2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
3 * Copyright (c) 2009 InMon Corp.
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.
19 #include "ofproto-dpif-sflow.h"
21 #include <sys/resource.h>
22 #include <sys/socket.h>
25 #include "collectors.h"
29 #include "openvswitch/hmap.h"
32 #include "openvswitch/ofpbuf.h"
35 #include "openvswitch/poll-loop.h"
36 #include "ovs-router.h"
37 #include "route-table.h"
38 #include "sflow_api.h"
39 #include "socket-util.h"
41 #include "openvswitch/vlog.h"
42 #include "lib/odp-util.h"
43 #include "lib/unaligned.h"
44 #include "ofproto-provider.h"
47 VLOG_DEFINE_THIS_MODULE(sflow
);
49 static struct ovs_mutex mutex
;
51 /* This global var is used to determine which sFlow
52 sub-agent should send the datapath counters. */
53 #define SFLOW_GC_SUBID_UNCLAIMED (uint32_t)-1
54 static uint32_t sflow_global_counters_subid
= SFLOW_GC_SUBID_UNCLAIMED
;
57 * The enum dpif_sflow_tunnel_type is to declare the types supported
59 enum dpif_sflow_tunnel_type
{
60 DPIF_SFLOW_TUNNEL_UNKNOWN
= 0,
61 DPIF_SFLOW_TUNNEL_VXLAN
,
62 DPIF_SFLOW_TUNNEL_GRE
,
63 DPIF_SFLOW_TUNNEL_LISP
,
64 DPIF_SFLOW_TUNNEL_GENEVE
67 struct dpif_sflow_port
{
68 struct hmap_node hmap_node
; /* In struct dpif_sflow's "ports" hmap. */
69 SFLDataSource_instance dsi
; /* sFlow library's notion of port number. */
70 struct ofport
*ofport
; /* To retrive port stats. */
72 enum dpif_sflow_tunnel_type tunnel_type
;
76 struct collectors
*collectors
;
77 SFLAgent
*sflow_agent
;
78 struct ofproto_sflow_options
*options
;
80 size_t n_flood
, n_all
;
81 struct hmap ports
; /* Contains "struct dpif_sflow_port"s. */
83 struct ovs_refcount ref_cnt
;
86 static void dpif_sflow_del_port__(struct dpif_sflow
*,
87 struct dpif_sflow_port
*);
89 #define RECEIVER_INDEX 1
91 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
94 ofproto_sflow_options_equal(const struct ofproto_sflow_options
*a
,
95 const struct ofproto_sflow_options
*b
)
97 return (sset_equals(&a
->targets
, &b
->targets
)
98 && a
->sampling_rate
== b
->sampling_rate
99 && a
->polling_interval
== b
->polling_interval
100 && a
->header_len
== b
->header_len
101 && a
->sub_id
== b
->sub_id
102 && nullable_string_is_equal(a
->agent_device
, b
->agent_device
)
103 && nullable_string_is_equal(a
->control_ip
, b
->control_ip
));
106 static struct ofproto_sflow_options
*
107 ofproto_sflow_options_clone(const struct ofproto_sflow_options
*old
)
109 struct ofproto_sflow_options
*new = xmemdup(old
, sizeof *old
);
110 sset_clone(&new->targets
, &old
->targets
);
111 new->agent_device
= nullable_xstrdup(old
->agent_device
);
112 new->control_ip
= nullable_xstrdup(old
->control_ip
);
117 ofproto_sflow_options_destroy(struct ofproto_sflow_options
*options
)
120 sset_destroy(&options
->targets
);
121 free(options
->agent_device
);
122 free(options
->control_ip
);
127 /* sFlow library callback to allocate memory. */
129 sflow_agent_alloc_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
132 return xzalloc(bytes
);
135 /* sFlow library callback to free memory. */
137 sflow_agent_free_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
144 /* sFlow library callback to report error. */
146 sflow_agent_error_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
149 VLOG_WARN("sFlow agent error: %s", msg
);
152 /* sFlow library callback to send datagram. */
154 sflow_agent_send_packet_cb(void *ds_
, SFLAgent
*agent OVS_UNUSED
,
155 SFLReceiver
*receiver OVS_UNUSED
, u_char
*pkt
,
158 struct dpif_sflow
*ds
= ds_
;
159 collectors_send(ds
->collectors
, pkt
, pktLen
);
162 static struct dpif_sflow_port
*
163 dpif_sflow_find_port(const struct dpif_sflow
*ds
, odp_port_t odp_port
)
166 struct dpif_sflow_port
*dsp
;
168 HMAP_FOR_EACH_IN_BUCKET (dsp
, hmap_node
, hash_odp_port(odp_port
),
170 if (dsp
->odp_port
== odp_port
) {
177 /* Call to get the datapath stats. Modeled after the dpctl utility.
179 * It might be more efficient for this module to be given a handle it can use
180 * to get these stats more efficiently, but this is only going to be called
181 * once every 20-30 seconds. Return number of datapaths found (normally expect
184 sflow_get_dp_stats(struct dpif_sflow
*ds OVS_UNUSED
,
185 struct dpif_dp_stats
*dp_totals
)
191 memset(dp_totals
, 0, sizeof *dp_totals
);
193 dp_enumerate_types(&types
);
194 SSET_FOR_EACH (type
, &types
) {
198 if (dp_enumerate_names(type
, &names
) == 0) {
199 SSET_FOR_EACH (name
, &names
) {
201 if (dpif_open(name
, type
, &dpif
) == 0) {
202 struct dpif_dp_stats dp_stats
;
203 if (dpif_get_dp_stats(dpif
, &dp_stats
) == 0) {
205 dp_totals
->n_hit
+= dp_stats
.n_hit
;
206 dp_totals
->n_missed
+= dp_stats
.n_missed
;
207 dp_totals
->n_lost
+= dp_stats
.n_lost
;
208 dp_totals
->n_flows
+= dp_stats
.n_flows
;
209 dp_totals
->n_mask_hit
+= dp_stats
.n_mask_hit
;
210 dp_totals
->n_masks
+= dp_stats
.n_masks
;
215 sset_destroy(&names
);
218 sset_destroy(&types
);
222 /* If there are multiple bridges defined then we need some
223 minimal artibration to decide which one should send the
224 global counters. This function allows each sub-agent to
225 ask if he should do it or not. */
227 sflow_global_counters_subid_test(uint32_t subid
)
230 if (sflow_global_counters_subid
== SFLOW_GC_SUBID_UNCLAIMED
) {
231 /* The role is up for grabs. */
232 sflow_global_counters_subid
= subid
;
234 return (sflow_global_counters_subid
== subid
);
238 sflow_global_counters_subid_clear(uint32_t subid
)
241 if (sflow_global_counters_subid
== subid
) {
242 /* The sub-agent that was sending global counters
243 is going away, so reset to allow another
245 sflow_global_counters_subid
= SFLOW_GC_SUBID_UNCLAIMED
;
250 sflow_agent_get_global_counters(void *ds_
, SFLPoller
*poller
,
251 SFL_COUNTERS_SAMPLE_TYPE
*cs
)
254 struct dpif_sflow
*ds
= ds_
;
255 SFLCounters_sample_element dp_elem
, res_elem
;
256 struct dpif_dp_stats dp_totals
;
259 if (!sflow_global_counters_subid_test(poller
->agent
->subId
)) {
260 /* Another sub-agent is currently responsible for this. */
265 if (sflow_get_dp_stats(ds
, &dp_totals
)) {
266 dp_elem
.tag
= SFLCOUNTERS_OVSDP
;
267 dp_elem
.counterBlock
.ovsdp
.n_hit
= dp_totals
.n_hit
;
268 dp_elem
.counterBlock
.ovsdp
.n_missed
= dp_totals
.n_missed
;
269 dp_elem
.counterBlock
.ovsdp
.n_lost
= dp_totals
.n_lost
;
270 dp_elem
.counterBlock
.ovsdp
.n_mask_hit
= dp_totals
.n_mask_hit
;
271 dp_elem
.counterBlock
.ovsdp
.n_flows
= dp_totals
.n_flows
;
272 dp_elem
.counterBlock
.ovsdp
.n_masks
= dp_totals
.n_masks
;
273 SFLADD_ELEMENT(cs
, &dp_elem
);
277 getrusage(RUSAGE_SELF
, &usage
);
278 res_elem
.tag
= SFLCOUNTERS_APP_RESOURCES
;
279 res_elem
.counterBlock
.appResources
.user_time
280 = timeval_to_msec(&usage
.ru_utime
);
281 res_elem
.counterBlock
.appResources
.system_time
282 = timeval_to_msec(&usage
.ru_stime
);
283 res_elem
.counterBlock
.appResources
.mem_used
= (usage
.ru_maxrss
* 1024);
284 SFL_UNDEF_GAUGE(res_elem
.counterBlock
.appResources
.mem_max
);
285 SFL_UNDEF_GAUGE(res_elem
.counterBlock
.appResources
.fd_open
);
286 SFL_UNDEF_GAUGE(res_elem
.counterBlock
.appResources
.fd_max
);
287 SFL_UNDEF_GAUGE(res_elem
.counterBlock
.appResources
.conn_open
);
288 SFL_UNDEF_GAUGE(res_elem
.counterBlock
.appResources
.conn_max
);
290 SFLADD_ELEMENT(cs
, &res_elem
);
291 sfl_poller_writeCountersSample(poller
, cs
);
295 sflow_agent_get_counters(void *ds_
, SFLPoller
*poller
,
296 SFL_COUNTERS_SAMPLE_TYPE
*cs
)
299 struct dpif_sflow
*ds
= ds_
;
300 SFLCounters_sample_element elem
, lacp_elem
, of_elem
, name_elem
;
301 SFLCounters_sample_element eth_elem
;
302 enum netdev_features current
;
303 struct dpif_sflow_port
*dsp
;
304 SFLIf_counters
*counters
;
305 SFLEthernet_counters
* eth_counters
;
306 struct netdev_stats stats
;
307 enum netdev_flags flags
;
308 struct lacp_slave_stats lacp_stats
;
311 dsp
= dpif_sflow_find_port(ds
, u32_to_odp(poller
->bridgePort
));
316 elem
.tag
= SFLCOUNTERS_GENERIC
;
317 counters
= &elem
.counterBlock
.generic
;
318 counters
->ifIndex
= SFL_DS_INDEX(poller
->dsi
);
319 counters
->ifType
= 6;
320 if (!netdev_get_features(dsp
->ofport
->netdev
, ¤t
, NULL
, NULL
, NULL
)) {
321 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
322 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
323 counters
->ifSpeed
= netdev_features_to_bps(current
, 0);
324 counters
->ifDirection
= (netdev_features_is_full_duplex(current
)
327 counters
->ifSpeed
= 100000000;
328 counters
->ifDirection
= 0;
330 if (!netdev_get_flags(dsp
->ofport
->netdev
, &flags
) && flags
& NETDEV_UP
) {
331 counters
->ifStatus
= 1; /* ifAdminStatus up. */
332 if (netdev_get_carrier(dsp
->ofport
->netdev
)) {
333 counters
->ifStatus
|= 2; /* ifOperStatus us. */
336 counters
->ifStatus
= 0; /* Down. */
340 1. Is the multicast counter filled in?
341 2. Does the multicast counter include broadcasts?
342 3. Does the rx_packets counter include multicasts/broadcasts?
344 ofproto_port_get_stats(dsp
->ofport
, &stats
);
345 counters
->ifInOctets
= stats
.rx_bytes
;
346 counters
->ifInUcastPkts
= stats
.rx_packets
;
347 counters
->ifInMulticastPkts
= stats
.multicast
;
348 counters
->ifInBroadcastPkts
= stats
.rx_broadcast_packets
;
349 counters
->ifInDiscards
= stats
.rx_dropped
;
350 counters
->ifInErrors
= stats
.rx_errors
;
351 counters
->ifInUnknownProtos
= -1;
352 counters
->ifOutOctets
= stats
.tx_bytes
;
353 counters
->ifOutUcastPkts
= stats
.tx_packets
;
354 counters
->ifOutMulticastPkts
= stats
.tx_multicast_packets
;
355 counters
->ifOutBroadcastPkts
= stats
.tx_broadcast_packets
;
356 counters
->ifOutDiscards
= stats
.tx_dropped
;
357 counters
->ifOutErrors
= stats
.tx_errors
;
358 counters
->ifPromiscuousMode
= 0;
360 SFLADD_ELEMENT(cs
, &elem
);
362 /* Include LACP counters and identifiers if this port is part of a LAG. */
363 if (ofproto_port_get_lacp_stats(dsp
->ofport
, &lacp_stats
) == 0) {
364 memset(&lacp_elem
, 0, sizeof lacp_elem
);
365 lacp_elem
.tag
= SFLCOUNTERS_LACP
;
366 lacp_elem
.counterBlock
.lacp
.actorSystemID
=
367 lacp_stats
.dot3adAggPortActorSystemID
;
368 lacp_elem
.counterBlock
.lacp
.partnerSystemID
=
369 lacp_stats
.dot3adAggPortPartnerOperSystemID
;
370 lacp_elem
.counterBlock
.lacp
.attachedAggID
=
371 lacp_stats
.dot3adAggPortAttachedAggID
;
372 lacp_elem
.counterBlock
.lacp
.portState
.v
.actorAdmin
=
373 lacp_stats
.dot3adAggPortActorAdminState
;
374 lacp_elem
.counterBlock
.lacp
.portState
.v
.actorOper
=
375 lacp_stats
.dot3adAggPortActorOperState
;
376 lacp_elem
.counterBlock
.lacp
.portState
.v
.partnerAdmin
=
377 lacp_stats
.dot3adAggPortPartnerAdminState
;
378 lacp_elem
.counterBlock
.lacp
.portState
.v
.partnerOper
=
379 lacp_stats
.dot3adAggPortPartnerOperState
;
380 lacp_elem
.counterBlock
.lacp
.LACPDUsRx
=
381 lacp_stats
.dot3adAggPortStatsLACPDUsRx
;
382 SFL_UNDEF_COUNTER(lacp_elem
.counterBlock
.lacp
.markerPDUsRx
);
383 SFL_UNDEF_COUNTER(lacp_elem
.counterBlock
.lacp
.markerResponsePDUsRx
);
384 SFL_UNDEF_COUNTER(lacp_elem
.counterBlock
.lacp
.unknownRx
);
385 lacp_elem
.counterBlock
.lacp
.illegalRx
=
386 lacp_stats
.dot3adAggPortStatsIllegalRx
;
387 lacp_elem
.counterBlock
.lacp
.LACPDUsTx
=
388 lacp_stats
.dot3adAggPortStatsLACPDUsTx
;
389 SFL_UNDEF_COUNTER(lacp_elem
.counterBlock
.lacp
.markerPDUsTx
);
390 SFL_UNDEF_COUNTER(lacp_elem
.counterBlock
.lacp
.markerResponsePDUsTx
);
391 SFLADD_ELEMENT(cs
, &lacp_elem
);
394 /* Include Port name. */
395 if ((ifName
= netdev_get_name(dsp
->ofport
->netdev
)) != NULL
) {
396 memset(&name_elem
, 0, sizeof name_elem
);
397 name_elem
.tag
= SFLCOUNTERS_PORTNAME
;
398 name_elem
.counterBlock
.portName
.portName
.str
= (char *)ifName
;
399 name_elem
.counterBlock
.portName
.portName
.len
= strlen(ifName
);
400 SFLADD_ELEMENT(cs
, &name_elem
);
403 /* Include OpenFlow DPID and openflow port number. */
404 memset(&of_elem
, 0, sizeof of_elem
);
405 of_elem
.tag
= SFLCOUNTERS_OPENFLOWPORT
;
406 of_elem
.counterBlock
.ofPort
.datapath_id
=
407 ofproto_get_datapath_id(dsp
->ofport
->ofproto
);
408 of_elem
.counterBlock
.ofPort
.port_no
=
409 (OVS_FORCE
uint32_t)dsp
->ofport
->ofp_port
;
410 SFLADD_ELEMENT(cs
, &of_elem
);
412 /* Include ethernet counters */
413 memset(ð_elem
, 0, sizeof eth_elem
);
414 eth_elem
.tag
= SFLCOUNTERS_ETHERNET
;
415 eth_counters
= ð_elem
.counterBlock
.ethernet
;
416 eth_counters
->dot3StatsAlignmentErrors
= stats
.rx_frame_errors
;
417 eth_counters
->dot3StatsFCSErrors
= stats
.rx_crc_errors
;
418 eth_counters
->dot3StatsFrameTooLongs
= stats
.rx_oversize_errors
;
419 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsSingleCollisionFrames
);
420 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsMultipleCollisionFrames
);
421 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsSQETestErrors
);
422 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsDeferredTransmissions
);
423 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsLateCollisions
);
424 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsExcessiveCollisions
);
425 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsInternalMacTransmitErrors
);
426 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsCarrierSenseErrors
);
427 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsInternalMacReceiveErrors
);
428 SFL_UNDEF_COUNTER(eth_counters
->dot3StatsSymbolErrors
);
429 SFLADD_ELEMENT(cs
, ð_elem
);
431 sfl_poller_writeCountersSample(poller
, cs
);
434 /* Obtains an address to use for the local sFlow agent and stores it into
435 * '*agent_addr'. Returns true if successful, false on failure.
437 * The sFlow agent address should be a local IP address that is persistent and
438 * reachable over the network, if possible. The IP address associated with
439 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
440 * address used to talk to the controller. If the agent device is not
441 * specified then it is figured out by taking a look at the routing table based
444 sflow_choose_agent_address(const char *agent_device
,
445 const struct sset
*targets
,
446 const char *control_ip
,
447 SFLAddress
*agent_addr
)
452 /* If 'agent_device' is the name of a network device, use its IP
454 if (!netdev_get_ip_by_name(agent_device
, &ip
)) {
458 /* If 'agent_device' is itself an IP address, use it. */
459 struct sockaddr_storage ss
;
460 if (inet_parse_address(agent_device
, &ss
)) {
461 ip
= ss_get_address(&ss
);
466 /* Otherwise, use an appropriate local IP address for one of the
467 * collectors' remote IP addresses. */
469 SSET_FOR_EACH (target
, targets
) {
470 struct sockaddr_storage ss
;
471 if (inet_parse_active(target
, SFL_DEFAULT_COLLECTOR_PORT
, &ss
, true)) {
472 /* sFlow only supports target in default routing table with
475 struct in6_addr target_ip
= ss_get_address(&ss
);
477 struct in6_addr gw
, src
= in6addr_any
;
479 if (ovs_router_lookup(0, &target_ip
, name
, &src
, &gw
)) {
486 struct sockaddr_storage ss
;
487 if (control_ip
&& inet_parse_address(control_ip
, &ss
)) {
488 ip
= ss_get_address(&ss
);
492 VLOG_ERR("could not determine IP address for sFlow agent");
496 memset(agent_addr
, 0, sizeof *agent_addr
);
497 if (IN6_IS_ADDR_V4MAPPED(&ip
)) {
498 agent_addr
->type
= SFLADDRESSTYPE_IP_V4
;
499 agent_addr
->address
.ip_v4
.addr
500 = (OVS_FORCE
uint32_t) in6_addr_get_mapped_ipv4(&ip
);
502 agent_addr
->type
= SFLADDRESSTYPE_IP_V6
;
503 memcpy(agent_addr
->address
.ip_v6
.addr
, ip
.s6_addr
,
504 sizeof agent_addr
->address
.ip_v6
.addr
);
510 dpif_sflow_clear__(struct dpif_sflow
*ds
) OVS_REQUIRES(mutex
)
512 if (ds
->sflow_agent
) {
513 sflow_global_counters_subid_clear(ds
->sflow_agent
->subId
);
514 sfl_agent_release(ds
->sflow_agent
);
515 free(ds
->sflow_agent
);
516 ds
->sflow_agent
= NULL
;
518 collectors_destroy(ds
->collectors
);
519 ds
->collectors
= NULL
;
520 ofproto_sflow_options_destroy(ds
->options
);
523 /* Turn off sampling to save CPU cycles. */
528 dpif_sflow_clear(struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
530 ovs_mutex_lock(&mutex
);
531 dpif_sflow_clear__(ds
);
532 ovs_mutex_unlock(&mutex
);
536 dpif_sflow_is_enabled(const struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
540 ovs_mutex_lock(&mutex
);
541 enabled
= ds
->collectors
!= NULL
;
542 ovs_mutex_unlock(&mutex
);
547 dpif_sflow_create(void)
549 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
550 struct dpif_sflow
*ds
;
552 if (ovsthread_once_start(&once
)) {
553 ovs_mutex_init_recursive(&mutex
);
554 ovsthread_once_done(&once
);
557 ds
= xcalloc(1, sizeof *ds
);
558 ds
->next_tick
= time_now() + 1;
559 hmap_init(&ds
->ports
);
561 ovs_refcount_init(&ds
->ref_cnt
);
567 dpif_sflow_ref(const struct dpif_sflow
*ds_
)
569 struct dpif_sflow
*ds
= CONST_CAST(struct dpif_sflow
*, ds_
);
571 ovs_refcount_ref(&ds
->ref_cnt
);
576 /* 32-bit fraction of packets to sample with. A value of 0 samples no packets,
577 * a value of %UINT32_MAX samples all packets and intermediate values sample
578 * intermediate fractions of packets. */
580 dpif_sflow_get_probability(const struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
582 uint32_t probability
;
583 ovs_mutex_lock(&mutex
);
584 probability
= ds
->probability
;
585 ovs_mutex_unlock(&mutex
);
590 dpif_sflow_unref(struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
592 if (ds
&& ovs_refcount_unref_relaxed(&ds
->ref_cnt
) == 1) {
593 struct dpif_sflow_port
*dsp
, *next
;
595 dpif_sflow_clear(ds
);
596 HMAP_FOR_EACH_SAFE (dsp
, next
, hmap_node
, &ds
->ports
) {
597 dpif_sflow_del_port__(ds
, dsp
);
599 hmap_destroy(&ds
->ports
);
605 dpif_sflow_add_poller(struct dpif_sflow
*ds
, struct dpif_sflow_port
*dsp
)
608 SFLPoller
*poller
= sfl_agent_addPoller(ds
->sflow_agent
, &dsp
->dsi
, ds
,
609 sflow_agent_get_counters
);
610 sfl_poller_set_sFlowCpInterval(poller
, ds
->options
->polling_interval
);
611 sfl_poller_set_sFlowCpReceiver(poller
, RECEIVER_INDEX
);
612 sfl_poller_set_bridgePort(poller
, odp_to_u32(dsp
->odp_port
));
615 static enum dpif_sflow_tunnel_type
616 dpif_sflow_tunnel_type(struct ofport
*ofport
) {
617 const char *type
= netdev_get_type(ofport
->netdev
);
619 if (strcmp(type
, "gre") == 0) {
620 return DPIF_SFLOW_TUNNEL_GRE
;
621 } else if (strcmp(type
, "vxlan") == 0) {
622 return DPIF_SFLOW_TUNNEL_VXLAN
;
623 } else if (strcmp(type
, "lisp") == 0) {
624 return DPIF_SFLOW_TUNNEL_LISP
;
625 } else if (strcmp(type
, "geneve") == 0) {
626 return DPIF_SFLOW_TUNNEL_GENEVE
;
629 return DPIF_SFLOW_TUNNEL_UNKNOWN
;
633 dpif_sflow_tunnel_proto(enum dpif_sflow_tunnel_type tunnel_type
)
635 /* Default to 0 (IPPROTO_IP), meaning "unknown". */
637 switch(tunnel_type
) {
639 case DPIF_SFLOW_TUNNEL_GRE
:
640 ipproto
= IPPROTO_GRE
;
643 case DPIF_SFLOW_TUNNEL_VXLAN
:
644 case DPIF_SFLOW_TUNNEL_LISP
:
645 case DPIF_SFLOW_TUNNEL_GENEVE
:
646 ipproto
= IPPROTO_UDP
;
648 case DPIF_SFLOW_TUNNEL_UNKNOWN
:
655 dpif_sflow_add_port(struct dpif_sflow
*ds
, struct ofport
*ofport
,
656 odp_port_t odp_port
) OVS_EXCLUDED(mutex
)
658 struct dpif_sflow_port
*dsp
;
660 enum dpif_sflow_tunnel_type tunnel_type
;
662 ovs_mutex_lock(&mutex
);
663 dpif_sflow_del_port(ds
, odp_port
);
665 tunnel_type
= dpif_sflow_tunnel_type(ofport
);
666 ifindex
= netdev_get_ifindex(ofport
->netdev
);
669 && tunnel_type
== DPIF_SFLOW_TUNNEL_UNKNOWN
) {
670 /* Not an ifindex port, and not a tunnel port either
671 * so do not add a cross-reference to it here.
676 /* Add to table of ports. */
677 dsp
= xmalloc(sizeof *dsp
);
678 dsp
->ofport
= ofport
;
679 dsp
->odp_port
= odp_port
;
680 dsp
->tunnel_type
= tunnel_type
;
681 hmap_insert(&ds
->ports
, &dsp
->hmap_node
, hash_odp_port(odp_port
));
684 /* Add poller for ports that have ifindex. */
685 SFL_DS_SET(dsp
->dsi
, SFL_DSCLASS_IFINDEX
, ifindex
, 0);
686 if (ds
->sflow_agent
) {
687 dpif_sflow_add_poller(ds
, dsp
);
690 /* Record "ifindex unknown" for the others */
691 SFL_DS_SET(dsp
->dsi
, SFL_DSCLASS_IFINDEX
, 0, 0);
695 ovs_mutex_unlock(&mutex
);
699 dpif_sflow_del_port__(struct dpif_sflow
*ds
, struct dpif_sflow_port
*dsp
)
703 && SFL_DS_INDEX(dsp
->dsi
)) {
704 sfl_agent_removePoller(ds
->sflow_agent
, &dsp
->dsi
);
705 sfl_agent_removeSampler(ds
->sflow_agent
, &dsp
->dsi
);
707 hmap_remove(&ds
->ports
, &dsp
->hmap_node
);
712 dpif_sflow_del_port(struct dpif_sflow
*ds
, odp_port_t odp_port
)
715 struct dpif_sflow_port
*dsp
;
717 ovs_mutex_lock(&mutex
);
718 dsp
= dpif_sflow_find_port(ds
, odp_port
);
720 dpif_sflow_del_port__(ds
, dsp
);
722 ovs_mutex_unlock(&mutex
);
726 dpif_sflow_set_options(struct dpif_sflow
*ds
,
727 const struct ofproto_sflow_options
*options
)
730 struct dpif_sflow_port
*dsp
;
731 bool options_changed
;
732 SFLReceiver
*receiver
;
735 SFLDataSource_instance dsi
;
740 ovs_mutex_lock(&mutex
);
741 if (sset_is_empty(&options
->targets
) || !options
->sampling_rate
) {
742 /* No point in doing any work if there are no targets or nothing to
744 dpif_sflow_clear__(ds
);
748 options_changed
= (!ds
->options
749 || !ofproto_sflow_options_equal(options
, ds
->options
));
751 /* Configure collectors if options have changed or if we're shortchanged in
752 * collectors (which indicates that opening one or more of the configured
753 * collectors failed, so that we should retry). */
755 || collectors_count(ds
->collectors
) < sset_count(&options
->targets
)) {
756 collectors_destroy(ds
->collectors
);
757 collectors_create(&options
->targets
, SFL_DEFAULT_COLLECTOR_PORT
,
759 if (ds
->collectors
== NULL
) {
760 VLOG_WARN_RL(&rl
, "no collectors could be initialized, "
762 dpif_sflow_clear__(ds
);
767 /* Choose agent IP address and agent device (if not yet setup) */
768 if (!sflow_choose_agent_address(options
->agent_device
,
770 options
->control_ip
, &agentIP
)) {
771 dpif_sflow_clear__(ds
);
775 /* Avoid reconfiguring if options didn't change. */
776 if (!options_changed
) {
779 ofproto_sflow_options_destroy(ds
->options
);
780 ds
->options
= ofproto_sflow_options_clone(options
);
783 VLOG_INFO("creating sFlow agent %d", options
->sub_id
);
784 if (ds
->sflow_agent
) {
785 sflow_global_counters_subid_clear(ds
->sflow_agent
->subId
);
786 sfl_agent_release(ds
->sflow_agent
);
788 ds
->sflow_agent
= xcalloc(1, sizeof *ds
->sflow_agent
);
790 sfl_agent_init(ds
->sflow_agent
,
793 now
, /* Boot time. */
794 now
, /* Current time. */
795 ds
, /* Pointer supplied to callbacks. */
796 sflow_agent_alloc_cb
,
798 sflow_agent_error_cb
,
799 sflow_agent_send_packet_cb
);
801 receiver
= sfl_agent_addReceiver(ds
->sflow_agent
);
802 sfl_receiver_set_sFlowRcvrOwner(receiver
, "Open vSwitch sFlow");
803 sfl_receiver_set_sFlowRcvrTimeout(receiver
, 0xffffffff);
805 /* Set the sampling_rate down in the datapath. */
806 ds
->probability
= MAX(1, UINT32_MAX
/ ds
->options
->sampling_rate
);
808 /* Add a single sampler for the bridge. This appears as a PHYSICAL_ENTITY
809 because it is associated with the hypervisor, and interacts with the server
810 hardware directly. The sub_id is used to distinguish this sampler from
811 others on other bridges within the same agent. */
812 dsIndex
= 1000 + options
->sub_id
;
813 SFL_DS_SET(dsi
, SFL_DSCLASS_PHYSICAL_ENTITY
, dsIndex
, 0);
814 sampler
= sfl_agent_addSampler(ds
->sflow_agent
, &dsi
);
815 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler
, ds
->options
->sampling_rate
);
816 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler
, ds
->options
->header_len
);
817 sfl_sampler_set_sFlowFsReceiver(sampler
, RECEIVER_INDEX
);
819 /* Add a counter poller for the bridge so we can use it to send
820 global counters such as datapath cache hit/miss stats. */
821 poller
= sfl_agent_addPoller(ds
->sflow_agent
, &dsi
, ds
,
822 sflow_agent_get_global_counters
);
823 sfl_poller_set_sFlowCpInterval(poller
, ds
->options
->polling_interval
);
824 sfl_poller_set_sFlowCpReceiver(poller
, RECEIVER_INDEX
);
826 /* Add pollers for the currently known ifindex-ports */
827 HMAP_FOR_EACH (dsp
, hmap_node
, &ds
->ports
) {
828 if (SFL_DS_INDEX(dsp
->dsi
)) {
829 dpif_sflow_add_poller(ds
, dsp
);
835 ovs_mutex_unlock(&mutex
);
839 dpif_sflow_odp_port_to_ifindex(const struct dpif_sflow
*ds
,
840 odp_port_t odp_port
) OVS_EXCLUDED(mutex
)
842 struct dpif_sflow_port
*dsp
;
845 ovs_mutex_lock(&mutex
);
846 dsp
= dpif_sflow_find_port(ds
, odp_port
);
847 ret
= dsp
? SFL_DS_INDEX(dsp
->dsi
) : 0;
848 ovs_mutex_unlock(&mutex
);
853 dpif_sflow_tunnel_v4(uint8_t tunnel_ipproto
,
854 const struct flow_tnl
*tunnel
,
855 SFLSampled_ipv4
*ipv4
)
858 ipv4
->protocol
= tunnel_ipproto
;
859 ipv4
->tos
= tunnel
->ip_tos
;
860 ipv4
->src_ip
.addr
= (OVS_FORCE
uint32_t) tunnel
->ip_src
;
861 ipv4
->dst_ip
.addr
= (OVS_FORCE
uint32_t) tunnel
->ip_dst
;
862 ipv4
->src_port
= (OVS_FORCE
uint16_t) tunnel
->tp_src
;
863 ipv4
->dst_port
= (OVS_FORCE
uint16_t) tunnel
->tp_dst
;
867 dpif_sflow_push_mpls_lse(struct dpif_sflow_actions
*sflow_actions
,
870 if (sflow_actions
->mpls_stack_depth
>= FLOW_MAX_MPLS_LABELS
) {
871 sflow_actions
->mpls_err
= true;
875 /* Record the new lse in host-byte-order. */
876 /* BOS flag will be fixed later when we send stack to sFlow library. */
877 sflow_actions
->mpls_lse
[sflow_actions
->mpls_stack_depth
++] = ntohl(lse
);
881 dpif_sflow_pop_mpls_lse(struct dpif_sflow_actions
*sflow_actions
)
883 if (sflow_actions
->mpls_stack_depth
== 0) {
884 sflow_actions
->mpls_err
= true;
887 sflow_actions
->mpls_stack_depth
--;
891 dpif_sflow_set_mpls(struct dpif_sflow_actions
*sflow_actions
,
892 const struct ovs_key_mpls
*mpls_key
, int n
)
895 if (n
> FLOW_MAX_MPLS_LABELS
) {
896 sflow_actions
->mpls_err
= true;
900 for (ii
= 0; ii
< n
; ii
++) {
901 /* Reverse stack order, and use host-byte-order for each lse. */
902 sflow_actions
->mpls_lse
[n
- ii
- 1] = ntohl(mpls_key
[ii
].mpls_lse
);
904 sflow_actions
->mpls_stack_depth
= n
;
908 sflow_read_tnl_push_action(const struct nlattr
*attr
,
909 struct dpif_sflow_actions
*sflow_actions
)
911 /* Modeled on lib/odp-util.c: format_odp_tnl_push_header */
912 const struct ovs_action_push_tnl
*data
= nl_attr_get(attr
);
913 const struct eth_header
*eth
= (const struct eth_header
*) data
->header
;
914 const struct ip_header
*ip
915 = ALIGNED_CAST(const struct ip_header
*, eth
+ 1);
917 sflow_actions
->out_port
= data
->out_port
;
920 /* TODO: SFlow does not currently define a MAC-in-MAC
921 * encapsulation structure. We could use an extension
922 * structure to report this.
926 /* Cannot assume alignment so just use memcpy. */
927 sflow_actions
->tunnel
.ip_src
= get_16aligned_be32(&ip
->ip_src
);
928 sflow_actions
->tunnel
.ip_dst
= get_16aligned_be32(&ip
->ip_dst
);
929 sflow_actions
->tunnel
.ip_tos
= ip
->ip_tos
;
930 sflow_actions
->tunnel
.ip_ttl
= ip
->ip_ttl
;
931 /* The tnl_push action can supply the ip_protocol too. */
932 sflow_actions
->tunnel_ipproto
= ip
->ip_proto
;
935 if (data
->tnl_type
== OVS_VPORT_TYPE_VXLAN
936 || data
->tnl_type
== OVS_VPORT_TYPE_GENEVE
) {
937 const struct udp_header
*udp
= (const struct udp_header
*) (ip
+ 1);
938 sflow_actions
->tunnel
.tp_src
= udp
->udp_src
;
939 sflow_actions
->tunnel
.tp_dst
= udp
->udp_dst
;
941 if (data
->tnl_type
== OVS_VPORT_TYPE_VXLAN
) {
942 const struct vxlanhdr
*vxh
= (const struct vxlanhdr
*) (udp
+ 1);
943 uint64_t tun_id
= ntohl(get_16aligned_be32(&vxh
->vx_vni
)) >> 8;
944 sflow_actions
->tunnel
.tun_id
= htonll(tun_id
);
946 const struct genevehdr
*gnh
= (const struct genevehdr
*) (udp
+ 1);
947 uint64_t tun_id
= ntohl(get_16aligned_be32(&gnh
->vni
)) >> 8;
948 sflow_actions
->tunnel
.tun_id
= htonll(tun_id
);
950 } else if (data
->tnl_type
== OVS_VPORT_TYPE_GRE
) {
951 const void *l4
= ip
+ 1;
952 const struct gre_base_hdr
*greh
= (const struct gre_base_hdr
*) l4
;
953 ovs_16aligned_be32
*options
= (ovs_16aligned_be32
*)(greh
+ 1);
954 if (greh
->flags
& htons(GRE_CSUM
)) {
957 if (greh
->flags
& htons(GRE_KEY
)) {
958 uint64_t tun_id
= ntohl(get_16aligned_be32(options
));
959 sflow_actions
->tunnel
.tun_id
= htonll(tun_id
);
965 sflow_read_set_action(const struct nlattr
*attr
,
966 struct dpif_sflow_actions
*sflow_actions
)
968 enum ovs_key_attr type
= nl_attr_type(attr
);
970 case OVS_KEY_ATTR_ENCAP
:
971 if (++sflow_actions
->encap_depth
> 1) {
972 /* Do not handle multi-encap for now. */
973 sflow_actions
->tunnel_err
= true;
975 dpif_sflow_read_actions(NULL
,
976 nl_attr_get(attr
), nl_attr_get_size(attr
),
977 sflow_actions
, true);
980 case OVS_KEY_ATTR_PRIORITY
:
981 case OVS_KEY_ATTR_SKB_MARK
:
982 case OVS_KEY_ATTR_DP_HASH
:
983 case OVS_KEY_ATTR_RECIRC_ID
:
986 case OVS_KEY_ATTR_TUNNEL
: {
987 if (++sflow_actions
->encap_depth
> 1) {
988 /* Do not handle multi-encap for now. */
989 sflow_actions
->tunnel_err
= true;
991 if (odp_tun_key_from_attr(attr
, &sflow_actions
->tunnel
)
993 /* Tunnel parsing error. */
994 sflow_actions
->tunnel_err
= true;
1000 case OVS_KEY_ATTR_IN_PORT
:
1001 case OVS_KEY_ATTR_ETHERNET
:
1002 case OVS_KEY_ATTR_VLAN
:
1005 case OVS_KEY_ATTR_MPLS
: {
1006 const struct ovs_key_mpls
*mpls_key
= nl_attr_get(attr
);
1007 size_t size
= nl_attr_get_size(attr
);
1008 dpif_sflow_set_mpls(sflow_actions
, mpls_key
, size
/ sizeof *mpls_key
);
1012 case OVS_KEY_ATTR_ETHERTYPE
:
1013 case OVS_KEY_ATTR_IPV4
:
1014 if (sflow_actions
->encap_depth
== 1) {
1015 const struct ovs_key_ipv4
*key
= nl_attr_get(attr
);
1016 if (key
->ipv4_src
) {
1017 sflow_actions
->tunnel
.ip_src
= key
->ipv4_src
;
1019 if (key
->ipv4_dst
) {
1020 sflow_actions
->tunnel
.ip_dst
= key
->ipv4_dst
;
1022 if (key
->ipv4_proto
) {
1023 sflow_actions
->tunnel_ipproto
= key
->ipv4_proto
;
1025 if (key
->ipv4_tos
) {
1026 sflow_actions
->tunnel
.ip_tos
= key
->ipv4_tos
;
1028 if (key
->ipv4_ttl
) {
1029 sflow_actions
->tunnel
.ip_tos
= key
->ipv4_ttl
;
1034 case OVS_KEY_ATTR_IPV6
:
1035 /* TODO: parse IPv6 encap. */
1038 /* These have the same structure and format. */
1039 case OVS_KEY_ATTR_TCP
:
1040 case OVS_KEY_ATTR_UDP
:
1041 case OVS_KEY_ATTR_SCTP
:
1042 if (sflow_actions
->encap_depth
== 1) {
1043 const struct ovs_key_tcp
*key
= nl_attr_get(attr
);
1045 sflow_actions
->tunnel
.tp_src
= key
->tcp_src
;
1048 sflow_actions
->tunnel
.tp_dst
= key
->tcp_dst
;
1053 case OVS_KEY_ATTR_TCP_FLAGS
:
1054 case OVS_KEY_ATTR_ICMP
:
1055 case OVS_KEY_ATTR_ICMPV6
:
1056 case OVS_KEY_ATTR_ARP
:
1057 case OVS_KEY_ATTR_ND
:
1058 case OVS_KEY_ATTR_CT_STATE
:
1059 case OVS_KEY_ATTR_CT_ZONE
:
1060 case OVS_KEY_ATTR_CT_MARK
:
1061 case OVS_KEY_ATTR_CT_LABELS
:
1062 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4
:
1063 case OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6
:
1064 case OVS_KEY_ATTR_UNSPEC
:
1065 case OVS_KEY_ATTR_PACKET_TYPE
:
1066 case OVS_KEY_ATTR_NSH
:
1067 case __OVS_KEY_ATTR_MAX
:
1074 dpif_sflow_capture_input_mpls(const struct flow
*flow
,
1075 struct dpif_sflow_actions
*sflow_actions
)
1077 if (eth_type_mpls(flow
->dl_type
)) {
1081 /* Calculate depth by detecting BOS. */
1082 for (ii
= 0; ii
< FLOW_MAX_MPLS_LABELS
; ii
++) {
1083 lse
= flow
->mpls_lse
[ii
];
1085 if (lse
& htonl(MPLS_BOS_MASK
)) {
1089 /* Capture stack, reversing stack order, and
1090 * using host-byte-order for each lse. BOS flag
1091 * is ignored for now. It is set later when
1092 * the output stack is encoded.
1094 for (ii
= 0; ii
< depth
; ii
++) {
1095 lse
= flow
->mpls_lse
[ii
];
1096 sflow_actions
->mpls_lse
[depth
- ii
- 1] = ntohl(lse
);
1098 sflow_actions
->mpls_stack_depth
= depth
;
1103 dpif_sflow_read_actions(const struct flow
*flow
,
1104 const struct nlattr
*actions
, size_t actions_len
,
1105 struct dpif_sflow_actions
*sflow_actions
,
1108 const struct nlattr
*a
;
1111 if (actions_len
== 0) {
1112 /* Packet dropped.*/
1116 if (flow
!= NULL
&& capture_mpls
== true) {
1117 /* Make sure the MPLS output stack
1118 * is seeded with the input stack.
1120 dpif_sflow_capture_input_mpls(flow
, sflow_actions
);
1122 /* XXX when 802.1AD(QinQ) is supported then
1123 * we can do the same with VLAN stacks here
1127 NL_ATTR_FOR_EACH (a
, left
, actions
, actions_len
) {
1128 enum ovs_action_attr type
= nl_attr_type(a
);
1130 case OVS_ACTION_ATTR_OUTPUT
:
1131 /* Capture the output port in case we need it
1132 * to get the output tunnel type.
1134 sflow_actions
->out_port
= nl_attr_get_odp_port(a
);
1137 case OVS_ACTION_ATTR_TUNNEL_POP
:
1138 /* XXX: Do not handle this for now. It's not clear
1139 * if we should start with encap_depth == 1 when we
1140 * see an input tunnel, or if we should assume
1141 * that the input tunnel was always "popped" if it
1142 * was presented to us decoded in flow->tunnel?
1144 * If we do handle this it might look like this,
1145 * as we clear the captured tunnel info and decrement
1148 * memset(&sflow_actions->tunnel, 0, sizeof struct flow_tnl);
1149 * sflow_actions->tunnel_ipproto = 0;
1150 * --sflow_actions->encap_depth;
1152 * but for now just disable the tunnel annotation:
1154 sflow_actions
->tunnel_err
= true;
1157 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1158 /* XXX: This actions appears to come with it's own
1159 * OUTPUT action, so should it be regarded as having
1160 * an implicit "pop" following it too? Put another
1161 * way, would two tnl_push() actions in succession
1162 * result in a packet with two layers of encap?
1164 if (++sflow_actions
->encap_depth
> 1) {
1165 /* Do not handle multi-encap for now. */
1166 sflow_actions
->tunnel_err
= true;
1168 sflow_read_tnl_push_action(a
, sflow_actions
);
1172 case OVS_ACTION_ATTR_TRUNC
:
1173 case OVS_ACTION_ATTR_USERSPACE
:
1174 case OVS_ACTION_ATTR_RECIRC
:
1175 case OVS_ACTION_ATTR_HASH
:
1176 case OVS_ACTION_ATTR_CT
:
1177 case OVS_ACTION_ATTR_CT_CLEAR
:
1178 case OVS_ACTION_ATTR_METER
:
1181 case OVS_ACTION_ATTR_SET_MASKED
:
1182 /* TODO: apply mask. XXX: Are we likely to see this? */
1185 case OVS_ACTION_ATTR_SET
:
1186 sflow_read_set_action(nl_attr_get(a
), sflow_actions
);
1189 case OVS_ACTION_ATTR_PUSH_VLAN
:
1190 case OVS_ACTION_ATTR_POP_VLAN
:
1191 /* TODO: 802.1AD(QinQ) is not supported by OVS (yet), so do not
1192 * construct a VLAN-stack. The sFlow user-action cookie already
1193 * captures the egress VLAN ID so there is nothing more to do here.
1197 case OVS_ACTION_ATTR_PUSH_MPLS
: {
1198 const struct ovs_action_push_mpls
*mpls
= nl_attr_get(a
);
1200 dpif_sflow_push_mpls_lse(sflow_actions
, mpls
->mpls_lse
);
1204 case OVS_ACTION_ATTR_POP_MPLS
: {
1205 dpif_sflow_pop_mpls_lse(sflow_actions
);
1208 case OVS_ACTION_ATTR_PUSH_ETH
:
1209 case OVS_ACTION_ATTR_POP_ETH
:
1210 /* TODO: SFlow does not currently define a MAC-in-MAC
1211 * encapsulation structure. We could use an extension
1212 * structure to report this.
1215 case OVS_ACTION_ATTR_CLONE
:
1217 dpif_sflow_read_actions(flow
, nl_attr_get(a
), nl_attr_get_size(a
),
1218 sflow_actions
, false);
1221 case OVS_ACTION_ATTR_SAMPLE
:
1222 case OVS_ACTION_ATTR_PUSH_NSH
:
1223 case OVS_ACTION_ATTR_POP_NSH
:
1224 case OVS_ACTION_ATTR_UNSPEC
:
1225 case __OVS_ACTION_ATTR_MAX
:
1233 dpif_sflow_encode_mpls_stack(SFLLabelStack
*stack
,
1234 uint32_t *mpls_lse_buf
,
1235 const struct dpif_sflow_actions
*sflow_actions
)
1237 /* Put the MPLS stack back into "packet header" order,
1238 * and make sure the BOS flag is set correctly on the last
1239 * one. Each lse is still in host-byte-order.
1243 stack
->depth
= sflow_actions
->mpls_stack_depth
;
1244 stack
->stack
= mpls_lse_buf
;
1245 for (ii
= 0; ii
< stack
->depth
; ii
++) {
1246 lse
= sflow_actions
->mpls_lse
[stack
->depth
- ii
- 1];
1247 stack
->stack
[ii
] = (lse
& ~MPLS_BOS_MASK
);
1249 stack
->stack
[stack
->depth
- 1] |= MPLS_BOS_MASK
;
1252 /* Extract the output port count from the user action cookie.
1253 * See http://sflow.org/sflow_version_5.txt "Input/Output port information"
1256 dpif_sflow_cookie_num_outputs(const struct user_action_cookie
*cookie
)
1258 uint32_t format
= cookie
->sflow
.output
& 0xC0000000;
1259 uint32_t port_n
= cookie
->sflow
.output
& 0x3FFFFFFF;
1261 return port_n
? 1 : 0;
1263 else if (format
== 0x80000000) {
1270 dpif_sflow_received(struct dpif_sflow
*ds
, const struct dp_packet
*packet
,
1271 const struct flow
*flow
, odp_port_t odp_in_port
,
1272 const struct user_action_cookie
*cookie
,
1273 const struct dpif_sflow_actions
*sflow_actions
)
1276 SFL_FLOW_SAMPLE_TYPE fs
;
1277 SFLFlow_sample_element hdrElem
;
1278 SFLSampled_header
*header
;
1279 SFLFlow_sample_element switchElem
;
1280 uint8_t tnlInProto
, tnlOutProto
;
1281 SFLFlow_sample_element tnlInElem
, tnlOutElem
;
1282 SFLFlow_sample_element vniInElem
, vniOutElem
;
1283 SFLFlow_sample_element mplsElem
;
1284 uint32_t mpls_lse_buf
[FLOW_MAX_MPLS_LABELS
];
1285 SFLSampler
*sampler
;
1286 struct dpif_sflow_port
*in_dsp
;
1287 struct dpif_sflow_port
*out_dsp
;
1290 ovs_mutex_lock(&mutex
);
1291 sampler
= ds
->sflow_agent
->samplers
;
1296 /* Build a flow sample. */
1297 memset(&fs
, 0, sizeof fs
);
1299 /* Look up the input ifIndex if this port has one. Otherwise just
1300 * leave it as 0 (meaning 'unknown') and continue. */
1301 in_dsp
= dpif_sflow_find_port(ds
, odp_in_port
);
1303 fs
.input
= SFL_DS_INDEX(in_dsp
->dsi
);
1306 /* Make the assumption that the random number generator in the
1307 * datapath converges to the configured mean, and just increment the
1308 * samplePool by the configured sampling rate every time. */
1309 sampler
->samplePool
+= sfl_sampler_get_sFlowFsPacketSamplingRate(sampler
);
1311 /* Sampled header. */
1312 memset(&hdrElem
, 0, sizeof hdrElem
);
1313 hdrElem
.tag
= SFLFLOW_HEADER
;
1314 header
= &hdrElem
.flowType
.header
;
1315 header
->header_protocol
= SFLHEADER_ETHERNET_ISO8023
;
1316 /* The frame_length should include the Ethernet FCS (4 bytes),
1317 * but it has already been stripped, so we need to add 4 here. */
1318 header
->frame_length
= dp_packet_size(packet
) + 4;
1319 /* Ethernet FCS stripped off. */
1320 header
->stripped
= 4;
1321 header
->header_length
= MIN(dp_packet_size(packet
),
1322 sampler
->sFlowFsMaximumHeaderSize
);
1323 header
->header_bytes
= dp_packet_data(packet
);
1325 /* Add extended switch element. */
1326 memset(&switchElem
, 0, sizeof(switchElem
));
1327 switchElem
.tag
= SFLFLOW_EX_SWITCH
;
1328 switchElem
.flowType
.sw
.src_vlan
= vlan_tci_to_vid(flow
->vlans
[0].tci
);
1329 switchElem
.flowType
.sw
.src_priority
= vlan_tci_to_pcp(flow
->vlans
[0].tci
);
1331 /* Retrieve data from user_action_cookie. */
1332 vlan_tci
= cookie
->sflow
.vlan_tci
;
1333 switchElem
.flowType
.sw
.dst_vlan
= vlan_tci_to_vid(vlan_tci
);
1334 switchElem
.flowType
.sw
.dst_priority
= vlan_tci_to_pcp(vlan_tci
);
1336 fs
.output
= cookie
->sflow
.output
;
1339 if (flow
->tunnel
.ip_dst
) {
1340 memset(&tnlInElem
, 0, sizeof(tnlInElem
));
1341 tnlInElem
.tag
= SFLFLOW_EX_IPV4_TUNNEL_INGRESS
;
1342 tnlInProto
= in_dsp
? dpif_sflow_tunnel_proto(in_dsp
->tunnel_type
) : 0;
1343 dpif_sflow_tunnel_v4(tnlInProto
,
1345 &tnlInElem
.flowType
.ipv4
);
1346 SFLADD_ELEMENT(&fs
, &tnlInElem
);
1347 if (flow
->tunnel
.tun_id
) {
1348 memset(&vniInElem
, 0, sizeof(vniInElem
));
1349 vniInElem
.tag
= SFLFLOW_EX_VNI_INGRESS
;
1350 vniInElem
.flowType
.tunnel_vni
.vni
1351 = ntohll(flow
->tunnel
.tun_id
);
1352 SFLADD_ELEMENT(&fs
, &vniInElem
);
1356 /* Output tunnel. */
1358 && sflow_actions
->encap_depth
== 1
1359 && !sflow_actions
->tunnel_err
1360 && dpif_sflow_cookie_num_outputs(cookie
) == 1) {
1361 tnlOutProto
= sflow_actions
->tunnel_ipproto
;
1362 if (tnlOutProto
== 0) {
1363 /* Try to infer the ip-protocol from the output port. */
1364 if (sflow_actions
->out_port
!= ODPP_NONE
) {
1365 out_dsp
= dpif_sflow_find_port(ds
, sflow_actions
->out_port
);
1367 tnlOutProto
= dpif_sflow_tunnel_proto(out_dsp
->tunnel_type
);
1371 memset(&tnlOutElem
, 0, sizeof(tnlOutElem
));
1372 tnlOutElem
.tag
= SFLFLOW_EX_IPV4_TUNNEL_EGRESS
;
1373 dpif_sflow_tunnel_v4(tnlOutProto
,
1374 &sflow_actions
->tunnel
,
1375 &tnlOutElem
.flowType
.ipv4
);
1376 SFLADD_ELEMENT(&fs
, &tnlOutElem
);
1377 if (sflow_actions
->tunnel
.tun_id
) {
1378 memset(&vniOutElem
, 0, sizeof(vniOutElem
));
1379 vniOutElem
.tag
= SFLFLOW_EX_VNI_EGRESS
;
1380 vniOutElem
.flowType
.tunnel_vni
.vni
1381 = ntohll(sflow_actions
->tunnel
.tun_id
);
1382 SFLADD_ELEMENT(&fs
, &vniOutElem
);
1386 /* MPLS output label stack. */
1388 && sflow_actions
->mpls_stack_depth
> 0
1389 && !sflow_actions
->mpls_err
1390 && dpif_sflow_cookie_num_outputs(cookie
) == 1) {
1391 memset(&mplsElem
, 0, sizeof(mplsElem
));
1392 mplsElem
.tag
= SFLFLOW_EX_MPLS
;
1393 dpif_sflow_encode_mpls_stack(&mplsElem
.flowType
.mpls
.out_stack
,
1396 SFLADD_ELEMENT(&fs
, &mplsElem
);
1399 /* Submit the flow sample to be encoded into the next datagram. */
1400 SFLADD_ELEMENT(&fs
, &hdrElem
);
1401 SFLADD_ELEMENT(&fs
, &switchElem
);
1402 sfl_sampler_writeFlowSample(sampler
, &fs
);
1405 ovs_mutex_unlock(&mutex
);
1409 dpif_sflow_run(struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
1411 ovs_mutex_lock(&mutex
);
1412 if (ds
->collectors
!= NULL
) {
1413 time_t now
= time_now();
1415 if (now
>= ds
->next_tick
) {
1416 sfl_agent_tick(ds
->sflow_agent
, time_wall());
1417 ds
->next_tick
= now
+ 1;
1420 ovs_mutex_unlock(&mutex
);
1424 dpif_sflow_wait(struct dpif_sflow
*ds
) OVS_EXCLUDED(mutex
)
1426 ovs_mutex_lock(&mutex
);
1427 if (ds
->collectors
!= NULL
) {
1428 poll_timer_wait_until(ds
->next_tick
* 1000LL);
1430 ovs_mutex_unlock(&mutex
);