2 * Copyright (c) 2009, 2010 InMon Corp.
3 * Copyright (c) 2009 Nicira Networks.
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-sflow.h"
22 #include "collectors.h"
28 #include "poll-loop.h"
29 #include "port-array.h"
30 #include "sflow_api.h"
31 #include "socket-util.h"
34 #define THIS_MODULE VLM_sflow
37 struct ofproto_sflow_port {
38 struct netdev *netdev; /* Underlying network device, for stats. */
39 SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
42 struct ofproto_sflow {
43 struct ofproto *ofproto;
44 struct collectors *collectors;
45 SFLAgent *sflow_agent;
46 struct ofproto_sflow_options *options;
49 size_t n_flood, n_all;
50 struct port_array ports; /* Indexed by ODP port number. */
53 #define RECEIVER_INDEX 1
55 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
58 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
59 const struct ofproto_sflow_options *b)
61 return (svec_equal(&a->targets, &b->targets)
62 && a->sampling_rate == b->sampling_rate
63 && a->polling_interval == b->polling_interval
64 && a->header_len == b->header_len
65 && a->sub_id == b->sub_id
66 && !strcmp(a->agent_device, b->agent_device)
67 && !strcmp(a->control_ip, b->control_ip));
70 static struct ofproto_sflow_options *
71 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
73 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
74 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
75 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
80 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
83 free(options->agent_device);
84 free(options->control_ip);
89 /* sFlow library callback to allocate memory. */
91 sflow_agent_alloc_cb(void *magic UNUSED, SFLAgent *agent UNUSED, size_t bytes)
93 return calloc(1, bytes);
96 /* sFlow library callback to free memory. */
98 sflow_agent_free_cb(void *magic UNUSED, SFLAgent *agent UNUSED, void *obj)
104 /* sFlow library callback to report error. */
106 sflow_agent_error_cb(void *magic UNUSED, SFLAgent *agent UNUSED, char *msg)
108 VLOG_WARN("sFlow agent error: %s", msg);
111 /* sFlow library callback to send datagram. */
113 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent UNUSED,
114 SFLReceiver *receiver UNUSED, u_char *pkt,
117 struct ofproto_sflow *os = os_;
118 collectors_send(os->collectors, pkt, pktLen);
122 sflow_agent_get_counters(void *os_, SFLPoller *poller,
123 SFL_COUNTERS_SAMPLE_TYPE *cs)
125 struct ofproto_sflow *os = os_;
126 SFLCounters_sample_element elem;
127 struct ofproto_sflow_port *osp;
128 SFLIf_counters *counters;
129 struct netdev_stats stats;
130 enum netdev_flags flags;
133 osp = port_array_get(&os->ports, poller->bridgePort);
138 elem.tag = SFLCOUNTERS_GENERIC;
139 counters = &elem.counterBlock.generic;
140 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
141 counters->ifType = 6;
142 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
143 counters->ifSpeed = netdev_features_to_bps(current);
144 counters->ifDirection = (netdev_features_is_full_duplex(current)
147 counters->ifSpeed = 100000000;
148 counters->ifDirection = 1;
150 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
153 counters->ifStatus = 1; /* ifAdminStatus up. */
154 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
155 counters->ifStatus |= 2; /* ifOperStatus us. */
158 counters->ifStatus = 0; /* Down. */
162 1. Is the multicast counter filled in?
163 2. Does the multicast counter include broadcasts?
164 3. Does the rx_packets counter include multicasts/broadcasts?
166 netdev_get_stats(osp->netdev, &stats);
167 counters->ifInOctets = stats.rx_bytes;
168 counters->ifInUcastPkts = stats.rx_packets;
169 counters->ifInMulticastPkts = stats.multicast;
170 counters->ifInBroadcastPkts = -1;
171 counters->ifInDiscards = stats.rx_dropped;
172 counters->ifInErrors = stats.rx_errors;
173 counters->ifInUnknownProtos = -1;
174 counters->ifOutOctets = stats.tx_bytes;
175 counters->ifOutUcastPkts = stats.tx_packets;
176 counters->ifOutMulticastPkts = -1;
177 counters->ifOutBroadcastPkts = -1;
178 counters->ifOutDiscards = stats.tx_dropped;
179 counters->ifOutErrors = stats.tx_errors;
180 counters->ifPromiscuousMode = 0;
182 SFLADD_ELEMENT(cs, &elem);
183 sfl_poller_writeCountersSample(poller, cs);
186 /* Obtains an address to use for the local sFlow agent and stores it into
187 * '*agent_addr'. Returns true if successful, false on failure.
189 * The sFlow agent address should be a local IP address that is persistent and
190 * reachable over the network, if possible. The IP address associated with
191 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
192 * address used to talk to the controller. */
194 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
195 SFLAddress *agent_addr)
199 memset(agent_addr, 0, sizeof *agent_addr);
200 agent_addr->type = SFLADDRESSTYPE_IP_V4;
203 struct netdev *netdev;
205 if (!netdev_open(agent_device, NETDEV_ETH_TYPE_NONE, &netdev)) {
206 int error = netdev_get_in4(netdev, &in4, NULL);
207 netdev_close(netdev);
214 if (control_ip && !lookup_ip(control_ip, &in4)) {
218 VLOG_ERR("could not determine IP address for sFlow agent");
222 agent_addr->address.ip_v4.addr = in4.s_addr;
227 ofproto_sflow_clear(struct ofproto_sflow *os)
229 struct ofproto_sflow_port *osp;
230 unsigned int odp_port;
232 if (os->sflow_agent) {
233 sfl_agent_release(os->sflow_agent);
234 os->sflow_agent = NULL;
236 collectors_destroy(os->collectors);
237 os->collectors = NULL;
238 ofproto_sflow_options_destroy(os->options);
241 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
242 ofproto_sflow_del_port(os, odp_port);
244 port_array_clear(&os->ports);
246 /* Turn off sampling to save CPU cycles. */
247 dpif_set_sflow_probability(os->dpif, 0);
251 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
253 return os->collectors != NULL;
256 struct ofproto_sflow *
257 ofproto_sflow_create(struct dpif *dpif)
259 struct ofproto_sflow *os;
261 os = xcalloc(1, sizeof *os);
263 os->next_tick = time_now() + 1;
264 port_array_init(&os->ports);
269 ofproto_sflow_destroy(struct ofproto_sflow *os)
272 ofproto_sflow_clear(os);
273 port_array_destroy(&os->ports);
279 ofproto_sflow_add_poller(struct ofproto_sflow *os,
280 struct ofproto_sflow_port *osp, uint16_t odp_port)
282 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
283 sflow_agent_get_counters);
284 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
285 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
286 sfl_poller_set_bridgePort(poller, odp_port);
290 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
291 const char *netdev_name)
293 struct ofproto_sflow_port *osp;
294 struct netdev *netdev;
298 ofproto_sflow_del_port(os, odp_port);
300 /* Open network device. */
301 error = netdev_open(netdev_name, NETDEV_ETH_TYPE_NONE, &netdev);
303 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
304 netdev_name, strerror(error));
308 /* Add to table of ports. */
309 osp = xmalloc(sizeof *osp);
310 osp->netdev = netdev;
311 ifindex = netdev_get_ifindex(netdev);
313 ifindex = (os->sflow_agent->subId << 16) + odp_port;
315 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
316 port_array_set(&os->ports, odp_port, osp);
319 if (os->sflow_agent) {
320 ofproto_sflow_add_poller(os, osp, odp_port);
325 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
327 struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
329 if (os->sflow_agent) {
330 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
332 netdev_close(osp->netdev);
334 port_array_set(&os->ports, odp_port, NULL);
339 ofproto_sflow_set_options(struct ofproto_sflow *os,
340 const struct ofproto_sflow_options *options)
342 struct ofproto_sflow_port *osp;
343 SFLDataSource_instance dsi;
344 bool options_changed;
346 SFLReceiver *receiver;
347 unsigned int odp_port;
352 options_changed = (!os->options
353 || !ofproto_sflow_options_equal(options, os->options));
355 /* Configure collectors if options have changed or if we're shortchanged in
356 * collectors (which indicates that opening one or more of the configured
357 * collectors failed, so that we should retry). */
359 || collectors_count(os->collectors) < options->targets.n) {
360 collectors_destroy(os->collectors);
361 error = collectors_create(&options->targets,
362 SFL_DEFAULT_COLLECTOR_PORT, &os->collectors);
363 if (os->collectors == NULL) {
364 VLOG_WARN_RL(&rl, "no configured collectors, sFlow disabled");
365 ofproto_sflow_clear(os);
370 /* Avoid reconfiguring if options didn't change. */
371 if (!options_changed) {
374 ofproto_sflow_options_destroy(os->options);
375 os->options = ofproto_sflow_options_clone(options);
377 /* Choose agent IP address. */
378 if (!sflow_choose_agent_address(options->agent_device,
379 options->control_ip, &agentIP)) {
380 ofproto_sflow_clear(os);
385 VLOG_INFO("creating sFlow agent %d", options->sub_id);
386 if (os->sflow_agent) {
387 sfl_agent_release(os->sflow_agent);
389 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
391 sfl_agent_init(os->sflow_agent,
394 now, /* Boot time. */
395 now, /* Current time. */
396 os, /* Pointer supplied to callbacks. */
397 sflow_agent_alloc_cb,
399 sflow_agent_error_cb,
400 sflow_agent_send_packet_cb);
402 receiver = sfl_agent_addReceiver(os->sflow_agent);
403 sfl_receiver_set_sFlowRcvrOwner(receiver, "OpenVSwitch sFlow");
404 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
406 /* Add a single sampler to represent the whole switch (special <ifIndex>:0
407 * datasource). The alternative is to model a physical switch more closely
408 * and instantiate a separate sampler object for each interface, but then
409 * unicasts would have to be offered to two samplers, and
410 * broadcasts/multicasts would have to be offered to all of them. Doing it
411 * this way with a single <ifindex>:0 sampler is much more efficient for a
412 * virtual switch, and is allowed by the sFlow standard.
414 SFL_DS_SET(dsi, 0, 0, 0);
415 sampler = sfl_agent_addSampler(os->sflow_agent, &dsi);
416 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
417 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, options->sampling_rate);
418 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, options->header_len);
420 /* Set the sampling_rate down in the datapath. */
421 dpif_set_sflow_probability(os->dpif,
422 MAX(1, UINT32_MAX / options->sampling_rate));
424 /* Add the currently known ports. */
425 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
426 ofproto_sflow_add_poller(os, osp, odp_port);
431 ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
433 SFL_FLOW_SAMPLE_TYPE fs;
434 SFLFlow_sample_element hdrElem;
435 SFLSampled_header *header;
436 SFLFlow_sample_element switchElem;
437 SFLSampler *sampler = os->sflow_agent->samplers;
438 const struct odp_sflow_sample_header *hdr;
439 const union odp_action *actions;
440 struct ofpbuf payload;
441 size_t n_actions, n_outputs;
446 /* Get odp_sflow_sample_header. */
447 min_size = sizeof *msg + sizeof *hdr;
448 if (min_size > msg->length) {
449 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
450 msg->length, min_size);
453 hdr = (const struct odp_sflow_sample_header *) (msg + 1);
456 n_actions = hdr->n_actions;
457 if (n_actions > 65536 / sizeof *actions) {
458 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%"PRIu32" > %zu)",
459 65536 / sizeof *actions, n_actions);
462 min_size += n_actions * sizeof *actions;
463 if (min_size > msg->length) {
464 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
466 n_actions, msg->length, min_size);
469 actions = (const union odp_action *) (hdr + 1);
471 /* Get packet payload and extract flow. */
472 payload.data = (union odp_action *) (actions + n_actions);
473 payload.size = msg->length - min_size;
474 flow_extract(&payload, msg->port, &flow);
476 /* Build a flow sample */
477 memset(&fs, 0, sizeof fs);
478 fs.input = msg->port == ODPP_LOCAL ? 0x3fffffff : msg->port;
479 fs.output = 0; /* Filled in correctly below. */
480 fs.sample_pool = hdr->sample_pool;
482 /* Sampled header. */
483 memset(&hdrElem, 0, sizeof hdrElem);
484 hdrElem.tag = SFLFLOW_HEADER;
485 header = &hdrElem.flowType.header;
486 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
487 header->frame_length = payload.size;
488 header->stripped = 4; /* Ethernet FCS stripped off. */
489 header->header_length = MIN(payload.size,
490 sampler->sFlowFsMaximumHeaderSize);
491 header->header_bytes = payload.data;
493 /* Add extended switch element. */
494 memset(&switchElem, 0, sizeof(switchElem));
495 switchElem.tag = SFLFLOW_EX_SWITCH;
496 switchElem.flowType.sw.src_vlan = flow.dl_vlan;
497 switchElem.flowType.sw.src_priority = -1; /* XXX */
498 switchElem.flowType.sw.dst_vlan = -1; /* Filled in correctly below. */
499 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
501 /* Figure out the output ports. */
503 for (i = 0; i < n_actions; i++) {
504 const union odp_action *a = &actions[i];
508 fs.output = a->output.port;
512 case ODPAT_OUTPUT_GROUP:
513 n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
514 : a->output_group.group == DP_GROUP_ALL ? os->n_all
518 case ODPAT_SET_VLAN_VID:
519 switchElem.flowType.sw.dst_vlan = a->vlan_vid.vlan_vid;
522 case ODPAT_SET_VLAN_PCP:
523 switchElem.flowType.sw.dst_priority = a->vlan_pcp.vlan_pcp;
530 if (n_outputs > 1 || !fs.output) {
531 /* Setting the high bit means "multiple output ports". */
532 fs.output = 0x80000000 | n_outputs;
535 /* Submit the flow sample to be encoded into the next datagram. */
536 SFLADD_ELEMENT(&fs, &hdrElem);
537 SFLADD_ELEMENT(&fs, &switchElem);
538 sfl_sampler_writeFlowSample(sampler, &fs);
542 ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
543 size_t n_flood, size_t n_all)
545 os->n_flood = n_flood;
550 ofproto_sflow_run(struct ofproto_sflow *os)
552 if (ofproto_sflow_is_enabled(os)) {
553 time_t now = time_now();
554 if (now >= os->next_tick) {
555 sfl_agent_tick(os->sflow_agent, now);
556 os->next_tick = now + 1;
562 ofproto_sflow_wait(struct ofproto_sflow *os)
564 if (ofproto_sflow_is_enabled(os)) {
565 poll_timer_wait(os->next_tick * 1000 - time_msec());