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 nullable_string_is_equal(const char *a, const char *b)
60 return a ? b && !strcmp(a, b) : !b;
64 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
65 const struct ofproto_sflow_options *b)
67 return (svec_equal(&a->targets, &b->targets)
68 && a->sampling_rate == b->sampling_rate
69 && a->polling_interval == b->polling_interval
70 && a->header_len == b->header_len
71 && a->sub_id == b->sub_id
72 && nullable_string_is_equal(a->agent_device, b->agent_device)
73 && nullable_string_is_equal(a->control_ip, b->control_ip));
76 static struct ofproto_sflow_options *
77 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
79 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
80 svec_clone(&new->targets, &old->targets);
81 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
82 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
87 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
90 svec_destroy(&options->targets);
91 free(options->agent_device);
92 free(options->control_ip);
97 /* sFlow library callback to allocate memory. */
99 sflow_agent_alloc_cb(void *magic UNUSED, SFLAgent *agent UNUSED, size_t bytes)
101 return calloc(1, bytes);
104 /* sFlow library callback to free memory. */
106 sflow_agent_free_cb(void *magic UNUSED, SFLAgent *agent UNUSED, void *obj)
112 /* sFlow library callback to report error. */
114 sflow_agent_error_cb(void *magic UNUSED, SFLAgent *agent UNUSED, char *msg)
116 VLOG_WARN("sFlow agent error: %s", msg);
119 /* sFlow library callback to send datagram. */
121 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent UNUSED,
122 SFLReceiver *receiver UNUSED, u_char *pkt,
125 struct ofproto_sflow *os = os_;
126 collectors_send(os->collectors, pkt, pktLen);
130 sflow_agent_get_counters(void *os_, SFLPoller *poller,
131 SFL_COUNTERS_SAMPLE_TYPE *cs)
133 struct ofproto_sflow *os = os_;
134 SFLCounters_sample_element elem;
135 struct ofproto_sflow_port *osp;
136 SFLIf_counters *counters;
137 struct netdev_stats stats;
138 enum netdev_flags flags;
141 osp = port_array_get(&os->ports, poller->bridgePort);
146 elem.tag = SFLCOUNTERS_GENERIC;
147 counters = &elem.counterBlock.generic;
148 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
149 counters->ifType = 6;
150 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
151 counters->ifSpeed = netdev_features_to_bps(current);
152 counters->ifDirection = (netdev_features_is_full_duplex(current)
155 counters->ifSpeed = 100000000;
156 counters->ifDirection = 1;
158 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
161 counters->ifStatus = 1; /* ifAdminStatus up. */
162 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
163 counters->ifStatus |= 2; /* ifOperStatus us. */
166 counters->ifStatus = 0; /* Down. */
170 1. Is the multicast counter filled in?
171 2. Does the multicast counter include broadcasts?
172 3. Does the rx_packets counter include multicasts/broadcasts?
174 netdev_get_stats(osp->netdev, &stats);
175 counters->ifInOctets = stats.rx_bytes;
176 counters->ifInUcastPkts = stats.rx_packets;
177 counters->ifInMulticastPkts = stats.multicast;
178 counters->ifInBroadcastPkts = -1;
179 counters->ifInDiscards = stats.rx_dropped;
180 counters->ifInErrors = stats.rx_errors;
181 counters->ifInUnknownProtos = -1;
182 counters->ifOutOctets = stats.tx_bytes;
183 counters->ifOutUcastPkts = stats.tx_packets;
184 counters->ifOutMulticastPkts = -1;
185 counters->ifOutBroadcastPkts = -1;
186 counters->ifOutDiscards = stats.tx_dropped;
187 counters->ifOutErrors = stats.tx_errors;
188 counters->ifPromiscuousMode = 0;
190 SFLADD_ELEMENT(cs, &elem);
191 sfl_poller_writeCountersSample(poller, cs);
194 /* Obtains an address to use for the local sFlow agent and stores it into
195 * '*agent_addr'. Returns true if successful, false on failure.
197 * The sFlow agent address should be a local IP address that is persistent and
198 * reachable over the network, if possible. The IP address associated with
199 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
200 * address used to talk to the controller. */
202 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
203 SFLAddress *agent_addr)
207 memset(agent_addr, 0, sizeof *agent_addr);
208 agent_addr->type = SFLADDRESSTYPE_IP_V4;
211 struct netdev *netdev;
213 if (!netdev_open(agent_device, NETDEV_ETH_TYPE_NONE, &netdev)) {
214 int error = netdev_get_in4(netdev, &in4, NULL);
215 netdev_close(netdev);
222 if (control_ip && !lookup_ip(control_ip, &in4)) {
226 VLOG_ERR("could not determine IP address for sFlow agent");
230 agent_addr->address.ip_v4.addr = in4.s_addr;
235 ofproto_sflow_clear(struct ofproto_sflow *os)
237 struct ofproto_sflow_port *osp;
238 unsigned int odp_port;
240 if (os->sflow_agent) {
241 sfl_agent_release(os->sflow_agent);
242 os->sflow_agent = NULL;
244 collectors_destroy(os->collectors);
245 os->collectors = NULL;
246 ofproto_sflow_options_destroy(os->options);
249 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
250 ofproto_sflow_del_port(os, odp_port);
252 port_array_clear(&os->ports);
254 /* Turn off sampling to save CPU cycles. */
255 dpif_set_sflow_probability(os->dpif, 0);
259 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
261 return os->collectors != NULL;
264 struct ofproto_sflow *
265 ofproto_sflow_create(struct dpif *dpif)
267 struct ofproto_sflow *os;
269 os = xcalloc(1, sizeof *os);
271 os->next_tick = time_now() + 1;
272 port_array_init(&os->ports);
277 ofproto_sflow_destroy(struct ofproto_sflow *os)
280 ofproto_sflow_clear(os);
281 port_array_destroy(&os->ports);
287 ofproto_sflow_add_poller(struct ofproto_sflow *os,
288 struct ofproto_sflow_port *osp, uint16_t odp_port)
290 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
291 sflow_agent_get_counters);
292 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
293 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
294 sfl_poller_set_bridgePort(poller, odp_port);
298 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
299 const char *netdev_name)
301 struct ofproto_sflow_port *osp;
302 struct netdev *netdev;
306 ofproto_sflow_del_port(os, odp_port);
308 /* Open network device. */
309 error = netdev_open(netdev_name, NETDEV_ETH_TYPE_NONE, &netdev);
311 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
312 netdev_name, strerror(error));
316 /* Add to table of ports. */
317 osp = xmalloc(sizeof *osp);
318 osp->netdev = netdev;
319 ifindex = netdev_get_ifindex(netdev);
321 ifindex = (os->sflow_agent->subId << 16) + odp_port;
323 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
324 port_array_set(&os->ports, odp_port, osp);
327 if (os->sflow_agent) {
328 ofproto_sflow_add_poller(os, osp, odp_port);
333 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
335 struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
337 if (os->sflow_agent) {
338 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
340 netdev_close(osp->netdev);
342 port_array_set(&os->ports, odp_port, NULL);
347 ofproto_sflow_set_options(struct ofproto_sflow *os,
348 const struct ofproto_sflow_options *options)
350 struct ofproto_sflow_port *osp;
351 SFLDataSource_instance dsi;
352 bool options_changed;
354 SFLReceiver *receiver;
355 unsigned int odp_port;
360 options_changed = (!os->options
361 || !ofproto_sflow_options_equal(options, os->options));
363 /* Configure collectors if options have changed or if we're shortchanged in
364 * collectors (which indicates that opening one or more of the configured
365 * collectors failed, so that we should retry). */
367 || collectors_count(os->collectors) < options->targets.n) {
368 collectors_destroy(os->collectors);
369 error = collectors_create(&options->targets,
370 SFL_DEFAULT_COLLECTOR_PORT, &os->collectors);
371 if (os->collectors == NULL) {
372 VLOG_WARN_RL(&rl, "no configured collectors, sFlow disabled");
373 ofproto_sflow_clear(os);
378 /* Avoid reconfiguring if options didn't change. */
379 if (!options_changed) {
382 ofproto_sflow_options_destroy(os->options);
383 os->options = ofproto_sflow_options_clone(options);
385 /* Choose agent IP address. */
386 if (!sflow_choose_agent_address(options->agent_device,
387 options->control_ip, &agentIP)) {
388 ofproto_sflow_clear(os);
393 VLOG_INFO("creating sFlow agent %d", options->sub_id);
394 if (os->sflow_agent) {
395 sfl_agent_release(os->sflow_agent);
397 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
399 sfl_agent_init(os->sflow_agent,
402 now, /* Boot time. */
403 now, /* Current time. */
404 os, /* Pointer supplied to callbacks. */
405 sflow_agent_alloc_cb,
407 sflow_agent_error_cb,
408 sflow_agent_send_packet_cb);
410 receiver = sfl_agent_addReceiver(os->sflow_agent);
411 sfl_receiver_set_sFlowRcvrOwner(receiver, "OpenVSwitch sFlow");
412 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
414 /* Add a single sampler to represent the whole switch (special <ifIndex>:0
415 * datasource). The alternative is to model a physical switch more closely
416 * and instantiate a separate sampler object for each interface, but then
417 * unicasts would have to be offered to two samplers, and
418 * broadcasts/multicasts would have to be offered to all of them. Doing it
419 * this way with a single <ifindex>:0 sampler is much more efficient for a
420 * virtual switch, and is allowed by the sFlow standard.
422 SFL_DS_SET(dsi, 0, 0, 0);
423 sampler = sfl_agent_addSampler(os->sflow_agent, &dsi);
424 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
425 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, options->sampling_rate);
426 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, options->header_len);
428 /* Set the sampling_rate down in the datapath. */
429 dpif_set_sflow_probability(os->dpif,
430 MAX(1, UINT32_MAX / options->sampling_rate));
432 /* Add the currently known ports. */
433 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
434 ofproto_sflow_add_poller(os, osp, odp_port);
439 ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os,
442 struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
443 return osp ? SFL_DS_INDEX(osp->dsi) : 0;
447 ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
449 SFL_FLOW_SAMPLE_TYPE fs;
450 SFLFlow_sample_element hdrElem;
451 SFLSampled_header *header;
452 SFLFlow_sample_element switchElem;
453 SFLSampler *sampler = os->sflow_agent->samplers;
454 const struct odp_sflow_sample_header *hdr;
455 const union odp_action *actions;
456 struct ofpbuf payload;
457 size_t n_actions, n_outputs;
462 /* Get odp_sflow_sample_header. */
463 min_size = sizeof *msg + sizeof *hdr;
464 if (min_size > msg->length) {
465 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
466 msg->length, min_size);
469 hdr = (const struct odp_sflow_sample_header *) (msg + 1);
472 n_actions = hdr->n_actions;
473 if (n_actions > 65536 / sizeof *actions) {
474 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%"PRIu32" > %zu)",
475 65536 / sizeof *actions, n_actions);
478 min_size += n_actions * sizeof *actions;
479 if (min_size > msg->length) {
480 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
482 n_actions, msg->length, min_size);
485 actions = (const union odp_action *) (hdr + 1);
487 /* Get packet payload and extract flow. */
488 payload.data = (union odp_action *) (actions + n_actions);
489 payload.size = msg->length - min_size;
490 flow_extract(&payload, msg->port, &flow);
492 /* Build a flow sample */
493 memset(&fs, 0, sizeof fs);
494 fs.input = ofproto_sflow_odp_port_to_ifindex(os, msg->port);
495 fs.output = 0; /* Filled in correctly below. */
496 fs.sample_pool = hdr->sample_pool;
498 /* Sampled header. */
499 memset(&hdrElem, 0, sizeof hdrElem);
500 hdrElem.tag = SFLFLOW_HEADER;
501 header = &hdrElem.flowType.header;
502 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
503 header->frame_length = payload.size;
504 header->stripped = 4; /* Ethernet FCS stripped off. */
505 header->header_length = MIN(payload.size,
506 sampler->sFlowFsMaximumHeaderSize);
507 header->header_bytes = payload.data;
509 /* Add extended switch element. */
510 memset(&switchElem, 0, sizeof(switchElem));
511 switchElem.tag = SFLFLOW_EX_SWITCH;
512 switchElem.flowType.sw.src_vlan = flow.dl_vlan;
513 switchElem.flowType.sw.src_priority = -1; /* XXX */
514 switchElem.flowType.sw.dst_vlan = -1; /* Filled in correctly below. */
515 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
517 /* Figure out the output ports. */
519 for (i = 0; i < n_actions; i++) {
520 const union odp_action *a = &actions[i];
524 fs.output = ofproto_sflow_odp_port_to_ifindex(os, a->output.port);
528 case ODPAT_OUTPUT_GROUP:
529 n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
530 : a->output_group.group == DP_GROUP_ALL ? os->n_all
534 case ODPAT_SET_VLAN_VID:
535 switchElem.flowType.sw.dst_vlan = a->vlan_vid.vlan_vid;
538 case ODPAT_SET_VLAN_PCP:
539 switchElem.flowType.sw.dst_priority = a->vlan_pcp.vlan_pcp;
547 /* This value indicates that the packet was dropped for an unknown
549 fs.output = 0x40000000 | 256;
550 } else if (n_outputs > 1 || !fs.output) {
551 /* Setting the high bit means "multiple output ports". */
552 fs.output = 0x80000000 | n_outputs;
555 /* Submit the flow sample to be encoded into the next datagram. */
556 SFLADD_ELEMENT(&fs, &hdrElem);
557 SFLADD_ELEMENT(&fs, &switchElem);
558 sfl_sampler_writeFlowSample(sampler, &fs);
562 ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
563 size_t n_flood, size_t n_all)
565 os->n_flood = n_flood;
570 ofproto_sflow_run(struct ofproto_sflow *os)
572 if (ofproto_sflow_is_enabled(os)) {
573 time_t now = time_now();
574 if (now >= os->next_tick) {
575 sfl_agent_tick(os->sflow_agent, now);
576 os->next_tick = now + 1;
582 ofproto_sflow_wait(struct ofproto_sflow *os)
584 if (ofproto_sflow_is_enabled(os)) {
585 poll_timer_wait(os->next_tick * 1000 - time_msec());