2 * Copyright (c) 2009, 2010 Nicira Networks.
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-sflow.h"
22 #include "collectors.h"
31 #include "poll-loop.h"
32 #include "sflow_api.h"
33 #include "socket-util.h"
37 VLOG_DEFINE_THIS_MODULE(sflow)
39 struct ofproto_sflow_port {
40 struct hmap_node hmap_node; /* In struct ofproto_sflow's "ports" hmap. */
41 struct netdev *netdev; /* Underlying network device, for stats. */
42 SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
43 uint16_t odp_port; /* ODP port number. */
46 struct ofproto_sflow {
47 struct ofproto *ofproto;
48 struct collectors *collectors;
49 SFLAgent *sflow_agent;
50 struct ofproto_sflow_options *options;
53 size_t n_flood, n_all;
54 struct hmap ports; /* Contains "struct ofproto_sflow_port"s. */
57 static void ofproto_sflow_del_port__(struct ofproto_sflow *,
58 struct ofproto_sflow_port *);
60 #define RECEIVER_INDEX 1
62 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
65 nullable_string_is_equal(const char *a, const char *b)
67 return a ? b && !strcmp(a, b) : !b;
71 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
72 const struct ofproto_sflow_options *b)
74 return (svec_equal(&a->targets, &b->targets)
75 && a->sampling_rate == b->sampling_rate
76 && a->polling_interval == b->polling_interval
77 && a->header_len == b->header_len
78 && a->sub_id == b->sub_id
79 && nullable_string_is_equal(a->agent_device, b->agent_device)
80 && nullable_string_is_equal(a->control_ip, b->control_ip));
83 static struct ofproto_sflow_options *
84 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
86 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
87 svec_clone(&new->targets, &old->targets);
88 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
89 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
94 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
97 svec_destroy(&options->targets);
98 free(options->agent_device);
99 free(options->control_ip);
104 /* sFlow library callback to allocate memory. */
106 sflow_agent_alloc_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
109 return calloc(1, bytes);
112 /* sFlow library callback to free memory. */
114 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
121 /* sFlow library callback to report error. */
123 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
126 VLOG_WARN("sFlow agent error: %s", msg);
129 /* sFlow library callback to send datagram. */
131 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent OVS_UNUSED,
132 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
135 struct ofproto_sflow *os = os_;
136 collectors_send(os->collectors, pkt, pktLen);
139 static struct ofproto_sflow_port *
140 ofproto_sflow_find_port(const struct ofproto_sflow *os, uint16_t odp_port)
142 struct ofproto_sflow_port *osp;
144 HMAP_FOR_EACH_IN_BUCKET (osp, hmap_node,
145 hash_int(odp_port, 0), &os->ports) {
146 if (osp->odp_port == odp_port) {
154 sflow_agent_get_counters(void *os_, SFLPoller *poller,
155 SFL_COUNTERS_SAMPLE_TYPE *cs)
157 struct ofproto_sflow *os = os_;
158 SFLCounters_sample_element elem;
159 struct ofproto_sflow_port *osp;
160 SFLIf_counters *counters;
161 struct netdev_stats stats;
162 enum netdev_flags flags;
165 osp = ofproto_sflow_find_port(os, poller->bridgePort);
170 elem.tag = SFLCOUNTERS_GENERIC;
171 counters = &elem.counterBlock.generic;
172 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
173 counters->ifType = 6;
174 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
175 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
176 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
177 counters->ifSpeed = netdev_features_to_bps(current);
178 counters->ifDirection = (netdev_features_is_full_duplex(current)
181 counters->ifSpeed = 100000000;
182 counters->ifDirection = 0;
184 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
187 counters->ifStatus = 1; /* ifAdminStatus up. */
188 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
189 counters->ifStatus |= 2; /* ifOperStatus us. */
192 counters->ifStatus = 0; /* Down. */
196 1. Is the multicast counter filled in?
197 2. Does the multicast counter include broadcasts?
198 3. Does the rx_packets counter include multicasts/broadcasts?
200 netdev_get_stats(osp->netdev, &stats);
201 counters->ifInOctets = stats.rx_bytes;
202 counters->ifInUcastPkts = stats.rx_packets;
203 counters->ifInMulticastPkts = stats.multicast;
204 counters->ifInBroadcastPkts = -1;
205 counters->ifInDiscards = stats.rx_dropped;
206 counters->ifInErrors = stats.rx_errors;
207 counters->ifInUnknownProtos = -1;
208 counters->ifOutOctets = stats.tx_bytes;
209 counters->ifOutUcastPkts = stats.tx_packets;
210 counters->ifOutMulticastPkts = -1;
211 counters->ifOutBroadcastPkts = -1;
212 counters->ifOutDiscards = stats.tx_dropped;
213 counters->ifOutErrors = stats.tx_errors;
214 counters->ifPromiscuousMode = 0;
216 SFLADD_ELEMENT(cs, &elem);
217 sfl_poller_writeCountersSample(poller, cs);
220 /* Obtains an address to use for the local sFlow agent and stores it into
221 * '*agent_addr'. Returns true if successful, false on failure.
223 * The sFlow agent address should be a local IP address that is persistent and
224 * reachable over the network, if possible. The IP address associated with
225 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
226 * address used to talk to the controller. */
228 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
229 SFLAddress *agent_addr)
233 memset(agent_addr, 0, sizeof *agent_addr);
234 agent_addr->type = SFLADDRESSTYPE_IP_V4;
237 struct netdev *netdev;
239 if (!netdev_open_default(agent_device, &netdev)) {
240 int error = netdev_get_in4(netdev, &in4, NULL);
241 netdev_close(netdev);
248 if (control_ip && !lookup_ip(control_ip, &in4)) {
252 VLOG_ERR("could not determine IP address for sFlow agent");
256 agent_addr->address.ip_v4.addr = in4.s_addr;
261 ofproto_sflow_clear(struct ofproto_sflow *os)
263 if (os->sflow_agent) {
264 sfl_agent_release(os->sflow_agent);
265 os->sflow_agent = NULL;
267 collectors_destroy(os->collectors);
268 os->collectors = NULL;
269 ofproto_sflow_options_destroy(os->options);
272 /* Turn off sampling to save CPU cycles. */
273 dpif_set_sflow_probability(os->dpif, 0);
277 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
279 return os->collectors != NULL;
282 struct ofproto_sflow *
283 ofproto_sflow_create(struct dpif *dpif)
285 struct ofproto_sflow *os;
287 os = xcalloc(1, sizeof *os);
289 os->next_tick = time_now() + 1;
290 hmap_init(&os->ports);
295 ofproto_sflow_destroy(struct ofproto_sflow *os)
298 struct ofproto_sflow_port *osp, *next;
300 ofproto_sflow_clear(os);
301 HMAP_FOR_EACH_SAFE (osp, next, hmap_node, &os->ports) {
302 ofproto_sflow_del_port__(os, osp);
304 hmap_destroy(&os->ports);
310 ofproto_sflow_add_poller(struct ofproto_sflow *os,
311 struct ofproto_sflow_port *osp, uint16_t odp_port)
313 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
314 sflow_agent_get_counters);
315 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
316 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
317 sfl_poller_set_bridgePort(poller, odp_port);
321 ofproto_sflow_add_sampler(struct ofproto_sflow *os,
322 struct ofproto_sflow_port *osp)
324 SFLSampler *sampler = sfl_agent_addSampler(os->sflow_agent, &osp->dsi);
325 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, os->options->sampling_rate);
326 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, os->options->header_len);
327 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
331 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t odp_port,
332 const char *netdev_name)
334 struct ofproto_sflow_port *osp;
335 struct netdev *netdev;
339 ofproto_sflow_del_port(os, odp_port);
341 /* Open network device. */
342 error = netdev_open_default(netdev_name, &netdev);
344 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
345 netdev_name, strerror(error));
349 /* Add to table of ports. */
350 osp = xmalloc(sizeof *osp);
351 osp->netdev = netdev;
352 ifindex = netdev_get_ifindex(netdev);
354 ifindex = (os->sflow_agent->subId << 16) + odp_port;
356 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
357 osp->odp_port = odp_port;
358 hmap_insert(&os->ports, &osp->hmap_node, hash_int(odp_port, 0));
360 /* Add poller and sampler. */
361 if (os->sflow_agent) {
362 ofproto_sflow_add_poller(os, osp, odp_port);
363 ofproto_sflow_add_sampler(os, osp);
368 ofproto_sflow_del_port__(struct ofproto_sflow *os,
369 struct ofproto_sflow_port *osp)
371 if (os->sflow_agent) {
372 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
373 sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
375 netdev_close(osp->netdev);
376 hmap_remove(&os->ports, &osp->hmap_node);
381 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t odp_port)
383 struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port);
385 ofproto_sflow_del_port__(os, osp);
390 ofproto_sflow_set_options(struct ofproto_sflow *os,
391 const struct ofproto_sflow_options *options)
393 struct ofproto_sflow_port *osp;
394 bool options_changed;
395 SFLReceiver *receiver;
399 if (!options->targets.n || !options->sampling_rate) {
400 /* No point in doing any work if there are no targets or nothing to
402 ofproto_sflow_clear(os);
406 options_changed = (!os->options
407 || !ofproto_sflow_options_equal(options, os->options));
409 /* Configure collectors if options have changed or if we're shortchanged in
410 * collectors (which indicates that opening one or more of the configured
411 * collectors failed, so that we should retry). */
413 || collectors_count(os->collectors) < options->targets.n) {
414 collectors_destroy(os->collectors);
415 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
417 if (os->collectors == NULL) {
418 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
420 ofproto_sflow_clear(os);
425 /* Avoid reconfiguring if options didn't change. */
426 if (!options_changed) {
429 ofproto_sflow_options_destroy(os->options);
430 os->options = ofproto_sflow_options_clone(options);
432 /* Choose agent IP address. */
433 if (!sflow_choose_agent_address(options->agent_device,
434 options->control_ip, &agentIP)) {
435 ofproto_sflow_clear(os);
440 VLOG_INFO("creating sFlow agent %d", options->sub_id);
441 if (os->sflow_agent) {
442 sfl_agent_release(os->sflow_agent);
444 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
446 sfl_agent_init(os->sflow_agent,
449 now, /* Boot time. */
450 now, /* Current time. */
451 os, /* Pointer supplied to callbacks. */
452 sflow_agent_alloc_cb,
454 sflow_agent_error_cb,
455 sflow_agent_send_packet_cb);
457 receiver = sfl_agent_addReceiver(os->sflow_agent);
458 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
459 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
461 /* Set the sampling_rate down in the datapath. */
462 dpif_set_sflow_probability(os->dpif,
463 MAX(1, UINT32_MAX / options->sampling_rate));
465 /* Add samplers and pollers for the currently known ports. */
466 HMAP_FOR_EACH (osp, hmap_node, &os->ports) {
467 ofproto_sflow_add_poller(os, osp, osp->odp_port);
468 ofproto_sflow_add_sampler(os, osp);
473 ofproto_sflow_odp_port_to_ifindex(const struct ofproto_sflow *os,
476 struct ofproto_sflow_port *osp = ofproto_sflow_find_port(os, odp_port);
477 return osp ? SFL_DS_INDEX(osp->dsi) : 0;
481 ofproto_sflow_received(struct ofproto_sflow *os, struct odp_msg *msg)
483 SFL_FLOW_SAMPLE_TYPE fs;
484 SFLFlow_sample_element hdrElem;
485 SFLSampled_header *header;
486 SFLFlow_sample_element switchElem;
488 const struct odp_sflow_sample_header *hdr;
489 const union odp_action *actions;
490 struct ofpbuf payload;
491 size_t n_actions, n_outputs;
496 /* Get odp_sflow_sample_header. */
497 min_size = sizeof *msg + sizeof *hdr;
498 if (min_size > msg->length) {
499 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
500 msg->length, min_size);
503 hdr = (const struct odp_sflow_sample_header *) (msg + 1);
506 n_actions = hdr->n_actions;
507 if (n_actions > 65536 / sizeof *actions) {
508 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%zu > %zu)",
509 65536 / sizeof *actions, n_actions);
512 min_size += n_actions * sizeof *actions;
513 if (min_size > msg->length) {
514 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
516 n_actions, msg->length, min_size);
519 actions = (const union odp_action *) (hdr + 1);
521 /* Get packet payload and extract flow. */
522 payload.data = (union odp_action *) (actions + n_actions);
523 payload.size = msg->length - min_size;
524 flow_extract(&payload, 0, msg->port, &flow);
526 /* Build a flow sample */
527 memset(&fs, 0, sizeof fs);
528 fs.input = ofproto_sflow_odp_port_to_ifindex(os, msg->port);
529 fs.output = 0; /* Filled in correctly below. */
530 fs.sample_pool = hdr->sample_pool;
532 /* We are going to give it to the sampler that represents this input port.
533 * By implementing "ingress-only" sampling like this we ensure that we
534 * never have to offer the same sample to more than one sampler. */
535 sampler = sfl_agent_getSamplerByIfIndex(os->sflow_agent, fs.input);
537 VLOG_WARN_RL(&rl, "no sampler for input ifIndex (%"PRIu32")",
542 /* Sampled header. */
543 memset(&hdrElem, 0, sizeof hdrElem);
544 hdrElem.tag = SFLFLOW_HEADER;
545 header = &hdrElem.flowType.header;
546 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
547 /* The frame_length should include the Ethernet FCS (4 bytes),
548 but it has already been stripped, so we need to add 4 here. */
549 header->frame_length = payload.size + 4;
550 /* Ethernet FCS stripped off. */
551 header->stripped = 4;
552 header->header_length = MIN(payload.size,
553 sampler->sFlowFsMaximumHeaderSize);
554 header->header_bytes = payload.data;
556 /* Add extended switch element. */
557 memset(&switchElem, 0, sizeof(switchElem));
558 switchElem.tag = SFLFLOW_EX_SWITCH;
559 switchElem.flowType.sw.src_vlan = ntohs(flow.dl_vlan);
560 switchElem.flowType.sw.src_priority = -1; /* XXX */
561 /* Initialize the output VLAN and priority to be the same as the input,
562 but these fields can be overriden below if affected by an action. */
563 switchElem.flowType.sw.dst_vlan = switchElem.flowType.sw.src_vlan;
564 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
566 /* Figure out the output ports. */
568 for (i = 0; i < n_actions; i++) {
569 const union odp_action *a = &actions[i];
574 fs.output = ofproto_sflow_odp_port_to_ifindex(os, a->output.port);
578 case ODPAT_SET_DL_TCI:
580 switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(tci);
581 switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(tci);
589 /* Set output port, as defined by http://www.sflow.org/sflow_version_5.txt
590 (search for "Input/output port information"). */
592 /* This value indicates that the packet was dropped for an unknown
594 fs.output = 0x40000000 | 256;
595 } else if (n_outputs > 1 || !fs.output) {
596 /* Setting the high bit means "multiple output ports". */
597 fs.output = 0x80000000 | n_outputs;
600 /* Submit the flow sample to be encoded into the next datagram. */
601 SFLADD_ELEMENT(&fs, &hdrElem);
602 SFLADD_ELEMENT(&fs, &switchElem);
603 sfl_sampler_writeFlowSample(sampler, &fs);
607 ofproto_sflow_run(struct ofproto_sflow *os)
609 if (ofproto_sflow_is_enabled(os)) {
610 time_t now = time_now();
611 if (now >= os->next_tick) {
612 sfl_agent_tick(os->sflow_agent, time_wall());
613 os->next_tick = now + 1;
619 ofproto_sflow_wait(struct ofproto_sflow *os)
621 if (ofproto_sflow_is_enabled(os)) {
622 poll_timer_wait_until(os->next_tick * 1000LL);