2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 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 static struct ovs_mutex mutex;
47 struct dpif_sflow_port {
48 struct hmap_node hmap_node; /* In struct dpif_sflow's "ports" hmap. */
49 SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
50 struct ofport *ofport; /* To retrive port stats. */
55 struct collectors *collectors;
56 SFLAgent *sflow_agent;
57 struct ofproto_sflow_options *options;
59 size_t n_flood, n_all;
60 struct hmap ports; /* Contains "struct dpif_sflow_port"s. */
65 static void dpif_sflow_del_port__(struct dpif_sflow *,
66 struct dpif_sflow_port *);
68 #define RECEIVER_INDEX 1
70 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
73 nullable_string_is_equal(const char *a, const char *b)
75 return a ? b && !strcmp(a, b) : !b;
79 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
80 const struct ofproto_sflow_options *b)
82 return (sset_equals(&a->targets, &b->targets)
83 && a->sampling_rate == b->sampling_rate
84 && a->polling_interval == b->polling_interval
85 && a->header_len == b->header_len
86 && a->sub_id == b->sub_id
87 && nullable_string_is_equal(a->agent_device, b->agent_device)
88 && nullable_string_is_equal(a->control_ip, b->control_ip));
91 static struct ofproto_sflow_options *
92 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
94 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
95 sset_clone(&new->targets, &old->targets);
96 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
97 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
102 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
105 sset_destroy(&options->targets);
106 free(options->agent_device);
107 free(options->control_ip);
112 /* sFlow library callback to allocate memory. */
114 sflow_agent_alloc_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
117 return calloc(1, bytes);
120 /* sFlow library callback to free memory. */
122 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
129 /* sFlow library callback to report error. */
131 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
134 VLOG_WARN("sFlow agent error: %s", msg);
137 /* sFlow library callback to send datagram. */
139 sflow_agent_send_packet_cb(void *ds_, SFLAgent *agent OVS_UNUSED,
140 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
143 struct dpif_sflow *ds = ds_;
144 collectors_send(ds->collectors, pkt, pktLen);
147 static struct dpif_sflow_port *
148 dpif_sflow_find_port(const struct dpif_sflow *ds, odp_port_t odp_port)
151 struct dpif_sflow_port *dsp;
153 HMAP_FOR_EACH_IN_BUCKET (dsp, hmap_node, hash_odp_port(odp_port),
155 if (dsp->odp_port == odp_port) {
163 sflow_agent_get_counters(void *ds_, SFLPoller *poller,
164 SFL_COUNTERS_SAMPLE_TYPE *cs)
167 struct dpif_sflow *ds = ds_;
168 SFLCounters_sample_element elem;
169 enum netdev_features current;
170 struct dpif_sflow_port *dsp;
171 SFLIf_counters *counters;
172 struct netdev_stats stats;
173 enum netdev_flags flags;
175 dsp = dpif_sflow_find_port(ds, u32_to_odp(poller->bridgePort));
180 elem.tag = SFLCOUNTERS_GENERIC;
181 counters = &elem.counterBlock.generic;
182 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
183 counters->ifType = 6;
184 if (!netdev_get_features(dsp->ofport->netdev, ¤t, NULL, NULL, NULL)) {
185 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
186 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
187 counters->ifSpeed = netdev_features_to_bps(current, 0);
188 counters->ifDirection = (netdev_features_is_full_duplex(current)
191 counters->ifSpeed = 100000000;
192 counters->ifDirection = 0;
194 if (!netdev_get_flags(dsp->ofport->netdev, &flags) && flags & NETDEV_UP) {
195 counters->ifStatus = 1; /* ifAdminStatus up. */
196 if (netdev_get_carrier(dsp->ofport->netdev)) {
197 counters->ifStatus |= 2; /* ifOperStatus us. */
200 counters->ifStatus = 0; /* Down. */
204 1. Is the multicast counter filled in?
205 2. Does the multicast counter include broadcasts?
206 3. Does the rx_packets counter include multicasts/broadcasts?
208 ofproto_port_get_stats(dsp->ofport, &stats);
209 counters->ifInOctets = stats.rx_bytes;
210 counters->ifInUcastPkts = stats.rx_packets;
211 counters->ifInMulticastPkts = stats.multicast;
212 counters->ifInBroadcastPkts = -1;
213 counters->ifInDiscards = stats.rx_dropped;
214 counters->ifInErrors = stats.rx_errors;
215 counters->ifInUnknownProtos = -1;
216 counters->ifOutOctets = stats.tx_bytes;
217 counters->ifOutUcastPkts = stats.tx_packets;
218 counters->ifOutMulticastPkts = -1;
219 counters->ifOutBroadcastPkts = -1;
220 counters->ifOutDiscards = stats.tx_dropped;
221 counters->ifOutErrors = stats.tx_errors;
222 counters->ifPromiscuousMode = 0;
224 SFLADD_ELEMENT(cs, &elem);
225 sfl_poller_writeCountersSample(poller, cs);
228 /* Obtains an address to use for the local sFlow agent and stores it into
229 * '*agent_addr'. Returns true if successful, false on failure.
231 * The sFlow agent address should be a local IP address that is persistent and
232 * reachable over the network, if possible. The IP address associated with
233 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
234 * address used to talk to the controller. If the agent device is not
235 * specified then it is figured out by taking a look at the routing table based
238 sflow_choose_agent_address(const char *agent_device,
239 const struct sset *targets,
240 const char *control_ip,
241 SFLAddress *agent_addr)
246 memset(agent_addr, 0, sizeof *agent_addr);
247 agent_addr->type = SFLADDRESSTYPE_IP_V4;
250 if (!netdev_get_in4_by_name(agent_device, &in4)) {
255 SSET_FOR_EACH (target, targets) {
256 struct sockaddr_in sin;
259 if (inet_parse_active(target, SFL_DEFAULT_COLLECTOR_PORT, &sin)
260 && route_table_get_name(sin.sin_addr.s_addr, name)
261 && !netdev_get_in4_by_name(name, &in4)) {
266 if (control_ip && !lookup_ip(control_ip, &in4)) {
270 VLOG_ERR("could not determine IP address for sFlow agent");
274 agent_addr->address.ip_v4.addr = (OVS_FORCE uint32_t) in4.s_addr;
279 dpif_sflow_clear__(struct dpif_sflow *ds) OVS_REQUIRES(mutex)
281 if (ds->sflow_agent) {
282 sfl_agent_release(ds->sflow_agent);
283 ds->sflow_agent = NULL;
285 collectors_destroy(ds->collectors);
286 ds->collectors = NULL;
287 ofproto_sflow_options_destroy(ds->options);
290 /* Turn off sampling to save CPU cycles. */
295 dpif_sflow_clear(struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
297 ovs_mutex_lock(&mutex);
298 dpif_sflow_clear__(ds);
299 ovs_mutex_unlock(&mutex);
303 dpif_sflow_is_enabled(const struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
307 ovs_mutex_lock(&mutex);
308 enabled = ds->collectors != NULL;
309 ovs_mutex_unlock(&mutex);
314 dpif_sflow_create(void)
316 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
317 struct dpif_sflow *ds;
319 if (ovsthread_once_start(&once)) {
320 ovs_mutex_init(&mutex, PTHREAD_MUTEX_RECURSIVE);
321 ovsthread_once_done(&once);
324 ds = xcalloc(1, sizeof *ds);
325 ds->next_tick = time_now() + 1;
326 hmap_init(&ds->ports);
328 route_table_register();
329 atomic_init(&ds->ref_cnt, 1);
335 dpif_sflow_ref(const struct dpif_sflow *ds_)
337 struct dpif_sflow *ds = CONST_CAST(struct dpif_sflow *, ds_);
340 atomic_add(&ds->ref_cnt, 1, &orig);
341 ovs_assert(orig > 0);
346 /* 32-bit fraction of packets to sample with. A value of 0 samples no packets,
347 * a value of %UINT32_MAX samples all packets and intermediate values sample
348 * intermediate fractions of packets. */
350 dpif_sflow_get_probability(const struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
352 uint32_t probability;
353 ovs_mutex_lock(&mutex);
354 probability = ds->probability;
355 ovs_mutex_unlock(&mutex);
360 dpif_sflow_unref(struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
368 atomic_sub(&ds->ref_cnt, 1, &orig);
369 ovs_assert(orig > 0);
371 struct dpif_sflow_port *dsp, *next;
373 route_table_unregister();
374 dpif_sflow_clear(ds);
375 HMAP_FOR_EACH_SAFE (dsp, next, hmap_node, &ds->ports) {
376 dpif_sflow_del_port__(ds, dsp);
378 hmap_destroy(&ds->ports);
384 dpif_sflow_add_poller(struct dpif_sflow *ds, struct dpif_sflow_port *dsp)
387 SFLPoller *poller = sfl_agent_addPoller(ds->sflow_agent, &dsp->dsi, ds,
388 sflow_agent_get_counters);
389 sfl_poller_set_sFlowCpInterval(poller, ds->options->polling_interval);
390 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
391 sfl_poller_set_bridgePort(poller, odp_to_u32(dsp->odp_port));
395 dpif_sflow_add_port(struct dpif_sflow *ds, struct ofport *ofport,
396 odp_port_t odp_port) OVS_EXCLUDED(mutex)
398 struct dpif_sflow_port *dsp;
401 ovs_mutex_lock(&mutex);
402 dpif_sflow_del_port(ds, odp_port);
404 ifindex = netdev_get_ifindex(ofport->netdev);
407 /* Not an ifindex port, so do not add a cross-reference to it here */
411 /* Add to table of ports. */
412 dsp = xmalloc(sizeof *dsp);
413 dsp->ofport = ofport;
414 dsp->odp_port = odp_port;
415 SFL_DS_SET(dsp->dsi, SFL_DSCLASS_IFINDEX, ifindex, 0);
416 hmap_insert(&ds->ports, &dsp->hmap_node, hash_odp_port(odp_port));
419 if (ds->sflow_agent) {
420 dpif_sflow_add_poller(ds, dsp);
424 ovs_mutex_unlock(&mutex);
428 dpif_sflow_del_port__(struct dpif_sflow *ds, struct dpif_sflow_port *dsp)
431 if (ds->sflow_agent) {
432 sfl_agent_removePoller(ds->sflow_agent, &dsp->dsi);
433 sfl_agent_removeSampler(ds->sflow_agent, &dsp->dsi);
435 hmap_remove(&ds->ports, &dsp->hmap_node);
440 dpif_sflow_del_port(struct dpif_sflow *ds, odp_port_t odp_port)
443 struct dpif_sflow_port *dsp;
445 ovs_mutex_lock(&mutex);
446 dsp = dpif_sflow_find_port(ds, odp_port);
448 dpif_sflow_del_port__(ds, dsp);
450 ovs_mutex_unlock(&mutex);
454 dpif_sflow_set_options(struct dpif_sflow *ds,
455 const struct ofproto_sflow_options *options)
458 struct dpif_sflow_port *dsp;
459 bool options_changed;
460 SFLReceiver *receiver;
463 SFLDataSource_instance dsi;
467 ovs_mutex_lock(&mutex);
468 if (sset_is_empty(&options->targets) || !options->sampling_rate) {
469 /* No point in doing any work if there are no targets or nothing to
471 dpif_sflow_clear__(ds);
475 options_changed = (!ds->options
476 || !ofproto_sflow_options_equal(options, ds->options));
478 /* Configure collectors if options have changed or if we're shortchanged in
479 * collectors (which indicates that opening one or more of the configured
480 * collectors failed, so that we should retry). */
482 || collectors_count(ds->collectors) < sset_count(&options->targets)) {
483 collectors_destroy(ds->collectors);
484 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
486 if (ds->collectors == NULL) {
487 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
489 dpif_sflow_clear__(ds);
494 /* Choose agent IP address and agent device (if not yet setup) */
495 if (!sflow_choose_agent_address(options->agent_device,
497 options->control_ip, &agentIP)) {
498 dpif_sflow_clear__(ds);
502 /* Avoid reconfiguring if options didn't change. */
503 if (!options_changed) {
506 ofproto_sflow_options_destroy(ds->options);
507 ds->options = ofproto_sflow_options_clone(options);
510 VLOG_INFO("creating sFlow agent %d", options->sub_id);
511 if (ds->sflow_agent) {
512 sfl_agent_release(ds->sflow_agent);
514 ds->sflow_agent = xcalloc(1, sizeof *ds->sflow_agent);
516 sfl_agent_init(ds->sflow_agent,
519 now, /* Boot time. */
520 now, /* Current time. */
521 ds, /* Pointer supplied to callbacks. */
522 sflow_agent_alloc_cb,
524 sflow_agent_error_cb,
525 sflow_agent_send_packet_cb);
527 receiver = sfl_agent_addReceiver(ds->sflow_agent);
528 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
529 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
531 /* Set the sampling_rate down in the datapath. */
532 ds->probability = MAX(1, UINT32_MAX / ds->options->sampling_rate);
534 /* Add a single sampler for the bridge. This appears as a PHYSICAL_ENTITY
535 because it is associated with the hypervisor, and interacts with the server
536 hardware directly. The sub_id is used to distinguish this sampler from
537 others on other bridges within the same agent. */
538 dsIndex = 1000 + options->sub_id;
539 SFL_DS_SET(dsi, SFL_DSCLASS_PHYSICAL_ENTITY, dsIndex, 0);
540 sampler = sfl_agent_addSampler(ds->sflow_agent, &dsi);
541 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, ds->options->sampling_rate);
542 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, ds->options->header_len);
543 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
545 /* Add pollers for the currently known ifindex-ports */
546 HMAP_FOR_EACH (dsp, hmap_node, &ds->ports) {
547 dpif_sflow_add_poller(ds, dsp);
552 ovs_mutex_unlock(&mutex);
556 dpif_sflow_odp_port_to_ifindex(const struct dpif_sflow *ds,
557 odp_port_t odp_port) OVS_EXCLUDED(mutex)
559 struct dpif_sflow_port *dsp;
562 ovs_mutex_lock(&mutex);
563 dsp = dpif_sflow_find_port(ds, odp_port);
564 ret = dsp ? SFL_DS_INDEX(dsp->dsi) : 0;
565 ovs_mutex_unlock(&mutex);
570 dpif_sflow_received(struct dpif_sflow *ds, struct ofpbuf *packet,
571 const struct flow *flow, odp_port_t odp_in_port,
572 const union user_action_cookie *cookie)
575 SFL_FLOW_SAMPLE_TYPE fs;
576 SFLFlow_sample_element hdrElem;
577 SFLSampled_header *header;
578 SFLFlow_sample_element switchElem;
580 struct dpif_sflow_port *in_dsp;
583 ovs_mutex_lock(&mutex);
584 sampler = ds->sflow_agent->samplers;
589 /* Build a flow sample. */
590 memset(&fs, 0, sizeof fs);
592 /* Look up the input ifIndex if this port has one. Otherwise just
593 * leave it as 0 (meaning 'unknown') and continue. */
594 in_dsp = dpif_sflow_find_port(ds, odp_in_port);
596 fs.input = SFL_DS_INDEX(in_dsp->dsi);
599 /* Make the assumption that the random number generator in the datapath converges
600 * to the configured mean, and just increment the samplePool by the configured
601 * sampling rate every time. */
602 sampler->samplePool += sfl_sampler_get_sFlowFsPacketSamplingRate(sampler);
604 /* Sampled header. */
605 memset(&hdrElem, 0, sizeof hdrElem);
606 hdrElem.tag = SFLFLOW_HEADER;
607 header = &hdrElem.flowType.header;
608 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
609 /* The frame_length should include the Ethernet FCS (4 bytes),
610 * but it has already been stripped, so we need to add 4 here. */
611 header->frame_length = packet->size + 4;
612 /* Ethernet FCS stripped off. */
613 header->stripped = 4;
614 header->header_length = MIN(packet->size,
615 sampler->sFlowFsMaximumHeaderSize);
616 header->header_bytes = packet->data;
618 /* Add extended switch element. */
619 memset(&switchElem, 0, sizeof(switchElem));
620 switchElem.tag = SFLFLOW_EX_SWITCH;
621 switchElem.flowType.sw.src_vlan = vlan_tci_to_vid(flow->vlan_tci);
622 switchElem.flowType.sw.src_priority = vlan_tci_to_pcp(flow->vlan_tci);
624 /* Retrieve data from user_action_cookie. */
625 vlan_tci = cookie->sflow.vlan_tci;
626 switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(vlan_tci);
627 switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(vlan_tci);
629 fs.output = cookie->sflow.output;
631 /* Submit the flow sample to be encoded into the next datagram. */
632 SFLADD_ELEMENT(&fs, &hdrElem);
633 SFLADD_ELEMENT(&fs, &switchElem);
634 sfl_sampler_writeFlowSample(sampler, &fs);
637 ovs_mutex_unlock(&mutex);
641 dpif_sflow_run(struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
643 ovs_mutex_lock(&mutex);
644 if (ds->collectors != NULL) {
645 time_t now = time_now();
647 if (now >= ds->next_tick) {
648 sfl_agent_tick(ds->sflow_agent, time_wall());
649 ds->next_tick = now + 1;
652 ovs_mutex_unlock(&mutex);
656 dpif_sflow_wait(struct dpif_sflow *ds) OVS_EXCLUDED(mutex)
658 ovs_mutex_lock(&mutex);
659 if (ds->collectors != NULL) {
660 poll_timer_wait_until(ds->next_tick * 1000LL);
662 ovs_mutex_unlock(&mutex);