- facet
->rs_packet_count
;
3903 rs_bytes
= facet
->byte_count
- facet
->rs_byte_count
;
3905 if (rs_packets
|| rs_bytes
|| facet
->used
> facet
->rs_used
) {
3906 facet
->rs_packet_count
= facet
->packet_count
;
3907 facet
->rs_byte_count
= facet
->byte_count
;
3908 facet
->rs_used
= facet
->used
;
3910 flow_push_stats(ofproto
, facet
->rule
, &facet
->flow
,
3911 rs_packets
, rs_bytes
, facet
->used
);
3915 struct ofproto_push
{
3916 struct action_xlate_ctx ctx
;
3923 push_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
3925 struct ofproto_push
*push
= CONTAINER_OF(ctx
, struct ofproto_push
, ctx
);
3928 rule
->packet_count
+= push
->packets
;
3929 rule
->byte_count
+= push
->bytes
;
3930 rule
->used
= MAX(push
->used
, rule
->used
);
3934 /* Pushes flow statistics to the rules which 'flow' resubmits into given
3935 * 'rule''s actions. */
3937 flow_push_stats(struct ofproto
*ofproto
, const struct rule
*rule
,
3938 struct flow
*flow
, uint64_t packets
, uint64_t bytes
,
3941 struct ofproto_push push
;
3943 push
.packets
= packets
;
3947 action_xlate_ctx_init(&push
.ctx
, ofproto
, flow
, NULL
);
3948 push
.ctx
.resubmit_hook
= push_resubmit
;
3949 ofpbuf_delete(xlate_actions(&push
.ctx
, rule
->actions
, rule
->n_actions
));
3952 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3953 * in which no matching flow already exists in the flow table.
3955 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3956 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3957 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3959 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3962 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3964 struct ofproto
*p
= ofconn
->ofproto
;
3965 struct ofpbuf
*packet
;
3970 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3971 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3972 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3976 if (fm
->buffer_id
!= UINT32_MAX
) {
3977 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3981 in_port
= UINT16_MAX
;
3984 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
3985 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3986 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3987 rule_insert(p
, rule
);
3989 rule_execute(p
, rule
, in_port
, packet
);
3994 static struct rule
*
3995 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3997 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
4001 send_buffered_packet(struct ofconn
*ofconn
,
4002 struct rule
*rule
, uint32_t buffer_id
)
4004 struct ofpbuf
*packet
;
4008 if (buffer_id
== UINT32_MAX
) {
4012 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
4017 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
4022 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4024 struct modify_flows_cbdata
{
4025 struct ofproto
*ofproto
;
4026 const struct flow_mod
*fm
;
4030 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
4033 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
4034 * encoded by ofp_mkerr() on failure.
4036 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4039 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4041 struct ofproto
*p
= ofconn
->ofproto
;
4042 struct rule
*match
= NULL
;
4043 struct cls_cursor cursor
;
4046 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4047 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
4048 if (!rule_is_hidden(rule
)) {
4050 modify_flow(p
, fm
, rule
);
4055 /* This credits the packet to whichever flow happened to match last.
4056 * That's weird. Maybe we should do a lookup for the flow that
4057 * actually matches the packet? Who knows. */
4058 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
4061 return add_flow(ofconn
, fm
);
4065 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4066 * code as encoded by ofp_mkerr() on failure.
4068 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4071 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4073 struct ofproto
*p
= ofconn
->ofproto
;
4074 struct rule
*rule
= find_flow_strict(p
, fm
);
4075 if (rule
&& !rule_is_hidden(rule
)) {
4076 modify_flow(p
, fm
, rule
);
4077 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
4079 return add_flow(ofconn
, fm
);
4083 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4084 * been identified as a flow in 'p''s flow table to be modified, by changing
4085 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4086 * ofp_action[] structures). */
4088 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
4090 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
4092 rule
->flow_cookie
= fm
->cookie
;
4094 /* If the actions are the same, do nothing. */
4095 if (fm
->n_actions
== rule
->n_actions
4097 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
4101 /* Replace actions. */
4102 free(rule
->actions
);
4103 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
4104 rule
->n_actions
= fm
->n_actions
;
4106 p
->need_revalidate
= true;
4111 /* OFPFC_DELETE implementation. */
4113 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
4115 /* Implements OFPFC_DELETE. */
4117 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
4119 struct rule
*rule
, *next_rule
;
4120 struct cls_cursor cursor
;
4122 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4123 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4124 delete_flow(p
, rule
, htons(fm
->out_port
));
4128 /* Implements OFPFC_DELETE_STRICT. */
4130 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4132 struct rule
*rule
= find_flow_strict(p
, fm
);
4134 delete_flow(p
, rule
, htons(fm
->out_port
));
4138 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4139 * been identified as a flow to delete from 'p''s flow table, by deleting the
4140 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4143 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4144 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4145 * specified 'out_port'. */
4147 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4149 if (rule_is_hidden(rule
)) {
4153 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4157 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4158 rule_remove(p
, rule
);
4162 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4164 struct ofproto
*p
= ofconn
->ofproto
;
4168 error
= reject_slave_controller(ofconn
, "flow_mod");
4173 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn
->flow_format
);
4178 /* We do not support the emergency flow cache. It will hopefully get
4179 * dropped from OpenFlow in the near future. */
4180 if (fm
.flags
& OFPFF_EMERG
) {
4181 /* There isn't a good fit for an error code, so just state that the
4182 * flow table is full. */
4183 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4186 error
= validate_actions(fm
.actions
, fm
.n_actions
,
4187 &fm
.cr
.flow
, p
->max_ports
);
4192 switch (fm
.command
) {
4194 return add_flow(ofconn
, &fm
);
4197 return modify_flows_loose(ofconn
, &fm
);
4199 case OFPFC_MODIFY_STRICT
:
4200 return modify_flow_strict(ofconn
, &fm
);
4203 delete_flows_loose(p
, &fm
);
4206 case OFPFC_DELETE_STRICT
:
4207 delete_flow_strict(p
, &fm
);
4211 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4216 handle_tun_id_from_cookie(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4218 const struct nxt_tun_id_cookie
*msg
4219 = (const struct nxt_tun_id_cookie
*) oh
;
4221 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4226 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4228 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
4229 struct nx_role_request
*reply
;
4233 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4234 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4236 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4239 role
= ntohl(nrr
->role
);
4240 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4241 && role
!= NX_ROLE_SLAVE
) {
4242 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4244 /* There's no good error code for this. */
4245 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4248 if (role
== NX_ROLE_MASTER
) {
4249 struct ofconn
*other
;
4251 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4252 if (other
->role
== NX_ROLE_MASTER
) {
4253 other
->role
= NX_ROLE_SLAVE
;
4257 ofconn
->role
= role
;
4259 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
4260 reply
->role
= htonl(role
);
4261 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4267 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4269 const struct nxt_set_flow_format
*msg
4270 = (const struct nxt_set_flow_format
*) oh
;
4273 format
= ntohl(msg
->format
);
4274 if (format
== NXFF_OPENFLOW10
4275 || format
== NXFF_TUN_ID_FROM_COOKIE
4276 || format
== NXFF_NXM
) {
4277 ofconn
->flow_format
= format
;
4280 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4285 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4287 struct ofp_header
*ob
;
4290 /* Currently, everything executes synchronously, so we can just
4291 * immediately send the barrier reply. */
4292 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4293 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4298 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4300 const struct ofp_header
*oh
= msg
->data
;
4301 const struct ofputil_msg_type
*type
;
4304 error
= ofputil_decode_msg_type(oh
, &type
);
4309 switch (ofputil_msg_type_code(type
)) {
4310 /* OpenFlow requests. */
4311 case OFPUTIL_OFPT_ECHO_REQUEST
:
4312 return handle_echo_request(ofconn
, oh
);
4314 case OFPUTIL_OFPT_FEATURES_REQUEST
:
4315 return handle_features_request(ofconn
, oh
);
4317 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
4318 return handle_get_config_request(ofconn
, oh
);
4320 case OFPUTIL_OFPT_SET_CONFIG
:
4321 return handle_set_config(ofconn
, msg
->data
);
4323 case OFPUTIL_OFPT_PACKET_OUT
:
4324 return handle_packet_out(ofconn
, oh
);
4326 case OFPUTIL_OFPT_PORT_MOD
:
4327 return handle_port_mod(ofconn
, oh
);
4329 case OFPUTIL_OFPT_FLOW_MOD
:
4330 return handle_flow_mod(ofconn
, oh
);
4332 case OFPUTIL_OFPT_BARRIER_REQUEST
:
4333 return handle_barrier_request(ofconn
, oh
);
4335 /* OpenFlow replies. */
4336 case OFPUTIL_OFPT_ECHO_REPLY
:
4339 /* Nicira extension requests. */
4340 case OFPUTIL_NXT_STATUS_REQUEST
:
4341 return switch_status_handle_request(
4342 ofconn
->ofproto
->switch_status
, ofconn
->rconn
, oh
);
4344 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE
:
4345 return handle_tun_id_from_cookie(ofconn
, oh
);
4347 case OFPUTIL_NXT_ROLE_REQUEST
:
4348 return handle_role_request(ofconn
, oh
);
4350 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
4351 return handle_nxt_set_flow_format(ofconn
, oh
);
4353 case OFPUTIL_NXT_FLOW_MOD
:
4354 return handle_flow_mod(ofconn
, oh
);
4356 /* OpenFlow statistics requests. */
4357 case OFPUTIL_OFPST_DESC_REQUEST
:
4358 return handle_desc_stats_request(ofconn
, oh
);
4360 case OFPUTIL_OFPST_FLOW_REQUEST
:
4361 return handle_flow_stats_request(ofconn
, oh
);
4363 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
4364 return handle_aggregate_stats_request(ofconn
, oh
);
4366 case OFPUTIL_OFPST_TABLE_REQUEST
:
4367 return handle_table_stats_request(ofconn
, oh
);
4369 case OFPUTIL_OFPST_PORT_REQUEST
:
4370 return handle_port_stats_request(ofconn
, oh
);
4372 case OFPUTIL_OFPST_QUEUE_REQUEST
:
4373 return handle_queue_stats_request(ofconn
, oh
);
4375 /* Nicira extension statistics requests. */
4376 case OFPUTIL_NXST_FLOW_REQUEST
:
4377 return handle_nxst_flow(ofconn
, oh
);
4379 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
4380 return handle_nxst_aggregate(ofconn
, oh
);
4382 case OFPUTIL_INVALID
:
4383 case OFPUTIL_OFPT_HELLO
:
4384 case OFPUTIL_OFPT_ERROR
:
4385 case OFPUTIL_OFPT_FEATURES_REPLY
:
4386 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
4387 case OFPUTIL_OFPT_PACKET_IN
:
4388 case OFPUTIL_OFPT_FLOW_REMOVED
:
4389 case OFPUTIL_OFPT_PORT_STATUS
:
4390 case OFPUTIL_OFPT_BARRIER_REPLY
:
4391 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
4392 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
4393 case OFPUTIL_OFPST_DESC_REPLY
:
4394 case OFPUTIL_OFPST_FLOW_REPLY
:
4395 case OFPUTIL_OFPST_QUEUE_REPLY
:
4396 case OFPUTIL_OFPST_PORT_REPLY
:
4397 case OFPUTIL_OFPST_TABLE_REPLY
:
4398 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
4399 case OFPUTIL_NXT_STATUS_REPLY
:
4400 case OFPUTIL_NXT_ROLE_REPLY
:
4401 case OFPUTIL_NXT_FLOW_REMOVED
:
4402 case OFPUTIL_NXST_FLOW_REPLY
:
4403 case OFPUTIL_NXST_AGGREGATE_REPLY
:
4405 if (VLOG_IS_WARN_ENABLED()) {
4406 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4407 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4410 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
4411 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
4413 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4419 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4421 int error
= handle_openflow__(ofconn
, ofp_msg
);
4423 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4425 COVERAGE_INC(ofproto_recv_openflow
);
4429 handle_miss_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4431 struct facet
*facet
;
4434 /* Obtain in_port and tun_id, at least. */
4435 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4437 /* Set header pointers in 'flow'. */
4438 flow_extract(upcall
->packet
, flow
.tun_id
, flow
.in_port
, &flow
);
4440 if (p
->ofhooks
->special_cb
4441 && !p
->ofhooks
->special_cb(&flow
, upcall
->packet
, p
->aux
)) {
4442 ofpbuf_delete(upcall
->packet
);
4446 /* Check with in-band control to see if this packet should be sent
4447 * to the local port regardless of the flow table. */
4448 if (in_band_msg_in_hook(p
->in_band
, &flow
, upcall
->packet
)) {
4449 struct ofpbuf odp_actions
;
4451 ofpbuf_init(&odp_actions
, 32);
4452 nl_msg_put_u32(&odp_actions
, ODP_ACTION_ATTR_OUTPUT
, ODPP_LOCAL
);
4453 dpif_execute(p
->dpif
, odp_actions
.data
, odp_actions
.size
,
4455 ofpbuf_uninit(&odp_actions
);
4458 facet
= facet_lookup_valid(p
, &flow
);
4460 struct rule
*rule
= rule_lookup(p
, &flow
);
4462 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4463 struct ofport
*port
= get_port(p
, flow
.in_port
);
4465 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4466 COVERAGE_INC(ofproto_no_packet_in
);
4467 /* XXX install 'drop' flow entry */
4468 ofpbuf_delete(upcall
->packet
);
4472 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4476 COVERAGE_INC(ofproto_packet_in
);
4477 send_packet_in(p
, upcall
, &flow
, false);
4481 facet
= facet_create(p
, rule
, &flow
, upcall
->packet
);
4482 } else if (!facet
->may_install
) {
4483 /* The facet is not installable, that is, we need to process every
4484 * packet, so process the current packet's actions into 'facet'. */
4485 facet_make_actions(p
, facet
, upcall
->packet
);
4488 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4490 * Extra-special case for fail-open mode.
4492 * We are in fail-open mode and the packet matched the fail-open rule,
4493 * but we are connected to a controller too. We should send the packet
4494 * up to the controller in the hope that it will try to set up a flow
4495 * and thereby allow us to exit fail-open.
4497 * See the top-level comment in fail-open.c for more information.
4499 send_packet_in(p
, upcall
, &flow
, true);
4502 facet_execute(p
, facet
, upcall
->packet
);
4503 facet_install(p
, facet
, false);
4507 handle_upcall(struct ofproto
*p
, struct dpif_upcall
*upcall
)
4511 switch (upcall
->type
) {
4512 case DPIF_UC_ACTION
:
4513 COVERAGE_INC(ofproto_ctlr_action
);
4514 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4515 send_packet_in(p
, upcall
, &flow
, false);
4518 case DPIF_UC_SAMPLE
:
4520 odp_flow_key_to_flow(upcall
->key
, upcall
->key_len
, &flow
);
4521 ofproto_sflow_received(p
->sflow
, upcall
, &flow
);
4523 ofpbuf_delete(upcall
->packet
);
4527 handle_miss_upcall(p
, upcall
);
4530 case DPIF_N_UC_TYPES
:
4532 VLOG_WARN_RL(&rl
, "upcall has unexpected type %"PRIu32
, upcall
->type
);
4537 /* Flow expiration. */
4539 static int ofproto_dp_max_idle(const struct ofproto
*);
4540 static void ofproto_update_stats(struct ofproto
*);
4541 static void rule_expire(struct ofproto
*, struct rule
*);
4542 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4544 /* This function is called periodically by ofproto_run(). Its job is to
4545 * collect updates for the flows that have been installed into the datapath,
4546 * most importantly when they last were used, and then use that information to
4547 * expire flows that have not been used recently.
4549 * Returns the number of milliseconds after which it should be called again. */
4551 ofproto_expire(struct ofproto
*ofproto
)
4553 struct rule
*rule
, *next_rule
;
4554 struct cls_cursor cursor
;
4557 /* Update stats for each flow in the datapath. */
4558 ofproto_update_stats(ofproto
);
4560 /* Expire facets that have been idle too long. */
4561 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4562 ofproto_expire_facets(ofproto
, dp_max_idle
);
4564 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4565 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4566 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4567 rule_expire(ofproto
, rule
);
4570 /* Let the hook know that we're at a stable point: all outstanding data
4571 * in existing flows has been accounted to the account_cb. Thus, the
4572 * hook can now reasonably do operations that depend on having accurate
4573 * flow volume accounting (currently, that's just bond rebalancing). */
4574 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4575 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4578 return MIN(dp_max_idle
, 1000);
4581 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4583 * This function also pushes statistics updates to rules which each facet
4584 * resubmits into. Generally these statistics will be accurate. However, if a
4585 * facet changes the rule it resubmits into at some time in between
4586 * ofproto_update_stats() runs, it is possible that statistics accrued to the
4587 * old rule will be incorrectly attributed to the new rule. This could be
4588 * avoided by calling ofproto_update_stats() whenever rules are created or
4589 * deleted. However, the performance impact of making so many calls to the
4590 * datapath do not justify the benefit of having perfectly accurate statistics.
4593 ofproto_update_stats(struct ofproto
*p
)
4595 const struct dpif_flow_stats
*stats
;
4596 struct dpif_flow_dump dump
;
4597 const struct nlattr
*key
;
4600 dpif_flow_dump_start(&dump
, p
->dpif
);
4601 while (dpif_flow_dump_next(&dump
, &key
, &key_len
, NULL
, NULL
, &stats
)) {
4602 struct facet
*facet
;
4605 if (odp_flow_key_to_flow(key
, key_len
, &flow
)) {
4609 odp_flow_key_format(key
, key_len
, &s
);
4610 VLOG_WARN_RL(&rl
, "failed to convert ODP flow key to flow: %s",
4616 facet
= facet_find(p
, &flow
);
4618 if (facet
&& facet
->installed
) {
4620 if (stats
->n_packets
>= facet
->dp_packet_count
) {
4621 facet
->packet_count
+= stats
->n_packets
- facet
->dp_packet_count
;
4623 VLOG_WARN_RL(&rl
, "unexpected packet count from the datapath");
4626 if (stats
->n_bytes
>= facet
->dp_byte_count
) {
4627 facet
->byte_count
+= stats
->n_bytes
- facet
->dp_byte_count
;
4629 VLOG_WARN_RL(&rl
, "unexpected byte count from datapath");
4632 facet
->dp_packet_count
= stats
->n_packets
;
4633 facet
->dp_byte_count
= stats
->n_bytes
;
4635 facet_update_time(p
, facet
, stats
->used
);
4636 facet_account(p
, facet
, stats
->n_bytes
);
4637 facet_push_stats(p
, facet
);
4639 /* There's a flow in the datapath that we know nothing about.
4641 COVERAGE_INC(ofproto_unexpected_rule
);
4642 dpif_flow_del(p
->dpif
, key
, key_len
, NULL
);
4645 dpif_flow_dump_done(&dump
);
4648 /* Calculates and returns the number of milliseconds of idle time after which
4649 * facets should expire from the datapath and we should fold their statistics
4650 * into their parent rules in userspace. */
4652 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4655 * Idle time histogram.
4657 * Most of the time a switch has a relatively small number of facets. When
4658 * this is the case we might as well keep statistics for all of them in
4659 * userspace and to cache them in the kernel datapath for performance as
4662 * As the number of facets increases, the memory required to maintain
4663 * statistics about them in userspace and in the kernel becomes
4664 * significant. However, with a large number of facets it is likely that
4665 * only a few of them are "heavy hitters" that consume a large amount of
4666 * bandwidth. At this point, only heavy hitters are worth caching in the
4667 * kernel and maintaining in userspaces; other facets we can discard.
4669 * The technique used to compute the idle time is to build a histogram with
4670 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4671 * that is installed in the kernel gets dropped in the appropriate bucket.
4672 * After the histogram has been built, we compute the cutoff so that only
4673 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4674 * cached. At least the most-recently-used bucket of facets is kept, so
4675 * actually an arbitrary number of facets can be kept in any given
4676 * expiration run (though the next run will delete most of those unless
4677 * they receive additional data).
4679 * This requires a second pass through the facets, in addition to the pass
4680 * made by ofproto_update_stats(), because the former function never looks
4681 * at uninstallable facets.
4683 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4684 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4685 int buckets
[N_BUCKETS
] = { 0 };
4686 struct facet
*facet
;
4691 total
= hmap_count(&ofproto
->facets
);
4692 if (total
<= 1000) {
4693 return N_BUCKETS
* BUCKET_WIDTH
;
4696 /* Build histogram. */
4698 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4699 long long int idle
= now
- facet
->used
;
4700 int bucket
= (idle
<= 0 ? 0
4701 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4702 : (unsigned int) idle
/ BUCKET_WIDTH
);
4706 /* Find the first bucket whose flows should be expired. */
4707 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4708 if (buckets
[bucket
]) {
4711 subtotal
+= buckets
[bucket
++];
4712 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4717 if (VLOG_IS_DBG_ENABLED()) {
4721 ds_put_cstr(&s
, "keep");
4722 for (i
= 0; i
< N_BUCKETS
; i
++) {
4724 ds_put_cstr(&s
, ", drop");
4727 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4730 VLOG_INFO("%s: %s (msec:count)",
4731 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4735 return bucket
* BUCKET_WIDTH
;
4739 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4741 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4742 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4743 struct ofexpired expired
;
4745 if (facet
->installed
) {
4746 struct dpif_flow_stats stats
;
4748 facet_put__(ofproto
, facet
, facet
->actions
, facet
->actions_len
,
4750 facet_update_stats(ofproto
, facet
, &stats
);
4753 expired
.flow
= facet
->flow
;
4754 expired
.packet_count
= facet
->packet_count
;
4755 expired
.byte_count
= facet
->byte_count
;
4756 expired
.used
= facet
->used
;
4757 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4762 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4764 long long int cutoff
= time_msec() - dp_max_idle
;
4765 struct facet
*facet
, *next_facet
;
4767 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4768 facet_active_timeout(ofproto
, facet
);
4769 if (facet
->used
< cutoff
) {
4770 facet_remove(ofproto
, facet
);
4775 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4776 * then delete it entirely. */
4778 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4780 struct facet
*facet
, *next_facet
;
4784 /* Has 'rule' expired? */
4786 if (rule
->hard_timeout
4787 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4788 reason
= OFPRR_HARD_TIMEOUT
;
4789 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4790 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4791 reason
= OFPRR_IDLE_TIMEOUT
;
4796 COVERAGE_INC(ofproto_expired
);
4798 /* Update stats. (This is a no-op if the rule expired due to an idle
4799 * timeout, because that only happens when the rule has no facets left.) */
4800 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4801 facet_remove(ofproto
, facet
);
4804 /* Get rid of the rule. */
4805 if (!rule_is_hidden(rule
)) {
4806 rule_send_removed(ofproto
, rule
, reason
);
4808 rule_remove(ofproto
, rule
);
4811 static struct ofpbuf
*
4812 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4815 struct ofp_flow_removed
*ofr
;
4818 ofr
= make_openflow_xid(sizeof *ofr
, OFPT_FLOW_REMOVED
, htonl(0), &buf
);
4819 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofr
->match
,
4820 rule
->flow_cookie
, &ofr
->cookie
);
4821 ofr
->priority
= htons(rule
->cr
.priority
);
4822 ofr
->reason
= reason
;
4823 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4824 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4825 ofr
->packet_count
= htonll(rule
->packet_count
);
4826 ofr
->byte_count
= htonll(rule
->byte_count
);
4831 static struct ofpbuf
*
4832 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4834 struct nx_flow_removed
*nfr
;
4838 make_nxmsg_xid(sizeof *nfr
, NXT_FLOW_REMOVED
, htonl(0), &buf
);
4839 match_len
= nx_put_match(buf
, &rule
->cr
);
4842 nfr
->cookie
= rule
->flow_cookie
;
4843 nfr
->priority
= htons(rule
->cr
.priority
);
4844 nfr
->reason
= reason
;
4845 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4846 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4847 nfr
->match_len
= htons(match_len
);
4848 nfr
->packet_count
= htonll(rule
->packet_count
);
4849 nfr
->byte_count
= htonll(rule
->byte_count
);
4855 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4857 struct ofconn
*ofconn
;
4859 if (!rule
->send_flow_removed
) {
4863 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4866 if (!rconn_is_connected(ofconn
->rconn
)
4867 || !ofconn_receives_async_msgs(ofconn
)) {
4871 msg
= (ofconn
->flow_format
== NXFF_NXM
4872 ? compose_nx_flow_removed(rule
, reason
)
4873 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4875 /* Account flow expirations under ofconn->reply_counter, the counter
4876 * for replies to OpenFlow requests. That works because preventing
4877 * OpenFlow requests from being processed also prevents new flows from
4878 * being added (and expiring). (It also prevents processing OpenFlow
4879 * requests that would not add new flows, so it is imperfect.) */
4880 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4884 /* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'.
4885 * The returned statistics include statistics for all of 'rule''s facets. */
4887 rule_get_stats(const struct rule
*rule
, uint64_t *packets
, uint64_t *bytes
)
4890 struct facet
*facet
;
4892 /* Start from historical data for 'rule' itself that are no longer tracked
4893 * in facets. This counts, for example, facets that have expired. */
4894 p
= rule
->packet_count
;
4895 b
= rule
->byte_count
;
4897 /* Add any statistics that are tracked by facets. This includes
4898 * statistical data recently updated by ofproto_update_stats() as well as
4899 * stats for packets that were executed "by hand" via dpif_execute(). */
4900 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
4901 p
+= facet
->packet_count
;
4902 b
+= facet
->byte_count
;
4909 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
4911 do_send_packet_in(struct ofpbuf
*ofp_packet_in
, void *ofconn_
)
4913 struct ofconn
*ofconn
= ofconn_
;
4915 rconn_send_with_limit(ofconn
->rconn
, ofp_packet_in
,
4916 ofconn
->packet_in_counter
, 100);
4919 /* Takes 'upcall', whose packet has the flow specified by 'flow', composes an
4920 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
4921 * scheduler for sending.
4923 * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
4924 * Otherwise, ownership is transferred to this function. */
4926 schedule_packet_in(struct ofconn
*ofconn
, struct dpif_upcall
*upcall
,
4927 const struct flow
*flow
, bool clone
)
4929 enum { OPI_SIZE
= offsetof(struct ofp_packet_in
, data
) };
4930 struct ofproto
*ofproto
= ofconn
->ofproto
;
4931 struct ofp_packet_in
*opi
;
4932 int total_len
, send_len
;
4933 struct ofpbuf
*packet
;
4937 /* Get OpenFlow buffer_id. */
4938 if (upcall
->type
== DPIF_UC_ACTION
) {
4939 buffer_id
= UINT32_MAX
;
4940 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4941 buffer_id
= pktbuf_get_null();
4942 } else if (!ofconn
->pktbuf
) {
4943 buffer_id
= UINT32_MAX
;
4945 buffer_id
= pktbuf_save(ofconn
->pktbuf
, upcall
->packet
, flow
->in_port
);
4948 /* Figure out how much of the packet to send. */
4949 total_len
= send_len
= upcall
->packet
->size
;
4950 if (buffer_id
!= UINT32_MAX
) {
4951 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4953 if (upcall
->type
== DPIF_UC_ACTION
) {
4954 send_len
= MIN(send_len
, upcall
->userdata
);
4957 /* Copy or steal buffer for OFPT_PACKET_IN. */
4959 packet
= ofpbuf_clone_data_with_headroom(upcall
->packet
->data
,
4960 send_len
, OPI_SIZE
);
4962 packet
= upcall
->packet
;
4963 packet
->size
= send_len
;
4966 /* Add OFPT_PACKET_IN. */
4967 opi
= ofpbuf_push_zeros(packet
, OPI_SIZE
);
4968 opi
->header
.version
= OFP_VERSION
;
4969 opi
->header
.type
= OFPT_PACKET_IN
;
4970 opi
->total_len
= htons(total_len
);
4971 opi
->in_port
= htons(odp_port_to_ofp_port(flow
->in_port
));
4972 opi
->reason
= upcall
->type
== DPIF_UC_MISS
? OFPR_NO_MATCH
: OFPR_ACTION
;
4973 opi
->buffer_id
= htonl(buffer_id
);
4974 update_openflow_length(packet
);
4976 /* Hand over to packet scheduler. It might immediately call into
4977 * do_send_packet_in() or it might buffer it for a while (until a later
4978 * call to pinsched_run()). */
4979 idx
= upcall
->type
== DPIF_UC_MISS
? 0 : 1;
4980 pinsched_send(ofconn
->schedulers
[idx
], flow
->in_port
,
4981 packet
, do_send_packet_in
, ofconn
);
4984 /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
4985 * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
4986 * their individual configurations.
4988 * Takes ownership of 'packet'. */
4990 send_packet_in(struct ofproto
*ofproto
, struct dpif_upcall
*upcall
,
4991 const struct flow
*flow
, bool clone
)
4993 struct ofconn
*ofconn
, *prev
;
4996 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4997 if (ofconn_receives_async_msgs(ofconn
)) {
4999 schedule_packet_in(prev
, upcall
, flow
, true);
5005 schedule_packet_in(prev
, upcall
, flow
, clone
);
5006 } else if (!clone
) {
5007 ofpbuf_delete(upcall
->packet
);
5012 pick_datapath_id(const struct ofproto
*ofproto
)
5014 const struct ofport
*port
;
5016 port
= get_port(ofproto
, ODPP_LOCAL
);
5018 uint8_t ea
[ETH_ADDR_LEN
];
5021 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5023 return eth_addr_to_uint64(ea
);
5025 VLOG_WARN("could not get MAC address for %s (%s)",
5026 netdev_get_name(port
->netdev
), strerror(error
));
5028 return ofproto
->fallback_dpid
;
5032 pick_fallback_dpid(void)
5034 uint8_t ea
[ETH_ADDR_LEN
];
5035 eth_addr_nicira_random(ea
);
5036 return eth_addr_to_uint64(ea
);
5040 ofproto_unixctl_list(struct unixctl_conn
*conn
, const char *arg OVS_UNUSED
,
5041 void *aux OVS_UNUSED
)
5043 const struct shash_node
*node
;
5047 SHASH_FOR_EACH (node
, &all_ofprotos
) {
5048 ds_put_format(&results
, "%s\n", node
->name
);
5050 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
5051 ds_destroy(&results
);
5054 struct ofproto_trace
{
5055 struct action_xlate_ctx ctx
;
5061 trace_format_rule(struct ds
*result
, int level
, const struct rule
*rule
)
5063 ds_put_char_multiple(result
, '\t', level
);
5065 ds_put_cstr(result
, "No match\n");
5069 ds_put_format(result
, "Rule: cookie=%#"PRIx64
" ",
5070 ntohll(rule
->flow_cookie
));
5071 cls_rule_format(&rule
->cr
, result
);
5072 ds_put_char(result
, '\n');
5074 ds_put_char_multiple(result
, '\t', level
);
5075 ds_put_cstr(result
, "OpenFlow ");
5076 ofp_print_actions(result
, (const struct ofp_action_header
*) rule
->actions
,
5077 rule
->n_actions
* sizeof *rule
->actions
);
5078 ds_put_char(result
, '\n');
5082 trace_format_flow(struct ds
*result
, int level
, const char *title
,
5083 struct ofproto_trace
*trace
)
5085 ds_put_char_multiple(result
, '\t', level
);
5086 ds_put_format(result
, "%s: ", title
);
5087 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
5088 ds_put_cstr(result
, "unchanged");
5090 flow_format(result
, &trace
->ctx
.flow
);
5091 trace
->flow
= trace
->ctx
.flow
;
5093 ds_put_char(result
, '\n');
5097 trace_resubmit(struct action_xlate_ctx
*ctx
, struct rule
*rule
)
5099 struct ofproto_trace
*trace
= CONTAINER_OF(ctx
, struct ofproto_trace
, ctx
);
5100 struct ds
*result
= trace
->result
;
5102 ds_put_char(result
, '\n');
5103 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
5104 trace_format_rule(result
, ctx
->recurse
+ 1, rule
);
5108 ofproto_unixctl_trace(struct unixctl_conn
*conn
, const char *args_
,
5109 void *aux OVS_UNUSED
)
5111 char *dpname
, *in_port_s
, *tun_id_s
, *packet_s
;
5112 char *args
= xstrdup(args_
);
5113 char *save_ptr
= NULL
;
5114 struct ofproto
*ofproto
;
5115 struct ofpbuf packet
;
5123 ofpbuf_init(&packet
, strlen(args
) / 2);
5126 dpname
= strtok_r(args
, " ", &save_ptr
);
5127 tun_id_s
= strtok_r(NULL
, " ", &save_ptr
);
5128 in_port_s
= strtok_r(NULL
, " ", &save_ptr
);
5129 packet_s
= strtok_r(NULL
, "", &save_ptr
); /* Get entire rest of line. */
5130 if (!dpname
|| !in_port_s
|| !packet_s
) {
5131 unixctl_command_reply(conn
, 501, "Bad command syntax");
5135 ofproto
= shash_find_data(&all_ofprotos
, dpname
);
5137 unixctl_command_reply(conn
, 501, "Unknown ofproto (use ofproto/list "
5142 tun_id
= htonll(strtoull(tun_id_s
, NULL
, 10));
5143 in_port
= ofp_port_to_odp_port(atoi(in_port_s
));
5145 packet_s
= ofpbuf_put_hex(&packet
, packet_s
, NULL
);
5146 packet_s
+= strspn(packet_s
, " ");
5147 if (*packet_s
!= '\0') {
5148 unixctl_command_reply(conn
, 501, "Trailing garbage in command");
5151 if (packet
.size
< ETH_HEADER_LEN
) {
5152 unixctl_command_reply(conn
, 501, "Packet data too short for Ethernet");
5156 ds_put_cstr(&result
, "Packet: ");
5157 s
= ofp_packet_to_string(packet
.data
, packet
.size
, packet
.size
);
5158 ds_put_cstr(&result
, s
);
5161 flow_extract(&packet
, tun_id
, in_port
, &flow
);
5162 ds_put_cstr(&result
, "Flow: ");
5163 flow_format(&result
, &flow
);
5164 ds_put_char(&result
, '\n');
5166 rule
= rule_lookup(ofproto
, &flow
);
5167 trace_format_rule(&result
, 0, rule
);
5169 struct ofproto_trace trace
;
5170 struct ofpbuf
*odp_actions
;
5172 trace
.result
= &result
;
5174 action_xlate_ctx_init(&trace
.ctx
, ofproto
, &flow
, &packet
);
5175 trace
.ctx
.resubmit_hook
= trace_resubmit
;
5176 odp_actions
= xlate_actions(&trace
.ctx
,
5177 rule
->actions
, rule
->n_actions
);
5179 ds_put_char(&result
, '\n');
5180 trace_format_flow(&result
, 0, "Final flow", &trace
);
5181 ds_put_cstr(&result
, "Datapath actions: ");
5182 format_odp_actions(&result
, odp_actions
->data
, odp_actions
->size
);
5183 ofpbuf_delete(odp_actions
);
5186 unixctl_command_reply(conn
, 200, ds_cstr(&result
));
5189 ds_destroy(&result
);
5190 ofpbuf_uninit(&packet
);
5195 ofproto_unixctl_init(void)
5197 static bool registered
;
5203 unixctl_command_register("ofproto/list", ofproto_unixctl_list
, NULL
);
5204 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace
, NULL
);
5208 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5209 struct ofpbuf
*odp_actions
, tag_type
*tags
,
5210 uint16_t *nf_output_iface
, void *ofproto_
)
5212 struct ofproto
*ofproto
= ofproto_
;
5215 /* Drop frames for reserved multicast addresses. */
5216 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5220 /* Learn source MAC (but don't try to learn from revalidation). */
5221 if (packet
!= NULL
) {
5222 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
5224 GRAT_ARP_LOCK_NONE
);
5226 /* The log messages here could actually be useful in debugging,
5227 * so keep the rate limit relatively high. */
5228 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5229 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5230 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5231 ofproto_revalidate(ofproto
, rev_tag
);
5235 /* Determine output port. */
5236 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
5239 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5240 nf_output_iface
, odp_actions
);
5241 } else if (out_port
!= flow
->in_port
) {
5242 nl_msg_put_u32(odp_actions
, ODP_ACTION_ATTR_OUTPUT
, out_port
);
5243 *nf_output_iface
= out_port
;
5251 static const struct ofhooks default_ofhooks
= {
5252 default_normal_ofhook_cb
,