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"
36 #define THIS_MODULE VLM_sflow
39 struct ofproto_sflow_port {
40 struct netdev *netdev; /* Underlying network device, for stats. */
41 SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
44 struct ofproto_sflow {
45 struct ofproto *ofproto;
46 struct collectors *collectors;
47 SFLAgent *sflow_agent;
48 struct ofproto_sflow_options *options;
51 size_t n_flood, n_all;
52 struct port_array ports; /* Indexed by XFLOW port number. */
55 #define RECEIVER_INDEX 1
57 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
60 nullable_string_is_equal(const char *a, const char *b)
62 return a ? b && !strcmp(a, b) : !b;
66 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
67 const struct ofproto_sflow_options *b)
69 return (svec_equal(&a->targets, &b->targets)
70 && a->sampling_rate == b->sampling_rate
71 && a->polling_interval == b->polling_interval
72 && a->header_len == b->header_len
73 && a->sub_id == b->sub_id
74 && nullable_string_is_equal(a->agent_device, b->agent_device)
75 && nullable_string_is_equal(a->control_ip, b->control_ip));
78 static struct ofproto_sflow_options *
79 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
81 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
82 svec_clone(&new->targets, &old->targets);
83 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
84 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
89 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
92 svec_destroy(&options->targets);
93 free(options->agent_device);
94 free(options->control_ip);
99 /* sFlow library callback to allocate memory. */
101 sflow_agent_alloc_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
104 return calloc(1, bytes);
107 /* sFlow library callback to free memory. */
109 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
116 /* sFlow library callback to report error. */
118 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
121 VLOG_WARN("sFlow agent error: %s", msg);
124 /* sFlow library callback to send datagram. */
126 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent OVS_UNUSED,
127 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
130 struct ofproto_sflow *os = os_;
131 collectors_send(os->collectors, pkt, pktLen);
135 sflow_agent_get_counters(void *os_, SFLPoller *poller,
136 SFL_COUNTERS_SAMPLE_TYPE *cs)
138 struct ofproto_sflow *os = os_;
139 SFLCounters_sample_element elem;
140 struct ofproto_sflow_port *osp;
141 SFLIf_counters *counters;
142 struct netdev_stats stats;
143 enum netdev_flags flags;
146 osp = port_array_get(&os->ports, poller->bridgePort);
151 elem.tag = SFLCOUNTERS_GENERIC;
152 counters = &elem.counterBlock.generic;
153 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
154 counters->ifType = 6;
155 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
156 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
157 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
158 counters->ifSpeed = netdev_features_to_bps(current);
159 counters->ifDirection = (netdev_features_is_full_duplex(current)
162 counters->ifSpeed = 100000000;
163 counters->ifDirection = 0;
165 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
168 counters->ifStatus = 1; /* ifAdminStatus up. */
169 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
170 counters->ifStatus |= 2; /* ifOperStatus us. */
173 counters->ifStatus = 0; /* Down. */
177 1. Is the multicast counter filled in?
178 2. Does the multicast counter include broadcasts?
179 3. Does the rx_packets counter include multicasts/broadcasts?
181 netdev_get_stats(osp->netdev, &stats);
182 counters->ifInOctets = stats.rx_bytes;
183 counters->ifInUcastPkts = stats.rx_packets;
184 counters->ifInMulticastPkts = stats.multicast;
185 counters->ifInBroadcastPkts = -1;
186 counters->ifInDiscards = stats.rx_dropped;
187 counters->ifInErrors = stats.rx_errors;
188 counters->ifInUnknownProtos = -1;
189 counters->ifOutOctets = stats.tx_bytes;
190 counters->ifOutUcastPkts = stats.tx_packets;
191 counters->ifOutMulticastPkts = -1;
192 counters->ifOutBroadcastPkts = -1;
193 counters->ifOutDiscards = stats.tx_dropped;
194 counters->ifOutErrors = stats.tx_errors;
195 counters->ifPromiscuousMode = 0;
197 SFLADD_ELEMENT(cs, &elem);
198 sfl_poller_writeCountersSample(poller, cs);
201 /* Obtains an address to use for the local sFlow agent and stores it into
202 * '*agent_addr'. Returns true if successful, false on failure.
204 * The sFlow agent address should be a local IP address that is persistent and
205 * reachable over the network, if possible. The IP address associated with
206 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
207 * address used to talk to the controller. */
209 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
210 SFLAddress *agent_addr)
214 memset(agent_addr, 0, sizeof *agent_addr);
215 agent_addr->type = SFLADDRESSTYPE_IP_V4;
218 struct netdev *netdev;
220 if (!netdev_open_default(agent_device, &netdev)) {
221 int error = netdev_get_in4(netdev, &in4, NULL);
222 netdev_close(netdev);
229 if (control_ip && !lookup_ip(control_ip, &in4)) {
233 VLOG_ERR("could not determine IP address for sFlow agent");
237 agent_addr->address.ip_v4.addr = in4.s_addr;
242 ofproto_sflow_clear(struct ofproto_sflow *os)
244 struct ofproto_sflow_port *osp;
245 unsigned int xflow_port;
247 if (os->sflow_agent) {
248 sfl_agent_release(os->sflow_agent);
249 os->sflow_agent = NULL;
251 collectors_destroy(os->collectors);
252 os->collectors = NULL;
253 ofproto_sflow_options_destroy(os->options);
256 PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
257 ofproto_sflow_del_port(os, xflow_port);
259 port_array_clear(&os->ports);
261 /* Turn off sampling to save CPU cycles. */
262 wdp_set_sflow_probability(os->wdp, 0);
266 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
268 return os->collectors != NULL;
271 struct ofproto_sflow *
272 ofproto_sflow_create(struct wdp *wdp)
274 struct ofproto_sflow *os;
276 os = xcalloc(1, sizeof *os);
278 os->next_tick = time_now() + 1;
279 port_array_init(&os->ports);
284 ofproto_sflow_destroy(struct ofproto_sflow *os)
287 ofproto_sflow_clear(os);
288 port_array_destroy(&os->ports);
294 ofproto_sflow_add_poller(struct ofproto_sflow *os,
295 struct ofproto_sflow_port *osp, uint16_t xflow_port)
297 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
298 sflow_agent_get_counters);
299 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
300 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
301 sfl_poller_set_bridgePort(poller, xflow_port);
305 ofproto_sflow_add_sampler(struct ofproto_sflow *os,
306 struct ofproto_sflow_port *osp,
307 u_int32_t sampling_rate, u_int32_t header_len)
309 SFLSampler *sampler = sfl_agent_addSampler(os->sflow_agent, &osp->dsi);
310 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, sampling_rate);
311 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, header_len);
312 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
316 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t xflow_port,
317 const char *netdev_name)
319 struct ofproto_sflow_port *osp;
320 struct netdev *netdev;
324 ofproto_sflow_del_port(os, xflow_port);
326 /* Open network device. */
327 error = netdev_open_default(netdev_name, &netdev);
329 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
330 netdev_name, strerror(error));
334 /* Add to table of ports. */
335 osp = xmalloc(sizeof *osp);
336 osp->netdev = netdev;
337 ifindex = netdev_get_ifindex(netdev);
339 ifindex = (os->sflow_agent->subId << 16) + xflow_port;
341 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
342 port_array_set(&os->ports, xflow_port, osp);
345 if (os->sflow_agent) {
346 ofproto_sflow_add_poller(os, osp, xflow_port);
351 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t xflow_port)
353 struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
355 if (os->sflow_agent) {
356 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
357 sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
359 netdev_close(osp->netdev);
361 port_array_set(&os->ports, xflow_port, NULL);
366 ofproto_sflow_set_options(struct ofproto_sflow *os,
367 const struct ofproto_sflow_options *options)
369 struct ofproto_sflow_port *osp;
370 bool options_changed;
371 SFLReceiver *receiver;
372 unsigned int xflow_port;
376 if (!options->targets.n || !options->sampling_rate) {
377 /* No point in doing any work if there are no targets or nothing to
379 ofproto_sflow_clear(os);
383 options_changed = (!os->options
384 || !ofproto_sflow_options_equal(options, os->options));
386 /* Configure collectors if options have changed or if we're shortchanged in
387 * collectors (which indicates that opening one or more of the configured
388 * collectors failed, so that we should retry). */
390 || collectors_count(os->collectors) < options->targets.n) {
391 collectors_destroy(os->collectors);
392 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
394 if (os->collectors == NULL) {
395 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
397 ofproto_sflow_clear(os);
402 /* Avoid reconfiguring if options didn't change. */
403 if (!options_changed) {
406 ofproto_sflow_options_destroy(os->options);
407 os->options = ofproto_sflow_options_clone(options);
409 /* Choose agent IP address. */
410 if (!sflow_choose_agent_address(options->agent_device,
411 options->control_ip, &agentIP)) {
412 ofproto_sflow_clear(os);
417 VLOG_INFO("creating sFlow agent %d", options->sub_id);
418 if (os->sflow_agent) {
419 sfl_agent_release(os->sflow_agent);
421 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
423 sfl_agent_init(os->sflow_agent,
426 now, /* Boot time. */
427 now, /* Current time. */
428 os, /* Pointer supplied to callbacks. */
429 sflow_agent_alloc_cb,
431 sflow_agent_error_cb,
432 sflow_agent_send_packet_cb);
434 receiver = sfl_agent_addReceiver(os->sflow_agent);
435 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
436 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
438 /* Set the sampling_rate down in the datapath. */
439 wdp_set_sflow_probability(os->wdp,
440 MAX(1, UINT32_MAX / options->sampling_rate));
442 /* Add samplers and pollers for the currently known ports. */
443 PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
444 ofproto_sflow_add_sampler(os, osp,
445 options->sampling_rate, options->header_len);
450 ofproto_sflow_xflow_port_to_ifindex(const struct ofproto_sflow *os,
453 struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
454 return osp ? SFL_DS_INDEX(osp->dsi) : 0;
458 ofproto_sflow_received(struct ofproto_sflow *os, struct xflow_msg *msg)
460 SFL_FLOW_SAMPLE_TYPE fs;
461 SFLFlow_sample_element hdrElem;
462 SFLSampled_header *header;
463 SFLFlow_sample_element switchElem;
465 const struct xflow_sflow_sample_header *hdr;
466 const union xflow_action *actions;
467 struct ofpbuf payload;
468 size_t n_actions, n_outputs;
473 /* Get xflow_sflow_sample_header. */
474 min_size = sizeof *msg + sizeof *hdr;
475 if (min_size > msg->length) {
476 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
477 msg->length, min_size);
480 hdr = (const struct xflow_sflow_sample_header *) (msg + 1);
483 n_actions = hdr->n_actions;
484 if (n_actions > 65536 / sizeof *actions) {
485 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%zu > %zu)",
486 65536 / sizeof *actions, n_actions);
489 min_size += n_actions * sizeof *actions;
490 if (min_size > msg->length) {
491 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
493 n_actions, msg->length, min_size);
496 actions = (const union xflow_action *) (hdr + 1);
498 /* Get packet payload and extract flow. */
499 payload.data = (union xflow_action *) (actions + n_actions);
500 payload.size = msg->length - min_size;
501 flow_extract(&payload, msg->port, &flow);
503 /* Build a flow sample */
504 memset(&fs, 0, sizeof fs);
505 fs.input = ofproto_sflow_xflow_port_to_ifindex(os, msg->port);
506 fs.output = 0; /* Filled in correctly below. */
507 fs.sample_pool = hdr->sample_pool;
509 /* We are going to give it to the sampler that represents this input port.
510 * By implementing "ingress-only" sampling like this we ensure that we
511 * never have to offer the same sample to more than one sampler. */
512 sampler = sfl_agent_getSamplerByIfIndex(os->sflow_agent, fs.input);
514 VLOG_WARN_RL(&rl, "no sampler for input ifIndex (%"PRIu32")",
519 /* Sampled header. */
520 memset(&hdrElem, 0, sizeof hdrElem);
521 hdrElem.tag = SFLFLOW_HEADER;
522 header = &hdrElem.flowType.header;
523 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
524 header->frame_length = payload.size;
525 header->stripped = 4; /* Ethernet FCS stripped off. */
526 header->header_length = MIN(payload.size,
527 sampler->sFlowFsMaximumHeaderSize);
528 header->header_bytes = payload.data;
530 /* Add extended switch element. */
531 memset(&switchElem, 0, sizeof(switchElem));
532 switchElem.tag = SFLFLOW_EX_SWITCH;
533 switchElem.flowType.sw.src_vlan = ntohs(flow.dl_vlan);
534 switchElem.flowType.sw.src_priority = -1; /* XXX */
535 switchElem.flowType.sw.dst_vlan = -1; /* Filled in correctly below. */
536 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
538 /* Figure out the output ports. */
540 for (i = 0; i < n_actions; i++) {
541 const union xflow_action *a = &actions[i];
545 fs.output = ofproto_sflow_xflow_port_to_ifindex(os, a->output.port);
549 case XFLOWAT_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
557 case XFLOWAT_SET_DL_TCI:
558 if (a->dl_tci.mask & htons(VLAN_VID_MASK)) {
559 switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(a->dl_tci.tci);
561 if (a->dl_tci.mask & htons(VLAN_PCP_MASK)) {
562 switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(a->dl_tci.tci);
571 /* Set output port, as defined by http://www.sflow.org/sflow_version_5.txt
572 (search for "Input/output port information"). */
574 /* This value indicates that the packet was dropped for an unknown
576 fs.output = 0x40000000 | 256;
577 } else if (n_outputs > 1 || !fs.output) {
578 /* Setting the high bit means "multiple output ports". */
579 fs.output = 0x80000000 | n_outputs;
582 /* Submit the flow sample to be encoded into the next datagram. */
583 SFLADD_ELEMENT(&fs, &hdrElem);
584 SFLADD_ELEMENT(&fs, &switchElem);
585 sfl_sampler_writeFlowSample(sampler, &fs);
589 ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
590 size_t n_flood, size_t n_all)
592 os->n_flood = n_flood;
597 ofproto_sflow_run(struct ofproto_sflow *os)
599 if (ofproto_sflow_is_enabled(os)) {
600 time_t now = time_now();
601 if (now >= os->next_tick) {
602 sfl_agent_tick(os->sflow_agent, now);
603 os->next_tick = now + 1;
609 ofproto_sflow_wait(struct ofproto_sflow *os)
611 if (ofproto_sflow_is_enabled(os)) {
612 poll_timer_wait(os->next_tick * 1000 - time_msec());