diff options
Diffstat (limited to 'external/unbound/daemon/remote.c')
-rw-r--r-- | external/unbound/daemon/remote.c | 2449 |
1 files changed, 2449 insertions, 0 deletions
diff --git a/external/unbound/daemon/remote.c b/external/unbound/daemon/remote.c new file mode 100644 index 000000000..100aa8be1 --- /dev/null +++ b/external/unbound/daemon/remote.c @@ -0,0 +1,2449 @@ +/* + * daemon/remote.c - remote control for the unbound daemon. + * + * Copyright (c) 2008, NLnet Labs. All rights reserved. + * + * This software is open source. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * Redistributions of source code must retain the above copyright notice, + * this list of conditions and the following disclaimer. + * + * Redistributions in binary form must reproduce the above copyright notice, + * this list of conditions and the following disclaimer in the documentation + * and/or other materials provided with the distribution. + * + * Neither the name of the NLNET LABS nor the names of its contributors may + * be used to endorse or promote products derived from this software without + * specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED + * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +/** + * \file + * + * This file contains the remote control functionality for the daemon. + * The remote control can be performed using either the commandline + * unbound-control tool, or a SSLv3/TLS capable web browser. + * The channel is secured using SSLv3 or TLSv1, and certificates. + * Both the server and the client(control tool) have their own keys. + */ +#include "config.h" +#ifdef HAVE_OPENSSL_ERR_H +#include <openssl/err.h> +#endif +#include <ctype.h> +#include "daemon/remote.h" +#include "daemon/worker.h" +#include "daemon/daemon.h" +#include "daemon/stats.h" +#include "daemon/cachedump.h" +#include "util/log.h" +#include "util/config_file.h" +#include "util/net_help.h" +#include "util/module.h" +#include "services/listen_dnsport.h" +#include "services/cache/rrset.h" +#include "services/cache/infra.h" +#include "services/mesh.h" +#include "services/localzone.h" +#include "util/storage/slabhash.h" +#include "util/fptr_wlist.h" +#include "util/data/dname.h" +#include "validator/validator.h" +#include "validator/val_kcache.h" +#include "validator/val_kentry.h" +#include "validator/val_anchor.h" +#include "iterator/iterator.h" +#include "iterator/iter_fwd.h" +#include "iterator/iter_hints.h" +#include "iterator/iter_delegpt.h" +#include "services/outbound_list.h" +#include "services/outside_network.h" +#include "ldns/str2wire.h" +#include "ldns/parseutil.h" +#include "ldns/wire2str.h" +#include "ldns/sbuffer.h" + +#ifdef HAVE_SYS_TYPES_H +# include <sys/types.h> +#endif +#ifdef HAVE_NETDB_H +#include <netdb.h> +#endif + +/* just for portability */ +#ifdef SQ +#undef SQ +#endif + +/** what to put on statistics lines between var and value, ": " or "=" */ +#define SQ "=" +/** if true, inhibits a lot of =0 lines from the stats output */ +static const int inhibit_zero = 1; + +/** subtract timers and the values do not overflow or become negative */ +static void +timeval_subtract(struct timeval* d, const struct timeval* end, + const struct timeval* start) +{ +#ifndef S_SPLINT_S + time_t end_usec = end->tv_usec; + d->tv_sec = end->tv_sec - start->tv_sec; + if(end_usec < start->tv_usec) { + end_usec += 1000000; + d->tv_sec--; + } + d->tv_usec = end_usec - start->tv_usec; +#endif +} + +/** divide sum of timers to get average */ +static void +timeval_divide(struct timeval* avg, const struct timeval* sum, size_t d) +{ +#ifndef S_SPLINT_S + size_t leftover; + if(d == 0) { + avg->tv_sec = 0; + avg->tv_usec = 0; + return; + } + avg->tv_sec = sum->tv_sec / d; + avg->tv_usec = sum->tv_usec / d; + /* handle fraction from seconds divide */ + leftover = sum->tv_sec - avg->tv_sec*d; + avg->tv_usec += (leftover*1000000)/d; +#endif +} + +struct daemon_remote* +daemon_remote_create(struct config_file* cfg) +{ + char* s_cert; + char* s_key; + struct daemon_remote* rc = (struct daemon_remote*)calloc(1, + sizeof(*rc)); + if(!rc) { + log_err("out of memory in daemon_remote_create"); + return NULL; + } + rc->max_active = 10; + + if(!cfg->remote_control_enable) { + rc->ctx = NULL; + return rc; + } + rc->ctx = SSL_CTX_new(SSLv23_server_method()); + if(!rc->ctx) { + log_crypto_err("could not SSL_CTX_new"); + free(rc); + return NULL; + } + /* no SSLv2 because has defects */ + if(!(SSL_CTX_set_options(rc->ctx, SSL_OP_NO_SSLv2) & SSL_OP_NO_SSLv2)){ + log_crypto_err("could not set SSL_OP_NO_SSLv2"); + daemon_remote_delete(rc); + return NULL; + } + s_cert = fname_after_chroot(cfg->server_cert_file, cfg, 1); + s_key = fname_after_chroot(cfg->server_key_file, cfg, 1); + if(!s_cert || !s_key) { + log_err("out of memory in remote control fname"); + goto setup_error; + } + verbose(VERB_ALGO, "setup SSL certificates"); + if (!SSL_CTX_use_certificate_file(rc->ctx,s_cert,SSL_FILETYPE_PEM)) { + log_err("Error for server-cert-file: %s", s_cert); + log_crypto_err("Error in SSL_CTX use_certificate_file"); + goto setup_error; + } + if(!SSL_CTX_use_PrivateKey_file(rc->ctx,s_key,SSL_FILETYPE_PEM)) { + log_err("Error for server-key-file: %s", s_key); + log_crypto_err("Error in SSL_CTX use_PrivateKey_file"); + goto setup_error; + } + if(!SSL_CTX_check_private_key(rc->ctx)) { + log_err("Error for server-key-file: %s", s_key); + log_crypto_err("Error in SSL_CTX check_private_key"); + goto setup_error; + } + if(!SSL_CTX_load_verify_locations(rc->ctx, s_cert, NULL)) { + log_crypto_err("Error setting up SSL_CTX verify locations"); + setup_error: + free(s_cert); + free(s_key); + daemon_remote_delete(rc); + return NULL; + } + SSL_CTX_set_client_CA_list(rc->ctx, SSL_load_client_CA_file(s_cert)); + SSL_CTX_set_verify(rc->ctx, SSL_VERIFY_PEER, NULL); + free(s_cert); + free(s_key); + + return rc; +} + +void daemon_remote_clear(struct daemon_remote* rc) +{ + struct rc_state* p, *np; + if(!rc) return; + /* but do not close the ports */ + listen_list_delete(rc->accept_list); + rc->accept_list = NULL; + /* do close these sockets */ + p = rc->busy_list; + while(p) { + np = p->next; + if(p->ssl) + SSL_free(p->ssl); + comm_point_delete(p->c); + free(p); + p = np; + } + rc->busy_list = NULL; + rc->active = 0; + rc->worker = NULL; +} + +void daemon_remote_delete(struct daemon_remote* rc) +{ + if(!rc) return; + daemon_remote_clear(rc); + if(rc->ctx) { + SSL_CTX_free(rc->ctx); + } + free(rc); +} + +/** + * Add and open a new control port + * @param ip: ip str + * @param nr: port nr + * @param list: list head + * @param noproto_is_err: if lack of protocol support is an error. + * @return false on failure. + */ +static int +add_open(const char* ip, int nr, struct listen_port** list, int noproto_is_err) +{ + struct addrinfo hints; + struct addrinfo* res; + struct listen_port* n; + int noproto; + int fd, r; + char port[15]; + snprintf(port, sizeof(port), "%d", nr); + port[sizeof(port)-1]=0; + memset(&hints, 0, sizeof(hints)); + hints.ai_socktype = SOCK_STREAM; + hints.ai_flags = AI_PASSIVE | AI_NUMERICHOST; + if((r = getaddrinfo(ip, port, &hints, &res)) != 0 || !res) { +#ifdef USE_WINSOCK + if(!noproto_is_err && r == EAI_NONAME) { + /* tried to lookup the address as name */ + return 1; /* return success, but do nothing */ + } +#endif /* USE_WINSOCK */ + log_err("control interface %s:%s getaddrinfo: %s %s", + ip?ip:"default", port, gai_strerror(r), +#ifdef EAI_SYSTEM + r==EAI_SYSTEM?(char*)strerror(errno):"" +#else + "" +#endif + ); + return 0; + } + + /* open fd */ + fd = create_tcp_accept_sock(res, 1, &noproto, 0); + freeaddrinfo(res); + if(fd == -1 && noproto) { + if(!noproto_is_err) + return 1; /* return success, but do nothing */ + log_err("cannot open control interface %s %d : " + "protocol not supported", ip, nr); + return 0; + } + if(fd == -1) { + log_err("cannot open control interface %s %d", ip, nr); + return 0; + } + + /* alloc */ + n = (struct listen_port*)calloc(1, sizeof(*n)); + if(!n) { +#ifndef USE_WINSOCK + close(fd); +#else + closesocket(fd); +#endif + log_err("out of memory"); + return 0; + } + n->next = *list; + *list = n; + n->fd = fd; + return 1; +} + +struct listen_port* daemon_remote_open_ports(struct config_file* cfg) +{ + struct listen_port* l = NULL; + log_assert(cfg->remote_control_enable && cfg->control_port); + if(cfg->control_ifs) { + struct config_strlist* p; + for(p = cfg->control_ifs; p; p = p->next) { + if(!add_open(p->str, cfg->control_port, &l, 1)) { + listening_ports_free(l); + return NULL; + } + } + } else { + /* defaults */ + if(cfg->do_ip6 && + !add_open("::1", cfg->control_port, &l, 0)) { + listening_ports_free(l); + return NULL; + } + if(cfg->do_ip4 && + !add_open("127.0.0.1", cfg->control_port, &l, 1)) { + listening_ports_free(l); + return NULL; + } + } + return l; +} + +/** open accept commpoint */ +static int +accept_open(struct daemon_remote* rc, int fd) +{ + struct listen_list* n = (struct listen_list*)malloc(sizeof(*n)); + if(!n) { + log_err("out of memory"); + return 0; + } + n->next = rc->accept_list; + rc->accept_list = n; + /* open commpt */ + n->com = comm_point_create_raw(rc->worker->base, fd, 0, + &remote_accept_callback, rc); + if(!n->com) + return 0; + /* keep this port open, its fd is kept in the rc portlist */ + n->com->do_not_close = 1; + return 1; +} + +int daemon_remote_open_accept(struct daemon_remote* rc, + struct listen_port* ports, struct worker* worker) +{ + struct listen_port* p; + rc->worker = worker; + for(p = ports; p; p = p->next) { + if(!accept_open(rc, p->fd)) { + log_err("could not create accept comm point"); + return 0; + } + } + return 1; +} + +void daemon_remote_stop_accept(struct daemon_remote* rc) +{ + struct listen_list* p; + for(p=rc->accept_list; p; p=p->next) { + comm_point_stop_listening(p->com); + } +} + +void daemon_remote_start_accept(struct daemon_remote* rc) +{ + struct listen_list* p; + for(p=rc->accept_list; p; p=p->next) { + comm_point_start_listening(p->com, -1, -1); + } +} + +int remote_accept_callback(struct comm_point* c, void* arg, int err, + struct comm_reply* ATTR_UNUSED(rep)) +{ + struct daemon_remote* rc = (struct daemon_remote*)arg; + struct sockaddr_storage addr; + socklen_t addrlen; + int newfd; + struct rc_state* n; + if(err != NETEVENT_NOERROR) { + log_err("error %d on remote_accept_callback", err); + return 0; + } + /* perform the accept */ + newfd = comm_point_perform_accept(c, &addr, &addrlen); + if(newfd == -1) + return 0; + /* create new commpoint unless we are servicing already */ + if(rc->active >= rc->max_active) { + log_warn("drop incoming remote control: too many connections"); + close_exit: +#ifndef USE_WINSOCK + close(newfd); +#else + closesocket(newfd); +#endif + return 0; + } + + /* setup commpoint to service the remote control command */ + n = (struct rc_state*)calloc(1, sizeof(*n)); + if(!n) { + log_err("out of memory"); + goto close_exit; + } + /* start in reading state */ + n->c = comm_point_create_raw(rc->worker->base, newfd, 0, + &remote_control_callback, n); + if(!n->c) { + log_err("out of memory"); + free(n); + goto close_exit; + } + log_addr(VERB_QUERY, "new control connection from", &addr, addrlen); + n->c->do_not_close = 0; + comm_point_stop_listening(n->c); + comm_point_start_listening(n->c, -1, REMOTE_CONTROL_TCP_TIMEOUT); + memcpy(&n->c->repinfo.addr, &addr, addrlen); + n->c->repinfo.addrlen = addrlen; + n->shake_state = rc_hs_read; + n->ssl = SSL_new(rc->ctx); + if(!n->ssl) { + log_crypto_err("could not SSL_new"); + comm_point_delete(n->c); + free(n); + goto close_exit; + } + SSL_set_accept_state(n->ssl); + (void)SSL_set_mode(n->ssl, SSL_MODE_AUTO_RETRY); + if(!SSL_set_fd(n->ssl, newfd)) { + log_crypto_err("could not SSL_set_fd"); + SSL_free(n->ssl); + comm_point_delete(n->c); + free(n); + goto close_exit; + } + + n->rc = rc; + n->next = rc->busy_list; + rc->busy_list = n; + rc->active ++; + + /* perform the first nonblocking read already, for windows, + * so it can return wouldblock. could be faster too. */ + (void)remote_control_callback(n->c, n, NETEVENT_NOERROR, NULL); + return 0; +} + +/** delete from list */ +static void +state_list_remove_elem(struct rc_state** list, struct comm_point* c) +{ + while(*list) { + if( (*list)->c == c) { + *list = (*list)->next; + return; + } + list = &(*list)->next; + } +} + +/** decrease active count and remove commpoint from busy list */ +static void +clean_point(struct daemon_remote* rc, struct rc_state* s) +{ + state_list_remove_elem(&rc->busy_list, s->c); + rc->active --; + if(s->ssl) { + SSL_shutdown(s->ssl); + SSL_free(s->ssl); + } + comm_point_delete(s->c); + free(s); +} + +int +ssl_print_text(SSL* ssl, const char* text) +{ + int r; + if(!ssl) + return 0; + ERR_clear_error(); + if((r=SSL_write(ssl, text, (int)strlen(text))) <= 0) { + if(SSL_get_error(ssl, r) == SSL_ERROR_ZERO_RETURN) { + verbose(VERB_QUERY, "warning, in SSL_write, peer " + "closed connection"); + return 0; + } + log_crypto_err("could not SSL_write"); + return 0; + } + return 1; +} + +/** print text over the ssl connection */ +static int +ssl_print_vmsg(SSL* ssl, const char* format, va_list args) +{ + char msg[1024]; + vsnprintf(msg, sizeof(msg), format, args); + return ssl_print_text(ssl, msg); +} + +/** printf style printing to the ssl connection */ +int ssl_printf(SSL* ssl, const char* format, ...) +{ + va_list args; + int ret; + va_start(args, format); + ret = ssl_print_vmsg(ssl, format, args); + va_end(args); + return ret; +} + +int +ssl_read_line(SSL* ssl, char* buf, size_t max) +{ + int r; + size_t len = 0; + if(!ssl) + return 0; + while(len < max) { + ERR_clear_error(); + if((r=SSL_read(ssl, buf+len, 1)) <= 0) { + if(SSL_get_error(ssl, r) == SSL_ERROR_ZERO_RETURN) { + buf[len] = 0; + return 1; + } + log_crypto_err("could not SSL_read"); + return 0; + } + if(buf[len] == '\n') { + /* return string without \n */ + buf[len] = 0; + return 1; + } + len++; + } + buf[max-1] = 0; + log_err("control line too long (%d): %s", (int)max, buf); + return 0; +} + +/** skip whitespace, return new pointer into string */ +static char* +skipwhite(char* str) +{ + /* EOS \0 is not a space */ + while( isspace(*str) ) + str++; + return str; +} + +/** send the OK to the control client */ +static void send_ok(SSL* ssl) +{ + (void)ssl_printf(ssl, "ok\n"); +} + +/** do the stop command */ +static void +do_stop(SSL* ssl, struct daemon_remote* rc) +{ + rc->worker->need_to_exit = 1; + comm_base_exit(rc->worker->base); + send_ok(ssl); +} + +/** do the reload command */ +static void +do_reload(SSL* ssl, struct daemon_remote* rc) +{ + rc->worker->need_to_exit = 0; + comm_base_exit(rc->worker->base); + send_ok(ssl); +} + +/** do the verbosity command */ +static void +do_verbosity(SSL* ssl, char* str) +{ + int val = atoi(str); + if(val == 0 && strcmp(str, "0") != 0) { + ssl_printf(ssl, "error in verbosity number syntax: %s\n", str); + return; + } + verbosity = val; + send_ok(ssl); +} + +/** print stats from statinfo */ +static int +print_stats(SSL* ssl, const char* nm, struct stats_info* s) +{ + struct timeval avg; + if(!ssl_printf(ssl, "%s.num.queries"SQ"%lu\n", nm, + (unsigned long)s->svr.num_queries)) return 0; + if(!ssl_printf(ssl, "%s.num.cachehits"SQ"%lu\n", nm, + (unsigned long)(s->svr.num_queries + - s->svr.num_queries_missed_cache))) return 0; + if(!ssl_printf(ssl, "%s.num.cachemiss"SQ"%lu\n", nm, + (unsigned long)s->svr.num_queries_missed_cache)) return 0; + if(!ssl_printf(ssl, "%s.num.prefetch"SQ"%lu\n", nm, + (unsigned long)s->svr.num_queries_prefetch)) return 0; + if(!ssl_printf(ssl, "%s.num.recursivereplies"SQ"%lu\n", nm, + (unsigned long)s->mesh_replies_sent)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.avg"SQ"%g\n", nm, + (s->svr.num_queries_missed_cache+s->svr.num_queries_prefetch)? + (double)s->svr.sum_query_list_size/ + (s->svr.num_queries_missed_cache+ + s->svr.num_queries_prefetch) : 0.0)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.max"SQ"%lu\n", nm, + (unsigned long)s->svr.max_query_list_size)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.overwritten"SQ"%lu\n", nm, + (unsigned long)s->mesh_jostled)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.exceeded"SQ"%lu\n", nm, + (unsigned long)s->mesh_dropped)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.current.all"SQ"%lu\n", nm, + (unsigned long)s->mesh_num_states)) return 0; + if(!ssl_printf(ssl, "%s.requestlist.current.user"SQ"%lu\n", nm, + (unsigned long)s->mesh_num_reply_states)) return 0; + timeval_divide(&avg, &s->mesh_replies_sum_wait, s->mesh_replies_sent); + if(!ssl_printf(ssl, "%s.recursion.time.avg"SQ ARG_LL "d.%6.6d\n", nm, + (long long)avg.tv_sec, (int)avg.tv_usec)) return 0; + if(!ssl_printf(ssl, "%s.recursion.time.median"SQ"%g\n", nm, + s->mesh_time_median)) return 0; + return 1; +} + +/** print stats for one thread */ +static int +print_thread_stats(SSL* ssl, int i, struct stats_info* s) +{ + char nm[16]; + snprintf(nm, sizeof(nm), "thread%d", i); + nm[sizeof(nm)-1]=0; + return print_stats(ssl, nm, s); +} + +/** print long number */ +static int +print_longnum(SSL* ssl, const char* desc, size_t x) +{ + if(x > 1024*1024*1024) { + /* more than a Gb */ + size_t front = x / (size_t)1000000; + size_t back = x % (size_t)1000000; + return ssl_printf(ssl, "%s%u%6.6u\n", desc, + (unsigned)front, (unsigned)back); + } else { + return ssl_printf(ssl, "%s%lu\n", desc, (unsigned long)x); + } +} + +/** print mem stats */ +static int +print_mem(SSL* ssl, struct worker* worker, struct daemon* daemon) +{ + int m; + size_t msg, rrset, val, iter; +#ifdef HAVE_SBRK + extern void* unbound_start_brk; + void* cur = sbrk(0); + if(!print_longnum(ssl, "mem.total.sbrk"SQ, + (size_t)((char*)cur - (char*)unbound_start_brk))) return 0; +#endif /* HAVE_SBRK */ + msg = slabhash_get_mem(daemon->env->msg_cache); + rrset = slabhash_get_mem(&daemon->env->rrset_cache->table); + val=0; + iter=0; + m = modstack_find(&worker->env.mesh->mods, "validator"); + if(m != -1) { + fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> + mods.mod[m]->get_mem)); + val = (*worker->env.mesh->mods.mod[m]->get_mem) + (&worker->env, m); + } + m = modstack_find(&worker->env.mesh->mods, "iterator"); + if(m != -1) { + fptr_ok(fptr_whitelist_mod_get_mem(worker->env.mesh-> + mods.mod[m]->get_mem)); + iter = (*worker->env.mesh->mods.mod[m]->get_mem) + (&worker->env, m); + } + + if(!print_longnum(ssl, "mem.cache.rrset"SQ, rrset)) + return 0; + if(!print_longnum(ssl, "mem.cache.message"SQ, msg)) + return 0; + if(!print_longnum(ssl, "mem.mod.iterator"SQ, iter)) + return 0; + if(!print_longnum(ssl, "mem.mod.validator"SQ, val)) + return 0; + return 1; +} + +/** print uptime stats */ +static int +print_uptime(SSL* ssl, struct worker* worker, int reset) +{ + struct timeval now = *worker->env.now_tv; + struct timeval up, dt; + timeval_subtract(&up, &now, &worker->daemon->time_boot); + timeval_subtract(&dt, &now, &worker->daemon->time_last_stat); + if(reset) + worker->daemon->time_last_stat = now; + if(!ssl_printf(ssl, "time.now"SQ ARG_LL "d.%6.6d\n", + (long long)now.tv_sec, (unsigned)now.tv_usec)) return 0; + if(!ssl_printf(ssl, "time.up"SQ ARG_LL "d.%6.6d\n", + (long long)up.tv_sec, (unsigned)up.tv_usec)) return 0; + if(!ssl_printf(ssl, "time.elapsed"SQ ARG_LL "d.%6.6d\n", + (long long)dt.tv_sec, (unsigned)dt.tv_usec)) return 0; + return 1; +} + +/** print extended histogram */ +static int +print_hist(SSL* ssl, struct stats_info* s) +{ + struct timehist* hist; + size_t i; + hist = timehist_setup(); + if(!hist) { + log_err("out of memory"); + return 0; + } + timehist_import(hist, s->svr.hist, NUM_BUCKETS_HIST); + for(i=0; i<hist->num; i++) { + if(!ssl_printf(ssl, + "histogram.%6.6d.%6.6d.to.%6.6d.%6.6d=%lu\n", + (int)hist->buckets[i].lower.tv_sec, + (int)hist->buckets[i].lower.tv_usec, + (int)hist->buckets[i].upper.tv_sec, + (int)hist->buckets[i].upper.tv_usec, + (unsigned long)hist->buckets[i].count)) { + timehist_delete(hist); + return 0; + } + } + timehist_delete(hist); + return 1; +} + +/** print extended stats */ +static int +print_ext(SSL* ssl, struct stats_info* s) +{ + int i; + char nm[16]; + const sldns_rr_descriptor* desc; + const sldns_lookup_table* lt; + /* TYPE */ + for(i=0; i<STATS_QTYPE_NUM; i++) { + if(inhibit_zero && s->svr.qtype[i] == 0) + continue; + desc = sldns_rr_descript((uint16_t)i); + if(desc && desc->_name) { + snprintf(nm, sizeof(nm), "%s", desc->_name); + } else if (i == LDNS_RR_TYPE_IXFR) { + snprintf(nm, sizeof(nm), "IXFR"); + } else if (i == LDNS_RR_TYPE_AXFR) { + snprintf(nm, sizeof(nm), "AXFR"); + } else if (i == LDNS_RR_TYPE_MAILA) { + snprintf(nm, sizeof(nm), "MAILA"); + } else if (i == LDNS_RR_TYPE_MAILB) { + snprintf(nm, sizeof(nm), "MAILB"); + } else if (i == LDNS_RR_TYPE_ANY) { + snprintf(nm, sizeof(nm), "ANY"); + } else { + snprintf(nm, sizeof(nm), "TYPE%d", i); + } + if(!ssl_printf(ssl, "num.query.type.%s"SQ"%lu\n", + nm, (unsigned long)s->svr.qtype[i])) return 0; + } + if(!inhibit_zero || s->svr.qtype_big) { + if(!ssl_printf(ssl, "num.query.type.other"SQ"%lu\n", + (unsigned long)s->svr.qtype_big)) return 0; + } + /* CLASS */ + for(i=0; i<STATS_QCLASS_NUM; i++) { + if(inhibit_zero && s->svr.qclass[i] == 0) + continue; + lt = sldns_lookup_by_id(sldns_rr_classes, i); + if(lt && lt->name) { + snprintf(nm, sizeof(nm), "%s", lt->name); + } else { + snprintf(nm, sizeof(nm), "CLASS%d", i); + } + if(!ssl_printf(ssl, "num.query.class.%s"SQ"%lu\n", + nm, (unsigned long)s->svr.qclass[i])) return 0; + } + if(!inhibit_zero || s->svr.qclass_big) { + if(!ssl_printf(ssl, "num.query.class.other"SQ"%lu\n", + (unsigned long)s->svr.qclass_big)) return 0; + } + /* OPCODE */ + for(i=0; i<STATS_OPCODE_NUM; i++) { + if(inhibit_zero && s->svr.qopcode[i] == 0) + continue; + lt = sldns_lookup_by_id(sldns_opcodes, i); + if(lt && lt->name) { + snprintf(nm, sizeof(nm), "%s", lt->name); + } else { + snprintf(nm, sizeof(nm), "OPCODE%d", i); + } + if(!ssl_printf(ssl, "num.query.opcode.%s"SQ"%lu\n", + nm, (unsigned long)s->svr.qopcode[i])) return 0; + } + /* transport */ + if(!ssl_printf(ssl, "num.query.tcp"SQ"%lu\n", + (unsigned long)s->svr.qtcp)) return 0; + if(!ssl_printf(ssl, "num.query.tcpout"SQ"%lu\n", + (unsigned long)s->svr.qtcp_outgoing)) return 0; + if(!ssl_printf(ssl, "num.query.ipv6"SQ"%lu\n", + (unsigned long)s->svr.qipv6)) return 0; + /* flags */ + if(!ssl_printf(ssl, "num.query.flags.QR"SQ"%lu\n", + (unsigned long)s->svr.qbit_QR)) return 0; + if(!ssl_printf(ssl, "num.query.flags.AA"SQ"%lu\n", + (unsigned long)s->svr.qbit_AA)) return 0; + if(!ssl_printf(ssl, "num.query.flags.TC"SQ"%lu\n", + (unsigned long)s->svr.qbit_TC)) return 0; + if(!ssl_printf(ssl, "num.query.flags.RD"SQ"%lu\n", + (unsigned long)s->svr.qbit_RD)) return 0; + if(!ssl_printf(ssl, "num.query.flags.RA"SQ"%lu\n", + (unsigned long)s->svr.qbit_RA)) return 0; + if(!ssl_printf(ssl, "num.query.flags.Z"SQ"%lu\n", + (unsigned long)s->svr.qbit_Z)) return 0; + if(!ssl_printf(ssl, "num.query.flags.AD"SQ"%lu\n", + (unsigned long)s->svr.qbit_AD)) return 0; + if(!ssl_printf(ssl, "num.query.flags.CD"SQ"%lu\n", + (unsigned long)s->svr.qbit_CD)) return 0; + if(!ssl_printf(ssl, "num.query.edns.present"SQ"%lu\n", + (unsigned long)s->svr.qEDNS)) return 0; + if(!ssl_printf(ssl, "num.query.edns.DO"SQ"%lu\n", + (unsigned long)s->svr.qEDNS_DO)) return 0; + + /* RCODE */ + for(i=0; i<STATS_RCODE_NUM; i++) { + if(inhibit_zero && s->svr.ans_rcode[i] == 0) + continue; + lt = sldns_lookup_by_id(sldns_rcodes, i); + if(lt && lt->name) { + snprintf(nm, sizeof(nm), "%s", lt->name); + } else { + snprintf(nm, sizeof(nm), "RCODE%d", i); + } + if(!ssl_printf(ssl, "num.answer.rcode.%s"SQ"%lu\n", + nm, (unsigned long)s->svr.ans_rcode[i])) return 0; + } + if(!inhibit_zero || s->svr.ans_rcode_nodata) { + if(!ssl_printf(ssl, "num.answer.rcode.nodata"SQ"%lu\n", + (unsigned long)s->svr.ans_rcode_nodata)) return 0; + } + /* validation */ + if(!ssl_printf(ssl, "num.answer.secure"SQ"%lu\n", + (unsigned long)s->svr.ans_secure)) return 0; + if(!ssl_printf(ssl, "num.answer.bogus"SQ"%lu\n", + (unsigned long)s->svr.ans_bogus)) return 0; + if(!ssl_printf(ssl, "num.rrset.bogus"SQ"%lu\n", + (unsigned long)s->svr.rrset_bogus)) return 0; + /* threat detection */ + if(!ssl_printf(ssl, "unwanted.queries"SQ"%lu\n", + (unsigned long)s->svr.unwanted_queries)) return 0; + if(!ssl_printf(ssl, "unwanted.replies"SQ"%lu\n", + (unsigned long)s->svr.unwanted_replies)) return 0; + /* cache counts */ + if(!ssl_printf(ssl, "msg.cache.count"SQ"%u\n", + (unsigned)s->svr.msg_cache_count)) return 0; + if(!ssl_printf(ssl, "rrset.cache.count"SQ"%u\n", + (unsigned)s->svr.rrset_cache_count)) return 0; + if(!ssl_printf(ssl, "infra.cache.count"SQ"%u\n", + (unsigned)s->svr.infra_cache_count)) return 0; + if(!ssl_printf(ssl, "key.cache.count"SQ"%u\n", + (unsigned)s->svr.key_cache_count)) return 0; + return 1; +} + +/** do the stats command */ +static void +do_stats(SSL* ssl, struct daemon_remote* rc, int reset) +{ + struct daemon* daemon = rc->worker->daemon; + struct stats_info total; + struct stats_info s; + int i; + log_assert(daemon->num > 0); + /* gather all thread statistics in one place */ + for(i=0; i<daemon->num; i++) { + server_stats_obtain(rc->worker, daemon->workers[i], &s, reset); + if(!print_thread_stats(ssl, i, &s)) + return; + if(i == 0) + total = s; + else server_stats_add(&total, &s); + } + /* print the thread statistics */ + total.mesh_time_median /= (double)daemon->num; + if(!print_stats(ssl, "total", &total)) + return; + if(!print_uptime(ssl, rc->worker, reset)) + return; + if(daemon->cfg->stat_extended) { + if(!print_mem(ssl, rc->worker, daemon)) + return; + if(!print_hist(ssl, &total)) + return; + if(!print_ext(ssl, &total)) + return; + } +} + +/** parse commandline argument domain name */ +static int +parse_arg_name(SSL* ssl, char* str, uint8_t** res, size_t* len, int* labs) +{ + uint8_t nm[LDNS_MAX_DOMAINLEN+1]; + size_t nmlen = sizeof(nm); + int status; + *res = NULL; + *len = 0; + *labs = 0; + status = sldns_str2wire_dname_buf(str, nm, &nmlen); + if(status != 0) { + ssl_printf(ssl, "error cannot parse name %s at %d: %s\n", str, + LDNS_WIREPARSE_OFFSET(status), + sldns_get_errorstr_parse(status)); + return 0; + } + *res = memdup(nm, nmlen); + if(!*res) { + ssl_printf(ssl, "error out of memory\n"); + return 0; + } + *labs = dname_count_size_labels(*res, len); + return 1; +} + +/** find second argument, modifies string */ +static int +find_arg2(SSL* ssl, char* arg, char** arg2) +{ + char* as = strchr(arg, ' '); + char* at = strchr(arg, '\t'); + if(as && at) { + if(at < as) + as = at; + as[0]=0; + *arg2 = skipwhite(as+1); + } else if(as) { + as[0]=0; + *arg2 = skipwhite(as+1); + } else if(at) { + at[0]=0; + *arg2 = skipwhite(at+1); + } else { + ssl_printf(ssl, "error could not find next argument " + "after %s\n", arg); + return 0; + } + return 1; +} + +/** Add a new zone */ +static void +do_zone_add(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + char* arg2; + enum localzone_type t; + struct local_zone* z; + if(!find_arg2(ssl, arg, &arg2)) + return; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + if(!local_zone_str2type(arg2, &t)) { + ssl_printf(ssl, "error not a zone type. %s\n", arg2); + free(nm); + return; + } + lock_rw_wrlock(&worker->daemon->local_zones->lock); + if((z=local_zones_find(worker->daemon->local_zones, nm, nmlen, + nmlabs, LDNS_RR_CLASS_IN))) { + /* already present in tree */ + lock_rw_wrlock(&z->lock); + z->type = t; /* update type anyway */ + lock_rw_unlock(&z->lock); + free(nm); + lock_rw_unlock(&worker->daemon->local_zones->lock); + send_ok(ssl); + return; + } + if(!local_zones_add_zone(worker->daemon->local_zones, nm, nmlen, + nmlabs, LDNS_RR_CLASS_IN, t)) { + lock_rw_unlock(&worker->daemon->local_zones->lock); + ssl_printf(ssl, "error out of memory\n"); + return; + } + lock_rw_unlock(&worker->daemon->local_zones->lock); + send_ok(ssl); +} + +/** Remove a zone */ +static void +do_zone_remove(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + struct local_zone* z; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + lock_rw_wrlock(&worker->daemon->local_zones->lock); + if((z=local_zones_find(worker->daemon->local_zones, nm, nmlen, + nmlabs, LDNS_RR_CLASS_IN))) { + /* present in tree */ + local_zones_del_zone(worker->daemon->local_zones, z); + } + lock_rw_unlock(&worker->daemon->local_zones->lock); + free(nm); + send_ok(ssl); +} + +/** Add new RR data */ +static void +do_data_add(SSL* ssl, struct worker* worker, char* arg) +{ + if(!local_zones_add_RR(worker->daemon->local_zones, arg)) { + ssl_printf(ssl,"error in syntax or out of memory, %s\n", arg); + return; + } + send_ok(ssl); +} + +/** Remove RR data */ +static void +do_data_remove(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + local_zones_del_data(worker->daemon->local_zones, nm, + nmlen, nmlabs, LDNS_RR_CLASS_IN); + free(nm); + send_ok(ssl); +} + +/** cache lookup of nameservers */ +static void +do_lookup(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + (void)print_deleg_lookup(ssl, worker, nm, nmlen, nmlabs); + free(nm); +} + +/** flush something from rrset and msg caches */ +static void +do_cache_remove(struct worker* worker, uint8_t* nm, size_t nmlen, + uint16_t t, uint16_t c) +{ + hashvalue_t h; + struct query_info k; + rrset_cache_remove(worker->env.rrset_cache, nm, nmlen, t, c, 0); + if(t == LDNS_RR_TYPE_SOA) + rrset_cache_remove(worker->env.rrset_cache, nm, nmlen, t, c, + PACKED_RRSET_SOA_NEG); + k.qname = nm; + k.qname_len = nmlen; + k.qtype = t; + k.qclass = c; + h = query_info_hash(&k); + slabhash_remove(worker->env.msg_cache, h, &k); +} + +/** flush a type */ +static void +do_flush_type(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + char* arg2; + uint16_t t; + if(!find_arg2(ssl, arg, &arg2)) + return; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + t = sldns_get_rr_type_by_name(arg2); + do_cache_remove(worker, nm, nmlen, t, LDNS_RR_CLASS_IN); + + free(nm); + send_ok(ssl); +} + +/** flush statistics */ +static void +do_flush_stats(SSL* ssl, struct worker* worker) +{ + worker_stats_clear(worker); + send_ok(ssl); +} + +/** + * Local info for deletion functions + */ +struct del_info { + /** worker */ + struct worker* worker; + /** name to delete */ + uint8_t* name; + /** length */ + size_t len; + /** labels */ + int labs; + /** now */ + time_t now; + /** time to invalidate to */ + time_t expired; + /** number of rrsets removed */ + size_t num_rrsets; + /** number of msgs removed */ + size_t num_msgs; + /** number of key entries removed */ + size_t num_keys; + /** length of addr */ + socklen_t addrlen; + /** socket address for host deletion */ + struct sockaddr_storage addr; +}; + +/** callback to delete hosts in infra cache */ +static void +infra_del_host(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct infra_key* k = (struct infra_key*)e->key; + if(sockaddr_cmp(&inf->addr, inf->addrlen, &k->addr, k->addrlen) == 0) { + struct infra_data* d = (struct infra_data*)e->data; + d->probedelay = 0; + d->timeout_A = 0; + d->timeout_AAAA = 0; + d->timeout_other = 0; + rtt_init(&d->rtt); + if(d->ttl >= inf->now) { + d->ttl = inf->expired; + inf->num_keys++; + } + } +} + +/** flush infra cache */ +static void +do_flush_infra(SSL* ssl, struct worker* worker, char* arg) +{ + struct sockaddr_storage addr; + socklen_t len; + struct del_info inf; + if(strcmp(arg, "all") == 0) { + slabhash_clear(worker->env.infra_cache->hosts); + send_ok(ssl); + return; + } + if(!ipstrtoaddr(arg, UNBOUND_DNS_PORT, &addr, &len)) { + (void)ssl_printf(ssl, "error parsing ip addr: '%s'\n", arg); + return; + } + /* delete all entries from cache */ + /* what we do is to set them all expired */ + inf.worker = worker; + inf.name = 0; + inf.len = 0; + inf.labs = 0; + inf.now = *worker->env.now; + inf.expired = *worker->env.now; + inf.expired -= 3; /* handle 3 seconds skew between threads */ + inf.num_rrsets = 0; + inf.num_msgs = 0; + inf.num_keys = 0; + inf.addrlen = len; + memmove(&inf.addr, &addr, len); + slabhash_traverse(worker->env.infra_cache->hosts, 1, &infra_del_host, + &inf); + send_ok(ssl); +} + +/** flush requestlist */ +static void +do_flush_requestlist(SSL* ssl, struct worker* worker) +{ + mesh_delete_all(worker->env.mesh); + send_ok(ssl); +} + +/** callback to delete rrsets in a zone */ +static void +zone_del_rrset(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct ub_packed_rrset_key* k = (struct ub_packed_rrset_key*)e->key; + if(dname_subdomain_c(k->rk.dname, inf->name)) { + struct packed_rrset_data* d = + (struct packed_rrset_data*)e->data; + if(d->ttl >= inf->now) { + d->ttl = inf->expired; + inf->num_rrsets++; + } + } +} + +/** callback to delete messages in a zone */ +static void +zone_del_msg(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct msgreply_entry* k = (struct msgreply_entry*)e->key; + if(dname_subdomain_c(k->key.qname, inf->name)) { + struct reply_info* d = (struct reply_info*)e->data; + if(d->ttl >= inf->now) { + d->ttl = inf->expired; + inf->num_msgs++; + } + } +} + +/** callback to delete keys in zone */ +static void +zone_del_kcache(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct key_entry_key* k = (struct key_entry_key*)e->key; + if(dname_subdomain_c(k->name, inf->name)) { + struct key_entry_data* d = (struct key_entry_data*)e->data; + if(d->ttl >= inf->now) { + d->ttl = inf->expired; + inf->num_keys++; + } + } +} + +/** remove all rrsets and keys from zone from cache */ +static void +do_flush_zone(SSL* ssl, struct worker* worker, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + struct del_info inf; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + /* delete all RRs and key entries from zone */ + /* what we do is to set them all expired */ + inf.worker = worker; + inf.name = nm; + inf.len = nmlen; + inf.labs = nmlabs; + inf.now = *worker->env.now; + inf.expired = *worker->env.now; + inf.expired -= 3; /* handle 3 seconds skew between threads */ + inf.num_rrsets = 0; + inf.num_msgs = 0; + inf.num_keys = 0; + slabhash_traverse(&worker->env.rrset_cache->table, 1, + &zone_del_rrset, &inf); + + slabhash_traverse(worker->env.msg_cache, 1, &zone_del_msg, &inf); + + /* and validator cache */ + if(worker->env.key_cache) { + slabhash_traverse(worker->env.key_cache->slab, 1, + &zone_del_kcache, &inf); + } + + free(nm); + + (void)ssl_printf(ssl, "ok removed %lu rrsets, %lu messages " + "and %lu key entries\n", (unsigned long)inf.num_rrsets, + (unsigned long)inf.num_msgs, (unsigned long)inf.num_keys); +} + +/** callback to delete bogus rrsets */ +static void +bogus_del_rrset(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct packed_rrset_data* d = (struct packed_rrset_data*)e->data; + if(d->security == sec_status_bogus) { + d->ttl = inf->expired; + inf->num_rrsets++; + } +} + +/** callback to delete bogus messages */ +static void +bogus_del_msg(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct reply_info* d = (struct reply_info*)e->data; + if(d->security == sec_status_bogus) { + d->ttl = inf->expired; + inf->num_msgs++; + } +} + +/** callback to delete bogus keys */ +static void +bogus_del_kcache(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct key_entry_data* d = (struct key_entry_data*)e->data; + if(d->isbad) { + d->ttl = inf->expired; + inf->num_keys++; + } +} + +/** remove all bogus rrsets, msgs and keys from cache */ +static void +do_flush_bogus(SSL* ssl, struct worker* worker) +{ + struct del_info inf; + /* what we do is to set them all expired */ + inf.worker = worker; + inf.now = *worker->env.now; + inf.expired = *worker->env.now; + inf.expired -= 3; /* handle 3 seconds skew between threads */ + inf.num_rrsets = 0; + inf.num_msgs = 0; + inf.num_keys = 0; + slabhash_traverse(&worker->env.rrset_cache->table, 1, + &bogus_del_rrset, &inf); + + slabhash_traverse(worker->env.msg_cache, 1, &bogus_del_msg, &inf); + + /* and validator cache */ + if(worker->env.key_cache) { + slabhash_traverse(worker->env.key_cache->slab, 1, + &bogus_del_kcache, &inf); + } + + (void)ssl_printf(ssl, "ok removed %lu rrsets, %lu messages " + "and %lu key entries\n", (unsigned long)inf.num_rrsets, + (unsigned long)inf.num_msgs, (unsigned long)inf.num_keys); +} + +/** callback to delete negative and servfail rrsets */ +static void +negative_del_rrset(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct ub_packed_rrset_key* k = (struct ub_packed_rrset_key*)e->key; + struct packed_rrset_data* d = (struct packed_rrset_data*)e->data; + /* delete the parentside negative cache rrsets, + * these are namerserver rrsets that failed lookup, rdata empty */ + if((k->rk.flags & PACKED_RRSET_PARENT_SIDE) && d->count == 1 && + d->rrsig_count == 0 && d->rr_len[0] == 0) { + d->ttl = inf->expired; + inf->num_rrsets++; + } +} + +/** callback to delete negative and servfail messages */ +static void +negative_del_msg(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct reply_info* d = (struct reply_info*)e->data; + /* rcode not NOERROR: NXDOMAIN, SERVFAIL, ..: an nxdomain or error + * or NOERROR rcode with ANCOUNT==0: a NODATA answer */ + if(FLAGS_GET_RCODE(d->flags) != 0 || d->an_numrrsets == 0) { + d->ttl = inf->expired; + inf->num_msgs++; + } +} + +/** callback to delete negative key entries */ +static void +negative_del_kcache(struct lruhash_entry* e, void* arg) +{ + /* entry is locked */ + struct del_info* inf = (struct del_info*)arg; + struct key_entry_data* d = (struct key_entry_data*)e->data; + /* could be bad because of lookup failure on the DS, DNSKEY, which + * was nxdomain or servfail, and thus a result of negative lookups */ + if(d->isbad) { + d->ttl = inf->expired; + inf->num_keys++; + } +} + +/** remove all negative(NODATA,NXDOMAIN), and servfail messages from cache */ +static void +do_flush_negative(SSL* ssl, struct worker* worker) +{ + struct del_info inf; + /* what we do is to set them all expired */ + inf.worker = worker; + inf.now = *worker->env.now; + inf.expired = *worker->env.now; + inf.expired -= 3; /* handle 3 seconds skew between threads */ + inf.num_rrsets = 0; + inf.num_msgs = 0; + inf.num_keys = 0; + slabhash_traverse(&worker->env.rrset_cache->table, 1, + &negative_del_rrset, &inf); + + slabhash_traverse(worker->env.msg_cache, 1, &negative_del_msg, &inf); + + /* and validator cache */ + if(worker->env.key_cache) { + slabhash_traverse(worker->env.key_cache->slab, 1, + &negative_del_kcache, &inf); + } + + (void)ssl_printf(ssl, "ok removed %lu rrsets, %lu messages " + "and %lu key entries\n", (unsigned long)inf.num_rrsets, + (unsigned long)inf.num_msgs, (unsigned long)inf.num_keys); +} + +/** remove name rrset from cache */ +static void +do_flush_name(SSL* ssl, struct worker* w, char* arg) +{ + uint8_t* nm; + int nmlabs; + size_t nmlen; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_A, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_AAAA, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_NS, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_SOA, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_CNAME, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_DNAME, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_MX, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_PTR, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_SRV, LDNS_RR_CLASS_IN); + do_cache_remove(w, nm, nmlen, LDNS_RR_TYPE_NAPTR, LDNS_RR_CLASS_IN); + + free(nm); + send_ok(ssl); +} + +/** printout a delegation point info */ +static int +ssl_print_name_dp(SSL* ssl, const char* str, uint8_t* nm, uint16_t dclass, + struct delegpt* dp) +{ + char buf[257]; + struct delegpt_ns* ns; + struct delegpt_addr* a; + int f = 0; + if(str) { /* print header for forward, stub */ + char* c = sldns_wire2str_class(dclass); + dname_str(nm, buf); + if(!ssl_printf(ssl, "%s %s %s ", buf, (c?c:"CLASS??"), str)) { + free(c); + return 0; + } + free(c); + } + for(ns = dp->nslist; ns; ns = ns->next) { + dname_str(ns->name, buf); + if(!ssl_printf(ssl, "%s%s", (f?" ":""), buf)) + return 0; + f = 1; + } + for(a = dp->target_list; a; a = a->next_target) { + addr_to_str(&a->addr, a->addrlen, buf, sizeof(buf)); + if(!ssl_printf(ssl, "%s%s", (f?" ":""), buf)) + return 0; + f = 1; + } + return ssl_printf(ssl, "\n"); +} + + +/** print root forwards */ +static int +print_root_fwds(SSL* ssl, struct iter_forwards* fwds, uint8_t* root) +{ + struct delegpt* dp; + dp = forwards_lookup(fwds, root, LDNS_RR_CLASS_IN); + if(!dp) + return ssl_printf(ssl, "off (using root hints)\n"); + /* if dp is returned it must be the root */ + log_assert(query_dname_compare(dp->name, root)==0); + return ssl_print_name_dp(ssl, NULL, root, LDNS_RR_CLASS_IN, dp); +} + +/** parse args into delegpt */ +static struct delegpt* +parse_delegpt(SSL* ssl, char* args, uint8_t* nm, int allow_names) +{ + /* parse args and add in */ + char* p = args; + char* todo; + struct delegpt* dp = delegpt_create_mlc(nm); + struct sockaddr_storage addr; + socklen_t addrlen; + if(!dp) { + (void)ssl_printf(ssl, "error out of memory\n"); + return NULL; + } + while(p) { + todo = p; + p = strchr(p, ' '); /* find next spot, if any */ + if(p) { + *p++ = 0; /* end this spot */ + p = skipwhite(p); /* position at next spot */ + } + /* parse address */ + if(!extstrtoaddr(todo, &addr, &addrlen)) { + if(allow_names) { + uint8_t* n = NULL; + size_t ln; + int lb; + if(!parse_arg_name(ssl, todo, &n, &ln, &lb)) { + (void)ssl_printf(ssl, "error cannot " + "parse IP address or name " + "'%s'\n", todo); + delegpt_free_mlc(dp); + return NULL; + } + if(!delegpt_add_ns_mlc(dp, n, 0)) { + (void)ssl_printf(ssl, "error out of memory\n"); + free(n); + delegpt_free_mlc(dp); + return NULL; + } + free(n); + + } else { + (void)ssl_printf(ssl, "error cannot parse" + " IP address '%s'\n", todo); + delegpt_free_mlc(dp); + return NULL; + } + } else { + /* add address */ + if(!delegpt_add_addr_mlc(dp, &addr, addrlen, 0, 0)) { + (void)ssl_printf(ssl, "error out of memory\n"); + delegpt_free_mlc(dp); + return NULL; + } + } + } + return dp; +} + +/** do the status command */ +static void +do_forward(SSL* ssl, struct worker* worker, char* args) +{ + struct iter_forwards* fwd = worker->env.fwds; + uint8_t* root = (uint8_t*)"\000"; + if(!fwd) { + (void)ssl_printf(ssl, "error: structure not allocated\n"); + return; + } + if(args == NULL || args[0] == 0) { + (void)print_root_fwds(ssl, fwd, root); + return; + } + /* set root forwards for this thread. since we are in remote control + * the actual mesh is not running, so we can freely edit it. */ + /* delete all the existing queries first */ + mesh_delete_all(worker->env.mesh); + if(strcmp(args, "off") == 0) { + forwards_delete_zone(fwd, LDNS_RR_CLASS_IN, root); + } else { + struct delegpt* dp; + if(!(dp = parse_delegpt(ssl, args, root, 0))) + return; + if(!forwards_add_zone(fwd, LDNS_RR_CLASS_IN, dp)) { + (void)ssl_printf(ssl, "error out of memory\n"); + return; + } + } + send_ok(ssl); +} + +static int +parse_fs_args(SSL* ssl, char* args, uint8_t** nm, struct delegpt** dp, + int* insecure, int* prime) +{ + char* zonename; + char* rest; + size_t nmlen; + int nmlabs; + /* parse all -x args */ + while(args[0] == '+') { + if(!find_arg2(ssl, args, &rest)) + return 0; + while(*(++args) != 0) { + if(*args == 'i' && insecure) + *insecure = 1; + else if(*args == 'p' && prime) + *prime = 1; + else { + (void)ssl_printf(ssl, "error: unknown option %s\n", args); + return 0; + } + } + args = rest; + } + /* parse name */ + if(dp) { + if(!find_arg2(ssl, args, &rest)) + return 0; + zonename = args; + args = rest; + } else zonename = args; + if(!parse_arg_name(ssl, zonename, nm, &nmlen, &nmlabs)) + return 0; + + /* parse dp */ + if(dp) { + if(!(*dp = parse_delegpt(ssl, args, *nm, 1))) { + free(*nm); + return 0; + } + } + return 1; +} + +/** do the forward_add command */ +static void +do_forward_add(SSL* ssl, struct worker* worker, char* args) +{ + struct iter_forwards* fwd = worker->env.fwds; + int insecure = 0; + uint8_t* nm = NULL; + struct delegpt* dp = NULL; + if(!parse_fs_args(ssl, args, &nm, &dp, &insecure, NULL)) + return; + if(insecure && worker->env.anchors) { + if(!anchors_add_insecure(worker->env.anchors, LDNS_RR_CLASS_IN, + nm)) { + (void)ssl_printf(ssl, "error out of memory\n"); + delegpt_free_mlc(dp); + free(nm); + return; + } + } + if(!forwards_add_zone(fwd, LDNS_RR_CLASS_IN, dp)) { + (void)ssl_printf(ssl, "error out of memory\n"); + free(nm); + return; + } + free(nm); + send_ok(ssl); +} + +/** do the forward_remove command */ +static void +do_forward_remove(SSL* ssl, struct worker* worker, char* args) +{ + struct iter_forwards* fwd = worker->env.fwds; + int insecure = 0; + uint8_t* nm = NULL; + if(!parse_fs_args(ssl, args, &nm, NULL, &insecure, NULL)) + return; + if(insecure && worker->env.anchors) + anchors_delete_insecure(worker->env.anchors, LDNS_RR_CLASS_IN, + nm); + forwards_delete_zone(fwd, LDNS_RR_CLASS_IN, nm); + free(nm); + send_ok(ssl); +} + +/** do the stub_add command */ +static void +do_stub_add(SSL* ssl, struct worker* worker, char* args) +{ + struct iter_forwards* fwd = worker->env.fwds; + int insecure = 0, prime = 0; + uint8_t* nm = NULL; + struct delegpt* dp = NULL; + if(!parse_fs_args(ssl, args, &nm, &dp, &insecure, &prime)) + return; + if(insecure && worker->env.anchors) { + if(!anchors_add_insecure(worker->env.anchors, LDNS_RR_CLASS_IN, + nm)) { + (void)ssl_printf(ssl, "error out of memory\n"); + delegpt_free_mlc(dp); + free(nm); + return; + } + } + if(!forwards_add_stub_hole(fwd, LDNS_RR_CLASS_IN, nm)) { + if(insecure && worker->env.anchors) + anchors_delete_insecure(worker->env.anchors, + LDNS_RR_CLASS_IN, nm); + (void)ssl_printf(ssl, "error out of memory\n"); + delegpt_free_mlc(dp); + free(nm); + return; + } + if(!hints_add_stub(worker->env.hints, LDNS_RR_CLASS_IN, dp, !prime)) { + (void)ssl_printf(ssl, "error out of memory\n"); + forwards_delete_stub_hole(fwd, LDNS_RR_CLASS_IN, nm); + if(insecure && worker->env.anchors) + anchors_delete_insecure(worker->env.anchors, + LDNS_RR_CLASS_IN, nm); + free(nm); + return; + } + free(nm); + send_ok(ssl); +} + +/** do the stub_remove command */ +static void +do_stub_remove(SSL* ssl, struct worker* worker, char* args) +{ + struct iter_forwards* fwd = worker->env.fwds; + int insecure = 0; + uint8_t* nm = NULL; + if(!parse_fs_args(ssl, args, &nm, NULL, &insecure, NULL)) + return; + if(insecure && worker->env.anchors) + anchors_delete_insecure(worker->env.anchors, LDNS_RR_CLASS_IN, + nm); + forwards_delete_stub_hole(fwd, LDNS_RR_CLASS_IN, nm); + hints_delete_stub(worker->env.hints, LDNS_RR_CLASS_IN, nm); + free(nm); + send_ok(ssl); +} + +/** do the insecure_add command */ +static void +do_insecure_add(SSL* ssl, struct worker* worker, char* arg) +{ + size_t nmlen; + int nmlabs; + uint8_t* nm = NULL; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + if(worker->env.anchors) { + if(!anchors_add_insecure(worker->env.anchors, + LDNS_RR_CLASS_IN, nm)) { + (void)ssl_printf(ssl, "error out of memory\n"); + free(nm); + return; + } + } + free(nm); + send_ok(ssl); +} + +/** do the insecure_remove command */ +static void +do_insecure_remove(SSL* ssl, struct worker* worker, char* arg) +{ + size_t nmlen; + int nmlabs; + uint8_t* nm = NULL; + if(!parse_arg_name(ssl, arg, &nm, &nmlen, &nmlabs)) + return; + if(worker->env.anchors) + anchors_delete_insecure(worker->env.anchors, + LDNS_RR_CLASS_IN, nm); + free(nm); + send_ok(ssl); +} + +/** do the status command */ +static void +do_status(SSL* ssl, struct worker* worker) +{ + int i; + time_t uptime; + if(!ssl_printf(ssl, "version: %s\n", PACKAGE_VERSION)) + return; + if(!ssl_printf(ssl, "verbosity: %d\n", verbosity)) + return; + if(!ssl_printf(ssl, "threads: %d\n", worker->daemon->num)) + return; + if(!ssl_printf(ssl, "modules: %d [", worker->daemon->mods.num)) + return; + for(i=0; i<worker->daemon->mods.num; i++) { + if(!ssl_printf(ssl, " %s", worker->daemon->mods.mod[i]->name)) + return; + } + if(!ssl_printf(ssl, " ]\n")) + return; + uptime = (time_t)time(NULL) - (time_t)worker->daemon->time_boot.tv_sec; + if(!ssl_printf(ssl, "uptime: " ARG_LL "d seconds\n", (long long)uptime)) + return; + if(!ssl_printf(ssl, "options:%s%s\n" , + (worker->daemon->reuseport?" reuseport":""), + (worker->daemon->rc->accept_list?" control(ssl)":""))) + return; + if(!ssl_printf(ssl, "unbound (pid %d) is running...\n", + (int)getpid())) + return; +} + +/** get age for the mesh state */ +static void +get_mesh_age(struct mesh_state* m, char* buf, size_t len, + struct module_env* env) +{ + if(m->reply_list) { + struct timeval d; + struct mesh_reply* r = m->reply_list; + /* last reply is the oldest */ + while(r && r->next) + r = r->next; + timeval_subtract(&d, env->now_tv, &r->start_time); + snprintf(buf, len, ARG_LL "d.%6.6d", + (long long)d.tv_sec, (int)d.tv_usec); + } else { + snprintf(buf, len, "-"); + } +} + +/** get status of a mesh state */ +static void +get_mesh_status(struct mesh_area* mesh, struct mesh_state* m, + char* buf, size_t len) +{ + enum module_ext_state s = m->s.ext_state[m->s.curmod]; + const char *modname = mesh->mods.mod[m->s.curmod]->name; + size_t l; + if(strcmp(modname, "iterator") == 0 && s == module_wait_reply && + m->s.minfo[m->s.curmod]) { + /* break into iterator to find out who its waiting for */ + struct iter_qstate* qstate = (struct iter_qstate*) + m->s.minfo[m->s.curmod]; + struct outbound_list* ol = &qstate->outlist; + struct outbound_entry* e; + snprintf(buf, len, "%s wait for", modname); + l = strlen(buf); + buf += l; len -= l; + if(ol->first == NULL) + snprintf(buf, len, " (empty_list)"); + for(e = ol->first; e; e = e->next) { + snprintf(buf, len, " "); + l = strlen(buf); + buf += l; len -= l; + addr_to_str(&e->qsent->addr, e->qsent->addrlen, + buf, len); + l = strlen(buf); + buf += l; len -= l; + } + } else if(s == module_wait_subquery) { + /* look in subs from mesh state to see what */ + char nm[257]; + struct mesh_state_ref* sub; + snprintf(buf, len, "%s wants", modname); + l = strlen(buf); + buf += l; len -= l; + if(m->sub_set.count == 0) + snprintf(buf, len, " (empty_list)"); + RBTREE_FOR(sub, struct mesh_state_ref*, &m->sub_set) { + char* t = sldns_wire2str_type(sub->s->s.qinfo.qtype); + char* c = sldns_wire2str_class(sub->s->s.qinfo.qclass); + dname_str(sub->s->s.qinfo.qname, nm); + snprintf(buf, len, " %s %s %s", (t?t:"TYPE??"), + (c?c:"CLASS??"), nm); + l = strlen(buf); + buf += l; len -= l; + free(t); + free(c); + } + } else { + snprintf(buf, len, "%s is %s", modname, strextstate(s)); + } +} + +/** do the dump_requestlist command */ +static void +do_dump_requestlist(SSL* ssl, struct worker* worker) +{ + struct mesh_area* mesh; + struct mesh_state* m; + int num = 0; + char buf[257]; + char timebuf[32]; + char statbuf[10240]; + if(!ssl_printf(ssl, "thread #%d\n", worker->thread_num)) + return; + if(!ssl_printf(ssl, "# type cl name seconds module status\n")) + return; + /* show worker mesh contents */ + mesh = worker->env.mesh; + if(!mesh) return; + RBTREE_FOR(m, struct mesh_state*, &mesh->all) { + char* t = sldns_wire2str_type(m->s.qinfo.qtype); + char* c = sldns_wire2str_class(m->s.qinfo.qclass); + dname_str(m->s.qinfo.qname, buf); + get_mesh_age(m, timebuf, sizeof(timebuf), &worker->env); + get_mesh_status(mesh, m, statbuf, sizeof(statbuf)); + if(!ssl_printf(ssl, "%3d %4s %2s %s %s %s\n", + num, (t?t:"TYPE??"), (c?c:"CLASS??"), buf, timebuf, + statbuf)) { + free(t); + free(c); + return; + } + num++; + free(t); + free(c); + } +} + +/** structure for argument data for dump infra host */ +struct infra_arg { + /** the infra cache */ + struct infra_cache* infra; + /** the SSL connection */ + SSL* ssl; + /** the time now */ + time_t now; + /** ssl failure? stop writing and skip the rest. If the tcp + * connection is broken, and writes fail, we then stop writing. */ + int ssl_failed; +}; + +/** callback for every host element in the infra cache */ +static void +dump_infra_host(struct lruhash_entry* e, void* arg) +{ + struct infra_arg* a = (struct infra_arg*)arg; + struct infra_key* k = (struct infra_key*)e->key; + struct infra_data* d = (struct infra_data*)e->data; + char ip_str[1024]; + char name[257]; + if(a->ssl_failed) + return; + addr_to_str(&k->addr, k->addrlen, ip_str, sizeof(ip_str)); + dname_str(k->zonename, name); + /* skip expired stuff (only backed off) */ + if(d->ttl < a->now) { + if(d->rtt.rto >= USEFUL_SERVER_TOP_TIMEOUT) { + if(!ssl_printf(a->ssl, "%s %s expired rto %d\n", ip_str, + name, d->rtt.rto)) { + a->ssl_failed = 1; + return; + } + } + return; + } + if(!ssl_printf(a->ssl, "%s %s ttl %lu ping %d var %d rtt %d rto %d " + "tA %d tAAAA %d tother %d " + "ednsknown %d edns %d delay %d lame dnssec %d rec %d A %d " + "other %d\n", ip_str, name, (unsigned long)(d->ttl - a->now), + d->rtt.srtt, d->rtt.rttvar, rtt_notimeout(&d->rtt), d->rtt.rto, + d->timeout_A, d->timeout_AAAA, d->timeout_other, + (int)d->edns_lame_known, (int)d->edns_version, + (int)(a->now<d->probedelay?d->probedelay-a->now:0), + (int)d->isdnsseclame, (int)d->rec_lame, (int)d->lame_type_A, + (int)d->lame_other)) { + a->ssl_failed = 1; + return; + } +} + +/** do the dump_infra command */ +static void +do_dump_infra(SSL* ssl, struct worker* worker) +{ + struct infra_arg arg; + arg.infra = worker->env.infra_cache; + arg.ssl = ssl; + arg.now = *worker->env.now; + arg.ssl_failed = 0; + slabhash_traverse(arg.infra->hosts, 0, &dump_infra_host, (void*)&arg); +} + +/** do the log_reopen command */ +static void +do_log_reopen(SSL* ssl, struct worker* worker) +{ + struct config_file* cfg = worker->env.cfg; + send_ok(ssl); + log_init(cfg->logfile, cfg->use_syslog, cfg->chrootdir); +} + +/** do the set_option command */ +static void +do_set_option(SSL* ssl, struct worker* worker, char* arg) +{ + char* arg2; + if(!find_arg2(ssl, arg, &arg2)) + return; + if(!config_set_option(worker->env.cfg, arg, arg2)) { + (void)ssl_printf(ssl, "error setting option\n"); + return; + } + send_ok(ssl); +} + +/* routine to printout option values over SSL */ +void remote_get_opt_ssl(char* line, void* arg) +{ + SSL* ssl = (SSL*)arg; + (void)ssl_printf(ssl, "%s\n", line); +} + +/** do the get_option command */ +static void +do_get_option(SSL* ssl, struct worker* worker, char* arg) +{ + int r; + r = config_get_option(worker->env.cfg, arg, remote_get_opt_ssl, ssl); + if(!r) { + (void)ssl_printf(ssl, "error unknown option\n"); + return; + } +} + +/** do the list_forwards command */ +static void +do_list_forwards(SSL* ssl, struct worker* worker) +{ + /* since its a per-worker structure no locks needed */ + struct iter_forwards* fwds = worker->env.fwds; + struct iter_forward_zone* z; + struct trust_anchor* a; + int insecure; + RBTREE_FOR(z, struct iter_forward_zone*, fwds->tree) { + if(!z->dp) continue; /* skip empty marker for stub */ + + /* see if it is insecure */ + insecure = 0; + if(worker->env.anchors && + (a=anchor_find(worker->env.anchors, z->name, + z->namelabs, z->namelen, z->dclass))) { + if(!a->keylist && !a->numDS && !a->numDNSKEY) + insecure = 1; + lock_basic_unlock(&a->lock); + } + + if(!ssl_print_name_dp(ssl, (insecure?"forward +i":"forward"), + z->name, z->dclass, z->dp)) + return; + } +} + +/** do the list_stubs command */ +static void +do_list_stubs(SSL* ssl, struct worker* worker) +{ + struct iter_hints_stub* z; + struct trust_anchor* a; + int insecure; + char str[32]; + RBTREE_FOR(z, struct iter_hints_stub*, &worker->env.hints->tree) { + + /* see if it is insecure */ + insecure = 0; + if(worker->env.anchors && + (a=anchor_find(worker->env.anchors, z->node.name, + z->node.labs, z->node.len, z->node.dclass))) { + if(!a->keylist && !a->numDS && !a->numDNSKEY) + insecure = 1; + lock_basic_unlock(&a->lock); + } + + snprintf(str, sizeof(str), "stub %sprime%s", + (z->noprime?"no":""), (insecure?" +i":"")); + if(!ssl_print_name_dp(ssl, str, z->node.name, + z->node.dclass, z->dp)) + return; + } +} + +/** do the list_local_zones command */ +static void +do_list_local_zones(SSL* ssl, struct worker* worker) +{ + struct local_zones* zones = worker->daemon->local_zones; + struct local_zone* z; + char buf[257]; + lock_rw_rdlock(&zones->lock); + RBTREE_FOR(z, struct local_zone*, &zones->ztree) { + lock_rw_rdlock(&z->lock); + dname_str(z->name, buf); + if(!ssl_printf(ssl, "%s %s\n", buf, + local_zone_type2str(z->type))) { + /* failure to print */ + lock_rw_unlock(&z->lock); + lock_rw_unlock(&zones->lock); + return; + } + lock_rw_unlock(&z->lock); + } + lock_rw_unlock(&zones->lock); +} + +/** do the list_local_data command */ +static void +do_list_local_data(SSL* ssl, struct worker* worker) +{ + struct local_zones* zones = worker->daemon->local_zones; + struct local_zone* z; + struct local_data* d; + struct local_rrset* p; + char* s = (char*)sldns_buffer_begin(worker->env.scratch_buffer); + size_t slen = sldns_buffer_capacity(worker->env.scratch_buffer); + lock_rw_rdlock(&zones->lock); + RBTREE_FOR(z, struct local_zone*, &zones->ztree) { + lock_rw_rdlock(&z->lock); + RBTREE_FOR(d, struct local_data*, &z->data) { + for(p = d->rrsets; p; p = p->next) { + struct packed_rrset_data* d = + (struct packed_rrset_data*)p->rrset->entry.data; + size_t i; + for(i=0; i<d->count + d->rrsig_count; i++) { + if(!packed_rr_to_string(p->rrset, i, + 0, s, slen)) { + if(!ssl_printf(ssl, "BADRR\n")) + return; + } + if(!ssl_printf(ssl, "%s\n", s)) + return; + } + } + } + lock_rw_unlock(&z->lock); + } + lock_rw_unlock(&zones->lock); +} + +/** tell other processes to execute the command */ +static void +distribute_cmd(struct daemon_remote* rc, SSL* ssl, char* cmd) +{ + int i; + if(!cmd || !ssl) + return; + /* skip i=0 which is me */ + for(i=1; i<rc->worker->daemon->num; i++) { + worker_send_cmd(rc->worker->daemon->workers[i], + worker_cmd_remote); + if(!tube_write_msg(rc->worker->daemon->workers[i]->cmd, + (uint8_t*)cmd, strlen(cmd)+1, 0)) { + ssl_printf(ssl, "error could not distribute cmd\n"); + return; + } + } +} + +/** check for name with end-of-string, space or tab after it */ +static int +cmdcmp(char* p, const char* cmd, size_t len) +{ + return strncmp(p,cmd,len)==0 && (p[len]==0||p[len]==' '||p[len]=='\t'); +} + +/** execute a remote control command */ +static void +execute_cmd(struct daemon_remote* rc, SSL* ssl, char* cmd, + struct worker* worker) +{ + char* p = skipwhite(cmd); + /* compare command */ + if(cmdcmp(p, "stop", 4)) { + do_stop(ssl, rc); + return; + } else if(cmdcmp(p, "reload", 6)) { + do_reload(ssl, rc); + return; + } else if(cmdcmp(p, "stats_noreset", 13)) { + do_stats(ssl, rc, 0); + return; + } else if(cmdcmp(p, "stats", 5)) { + do_stats(ssl, rc, 1); + return; + } else if(cmdcmp(p, "status", 6)) { + do_status(ssl, worker); + return; + } else if(cmdcmp(p, "dump_cache", 10)) { + (void)dump_cache(ssl, worker); + return; + } else if(cmdcmp(p, "load_cache", 10)) { + if(load_cache(ssl, worker)) send_ok(ssl); + return; + } else if(cmdcmp(p, "list_forwards", 13)) { + do_list_forwards(ssl, worker); + return; + } else if(cmdcmp(p, "list_stubs", 10)) { + do_list_stubs(ssl, worker); + return; + } else if(cmdcmp(p, "list_local_zones", 16)) { + do_list_local_zones(ssl, worker); + return; + } else if(cmdcmp(p, "list_local_data", 15)) { + do_list_local_data(ssl, worker); + return; + } else if(cmdcmp(p, "stub_add", 8)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_stub_add(ssl, worker, skipwhite(p+8)); + return; + } else if(cmdcmp(p, "stub_remove", 11)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_stub_remove(ssl, worker, skipwhite(p+11)); + return; + } else if(cmdcmp(p, "forward_add", 11)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_forward_add(ssl, worker, skipwhite(p+11)); + return; + } else if(cmdcmp(p, "forward_remove", 14)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_forward_remove(ssl, worker, skipwhite(p+14)); + return; + } else if(cmdcmp(p, "insecure_add", 12)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_insecure_add(ssl, worker, skipwhite(p+12)); + return; + } else if(cmdcmp(p, "insecure_remove", 15)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_insecure_remove(ssl, worker, skipwhite(p+15)); + return; + } else if(cmdcmp(p, "forward", 7)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_forward(ssl, worker, skipwhite(p+7)); + return; + } else if(cmdcmp(p, "flush_stats", 11)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_flush_stats(ssl, worker); + return; + } else if(cmdcmp(p, "flush_requestlist", 17)) { + /* must always distribute this cmd */ + if(rc) distribute_cmd(rc, ssl, cmd); + do_flush_requestlist(ssl, worker); + return; + } else if(cmdcmp(p, "lookup", 6)) { + do_lookup(ssl, worker, skipwhite(p+6)); + return; + } + +#ifdef THREADS_DISABLED + /* other processes must execute the command as well */ + /* commands that should not be distributed, returned above. */ + if(rc) { /* only if this thread is the master (rc) thread */ + /* done before the code below, which may split the string */ + distribute_cmd(rc, ssl, cmd); + } +#endif + if(cmdcmp(p, "verbosity", 9)) { + do_verbosity(ssl, skipwhite(p+9)); + } else if(cmdcmp(p, "local_zone_remove", 17)) { + do_zone_remove(ssl, worker, skipwhite(p+17)); + } else if(cmdcmp(p, "local_zone", 10)) { + do_zone_add(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "local_data_remove", 17)) { + do_data_remove(ssl, worker, skipwhite(p+17)); + } else if(cmdcmp(p, "local_data", 10)) { + do_data_add(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "flush_zone", 10)) { + do_flush_zone(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "flush_type", 10)) { + do_flush_type(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "flush_infra", 11)) { + do_flush_infra(ssl, worker, skipwhite(p+11)); + } else if(cmdcmp(p, "flush", 5)) { + do_flush_name(ssl, worker, skipwhite(p+5)); + } else if(cmdcmp(p, "dump_requestlist", 16)) { + do_dump_requestlist(ssl, worker); + } else if(cmdcmp(p, "dump_infra", 10)) { + do_dump_infra(ssl, worker); + } else if(cmdcmp(p, "log_reopen", 10)) { + do_log_reopen(ssl, worker); + } else if(cmdcmp(p, "set_option", 10)) { + do_set_option(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "get_option", 10)) { + do_get_option(ssl, worker, skipwhite(p+10)); + } else if(cmdcmp(p, "flush_bogus", 11)) { + do_flush_bogus(ssl, worker); + } else if(cmdcmp(p, "flush_negative", 14)) { + do_flush_negative(ssl, worker); + } else { + (void)ssl_printf(ssl, "error unknown command '%s'\n", p); + } +} + +void +daemon_remote_exec(struct worker* worker) +{ + /* read the cmd string */ + uint8_t* msg = NULL; + uint32_t len = 0; + if(!tube_read_msg(worker->cmd, &msg, &len, 0)) { + log_err("daemon_remote_exec: tube_read_msg failed"); + return; + } + verbose(VERB_ALGO, "remote exec distributed: %s", (char*)msg); + execute_cmd(NULL, NULL, (char*)msg, worker); + free(msg); +} + +/** handle remote control request */ +static void +handle_req(struct daemon_remote* rc, struct rc_state* s, SSL* ssl) +{ + int r; + char pre[10]; + char magic[7]; + char buf[1024]; +#ifdef USE_WINSOCK + /* makes it possible to set the socket blocking again. */ + /* basically removes it from winsock_event ... */ + WSAEventSelect(s->c->fd, NULL, 0); +#endif + fd_set_block(s->c->fd); + + /* try to read magic UBCT[version]_space_ string */ + ERR_clear_error(); + if((r=SSL_read(ssl, magic, (int)sizeof(magic)-1)) <= 0) { + if(SSL_get_error(ssl, r) == SSL_ERROR_ZERO_RETURN) + return; + log_crypto_err("could not SSL_read"); + return; + } + magic[6] = 0; + if( r != 6 || strncmp(magic, "UBCT", 4) != 0) { + verbose(VERB_QUERY, "control connection has bad magic string"); + /* probably wrong tool connected, ignore it completely */ + return; + } + + /* read the command line */ + if(!ssl_read_line(ssl, buf, sizeof(buf))) { + return; + } + snprintf(pre, sizeof(pre), "UBCT%d ", UNBOUND_CONTROL_VERSION); + if(strcmp(magic, pre) != 0) { + verbose(VERB_QUERY, "control connection had bad " + "version %s, cmd: %s", magic, buf); + ssl_printf(ssl, "error version mismatch\n"); + return; + } + verbose(VERB_DETAIL, "control cmd: %s", buf); + + /* figure out what to do */ + execute_cmd(rc, ssl, buf, rc->worker); +} + +int remote_control_callback(struct comm_point* c, void* arg, int err, + struct comm_reply* ATTR_UNUSED(rep)) +{ + struct rc_state* s = (struct rc_state*)arg; + struct daemon_remote* rc = s->rc; + int r; + if(err != NETEVENT_NOERROR) { + if(err==NETEVENT_TIMEOUT) + log_err("remote control timed out"); + clean_point(rc, s); + return 0; + } + /* (continue to) setup the SSL connection */ + ERR_clear_error(); + r = SSL_do_handshake(s->ssl); + if(r != 1) { + int r2 = SSL_get_error(s->ssl, r); + if(r2 == SSL_ERROR_WANT_READ) { + if(s->shake_state == rc_hs_read) { + /* try again later */ + return 0; + } + s->shake_state = rc_hs_read; + comm_point_listen_for_rw(c, 1, 0); + return 0; + } else if(r2 == SSL_ERROR_WANT_WRITE) { + if(s->shake_state == rc_hs_write) { + /* try again later */ + return 0; + } + s->shake_state = rc_hs_write; + comm_point_listen_for_rw(c, 0, 1); + return 0; + } else { + if(r == 0) + log_err("remote control connection closed prematurely"); + log_addr(1, "failed connection from", + &s->c->repinfo.addr, s->c->repinfo.addrlen); + log_crypto_err("remote control failed ssl"); + clean_point(rc, s); + return 0; + } + } + s->shake_state = rc_none; + + /* once handshake has completed, check authentication */ + if(SSL_get_verify_result(s->ssl) == X509_V_OK) { + X509* x = SSL_get_peer_certificate(s->ssl); + if(!x) { + verbose(VERB_DETAIL, "remote control connection " + "provided no client certificate"); + clean_point(rc, s); + return 0; + } + verbose(VERB_ALGO, "remote control connection authenticated"); + X509_free(x); + } else { + verbose(VERB_DETAIL, "remote control connection failed to " + "authenticate with client certificate"); + clean_point(rc, s); + return 0; + } + + /* if OK start to actually handle the request */ + handle_req(rc, s, s->ssl); + + verbose(VERB_ALGO, "remote control operation completed"); + clean_point(rc, s); + return 0; +} |