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 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. */
63 static void dpif_sflow_del_port__(struct dpif_sflow *,
64 struct dpif_sflow_port *);
66 #define RECEIVER_INDEX 1
68 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
71 nullable_string_is_equal(const char *a, const char *b)
73 return a ? b && !strcmp(a, b) : !b;
77 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
78 const struct ofproto_sflow_options *b)
80 return (sset_equals(&a->targets, &b->targets)
81 && a->sampling_rate == b->sampling_rate
82 && a->polling_interval == b->polling_interval
83 && a->header_len == b->header_len
84 && a->sub_id == b->sub_id
85 && nullable_string_is_equal(a->agent_device, b->agent_device)
86 && nullable_string_is_equal(a->control_ip, b->control_ip));
89 static struct ofproto_sflow_options *
90 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
92 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
93 sset_clone(&new->targets, &old->targets);
94 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
95 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
100 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
103 sset_destroy(&options->targets);
104 free(options->agent_device);
105 free(options->control_ip);
110 /* sFlow library callback to allocate memory. */
112 sflow_agent_alloc_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
115 return calloc(1, bytes);
118 /* sFlow library callback to free memory. */
120 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
127 /* sFlow library callback to report error. */
129 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
132 VLOG_WARN("sFlow agent error: %s", msg);
135 /* sFlow library callback to send datagram. */
137 sflow_agent_send_packet_cb(void *ds_, SFLAgent *agent OVS_UNUSED,
138 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
141 struct dpif_sflow *ds = ds_;
142 collectors_send(ds->collectors, pkt, pktLen);
145 static struct dpif_sflow_port *
146 dpif_sflow_find_port(const struct dpif_sflow *ds, odp_port_t odp_port)
148 struct dpif_sflow_port *dsp;
150 HMAP_FOR_EACH_IN_BUCKET (dsp, hmap_node, hash_odp_port(odp_port),
152 if (dsp->odp_port == odp_port) {
160 sflow_agent_get_counters(void *ds_, SFLPoller *poller,
161 SFL_COUNTERS_SAMPLE_TYPE *cs)
163 struct dpif_sflow *ds = ds_;
164 SFLCounters_sample_element elem;
165 enum netdev_features current;
166 struct dpif_sflow_port *dsp;
167 SFLIf_counters *counters;
168 struct netdev_stats stats;
169 enum netdev_flags flags;
171 dsp = dpif_sflow_find_port(ds, u32_to_odp(poller->bridgePort));
176 elem.tag = SFLCOUNTERS_GENERIC;
177 counters = &elem.counterBlock.generic;
178 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
179 counters->ifType = 6;
180 if (!netdev_get_features(dsp->ofport->netdev, ¤t, NULL, NULL, NULL)) {
181 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
182 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
183 counters->ifSpeed = netdev_features_to_bps(current, 0);
184 counters->ifDirection = (netdev_features_is_full_duplex(current)
187 counters->ifSpeed = 100000000;
188 counters->ifDirection = 0;
190 if (!netdev_get_flags(dsp->ofport->netdev, &flags) && flags & NETDEV_UP) {
191 counters->ifStatus = 1; /* ifAdminStatus up. */
192 if (netdev_get_carrier(dsp->ofport->netdev)) {
193 counters->ifStatus |= 2; /* ifOperStatus us. */
196 counters->ifStatus = 0; /* Down. */
200 1. Is the multicast counter filled in?
201 2. Does the multicast counter include broadcasts?
202 3. Does the rx_packets counter include multicasts/broadcasts?
204 ofproto_port_get_stats(dsp->ofport, &stats);
205 counters->ifInOctets = stats.rx_bytes;
206 counters->ifInUcastPkts = stats.rx_packets;
207 counters->ifInMulticastPkts = stats.multicast;
208 counters->ifInBroadcastPkts = -1;
209 counters->ifInDiscards = stats.rx_dropped;
210 counters->ifInErrors = stats.rx_errors;
211 counters->ifInUnknownProtos = -1;
212 counters->ifOutOctets = stats.tx_bytes;
213 counters->ifOutUcastPkts = stats.tx_packets;
214 counters->ifOutMulticastPkts = -1;
215 counters->ifOutBroadcastPkts = -1;
216 counters->ifOutDiscards = stats.tx_dropped;
217 counters->ifOutErrors = stats.tx_errors;
218 counters->ifPromiscuousMode = 0;
220 SFLADD_ELEMENT(cs, &elem);
221 sfl_poller_writeCountersSample(poller, cs);
224 /* Obtains an address to use for the local sFlow agent and stores it into
225 * '*agent_addr'. Returns true if successful, false on failure.
227 * The sFlow agent address should be a local IP address that is persistent and
228 * reachable over the network, if possible. The IP address associated with
229 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
230 * address used to talk to the controller. If the agent device is not
231 * specified then it is figured out by taking a look at the routing table based
234 sflow_choose_agent_address(const char *agent_device,
235 const struct sset *targets,
236 const char *control_ip,
237 SFLAddress *agent_addr)
242 memset(agent_addr, 0, sizeof *agent_addr);
243 agent_addr->type = SFLADDRESSTYPE_IP_V4;
246 if (!netdev_get_in4_by_name(agent_device, &in4)) {
251 SSET_FOR_EACH (target, targets) {
252 struct sockaddr_in sin;
255 if (inet_parse_active(target, SFL_DEFAULT_COLLECTOR_PORT, &sin)
256 && route_table_get_name(sin.sin_addr.s_addr, name)
257 && !netdev_get_in4_by_name(name, &in4)) {
262 if (control_ip && !lookup_ip(control_ip, &in4)) {
266 VLOG_ERR("could not determine IP address for sFlow agent");
270 agent_addr->address.ip_v4.addr = (OVS_FORCE uint32_t) in4.s_addr;
275 dpif_sflow_clear(struct dpif_sflow *ds)
277 if (ds->sflow_agent) {
278 sfl_agent_release(ds->sflow_agent);
279 ds->sflow_agent = NULL;
281 collectors_destroy(ds->collectors);
282 ds->collectors = NULL;
283 ofproto_sflow_options_destroy(ds->options);
286 /* Turn off sampling to save CPU cycles. */
291 dpif_sflow_is_enabled(const struct dpif_sflow *ds)
293 return ds->collectors != NULL;
297 dpif_sflow_create(void)
299 struct dpif_sflow *ds;
301 ds = xcalloc(1, sizeof *ds);
302 ds->next_tick = time_now() + 1;
303 hmap_init(&ds->ports);
305 route_table_register();
312 dpif_sflow_ref(const struct dpif_sflow *ds_)
314 struct dpif_sflow *ds = CONST_CAST(struct dpif_sflow *, ds_);
316 ovs_assert(ds->ref_cnt > 0);
322 /* 32-bit fraction of packets to sample with. A value of 0 samples no packets,
323 * a value of %UINT32_MAX samples all packets and intermediate values sample
324 * intermediate fractions of packets. */
326 dpif_sflow_get_probability(const struct dpif_sflow *ds)
328 return ds->probability;
332 dpif_sflow_unref(struct dpif_sflow *ds)
338 ovs_assert(ds->ref_cnt > 0);
339 if (!--ds->ref_cnt) {
340 struct dpif_sflow_port *dsp, *next;
342 route_table_unregister();
343 dpif_sflow_clear(ds);
344 HMAP_FOR_EACH_SAFE (dsp, next, hmap_node, &ds->ports) {
345 dpif_sflow_del_port__(ds, dsp);
347 hmap_destroy(&ds->ports);
353 dpif_sflow_add_poller(struct dpif_sflow *ds, struct dpif_sflow_port *dsp)
355 SFLPoller *poller = sfl_agent_addPoller(ds->sflow_agent, &dsp->dsi, ds,
356 sflow_agent_get_counters);
357 sfl_poller_set_sFlowCpInterval(poller, ds->options->polling_interval);
358 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
359 sfl_poller_set_bridgePort(poller, odp_to_u32(dsp->odp_port));
363 dpif_sflow_add_port(struct dpif_sflow *ds, struct ofport *ofport,
366 struct dpif_sflow_port *dsp;
369 dpif_sflow_del_port(ds, odp_port);
371 ifindex = netdev_get_ifindex(ofport->netdev);
374 /* Not an ifindex port, so do not add a cross-reference to it here */
378 /* Add to table of ports. */
379 dsp = xmalloc(sizeof *dsp);
380 dsp->ofport = ofport;
381 dsp->odp_port = odp_port;
382 SFL_DS_SET(dsp->dsi, SFL_DSCLASS_IFINDEX, ifindex, 0);
383 hmap_insert(&ds->ports, &dsp->hmap_node, hash_odp_port(odp_port));
386 if (ds->sflow_agent) {
387 dpif_sflow_add_poller(ds, dsp);
392 dpif_sflow_del_port__(struct dpif_sflow *ds, struct dpif_sflow_port *dsp)
394 if (ds->sflow_agent) {
395 sfl_agent_removePoller(ds->sflow_agent, &dsp->dsi);
396 sfl_agent_removeSampler(ds->sflow_agent, &dsp->dsi);
398 hmap_remove(&ds->ports, &dsp->hmap_node);
403 dpif_sflow_del_port(struct dpif_sflow *ds, odp_port_t odp_port)
405 struct dpif_sflow_port *dsp = dpif_sflow_find_port(ds, odp_port);
407 dpif_sflow_del_port__(ds, dsp);
412 dpif_sflow_set_options(struct dpif_sflow *ds,
413 const struct ofproto_sflow_options *options)
415 struct dpif_sflow_port *dsp;
416 bool options_changed;
417 SFLReceiver *receiver;
420 SFLDataSource_instance dsi;
424 if (sset_is_empty(&options->targets) || !options->sampling_rate) {
425 /* No point in doing any work if there are no targets or nothing to
427 dpif_sflow_clear(ds);
431 options_changed = (!ds->options
432 || !ofproto_sflow_options_equal(options, ds->options));
434 /* Configure collectors if options have changed or if we're shortchanged in
435 * collectors (which indicates that opening one or more of the configured
436 * collectors failed, so that we should retry). */
438 || collectors_count(ds->collectors) < sset_count(&options->targets)) {
439 collectors_destroy(ds->collectors);
440 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
442 if (ds->collectors == NULL) {
443 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
445 dpif_sflow_clear(ds);
450 /* Choose agent IP address and agent device (if not yet setup) */
451 if (!sflow_choose_agent_address(options->agent_device,
453 options->control_ip, &agentIP)) {
454 dpif_sflow_clear(ds);
458 /* Avoid reconfiguring if options didn't change. */
459 if (!options_changed) {
462 ofproto_sflow_options_destroy(ds->options);
463 ds->options = ofproto_sflow_options_clone(options);
466 VLOG_INFO("creating sFlow agent %d", options->sub_id);
467 if (ds->sflow_agent) {
468 sfl_agent_release(ds->sflow_agent);
470 ds->sflow_agent = xcalloc(1, sizeof *ds->sflow_agent);
472 sfl_agent_init(ds->sflow_agent,
475 now, /* Boot time. */
476 now, /* Current time. */
477 ds, /* Pointer supplied to callbacks. */
478 sflow_agent_alloc_cb,
480 sflow_agent_error_cb,
481 sflow_agent_send_packet_cb);
483 receiver = sfl_agent_addReceiver(ds->sflow_agent);
484 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
485 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
487 /* Set the sampling_rate down in the datapath. */
488 ds->probability = MAX(1, UINT32_MAX / ds->options->sampling_rate);
490 /* Add a single sampler for the bridge. This appears as a PHYSICAL_ENTITY
491 because it is associated with the hypervisor, and interacts with the server
492 hardware directly. The sub_id is used to distinguish this sampler from
493 others on other bridges within the same agent. */
494 dsIndex = 1000 + options->sub_id;
495 SFL_DS_SET(dsi, SFL_DSCLASS_PHYSICAL_ENTITY, dsIndex, 0);
496 sampler = sfl_agent_addSampler(ds->sflow_agent, &dsi);
497 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, ds->options->sampling_rate);
498 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, ds->options->header_len);
499 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
501 /* Add pollers for the currently known ifindex-ports */
502 HMAP_FOR_EACH (dsp, hmap_node, &ds->ports) {
503 dpif_sflow_add_poller(ds, dsp);
508 dpif_sflow_odp_port_to_ifindex(const struct dpif_sflow *ds,
511 struct dpif_sflow_port *dsp = dpif_sflow_find_port(ds, odp_port);
512 return dsp ? SFL_DS_INDEX(dsp->dsi) : 0;
516 dpif_sflow_received(struct dpif_sflow *ds, struct ofpbuf *packet,
517 const struct flow *flow, odp_port_t odp_in_port,
518 const union user_action_cookie *cookie)
520 SFL_FLOW_SAMPLE_TYPE fs;
521 SFLFlow_sample_element hdrElem;
522 SFLSampled_header *header;
523 SFLFlow_sample_element switchElem;
525 struct dpif_sflow_port *in_dsp;
528 sampler = ds->sflow_agent->samplers;
533 /* Build a flow sample. */
534 memset(&fs, 0, sizeof fs);
536 /* Look up the input ifIndex if this port has one. Otherwise just
537 * leave it as 0 (meaning 'unknown') and continue. */
538 in_dsp = dpif_sflow_find_port(ds, odp_in_port);
540 fs.input = SFL_DS_INDEX(in_dsp->dsi);
543 /* Make the assumption that the random number generator in the datapath converges
544 * to the configured mean, and just increment the samplePool by the configured
545 * sampling rate every time. */
546 sampler->samplePool += sfl_sampler_get_sFlowFsPacketSamplingRate(sampler);
548 /* Sampled header. */
549 memset(&hdrElem, 0, sizeof hdrElem);
550 hdrElem.tag = SFLFLOW_HEADER;
551 header = &hdrElem.flowType.header;
552 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
553 /* The frame_length should include the Ethernet FCS (4 bytes),
554 * but it has already been stripped, so we need to add 4 here. */
555 header->frame_length = packet->size + 4;
556 /* Ethernet FCS stripped off. */
557 header->stripped = 4;
558 header->header_length = MIN(packet->size,
559 sampler->sFlowFsMaximumHeaderSize);
560 header->header_bytes = packet->data;
562 /* Add extended switch element. */
563 memset(&switchElem, 0, sizeof(switchElem));
564 switchElem.tag = SFLFLOW_EX_SWITCH;
565 switchElem.flowType.sw.src_vlan = vlan_tci_to_vid(flow->vlan_tci);
566 switchElem.flowType.sw.src_priority = vlan_tci_to_pcp(flow->vlan_tci);
568 /* Retrieve data from user_action_cookie. */
569 vlan_tci = cookie->sflow.vlan_tci;
570 switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(vlan_tci);
571 switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(vlan_tci);
573 fs.output = cookie->sflow.output;
575 /* Submit the flow sample to be encoded into the next datagram. */
576 SFLADD_ELEMENT(&fs, &hdrElem);
577 SFLADD_ELEMENT(&fs, &switchElem);
578 sfl_sampler_writeFlowSample(sampler, &fs);
582 dpif_sflow_run(struct dpif_sflow *ds)
584 if (dpif_sflow_is_enabled(ds)) {
585 time_t now = time_now();
587 if (now >= ds->next_tick) {
588 sfl_agent_tick(ds->sflow_agent, time_wall());
589 ds->next_tick = now + 1;
595 dpif_sflow_wait(struct dpif_sflow *ds)
597 if (dpif_sflow_is_enabled(ds)) {
598 poll_timer_wait_until(ds->next_tick * 1000LL);