mirror of
https://github.com/zeek/zeek.git
synced 2025-10-01 22:28:20 +00:00
1730 lines
56 KiB
C++
1730 lines
56 KiB
C++
// See the file "COPYING" in the main distribution directory for copyright.
|
|
|
|
#include "zeek/DNS_Mgr.h"
|
|
|
|
#include "zeek/zeek-config.h"
|
|
|
|
#include <netdb.h>
|
|
#include <netinet/in.h>
|
|
#include <poll.h>
|
|
#include <sys/socket.h>
|
|
#include <sys/stat.h>
|
|
#include <sys/types.h>
|
|
#include <unistd.h>
|
|
#include <algorithm>
|
|
#include <cerrno>
|
|
#include <cstdlib>
|
|
#include <vector>
|
|
|
|
#ifdef TIME_WITH_SYS_TIME
|
|
#include <sys/time.h>
|
|
#include <ctime>
|
|
#elif defined(HAVE_SYS_TIME_H)
|
|
#include <sys/time.h>
|
|
#else
|
|
#include <ctime>
|
|
#endif
|
|
|
|
#include <ztd/out_ptr.hpp>
|
|
using ztd::out_ptr::out_ptr;
|
|
|
|
#include <ares.h>
|
|
#include <ares_dns.h>
|
|
#include <ares_nameser.h>
|
|
|
|
#include "zeek/DNS_Mapping.h"
|
|
#include "zeek/Event.h"
|
|
#include "zeek/Expr.h"
|
|
#include "zeek/Hash.h"
|
|
#include "zeek/ID.h"
|
|
#include "zeek/IntrusivePtr.h"
|
|
#include "zeek/Reporter.h"
|
|
#include "zeek/RunState.h"
|
|
#include "zeek/Val.h"
|
|
#include "zeek/ZeekString.h"
|
|
#include "zeek/iosource/Manager.h"
|
|
#include "zeek/telemetry/Manager.h"
|
|
|
|
#include "zeek/3rdparty/doctest.h"
|
|
|
|
// Number of seconds we'll wait for a reply.
|
|
constexpr int DNS_TIMEOUT = 5;
|
|
|
|
// The maximum allowed number of pending asynchronous requests.
|
|
constexpr int MAX_PENDING_REQUESTS = 20;
|
|
|
|
// The maximum number of bytes requested via UDP. TCP fallback won't happen on
|
|
// requests until a response is larger than this.
|
|
constexpr int MAX_UDP_BUFFER_SIZE = 4096;
|
|
|
|
// This unfortunately doesn't exist in c-ares, even though it seems rather useful.
|
|
static const char* request_type_string(int request_type) {
|
|
switch ( request_type ) {
|
|
case T_A: return "T_A";
|
|
case T_NS: return "T_NS";
|
|
case T_MD: return "T_MD";
|
|
case T_MF: return "T_MF";
|
|
case T_CNAME: return "T_CNAME";
|
|
case T_SOA: return "T_SOA";
|
|
case T_MB: return "T_MB";
|
|
case T_MG: return "T_MG";
|
|
case T_MR: return "T_MR";
|
|
case T_NULL: return "T_NULL";
|
|
case T_WKS: return "T_WKS";
|
|
case T_PTR: return "T_PTR";
|
|
case T_HINFO: return "T_HINFO";
|
|
case T_MINFO: return "T_MINFO";
|
|
case T_MX: return "T_MX";
|
|
case T_TXT: return "T_TXT";
|
|
case T_RP: return "T_RP";
|
|
case T_AFSDB: return "T_AFSDB";
|
|
case T_X25: return "T_X25";
|
|
case T_ISDN: return "T_ISDN";
|
|
case T_RT: return "T_RT";
|
|
case T_NSAP: return "T_NSAP";
|
|
case T_NSAP_PTR: return "T_NSAP_PTR";
|
|
case T_SIG: return "T_SIG";
|
|
case T_KEY: return "T_KEY";
|
|
case T_PX: return "T_PX";
|
|
case T_GPOS: return "T_GPOS";
|
|
case T_AAAA: return "T_AAAA";
|
|
case T_LOC: return "T_LOC";
|
|
case T_NXT: return "T_NXT";
|
|
case T_EID: return "T_EID";
|
|
case T_NIMLOC: return "T_NIMLOC";
|
|
case T_SRV: return "T_SRV";
|
|
case T_ATMA: return "T_ATMA";
|
|
case T_NAPTR: return "T_NAPTR";
|
|
case T_KX: return "T_KX";
|
|
case T_CERT: return "T_CERT";
|
|
case T_A6: return "T_A6";
|
|
case T_DNAME: return "T_DNAME";
|
|
case T_SINK: return "T_SINK";
|
|
case T_OPT: return "T_OPT";
|
|
case T_APL: return "T_APL";
|
|
case T_DS: return "T_DS";
|
|
case T_SSHFP: return "T_SSHFP";
|
|
case T_RRSIG: return "T_RRSIG";
|
|
case T_NSEC: return "T_NSEC";
|
|
case T_DNSKEY: return "T_DNSKEY";
|
|
case T_TKEY: return "T_TKEY";
|
|
case T_TSIG: return "T_TSIG";
|
|
case T_IXFR: return "T_IXFR";
|
|
case T_AXFR: return "T_AXFR";
|
|
case T_MAILB: return "T_MAILB";
|
|
case T_MAILA: return "T_MAILA";
|
|
case T_ANY: return "T_ANY";
|
|
case T_URI: return "T_URI";
|
|
case T_CAA: return "T_CAA";
|
|
case T_MAX: return "T_MAX";
|
|
default: return "";
|
|
}
|
|
}
|
|
|
|
struct ares_deleter {
|
|
void operator()(char* s) const { ares_free_string(s); }
|
|
void operator()(ares_addrinfo* s) const { ares_freeaddrinfo(s); }
|
|
void operator()(ares_dns_record_t* r) const { ares_dns_record_destroy(r); }
|
|
};
|
|
|
|
namespace zeek::detail {
|
|
static void addrinfo_cb(void* arg, int status, int timeouts, struct ares_addrinfo* result);
|
|
static void query_cb(void* arg, ares_status_t status, size_t timeouts, const ares_dns_record* dnsrec);
|
|
static void sock_cb(void* data, int s, int read, int write);
|
|
|
|
struct CallbackArgs {
|
|
DNS_Request* req;
|
|
DNS_Mgr* mgr;
|
|
};
|
|
|
|
class DNS_Request {
|
|
public:
|
|
DNS_Request(std::string host, int request_type, bool async = false);
|
|
DNS_Request(const IPAddr& addr, bool async = false);
|
|
~DNS_Request() = default;
|
|
|
|
std::string Host() const { return host; }
|
|
const IPAddr& Addr() const { return addr; }
|
|
int RequestType() const { return request_type; }
|
|
bool IsTxt() const { return request_type == 16; }
|
|
|
|
void MakeRequest(ares_channel channel, DNS_Mgr* mgr);
|
|
void ProcessAsyncResult(bool timed_out, DNS_Mgr* mgr);
|
|
|
|
private:
|
|
std::string host;
|
|
IPAddr addr;
|
|
int request_type = 0; // Query type
|
|
bool async = false;
|
|
std::unique_ptr<ares_dns_record_t, ares_deleter> query_rec;
|
|
static uint16_t request_id;
|
|
};
|
|
|
|
uint16_t DNS_Request::request_id = 0;
|
|
|
|
DNS_Request::DNS_Request(std::string host, int request_type, bool async)
|
|
: host(std::move(host)), request_type(request_type), async(async) {
|
|
// We combine the T_A and T_AAAA requests together in one request, so set the type
|
|
// to T_A to make things easier in other parts of the code (mostly around lookups).
|
|
if ( request_type == T_AAAA )
|
|
request_type = T_A;
|
|
}
|
|
|
|
DNS_Request::DNS_Request(const IPAddr& addr, bool async) : addr(addr), async(async) { request_type = T_PTR; }
|
|
|
|
void DNS_Request::MakeRequest(ares_channel channel, DNS_Mgr* mgr) {
|
|
// This needs to get deleted at the end of the callback method.
|
|
auto req_data = std::make_unique<CallbackArgs>();
|
|
req_data->req = this;
|
|
req_data->mgr = mgr;
|
|
|
|
// It's completely fine if this rolls over. It's just to keep the query ID different
|
|
// from one query to the next, and it's unlikely we'd do 2^16 queries so fast that
|
|
// all of them would be in flight at the same time.
|
|
DNS_Request::request_id++;
|
|
|
|
if ( request_type == T_A ) {
|
|
// For A/AAAA requests, we use a different method than the other requests. Since
|
|
// we're using the AF_UNSPEC family, we get both the ipv4 and ipv6 responses
|
|
// back in the same request if use ares_getaddrinfo() so we can store them both
|
|
// in the same mapping.
|
|
ares_addrinfo_hints hints = {ARES_AI_CANONNAME, AF_UNSPEC, 0, 0};
|
|
ares_getaddrinfo(channel, host.c_str(), nullptr, &hints, addrinfo_cb, req_data.release());
|
|
}
|
|
else {
|
|
std::string query_host;
|
|
if ( request_type == T_PTR )
|
|
query_host = addr.PtrName();
|
|
else
|
|
query_host = host;
|
|
|
|
std::unique_ptr<ares_dns_record_t, ares_deleter> dnsrec;
|
|
int len = 0;
|
|
ares_status_t status = ares_dns_record_create(out_ptr<ares_dns_record_t*>(dnsrec), DNS_Request::request_id,
|
|
ARES_FLAG_RD, ARES_OPCODE_QUERY, ARES_RCODE_NOERROR);
|
|
|
|
if ( status != ARES_SUCCESS || dnsrec == nullptr )
|
|
return;
|
|
|
|
status = ares_dns_record_query_add(dnsrec.get(), query_host.c_str(),
|
|
static_cast<ares_dns_rec_type_t>(request_type), ARES_CLASS_IN);
|
|
|
|
if ( status != ARES_SUCCESS )
|
|
return;
|
|
|
|
// Store this so it can be destroyed when the request is destroyed.
|
|
this->query_rec = std::move(dnsrec);
|
|
ares_send_dnsrec(channel, query_rec.get(), query_cb, req_data.release(), nullptr);
|
|
}
|
|
}
|
|
|
|
void DNS_Request::ProcessAsyncResult(bool timed_out, DNS_Mgr* mgr) {
|
|
if ( ! async )
|
|
return;
|
|
|
|
if ( request_type == T_A )
|
|
mgr->CheckAsyncHostRequest(host, timed_out);
|
|
else if ( request_type == T_PTR )
|
|
mgr->CheckAsyncAddrRequest(addr, timed_out);
|
|
else
|
|
mgr->CheckAsyncOtherRequest(host, timed_out, request_type);
|
|
}
|
|
|
|
/**
|
|
* Retrieves the TTL value from the first RR in the response.
|
|
*
|
|
* This code is adapted from an internal c-ares method called * ares__parse_into_addrinfo,
|
|
* which is used for ares_getaddrinfo callbacks. It's also the only method that properly
|
|
* parses out TTL data currently. This skips over the question and the first bit of the
|
|
* response to get to the first RR, and then returns the TTL from that RR. We only use the
|
|
* first RR because it's a good approximation for now, at least until the work in c-ares
|
|
* lands to add TTL support to the other RR-parsing methods.
|
|
*
|
|
* @param abuf The buffer containing the entire response from the server.
|
|
* @param alen The length of the buffer
|
|
* @param ttl An out param for returning the TTL value.
|
|
* @return A status code from c-ares. This will be ARES_SUCCESS on success, or some other
|
|
* code on failure.
|
|
*/
|
|
static int get_ttl(unsigned char* abuf, int alen, int* ttl) {
|
|
int status;
|
|
long len;
|
|
std::unique_ptr<char, ares_deleter> hostname;
|
|
|
|
*ttl = DNS_TIMEOUT;
|
|
|
|
unsigned char* aptr = abuf + HFIXEDSZ;
|
|
status = ares_expand_name(aptr, abuf, alen, out_ptr<char*>(hostname), &len);
|
|
if ( status != ARES_SUCCESS )
|
|
return status;
|
|
|
|
if ( aptr + len + QFIXEDSZ > abuf + alen )
|
|
return ARES_EBADRESP;
|
|
|
|
aptr += len + QFIXEDSZ;
|
|
hostname.reset();
|
|
|
|
status = ares_expand_name(aptr, abuf, alen, out_ptr<char*>(hostname), &len);
|
|
if ( status != ARES_SUCCESS )
|
|
return status;
|
|
|
|
if ( aptr + RRFIXEDSZ > abuf + alen )
|
|
return ARES_EBADRESP;
|
|
|
|
aptr += len;
|
|
*ttl = DNS_RR_TTL(aptr);
|
|
|
|
return status;
|
|
}
|
|
|
|
/**
|
|
* Called in response to ares_getaddrinfo requests. Builds a hostent structure from
|
|
* the result data and sends it to the DNS manager via AddResult().
|
|
*/
|
|
static void addrinfo_cb(void* arg, int status, int timeouts, struct ares_addrinfo* result) {
|
|
auto arg_data = reinterpret_cast<CallbackArgs*>(arg);
|
|
const auto [req, mgr] = *arg_data;
|
|
std::unique_ptr<ares_addrinfo, ares_deleter> res_ptr(result);
|
|
|
|
if ( status != ARES_SUCCESS ) {
|
|
// These two statuses should only ever be sent if we're shutting down everything
|
|
// and all of the existing queries are being cancelled. There's no reason to
|
|
// store a status that's just going to get deleted, nor is there a reason to log
|
|
// anything.
|
|
if ( status != ARES_ECANCELLED && status != ARES_EDESTRUCTION ) {
|
|
// Insert something into the cache so that the request loop will end correctly.
|
|
// We use the DNS_TIMEOUT value as the TTL here since it's small enough that the
|
|
// failed response will expire soon, and because we don't have the TTL from the
|
|
// response data.
|
|
mgr->AddResult(req, nullptr, DNS_TIMEOUT);
|
|
}
|
|
}
|
|
else {
|
|
std::vector<in_addr*> addrs;
|
|
std::vector<in6_addr*> addrs6;
|
|
for ( ares_addrinfo_node* entry = result->nodes; entry != nullptr; entry = entry->ai_next ) {
|
|
if ( entry->ai_family == AF_INET ) {
|
|
struct sockaddr_in* addr = reinterpret_cast<sockaddr_in*>(entry->ai_addr);
|
|
addrs.push_back(&addr->sin_addr);
|
|
}
|
|
else if ( entry->ai_family == AF_INET6 ) {
|
|
struct sockaddr_in6* addr = (struct sockaddr_in6*)(entry->ai_addr);
|
|
addrs6.push_back(&addr->sin6_addr);
|
|
}
|
|
}
|
|
|
|
if ( ! addrs.empty() ) {
|
|
// Push a null on the end so the addr list has a final point during later parsing.
|
|
addrs.push_back(nullptr);
|
|
|
|
struct hostent he{};
|
|
he.h_name = util::copy_string(result->name);
|
|
he.h_addrtype = AF_INET;
|
|
he.h_length = sizeof(in_addr);
|
|
he.h_addr_list = reinterpret_cast<char**>(addrs.data());
|
|
|
|
mgr->AddResult(req, &he, result->nodes[0].ai_ttl);
|
|
|
|
delete[] he.h_name;
|
|
}
|
|
|
|
if ( ! addrs6.empty() ) {
|
|
// Push a null on the end so the addr list has a final point during later parsing.
|
|
addrs6.push_back(nullptr);
|
|
|
|
struct hostent he{};
|
|
he.h_name = util::copy_string(result->name);
|
|
he.h_addrtype = AF_INET6;
|
|
he.h_length = sizeof(in6_addr);
|
|
he.h_addr_list = reinterpret_cast<char**>(addrs6.data());
|
|
|
|
mgr->AddResult(req, &he, result->nodes[0].ai_ttl, true);
|
|
|
|
delete[] he.h_name;
|
|
}
|
|
}
|
|
|
|
req->ProcessAsyncResult(timeouts > 0, mgr);
|
|
|
|
// TODO: might need to turn these into unique_ptr as well?
|
|
delete req;
|
|
delete arg_data;
|
|
}
|
|
|
|
static void query_cb(void* arg, ares_status_t status, size_t timeouts, const ares_dns_record_t* dnsrec) {
|
|
auto arg_data = reinterpret_cast<CallbackArgs*>(arg);
|
|
const auto [req, mgr] = *arg_data;
|
|
|
|
if ( status != ARES_SUCCESS ) {
|
|
// These two statuses should only ever be sent if we're shutting down everything
|
|
// and all of the existing queries are being cancelled. There's no reason to
|
|
// store a status that's just going to get deleted, nor is there a reason to log
|
|
// anything.
|
|
if ( status != ARES_ECANCELLED && status != ARES_EDESTRUCTION ) {
|
|
// Insert something into the cache so that the request loop will end correctly.
|
|
// We use the DNS_TIMEOUT value as the TTL here since it's small enough that the
|
|
// failed response will expire soon, and because we don't have the TTL from the
|
|
// response data.
|
|
mgr->AddResult(req, nullptr, DNS_TIMEOUT);
|
|
}
|
|
}
|
|
else {
|
|
struct hostent he{};
|
|
|
|
uint32_t ttl = 0;
|
|
size_t rr_cnt = ares_dns_record_rr_cnt(dnsrec, ARES_SECTION_ANSWER);
|
|
bool error = false;
|
|
|
|
for ( size_t idx = 0; idx < rr_cnt; idx++ ) {
|
|
const ares_dns_rr_t* rr = ares_dns_record_rr_get_const(dnsrec, ARES_SECTION_ANSWER, 0);
|
|
|
|
// Use the ttl from the first record, since we don't keep track of the TTLs for each
|
|
// individually.
|
|
if ( idx == 0 )
|
|
ttl = ares_dns_rr_get_ttl(rr);
|
|
|
|
ares_dns_rec_type_t type = ares_dns_rr_get_type(rr);
|
|
if ( req->RequestType() != type )
|
|
continue;
|
|
|
|
if ( type == ARES_REC_TYPE_PTR ) {
|
|
const char* txt = ares_dns_rr_get_str(rr, ARES_RR_PTR_DNAME);
|
|
if ( txt == nullptr ) {
|
|
// According to the c-ares docs, this can happen but only in cases of "misuse". We
|
|
// still need to check for it though.
|
|
error = true;
|
|
break;
|
|
}
|
|
|
|
// TODO: it's possible that a response has multiple aliases. We
|
|
// don't handle those so we can just break here after setting
|
|
// h_aliases to null.
|
|
he.h_name = util::copy_string(txt);
|
|
he.h_aliases = nullptr;
|
|
break;
|
|
}
|
|
else if ( type == ARES_REC_TYPE_TXT ) {
|
|
size_t abin_cnt = ares_dns_rr_get_abin_cnt(rr, ARES_RR_TXT_DATA);
|
|
if ( abin_cnt == 0 )
|
|
break;
|
|
|
|
// TODO: We only process the first abin in the response. There might be more.
|
|
size_t abin_len;
|
|
const unsigned char* abin = ares_dns_rr_get_abin(rr, ARES_RR_TXT_DATA, 0, &abin_len);
|
|
if ( abin == nullptr ) {
|
|
// According to the c-ares docs, this can happen but only in cases of "misuse". We
|
|
// still need to check for it though.
|
|
error = true;
|
|
break;
|
|
}
|
|
|
|
he.h_name = new char[abin_len + 1];
|
|
strncpy(he.h_name, reinterpret_cast<const char*>(abin), abin_len);
|
|
he.h_name[abin_len] = 0;
|
|
he.h_aliases = nullptr;
|
|
|
|
// TODO: We only process the first RR for a TXT query, even if there are more of them.
|
|
break;
|
|
}
|
|
else {
|
|
error = true;
|
|
reporter->Error("Requests of type %d (%s) are unsupported", type, ares_dns_rec_type_tostr(type));
|
|
break;
|
|
}
|
|
}
|
|
|
|
if ( rr_cnt != 0 && ! error )
|
|
mgr->AddResult(req, &he, ttl);
|
|
else
|
|
// See above for why DNS_TIMEOUT here.
|
|
mgr->AddResult(req, nullptr, DNS_TIMEOUT);
|
|
|
|
delete[] he.h_name;
|
|
}
|
|
|
|
req->ProcessAsyncResult(timeouts > 0, mgr);
|
|
delete arg_data;
|
|
delete req;
|
|
}
|
|
|
|
/**
|
|
* Called when the c-ares socket changes state, which indicates that it's connected to
|
|
* some source of data (either a host file or a DNS server). This indicates that we're
|
|
* able to do lookups against c-ares now and should activate the IOSource.
|
|
*/
|
|
static void sock_cb(void* data, ares_socket_t s, int read, int write) {
|
|
auto mgr = reinterpret_cast<DNS_Mgr*>(data);
|
|
mgr->RegisterSocket((int)s, read == 1, write == 1);
|
|
}
|
|
|
|
DNS_Mgr::DNS_Mgr(DNS_MgrMode arg_mode) : IOSource(true), mode(arg_mode) { ares_library_init(ARES_LIB_INIT_ALL); }
|
|
|
|
DNS_Mgr::~DNS_Mgr() {
|
|
Flush();
|
|
|
|
ares_cancel(channel);
|
|
ares_destroy(channel);
|
|
ares_library_cleanup();
|
|
}
|
|
|
|
void DNS_Mgr::Done() {
|
|
shutting_down = true;
|
|
Flush();
|
|
}
|
|
|
|
void DNS_Mgr::RegisterSocket(int fd, bool read, bool write) {
|
|
if ( read && ! socket_fds.contains(fd) ) {
|
|
socket_fds.insert(fd);
|
|
iosource_mgr->RegisterFd(fd, this, IOSource::READ);
|
|
}
|
|
else if ( ! read && socket_fds.contains(fd) ) {
|
|
socket_fds.erase(fd);
|
|
iosource_mgr->UnregisterFd(fd, this, IOSource::READ);
|
|
}
|
|
|
|
if ( write && ! write_socket_fds.contains(fd) ) {
|
|
write_socket_fds.insert(fd);
|
|
iosource_mgr->RegisterFd(fd, this, IOSource::WRITE);
|
|
}
|
|
else if ( ! write && write_socket_fds.contains(fd) ) {
|
|
write_socket_fds.erase(fd);
|
|
iosource_mgr->UnregisterFd(fd, this, IOSource::WRITE);
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::InitSource() {
|
|
if ( did_init )
|
|
return;
|
|
|
|
ares_options options;
|
|
int optmask = 0;
|
|
|
|
// Enable an EDNS option to be sent with the requests. This allows us to set
|
|
// a bigger UDP buffer size in the request, which prevents fallback to TCP
|
|
// at least up to that size.
|
|
options.flags = ARES_FLAG_EDNS;
|
|
optmask |= ARES_OPT_FLAGS;
|
|
|
|
options.ednspsz = MAX_UDP_BUFFER_SIZE;
|
|
optmask |= ARES_OPT_EDNSPSZ;
|
|
|
|
options.socket_receive_buffer_size = MAX_UDP_BUFFER_SIZE;
|
|
optmask |= ARES_OPT_SOCK_RCVBUF;
|
|
|
|
// This option is in milliseconds.
|
|
options.timeout = DNS_TIMEOUT * 1000;
|
|
optmask |= ARES_OPT_TIMEOUTMS;
|
|
|
|
// This causes c-ares to only attempt each server twice before
|
|
// giving up.
|
|
options.tries = 2;
|
|
optmask |= ARES_OPT_TRIES;
|
|
|
|
// See the comment on sock_cb for how this gets used.
|
|
options.sock_state_cb = sock_cb;
|
|
options.sock_state_cb_data = this;
|
|
optmask |= ARES_OPT_SOCK_STATE_CB;
|
|
|
|
int status = ares_init_options(&channel, &options, optmask);
|
|
if ( status != ARES_SUCCESS )
|
|
reporter->FatalError("Failed to initialize c-ares for DNS resolution: %s", ares_strerror(status));
|
|
|
|
// Note that Init() may be called by way of LookupHost() during the act of
|
|
// parsing a hostname literal (e.g. google.com), so we can't use a
|
|
// script-layer option to configure the DNS resolver as it may not be
|
|
// configured to the user's desired address at the time when we need to to
|
|
// the lookup.
|
|
auto dns_resolver = getenv("ZEEK_DNS_RESOLVER");
|
|
if ( dns_resolver ) {
|
|
ares_set_servers_csv(channel, dns_resolver);
|
|
}
|
|
|
|
did_init = true;
|
|
}
|
|
|
|
void DNS_Mgr::InitPostScript() {
|
|
num_requests_metric =
|
|
telemetry_mgr->CounterInstance("zeek", "dnsmgr_requests", {}, "Total number of requests through DNS_Mgr");
|
|
successful_metric = telemetry_mgr->CounterInstance("zeek", "dnsmgr_successful_requests", {},
|
|
"Total number of successful requests through DNS_Mgr");
|
|
failed_metric = telemetry_mgr->CounterInstance("zeek", "dnsmgr_failed_requests", {},
|
|
"Total number of failed requests through DNS_Mgr");
|
|
asyncs_pending_metric = telemetry_mgr->GaugeInstance("zeek", "dnsmgr_pending_asyncs_requests", {},
|
|
"Number of pending async requests through DNS_Mgr");
|
|
|
|
cached_hosts_metric =
|
|
telemetry_mgr->GaugeInstance("zeek", "dnsmgr_cache_entries", {{"type", "host"}},
|
|
"Number of cached hosts in DNS_Mgr", "", []() {
|
|
double value = 0;
|
|
|
|
if ( dns_mgr ) {
|
|
dns_mgr->UpdateCachedStats(false);
|
|
value = static_cast<double>(dns_mgr->last_cached_stats.hosts);
|
|
}
|
|
return value;
|
|
});
|
|
|
|
cached_addresses_metric =
|
|
telemetry_mgr->GaugeInstance("zeek", "dnsmgr_cache_entries", {{"type", "address"}},
|
|
"Number of cached addresses in DNS_Mgr", "", []() {
|
|
double value = 0;
|
|
|
|
if ( dns_mgr ) {
|
|
dns_mgr->UpdateCachedStats(false);
|
|
value = static_cast<double>(dns_mgr->last_cached_stats.addresses);
|
|
}
|
|
return value;
|
|
});
|
|
|
|
cached_texts_metric =
|
|
telemetry_mgr->GaugeInstance("zeek", "dnsmgr_cache_entries", {{"type", "text"}},
|
|
"Number of cached texts in DNS_Mgr", "", []() {
|
|
double value = 0;
|
|
|
|
if ( dns_mgr ) {
|
|
dns_mgr->UpdateCachedStats(false);
|
|
value = static_cast<double>(dns_mgr->last_cached_stats.texts);
|
|
}
|
|
return value;
|
|
});
|
|
|
|
if ( ! doctest::is_running_in_test ) {
|
|
dm_rec = id::find_type<RecordType>("dns_mapping");
|
|
|
|
// Registering will call InitSource(), which sets up all of the DNS library stuff
|
|
iosource_mgr->Register(this, true);
|
|
}
|
|
else {
|
|
// This would normally be called when registering the iosource above.
|
|
InitSource();
|
|
}
|
|
|
|
// Load the DNS cache from disk, if it exists.
|
|
std::string cache_dir = dir.empty() ? "." : dir;
|
|
cache_name = util::fmt("%s/%s", cache_dir.c_str(), ".zeek-dns-cache");
|
|
LoadCache(cache_name);
|
|
}
|
|
|
|
static TableValPtr fake_name_lookup_result(const std::string& name) {
|
|
hash128_t hash;
|
|
KeyedHash::StaticHash128(name.c_str(), name.size(), &hash);
|
|
auto hv = make_intrusive<ListVal>(TYPE_ADDR);
|
|
hv->Append(make_intrusive<AddrVal>(reinterpret_cast<const uint32_t*>(&hash)));
|
|
return hv->ToSetVal();
|
|
}
|
|
|
|
static std::string fake_lookup_result(const std::string& name, int request_type) {
|
|
return util::fmt("fake_lookup_result_%s_%s", request_type_string(request_type), name.c_str());
|
|
}
|
|
|
|
static std::string fake_addr_lookup_result(const IPAddr& addr) {
|
|
return util::fmt("fake_addr_lookup_result_%s", addr.AsString().c_str());
|
|
}
|
|
|
|
static void resolve_lookup_cb(DNS_Mgr::LookupCallback* callback, TableValPtr result) {
|
|
callback->Resolved(std::move(result));
|
|
delete callback;
|
|
}
|
|
|
|
static void resolve_lookup_cb(DNS_Mgr::LookupCallback* callback, const std::string& result) {
|
|
callback->Resolved(result);
|
|
delete callback;
|
|
}
|
|
|
|
ValPtr DNS_Mgr::Lookup(const std::string& name, int request_type) {
|
|
if ( shutting_down )
|
|
return nullptr;
|
|
|
|
if ( request_type == T_A || request_type == T_AAAA )
|
|
return LookupHost(name);
|
|
|
|
if ( mode == DNS_FAKE )
|
|
return make_intrusive<StringVal>(fake_lookup_result(name, request_type));
|
|
|
|
InitSource();
|
|
|
|
if ( mode != DNS_PRIME ) {
|
|
if ( auto val = LookupOtherInCache(name, request_type, false) )
|
|
return val;
|
|
}
|
|
|
|
switch ( mode ) {
|
|
case DNS_PRIME: {
|
|
auto req = new DNS_Request(name, request_type);
|
|
req->MakeRequest(channel, this);
|
|
return empty_addr_set();
|
|
}
|
|
|
|
case DNS_FORCE:
|
|
reporter->FatalError("can't find DNS entry for %s (req type %d / %s) in cache", name.c_str(), request_type,
|
|
request_type_string(request_type));
|
|
return nullptr;
|
|
|
|
case DNS_DEFAULT: {
|
|
auto req = new DNS_Request(name, request_type);
|
|
req->MakeRequest(channel, this);
|
|
Resolve();
|
|
|
|
// Call LookupHost() a second time to get the newly stored value out of the cache.
|
|
return Lookup(name, request_type);
|
|
}
|
|
|
|
default: reporter->InternalError("bad mode %d in DNS_Mgr::Lookup", mode); return nullptr;
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
TableValPtr DNS_Mgr::LookupHost(const std::string& name) {
|
|
if ( shutting_down )
|
|
return nullptr;
|
|
|
|
if ( mode == DNS_FAKE )
|
|
return fake_name_lookup_result(name);
|
|
|
|
InitSource();
|
|
|
|
// Check the cache before attempting to look up the name remotely.
|
|
if ( mode != DNS_PRIME ) {
|
|
if ( auto val = LookupNameInCache(name, false, true) )
|
|
return val;
|
|
}
|
|
|
|
// Not found, or priming.
|
|
switch ( mode ) {
|
|
case DNS_PRIME: {
|
|
// We pass T_A here, but DNSRequest::MakeRequest() will special-case that in
|
|
// a request that gets both T_A and T_AAAA results at one time.
|
|
auto req = new DNS_Request(name, T_A);
|
|
req->MakeRequest(channel, this);
|
|
return empty_addr_set();
|
|
}
|
|
|
|
case DNS_FORCE: reporter->FatalError("can't find DNS entry for %s in cache", name.c_str()); return nullptr;
|
|
|
|
case DNS_DEFAULT: {
|
|
// We pass T_A here, but DNSRequest::MakeRequest() will special-case that in
|
|
// a request that gets both T_A and T_AAAA results at one time.
|
|
auto req = new DNS_Request(name, T_A);
|
|
req->MakeRequest(channel, this);
|
|
Resolve();
|
|
|
|
// Call LookupHost() a second time to get the newly stored value out of the cache.
|
|
return LookupHost(name);
|
|
}
|
|
|
|
default: reporter->InternalError("bad mode in DNS_Mgr::LookupHost"); return nullptr;
|
|
}
|
|
}
|
|
|
|
StringValPtr DNS_Mgr::LookupAddr(const IPAddr& addr) {
|
|
if ( shutting_down )
|
|
return nullptr;
|
|
|
|
if ( mode == DNS_FAKE )
|
|
return make_intrusive<StringVal>(fake_addr_lookup_result(addr));
|
|
|
|
InitSource();
|
|
|
|
// Check the cache before attempting to look up the name remotely.
|
|
if ( mode != DNS_PRIME ) {
|
|
if ( auto val = LookupAddrInCache(addr, false, true) )
|
|
return val;
|
|
}
|
|
|
|
// Not found, or priming.
|
|
switch ( mode ) {
|
|
case DNS_PRIME: {
|
|
auto req = new DNS_Request(addr);
|
|
req->MakeRequest(channel, this);
|
|
return make_intrusive<StringVal>("<none>");
|
|
}
|
|
|
|
case DNS_FORCE:
|
|
reporter->FatalError("can't find DNS entry for %s in cache", addr.AsString().c_str());
|
|
return nullptr;
|
|
|
|
case DNS_DEFAULT: {
|
|
auto req = new DNS_Request(addr);
|
|
req->MakeRequest(channel, this);
|
|
Resolve();
|
|
|
|
// Call LookupAddr() a second time to get the newly stored value out of the cache.
|
|
return LookupAddr(addr);
|
|
}
|
|
|
|
default: reporter->InternalError("bad mode in DNS_Mgr::LookupAddr"); return nullptr;
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::LookupHost(const std::string& name, LookupCallback* callback) {
|
|
if ( shutting_down )
|
|
return;
|
|
|
|
if ( mode == DNS_FAKE ) {
|
|
resolve_lookup_cb(callback, fake_name_lookup_result(name));
|
|
return;
|
|
}
|
|
|
|
// Do we already know the answer?
|
|
if ( auto addrs = LookupNameInCache(name, true, false) ) {
|
|
resolve_lookup_cb(callback, std::move(addrs));
|
|
return;
|
|
}
|
|
|
|
AsyncRequest* req = nullptr;
|
|
|
|
// If we already have a request waiting for this host, we don't need to make
|
|
// another one. We can just add the callback to it and it'll get handled
|
|
// when the first request comes back.
|
|
auto key = std::make_pair(T_A, name);
|
|
auto i = asyncs.find(key);
|
|
if ( i != asyncs.end() )
|
|
req = i->second;
|
|
else {
|
|
// A new one.
|
|
req = new AsyncRequest{name, T_A};
|
|
asyncs_queued.push_back(req);
|
|
asyncs.emplace_hint(i, std::move(key), req);
|
|
}
|
|
|
|
req->callbacks.push_back(callback);
|
|
|
|
// There may be requests in the queue that haven't been processed yet
|
|
// so go ahead and reissue them, even if this method didn't change
|
|
// anything.
|
|
IssueAsyncRequests();
|
|
}
|
|
|
|
void DNS_Mgr::LookupAddr(const IPAddr& addr, LookupCallback* callback) {
|
|
if ( shutting_down )
|
|
return;
|
|
|
|
if ( mode == DNS_FAKE ) {
|
|
resolve_lookup_cb(callback, fake_addr_lookup_result(addr));
|
|
return;
|
|
}
|
|
|
|
// Do we already know the answer?
|
|
if ( auto name = LookupAddrInCache(addr, true, false) ) {
|
|
resolve_lookup_cb(callback, name->CheckString());
|
|
return;
|
|
}
|
|
|
|
AsyncRequest* req = nullptr;
|
|
|
|
// If we already have a request waiting for this host, we don't need to make
|
|
// another one. We can just add the callback to it and it'll get handled
|
|
// when the first request comes back.
|
|
auto i = asyncs.find(addr);
|
|
if ( i != asyncs.end() )
|
|
req = i->second;
|
|
else {
|
|
// A new one.
|
|
req = new AsyncRequest{addr};
|
|
asyncs_queued.push_back(req);
|
|
asyncs.emplace_hint(i, addr, req);
|
|
}
|
|
|
|
req->callbacks.push_back(callback);
|
|
|
|
// There may be requests in the queue that haven't been processed yet
|
|
// so go ahead and reissue them, even if this method didn't change
|
|
// anything.
|
|
IssueAsyncRequests();
|
|
}
|
|
|
|
void DNS_Mgr::Lookup(const std::string& name, int request_type, LookupCallback* callback) {
|
|
if ( shutting_down )
|
|
return;
|
|
|
|
if ( mode == DNS_FAKE ) {
|
|
resolve_lookup_cb(callback, fake_lookup_result(name, request_type));
|
|
return;
|
|
}
|
|
|
|
// Do we already know the answer?
|
|
if ( auto txt = LookupOtherInCache(name, request_type, true) ) {
|
|
resolve_lookup_cb(callback, txt->CheckString());
|
|
return;
|
|
}
|
|
|
|
AsyncRequest* req = nullptr;
|
|
|
|
// If we already have a request waiting for this host, we don't need to make
|
|
// another one. We can just add the callback to it and it'll get handled
|
|
// when the first request comes back.
|
|
auto key = std::make_pair(request_type, name);
|
|
auto i = asyncs.find(key);
|
|
if ( i != asyncs.end() )
|
|
req = i->second;
|
|
else {
|
|
// A new one.
|
|
req = new AsyncRequest{name, request_type};
|
|
asyncs_queued.push_back(req);
|
|
asyncs.emplace_hint(i, std::move(key), req);
|
|
}
|
|
|
|
req->callbacks.push_back(callback);
|
|
|
|
IssueAsyncRequests();
|
|
}
|
|
|
|
void DNS_Mgr::Resolve() {
|
|
int nfds = 0;
|
|
struct pollfd pollfds[1024];
|
|
|
|
struct timeval tv;
|
|
tv.tv_sec = DNS_TIMEOUT;
|
|
tv.tv_usec = 0;
|
|
|
|
for ( int i = 0; i < MAX_PENDING_REQUESTS; i++ ) {
|
|
if ( socket_fds.empty() && write_socket_fds.empty() )
|
|
break;
|
|
|
|
memset(pollfds, 0, sizeof(pollfd) * 1024);
|
|
|
|
for ( int fd : socket_fds ) {
|
|
if ( nfds == 1024 )
|
|
break;
|
|
|
|
pollfds[nfds].fd = fd;
|
|
pollfds[nfds].events = POLLIN;
|
|
++nfds;
|
|
}
|
|
|
|
for ( int fd : write_socket_fds ) {
|
|
if ( nfds == 1024 )
|
|
break;
|
|
|
|
pollfds[nfds].fd = fd;
|
|
pollfds[nfds].events = POLLIN | POLLOUT;
|
|
++nfds;
|
|
}
|
|
|
|
// poll() timeout is in milliseconds.
|
|
struct timeval* tvp = ares_timeout(channel, &tv, &tv);
|
|
int timeout_ms = tvp->tv_sec * 1000 + tvp->tv_usec / 1000;
|
|
|
|
int res = poll(pollfds, nfds, timeout_ms);
|
|
|
|
if ( res > 0 ) {
|
|
for ( int i = 0; i < nfds; i++ ) {
|
|
int rdfd = pollfds[i].revents & POLLIN ? pollfds[i].fd : ARES_SOCKET_BAD;
|
|
int wrfd = pollfds[i].revents & POLLOUT ? pollfds[i].fd : ARES_SOCKET_BAD;
|
|
|
|
if ( rdfd != ARES_SOCKET_BAD || wrfd != ARES_SOCKET_BAD )
|
|
ares_process_fd(channel, rdfd, wrfd);
|
|
}
|
|
}
|
|
else if ( res == 0 )
|
|
// Do timeout processing when poll() timed out.
|
|
ares_process_fd(channel, ARES_SOCKET_BAD, ARES_SOCKET_BAD);
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::Event(EventHandlerPtr e, const DNS_MappingPtr& dm) {
|
|
if ( e )
|
|
event_mgr.Enqueue(e, BuildMappingVal(dm));
|
|
}
|
|
|
|
void DNS_Mgr::Event(EventHandlerPtr e, const DNS_MappingPtr& dm, ListValPtr l1, ListValPtr l2) {
|
|
if ( e )
|
|
event_mgr.Enqueue(e, BuildMappingVal(dm), l1->ToSetVal(), l2->ToSetVal());
|
|
}
|
|
|
|
void DNS_Mgr::Event(EventHandlerPtr e, const DNS_MappingPtr& old_dm, DNS_MappingPtr new_dm) {
|
|
if ( e )
|
|
event_mgr.Enqueue(e, BuildMappingVal(old_dm), BuildMappingVal(new_dm));
|
|
}
|
|
|
|
RecordValPtr DNS_Mgr::BuildMappingVal(const DNS_MappingPtr& dm) {
|
|
if ( ! dm_rec )
|
|
return nullptr;
|
|
|
|
auto r = make_intrusive<RecordVal>(dm_rec);
|
|
|
|
r->AssignTime(0, dm->CreationTime());
|
|
r->Assign(1, dm->ReqHost() ? dm->ReqHost() : "");
|
|
r->Assign(2, make_intrusive<AddrVal>(dm->ReqAddr()));
|
|
r->Assign(3, dm->Valid());
|
|
|
|
auto h = dm->Host();
|
|
r->Assign(4, h ? std::move(h) : make_intrusive<StringVal>("<none>"));
|
|
r->Assign(5, dm->AddrsSet());
|
|
|
|
return r;
|
|
}
|
|
|
|
void DNS_Mgr::AddResult(DNS_Request* dr, struct hostent* h, uint32_t ttl, bool merge) {
|
|
// TODO: Should we handle hostname aliases here somehow?
|
|
|
|
DNS_MappingPtr new_mapping = nullptr;
|
|
DNS_MappingPtr prev_mapping = nullptr;
|
|
bool keep_prev = true;
|
|
|
|
MappingMap::iterator it;
|
|
if ( dr->RequestType() == T_PTR ) {
|
|
new_mapping = std::make_shared<DNS_Mapping>(dr->Addr(), h, ttl);
|
|
it = all_mappings.find(dr->Addr());
|
|
if ( it == all_mappings.end() ) {
|
|
auto result = all_mappings.emplace(dr->Addr(), new_mapping);
|
|
it = result.first;
|
|
}
|
|
else
|
|
prev_mapping = it->second;
|
|
}
|
|
else {
|
|
new_mapping = std::make_shared<DNS_Mapping>(dr->Host(), h, ttl, dr->RequestType());
|
|
auto key = std::make_pair(dr->RequestType(), dr->Host());
|
|
|
|
it = all_mappings.find(key);
|
|
if ( it == all_mappings.end() ) {
|
|
auto result = all_mappings.emplace(std::move(key), new_mapping);
|
|
it = result.first;
|
|
}
|
|
else
|
|
prev_mapping = it->second;
|
|
}
|
|
|
|
if ( prev_mapping && prev_mapping->Valid() ) {
|
|
if ( new_mapping->Valid() ) {
|
|
if ( merge )
|
|
new_mapping->Merge(prev_mapping);
|
|
|
|
it->second = new_mapping;
|
|
keep_prev = false;
|
|
}
|
|
}
|
|
else {
|
|
it->second = new_mapping;
|
|
keep_prev = false;
|
|
}
|
|
|
|
if ( prev_mapping && ! dr->IsTxt() )
|
|
CompareMappings(prev_mapping, new_mapping);
|
|
|
|
if ( keep_prev )
|
|
new_mapping.reset();
|
|
else
|
|
prev_mapping.reset();
|
|
}
|
|
|
|
void DNS_Mgr::CompareMappings(const DNS_MappingPtr& prev_mapping, const DNS_MappingPtr& new_mapping) {
|
|
if ( prev_mapping->Failed() ) {
|
|
if ( new_mapping->Failed() )
|
|
// Nothing changed.
|
|
return;
|
|
|
|
Event(dns_mapping_valid, new_mapping);
|
|
return;
|
|
}
|
|
|
|
else if ( new_mapping->Failed() ) {
|
|
Event(dns_mapping_unverified, prev_mapping);
|
|
return;
|
|
}
|
|
|
|
auto prev_s = prev_mapping->Host();
|
|
auto new_s = new_mapping->Host();
|
|
|
|
if ( prev_s || new_s ) {
|
|
if ( ! prev_s )
|
|
Event(dns_mapping_new_name, new_mapping);
|
|
else if ( ! new_s )
|
|
Event(dns_mapping_lost_name, prev_mapping);
|
|
else if ( ! Bstr_eq(new_s->AsString(), prev_s->AsString()) )
|
|
Event(dns_mapping_name_changed, prev_mapping, new_mapping);
|
|
}
|
|
|
|
auto prev_a = prev_mapping->Addrs();
|
|
auto new_a = new_mapping->Addrs();
|
|
|
|
if ( ! prev_a || ! new_a ) {
|
|
reporter->InternalWarning("confused in DNS_Mgr::CompareMappings");
|
|
return;
|
|
}
|
|
|
|
auto prev_delta = AddrListDelta(prev_a, new_a);
|
|
auto new_delta = AddrListDelta(new_a, prev_a);
|
|
|
|
if ( prev_delta->Length() > 0 || new_delta->Length() > 0 )
|
|
Event(dns_mapping_altered, new_mapping, std::move(prev_delta), std::move(new_delta));
|
|
}
|
|
|
|
ListValPtr DNS_Mgr::AddrListDelta(ListValPtr al1, ListValPtr al2) {
|
|
auto delta = make_intrusive<ListVal>(TYPE_ADDR);
|
|
|
|
for ( int i = 0; i < al1->Length(); ++i ) {
|
|
const IPAddr& al1_i = al1->Idx(i)->AsAddr();
|
|
|
|
int j;
|
|
for ( j = 0; j < al2->Length(); ++j ) {
|
|
const IPAddr& al2_j = al2->Idx(j)->AsAddr();
|
|
if ( al1_i == al2_j )
|
|
break;
|
|
}
|
|
|
|
if ( j >= al2->Length() )
|
|
// Didn't find it.
|
|
delta->Append(al1->Idx(i));
|
|
}
|
|
|
|
return delta;
|
|
}
|
|
|
|
void DNS_Mgr::LoadCache(const std::string& path) {
|
|
FILE* f = fopen(path.c_str(), "r");
|
|
|
|
if ( ! f )
|
|
return;
|
|
|
|
if ( ! DNS_Mapping::ValidateCacheVersion(f) ) {
|
|
fclose(f);
|
|
return;
|
|
}
|
|
|
|
// Loop until we find a mapping that doesn't initialize correctly.
|
|
auto m = std::make_shared<DNS_Mapping>(f);
|
|
for ( ; ! m->NoMapping() && ! m->InitFailed(); m = std::make_shared<DNS_Mapping>(f) ) {
|
|
if ( m->ReqHost() )
|
|
all_mappings.insert_or_assign(std::make_pair(m->ReqType(), m->ReqHost()), m);
|
|
else
|
|
all_mappings.insert_or_assign(m->ReqAddr(), m);
|
|
}
|
|
|
|
if ( ! m->NoMapping() )
|
|
reporter->FatalError("DNS cache corrupted");
|
|
|
|
fclose(f);
|
|
}
|
|
|
|
bool DNS_Mgr::Save() {
|
|
if ( cache_name.empty() )
|
|
return false;
|
|
|
|
FILE* f = fopen(cache_name.c_str(), "w");
|
|
|
|
if ( ! f )
|
|
return false;
|
|
|
|
DNS_Mapping::InitializeCache(f);
|
|
Save(f, all_mappings);
|
|
|
|
fclose(f);
|
|
|
|
return true;
|
|
}
|
|
|
|
void DNS_Mgr::Save(FILE* f, const MappingMap& m) {
|
|
for ( const auto& [key, mapping] : m ) {
|
|
if ( mapping )
|
|
mapping->Save(f);
|
|
}
|
|
}
|
|
|
|
TableValPtr DNS_Mgr::LookupNameInCache(const std::string& name, bool cleanup_expired, bool check_failed) {
|
|
auto it = all_mappings.find(std::make_pair(T_A, name));
|
|
if ( it == all_mappings.end() )
|
|
return nullptr;
|
|
|
|
auto d = it->second;
|
|
|
|
if ( ! d || d->names.empty() )
|
|
return nullptr;
|
|
|
|
if ( cleanup_expired && (d && d->Expired()) ) {
|
|
all_mappings.erase(it);
|
|
|
|
// If the TTL is zero, we're immediately expiring the response. We don't want
|
|
// to return though because the response was valid for a brief moment in time.
|
|
if ( d->TTL() != 0 )
|
|
return nullptr;
|
|
}
|
|
|
|
if ( check_failed && (d && d->Failed()) ) {
|
|
reporter->Warning("Can't resolve host: %s", name.c_str());
|
|
return empty_addr_set();
|
|
}
|
|
|
|
return d->AddrsSet();
|
|
}
|
|
|
|
StringValPtr DNS_Mgr::LookupAddrInCache(const IPAddr& addr, bool cleanup_expired, bool check_failed) {
|
|
auto it = all_mappings.find(addr);
|
|
if ( it == all_mappings.end() )
|
|
return nullptr;
|
|
|
|
auto d = it->second;
|
|
|
|
if ( cleanup_expired && d->Expired() ) {
|
|
all_mappings.erase(it);
|
|
|
|
// If the TTL is zero, we're immediately expiring the response. We don't want
|
|
// to return though because the response was valid for a brief moment in time.
|
|
if ( d->TTL() != 0 )
|
|
return nullptr;
|
|
}
|
|
else if ( check_failed && d->Failed() ) {
|
|
std::string s(addr);
|
|
reporter->Warning("can't resolve IP address: %s", s.c_str());
|
|
return make_intrusive<StringVal>(s);
|
|
}
|
|
|
|
if ( d->Host() )
|
|
return d->Host();
|
|
|
|
return make_intrusive<StringVal>("<\?\?\?>");
|
|
}
|
|
|
|
StringValPtr DNS_Mgr::LookupOtherInCache(const std::string& name, int request_type, bool cleanup_expired) {
|
|
auto it = all_mappings.find(std::make_pair(request_type, name));
|
|
if ( it == all_mappings.end() )
|
|
return nullptr;
|
|
|
|
auto d = it->second;
|
|
|
|
if ( cleanup_expired && d->Expired() ) {
|
|
all_mappings.erase(it);
|
|
|
|
// If the TTL is zero, we're immediately expiring the response. We don't want
|
|
// to return though because the response was valid for a brief moment in time.
|
|
if ( d->TTL() != 0 )
|
|
return nullptr;
|
|
}
|
|
|
|
if ( d->Host() )
|
|
return d->Host();
|
|
|
|
return make_intrusive<StringVal>("<\?\?\?>");
|
|
}
|
|
|
|
void DNS_Mgr::IssueAsyncRequests() {
|
|
while ( ! asyncs_queued.empty() && asyncs_pending < MAX_PENDING_REQUESTS ) {
|
|
DNS_Request* dns_req = nullptr;
|
|
AsyncRequest* req = asyncs_queued.front();
|
|
asyncs_queued.pop_front();
|
|
|
|
num_requests_metric->Inc();
|
|
req->time = util::current_time();
|
|
|
|
if ( req->type == T_PTR )
|
|
dns_req = new DNS_Request(req->addr, true);
|
|
else if ( req->type == T_A || req->type == T_AAAA )
|
|
// We pass T_A here, but DNSRequest::MakeRequest() will special-case that in
|
|
// a request that gets both T_A and T_AAAA results at one time.
|
|
dns_req = new DNS_Request(req->host.c_str(), T_A, true);
|
|
else
|
|
dns_req = new DNS_Request(req->host.c_str(), req->type, true);
|
|
|
|
dns_req->MakeRequest(channel, this);
|
|
|
|
++asyncs_pending;
|
|
asyncs_pending_metric->Inc();
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::CheckAsyncHostRequest(const std::string& host, bool timeout) {
|
|
// Note that this code is a mirror of that for CheckAsyncAddrRequest.
|
|
auto i = asyncs.find(std::make_pair(T_A, host));
|
|
|
|
if ( i != asyncs.end() ) {
|
|
if ( timeout ) {
|
|
failed_metric->Inc();
|
|
i->second->Timeout();
|
|
}
|
|
else if ( auto addrs = LookupNameInCache(host, true, false) ) {
|
|
successful_metric->Inc();
|
|
i->second->Resolved(addrs);
|
|
}
|
|
else
|
|
return;
|
|
|
|
delete i->second;
|
|
asyncs.erase(i);
|
|
--asyncs_pending;
|
|
asyncs_pending_metric->Dec();
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::CheckAsyncAddrRequest(const IPAddr& addr, bool timeout) {
|
|
// Note that this code is a mirror of that for CheckAsyncHostRequest.
|
|
|
|
// In the following, if it's not in the respective map anymore, we've
|
|
// already finished it earlier and don't have anything to do.
|
|
auto i = asyncs.find(addr);
|
|
|
|
if ( i != asyncs.end() ) {
|
|
if ( timeout ) {
|
|
failed_metric->Inc();
|
|
i->second->Timeout();
|
|
}
|
|
else if ( auto name = LookupAddrInCache(addr, true, false) ) {
|
|
successful_metric->Inc();
|
|
i->second->Resolved(name->CheckString());
|
|
}
|
|
else
|
|
return;
|
|
|
|
delete i->second;
|
|
asyncs.erase(i);
|
|
--asyncs_pending;
|
|
asyncs_pending_metric->Dec();
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::CheckAsyncOtherRequest(const std::string& host, bool timeout, int request_type) {
|
|
// Note that this code is a mirror of that for CheckAsyncAddrRequest.
|
|
|
|
auto i = asyncs.find(std::make_pair(request_type, host));
|
|
if ( i != asyncs.end() ) {
|
|
if ( timeout ) {
|
|
failed_metric->Inc();
|
|
i->second->Timeout();
|
|
}
|
|
else if ( auto name = LookupOtherInCache(host, request_type, true) ) {
|
|
successful_metric->Inc();
|
|
i->second->Resolved(name->CheckString());
|
|
}
|
|
else
|
|
return;
|
|
|
|
delete i->second;
|
|
asyncs.erase(i);
|
|
--asyncs_pending;
|
|
asyncs_pending_metric->Dec();
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::Flush() {
|
|
Resolve();
|
|
all_mappings.clear();
|
|
}
|
|
|
|
double DNS_Mgr::GetNextTimeout() {
|
|
if ( asyncs_pending == 0 )
|
|
return -1;
|
|
|
|
struct timeval tv;
|
|
struct timeval* tvp = ares_timeout(channel, nullptr, &tv);
|
|
|
|
// If you pass nullptr as the max time argument to ares_timeout, it will return null if there
|
|
// isn't anything waiting to be processed.
|
|
if ( ! tvp )
|
|
return -1;
|
|
|
|
// Clamp the timeout to our desired max, since we passed NULl to ares_timeout.
|
|
if ( tvp->tv_sec > DNS_TIMEOUT ) {
|
|
tvp->tv_sec = DNS_TIMEOUT;
|
|
tvp->tv_usec = 0;
|
|
}
|
|
|
|
return static_cast<double>(tvp->tv_sec) + (static_cast<double>(tvp->tv_usec) / 1e6);
|
|
}
|
|
|
|
void DNS_Mgr::ProcessFd(int fd, int flags) {
|
|
if ( socket_fds.contains(fd) ) {
|
|
int read_fd = (flags & IOSource::ProcessFlags::READ) != 0 ? fd : ARES_SOCKET_BAD;
|
|
int write_fd = (flags & IOSource::ProcessFlags::WRITE) != 0 ? fd : ARES_SOCKET_BAD;
|
|
ares_process_fd(channel, read_fd, write_fd);
|
|
}
|
|
|
|
IssueAsyncRequests();
|
|
}
|
|
|
|
void DNS_Mgr::Process() {
|
|
// Process() is called when DNS_Mgr is found "ready" when its
|
|
// GetNextTimeout() returns 0.0, but there's no active FD.
|
|
//
|
|
// Kick off timeouts at least.
|
|
ares_process_fd(channel, ARES_SOCKET_BAD, ARES_SOCKET_BAD);
|
|
}
|
|
|
|
void DNS_Mgr::UpdateCachedStats(bool force) {
|
|
double now = util::current_time();
|
|
if ( force || last_cached_stats_update < now - 0.01 ) {
|
|
last_cached_stats.hosts = 0;
|
|
last_cached_stats.addresses = 0;
|
|
last_cached_stats.texts = 0;
|
|
last_cached_stats.total = all_mappings.size();
|
|
|
|
for ( const auto& [key, mapping] : all_mappings ) {
|
|
if ( mapping->ReqType() == T_PTR )
|
|
last_cached_stats.addresses++;
|
|
else if ( mapping->ReqType() == T_A )
|
|
last_cached_stats.hosts++;
|
|
else
|
|
last_cached_stats.texts++;
|
|
}
|
|
|
|
last_cached_stats_update = now;
|
|
}
|
|
}
|
|
|
|
void DNS_Mgr::GetStats(Stats* stats) {
|
|
stats->requests = static_cast<unsigned long>(num_requests_metric->Value());
|
|
stats->successful = static_cast<unsigned long>(successful_metric->Value());
|
|
stats->failed = static_cast<unsigned long>(failed_metric->Value());
|
|
stats->pending = asyncs_pending;
|
|
|
|
UpdateCachedStats(true);
|
|
stats->cached = last_cached_stats;
|
|
}
|
|
|
|
void DNS_Mgr::AsyncRequest::Resolved(const std::string& name) {
|
|
for ( const auto& cb : callbacks ) {
|
|
cb->Resolved(name);
|
|
if ( ! doctest::is_running_in_test )
|
|
delete cb;
|
|
}
|
|
|
|
callbacks.clear();
|
|
processed = true;
|
|
}
|
|
|
|
void DNS_Mgr::AsyncRequest::Resolved(TableValPtr addrs) {
|
|
for ( const auto& cb : callbacks ) {
|
|
cb->Resolved(addrs);
|
|
if ( ! doctest::is_running_in_test )
|
|
delete cb;
|
|
}
|
|
|
|
callbacks.clear();
|
|
processed = true;
|
|
}
|
|
|
|
void DNS_Mgr::AsyncRequest::Timeout() {
|
|
for ( const auto& cb : callbacks ) {
|
|
cb->Timeout();
|
|
if ( ! doctest::is_running_in_test )
|
|
delete cb;
|
|
}
|
|
|
|
callbacks.clear();
|
|
processed = true;
|
|
}
|
|
|
|
TableValPtr DNS_Mgr::empty_addr_set() {
|
|
// TODO: can this be returned statically as well? Does the result get used in a way
|
|
// that would modify the same value being returned repeatedly?
|
|
auto addr_t = base_type(TYPE_ADDR);
|
|
auto set_index = make_intrusive<TypeList>(addr_t);
|
|
set_index->Append(std::move(addr_t));
|
|
auto s = make_intrusive<SetType>(std::move(set_index), nullptr);
|
|
return make_intrusive<TableVal>(std::move(s));
|
|
}
|
|
|
|
DNS_Mgr::AsyncRequest::AsyncRequest(const IPAddr& addr) : addr(addr), type(T_PTR) {}
|
|
|
|
//////////////////////////////////////////////////////////////////////////////////////////
|
|
//////////////////////////////////////////////////////////////////////////////////////////
|
|
//////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// Unit testing coverage for the DNS_Mgr code, including making actual DNS requests to
|
|
// test responses and timeouts. Note that all of these tests are marked with the skip
|
|
// decorator, since they take some time to run and this slows down local development. To
|
|
// run them manually, pass the --no-skip flag when running tests. These tests are
|
|
// run automatically as part of CI builds.
|
|
|
|
static std::vector<IPAddr> get_result_addresses(TableValPtr addrs) {
|
|
std::vector<IPAddr> results;
|
|
|
|
auto m = addrs->ToMap();
|
|
for ( const auto& [k, v] : m ) {
|
|
auto lv = cast_intrusive<ListVal>(k);
|
|
auto lvv = lv->Vals();
|
|
for ( const auto& addr : lvv ) {
|
|
auto addr_ptr = cast_intrusive<AddrVal>(addr);
|
|
results.push_back(addr_ptr->Get());
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
class TestCallback : public DNS_Mgr::LookupCallback {
|
|
public:
|
|
TestCallback() {}
|
|
void Resolved(const std::string& name) override {
|
|
host_result = name;
|
|
done = true;
|
|
}
|
|
void Resolved(TableValPtr addrs) override {
|
|
addr_results = get_result_addresses(addrs);
|
|
done = true;
|
|
}
|
|
void Timeout() override {
|
|
timeout = true;
|
|
done = true;
|
|
}
|
|
|
|
std::string host_result;
|
|
std::vector<IPAddr> addr_results;
|
|
bool done = false;
|
|
bool timeout = false;
|
|
};
|
|
|
|
/**
|
|
* Derived testing version of DNS_Mgr so that the Process() method can be exposed
|
|
* publicly. If new unit tests are added, this class should be used over using
|
|
* DNS_Mgr directly.
|
|
*/
|
|
class TestDNS_Mgr final : public DNS_Mgr {
|
|
public:
|
|
explicit TestDNS_Mgr(DNS_MgrMode mode) : DNS_Mgr(mode) {}
|
|
void Process() override;
|
|
};
|
|
|
|
void TestDNS_Mgr::Process() {
|
|
// Only allow usage of this method when running unit tests.
|
|
assert(doctest::is_running_in_test);
|
|
Resolve();
|
|
IssueAsyncRequests();
|
|
}
|
|
|
|
TEST_CASE("dns_mgr priming" * doctest::skip(true)) {
|
|
// TODO: This test uses mkdtemp, which isn't available on Windows.
|
|
#ifndef _MSC_VER
|
|
char prefix[] = "/tmp/zeek-unit-test-XXXXXX";
|
|
auto tmpdir = mkdtemp(prefix);
|
|
|
|
// Create a manager to prime the cache, make a few requests, and the save
|
|
// the result. This tests that the priming code will create the requests but
|
|
// wait for Resolve() to actually make the requests.
|
|
TestDNS_Mgr mgr(DNS_PRIME);
|
|
mgr.SetDir(tmpdir);
|
|
mgr.InitPostScript();
|
|
|
|
auto host_result = mgr.LookupHost("one.one.one.one");
|
|
REQUIRE(host_result != nullptr);
|
|
CHECK(host_result->EqualTo(TestDNS_Mgr::empty_addr_set()));
|
|
|
|
IPAddr ones("1.1.1.1");
|
|
auto addr_result = mgr.LookupAddr(ones);
|
|
CHECK(strcmp(addr_result->CheckString(), "<none>") == 0);
|
|
|
|
// This should wait until we have all of the results back from the above
|
|
// requests.
|
|
mgr.Resolve();
|
|
|
|
// Save off the resulting values from Resolve() into a file on disk
|
|
// in the tmpdir created by mkdtemp.
|
|
REQUIRE(mgr.Save());
|
|
|
|
// Make a second DNS manager and reload the cache that we just saved.
|
|
TestDNS_Mgr mgr2(DNS_FORCE);
|
|
dns_mgr = &mgr2;
|
|
mgr2.SetDir(tmpdir);
|
|
mgr2.InitPostScript();
|
|
|
|
// Make the same two requests, but verify that we're correctly getting
|
|
// data out of the cache.
|
|
host_result = mgr2.LookupHost("one.one.one.one");
|
|
REQUIRE(host_result != nullptr);
|
|
CHECK_FALSE(host_result->EqualTo(TestDNS_Mgr::empty_addr_set()));
|
|
|
|
addr_result = mgr2.LookupAddr(ones);
|
|
REQUIRE(addr_result != nullptr);
|
|
CHECK(strcmp(addr_result->CheckString(), "one.one.one.one") == 0);
|
|
|
|
// Clean up cache file and the temp directory
|
|
unlink(mgr2.CacheFile().c_str());
|
|
rmdir(tmpdir);
|
|
#endif
|
|
}
|
|
|
|
TEST_CASE("dns_mgr alternate server" * doctest::skip(true)) {
|
|
char* old_server = getenv("ZEEK_DNS_RESOLVER");
|
|
|
|
setenv("ZEEK_DNS_RESOLVER", "1.1.1.1", 1);
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
|
|
mgr.InitPostScript();
|
|
|
|
auto result = mgr.LookupAddr("1.1.1.1");
|
|
REQUIRE(result != nullptr);
|
|
CHECK(strcmp(result->CheckString(), "one.one.one.one") == 0);
|
|
|
|
// FIXME: This won't run on systems without IPv6 connectivity.
|
|
// setenv("ZEEK_DNS_RESOLVER", "2606:4700:4700::1111", 1);
|
|
// TestDNS_Mgr mgr2(DNS_DEFAULT, true);
|
|
// mgr2.InitPostScript();
|
|
// result = mgr2.LookupAddr("1.1.1.1");
|
|
// mgr2.Resolve();
|
|
|
|
// result = mgr2.LookupAddr("1.1.1.1");
|
|
// CHECK(strcmp(result->CheckString(), "one.one.one.one") == 0);
|
|
|
|
if ( old_server )
|
|
setenv("ZEEK_DNS_RESOLVER", old_server, 1);
|
|
else
|
|
unsetenv("ZEEK_DNS_RESOLVER");
|
|
}
|
|
|
|
TEST_CASE("dns_mgr default mode" * doctest::skip(true)) {
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
mgr.InitPostScript();
|
|
|
|
IPAddr ones4("1.1.1.1");
|
|
IPAddr ones6("2606:4700:4700::1111");
|
|
|
|
auto host_result = mgr.LookupHost("one.one.one.one");
|
|
REQUIRE(host_result != nullptr);
|
|
CHECK_FALSE(host_result->EqualTo(TestDNS_Mgr::empty_addr_set()));
|
|
|
|
auto addrs_from_request = get_result_addresses(host_result);
|
|
auto it = std::ranges::find(addrs_from_request, ones4);
|
|
CHECK(it != addrs_from_request.end());
|
|
it = std::ranges::find(addrs_from_request, ones6);
|
|
CHECK(it != addrs_from_request.end());
|
|
|
|
auto addr_result = mgr.LookupAddr(ones4);
|
|
REQUIRE(addr_result != nullptr);
|
|
CHECK(strcmp(addr_result->CheckString(), "one.one.one.one") == 0);
|
|
|
|
addr_result = mgr.LookupAddr(ones6);
|
|
REQUIRE(addr_result != nullptr);
|
|
CHECK(strcmp(addr_result->CheckString(), "one.one.one.one") == 0);
|
|
|
|
IPAddr bad("240.0.0.0");
|
|
addr_result = mgr.LookupAddr(bad);
|
|
REQUIRE(addr_result != nullptr);
|
|
CHECK(strcmp(addr_result->CheckString(), "240.0.0.0") == 0);
|
|
}
|
|
|
|
TEST_CASE("dns_mgr async host" * doctest::skip(true)) {
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
mgr.InitPostScript();
|
|
|
|
TestCallback cb{};
|
|
mgr.LookupHost("one.one.one.one", &cb);
|
|
|
|
// This shouldn't take any longer than DNS_TIMEOUT+1 seconds, so bound it
|
|
// just in case of some failure we're not aware of yet.
|
|
int count = 0;
|
|
while ( ! cb.done && (count < DNS_TIMEOUT + 1) ) {
|
|
mgr.Process();
|
|
sleep(1);
|
|
if ( ! cb.timeout )
|
|
count++;
|
|
}
|
|
|
|
REQUIRE(count < (DNS_TIMEOUT + 1));
|
|
if ( ! cb.timeout ) {
|
|
REQUIRE_FALSE(cb.addr_results.empty());
|
|
IPAddr ones("1.1.1.1");
|
|
auto it = std::ranges::find(cb.addr_results, ones);
|
|
CHECK(it != cb.addr_results.end());
|
|
}
|
|
|
|
mgr.Flush();
|
|
}
|
|
|
|
TEST_CASE("dns_mgr async addr" * doctest::skip(true)) {
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
mgr.InitPostScript();
|
|
|
|
TestCallback cb{};
|
|
mgr.LookupAddr(IPAddr{"1.1.1.1"}, &cb);
|
|
|
|
// This shouldn't take any longer than DNS_TIMEOUT +1 seconds, so bound it
|
|
// just in case of some failure we're not aware of yet.
|
|
int count = 0;
|
|
while ( ! cb.done && (count < DNS_TIMEOUT + 1) ) {
|
|
mgr.Process();
|
|
sleep(1);
|
|
if ( ! cb.timeout )
|
|
count++;
|
|
}
|
|
|
|
REQUIRE(count < (DNS_TIMEOUT + 1));
|
|
if ( ! cb.timeout )
|
|
REQUIRE(cb.host_result == "one.one.one.one");
|
|
|
|
mgr.Flush();
|
|
}
|
|
|
|
TEST_CASE("dns_mgr async text" * doctest::skip(true)) {
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
mgr.InitPostScript();
|
|
|
|
TestCallback cb{};
|
|
mgr.Lookup("unittest.zeek.org", T_TXT, &cb);
|
|
|
|
// This shouldn't take any longer than DNS_TIMEOUT +1 seconds, so bound it
|
|
// just in case of some failure we're not aware of yet.
|
|
int count = 0;
|
|
while ( ! cb.done && (count < DNS_TIMEOUT + 1) ) {
|
|
mgr.Process();
|
|
sleep(1);
|
|
if ( ! cb.timeout )
|
|
count++;
|
|
}
|
|
|
|
REQUIRE(count < (DNS_TIMEOUT + 1));
|
|
if ( ! cb.timeout )
|
|
REQUIRE(cb.host_result == "testing dns_mgr");
|
|
|
|
mgr.Flush();
|
|
}
|
|
|
|
TEST_CASE("dns_mgr timeouts" * doctest::skip(true)) {
|
|
char* old_server = getenv("ZEEK_DNS_RESOLVER");
|
|
|
|
// This is the address for blackhole.webpagetest.org, which provides a DNS
|
|
// server that lets you connect but never returns any responses, always
|
|
// resulting in a timeout.
|
|
setenv("ZEEK_DNS_RESOLVER", "3.219.212.117", 1);
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
|
|
mgr.InitPostScript();
|
|
auto addr_result = mgr.LookupAddr("1.1.1.1");
|
|
REQUIRE(addr_result != nullptr);
|
|
CHECK(strcmp(addr_result->CheckString(), "1.1.1.1") == 0);
|
|
|
|
auto host_result = mgr.LookupHost("one.one.one.one");
|
|
REQUIRE(host_result != nullptr);
|
|
auto addresses = get_result_addresses(host_result);
|
|
CHECK(addresses.size() == 0);
|
|
|
|
if ( old_server )
|
|
setenv("ZEEK_DNS_RESOLVER", old_server, 1);
|
|
else
|
|
unsetenv("ZEEK_DNS_RESOLVER");
|
|
}
|
|
|
|
TEST_CASE("dns_mgr async timeouts" * doctest::skip(true)) {
|
|
char* old_server = getenv("ZEEK_DNS_RESOLVER");
|
|
|
|
// This is the address for blackhole.webpagetest.org, which provides a DNS
|
|
// server that lets you connect but never returns any responses, always
|
|
// resulting in a timeout.
|
|
setenv("ZEEK_DNS_RESOLVER", "3.219.212.117", 1);
|
|
TestDNS_Mgr mgr(DNS_DEFAULT);
|
|
mgr.InitPostScript();
|
|
|
|
TestCallback cb{};
|
|
mgr.Lookup("unittest.zeek.org", T_TXT, &cb);
|
|
|
|
// This shouldn't take any longer than DNS_TIMEOUT +1 seconds, so bound it
|
|
// just in case of some failure we're not aware of yet.
|
|
int count = 0;
|
|
while ( ! cb.done && (count < DNS_TIMEOUT + 1) ) {
|
|
mgr.Process();
|
|
sleep(1);
|
|
if ( ! cb.timeout )
|
|
count++;
|
|
}
|
|
|
|
REQUIRE(count < (DNS_TIMEOUT + 1));
|
|
CHECK(cb.timeout);
|
|
|
|
mgr.Flush();
|
|
|
|
if ( old_server )
|
|
setenv("ZEEK_DNS_RESOLVER", old_server, 1);
|
|
else
|
|
unsetenv("ZEEK_DNS_RESOLVER");
|
|
}
|
|
|
|
} // namespace zeek::detail
|