2 * Copyright (c) 2009, 2010, 2011, 2012 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/socket.h>
24 #include "collectors.h"
34 #include "poll-loop.h"
35 #include "route-table.h"
36 #include "sflow_api.h"
37 #include "socket-util.h"
40 #include "lib/odp-util.h"
41 #include "ofproto-provider.h"
43 VLOG_DEFINE_THIS_MODULE(sflow
);
45 struct dpif_sflow_port
{
46 struct hmap_node hmap_node
; /* In struct dpif_sflow's "ports" hmap. */
47 SFLDataSource_instance dsi
; /* sFlow library's notion of port number. */
48 struct ofport
*ofport
; /* To retrive port stats. */
53 struct collectors
*collectors
;
54 SFLAgent
*sflow_agent
;
55 struct ofproto_sflow_options
*options
;
57 size_t n_flood
, n_all
;
58 struct hmap ports
; /* Contains "struct dpif_sflow_port"s. */
62 static void dpif_sflow_del_port__(struct dpif_sflow
*,
63 struct dpif_sflow_port
*);
65 #define RECEIVER_INDEX 1
67 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
70 nullable_string_is_equal(const char *a
, const char *b
)
72 return a
? b
&& !strcmp(a
, b
) : !b
;
76 ofproto_sflow_options_equal(const struct ofproto_sflow_options
*a
,
77 const struct ofproto_sflow_options
*b
)
79 return (sset_equals(&a
->targets
, &b
->targets
)
80 && a
->sampling_rate
== b
->sampling_rate
81 && a
->polling_interval
== b
->polling_interval
82 && a
->header_len
== b
->header_len
83 && a
->sub_id
== b
->sub_id
84 && nullable_string_is_equal(a
->agent_device
, b
->agent_device
)
85 && nullable_string_is_equal(a
->control_ip
, b
->control_ip
));
88 static struct ofproto_sflow_options
*
89 ofproto_sflow_options_clone(const struct ofproto_sflow_options
*old
)
91 struct ofproto_sflow_options
*new = xmemdup(old
, sizeof *old
);
92 sset_clone(&new->targets
, &old
->targets
);
93 new->agent_device
= old
->agent_device
? xstrdup(old
->agent_device
) : NULL
;
94 new->control_ip
= old
->control_ip
? xstrdup(old
->control_ip
) : NULL
;
99 ofproto_sflow_options_destroy(struct ofproto_sflow_options
*options
)
102 sset_destroy(&options
->targets
);
103 free(options
->agent_device
);
104 free(options
->control_ip
);
109 /* sFlow library callback to allocate memory. */
111 sflow_agent_alloc_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
114 return calloc(1, bytes
);
117 /* sFlow library callback to free memory. */
119 sflow_agent_free_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
126 /* sFlow library callback to report error. */
128 sflow_agent_error_cb(void *magic OVS_UNUSED
, SFLAgent
*agent OVS_UNUSED
,
131 VLOG_WARN("sFlow agent error: %s", msg
);
134 /* sFlow library callback to send datagram. */
136 sflow_agent_send_packet_cb(void *ds_
, SFLAgent
*agent OVS_UNUSED
,
137 SFLReceiver
*receiver OVS_UNUSED
, u_char
*pkt
,
140 struct dpif_sflow
*ds
= ds_
;
141 collectors_send(ds
->collectors
, pkt
, pktLen
);
144 static struct dpif_sflow_port
*
145 dpif_sflow_find_port(const struct dpif_sflow
*ds
, uint32_t odp_port
)
147 struct dpif_sflow_port
*dsp
;
149 HMAP_FOR_EACH_IN_BUCKET (dsp
, hmap_node
,
150 hash_int(odp_port
, 0), &ds
->ports
) {
151 if (dsp
->odp_port
== odp_port
) {
159 sflow_agent_get_counters(void *ds_
, SFLPoller
*poller
,
160 SFL_COUNTERS_SAMPLE_TYPE
*cs
)
162 struct dpif_sflow
*ds
= ds_
;
163 SFLCounters_sample_element elem
;
164 enum netdev_features current
;
165 struct dpif_sflow_port
*dsp
;
166 SFLIf_counters
*counters
;
167 struct netdev_stats stats
;
168 enum netdev_flags flags
;
170 dsp
= dpif_sflow_find_port(ds
, poller
->bridgePort
);
175 elem
.tag
= SFLCOUNTERS_GENERIC
;
176 counters
= &elem
.counterBlock
.generic
;
177 counters
->ifIndex
= SFL_DS_INDEX(poller
->dsi
);
178 counters
->ifType
= 6;
179 if (!netdev_get_features(dsp
->ofport
->netdev
, ¤t
, NULL
, NULL
, NULL
)) {
180 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
181 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
182 counters
->ifSpeed
= netdev_features_to_bps(current
, 0);
183 counters
->ifDirection
= (netdev_features_is_full_duplex(current
)
186 counters
->ifSpeed
= 100000000;
187 counters
->ifDirection
= 0;
189 if (!netdev_get_flags(dsp
->ofport
->netdev
, &flags
) && flags
& NETDEV_UP
) {
190 counters
->ifStatus
= 1; /* ifAdminStatus up. */
191 if (netdev_get_carrier(dsp
->ofport
->netdev
)) {
192 counters
->ifStatus
|= 2; /* ifOperStatus us. */
195 counters
->ifStatus
= 0; /* Down. */
199 1. Is the multicast counter filled in?
200 2. Does the multicast counter include broadcasts?
201 3. Does the rx_packets counter include multicasts/broadcasts?
203 ofproto_port_get_stats(dsp
->ofport
, &stats
);
204 counters
->ifInOctets
= stats
.rx_bytes
;
205 counters
->ifInUcastPkts
= stats
.rx_packets
;
206 counters
->ifInMulticastPkts
= stats
.multicast
;
207 counters
->ifInBroadcastPkts
= -1;
208 counters
->ifInDiscards
= stats
.rx_dropped
;
209 counters
->ifInErrors
= stats
.rx_errors
;
210 counters
->ifInUnknownProtos
= -1;
211 counters
->ifOutOctets
= stats
.tx_bytes
;
212 counters
->ifOutUcastPkts
= stats
.tx_packets
;
213 counters
->ifOutMulticastPkts
= -1;
214 counters
->ifOutBroadcastPkts
= -1;
215 counters
->ifOutDiscards
= stats
.tx_dropped
;
216 counters
->ifOutErrors
= stats
.tx_errors
;
217 counters
->ifPromiscuousMode
= 0;
219 SFLADD_ELEMENT(cs
, &elem
);
220 sfl_poller_writeCountersSample(poller
, cs
);
223 /* Obtains an address to use for the local sFlow agent and stores it into
224 * '*agent_addr'. Returns true if successful, false on failure.
226 * The sFlow agent address should be a local IP address that is persistent and
227 * reachable over the network, if possible. The IP address associated with
228 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
229 * address used to talk to the controller. If the agent device is not
230 * specified then it is figured out by taking a look at the routing table based
233 sflow_choose_agent_address(const char *agent_device
,
234 const struct sset
*targets
,
235 const char *control_ip
,
236 SFLAddress
*agent_addr
)
241 memset(agent_addr
, 0, sizeof *agent_addr
);
242 agent_addr
->type
= SFLADDRESSTYPE_IP_V4
;
245 if (!netdev_get_in4_by_name(agent_device
, &in4
)) {
250 SSET_FOR_EACH (target
, targets
) {
251 struct sockaddr_in sin
;
254 if (inet_parse_active(target
, SFL_DEFAULT_COLLECTOR_PORT
, &sin
)
255 && route_table_get_name(sin
.sin_addr
.s_addr
, name
)
256 && !netdev_get_in4_by_name(name
, &in4
)) {
261 if (control_ip
&& !lookup_ip(control_ip
, &in4
)) {
265 VLOG_ERR("could not determine IP address for sFlow agent");
269 agent_addr
->address
.ip_v4
.addr
= (OVS_FORCE
uint32_t) in4
.s_addr
;
274 dpif_sflow_clear(struct dpif_sflow
*ds
)
276 if (ds
->sflow_agent
) {
277 sfl_agent_release(ds
->sflow_agent
);
278 ds
->sflow_agent
= NULL
;
280 collectors_destroy(ds
->collectors
);
281 ds
->collectors
= NULL
;
282 ofproto_sflow_options_destroy(ds
->options
);
285 /* Turn off sampling to save CPU cycles. */
290 dpif_sflow_is_enabled(const struct dpif_sflow
*ds
)
292 return ds
->collectors
!= NULL
;
296 dpif_sflow_create(void)
298 struct dpif_sflow
*ds
;
300 ds
= xcalloc(1, sizeof *ds
);
301 ds
->next_tick
= time_now() + 1;
302 hmap_init(&ds
->ports
);
304 route_table_register();
309 /* 32-bit fraction of packets to sample with. A value of 0 samples no packets,
310 * a value of %UINT32_MAX samples all packets and intermediate values sample
311 * intermediate fractions of packets. */
313 dpif_sflow_get_probability(const struct dpif_sflow
*ds
)
315 return ds
->probability
;
319 dpif_sflow_destroy(struct dpif_sflow
*ds
)
322 struct dpif_sflow_port
*dsp
, *next
;
324 route_table_unregister();
325 dpif_sflow_clear(ds
);
326 HMAP_FOR_EACH_SAFE (dsp
, next
, hmap_node
, &ds
->ports
) {
327 dpif_sflow_del_port__(ds
, dsp
);
329 hmap_destroy(&ds
->ports
);
335 dpif_sflow_add_poller(struct dpif_sflow
*ds
, struct dpif_sflow_port
*dsp
)
337 SFLPoller
*poller
= sfl_agent_addPoller(ds
->sflow_agent
, &dsp
->dsi
, ds
,
338 sflow_agent_get_counters
);
339 sfl_poller_set_sFlowCpInterval(poller
, ds
->options
->polling_interval
);
340 sfl_poller_set_sFlowCpReceiver(poller
, RECEIVER_INDEX
);
341 sfl_poller_set_bridgePort(poller
, dsp
->odp_port
);
345 dpif_sflow_add_sampler(struct dpif_sflow
*ds
, struct dpif_sflow_port
*dsp
)
347 SFLSampler
*sampler
= sfl_agent_addSampler(ds
->sflow_agent
, &dsp
->dsi
);
348 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler
, ds
->options
->sampling_rate
);
349 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler
, ds
->options
->header_len
);
350 sfl_sampler_set_sFlowFsReceiver(sampler
, RECEIVER_INDEX
);
354 dpif_sflow_add_port(struct dpif_sflow
*ds
, struct ofport
*ofport
,
357 struct dpif_sflow_port
*dsp
;
360 dpif_sflow_del_port(ds
, odp_port
);
362 /* Add to table of ports. */
363 dsp
= xmalloc(sizeof *dsp
);
364 ifindex
= netdev_get_ifindex(ofport
->netdev
);
366 ifindex
= (ds
->sflow_agent
->subId
<< 16) + odp_port
;
368 dsp
->ofport
= ofport
;
369 dsp
->odp_port
= odp_port
;
370 SFL_DS_SET(dsp
->dsi
, 0, ifindex
, 0);
371 hmap_insert(&ds
->ports
, &dsp
->hmap_node
, hash_int(odp_port
, 0));
373 /* Add poller and sampler. */
374 if (ds
->sflow_agent
) {
375 dpif_sflow_add_poller(ds
, dsp
);
376 dpif_sflow_add_sampler(ds
, dsp
);
381 dpif_sflow_del_port__(struct dpif_sflow
*ds
, struct dpif_sflow_port
*dsp
)
383 if (ds
->sflow_agent
) {
384 sfl_agent_removePoller(ds
->sflow_agent
, &dsp
->dsi
);
385 sfl_agent_removeSampler(ds
->sflow_agent
, &dsp
->dsi
);
387 hmap_remove(&ds
->ports
, &dsp
->hmap_node
);
392 dpif_sflow_del_port(struct dpif_sflow
*ds
, uint32_t odp_port
)
394 struct dpif_sflow_port
*dsp
= dpif_sflow_find_port(ds
, odp_port
);
396 dpif_sflow_del_port__(ds
, dsp
);
401 dpif_sflow_set_options(struct dpif_sflow
*ds
,
402 const struct ofproto_sflow_options
*options
)
404 struct dpif_sflow_port
*dsp
;
405 bool options_changed
;
406 SFLReceiver
*receiver
;
410 if (sset_is_empty(&options
->targets
) || !options
->sampling_rate
) {
411 /* No point in doing any work if there are no targets or nothing to
413 dpif_sflow_clear(ds
);
417 options_changed
= (!ds
->options
418 || !ofproto_sflow_options_equal(options
, ds
->options
));
420 /* Configure collectors if options have changed or if we're shortchanged in
421 * collectors (which indicates that opening one or more of the configured
422 * collectors failed, so that we should retry). */
424 || collectors_count(ds
->collectors
) < sset_count(&options
->targets
)) {
425 collectors_destroy(ds
->collectors
);
426 collectors_create(&options
->targets
, SFL_DEFAULT_COLLECTOR_PORT
,
428 if (ds
->collectors
== NULL
) {
429 VLOG_WARN_RL(&rl
, "no collectors could be initialized, "
431 dpif_sflow_clear(ds
);
436 /* Choose agent IP address and agent device (if not yet setup) */
437 if (!sflow_choose_agent_address(options
->agent_device
,
439 options
->control_ip
, &agentIP
)) {
440 dpif_sflow_clear(ds
);
444 /* Avoid reconfiguring if options didn't change. */
445 if (!options_changed
) {
448 ofproto_sflow_options_destroy(ds
->options
);
449 ds
->options
= ofproto_sflow_options_clone(options
);
452 VLOG_INFO("creating sFlow agent %d", options
->sub_id
);
453 if (ds
->sflow_agent
) {
454 sfl_agent_release(ds
->sflow_agent
);
456 ds
->sflow_agent
= xcalloc(1, sizeof *ds
->sflow_agent
);
458 sfl_agent_init(ds
->sflow_agent
,
461 now
, /* Boot time. */
462 now
, /* Current time. */
463 ds
, /* Pointer supplied to callbacks. */
464 sflow_agent_alloc_cb
,
466 sflow_agent_error_cb
,
467 sflow_agent_send_packet_cb
);
469 receiver
= sfl_agent_addReceiver(ds
->sflow_agent
);
470 sfl_receiver_set_sFlowRcvrOwner(receiver
, "Open vSwitch sFlow");
471 sfl_receiver_set_sFlowRcvrTimeout(receiver
, 0xffffffff);
473 /* Set the sampling_rate down in the datapath. */
474 ds
->probability
= MAX(1, UINT32_MAX
/ ds
->options
->sampling_rate
);
476 /* Add samplers and pollers for the currently known ports. */
477 HMAP_FOR_EACH (dsp
, hmap_node
, &ds
->ports
) {
478 dpif_sflow_add_poller(ds
, dsp
);
479 dpif_sflow_add_sampler(ds
, dsp
);
484 dpif_sflow_odp_port_to_ifindex(const struct dpif_sflow
*ds
,
487 struct dpif_sflow_port
*dsp
= dpif_sflow_find_port(ds
, odp_port
);
488 return dsp
? SFL_DS_INDEX(dsp
->dsi
) : 0;
492 dpif_sflow_received(struct dpif_sflow
*ds
, struct ofpbuf
*packet
,
493 const struct flow
*flow
, uint32_t odp_in_port
,
494 const union user_action_cookie
*cookie
)
496 SFL_FLOW_SAMPLE_TYPE fs
;
497 SFLFlow_sample_element hdrElem
;
498 SFLSampled_header
*header
;
499 SFLFlow_sample_element switchElem
;
501 struct dpif_sflow_port
*in_dsp
;
502 struct netdev_stats stats
;
506 /* Build a flow sample */
507 memset(&fs
, 0, sizeof fs
);
509 in_dsp
= dpif_sflow_find_port(ds
, odp_in_port
);
513 fs
.input
= SFL_DS_INDEX(in_dsp
->dsi
);
515 error
= ofproto_port_get_stats(in_dsp
->ofport
, &stats
);
517 VLOG_WARN_RL(&rl
, "netdev get-stats error %s", strerror(error
));
520 fs
.sample_pool
= stats
.rx_packets
;
522 /* We are going to give it to the sampler that represents this input port.
523 * By implementing "ingress-only" sampling like this we ensure that we
524 * never have to offer the same sample to more than one sampler. */
525 sampler
= sfl_agent_getSamplerByIfIndex(ds
->sflow_agent
, fs
.input
);
527 VLOG_WARN_RL(&rl
, "no sampler for input ifIndex (%"PRIu32
")",
532 /* Sampled header. */
533 memset(&hdrElem
, 0, sizeof hdrElem
);
534 hdrElem
.tag
= SFLFLOW_HEADER
;
535 header
= &hdrElem
.flowType
.header
;
536 header
->header_protocol
= SFLHEADER_ETHERNET_ISO8023
;
537 /* The frame_length should include the Ethernet FCS (4 bytes),
538 but it has already been stripped, so we need to add 4 here. */
539 header
->frame_length
= packet
->size
+ 4;
540 /* Ethernet FCS stripped off. */
541 header
->stripped
= 4;
542 header
->header_length
= MIN(packet
->size
,
543 sampler
->sFlowFsMaximumHeaderSize
);
544 header
->header_bytes
= packet
->data
;
546 /* Add extended switch element. */
547 memset(&switchElem
, 0, sizeof(switchElem
));
548 switchElem
.tag
= SFLFLOW_EX_SWITCH
;
549 switchElem
.flowType
.sw
.src_vlan
= vlan_tci_to_vid(flow
->vlan_tci
);
550 switchElem
.flowType
.sw
.src_priority
= vlan_tci_to_pcp(flow
->vlan_tci
);
552 /* Retrieve data from user_action_cookie. */
553 vlan_tci
= cookie
->sflow
.vlan_tci
;
554 switchElem
.flowType
.sw
.dst_vlan
= vlan_tci_to_vid(vlan_tci
);
555 switchElem
.flowType
.sw
.dst_priority
= vlan_tci_to_pcp(vlan_tci
);
557 fs
.output
= cookie
->sflow
.output
;
559 /* Submit the flow sample to be encoded into the next datagram. */
560 SFLADD_ELEMENT(&fs
, &hdrElem
);
561 SFLADD_ELEMENT(&fs
, &switchElem
);
562 sfl_sampler_writeFlowSample(sampler
, &fs
);
566 dpif_sflow_run(struct dpif_sflow
*ds
)
568 if (dpif_sflow_is_enabled(ds
)) {
569 time_t now
= time_now();
571 if (now
>= ds
->next_tick
) {
572 sfl_agent_tick(ds
->sflow_agent
, time_wall());
573 ds
->next_tick
= now
+ 1;
579 dpif_sflow_wait(struct dpif_sflow
*ds
)
581 if (dpif_sflow_is_enabled(ds
)) {
582 poll_timer_wait_until(ds
->next_tick
* 1000LL);