2 * daemon/stats.c - collect runtime performance indicators.
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
6 * This software is open source.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 * This file describes the data structure used to collect runtime performance
40 * numbers. These 'statistics' may be of interest to the operator.
47 #include <sys/types.h>
48 #include "daemon/stats.h"
49 #include "daemon/worker.h"
50 #include "daemon/daemon.h"
51 #include "services/mesh.h"
52 #include "services/outside_network.h"
53 #include "services/listen_dnsport.h"
54 #include "util/config_file.h"
55 #include "util/tube.h"
56 #include "util/timehist.h"
57 #include "util/net_help.h"
58 #include "validator/validator.h"
59 #include "iterator/iterator.h"
60 #include "sldns/sbuffer.h"
61 #include "services/cache/rrset.h"
62 #include "services/cache/infra.h"
63 #include "services/authzone.h"
64 #include "validator/val_kcache.h"
65 #include "validator/val_neg.h"
67 #include "edns-subnet/subnetmod.h"
70 #include <openssl/ssl.h>
73 /** add timers and the values do not overflow or become negative */
75 stats_timeval_add(long long* d_sec, long long* d_usec, long long add_sec, long long add_usec)
79 (*d_usec) += add_usec;
80 if((*d_usec) >= 1000000) {
87 void server_stats_init(struct ub_server_stats* stats, struct config_file* cfg)
89 memset(stats, 0, sizeof(*stats));
90 stats->extended = cfg->stat_extended;
93 void server_stats_querymiss(struct ub_server_stats* stats, struct worker* worker)
95 stats->num_queries_missed_cache++;
96 stats->sum_query_list_size += worker->env.mesh->all.count;
97 if((long long)worker->env.mesh->all.count > stats->max_query_list_size)
98 stats->max_query_list_size = (long long)worker->env.mesh->all.count;
101 void server_stats_prefetch(struct ub_server_stats* stats, struct worker* worker)
103 stats->num_queries_prefetch++;
104 /* changes the query list size so account that, like a querymiss */
105 stats->sum_query_list_size += worker->env.mesh->all.count;
106 if((long long)worker->env.mesh->all.count > stats->max_query_list_size)
107 stats->max_query_list_size = (long long)worker->env.mesh->all.count;
110 void server_stats_log(struct ub_server_stats* stats, struct worker* worker,
113 log_info("server stats for thread %d: %u queries, "
114 "%u answers from cache, %u recursions, %u prefetch, %u rejected by "
116 threadnum, (unsigned)stats->num_queries,
117 (unsigned)(stats->num_queries -
118 stats->num_queries_missed_cache),
119 (unsigned)stats->num_queries_missed_cache,
120 (unsigned)stats->num_queries_prefetch,
121 (unsigned)stats->num_queries_ip_ratelimited);
122 log_info("server stats for thread %d: requestlist max %u avg %g "
123 "exceeded %u jostled %u", threadnum,
124 (unsigned)stats->max_query_list_size,
125 (stats->num_queries_missed_cache+stats->num_queries_prefetch)?
126 (double)stats->sum_query_list_size/
127 (double)(stats->num_queries_missed_cache+
128 stats->num_queries_prefetch) : 0.0,
129 (unsigned)worker->env.mesh->stats_dropped,
130 (unsigned)worker->env.mesh->stats_jostled);
135 /** Set the EDNS Subnet stats. */
137 set_subnet_stats(struct worker* worker, struct ub_server_stats* svr,
140 int m = modstack_find(&worker->env.mesh->mods, "subnet");
141 struct subnet_env* sne;
144 sne = (struct subnet_env*)worker->env.modinfo[m];
145 if(reset && !worker->env.cfg->stat_cumulative) {
146 lock_rw_wrlock(&sne->biglock);
148 lock_rw_rdlock(&sne->biglock);
150 svr->num_query_subnet = (long long)(sne->num_msg_nocache + sne->num_msg_cache);
151 svr->num_query_subnet_cache = (long long)sne->num_msg_cache;
152 if(reset && !worker->env.cfg->stat_cumulative) {
153 sne->num_msg_cache = 0;
154 sne->num_msg_nocache = 0;
156 lock_rw_unlock(&sne->biglock);
158 #endif /* CLIENT_SUBNET */
160 /** Set the neg cache stats. */
162 set_neg_cache_stats(struct worker* worker, struct ub_server_stats* svr,
165 int m = modstack_find(&worker->env.mesh->mods, "validator");
167 struct val_neg_cache* neg;
170 ve = (struct val_env*)worker->env.modinfo[m];
174 lock_basic_lock(&neg->lock);
175 svr->num_neg_cache_noerror = (long long)neg->num_neg_cache_noerror;
176 svr->num_neg_cache_nxdomain = (long long)neg->num_neg_cache_nxdomain;
177 if(reset && !worker->env.cfg->stat_cumulative) {
178 neg->num_neg_cache_noerror = 0;
179 neg->num_neg_cache_nxdomain = 0;
181 lock_basic_unlock(&neg->lock);
184 /** get rrsets bogus number from validator */
186 get_rrset_bogus(struct worker* worker, int reset)
188 int m = modstack_find(&worker->env.mesh->mods, "validator");
193 ve = (struct val_env*)worker->env.modinfo[m];
194 lock_basic_lock(&ve->bogus_lock);
195 r = ve->num_rrset_bogus;
196 if(reset && !worker->env.cfg->stat_cumulative)
197 ve->num_rrset_bogus = 0;
198 lock_basic_unlock(&ve->bogus_lock);
202 /** get number of ratelimited queries from iterator */
204 get_queries_ratelimit(struct worker* worker, int reset)
206 int m = modstack_find(&worker->env.mesh->mods, "iterator");
211 ie = (struct iter_env*)worker->env.modinfo[m];
212 lock_basic_lock(&ie->queries_ratelimit_lock);
213 r = ie->num_queries_ratelimited;
214 if(reset && !worker->env.cfg->stat_cumulative)
215 ie->num_queries_ratelimited = 0;
216 lock_basic_unlock(&ie->queries_ratelimit_lock);
221 /** get the number of shared secret cache miss */
223 get_dnscrypt_cache_miss(struct worker* worker, int reset)
226 struct dnsc_env* de = worker->daemon->dnscenv;
229 lock_basic_lock(&de->shared_secrets_cache_lock);
230 r = de->num_query_dnscrypt_secret_missed_cache;
231 if(reset && !worker->env.cfg->stat_cumulative)
232 de->num_query_dnscrypt_secret_missed_cache = 0;
233 lock_basic_unlock(&de->shared_secrets_cache_lock);
237 /** get the number of replayed queries */
239 get_dnscrypt_replay(struct worker* worker, int reset)
242 struct dnsc_env* de = worker->daemon->dnscenv;
244 lock_basic_lock(&de->nonces_cache_lock);
245 r = de->num_query_dnscrypt_replay;
246 if(reset && !worker->env.cfg->stat_cumulative)
247 de->num_query_dnscrypt_replay = 0;
248 lock_basic_unlock(&de->nonces_cache_lock);
251 #endif /* USE_DNSCRYPT */
254 server_stats_compile(struct worker* worker, struct ub_stats_info* s, int reset)
257 struct listen_list* lp;
259 s->svr = worker->stats;
260 s->mesh_num_states = (long long)worker->env.mesh->all.count;
261 s->mesh_num_reply_states = (long long)worker->env.mesh->num_reply_states;
262 s->mesh_jostled = (long long)worker->env.mesh->stats_jostled;
263 s->mesh_dropped = (long long)worker->env.mesh->stats_dropped;
264 s->mesh_replies_sent = (long long)worker->env.mesh->replies_sent;
265 s->mesh_replies_sum_wait_sec = (long long)worker->env.mesh->replies_sum_wait.tv_sec;
266 s->mesh_replies_sum_wait_usec = (long long)worker->env.mesh->replies_sum_wait.tv_usec;
267 s->mesh_time_median = timehist_quartile(worker->env.mesh->histogram,
270 /* add in the values from the mesh */
271 s->svr.ans_secure += (long long)worker->env.mesh->ans_secure;
272 s->svr.ans_bogus += (long long)worker->env.mesh->ans_bogus;
273 s->svr.ans_rcode_nodata += (long long)worker->env.mesh->ans_nodata;
274 s->svr.ans_expired += (long long)worker->env.mesh->ans_expired;
275 for(i=0; i<UB_STATS_RCODE_NUM; i++)
276 s->svr.ans_rcode[i] += (long long)worker->env.mesh->ans_rcode[i];
277 for(i=0; i<UB_STATS_RPZ_ACTION_NUM; i++)
278 s->svr.rpz_action[i] += (long long)worker->env.mesh->rpz_action[i];
279 timehist_export(worker->env.mesh->histogram, s->svr.hist,
281 /* values from outside network */
282 s->svr.unwanted_replies = (long long)worker->back->unwanted_replies;
283 s->svr.qtcp_outgoing = (long long)worker->back->num_tcp_outgoing;
285 /* get and reset validator rrset bogus number */
286 s->svr.rrset_bogus = (long long)get_rrset_bogus(worker, reset);
288 /* get and reset iterator query ratelimit number */
289 s->svr.queries_ratelimited = (long long)get_queries_ratelimit(worker, reset);
291 /* get cache sizes */
292 s->svr.msg_cache_count = (long long)count_slabhash_entries(worker->env.msg_cache);
293 s->svr.rrset_cache_count = (long long)count_slabhash_entries(&worker->env.rrset_cache->table);
294 s->svr.infra_cache_count = (long long)count_slabhash_entries(worker->env.infra_cache->hosts);
295 if(worker->env.key_cache)
296 s->svr.key_cache_count = (long long)count_slabhash_entries(worker->env.key_cache->slab);
297 else s->svr.key_cache_count = 0;
300 if(worker->daemon->dnscenv) {
301 s->svr.num_query_dnscrypt_secret_missed_cache =
302 (long long)get_dnscrypt_cache_miss(worker, reset);
303 s->svr.shared_secret_cache_count = (long long)count_slabhash_entries(
304 worker->daemon->dnscenv->shared_secrets_cache);
305 s->svr.nonce_cache_count = (long long)count_slabhash_entries(
306 worker->daemon->dnscenv->nonces_cache);
307 s->svr.num_query_dnscrypt_replay =
308 (long long)get_dnscrypt_replay(worker, reset);
310 s->svr.num_query_dnscrypt_secret_missed_cache = 0;
311 s->svr.shared_secret_cache_count = 0;
312 s->svr.nonce_cache_count = 0;
313 s->svr.num_query_dnscrypt_replay = 0;
316 s->svr.num_query_dnscrypt_secret_missed_cache = 0;
317 s->svr.shared_secret_cache_count = 0;
318 s->svr.nonce_cache_count = 0;
319 s->svr.num_query_dnscrypt_replay = 0;
320 #endif /* USE_DNSCRYPT */
321 if(worker->env.auth_zones) {
322 if(reset && !worker->env.cfg->stat_cumulative) {
323 lock_rw_wrlock(&worker->env.auth_zones->lock);
325 lock_rw_rdlock(&worker->env.auth_zones->lock);
327 s->svr.num_query_authzone_up = (long long)worker->env.
328 auth_zones->num_query_up;
329 s->svr.num_query_authzone_down = (long long)worker->env.
330 auth_zones->num_query_down;
331 if(reset && !worker->env.cfg->stat_cumulative) {
332 worker->env.auth_zones->num_query_up = 0;
333 worker->env.auth_zones->num_query_down = 0;
335 lock_rw_unlock(&worker->env.auth_zones->lock);
337 s->svr.mem_stream_wait =
338 (long long)tcp_req_info_get_stream_buffer_size();
339 s->svr.mem_http2_query_buffer =
340 (long long)http2_get_query_buffer_size();
341 s->svr.mem_http2_response_buffer =
342 (long long)http2_get_response_buffer_size();
344 /* Set neg cache usage numbers */
345 set_neg_cache_stats(worker, &s->svr, reset);
347 /* EDNS Subnet usage numbers */
348 set_subnet_stats(worker, &s->svr, reset);
350 s->svr.num_query_subnet = 0;
351 s->svr.num_query_subnet_cache = 0;
354 /* get tcp accept usage */
355 s->svr.tcp_accept_usage = 0;
356 for(lp = worker->front->cps; lp; lp = lp->next) {
357 if(lp->com->type == comm_tcp_accept)
358 s->svr.tcp_accept_usage += (long long)lp->com->cur_tcp_count;
361 if(reset && !worker->env.cfg->stat_cumulative) {
362 worker_stats_clear(worker);
366 void server_stats_obtain(struct worker* worker, struct worker* who,
367 struct ub_stats_info* s, int reset)
369 uint8_t *reply = NULL;
372 /* just fill it in */
373 server_stats_compile(worker, s, reset);
376 /* communicate over tube */
377 verbose(VERB_ALGO, "write stats cmd");
379 worker_send_cmd(who, worker_cmd_stats);
380 else worker_send_cmd(who, worker_cmd_stats_noreset);
381 verbose(VERB_ALGO, "wait for stats reply");
382 if(!tube_read_msg(worker->cmd, &reply, &len, 0))
383 fatal_exit("failed to read stats over cmd channel");
384 if(len != (uint32_t)sizeof(*s))
385 fatal_exit("stats on cmd channel wrong length %d %d",
386 (int)len, (int)sizeof(*s));
387 memcpy(s, reply, (size_t)len);
391 void server_stats_reply(struct worker* worker, int reset)
393 struct ub_stats_info s;
394 server_stats_compile(worker, &s, reset);
395 verbose(VERB_ALGO, "write stats replymsg");
396 if(!tube_write_msg(worker->daemon->workers[0]->cmd,
397 (uint8_t*)&s, sizeof(s), 0))
398 fatal_exit("could not write stat values over cmd channel");
401 void server_stats_add(struct ub_stats_info* total, struct ub_stats_info* a)
403 total->svr.num_queries += a->svr.num_queries;
404 total->svr.num_queries_ip_ratelimited += a->svr.num_queries_ip_ratelimited;
405 total->svr.num_queries_missed_cache += a->svr.num_queries_missed_cache;
406 total->svr.num_queries_prefetch += a->svr.num_queries_prefetch;
407 total->svr.sum_query_list_size += a->svr.sum_query_list_size;
408 total->svr.ans_expired += a->svr.ans_expired;
410 total->svr.num_query_dnscrypt_crypted += a->svr.num_query_dnscrypt_crypted;
411 total->svr.num_query_dnscrypt_cert += a->svr.num_query_dnscrypt_cert;
412 total->svr.num_query_dnscrypt_cleartext += \
413 a->svr.num_query_dnscrypt_cleartext;
414 total->svr.num_query_dnscrypt_crypted_malformed += \
415 a->svr.num_query_dnscrypt_crypted_malformed;
416 #endif /* USE_DNSCRYPT */
417 /* the max size reached is upped to higher of both */
418 if(a->svr.max_query_list_size > total->svr.max_query_list_size)
419 total->svr.max_query_list_size = a->svr.max_query_list_size;
421 if(a->svr.extended) {
423 total->svr.qtype_big += a->svr.qtype_big;
424 total->svr.qclass_big += a->svr.qclass_big;
425 total->svr.qtcp += a->svr.qtcp;
426 total->svr.qtcp_outgoing += a->svr.qtcp_outgoing;
427 total->svr.qtls += a->svr.qtls;
428 total->svr.qtls_resume += a->svr.qtls_resume;
429 total->svr.qhttps += a->svr.qhttps;
430 total->svr.qipv6 += a->svr.qipv6;
431 total->svr.qbit_QR += a->svr.qbit_QR;
432 total->svr.qbit_AA += a->svr.qbit_AA;
433 total->svr.qbit_TC += a->svr.qbit_TC;
434 total->svr.qbit_RD += a->svr.qbit_RD;
435 total->svr.qbit_RA += a->svr.qbit_RA;
436 total->svr.qbit_Z += a->svr.qbit_Z;
437 total->svr.qbit_AD += a->svr.qbit_AD;
438 total->svr.qbit_CD += a->svr.qbit_CD;
439 total->svr.qEDNS += a->svr.qEDNS;
440 total->svr.qEDNS_DO += a->svr.qEDNS_DO;
441 total->svr.ans_rcode_nodata += a->svr.ans_rcode_nodata;
442 total->svr.ans_secure += a->svr.ans_secure;
443 total->svr.ans_bogus += a->svr.ans_bogus;
444 total->svr.unwanted_replies += a->svr.unwanted_replies;
445 total->svr.unwanted_queries += a->svr.unwanted_queries;
446 total->svr.tcp_accept_usage += a->svr.tcp_accept_usage;
447 for(i=0; i<UB_STATS_QTYPE_NUM; i++)
448 total->svr.qtype[i] += a->svr.qtype[i];
449 for(i=0; i<UB_STATS_QCLASS_NUM; i++)
450 total->svr.qclass[i] += a->svr.qclass[i];
451 for(i=0; i<UB_STATS_OPCODE_NUM; i++)
452 total->svr.qopcode[i] += a->svr.qopcode[i];
453 for(i=0; i<UB_STATS_RCODE_NUM; i++)
454 total->svr.ans_rcode[i] += a->svr.ans_rcode[i];
455 for(i=0; i<NUM_BUCKETS_HIST; i++)
456 total->svr.hist[i] += a->svr.hist[i];
457 for(i=0; i<UB_STATS_RPZ_ACTION_NUM; i++)
458 total->svr.rpz_action[i] += a->svr.rpz_action[i];
461 total->mesh_num_states += a->mesh_num_states;
462 total->mesh_num_reply_states += a->mesh_num_reply_states;
463 total->mesh_jostled += a->mesh_jostled;
464 total->mesh_dropped += a->mesh_dropped;
465 total->mesh_replies_sent += a->mesh_replies_sent;
466 stats_timeval_add(&total->mesh_replies_sum_wait_sec, &total->mesh_replies_sum_wait_usec, a->mesh_replies_sum_wait_sec, a->mesh_replies_sum_wait_usec);
467 /* the medians are averaged together, this is not as accurate as
468 * taking the median over all of the data, but is good and fast
469 * added up here, division later*/
470 total->mesh_time_median += a->mesh_time_median;
473 void server_stats_insquery(struct ub_server_stats* stats, struct comm_point* c,
474 uint16_t qtype, uint16_t qclass, struct edns_data* edns,
475 struct comm_reply* repinfo)
477 uint16_t flags = sldns_buffer_read_u16_at(c->buffer, 2);
478 if(qtype < UB_STATS_QTYPE_NUM)
479 stats->qtype[qtype]++;
480 else stats->qtype_big++;
481 if(qclass < UB_STATS_QCLASS_NUM)
482 stats->qclass[qclass]++;
483 else stats->qclass_big++;
484 stats->qopcode[ LDNS_OPCODE_WIRE(sldns_buffer_begin(c->buffer)) ]++;
485 if(c->type != comm_udp) {
490 if(SSL_session_reused(c->ssl))
491 stats->qtls_resume++;
493 if(c->type == comm_http)
497 if(repinfo && addr_is_ip6(&repinfo->addr, repinfo->addrlen))
515 if(edns->edns_present) {
517 if( (edns->bits & EDNS_DO) )
522 void server_stats_insrcode(struct ub_server_stats* stats, sldns_buffer* buf)
524 if(stats->extended && sldns_buffer_limit(buf) != 0) {
525 int r = (int)LDNS_RCODE_WIRE( sldns_buffer_begin(buf) );
526 stats->ans_rcode[r] ++;
527 if(r == 0 && LDNS_ANCOUNT( sldns_buffer_begin(buf) ) == 0)
528 stats->ans_rcode_nodata ++;