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"
35 VLOG_DEFINE_THIS_MODULE(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 OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
102 return calloc(1, bytes);
105 /* sFlow library callback to free memory. */
107 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
114 /* sFlow library callback to report error. */
116 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
119 VLOG_WARN("sFlow agent error: %s", msg);
122 /* sFlow library callback to send datagram. */
124 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent OVS_UNUSED,
125 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
128 struct ofproto_sflow *os = os_;
129 collectors_send(os->collectors, pkt, pktLen);
133 sflow_agent_get_counters(void *os_, SFLPoller *poller,
134 SFL_COUNTERS_SAMPLE_TYPE *cs)
136 struct ofproto_sflow *os = os_;
137 SFLCounters_sample_element elem;
138 struct ofproto_sflow_port *osp;
139 SFLIf_counters *counters;
140 struct netdev_stats stats;
141 enum netdev_flags flags;
144 osp = port_array_get(&os->ports, poller->bridgePort);
149 elem.tag = SFLCOUNTERS_GENERIC;
150 counters = &elem.counterBlock.generic;
151 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
152 counters->ifType = 6;
153 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
154 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
155 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
156 counters->ifSpeed = netdev_features_to_bps(current);
157 counters->ifDirection = (netdev_features_is_full_duplex(current)
160 counters->ifSpeed = 100000000;
161 counters->ifDirection = 0;
163 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
166 counters->ifStatus = 1; /* ifAdminStatus up. */
167 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
168 counters->ifStatus |= 2; /* ifOperStatus us. */
171 counters->ifStatus = 0; /* Down. */
175 1. Is the multicast counter filled in?
176 2. Does the multicast counter include broadcasts?
177 3. Does the rx_packets counter include multicasts/broadcasts?
179 netdev_get_stats(osp->netdev, &stats);
180 counters->ifInOctets = stats.rx_bytes;
181 counters->ifInUcastPkts = stats.rx_packets;
182 counters->ifInMulticastPkts = stats.multicast;
183 counters->ifInBroadcastPkts = -1;
184 counters->ifInDiscards = stats.rx_dropped;
185 counters->ifInErrors = stats.rx_errors;
186 counters->ifInUnknownProtos = -1;
187 counters->ifOutOctets = stats.tx_bytes;
188 counters->ifOutUcastPkts = stats.tx_packets;
189 counters->ifOutMulticastPkts = -1;
190 counters->ifOutBroadcastPkts = -1;
191 counters->ifOutDiscards = stats.tx_dropped;
192 counters->ifOutErrors = stats.tx_errors;
193 counters->ifPromiscuousMode = 0;
195 SFLADD_ELEMENT(cs, &elem);
196 sfl_poller_writeCountersSample(poller, cs);
199 /* Obtains an address to use for the local sFlow agent and stores it into
200 * '*agent_addr'. Returns true if successful, false on failure.
202 * The sFlow agent address should be a local IP address that is persistent and
203 * reachable over the network, if possible. The IP address associated with
204 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
205 * address used to talk to the controller. */
207 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
208 SFLAddress *agent_addr)
212 memset(agent_addr, 0, sizeof *agent_addr);
213 agent_addr->type = SFLADDRESSTYPE_IP_V4;
216 struct netdev *netdev;
218 if (!netdev_open_default(agent_device, &netdev)) {
219 int error = netdev_get_in4(netdev, &in4, NULL);
220 netdev_close(netdev);
227 if (control_ip && !lookup_ip(control_ip, &in4)) {
231 VLOG_ERR("could not determine IP address for sFlow agent");
235 agent_addr->address.ip_v4.addr = in4.s_addr;
240 ofproto_sflow_clear(struct ofproto_sflow *os)
242 if (os->sflow_agent) {
243 sfl_agent_release(os->sflow_agent);
244 os->sflow_agent = NULL;
246 collectors_destroy(os->collectors);
247 os->collectors = NULL;
248 ofproto_sflow_options_destroy(os->options);
251 /* Turn off sampling to save CPU cycles. */
252 dpif_set_sflow_probability(os->dpif, 0);
256 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
258 return os->collectors != NULL;
261 struct ofproto_sflow *
262 ofproto_sflow_create(struct dpif *dpif)
264 struct ofproto_sflow *os;
266 os = xcalloc(1, sizeof *os);
268 os->next_tick = time_now() + 1;
269 port_array_init(&os->ports);
274 ofproto_sflow_destroy(struct ofproto_sflow *os)
277 struct ofproto_sflow_port *osp;
278 unsigned int odp_port;
280 ofproto_sflow_clear(os);
281 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
282 ofproto_sflow_del_port(os, odp_port);
284 port_array_destroy(&os->ports);
290 ofproto_sflow_add_poller(struct ofproto_sflow *os,
291 struct ofproto_sflow_port *osp, uint16_t odp_port)
293 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
294 sflow_agent_get_counters);
295 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
296 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
297 sfl_poller_set_bridgePort(poller, odp_port);
301 ofproto_sflow_add_sampler(struct ofproto_sflow *os,
302 struct ofproto_sflow_port *osp)
304 SFLSampler *sampler = sfl_agent_addSampler(os->sflow_agent, &osp->dsi);
305 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, os->options->sampling_rate);
306 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, os->options->header_len);
307 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
311 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
312 const char *netdev_name)
314 struct ofproto_sflow_port *osp;
315 struct netdev *netdev;
319 ofproto_sflow_del_port(os, odp_port);
321 /* Open network device. */
322 error = netdev_open_default(netdev_name, &netdev);
324 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
325 netdev_name, strerror(error));
329 /* Add to table of ports. */
330 osp = xmalloc(sizeof *osp);
331 osp->netdev = netdev;
332 ifindex = netdev_get_ifindex(netdev);
334 ifindex = (os->sflow_agent->subId << 16) + odp_port;
336 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
337 port_array_set(&os->ports, odp_port, osp);
339 /* Add poller and sampler. */
340 if (os->sflow_agent) {
341 ofproto_sflow_add_poller(os, osp, odp_port);
342 ofproto_sflow_add_sampler(os, osp);
347 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
349 struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
351 if (os->sflow_agent) {
352 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
353 sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
355 netdev_close(osp->netdev);
357 port_array_delete(&os->ports, odp_port);
362 ofproto_sflow_set_options(struct ofproto_sflow *os,
363 const struct ofproto_sflow_options *options)
365 struct ofproto_sflow_port *osp;
366 bool options_changed;
367 SFLReceiver *receiver;
368 unsigned int odp_port;
372 if (!options->targets.n || !options->sampling_rate) {
373 /* No point in doing any work if there are no targets or nothing to
375 ofproto_sflow_clear(os);
379 options_changed = (!os->options
380 || !ofproto_sflow_options_equal(options, os->options));
382 /* Configure collectors if options have changed or if we're shortchanged in
383 * collectors (which indicates that opening one or more of the configured
384 * collectors failed, so that we should retry). */
386 || collectors_count(os->collectors) < options->targets.n) {
387 collectors_destroy(os->collectors);
388 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
390 if (os->collectors == NULL) {
391 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
393 ofproto_sflow_clear(os);
398 /* Avoid reconfiguring if options didn't change. */
399 if (!options_changed) {
402 ofproto_sflow_options_destroy(os->options);
403 os->options = ofproto_sflow_options_clone(options);
405 /* Choose agent IP address. */
406 if (!sflow_choose_agent_address(options->agent_device,
407 options->control_ip, &agentIP)) {
408 ofproto_sflow_clear(os);
413 VLOG_INFO("creating sFlow agent %d", options->sub_id);
414 if (os->sflow_agent) {
415 sfl_agent_release(os->sflow_agent);
417 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
419 sfl_agent_init(os->sflow_agent,
422 now, /* Boot time. */
423 now, /* Current time. */
424 os, /* Pointer supplied to callbacks. */
425 sflow_agent_alloc_cb,
427 sflow_agent_error_cb,
428 sflow_agent_send_packet_cb);
430 receiver = sfl_agent_addReceiver(os->sflow_agent);
431 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
432 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
434 /* Set the sampling_rate down in the datapath. */
435 dpif_set_sflow_probability(os->dpif,
436 MAX(1, UINT32_MAX / options->sampling_rate));
438 /* Add samplers and pollers for the currently known ports. */
439 PORT_ARRAY_FOR_EACH (osp, &os->ports, odp_port) {
440 ofproto_sflow_add_poller(os, osp, odp_port);
441 ofproto_sflow_add_sampler(os, osp);
446 ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os,
449 struct ofproto_sflow_port *osp = port_array_get(&os->ports, odp_port);
450 return osp ? SFL_DS_INDEX(osp->dsi) : 0;
454 ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
456 SFL_FLOW_SAMPLE_TYPE fs;
457 SFLFlow_sample_element hdrElem;
458 SFLSampled_header *header;
459 SFLFlow_sample_element switchElem;
461 const struct odp_sflow_sample_header *hdr;
462 const union odp_action *actions;
463 struct ofpbuf payload;
464 size_t n_actions, n_outputs;
469 /* Get odp_sflow_sample_header. */
470 min_size = sizeof *msg + sizeof *hdr;
471 if (min_size > msg->length) {
472 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
473 msg->length, min_size);
476 hdr = (const struct odp_sflow_sample_header *) (msg + 1);
479 n_actions = hdr->n_actions;
480 if (n_actions > 65536 / sizeof *actions) {
481 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%zu > %zu)",
482 65536 / sizeof *actions, n_actions);
485 min_size += n_actions * sizeof *actions;
486 if (min_size > msg->length) {
487 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
489 n_actions, msg->length, min_size);
492 actions = (const union odp_action *) (hdr + 1);
494 /* Get packet payload and extract flow. */
495 payload.data = (union odp_action *) (actions + n_actions);
496 payload.size = msg->length - min_size;
497 flow_extract(&payload, 0, msg->port, &flow);
499 /* Build a flow sample */
500 memset(&fs, 0, sizeof fs);
501 fs.input = ofproto_sflow_odp_port_to_ifindex(os, msg->port);
502 fs.output = 0; /* Filled in correctly below. */
503 fs.sample_pool = hdr->sample_pool;
505 /* We are going to give it to the sampler that represents this input port.
506 * By implementing "ingress-only" sampling like this we ensure that we
507 * never have to offer the same sample to more than one sampler. */
508 sampler = sfl_agent_getSamplerByIfIndex(os->sflow_agent, fs.input);
510 VLOG_WARN_RL(&rl, "no sampler for input ifIndex (%"PRIu32")",
515 /* Sampled header. */
516 memset(&hdrElem, 0, sizeof hdrElem);
517 hdrElem.tag = SFLFLOW_HEADER;
518 header = &hdrElem.flowType.header;
519 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
520 /* The frame_length should include the Ethernet FCS (4 bytes),
521 but it has already been stripped, so we need to add 4 here. */
522 header->frame_length = payload.size + 4;
523 /* Ethernet FCS stripped off. */
524 header->stripped = 4;
525 header->header_length = MIN(payload.size,
526 sampler->sFlowFsMaximumHeaderSize);
527 header->header_bytes = payload.data;
529 /* Add extended switch element. */
530 memset(&switchElem, 0, sizeof(switchElem));
531 switchElem.tag = SFLFLOW_EX_SWITCH;
532 switchElem.flowType.sw.src_vlan = ntohs(flow.dl_vlan);
533 switchElem.flowType.sw.src_priority = -1; /* XXX */
534 /* Initialize the output VLAN and priority to be the same as the input,
535 but these fields can be overriden below if affected by an action. */
536 switchElem.flowType.sw.dst_vlan = switchElem.flowType.sw.src_vlan;
537 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
539 /* Figure out the output ports. */
541 for (i = 0; i < n_actions; i++) {
542 const union odp_action *a = &actions[i];
546 fs.output = ofproto_sflow_odp_port_to_ifindex(os, a->output.port);
550 case ODPAT_OUTPUT_GROUP:
551 n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
552 : a->output_group.group == DP_GROUP_ALL ? os->n_all
556 case ODPAT_SET_VLAN_VID:
557 switchElem.flowType.sw.dst_vlan = ntohs(a->vlan_vid.vlan_vid);
560 case ODPAT_SET_VLAN_PCP:
561 switchElem.flowType.sw.dst_priority = a->vlan_pcp.vlan_pcp;
569 /* Set output port, as defined by http://www.sflow.org/sflow_version_5.txt
570 (search for "Input/output port information"). */
572 /* This value indicates that the packet was dropped for an unknown
574 fs.output = 0x40000000 | 256;
575 } else if (n_outputs > 1 || !fs.output) {
576 /* Setting the high bit means "multiple output ports". */
577 fs.output = 0x80000000 | n_outputs;
580 /* Submit the flow sample to be encoded into the next datagram. */
581 SFLADD_ELEMENT(&fs, &hdrElem);
582 SFLADD_ELEMENT(&fs, &switchElem);
583 sfl_sampler_writeFlowSample(sampler, &fs);
587 ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
588 size_t n_flood, size_t n_all)
590 os->n_flood = n_flood;
595 ofproto_sflow_run(struct ofproto_sflow *os)
597 if (ofproto_sflow_is_enabled(os)) {
598 time_t now = time_now();
599 if (now >= os->next_tick) {
600 sfl_agent_tick(os->sflow_agent, time_wall());
601 os->next_tick = now + 1;
607 ofproto_sflow_wait(struct ofproto_sflow *os)
609 if (ofproto_sflow_is_enabled(os)) {
610 poll_timer_wait_until(os->next_tick * 1000LL);