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 #define THIS_MODULE VLM_sflow
38 struct ofproto_sflow_port {
39 struct netdev *netdev; /* Underlying network device, for stats. */
40 SFLDataSource_instance dsi; /* sFlow library's notion of port number. */
43 struct ofproto_sflow {
44 struct ofproto *ofproto;
45 struct collectors *collectors;
46 SFLAgent *sflow_agent;
47 struct ofproto_sflow_options *options;
50 size_t n_flood, n_all;
51 struct port_array ports; /* Indexed by XFLOW port number. */
54 #define RECEIVER_INDEX 1
56 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
59 nullable_string_is_equal(const char *a, const char *b)
61 return a ? b && !strcmp(a, b) : !b;
65 ofproto_sflow_options_equal(const struct ofproto_sflow_options *a,
66 const struct ofproto_sflow_options *b)
68 return (svec_equal(&a->targets, &b->targets)
69 && a->sampling_rate == b->sampling_rate
70 && a->polling_interval == b->polling_interval
71 && a->header_len == b->header_len
72 && a->sub_id == b->sub_id
73 && nullable_string_is_equal(a->agent_device, b->agent_device)
74 && nullable_string_is_equal(a->control_ip, b->control_ip));
77 static struct ofproto_sflow_options *
78 ofproto_sflow_options_clone(const struct ofproto_sflow_options *old)
80 struct ofproto_sflow_options *new = xmemdup(old, sizeof *old);
81 svec_clone(&new->targets, &old->targets);
82 new->agent_device = old->agent_device ? xstrdup(old->agent_device) : NULL;
83 new->control_ip = old->control_ip ? xstrdup(old->control_ip) : NULL;
88 ofproto_sflow_options_destroy(struct ofproto_sflow_options *options)
91 svec_destroy(&options->targets);
92 free(options->agent_device);
93 free(options->control_ip);
98 /* sFlow library callback to allocate memory. */
100 sflow_agent_alloc_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
103 return calloc(1, bytes);
106 /* sFlow library callback to free memory. */
108 sflow_agent_free_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
115 /* sFlow library callback to report error. */
117 sflow_agent_error_cb(void *magic OVS_UNUSED, SFLAgent *agent OVS_UNUSED,
120 VLOG_WARN("sFlow agent error: %s", msg);
123 /* sFlow library callback to send datagram. */
125 sflow_agent_send_packet_cb(void *os_, SFLAgent *agent OVS_UNUSED,
126 SFLReceiver *receiver OVS_UNUSED, u_char *pkt,
129 struct ofproto_sflow *os = os_;
130 collectors_send(os->collectors, pkt, pktLen);
134 sflow_agent_get_counters(void *os_, SFLPoller *poller,
135 SFL_COUNTERS_SAMPLE_TYPE *cs)
137 struct ofproto_sflow *os = os_;
138 SFLCounters_sample_element elem;
139 struct ofproto_sflow_port *osp;
140 SFLIf_counters *counters;
141 struct netdev_stats stats;
142 enum netdev_flags flags;
145 osp = port_array_get(&os->ports, poller->bridgePort);
150 elem.tag = SFLCOUNTERS_GENERIC;
151 counters = &elem.counterBlock.generic;
152 counters->ifIndex = SFL_DS_INDEX(poller->dsi);
153 counters->ifType = 6;
154 if (!netdev_get_features(osp->netdev, ¤t, NULL, NULL, NULL)) {
155 /* The values of ifDirection come from MAU MIB (RFC 2668): 0 = unknown,
156 1 = full-duplex, 2 = half-duplex, 3 = in, 4=out */
157 counters->ifSpeed = netdev_features_to_bps(current);
158 counters->ifDirection = (netdev_features_is_full_duplex(current)
161 counters->ifSpeed = 100000000;
162 counters->ifDirection = 0;
164 if (!netdev_get_flags(osp->netdev, &flags) && flags & NETDEV_UP) {
167 counters->ifStatus = 1; /* ifAdminStatus up. */
168 if (!netdev_get_carrier(osp->netdev, &carrier) && carrier) {
169 counters->ifStatus |= 2; /* ifOperStatus us. */
172 counters->ifStatus = 0; /* Down. */
176 1. Is the multicast counter filled in?
177 2. Does the multicast counter include broadcasts?
178 3. Does the rx_packets counter include multicasts/broadcasts?
180 netdev_get_stats(osp->netdev, &stats);
181 counters->ifInOctets = stats.rx_bytes;
182 counters->ifInUcastPkts = stats.rx_packets;
183 counters->ifInMulticastPkts = stats.multicast;
184 counters->ifInBroadcastPkts = -1;
185 counters->ifInDiscards = stats.rx_dropped;
186 counters->ifInErrors = stats.rx_errors;
187 counters->ifInUnknownProtos = -1;
188 counters->ifOutOctets = stats.tx_bytes;
189 counters->ifOutUcastPkts = stats.tx_packets;
190 counters->ifOutMulticastPkts = -1;
191 counters->ifOutBroadcastPkts = -1;
192 counters->ifOutDiscards = stats.tx_dropped;
193 counters->ifOutErrors = stats.tx_errors;
194 counters->ifPromiscuousMode = 0;
196 SFLADD_ELEMENT(cs, &elem);
197 sfl_poller_writeCountersSample(poller, cs);
200 /* Obtains an address to use for the local sFlow agent and stores it into
201 * '*agent_addr'. Returns true if successful, false on failure.
203 * The sFlow agent address should be a local IP address that is persistent and
204 * reachable over the network, if possible. The IP address associated with
205 * 'agent_device' is used if it has one, and otherwise 'control_ip', the IP
206 * address used to talk to the controller. */
208 sflow_choose_agent_address(const char *agent_device, const char *control_ip,
209 SFLAddress *agent_addr)
213 memset(agent_addr, 0, sizeof *agent_addr);
214 agent_addr->type = SFLADDRESSTYPE_IP_V4;
217 struct netdev *netdev;
219 if (!netdev_open_default(agent_device, &netdev)) {
220 int error = netdev_get_in4(netdev, &in4, NULL);
221 netdev_close(netdev);
228 if (control_ip && !lookup_ip(control_ip, &in4)) {
232 VLOG_ERR("could not determine IP address for sFlow agent");
236 agent_addr->address.ip_v4.addr = in4.s_addr;
241 ofproto_sflow_clear(struct ofproto_sflow *os)
243 struct ofproto_sflow_port *osp;
244 unsigned int xflow_port;
246 if (os->sflow_agent) {
247 sfl_agent_release(os->sflow_agent);
248 os->sflow_agent = NULL;
250 collectors_destroy(os->collectors);
251 os->collectors = NULL;
252 ofproto_sflow_options_destroy(os->options);
255 PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
256 ofproto_sflow_del_port(os, xflow_port);
258 port_array_clear(&os->ports);
260 /* Turn off sampling to save CPU cycles. */
261 xfif_set_sflow_probability(os->xfif, 0);
265 ofproto_sflow_is_enabled(const struct ofproto_sflow *os)
267 return os->collectors != NULL;
270 struct ofproto_sflow *
271 ofproto_sflow_create(struct xfif *xfif)
273 struct ofproto_sflow *os;
275 os = xcalloc(1, sizeof *os);
277 os->next_tick = time_now() + 1;
278 port_array_init(&os->ports);
283 ofproto_sflow_destroy(struct ofproto_sflow *os)
286 ofproto_sflow_clear(os);
287 port_array_destroy(&os->ports);
293 ofproto_sflow_add_poller(struct ofproto_sflow *os,
294 struct ofproto_sflow_port *osp, uint16_t xflow_port)
296 SFLPoller *poller = sfl_agent_addPoller(os->sflow_agent, &osp->dsi, os,
297 sflow_agent_get_counters);
298 sfl_poller_set_sFlowCpInterval(poller, os->options->polling_interval);
299 sfl_poller_set_sFlowCpReceiver(poller, RECEIVER_INDEX);
300 sfl_poller_set_bridgePort(poller, xflow_port);
304 ofproto_sflow_add_sampler(struct ofproto_sflow *os,
305 struct ofproto_sflow_port *osp,
306 u_int32_t sampling_rate, u_int32_t header_len)
308 SFLSampler *sampler = sfl_agent_addSampler(os->sflow_agent, &osp->dsi);
309 sfl_sampler_set_sFlowFsPacketSamplingRate(sampler, sampling_rate);
310 sfl_sampler_set_sFlowFsMaximumHeaderSize(sampler, header_len);
311 sfl_sampler_set_sFlowFsReceiver(sampler, RECEIVER_INDEX);
315 ofproto_sflow_add_port(struct ofproto_sflow *os, uint16_t xflow_port,
316 const char *netdev_name)
318 struct ofproto_sflow_port *osp;
319 struct netdev *netdev;
323 ofproto_sflow_del_port(os, xflow_port);
325 /* Open network device. */
326 error = netdev_open_default(netdev_name, &netdev);
328 VLOG_WARN_RL(&rl, "failed to open network device \"%s\": %s",
329 netdev_name, strerror(error));
333 /* Add to table of ports. */
334 osp = xmalloc(sizeof *osp);
335 osp->netdev = netdev;
336 ifindex = netdev_get_ifindex(netdev);
338 ifindex = (os->sflow_agent->subId << 16) + xflow_port;
340 SFL_DS_SET(osp->dsi, 0, ifindex, 0);
341 port_array_set(&os->ports, xflow_port, osp);
344 if (os->sflow_agent) {
345 ofproto_sflow_add_poller(os, osp, xflow_port);
350 ofproto_sflow_del_port(struct ofproto_sflow *os, uint16_t xflow_port)
352 struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
354 if (os->sflow_agent) {
355 sfl_agent_removePoller(os->sflow_agent, &osp->dsi);
356 sfl_agent_removeSampler(os->sflow_agent, &osp->dsi);
358 netdev_close(osp->netdev);
360 port_array_set(&os->ports, xflow_port, NULL);
365 ofproto_sflow_set_options(struct ofproto_sflow *os,
366 const struct ofproto_sflow_options *options)
368 struct ofproto_sflow_port *osp;
369 bool options_changed;
370 SFLReceiver *receiver;
371 unsigned int xflow_port;
375 if (!options->targets.n || !options->sampling_rate) {
376 /* No point in doing any work if there are no targets or nothing to
378 ofproto_sflow_clear(os);
382 options_changed = (!os->options
383 || !ofproto_sflow_options_equal(options, os->options));
385 /* Configure collectors if options have changed or if we're shortchanged in
386 * collectors (which indicates that opening one or more of the configured
387 * collectors failed, so that we should retry). */
389 || collectors_count(os->collectors) < options->targets.n) {
390 collectors_destroy(os->collectors);
391 collectors_create(&options->targets, SFL_DEFAULT_COLLECTOR_PORT,
393 if (os->collectors == NULL) {
394 VLOG_WARN_RL(&rl, "no collectors could be initialized, "
396 ofproto_sflow_clear(os);
401 /* Avoid reconfiguring if options didn't change. */
402 if (!options_changed) {
405 ofproto_sflow_options_destroy(os->options);
406 os->options = ofproto_sflow_options_clone(options);
408 /* Choose agent IP address. */
409 if (!sflow_choose_agent_address(options->agent_device,
410 options->control_ip, &agentIP)) {
411 ofproto_sflow_clear(os);
416 VLOG_INFO("creating sFlow agent %d", options->sub_id);
417 if (os->sflow_agent) {
418 sfl_agent_release(os->sflow_agent);
420 os->sflow_agent = xcalloc(1, sizeof *os->sflow_agent);
422 sfl_agent_init(os->sflow_agent,
425 now, /* Boot time. */
426 now, /* Current time. */
427 os, /* Pointer supplied to callbacks. */
428 sflow_agent_alloc_cb,
430 sflow_agent_error_cb,
431 sflow_agent_send_packet_cb);
433 receiver = sfl_agent_addReceiver(os->sflow_agent);
434 sfl_receiver_set_sFlowRcvrOwner(receiver, "Open vSwitch sFlow");
435 sfl_receiver_set_sFlowRcvrTimeout(receiver, 0xffffffff);
437 /* Set the sampling_rate down in the datapath. */
438 xfif_set_sflow_probability(os->xfif,
439 MAX(1, UINT32_MAX / options->sampling_rate));
441 /* Add samplers and pollers for the currently known ports. */
442 PORT_ARRAY_FOR_EACH (osp, &os->ports, xflow_port) {
443 ofproto_sflow_add_sampler(os, osp,
444 options->sampling_rate, options->header_len);
449 ofproto_sflow_xflow_port_to_ifindex(const struct ofproto_sflow *os,
452 struct ofproto_sflow_port *osp = port_array_get(&os->ports, xflow_port);
453 return osp ? SFL_DS_INDEX(osp->dsi) : 0;
457 ofproto_sflow_received(struct ofproto_sflow *os, struct xflow_msg *msg)
459 SFL_FLOW_SAMPLE_TYPE fs;
460 SFLFlow_sample_element hdrElem;
461 SFLSampled_header *header;
462 SFLFlow_sample_element switchElem;
464 const struct xflow_sflow_sample_header *hdr;
465 const union xflow_action *actions;
466 struct ofpbuf payload;
467 size_t n_actions, n_outputs;
472 /* Get xflow_sflow_sample_header. */
473 min_size = sizeof *msg + sizeof *hdr;
474 if (min_size > msg->length) {
475 VLOG_WARN_RL(&rl, "sFlow packet too small (%"PRIu32" < %zu)",
476 msg->length, min_size);
479 hdr = (const struct xflow_sflow_sample_header *) (msg + 1);
482 n_actions = hdr->n_actions;
483 if (n_actions > 65536 / sizeof *actions) {
484 VLOG_WARN_RL(&rl, "too many actions in sFlow packet (%zu > %zu)",
485 65536 / sizeof *actions, n_actions);
488 min_size += n_actions * sizeof *actions;
489 if (min_size > msg->length) {
490 VLOG_WARN_RL(&rl, "sFlow packet with %zu actions too small "
492 n_actions, msg->length, min_size);
495 actions = (const union xflow_action *) (hdr + 1);
497 /* Get packet payload and extract flow. */
498 payload.data = (union xflow_action *) (actions + n_actions);
499 payload.size = msg->length - min_size;
500 flow_extract(&payload, msg->port, &flow);
502 /* Build a flow sample */
503 memset(&fs, 0, sizeof fs);
504 fs.input = ofproto_sflow_xflow_port_to_ifindex(os, msg->port);
505 fs.output = 0; /* Filled in correctly below. */
506 fs.sample_pool = hdr->sample_pool;
508 /* We are going to give it to the sampler that represents this input port.
509 * By implementing "ingress-only" sampling like this we ensure that we
510 * never have to offer the same sample to more than one sampler. */
511 sampler = sfl_agent_getSamplerByIfIndex(os->sflow_agent, fs.input);
513 VLOG_WARN_RL(&rl, "no sampler for input ifIndex (%"PRIu32")",
518 /* Sampled header. */
519 memset(&hdrElem, 0, sizeof hdrElem);
520 hdrElem.tag = SFLFLOW_HEADER;
521 header = &hdrElem.flowType.header;
522 header->header_protocol = SFLHEADER_ETHERNET_ISO8023;
523 header->frame_length = payload.size;
524 header->stripped = 4; /* Ethernet FCS stripped off. */
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 switchElem.flowType.sw.dst_vlan = -1; /* Filled in correctly below. */
535 switchElem.flowType.sw.dst_priority = switchElem.flowType.sw.src_priority;
537 /* Figure out the output ports. */
539 for (i = 0; i < n_actions; i++) {
540 const union xflow_action *a = &actions[i];
544 fs.output = ofproto_sflow_xflow_port_to_ifindex(os, a->output.port);
548 case XFLOWAT_OUTPUT_GROUP:
549 n_outputs += (a->output_group.group == DP_GROUP_FLOOD ? os->n_flood
550 : a->output_group.group == DP_GROUP_ALL ? os->n_all
554 case XFLOWAT_SET_DL_TCI:
555 if (a->dl_tci.mask & htons(VLAN_VID_MASK)) {
556 switchElem.flowType.sw.dst_vlan = vlan_tci_to_vid(a->dl_tci.tci);
558 if (a->dl_tci.mask & htons(VLAN_PCP_MASK)) {
559 switchElem.flowType.sw.dst_priority = vlan_tci_to_pcp(a->dl_tci.tci);
568 /* Set output port, as defined by http://www.sflow.org/sflow_version_5.txt
569 (search for "Input/output port information"). */
571 /* This value indicates that the packet was dropped for an unknown
573 fs.output = 0x40000000 | 256;
574 } else if (n_outputs > 1 || !fs.output) {
575 /* Setting the high bit means "multiple output ports". */
576 fs.output = 0x80000000 | n_outputs;
579 /* Submit the flow sample to be encoded into the next datagram. */
580 SFLADD_ELEMENT(&fs, &hdrElem);
581 SFLADD_ELEMENT(&fs, &switchElem);
582 sfl_sampler_writeFlowSample(sampler, &fs);
586 ofproto_sflow_set_group_sizes(struct ofproto_sflow *os,
587 size_t n_flood, size_t n_all)
589 os->n_flood = n_flood;
594 ofproto_sflow_run(struct ofproto_sflow *os)
596 if (ofproto_sflow_is_enabled(os)) {
597 time_t now = time_now();
598 if (now >= os->next_tick) {
599 sfl_agent_tick(os->sflow_agent, now);
600 os->next_tick = now + 1;
606 ofproto_sflow_wait(struct ofproto_sflow *os)
608 if (ofproto_sflow_is_enabled(os)) {
609 poll_timer_wait(os->next_tick * 1000 - time_msec());