xref: /freebsd/contrib/unbound/services/listen_dnsport.c (revision 1fb9c5ffe25fcba0857c4fdf6ed19ca5bf6bc858)
1 /*
2  * services/listen_dnsport.c - listen on port 53 for incoming DNS queries.
3  *
4  * Copyright (c) 2007, NLnet Labs. All rights reserved.
5  *
6  * This software is open source.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * Redistributions of source code must retain the above copyright notice,
13  * this list of conditions and the following disclaimer.
14  *
15  * Redistributions in binary form must reproduce the above copyright notice,
16  * this list of conditions and the following disclaimer in the documentation
17  * and/or other materials provided with the distribution.
18  *
19  * Neither the name of the NLNET LABS nor the names of its contributors may
20  * be used to endorse or promote products derived from this software without
21  * specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 /**
37  * \file
38  *
39  * This file has functions to get queries from clients.
40  */
41 #include "config.h"
42 #ifdef HAVE_SYS_TYPES_H
43 #  include <sys/types.h>
44 #endif
45 #include <limits.h>
46 #ifdef USE_TCP_FASTOPEN
47 #include <netinet/tcp.h>
48 #endif
49 #include <ctype.h>
50 #include "services/listen_dnsport.h"
51 #include "services/outside_network.h"
52 #include "util/netevent.h"
53 #include "util/log.h"
54 #include "util/config_file.h"
55 #include "util/net_help.h"
56 #include "sldns/sbuffer.h"
57 #include "sldns/parseutil.h"
58 #include "sldns/wire2str.h"
59 #include "services/mesh.h"
60 #include "util/fptr_wlist.h"
61 #include "util/locks.h"
62 #include "util/timeval_func.h"
63 
64 #ifdef HAVE_NETDB_H
65 #include <netdb.h>
66 #endif
67 #include <fcntl.h>
68 
69 #ifdef HAVE_SYS_UN_H
70 #include <sys/un.h>
71 #endif
72 
73 #ifdef HAVE_SYSTEMD
74 #include <systemd/sd-daemon.h>
75 #endif
76 
77 #ifdef HAVE_IFADDRS_H
78 #include <ifaddrs.h>
79 #endif
80 #ifdef HAVE_NET_IF_H
81 #include <net/if.h>
82 #endif
83 
84 #ifdef HAVE_TIME_H
85 #include <time.h>
86 #endif
87 #include <sys/time.h>
88 
89 #ifdef HAVE_NGTCP2
90 #include <ngtcp2/ngtcp2.h>
91 #include <ngtcp2/ngtcp2_crypto.h>
92 #ifdef HAVE_NGTCP2_NGTCP2_CRYPTO_OSSL_H
93 #include <ngtcp2/ngtcp2_crypto_ossl.h>
94 #elif defined(HAVE_NGTCP2_NGTCP2_CRYPTO_QUICTLS_H)
95 #include <ngtcp2/ngtcp2_crypto_quictls.h>
96 #elif defined(HAVE_NGTCP2_NGTCP2_CRYPTO_OPENSSL_H)
97 #include <ngtcp2/ngtcp2_crypto_openssl.h>
98 #define MAKE_QUIC_METHOD 1
99 #endif
100 #endif
101 
102 #ifdef HAVE_OPENSSL_SSL_H
103 #include <openssl/ssl.h>
104 #endif
105 
106 #ifdef HAVE_LINUX_NET_TSTAMP_H
107 #include <linux/net_tstamp.h>
108 #endif
109 
110 /** number of queued TCP connections for listen() */
111 #define TCP_BACKLOG 256
112 
113 #ifndef THREADS_DISABLED
114 /** lock on the counter of stream buffer memory */
115 static lock_basic_type stream_wait_count_lock;
116 /** lock on the counter of HTTP2 query buffer memory */
117 static lock_basic_type http2_query_buffer_count_lock;
118 /** lock on the counter of HTTP2 response buffer memory */
119 static lock_basic_type http2_response_buffer_count_lock;
120 #endif
121 /** size (in bytes) of stream wait buffers */
122 static size_t stream_wait_count = 0;
123 /** is the lock initialised for stream wait buffers */
124 static int stream_wait_lock_inited = 0;
125 /** size (in bytes) of HTTP2 query buffers */
126 static size_t http2_query_buffer_count = 0;
127 /** is the lock initialised for HTTP2 query buffers */
128 static int http2_query_buffer_lock_inited = 0;
129 /** size (in bytes) of HTTP2 response buffers */
130 static size_t http2_response_buffer_count = 0;
131 /** is the lock initialised for HTTP2 response buffers */
132 static int http2_response_buffer_lock_inited = 0;
133 
134 /**
135  * Debug print of the getaddrinfo returned address.
136  * @param addr: the address returned.
137  * @param additional: additional text that describes the type of socket,
138  * 	or NULL for no text.
139  */
140 static void
141 verbose_print_addr(struct addrinfo *addr, const char* additional)
142 {
143 	if(verbosity >= VERB_ALGO) {
144 		char buf[100];
145 		void* sinaddr = &((struct sockaddr_in*)addr->ai_addr)->sin_addr;
146 #ifdef INET6
147 		if(addr->ai_family == AF_INET6)
148 			sinaddr = &((struct sockaddr_in6*)addr->ai_addr)->
149 				sin6_addr;
150 #endif /* INET6 */
151 		if(inet_ntop(addr->ai_family, sinaddr, buf,
152 			(socklen_t)sizeof(buf)) == 0) {
153 			(void)strlcpy(buf, "(null)", sizeof(buf));
154 		}
155 		buf[sizeof(buf)-1] = 0;
156 		verbose(VERB_ALGO, "creating %s%s socket %s %d%s%s",
157 			addr->ai_socktype==SOCK_DGRAM?"udp":
158 			addr->ai_socktype==SOCK_STREAM?"tcp":"otherproto",
159 			addr->ai_family==AF_INET?"4":
160 			addr->ai_family==AF_INET6?"6":
161 			"_otherfam", buf,
162 			ntohs(((struct sockaddr_in*)addr->ai_addr)->sin_port),
163 			(additional?" ":""), (additional?additional:""));
164 	}
165 }
166 
167 void
168 verbose_print_unbound_socket(struct unbound_socket* ub_sock)
169 {
170 	if(verbosity >= VERB_ALGO) {
171 		char buf[256];
172 		log_info("listing of unbound_socket structure:");
173 		addr_to_str((void*)ub_sock->addr, ub_sock->addrlen, buf,
174 			sizeof(buf));
175 		log_info("%s s is: %d, fam is: %s, acl: %s", buf, ub_sock->s,
176 			ub_sock->fam == AF_INET?"AF_INET":"AF_INET6",
177 			ub_sock->acl?"yes":"no");
178 	}
179 }
180 
181 #ifdef HAVE_SYSTEMD
182 static int
183 systemd_get_activated(int family, int socktype, int listen,
184 		      struct sockaddr *addr, socklen_t addrlen,
185 		      const char *path)
186 {
187 	int i = 0;
188 	int r = 0;
189 	int s = -1;
190 	const char* listen_pid, *listen_fds;
191 
192 	/* We should use "listen" option only for stream protocols. For UDP it should be -1 */
193 
194 	if((r = sd_booted()) < 1) {
195 		if(r == 0)
196 			log_warn("systemd is not running");
197 		else
198 			log_err("systemd sd_booted(): %s", strerror(-r));
199 		return -1;
200 	}
201 
202 	listen_pid = getenv("LISTEN_PID");
203 	listen_fds = getenv("LISTEN_FDS");
204 
205 	if (!listen_pid) {
206 		log_warn("Systemd mandatory ENV variable is not defined: LISTEN_PID");
207 		return -1;
208 	}
209 
210 	if (!listen_fds) {
211 		log_warn("Systemd mandatory ENV variable is not defined: LISTEN_FDS");
212 		return -1;
213 	}
214 
215 	if((r = sd_listen_fds(0)) < 1) {
216 		if(r == 0)
217 			log_warn("systemd: did not return socket, check unit configuration");
218 		else
219 			log_err("systemd sd_listen_fds(): %s", strerror(-r));
220 		return -1;
221 	}
222 
223 	for(i = 0; i < r; i++) {
224 		if(sd_is_socket(SD_LISTEN_FDS_START + i, family, socktype, listen)) {
225 			s = SD_LISTEN_FDS_START + i;
226 			break;
227 		}
228 	}
229 	if (s == -1) {
230 		if (addr)
231 			log_err_addr("systemd sd_listen_fds()",
232 				     "no such socket",
233 				     (struct sockaddr_storage *)addr, addrlen);
234 		else
235 			log_err("systemd sd_listen_fds(): %s", path);
236 	}
237 	return s;
238 }
239 #endif
240 
241 int
242 create_udp_sock(int family, int socktype, struct sockaddr* addr,
243         socklen_t addrlen, int v6only, int* inuse, int* noproto,
244 	int rcv, int snd, int listen, int* reuseport, int transparent,
245 	int freebind, int use_systemd, int dscp)
246 {
247 	int s;
248 	char* err;
249 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_USE_MIN_MTU)  || defined(IP_TRANSPARENT) || defined(IP_BINDANY) || defined(IP_FREEBIND) || defined (SO_BINDANY)
250 	int on=1;
251 #endif
252 #ifdef IPV6_MTU
253 	int mtu = IPV6_MIN_MTU;
254 #endif
255 #if !defined(SO_RCVBUFFORCE) && !defined(SO_RCVBUF)
256 	(void)rcv;
257 #endif
258 #if !defined(SO_SNDBUFFORCE) && !defined(SO_SNDBUF)
259 	(void)snd;
260 #endif
261 #ifndef IPV6_V6ONLY
262 	(void)v6only;
263 #endif
264 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
265 	(void)transparent;
266 #endif
267 #if !defined(IP_FREEBIND)
268 	(void)freebind;
269 #endif
270 #ifdef HAVE_SYSTEMD
271 	int got_fd_from_systemd = 0;
272 
273 	if (!use_systemd
274 	    || (use_systemd
275 		&& (s = systemd_get_activated(family, socktype, -1, addr,
276 					      addrlen, NULL)) == -1)) {
277 #else
278 	(void)use_systemd;
279 #endif
280 	if((s = socket(family, socktype, 0)) == -1) {
281 		*inuse = 0;
282 #ifndef USE_WINSOCK
283 		if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
284 			*noproto = 1;
285 			return -1;
286 		}
287 #else
288 		if(WSAGetLastError() == WSAEAFNOSUPPORT ||
289 			WSAGetLastError() == WSAEPROTONOSUPPORT) {
290 			*noproto = 1;
291 			return -1;
292 		}
293 #endif
294 		log_err("can't create socket: %s", sock_strerror(errno));
295 		*noproto = 0;
296 		return -1;
297 	}
298 #ifdef HAVE_SYSTEMD
299 	} else {
300 		got_fd_from_systemd = 1;
301 	}
302 #endif
303 	if(listen) {
304 #ifdef SO_REUSEADDR
305 		if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
306 			(socklen_t)sizeof(on)) < 0) {
307 			log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
308 				sock_strerror(errno));
309 #ifndef USE_WINSOCK
310 			if(errno != ENOSYS) {
311 				close(s);
312 				*noproto = 0;
313 				*inuse = 0;
314 				return -1;
315 			}
316 #else
317 			closesocket(s);
318 			*noproto = 0;
319 			*inuse = 0;
320 			return -1;
321 #endif
322 		}
323 #endif /* SO_REUSEADDR */
324 #ifdef SO_REUSEPORT
325 #  ifdef SO_REUSEPORT_LB
326 		/* on FreeBSD 12 we have SO_REUSEPORT_LB that does loadbalance
327 		 * like SO_REUSEPORT on Linux.  This is what the users want
328 		 * with the config option in unbound.conf; if we actually
329 		 * need local address and port reuse they'll also need to
330 		 * have SO_REUSEPORT set for them, assume it was _LB they want.
331 		 */
332 		if (reuseport && *reuseport &&
333 		    setsockopt(s, SOL_SOCKET, SO_REUSEPORT_LB, (void*)&on,
334 			(socklen_t)sizeof(on)) < 0) {
335 #ifdef ENOPROTOOPT
336 			if(errno != ENOPROTOOPT || verbosity >= 3)
337 				log_warn("setsockopt(.. SO_REUSEPORT_LB ..) failed: %s",
338 					strerror(errno));
339 #endif
340 			/* this option is not essential, we can continue */
341 			*reuseport = 0;
342 		}
343 #  else /* no SO_REUSEPORT_LB */
344 
345 		/* try to set SO_REUSEPORT so that incoming
346 		 * queries are distributed evenly among the receiving threads.
347 		 * Each thread must have its own socket bound to the same port,
348 		 * with SO_REUSEPORT set on each socket.
349 		 */
350 		if (reuseport && *reuseport &&
351 		    setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
352 			(socklen_t)sizeof(on)) < 0) {
353 #ifdef ENOPROTOOPT
354 			if(errno != ENOPROTOOPT || verbosity >= 3)
355 				log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
356 					strerror(errno));
357 #endif
358 			/* this option is not essential, we can continue */
359 			*reuseport = 0;
360 		}
361 #  endif /* SO_REUSEPORT_LB */
362 #else
363 		(void)reuseport;
364 #endif /* defined(SO_REUSEPORT) */
365 #ifdef IP_TRANSPARENT
366 		if (transparent &&
367 		    setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
368 		    (socklen_t)sizeof(on)) < 0) {
369 			log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
370 			strerror(errno));
371 		}
372 #elif defined(IP_BINDANY)
373 		if (transparent &&
374 		    setsockopt(s, (family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
375 		    (family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
376 		    (void*)&on, (socklen_t)sizeof(on)) < 0) {
377 			log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
378 			(family==AF_INET6?"V6":""), strerror(errno));
379 		}
380 #elif defined(SO_BINDANY)
381 		if (transparent &&
382 		    setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on,
383 		    (socklen_t)sizeof(on)) < 0) {
384 			log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
385 			strerror(errno));
386 		}
387 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
388 	}
389 #ifdef IP_FREEBIND
390 	if(freebind &&
391 	    setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
392 	    (socklen_t)sizeof(on)) < 0) {
393 		log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
394 		strerror(errno));
395 	}
396 #endif /* IP_FREEBIND */
397 	if(rcv) {
398 #ifdef SO_RCVBUF
399 		int got;
400 		socklen_t slen = (socklen_t)sizeof(got);
401 #  ifdef SO_RCVBUFFORCE
402 		/* Linux specific: try to use root permission to override
403 		 * system limits on rcvbuf. The limit is stored in
404 		 * /proc/sys/net/core/rmem_max or sysctl net.core.rmem_max */
405 		if(setsockopt(s, SOL_SOCKET, SO_RCVBUFFORCE, (void*)&rcv,
406 			(socklen_t)sizeof(rcv)) < 0) {
407 			if(errno != EPERM) {
408 				log_err("setsockopt(..., SO_RCVBUFFORCE, "
409 					"...) failed: %s", sock_strerror(errno));
410 				sock_close(s);
411 				*noproto = 0;
412 				*inuse = 0;
413 				return -1;
414 			}
415 #  endif /* SO_RCVBUFFORCE */
416 			if(setsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&rcv,
417 				(socklen_t)sizeof(rcv)) < 0) {
418 				log_err("setsockopt(..., SO_RCVBUF, "
419 					"...) failed: %s", sock_strerror(errno));
420 				sock_close(s);
421 				*noproto = 0;
422 				*inuse = 0;
423 				return -1;
424 			}
425 			/* check if we got the right thing or if system
426 			 * reduced to some system max.  Warn if so */
427 			if(getsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&got,
428 				&slen) >= 0 && got < rcv/2) {
429 				log_warn("so-rcvbuf %u was not granted. "
430 					"Got %u. To fix: start with "
431 					"root permissions(linux) or sysctl "
432 					"bigger net.core.rmem_max(linux) or "
433 					"kern.ipc.maxsockbuf(bsd) values.",
434 					(unsigned)rcv, (unsigned)got);
435 			}
436 #  ifdef SO_RCVBUFFORCE
437 		}
438 #  endif
439 #endif /* SO_RCVBUF */
440 	}
441 	/* first do RCVBUF as the receive buffer is more important */
442 	if(snd) {
443 #ifdef SO_SNDBUF
444 		int got;
445 		socklen_t slen = (socklen_t)sizeof(got);
446 #  ifdef SO_SNDBUFFORCE
447 		/* Linux specific: try to use root permission to override
448 		 * system limits on sndbuf. The limit is stored in
449 		 * /proc/sys/net/core/wmem_max or sysctl net.core.wmem_max */
450 		if(setsockopt(s, SOL_SOCKET, SO_SNDBUFFORCE, (void*)&snd,
451 			(socklen_t)sizeof(snd)) < 0) {
452 			if(errno != EPERM && errno != ENOBUFS) {
453 				log_err("setsockopt(..., SO_SNDBUFFORCE, "
454 					"...) failed: %s", sock_strerror(errno));
455 				sock_close(s);
456 				*noproto = 0;
457 				*inuse = 0;
458 				return -1;
459 			}
460 			if(errno != EPERM) {
461 				verbose(VERB_ALGO, "setsockopt(..., SO_SNDBUFFORCE, "
462 					"...) was not granted: %s", sock_strerror(errno));
463 			}
464 #  endif /* SO_SNDBUFFORCE */
465 			if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&snd,
466 				(socklen_t)sizeof(snd)) < 0) {
467 				if(errno != ENOSYS && errno != ENOBUFS) {
468 					log_err("setsockopt(..., SO_SNDBUF, "
469 						"...) failed: %s", sock_strerror(errno));
470 					sock_close(s);
471 					*noproto = 0;
472 					*inuse = 0;
473 					return -1;
474 				}
475 				log_warn("setsockopt(..., SO_SNDBUF, "
476 					"...) was not granted: %s", sock_strerror(errno));
477 			}
478 			/* check if we got the right thing or if system
479 			 * reduced to some system max.  Warn if so */
480 			if(getsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&got,
481 				&slen) >= 0 && got < snd/2) {
482 				log_warn("so-sndbuf %u was not granted. "
483 					"Got %u. To fix: start with "
484 					"root permissions(linux) or sysctl "
485 					"bigger net.core.wmem_max(linux) or "
486 					"kern.ipc.maxsockbuf(bsd) values. or "
487 					"set so-sndbuf: 0 (use system value).",
488 					(unsigned)snd, (unsigned)got);
489 			}
490 #  ifdef SO_SNDBUFFORCE
491 		}
492 #  endif
493 #endif /* SO_SNDBUF */
494 	}
495 	err = set_ip_dscp(s, family, dscp);
496 	if(err != NULL)
497 		log_warn("error setting IP DiffServ codepoint %d on UDP socket: %s", dscp, err);
498 	if(family == AF_INET6) {
499 # if defined(IPV6_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
500 		int omit6_set = 0;
501 		int action;
502 # endif
503 # if defined(IPV6_V6ONLY)
504 		if(v6only
505 #   ifdef HAVE_SYSTEMD
506 			/* Systemd wants to control if the socket is v6 only
507 			 * or both, with BindIPv6Only=default, ipv6-only or
508 			 * both in systemd.socket, so it is not set here. */
509 			&& !got_fd_from_systemd
510 #   endif
511 			) {
512 			int val=(v6only==2)?0:1;
513 			if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
514 				(void*)&val, (socklen_t)sizeof(val)) < 0) {
515 				log_err("setsockopt(..., IPV6_V6ONLY"
516 					", ...) failed: %s", sock_strerror(errno));
517 				sock_close(s);
518 				*noproto = 0;
519 				*inuse = 0;
520 				return -1;
521 			}
522 		}
523 # endif
524 # if defined(IPV6_USE_MIN_MTU)
525 		/*
526 		 * There is no fragmentation of IPv6 datagrams
527 		 * during forwarding in the network. Therefore
528 		 * we do not send UDP datagrams larger than
529 		 * the minimum IPv6 MTU of 1280 octets. The
530 		 * EDNS0 message length can be larger if the
531 		 * network stack supports IPV6_USE_MIN_MTU.
532 		 */
533 		if (setsockopt(s, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
534 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
535 			log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
536 				"...) failed: %s", sock_strerror(errno));
537 			sock_close(s);
538 			*noproto = 0;
539 			*inuse = 0;
540 			return -1;
541 		}
542 # elif defined(IPV6_MTU)
543 #   ifndef USE_WINSOCK
544 		/*
545 		 * On Linux, to send no larger than 1280, the PMTUD is
546 		 * disabled by default for datagrams anyway, so we set
547 		 * the MTU to use.
548 		 */
549 		if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU,
550 			(void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
551 			log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
552 				sock_strerror(errno));
553 			sock_close(s);
554 			*noproto = 0;
555 			*inuse = 0;
556 			return -1;
557 		}
558 #   elif defined(IPV6_USER_MTU)
559 		/* As later versions of the mingw crosscompiler define
560 		 * IPV6_MTU, do the same for windows but use IPV6_USER_MTU
561 		 * instead which is writable; IPV6_MTU is readonly there. */
562 		if (setsockopt(s, IPPROTO_IPV6, IPV6_USER_MTU,
563 			(void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
564 			if (WSAGetLastError() != WSAENOPROTOOPT) {
565 				log_err("setsockopt(..., IPV6_USER_MTU, ...) failed: %s",
566 					wsa_strerror(WSAGetLastError()));
567 				sock_close(s);
568 				*noproto = 0;
569 				*inuse = 0;
570 				return -1;
571 			}
572 		}
573 #   endif /* USE_WINSOCK */
574 # endif /* IPv6 MTU */
575 # if defined(IPV6_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
576 #  if defined(IP_PMTUDISC_OMIT)
577 		action = IP_PMTUDISC_OMIT;
578 		if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU_DISCOVER,
579 			&action, (socklen_t)sizeof(action)) < 0) {
580 
581 			if (errno != EINVAL) {
582 				log_err("setsockopt(..., IPV6_MTU_DISCOVER, IP_PMTUDISC_OMIT...) failed: %s",
583 					strerror(errno));
584 				sock_close(s);
585 				*noproto = 0;
586 				*inuse = 0;
587 				return -1;
588 			}
589 		}
590 		else
591 		{
592 		    omit6_set = 1;
593 		}
594 #  endif
595 		if (omit6_set == 0) {
596 			action = IP_PMTUDISC_DONT;
597 			if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU_DISCOVER,
598 				&action, (socklen_t)sizeof(action)) < 0) {
599 				log_err("setsockopt(..., IPV6_MTU_DISCOVER, IP_PMTUDISC_DONT...) failed: %s",
600 					strerror(errno));
601 				sock_close(s);
602 				*noproto = 0;
603 				*inuse = 0;
604 				return -1;
605 			}
606 		}
607 # endif /* IPV6_MTU_DISCOVER */
608 	} else if(family == AF_INET) {
609 #  if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
610 /* linux 3.15 has IP_PMTUDISC_OMIT, Hannes Frederic Sowa made it so that
611  * PMTU information is not accepted, but fragmentation is allowed
612  * if and only if the packet size exceeds the outgoing interface MTU
613  * (and also uses the interface mtu to determine the size of the packets).
614  * So there won't be any EMSGSIZE error.  Against DNS fragmentation attacks.
615  * FreeBSD already has same semantics without setting the option. */
616 		int omit_set = 0;
617 		int action;
618 #   if defined(IP_PMTUDISC_OMIT)
619 		action = IP_PMTUDISC_OMIT;
620 		if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
621 			&action, (socklen_t)sizeof(action)) < 0) {
622 
623 			if (errno != EINVAL) {
624 				log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_OMIT...) failed: %s",
625 					strerror(errno));
626 				sock_close(s);
627 				*noproto = 0;
628 				*inuse = 0;
629 				return -1;
630 			}
631 		}
632 		else
633 		{
634 		    omit_set = 1;
635 		}
636 #   endif
637 		if (omit_set == 0) {
638    			action = IP_PMTUDISC_DONT;
639 			if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
640 				&action, (socklen_t)sizeof(action)) < 0) {
641 				log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_DONT...) failed: %s",
642 					strerror(errno));
643 				sock_close(s);
644 				*noproto = 0;
645 				*inuse = 0;
646 				return -1;
647 			}
648 		}
649 #  elif defined(IP_DONTFRAG) && !defined(__APPLE__)
650 		/* the IP_DONTFRAG option if defined in the 11.0 OSX headers,
651 		 * but does not work on that version, so we exclude it */
652 		/* a nonzero value disables fragmentation, according to
653 		 * docs.oracle.com for ip(4). */
654 		int off = 1;
655 		if (setsockopt(s, IPPROTO_IP, IP_DONTFRAG,
656 			&off, (socklen_t)sizeof(off)) < 0) {
657 			log_err("setsockopt(..., IP_DONTFRAG, ...) failed: %s",
658 				strerror(errno));
659 			sock_close(s);
660 			*noproto = 0;
661 			*inuse = 0;
662 			return -1;
663 		}
664 #  endif /* IPv4 MTU */
665 	}
666 	if(
667 #ifdef HAVE_SYSTEMD
668 		!got_fd_from_systemd &&
669 #endif
670 		bind(s, (struct sockaddr*)addr, addrlen) != 0) {
671 		*noproto = 0;
672 		*inuse = 0;
673 #ifndef USE_WINSOCK
674 #ifdef EADDRINUSE
675 		*inuse = (errno == EADDRINUSE);
676 		/* detect freebsd jail with no ipv6 permission */
677 		if(family==AF_INET6 && errno==EINVAL)
678 			*noproto = 1;
679 		else if(errno != EADDRINUSE &&
680 			!(errno == EACCES && verbosity < 4 && !listen)
681 #ifdef EADDRNOTAVAIL
682 			&& !(errno == EADDRNOTAVAIL && verbosity < 4 && !listen)
683 #endif
684 			) {
685 			log_err_addr("can't bind socket", strerror(errno),
686 				(struct sockaddr_storage*)addr, addrlen);
687 		}
688 #endif /* EADDRINUSE */
689 #else /* USE_WINSOCK */
690 		if(WSAGetLastError() != WSAEADDRINUSE &&
691 			WSAGetLastError() != WSAEADDRNOTAVAIL &&
692 			!(WSAGetLastError() == WSAEACCES && verbosity < 4 && !listen)) {
693 			log_err_addr("can't bind socket",
694 				wsa_strerror(WSAGetLastError()),
695 				(struct sockaddr_storage*)addr, addrlen);
696 		}
697 #endif /* USE_WINSOCK */
698 		sock_close(s);
699 		return -1;
700 	}
701 	if(!fd_set_nonblock(s)) {
702 		*noproto = 0;
703 		*inuse = 0;
704 		sock_close(s);
705 		return -1;
706 	}
707 	return s;
708 }
709 
710 int
711 create_tcp_accept_sock(struct addrinfo *addr, int v6only, int* noproto,
712 	int* reuseport, int transparent, int mss, int nodelay, int freebind,
713 	int use_systemd, int dscp, const char* additional)
714 {
715 	int s = -1;
716 	char* err;
717 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT)		\
718 	|| defined(IPV6_V6ONLY) || defined(IP_TRANSPARENT)	\
719 	|| defined(IP_BINDANY) || defined(IP_FREEBIND)		\
720 	|| defined(SO_BINDANY) || defined(TCP_NODELAY)
721 	int on = 1;
722 #endif
723 #ifdef HAVE_SYSTEMD
724 	int got_fd_from_systemd = 0;
725 #endif
726 #ifdef USE_TCP_FASTOPEN
727 	int qlen;
728 #endif
729 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
730 	(void)transparent;
731 #endif
732 #if !defined(IP_FREEBIND)
733 	(void)freebind;
734 #endif
735 	verbose_print_addr(addr, additional);
736 	*noproto = 0;
737 #ifdef HAVE_SYSTEMD
738 	if (!use_systemd ||
739 	    (use_systemd
740 	     && (s = systemd_get_activated(addr->ai_family, addr->ai_socktype, 1,
741 					   addr->ai_addr, addr->ai_addrlen,
742 					   NULL)) == -1)) {
743 #else
744 	(void)use_systemd;
745 #endif
746 	if((s = socket(addr->ai_family, addr->ai_socktype, 0)) == -1) {
747 #ifndef USE_WINSOCK
748 		if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
749 			*noproto = 1;
750 			return -1;
751 		}
752 #else
753 		if(WSAGetLastError() == WSAEAFNOSUPPORT ||
754 			WSAGetLastError() == WSAEPROTONOSUPPORT) {
755 			*noproto = 1;
756 			return -1;
757 		}
758 #endif
759 		log_err("can't create socket: %s", sock_strerror(errno));
760 		return -1;
761 	}
762 	if(nodelay) {
763 #if defined(IPPROTO_TCP) && defined(TCP_NODELAY)
764 		if(setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void*)&on,
765 			(socklen_t)sizeof(on)) < 0) {
766 			#ifndef USE_WINSOCK
767 			log_err(" setsockopt(.. TCP_NODELAY ..) failed: %s",
768 				strerror(errno));
769 			#else
770 			log_err(" setsockopt(.. TCP_NODELAY ..) failed: %s",
771 				wsa_strerror(WSAGetLastError()));
772 			#endif
773 		}
774 #else
775 		log_warn(" setsockopt(TCP_NODELAY) unsupported");
776 #endif /* defined(IPPROTO_TCP) && defined(TCP_NODELAY) */
777 	}
778 	if (mss > 0) {
779 #if defined(IPPROTO_TCP) && defined(TCP_MAXSEG)
780 		if(setsockopt(s, IPPROTO_TCP, TCP_MAXSEG, (void*)&mss,
781 			(socklen_t)sizeof(mss)) < 0) {
782 			log_err(" setsockopt(.. TCP_MAXSEG ..) failed: %s",
783 				sock_strerror(errno));
784 		} else {
785 			verbose(VERB_ALGO,
786 				" tcp socket mss set to %d", mss);
787 		}
788 #else
789 		log_warn(" setsockopt(TCP_MAXSEG) unsupported");
790 #endif /* defined(IPPROTO_TCP) && defined(TCP_MAXSEG) */
791 	}
792 #ifdef HAVE_SYSTEMD
793 	} else {
794 		got_fd_from_systemd = 1;
795     }
796 #endif
797 #ifdef SO_REUSEADDR
798 	if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
799 		(socklen_t)sizeof(on)) < 0) {
800 		log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
801 			sock_strerror(errno));
802 		sock_close(s);
803 		return -1;
804 	}
805 #endif /* SO_REUSEADDR */
806 #ifdef IP_FREEBIND
807 	if (freebind && setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
808 	    (socklen_t)sizeof(on)) < 0) {
809 		log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
810 		strerror(errno));
811 	}
812 #endif /* IP_FREEBIND */
813 #ifdef SO_REUSEPORT
814 	/* try to set SO_REUSEPORT so that incoming
815 	 * connections are distributed evenly among the receiving threads.
816 	 * Each thread must have its own socket bound to the same port,
817 	 * with SO_REUSEPORT set on each socket.
818 	 */
819 	if (reuseport && *reuseport &&
820 		setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
821 		(socklen_t)sizeof(on)) < 0) {
822 #ifdef ENOPROTOOPT
823 		if(errno != ENOPROTOOPT || verbosity >= 3)
824 			log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
825 				strerror(errno));
826 #endif
827 		/* this option is not essential, we can continue */
828 		*reuseport = 0;
829 	}
830 #else
831 	(void)reuseport;
832 #endif /* defined(SO_REUSEPORT) */
833 #if defined(IPV6_V6ONLY)
834 	if(addr->ai_family == AF_INET6 && v6only
835 #  ifdef HAVE_SYSTEMD
836 		/* Systemd wants to control if the socket is v6 only
837 		 * or both, with BindIPv6Only=default, ipv6-only or
838 		 * both in systemd.socket, so it is not set here. */
839 		&& !got_fd_from_systemd
840 #  endif
841 		) {
842 		if(setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
843 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
844 			log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
845 				sock_strerror(errno));
846 			sock_close(s);
847 			return -1;
848 		}
849 	}
850 #else
851 	(void)v6only;
852 #endif /* IPV6_V6ONLY */
853 #ifdef IP_TRANSPARENT
854 	if (transparent &&
855 	    setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
856 	    (socklen_t)sizeof(on)) < 0) {
857 		log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
858 			strerror(errno));
859 	}
860 #elif defined(IP_BINDANY)
861 	if (transparent &&
862 	    setsockopt(s, (addr->ai_family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
863 	    (addr->ai_family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
864 	    (void*)&on, (socklen_t)sizeof(on)) < 0) {
865 		log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
866 		(addr->ai_family==AF_INET6?"V6":""), strerror(errno));
867 	}
868 #elif defined(SO_BINDANY)
869 	if (transparent &&
870 	    setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on, (socklen_t)
871 	    sizeof(on)) < 0) {
872 		log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
873 		strerror(errno));
874 	}
875 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
876 	err = set_ip_dscp(s, addr->ai_family, dscp);
877 	if(err != NULL)
878 		log_warn("error setting IP DiffServ codepoint %d on TCP socket: %s", dscp, err);
879 	if(
880 #ifdef HAVE_SYSTEMD
881 		!got_fd_from_systemd &&
882 #endif
883         bind(s, addr->ai_addr, addr->ai_addrlen) != 0) {
884 #ifndef USE_WINSOCK
885 		/* detect freebsd jail with no ipv6 permission */
886 		if(addr->ai_family==AF_INET6 && errno==EINVAL)
887 			*noproto = 1;
888 		else {
889 			log_err_addr("can't bind socket", strerror(errno),
890 				(struct sockaddr_storage*)addr->ai_addr,
891 				addr->ai_addrlen);
892 		}
893 #else
894 		log_err_addr("can't bind socket",
895 			wsa_strerror(WSAGetLastError()),
896 			(struct sockaddr_storage*)addr->ai_addr,
897 			addr->ai_addrlen);
898 #endif
899 		sock_close(s);
900 		return -1;
901 	}
902 	if(!fd_set_nonblock(s)) {
903 		sock_close(s);
904 		return -1;
905 	}
906 	if(listen(s, TCP_BACKLOG) == -1) {
907 		log_err("can't listen: %s", sock_strerror(errno));
908 		sock_close(s);
909 		return -1;
910 	}
911 #ifdef USE_TCP_FASTOPEN
912 	/* qlen specifies how many outstanding TFO requests to allow. Limit is a defense
913 	   against IP spoofing attacks as suggested in RFC7413 */
914 #ifdef __APPLE__
915 	/* OS X implementation only supports qlen of 1 via this call. Actual
916 	   value is configured by the net.inet.tcp.fastopen_backlog kernel param. */
917 	qlen = 1;
918 #else
919 	/* 5 is recommended on linux */
920 	qlen = 5;
921 #endif
922 	if ((setsockopt(s, IPPROTO_TCP, TCP_FASTOPEN, &qlen,
923 		  sizeof(qlen))) == -1 ) {
924 #ifdef ENOPROTOOPT
925 		/* squelch ENOPROTOOPT: freebsd server mode with kernel support
926 		   disabled, except when verbosity enabled for debugging */
927 		if(errno != ENOPROTOOPT || verbosity >= 3) {
928 #endif
929 		  if(errno == EPERM) {
930 		  	log_warn("Setting TCP Fast Open as server failed: %s ; this could likely be because sysctl net.inet.tcp.fastopen.enabled, net.inet.tcp.fastopen.server_enable, or net.ipv4.tcp_fastopen is disabled", strerror(errno));
931 		  } else {
932 		  	log_err("Setting TCP Fast Open as server failed: %s", strerror(errno));
933 		  }
934 #ifdef ENOPROTOOPT
935 		}
936 #endif
937 	}
938 #endif
939 	return s;
940 }
941 
942 char*
943 set_ip_dscp(int socket, int addrfamily, int dscp)
944 {
945 	int ds;
946 
947 	if(dscp == 0)
948 		return NULL;
949 	ds = dscp << 2;
950 	switch(addrfamily) {
951 	case AF_INET6:
952 	#ifdef IPV6_TCLASS
953 		if(setsockopt(socket, IPPROTO_IPV6, IPV6_TCLASS, (void*)&ds,
954 			sizeof(ds)) < 0)
955 			return sock_strerror(errno);
956 		break;
957 	#else
958 		return "IPV6_TCLASS not defined on this system";
959 	#endif
960 	default:
961 		if(setsockopt(socket, IPPROTO_IP, IP_TOS, (void*)&ds, sizeof(ds)) < 0)
962 			return sock_strerror(errno);
963 		break;
964 	}
965 	return NULL;
966 }
967 
968 int
969 create_local_accept_sock(const char *path, int* noproto, int use_systemd)
970 {
971 #ifdef HAVE_SYSTEMD
972 	int ret;
973 
974 	if (use_systemd && (ret = systemd_get_activated(AF_LOCAL, SOCK_STREAM, 1, NULL, 0, path)) != -1)
975 		return ret;
976 	else {
977 #endif
978 #ifdef HAVE_SYS_UN_H
979 	int s;
980 	struct sockaddr_un usock;
981 #ifndef HAVE_SYSTEMD
982 	(void)use_systemd;
983 #endif
984 
985 	verbose(VERB_ALGO, "creating unix socket %s", path);
986 #ifdef HAVE_STRUCT_SOCKADDR_UN_SUN_LEN
987 	/* this member exists on BSDs, not Linux */
988 	usock.sun_len = (unsigned)sizeof(usock);
989 #endif
990 	usock.sun_family = AF_LOCAL;
991 	/* length is 92-108, 104 on FreeBSD */
992 	(void)strlcpy(usock.sun_path, path, sizeof(usock.sun_path));
993 
994 	if ((s = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) {
995 		log_err("Cannot create local socket %s (%s)",
996 			path, strerror(errno));
997 		return -1;
998 	}
999 
1000 	if (unlink(path) && errno != ENOENT) {
1001 		/* The socket already exists and cannot be removed */
1002 		log_err("Cannot remove old local socket %s (%s)",
1003 			path, strerror(errno));
1004 		goto err;
1005 	}
1006 
1007 	if (bind(s, (struct sockaddr *)&usock,
1008 		(socklen_t)sizeof(struct sockaddr_un)) == -1) {
1009 		log_err("Cannot bind local socket %s (%s)",
1010 			path, strerror(errno));
1011 		goto err;
1012 	}
1013 
1014 	if (!fd_set_nonblock(s)) {
1015 		log_err("Cannot set non-blocking mode");
1016 		goto err;
1017 	}
1018 
1019 	if (listen(s, TCP_BACKLOG) == -1) {
1020 		log_err("can't listen: %s", strerror(errno));
1021 		goto err;
1022 	}
1023 
1024 	(void)noproto; /*unused*/
1025 	return s;
1026 
1027 err:
1028 	sock_close(s);
1029 	return -1;
1030 
1031 #ifdef HAVE_SYSTEMD
1032 	}
1033 #endif
1034 #else
1035 	(void)use_systemd;
1036 	(void)path;
1037 	log_err("Local sockets are not supported");
1038 	*noproto = 1;
1039 	return -1;
1040 #endif
1041 }
1042 
1043 
1044 /**
1045  * Create socket from getaddrinfo results
1046  */
1047 static int
1048 make_sock(int stype, const char* ifname, int port,
1049 	struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
1050 	int* reuseport, int transparent, int tcp_mss, int nodelay, int freebind,
1051 	int use_systemd, int dscp, struct unbound_socket* ub_sock,
1052 	const char* additional)
1053 {
1054 	struct addrinfo *res = NULL;
1055 	int r, s, inuse, noproto;
1056 	char portbuf[32];
1057 	snprintf(portbuf, sizeof(portbuf), "%d", port);
1058 	hints->ai_socktype = stype;
1059 	*noip6 = 0;
1060 	if((r=getaddrinfo(ifname, portbuf, hints, &res)) != 0 || !res) {
1061 #ifdef USE_WINSOCK
1062 		if(r == EAI_NONAME && hints->ai_family == AF_INET6){
1063 			*noip6 = 1; /* 'Host not found' for IP6 on winXP */
1064 			return -1;
1065 		}
1066 #endif
1067 		log_err("node %s:%s getaddrinfo: %s %s",
1068 			ifname?ifname:"default", portbuf, gai_strerror(r),
1069 #ifdef EAI_SYSTEM
1070 			(r==EAI_SYSTEM?(char*)strerror(errno):"")
1071 #else
1072 			""
1073 #endif
1074 		);
1075 		return -1;
1076 	}
1077 	if(stype == SOCK_DGRAM) {
1078 		verbose_print_addr(res, additional);
1079 		s = create_udp_sock(res->ai_family, res->ai_socktype,
1080 			(struct sockaddr*)res->ai_addr, res->ai_addrlen,
1081 			v6only, &inuse, &noproto, (int)rcv, (int)snd, 1,
1082 			reuseport, transparent, freebind, use_systemd, dscp);
1083 		if(s == -1 && inuse) {
1084 			log_err("bind: address already in use");
1085 		} else if(s == -1 && noproto && hints->ai_family == AF_INET6){
1086 			*noip6 = 1;
1087 		}
1088 	} else	{
1089 		s = create_tcp_accept_sock(res, v6only, &noproto, reuseport,
1090 			transparent, tcp_mss, nodelay, freebind, use_systemd,
1091 			dscp, additional);
1092 		if(s == -1 && noproto && hints->ai_family == AF_INET6){
1093 			*noip6 = 1;
1094 		}
1095 	}
1096 
1097 	if(!res->ai_addr) {
1098 		log_err("getaddrinfo returned no address");
1099 		freeaddrinfo(res);
1100 		sock_close(s);
1101 		return -1;
1102 	}
1103 	ub_sock->addr = memdup(res->ai_addr, res->ai_addrlen);
1104 	ub_sock->addrlen = res->ai_addrlen;
1105 	if(!ub_sock->addr) {
1106 		log_err("out of memory: allocate listening address");
1107 		freeaddrinfo(res);
1108 		sock_close(s);
1109 		return -1;
1110 	}
1111 	freeaddrinfo(res);
1112 
1113 	ub_sock->s = s;
1114 	ub_sock->fam = hints->ai_family;
1115 	ub_sock->acl = NULL;
1116 
1117 	return s;
1118 }
1119 
1120 /** make socket and first see if ifname contains port override info */
1121 static int
1122 make_sock_port(int stype, const char* ifname, int port,
1123 	struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
1124 	int* reuseport, int transparent, int tcp_mss, int nodelay, int freebind,
1125 	int use_systemd, int dscp, struct unbound_socket* ub_sock,
1126 	const char* additional)
1127 {
1128 	const char* s = strchr(ifname, '@');
1129 	if(s) {
1130 		/* override port with ifspec@port */
1131 		int port;
1132 		char newif[128];
1133 		if((size_t)(s-ifname) >= sizeof(newif)) {
1134 			log_err("ifname too long: %s", ifname);
1135 			*noip6 = 0;
1136 			return -1;
1137 		}
1138 		port = atoi(s+1);
1139 		if(port < 0 || 0 == port || port > 65535) {
1140 			log_err("invalid portnumber in interface: %s", ifname);
1141 			*noip6 = 0;
1142 			return -1;
1143 		}
1144 		(void)strlcpy(newif, ifname, sizeof(newif));
1145 		newif[s-ifname] = 0;
1146 		return make_sock(stype, newif, port, hints, v6only, noip6, rcv,
1147 			snd, reuseport, transparent, tcp_mss, nodelay, freebind,
1148 			use_systemd, dscp, ub_sock, additional);
1149 	}
1150 	return make_sock(stype, ifname, port, hints, v6only, noip6, rcv, snd,
1151 		reuseport, transparent, tcp_mss, nodelay, freebind, use_systemd,
1152 		dscp, ub_sock, additional);
1153 }
1154 
1155 /**
1156  * Add port to open ports list.
1157  * @param list: list head. changed.
1158  * @param s: fd.
1159  * @param ftype: if fd is UDP.
1160  * @param pp2_enabled: if PROXYv2 is enabled for this port.
1161  * @param ub_sock: socket with address.
1162  * @return false on failure. list in unchanged then.
1163  */
1164 static int
1165 port_insert(struct listen_port** list, int s, enum listen_type ftype,
1166 	int pp2_enabled, struct unbound_socket* ub_sock)
1167 {
1168 	struct listen_port* item = (struct listen_port*)malloc(
1169 		sizeof(struct listen_port));
1170 	if(!item)
1171 		return 0;
1172 	item->next = *list;
1173 	item->fd = s;
1174 	item->ftype = ftype;
1175 	item->pp2_enabled = pp2_enabled;
1176 	item->socket = ub_sock;
1177 	*list = item;
1178 	return 1;
1179 }
1180 
1181 /** set fd to receive software timestamps */
1182 static int
1183 set_recvtimestamp(int s)
1184 {
1185 #ifdef HAVE_LINUX_NET_TSTAMP_H
1186 	int opt = SOF_TIMESTAMPING_RX_SOFTWARE | SOF_TIMESTAMPING_SOFTWARE;
1187 	if (setsockopt(s, SOL_SOCKET, SO_TIMESTAMPNS, (void*)&opt, (socklen_t)sizeof(opt)) < 0) {
1188 		log_err("setsockopt(..., SO_TIMESTAMPNS, ...) failed: %s",
1189 			strerror(errno));
1190 		return 0;
1191 	}
1192 	return 1;
1193 #elif defined(SO_TIMESTAMP) && defined(SCM_TIMESTAMP)
1194 	int on = 1;
1195 	/* FreeBSD and also Linux. */
1196 	if (setsockopt(s, SOL_SOCKET, SO_TIMESTAMP, (void*)&on, (socklen_t)sizeof(on)) < 0) {
1197 		log_err("setsockopt(..., SO_TIMESTAMP, ...) failed: %s",
1198 			strerror(errno));
1199 		return 0;
1200 	}
1201 	return 1;
1202 #else
1203 	log_err("packets timestamping is not supported on this platform");
1204 	(void)s;
1205 	return 0;
1206 #endif
1207 }
1208 
1209 /** set fd to receive source address packet info */
1210 static int
1211 set_recvpktinfo(int s, int family)
1212 {
1213 #if defined(IPV6_RECVPKTINFO) || defined(IPV6_PKTINFO) || (defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)) || defined(IP_PKTINFO)
1214 	int on = 1;
1215 #else
1216 	(void)s;
1217 #endif
1218 	if(family == AF_INET6) {
1219 #           ifdef IPV6_RECVPKTINFO
1220 		if(setsockopt(s, IPPROTO_IPV6, IPV6_RECVPKTINFO,
1221 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
1222 			log_err("setsockopt(..., IPV6_RECVPKTINFO, ...) failed: %s",
1223 				strerror(errno));
1224 			return 0;
1225 		}
1226 #           elif defined(IPV6_PKTINFO)
1227 		if(setsockopt(s, IPPROTO_IPV6, IPV6_PKTINFO,
1228 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
1229 			log_err("setsockopt(..., IPV6_PKTINFO, ...) failed: %s",
1230 				strerror(errno));
1231 			return 0;
1232 		}
1233 #           else
1234 		log_err("no IPV6_RECVPKTINFO and IPV6_PKTINFO options, please "
1235 			"disable interface-automatic or do-ip6 in config");
1236 		return 0;
1237 #           endif /* defined IPV6_RECVPKTINFO */
1238 
1239 	} else if(family == AF_INET) {
1240 #           ifdef IP_PKTINFO
1241 		if(setsockopt(s, IPPROTO_IP, IP_PKTINFO,
1242 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
1243 			log_err("setsockopt(..., IP_PKTINFO, ...) failed: %s",
1244 				strerror(errno));
1245 			return 0;
1246 		}
1247 #           elif defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)
1248 		if(setsockopt(s, IPPROTO_IP, IP_RECVDSTADDR,
1249 			(void*)&on, (socklen_t)sizeof(on)) < 0) {
1250 			log_err("setsockopt(..., IP_RECVDSTADDR, ...) failed: %s",
1251 				strerror(errno));
1252 			return 0;
1253 		}
1254 #           else
1255 		log_err("no IP_SENDSRCADDR or IP_PKTINFO option, please disable "
1256 			"interface-automatic or do-ip4 in config");
1257 		return 0;
1258 #           endif /* IP_PKTINFO */
1259 
1260 	}
1261 	return 1;
1262 }
1263 
1264 /**
1265  * Helper for ports_open. Creates one interface (or NULL for default).
1266  * @param ifname: The interface ip address.
1267  * @param do_auto: use automatic interface detection.
1268  * 	If enabled, then ifname must be the wildcard name.
1269  * @param do_udp: if udp should be used.
1270  * @param do_tcp: if tcp should be used.
1271  * @param hints: for getaddrinfo. family and flags have to be set by caller.
1272  * @param port: Port number to use.
1273  * @param list: list of open ports, appended to, changed to point to list head.
1274  * @param rcv: receive buffer size for UDP
1275  * @param snd: send buffer size for UDP
1276  * @param ssl_port: ssl service port number
1277  * @param tls_additional_port: list of additional ssl service port numbers.
1278  * @param https_port: DoH service port number
1279  * @param proxy_protocol_port: list of PROXYv2 port numbers.
1280  * @param reuseport: try to set SO_REUSEPORT if nonNULL and true.
1281  * 	set to false on exit if reuseport failed due to no kernel support.
1282  * @param transparent: set IP_TRANSPARENT socket option.
1283  * @param tcp_mss: maximum segment size of tcp socket. default if zero.
1284  * @param freebind: set IP_FREEBIND socket option.
1285  * @param http2_nodelay: set TCP_NODELAY on HTTP/2 connection
1286  * @param use_systemd: if true, fetch sockets from systemd.
1287  * @param dnscrypt_port: dnscrypt service port number
1288  * @param dscp: DSCP to use.
1289  * @param quic_port: dns over quic port number.
1290  * @param http_notls_downstream: if no tls is used for https downstream.
1291  * @param sock_queue_timeout: the sock_queue_timeout from config. Seconds to
1292  * 	wait to discard if UDP packets have waited for long in the socket
1293  * 	buffer.
1294  * @return: returns false on error.
1295  */
1296 static int
1297 ports_create_if(const char* ifname, int do_auto, int do_udp, int do_tcp,
1298 	struct addrinfo *hints, int port, struct listen_port** list,
1299 	size_t rcv, size_t snd, int ssl_port,
1300 	struct config_strlist* tls_additional_port, int https_port,
1301 	struct config_strlist* proxy_protocol_port,
1302 	int* reuseport, int transparent, int tcp_mss, int freebind,
1303 	int http2_nodelay, int use_systemd, int dnscrypt_port, int dscp,
1304 	int quic_port, int http_notls_downstream, int sock_queue_timeout)
1305 {
1306 	int s, noip6=0;
1307 	int is_ssl = if_is_ssl(ifname, port, ssl_port, tls_additional_port);
1308 	int is_https = if_is_https(ifname, port, https_port);
1309 	int is_dnscrypt = if_is_dnscrypt(ifname, port, dnscrypt_port);
1310 	int is_pp2 = if_is_pp2(ifname, port, proxy_protocol_port);
1311 	int is_doq = if_is_quic(ifname, port, quic_port);
1312 	/* Always set TCP_NODELAY on TLS connection as it speeds up the TLS
1313 	 * handshake. DoH had already such option so we respect it.
1314 	 * Otherwise the server waits before sending more handshake data for
1315 	 * the client ACK (Nagle's algorithm), which is delayed because the
1316 	 * client waits for more data before ACKing (delayed ACK). */
1317 	int nodelay = is_https?http2_nodelay:is_ssl;
1318 	struct unbound_socket* ub_sock;
1319 	const char* add = NULL;
1320 
1321 	if(!do_udp && !do_tcp)
1322 		return 0;
1323 
1324 	if(is_pp2) {
1325 		if(is_dnscrypt) {
1326 			fatal_exit("PROXYv2 and DNSCrypt combination not "
1327 				"supported!");
1328 		} else if(is_https) {
1329 			fatal_exit("PROXYv2 and DoH combination not "
1330 				"supported!");
1331 		} else if(is_doq) {
1332 			fatal_exit("PROXYv2 and DoQ combination not "
1333 				"supported!");
1334 		}
1335 	}
1336 
1337 	/* Check if both UDP and TCP ports should be open.
1338 	 * In the case of encrypted channels, probably an unencrypted channel
1339 	 * at the same port is not desired. */
1340 	if((is_ssl || is_https) && !is_doq) do_udp = do_auto = 0;
1341 	if((is_doq) && !(is_https || is_ssl)) do_tcp = 0;
1342 
1343 	if(do_auto) {
1344 		ub_sock = calloc(1, sizeof(struct unbound_socket));
1345 		if(!ub_sock)
1346 			return 0;
1347 		if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1348 			&noip6, rcv, snd, reuseport, transparent,
1349 			tcp_mss, nodelay, freebind, use_systemd, dscp, ub_sock,
1350 			(is_dnscrypt?"udpancil_dnscrypt":"udpancil"))) == -1) {
1351 			free(ub_sock->addr);
1352 			free(ub_sock);
1353 			if(noip6) {
1354 				log_warn("IPv6 protocol not available");
1355 				return 1;
1356 			}
1357 			return 0;
1358 		}
1359 		/* getting source addr packet info is highly non-portable */
1360 		if(!set_recvpktinfo(s, hints->ai_family)) {
1361 			sock_close(s);
1362 			free(ub_sock->addr);
1363 			free(ub_sock);
1364 			return 0;
1365 		}
1366 		if (sock_queue_timeout && !set_recvtimestamp(s)) {
1367 			log_warn("socket timestamping is not available");
1368 		}
1369 		if(!port_insert(list, s, is_dnscrypt
1370 			?listen_type_udpancil_dnscrypt:listen_type_udpancil,
1371 			is_pp2, ub_sock)) {
1372 			sock_close(s);
1373 			free(ub_sock->addr);
1374 			free(ub_sock);
1375 			return 0;
1376 		}
1377 	} else if(do_udp) {
1378 		enum listen_type udp_port_type;
1379 		ub_sock = calloc(1, sizeof(struct unbound_socket));
1380 		if(!ub_sock)
1381 			return 0;
1382 		if(is_dnscrypt) {
1383 			udp_port_type = listen_type_udp_dnscrypt;
1384 			add = "dnscrypt";
1385 		} else if(is_doq) {
1386 			udp_port_type = listen_type_doq;
1387 			add = "doq";
1388 			if(if_listens_on(ifname, port, 53, NULL)) {
1389 				log_err("DNS over QUIC is strictly not "
1390 					"allowed on port 53 as per RFC 9250. "
1391 					"Port 53 is for DNS datagrams. Error "
1392 					"for interface '%s'.", ifname);
1393 				free(ub_sock->addr);
1394 				free(ub_sock);
1395 				return 0;
1396 			}
1397 		} else {
1398 			udp_port_type = listen_type_udp;
1399 			add = NULL;
1400 		}
1401 		/* regular udp socket */
1402 		if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1403 			&noip6, rcv, snd, reuseport, transparent,
1404 			tcp_mss, nodelay, freebind, use_systemd, dscp, ub_sock,
1405 			add)) == -1) {
1406 			free(ub_sock->addr);
1407 			free(ub_sock);
1408 			if(noip6) {
1409 				log_warn("IPv6 protocol not available");
1410 				return 1;
1411 			}
1412 			return 0;
1413 		}
1414 		if(udp_port_type == listen_type_doq) {
1415 			if(!set_recvpktinfo(s, hints->ai_family)) {
1416 				sock_close(s);
1417 				free(ub_sock->addr);
1418 				free(ub_sock);
1419 				return 0;
1420 			}
1421 		}
1422 		if(udp_port_type == listen_type_udp && sock_queue_timeout)
1423 			udp_port_type = listen_type_udpancil;
1424 		if (sock_queue_timeout) {
1425 			if(!set_recvtimestamp(s)) {
1426 				log_warn("socket timestamping is not available");
1427 			} else {
1428 				if(udp_port_type == listen_type_udp)
1429 					udp_port_type = listen_type_udpancil;
1430 			}
1431 		}
1432 		if(!port_insert(list, s, udp_port_type, is_pp2, ub_sock)) {
1433 			sock_close(s);
1434 			free(ub_sock->addr);
1435 			free(ub_sock);
1436 			return 0;
1437 		}
1438 	}
1439 	if(do_tcp) {
1440 		enum listen_type port_type;
1441 		ub_sock = calloc(1, sizeof(struct unbound_socket));
1442 		if(!ub_sock)
1443 			return 0;
1444 		if(is_ssl) {
1445 			port_type = listen_type_ssl;
1446 			add = "tls";
1447 		} else if(is_https) {
1448 			port_type = listen_type_http;
1449 			add = "https";
1450 			if(http_notls_downstream)
1451 				add = "http";
1452 		} else if(is_dnscrypt) {
1453 			port_type = listen_type_tcp_dnscrypt;
1454 			add = "dnscrypt";
1455 		} else {
1456 			port_type = listen_type_tcp;
1457 			add = NULL;
1458 		}
1459 		if((s = make_sock_port(SOCK_STREAM, ifname, port, hints, 1,
1460 			&noip6, 0, 0, reuseport, transparent, tcp_mss, nodelay,
1461 			freebind, use_systemd, dscp, ub_sock, add)) == -1) {
1462 			free(ub_sock->addr);
1463 			free(ub_sock);
1464 			if(noip6) {
1465 				/*log_warn("IPv6 protocol not available");*/
1466 				return 1;
1467 			}
1468 			return 0;
1469 		}
1470 		if(is_ssl)
1471 			verbose(VERB_ALGO, "setup TCP for SSL service");
1472 		if(!port_insert(list, s, port_type, is_pp2, ub_sock)) {
1473 			sock_close(s);
1474 			free(ub_sock->addr);
1475 			free(ub_sock);
1476 			return 0;
1477 		}
1478 	}
1479 	return 1;
1480 }
1481 
1482 /**
1483  * Add items to commpoint list in front.
1484  * @param c: commpoint to add.
1485  * @param front: listen struct.
1486  * @return: false on failure.
1487  */
1488 static int
1489 listen_cp_insert(struct comm_point* c, struct listen_dnsport* front)
1490 {
1491 	struct listen_list* item = (struct listen_list*)malloc(
1492 		sizeof(struct listen_list));
1493 	if(!item)
1494 		return 0;
1495 	item->com = c;
1496 	item->next = front->cps;
1497 	front->cps = item;
1498 	return 1;
1499 }
1500 
1501 void listen_setup_locks(void)
1502 {
1503 	if(!stream_wait_lock_inited) {
1504 		lock_basic_init(&stream_wait_count_lock);
1505 		stream_wait_lock_inited = 1;
1506 	}
1507 	if(!http2_query_buffer_lock_inited) {
1508 		lock_basic_init(&http2_query_buffer_count_lock);
1509 		http2_query_buffer_lock_inited = 1;
1510 	}
1511 	if(!http2_response_buffer_lock_inited) {
1512 		lock_basic_init(&http2_response_buffer_count_lock);
1513 		http2_response_buffer_lock_inited = 1;
1514 	}
1515 }
1516 
1517 void listen_desetup_locks(void)
1518 {
1519 	if(stream_wait_lock_inited) {
1520 		stream_wait_lock_inited = 0;
1521 		lock_basic_destroy(&stream_wait_count_lock);
1522 	}
1523 	if(http2_query_buffer_lock_inited) {
1524 		http2_query_buffer_lock_inited = 0;
1525 		lock_basic_destroy(&http2_query_buffer_count_lock);
1526 	}
1527 	if(http2_response_buffer_lock_inited) {
1528 		http2_response_buffer_lock_inited = 0;
1529 		lock_basic_destroy(&http2_response_buffer_count_lock);
1530 	}
1531 }
1532 
1533 struct listen_dnsport*
1534 listen_create(struct comm_base* base, struct listen_port* ports,
1535 	size_t bufsize, int tcp_accept_count, int tcp_idle_timeout,
1536 	int harden_large_queries, uint32_t http_max_streams,
1537 	char* http_endpoint, int http_notls, struct tcl_list* tcp_conn_limit,
1538 	void* dot_sslctx, void* doh_sslctx, void* quic_sslctx,
1539 	struct dt_env* dtenv,
1540 	struct doq_table* doq_table,
1541 	struct ub_randstate* rnd,struct config_file* cfg,
1542 	comm_point_callback_type* cb, void *cb_arg)
1543 {
1544 	struct listen_dnsport* front = (struct listen_dnsport*)
1545 		malloc(sizeof(struct listen_dnsport));
1546 	if(!front)
1547 		return NULL;
1548 	front->cps = NULL;
1549 	front->udp_buff = sldns_buffer_new(bufsize);
1550 #ifdef USE_DNSCRYPT
1551 	front->dnscrypt_udp_buff = NULL;
1552 #endif
1553 	if(!front->udp_buff) {
1554 		free(front);
1555 		return NULL;
1556 	}
1557 
1558 	/* create comm points as needed */
1559 	while(ports) {
1560 		struct comm_point* cp = NULL;
1561 		if(ports->ftype == listen_type_udp ||
1562 		   ports->ftype == listen_type_udp_dnscrypt) {
1563 			cp = comm_point_create_udp(base, ports->fd,
1564 				front->udp_buff, ports->pp2_enabled, cb,
1565 				cb_arg, ports->socket);
1566 		} else if(ports->ftype == listen_type_doq && doq_table) {
1567 #ifndef HAVE_NGTCP2
1568 			log_warn("Unbound is not compiled with "
1569 				"ngtcp2. This is required to use DNS "
1570 				"over QUIC.");
1571 #endif
1572 			cp = comm_point_create_doq(base, ports->fd,
1573 				front->udp_buff, cb, cb_arg, ports->socket,
1574 				doq_table, rnd, quic_sslctx, cfg);
1575 		} else if(ports->ftype == listen_type_tcp ||
1576 				ports->ftype == listen_type_tcp_dnscrypt) {
1577 			cp = comm_point_create_tcp(base, ports->fd,
1578 				tcp_accept_count, tcp_idle_timeout,
1579 				harden_large_queries, 0, NULL,
1580 				tcp_conn_limit, bufsize, front->udp_buff,
1581 				ports->ftype, ports->pp2_enabled, cb, cb_arg,
1582 				ports->socket);
1583 		} else if(ports->ftype == listen_type_ssl ||
1584 			ports->ftype == listen_type_http) {
1585 			cp = comm_point_create_tcp(base, ports->fd,
1586 				tcp_accept_count, tcp_idle_timeout,
1587 				harden_large_queries,
1588 				http_max_streams, http_endpoint,
1589 				tcp_conn_limit, bufsize, front->udp_buff,
1590 				ports->ftype, ports->pp2_enabled, cb, cb_arg,
1591 				ports->socket);
1592 			if(ports->ftype == listen_type_http) {
1593 				if(!doh_sslctx && !http_notls) {
1594 					log_warn("HTTPS port configured, but "
1595 						"no TLS tls-service-key or "
1596 						"tls-service-pem set");
1597 				}
1598 #ifndef HAVE_SSL_CTX_SET_ALPN_SELECT_CB
1599 				if(!http_notls) {
1600 					log_warn("Unbound is not compiled "
1601 						"with an OpenSSL version "
1602 						"supporting ALPN "
1603 						"(OpenSSL >= 1.0.2). This "
1604 						"is required to use "
1605 						"DNS-over-HTTPS");
1606 				}
1607 #endif
1608 #ifndef HAVE_NGHTTP2_NGHTTP2_H
1609 				log_warn("Unbound is not compiled with "
1610 					"nghttp2. This is required to use "
1611 					"DNS-over-HTTPS.");
1612 #endif
1613 			}
1614 		} else if(ports->ftype == listen_type_udpancil ||
1615 				  ports->ftype == listen_type_udpancil_dnscrypt) {
1616 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
1617 			cp = comm_point_create_udp_ancil(base, ports->fd,
1618 				front->udp_buff, ports->pp2_enabled, cb,
1619 				cb_arg, ports->socket);
1620 #else
1621 			log_warn("This system does not support UDP ancillary data.");
1622 #endif
1623 		}
1624 		if(!cp) {
1625 			log_err("can't create commpoint");
1626 			listen_delete(front);
1627 			return NULL;
1628 		}
1629 		if((http_notls && ports->ftype == listen_type_http) ||
1630 			(ports->ftype == listen_type_tcp) ||
1631 			(ports->ftype == listen_type_udp) ||
1632 			(ports->ftype == listen_type_udpancil) ||
1633 			(ports->ftype == listen_type_tcp_dnscrypt) ||
1634 			(ports->ftype == listen_type_udp_dnscrypt) ||
1635 			(ports->ftype == listen_type_udpancil_dnscrypt)) {
1636 			cp->ssl = NULL;
1637 		} else if(ports->ftype == listen_type_doq) {
1638 			cp->ssl = quic_sslctx;
1639 		} else if(ports->ftype == listen_type_http) {
1640 			cp->ssl = doh_sslctx;
1641 		} else {
1642 			cp->ssl = dot_sslctx;
1643 		}
1644 		cp->dtenv = dtenv;
1645 		cp->do_not_close = 1;
1646 #ifdef USE_DNSCRYPT
1647 		if (ports->ftype == listen_type_udp_dnscrypt ||
1648 			ports->ftype == listen_type_tcp_dnscrypt ||
1649 			ports->ftype == listen_type_udpancil_dnscrypt) {
1650 			cp->dnscrypt = 1;
1651 			cp->dnscrypt_buffer = sldns_buffer_new(bufsize);
1652 			if(!cp->dnscrypt_buffer) {
1653 				log_err("can't alloc dnscrypt_buffer");
1654 				comm_point_delete(cp);
1655 				listen_delete(front);
1656 				return NULL;
1657 			}
1658 			front->dnscrypt_udp_buff = cp->dnscrypt_buffer;
1659 		}
1660 #endif
1661 		if(!listen_cp_insert(cp, front)) {
1662 			log_err("malloc failed");
1663 			comm_point_delete(cp);
1664 			listen_delete(front);
1665 			return NULL;
1666 		}
1667 		ports = ports->next;
1668 	}
1669 	if(!front->cps) {
1670 		log_err("Could not open sockets to accept queries.");
1671 		listen_delete(front);
1672 		return NULL;
1673 	}
1674 
1675 	return front;
1676 }
1677 
1678 void
1679 listen_list_delete(struct listen_list* list)
1680 {
1681 	struct listen_list *p = list, *pn;
1682 	while(p) {
1683 		pn = p->next;
1684 		comm_point_delete(p->com);
1685 		free(p);
1686 		p = pn;
1687 	}
1688 }
1689 
1690 void
1691 listen_delete(struct listen_dnsport* front)
1692 {
1693 	if(!front)
1694 		return;
1695 	listen_list_delete(front->cps);
1696 #ifdef USE_DNSCRYPT
1697 	if(front->dnscrypt_udp_buff &&
1698 		front->udp_buff != front->dnscrypt_udp_buff) {
1699 		sldns_buffer_free(front->dnscrypt_udp_buff);
1700 	}
1701 #endif
1702 	sldns_buffer_free(front->udp_buff);
1703 	free(front);
1704 }
1705 
1706 #ifdef HAVE_GETIFADDRS
1707 static int
1708 resolve_ifa_name(struct ifaddrs *ifas, const char *search_ifa, char ***ip_addresses, int *ip_addresses_size)
1709 {
1710 	struct ifaddrs *ifa;
1711 	void *tmpbuf;
1712 	int last_ip_addresses_size = *ip_addresses_size;
1713 
1714 	for(ifa = ifas; ifa != NULL; ifa = ifa->ifa_next) {
1715 		sa_family_t family;
1716 		const char* atsign;
1717 #ifdef INET6      /* |   address ip    | % |  ifa name  | @ |  port  | nul */
1718 		char addr_buf[INET6_ADDRSTRLEN + 1 + IF_NAMESIZE + 1 + 16 + 1];
1719 #else
1720 		char addr_buf[INET_ADDRSTRLEN + 1 + 16 + 1];
1721 #endif
1722 
1723 		if((atsign=strrchr(search_ifa, '@')) != NULL) {
1724 			if(strlen(ifa->ifa_name) != (size_t)(atsign-search_ifa)
1725 			   || strncmp(ifa->ifa_name, search_ifa,
1726 			   atsign-search_ifa) != 0)
1727 				continue;
1728 		} else {
1729 			if(strcmp(ifa->ifa_name, search_ifa) != 0)
1730 				continue;
1731 			atsign = "";
1732 		}
1733 
1734 		if(ifa->ifa_addr == NULL)
1735 			continue;
1736 
1737 		family = ifa->ifa_addr->sa_family;
1738 		if(family == AF_INET) {
1739 			char a4[INET_ADDRSTRLEN + 1];
1740 			struct sockaddr_in *in4 = (struct sockaddr_in *)
1741 				ifa->ifa_addr;
1742 			if(!inet_ntop(family, &in4->sin_addr, a4, sizeof(a4))) {
1743 				log_err("inet_ntop failed");
1744 				return 0;
1745 			}
1746 			snprintf(addr_buf, sizeof(addr_buf), "%s%s",
1747 				a4, atsign);
1748 		}
1749 #ifdef INET6
1750 		else if(family == AF_INET6) {
1751 			struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)
1752 				ifa->ifa_addr;
1753 			char a6[INET6_ADDRSTRLEN + 1];
1754 			char if_index_name[IF_NAMESIZE + 1];
1755 			if_index_name[0] = 0;
1756 			if(!inet_ntop(family, &in6->sin6_addr, a6, sizeof(a6))) {
1757 				log_err("inet_ntop failed");
1758 				return 0;
1759 			}
1760 			(void)if_indextoname(in6->sin6_scope_id,
1761 				(char *)if_index_name);
1762 			if (strlen(if_index_name) != 0) {
1763 				snprintf(addr_buf, sizeof(addr_buf),
1764 					"%s%%%s%s", a6, if_index_name, atsign);
1765 			} else {
1766 				snprintf(addr_buf, sizeof(addr_buf), "%s%s",
1767 					a6, atsign);
1768 			}
1769 		}
1770 #endif
1771 		else {
1772 			continue;
1773 		}
1774 		verbose(4, "interface %s has address %s", search_ifa, addr_buf);
1775 
1776 		tmpbuf = realloc(*ip_addresses, sizeof(char *) * (*ip_addresses_size + 1));
1777 		if(!tmpbuf) {
1778 			log_err("realloc failed: out of memory");
1779 			return 0;
1780 		} else {
1781 			*ip_addresses = tmpbuf;
1782 		}
1783 		(*ip_addresses)[*ip_addresses_size] = strdup(addr_buf);
1784 		if(!(*ip_addresses)[*ip_addresses_size]) {
1785 			log_err("strdup failed: out of memory");
1786 			return 0;
1787 		}
1788 		(*ip_addresses_size)++;
1789 	}
1790 
1791 	if (*ip_addresses_size == last_ip_addresses_size) {
1792 		tmpbuf = realloc(*ip_addresses, sizeof(char *) * (*ip_addresses_size + 1));
1793 		if(!tmpbuf) {
1794 			log_err("realloc failed: out of memory");
1795 			return 0;
1796 		} else {
1797 			*ip_addresses = tmpbuf;
1798 		}
1799 		(*ip_addresses)[*ip_addresses_size] = strdup(search_ifa);
1800 		if(!(*ip_addresses)[*ip_addresses_size]) {
1801 			log_err("strdup failed: out of memory");
1802 			return 0;
1803 		}
1804 		(*ip_addresses_size)++;
1805 	}
1806 	return 1;
1807 }
1808 #endif /* HAVE_GETIFADDRS */
1809 
1810 int resolve_interface_names(char** ifs, int num_ifs,
1811 	struct config_strlist* list, char*** resif, int* num_resif)
1812 {
1813 #ifdef HAVE_GETIFADDRS
1814 	struct ifaddrs *addrs = NULL;
1815 	if(num_ifs == 0 && list == NULL) {
1816 		*resif = NULL;
1817 		*num_resif = 0;
1818 		return 1;
1819 	}
1820 	if(getifaddrs(&addrs) == -1) {
1821 		log_err("failed to list interfaces: getifaddrs: %s",
1822 			strerror(errno));
1823 		freeifaddrs(addrs);
1824 		return 0;
1825 	}
1826 	if(ifs) {
1827 		int i;
1828 		for(i=0; i<num_ifs; i++) {
1829 			if(!resolve_ifa_name(addrs, ifs[i], resif, num_resif)) {
1830 				freeifaddrs(addrs);
1831 				config_del_strarray(*resif, *num_resif);
1832 				*resif = NULL;
1833 				*num_resif = 0;
1834 				return 0;
1835 			}
1836 		}
1837 	}
1838 	if(list) {
1839 		struct config_strlist* p;
1840 		for(p = list; p; p = p->next) {
1841 			if(!resolve_ifa_name(addrs, p->str, resif, num_resif)) {
1842 				freeifaddrs(addrs);
1843 				config_del_strarray(*resif, *num_resif);
1844 				*resif = NULL;
1845 				*num_resif = 0;
1846 				return 0;
1847 			}
1848 }
1849 	}
1850 	freeifaddrs(addrs);
1851 	return 1;
1852 #else
1853 	struct config_strlist* p;
1854 	if(num_ifs == 0 && list == NULL) {
1855 		*resif = NULL;
1856 		*num_resif = 0;
1857 		return 1;
1858 	}
1859 	*num_resif = num_ifs;
1860 	for(p = list; p; p = p->next) {
1861 		(*num_resif)++;
1862 	}
1863 	*resif = calloc(*num_resif, sizeof(**resif));
1864 	if(!*resif) {
1865 		log_err("out of memory");
1866 		return 0;
1867 	}
1868 	if(ifs) {
1869 		int i;
1870 		for(i=0; i<num_ifs; i++) {
1871 			(*resif)[i] = strdup(ifs[i]);
1872 			if(!((*resif)[i])) {
1873 				log_err("out of memory");
1874 				config_del_strarray(*resif, *num_resif);
1875 				*resif = NULL;
1876 				*num_resif = 0;
1877 				return 0;
1878 			}
1879 		}
1880 	}
1881 	if(list) {
1882 		int idx = num_ifs;
1883 		for(p = list; p; p = p->next) {
1884 			(*resif)[idx] = strdup(p->str);
1885 			if(!((*resif)[idx])) {
1886 				log_err("out of memory");
1887 				config_del_strarray(*resif, *num_resif);
1888 				*resif = NULL;
1889 				*num_resif = 0;
1890 				return 0;
1891 			}
1892 			idx++;
1893 		}
1894 	}
1895 	return 1;
1896 #endif /* HAVE_GETIFADDRS */
1897 }
1898 
1899 struct listen_port*
1900 listening_ports_open(struct config_file* cfg, char** ifs, int num_ifs,
1901 	int* reuseport)
1902 {
1903 	struct listen_port* list = NULL;
1904 	struct addrinfo hints;
1905 	int i, do_ip4, do_ip6;
1906 	int do_tcp, do_auto;
1907 	do_ip4 = cfg->do_ip4;
1908 	do_ip6 = cfg->do_ip6;
1909 	do_tcp = cfg->do_tcp;
1910 	do_auto = cfg->if_automatic && cfg->do_udp;
1911 	if(cfg->incoming_num_tcp == 0)
1912 		do_tcp = 0;
1913 
1914 	/* getaddrinfo */
1915 	memset(&hints, 0, sizeof(hints));
1916 	hints.ai_flags = AI_PASSIVE;
1917 	/* no name lookups on our listening ports */
1918 	if(num_ifs > 0)
1919 		hints.ai_flags |= AI_NUMERICHOST;
1920 	hints.ai_family = AF_UNSPEC;
1921 #ifndef INET6
1922 	do_ip6 = 0;
1923 #endif
1924 	if(!do_ip4 && !do_ip6) {
1925 		return NULL;
1926 	}
1927 	/* create ip4 and ip6 ports so that return addresses are nice. */
1928 	if(do_auto || num_ifs == 0) {
1929 		if(do_auto && cfg->if_automatic_ports &&
1930 			cfg->if_automatic_ports[0]!=0) {
1931 			char* now = cfg->if_automatic_ports;
1932 			while(now && *now) {
1933 				char* after;
1934 				int extraport;
1935 				while(isspace((unsigned char)*now))
1936 					now++;
1937 				if(!*now)
1938 					break;
1939 				after = now;
1940 				extraport = (int)strtol(now, &after, 10);
1941 				if(extraport < 0 || extraport > 65535) {
1942 					log_err("interface-automatic-ports port number out of range, at position %d of '%s'", (int)(now-cfg->if_automatic_ports)+1, cfg->if_automatic_ports);
1943 					listening_ports_free(list);
1944 					return NULL;
1945 				}
1946 				if(extraport == 0 && now == after) {
1947 					log_err("interface-automatic-ports could not be parsed, at position %d of '%s'", (int)(now-cfg->if_automatic_ports)+1, cfg->if_automatic_ports);
1948 					listening_ports_free(list);
1949 					return NULL;
1950 				}
1951 				now = after;
1952 				if(do_ip6) {
1953 					hints.ai_family = AF_INET6;
1954 					if(!ports_create_if("::0",
1955 						do_auto, cfg->do_udp, do_tcp,
1956 						&hints, extraport, &list,
1957 						cfg->so_rcvbuf, cfg->so_sndbuf,
1958 						cfg->ssl_port, cfg->tls_additional_port,
1959 						cfg->https_port,
1960 						cfg->proxy_protocol_port,
1961 						reuseport, cfg->ip_transparent,
1962 						cfg->tcp_mss, cfg->ip_freebind,
1963 						cfg->http_nodelay, cfg->use_systemd,
1964 						cfg->dnscrypt_port, cfg->ip_dscp,
1965 						cfg->quic_port, cfg->http_notls_downstream,
1966 						cfg->sock_queue_timeout)) {
1967 						listening_ports_free(list);
1968 						return NULL;
1969 					}
1970 				}
1971 				if(do_ip4) {
1972 					hints.ai_family = AF_INET;
1973 					if(!ports_create_if("0.0.0.0",
1974 						do_auto, cfg->do_udp, do_tcp,
1975 						&hints, extraport, &list,
1976 						cfg->so_rcvbuf, cfg->so_sndbuf,
1977 						cfg->ssl_port, cfg->tls_additional_port,
1978 						cfg->https_port,
1979 						cfg->proxy_protocol_port,
1980 						reuseport, cfg->ip_transparent,
1981 						cfg->tcp_mss, cfg->ip_freebind,
1982 						cfg->http_nodelay, cfg->use_systemd,
1983 						cfg->dnscrypt_port, cfg->ip_dscp,
1984 						cfg->quic_port, cfg->http_notls_downstream,
1985 						cfg->sock_queue_timeout)) {
1986 						listening_ports_free(list);
1987 						return NULL;
1988 					}
1989 				}
1990 			}
1991 			return list;
1992 		}
1993 		if(do_ip6) {
1994 			hints.ai_family = AF_INET6;
1995 			if(!ports_create_if(do_auto?"::0":"::1",
1996 				do_auto, cfg->do_udp, do_tcp,
1997 				&hints, cfg->port, &list,
1998 				cfg->so_rcvbuf, cfg->so_sndbuf,
1999 				cfg->ssl_port, cfg->tls_additional_port,
2000 				cfg->https_port, cfg->proxy_protocol_port,
2001 				reuseport, cfg->ip_transparent,
2002 				cfg->tcp_mss, cfg->ip_freebind,
2003 				cfg->http_nodelay, cfg->use_systemd,
2004 				cfg->dnscrypt_port, cfg->ip_dscp,
2005 				cfg->quic_port, cfg->http_notls_downstream,
2006 				cfg->sock_queue_timeout)) {
2007 				listening_ports_free(list);
2008 				return NULL;
2009 			}
2010 		}
2011 		if(do_ip4) {
2012 			hints.ai_family = AF_INET;
2013 			if(!ports_create_if(do_auto?"0.0.0.0":"127.0.0.1",
2014 				do_auto, cfg->do_udp, do_tcp,
2015 				&hints, cfg->port, &list,
2016 				cfg->so_rcvbuf, cfg->so_sndbuf,
2017 				cfg->ssl_port, cfg->tls_additional_port,
2018 				cfg->https_port, cfg->proxy_protocol_port,
2019 				reuseport, cfg->ip_transparent,
2020 				cfg->tcp_mss, cfg->ip_freebind,
2021 				cfg->http_nodelay, cfg->use_systemd,
2022 				cfg->dnscrypt_port, cfg->ip_dscp,
2023 				cfg->quic_port, cfg->http_notls_downstream,
2024 				cfg->sock_queue_timeout)) {
2025 				listening_ports_free(list);
2026 				return NULL;
2027 			}
2028 		}
2029 	} else for(i = 0; i<num_ifs; i++) {
2030 		if(str_is_ip6(ifs[i])) {
2031 			if(!do_ip6)
2032 				continue;
2033 			hints.ai_family = AF_INET6;
2034 			if(!ports_create_if(ifs[i], 0, cfg->do_udp,
2035 				do_tcp, &hints, cfg->port, &list,
2036 				cfg->so_rcvbuf, cfg->so_sndbuf,
2037 				cfg->ssl_port, cfg->tls_additional_port,
2038 				cfg->https_port, cfg->proxy_protocol_port,
2039 				reuseport, cfg->ip_transparent,
2040 				cfg->tcp_mss, cfg->ip_freebind,
2041 				cfg->http_nodelay, cfg->use_systemd,
2042 				cfg->dnscrypt_port, cfg->ip_dscp,
2043 				cfg->quic_port, cfg->http_notls_downstream,
2044 				cfg->sock_queue_timeout)) {
2045 				listening_ports_free(list);
2046 				return NULL;
2047 			}
2048 		} else {
2049 			if(!do_ip4)
2050 				continue;
2051 			hints.ai_family = AF_INET;
2052 			if(!ports_create_if(ifs[i], 0, cfg->do_udp,
2053 				do_tcp, &hints, cfg->port, &list,
2054 				cfg->so_rcvbuf, cfg->so_sndbuf,
2055 				cfg->ssl_port, cfg->tls_additional_port,
2056 				cfg->https_port, cfg->proxy_protocol_port,
2057 				reuseport, cfg->ip_transparent,
2058 				cfg->tcp_mss, cfg->ip_freebind,
2059 				cfg->http_nodelay, cfg->use_systemd,
2060 				cfg->dnscrypt_port, cfg->ip_dscp,
2061 				cfg->quic_port, cfg->http_notls_downstream,
2062 				cfg->sock_queue_timeout)) {
2063 				listening_ports_free(list);
2064 				return NULL;
2065 			}
2066 		}
2067 	}
2068 
2069 	return list;
2070 }
2071 
2072 void listening_ports_free(struct listen_port* list)
2073 {
2074 	struct listen_port* nx;
2075 	while(list) {
2076 		nx = list->next;
2077 		if(list->fd != -1) {
2078 			sock_close(list->fd);
2079 		}
2080 		/* rc_ports don't have ub_socket */
2081 		if(list->socket) {
2082 			free(list->socket->addr);
2083 			free(list->socket);
2084 		}
2085 		free(list);
2086 		list = nx;
2087 	}
2088 }
2089 
2090 size_t listen_get_mem(struct listen_dnsport* listen)
2091 {
2092 	struct listen_list* p;
2093 	size_t s = sizeof(*listen) + sizeof(*listen->base) +
2094 		sizeof(*listen->udp_buff) +
2095 		sldns_buffer_capacity(listen->udp_buff);
2096 #ifdef USE_DNSCRYPT
2097 	s += sizeof(*listen->dnscrypt_udp_buff);
2098 	if(listen->udp_buff != listen->dnscrypt_udp_buff){
2099 		s += sldns_buffer_capacity(listen->dnscrypt_udp_buff);
2100 	}
2101 #endif
2102 	for(p = listen->cps; p; p = p->next) {
2103 		s += sizeof(*p);
2104 		s += comm_point_get_mem(p->com);
2105 	}
2106 	return s;
2107 }
2108 
2109 void listen_stop_accept(struct listen_dnsport* listen)
2110 {
2111 	/* do not stop the ones that have no tcp_free list
2112 	 * (they have already stopped listening) */
2113 	struct listen_list* p;
2114 	for(p=listen->cps; p; p=p->next) {
2115 		if(p->com->type == comm_tcp_accept &&
2116 			p->com->tcp_free != NULL) {
2117 			comm_point_stop_listening(p->com);
2118 		}
2119 	}
2120 }
2121 
2122 void listen_start_accept(struct listen_dnsport* listen)
2123 {
2124 	/* do not start the ones that have no tcp_free list, it is no
2125 	 * use to listen to them because they have no free tcp handlers */
2126 	struct listen_list* p;
2127 	for(p=listen->cps; p; p=p->next) {
2128 		if(p->com->type == comm_tcp_accept &&
2129 			p->com->tcp_free != NULL) {
2130 			comm_point_start_listening(p->com, -1, -1);
2131 		}
2132 	}
2133 }
2134 
2135 struct tcp_req_info*
2136 tcp_req_info_create(struct comm_base* base, struct sldns_buffer* spoolbuf)
2137 {
2138 	struct tcp_req_info* req = (struct tcp_req_info*)malloc(sizeof(*req));
2139 	if(!req) {
2140 		log_err("malloc failure for new stream outoforder processing structure");
2141 		return NULL;
2142 	}
2143 	memset(req, 0, sizeof(*req));
2144 	req->read_again_timer = comm_timer_create(base, tcp_read_again_cb, req);
2145 	if(!req->read_again_timer) {
2146 		log_err("malloc failure");
2147 		free(req);
2148 		return NULL;
2149 	}
2150 	req->spool_buffer = spoolbuf;
2151 	return req;
2152 }
2153 
2154 void
2155 tcp_req_info_delete(struct tcp_req_info* req)
2156 {
2157 	if(!req) return;
2158 	tcp_req_info_clear(req);
2159 	comm_timer_delete(req->read_again_timer);
2160 	/* cp is pointer back to commpoint that owns this struct and
2161 	 * called delete on us */
2162 	/* spool_buffer is shared udp buffer, not deleted here */
2163 	free(req);
2164 }
2165 
2166 void tcp_req_info_clear(struct tcp_req_info* req)
2167 {
2168 	struct tcp_req_open_item* open, *nopen;
2169 	struct tcp_req_done_item* item, *nitem;
2170 	if(!req) return;
2171 
2172 	/* free outstanding request mesh reply entries */
2173 	open = req->open_req_list;
2174 	while(open) {
2175 		nopen = open->next;
2176 		mesh_state_remove_reply(open->mesh, open->mesh_state, req->cp,
2177 			NULL, NULL);
2178 		free(open);
2179 		open = nopen;
2180 	}
2181 	req->open_req_list = NULL;
2182 	req->num_open_req = 0;
2183 
2184 	/* free pending writable result packets */
2185 	item = req->done_req_list;
2186 	while(item) {
2187 		nitem = item->next;
2188 		lock_basic_lock(&stream_wait_count_lock);
2189 		stream_wait_count -= (sizeof(struct tcp_req_done_item)
2190 			+item->len);
2191 		lock_basic_unlock(&stream_wait_count_lock);
2192 		free(item->buf);
2193 		free(item);
2194 		item = nitem;
2195 	}
2196 	req->done_req_list = NULL;
2197 	req->num_done_req = 0;
2198 	req->read_is_closed = 0;
2199 
2200 	if(comm_timer_is_set(req->read_again_timer))
2201 		comm_timer_disable(req->read_again_timer);
2202 }
2203 
2204 void
2205 tcp_req_info_remove_mesh_state(struct tcp_req_info* req, struct mesh_state* m)
2206 {
2207 	struct tcp_req_open_item* open, *prev = NULL;
2208 	if(!req || !m) return;
2209 	open = req->open_req_list;
2210 	while(open) {
2211 		if(open->mesh_state == m) {
2212 			struct tcp_req_open_item* next;
2213 			if(prev) prev->next = open->next;
2214 			else req->open_req_list = open->next;
2215 			/* caller has to manage the mesh state reply entry */
2216 			next = open->next;
2217 			free(open);
2218 			req->num_open_req --;
2219 
2220 			/* prev = prev; */
2221 			open = next;
2222 			continue;
2223 		}
2224 		prev = open;
2225 		open = open->next;
2226 	}
2227 }
2228 
2229 /** setup listening for read or write */
2230 static void
2231 tcp_req_info_setup_listen(struct tcp_req_info* req)
2232 {
2233 	int wr = 0;
2234 	int rd = 0;
2235 
2236 	if(req->cp->tcp_byte_count != 0) {
2237 		/* cannot change, halfway through */
2238 		return;
2239 	}
2240 
2241 	if(!req->cp->tcp_is_reading)
2242 		wr = 1;
2243 	if(!req->read_is_closed)
2244 		rd = 1;
2245 
2246 	if(wr) {
2247 		req->cp->tcp_is_reading = 0;
2248 		comm_point_stop_listening(req->cp);
2249 		comm_point_start_listening(req->cp, -1,
2250 			adjusted_tcp_timeout(req->cp));
2251 	} else if(rd) {
2252 		req->cp->tcp_is_reading = 1;
2253 		comm_point_stop_listening(req->cp);
2254 		comm_point_start_listening(req->cp, -1,
2255 			adjusted_tcp_timeout(req->cp));
2256 		/* and also read it (from SSL stack buffers), so
2257 		 * no event read event is expected since the remainder of
2258 		 * the TLS frame is sitting in the buffers. */
2259 		req->read_again = 1;
2260 	} else {
2261 		comm_point_stop_listening(req->cp);
2262 		comm_point_start_listening(req->cp, -1,
2263 			adjusted_tcp_timeout(req->cp));
2264 		comm_point_listen_for_rw(req->cp, 0, 0);
2265 	}
2266 }
2267 
2268 /** remove first item from list of pending results */
2269 static struct tcp_req_done_item*
2270 tcp_req_info_pop_done(struct tcp_req_info* req)
2271 {
2272 	struct tcp_req_done_item* item;
2273 	log_assert(req->num_done_req > 0 && req->done_req_list);
2274 	item = req->done_req_list;
2275 	lock_basic_lock(&stream_wait_count_lock);
2276 	stream_wait_count -= (sizeof(struct tcp_req_done_item)+item->len);
2277 	lock_basic_unlock(&stream_wait_count_lock);
2278 	req->done_req_list = req->done_req_list->next;
2279 	req->num_done_req --;
2280 	return item;
2281 }
2282 
2283 /** Send given buffer and setup to write */
2284 static void
2285 tcp_req_info_start_write_buf(struct tcp_req_info* req, uint8_t* buf,
2286 	size_t len)
2287 {
2288 	sldns_buffer_clear(req->cp->buffer);
2289 	sldns_buffer_write(req->cp->buffer, buf, len);
2290 	sldns_buffer_flip(req->cp->buffer);
2291 
2292 	req->cp->tcp_is_reading = 0; /* we are now writing */
2293 }
2294 
2295 /** pick up the next result and start writing it to the channel */
2296 static void
2297 tcp_req_pickup_next_result(struct tcp_req_info* req)
2298 {
2299 	if(req->num_done_req > 0) {
2300 		/* unlist the done item from the list of pending results */
2301 		struct tcp_req_done_item* item = tcp_req_info_pop_done(req);
2302 		tcp_req_info_start_write_buf(req, item->buf, item->len);
2303 		free(item->buf);
2304 		free(item);
2305 	}
2306 }
2307 
2308 /** the read channel has closed */
2309 int
2310 tcp_req_info_handle_read_close(struct tcp_req_info* req)
2311 {
2312 	verbose(VERB_ALGO, "tcp channel read side closed %d", req->cp->fd);
2313 	/* RFC 7766 6.2.4 says to drop pending replies when client closes. */
2314 	return 0; /* drop connection */
2315 }
2316 
2317 void
2318 tcp_req_info_handle_writedone(struct tcp_req_info* req)
2319 {
2320 	/* back to reading state, we finished this write event */
2321 	sldns_buffer_clear(req->cp->buffer);
2322 	if(req->num_done_req == 0 && req->read_is_closed) {
2323 		/* no more to write and nothing to read, close it */
2324 		comm_point_drop_reply(&req->cp->repinfo);
2325 		return;
2326 	}
2327 	req->cp->tcp_is_reading = 1;
2328 	/* see if another result needs writing */
2329 	tcp_req_pickup_next_result(req);
2330 
2331 	/* see if there is more to write, if not stop_listening for writing */
2332 	/* see if new requests are allowed, if so, start_listening
2333 	 * for reading */
2334 	tcp_req_info_setup_listen(req);
2335 }
2336 
2337 void
2338 tcp_req_info_handle_readdone(struct tcp_req_info* req)
2339 {
2340 	struct comm_point* c = req->cp;
2341 
2342 	/* we want to read up several requests, unless there are
2343 	 * pending answers */
2344 
2345 	req->is_drop = 0;
2346 	req->is_reply = 0;
2347 	req->in_worker_handle = 1;
2348 	sldns_buffer_set_limit(req->spool_buffer, 0);
2349 	/* handle the current request */
2350 	/* this calls the worker handle request routine that could give
2351 	 * a cache response, or localdata response, or drop the reply,
2352 	 * or schedule a mesh entry for later */
2353 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2354 	if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
2355 		req->in_worker_handle = 0;
2356 		/* there is an answer, put it up.  It is already in the
2357 		 * c->buffer, just send it. */
2358 		/* since we were just reading a query, the channel is
2359 		 * clear to write to */
2360 	send_it:
2361 		c->tcp_is_reading = 0;
2362 		comm_point_stop_listening(c);
2363 		comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
2364 		return;
2365 	}
2366 	req->in_worker_handle = 0;
2367 	/* it should be waiting in the mesh for recursion.
2368 	 * If mesh failed to add a new entry and called commpoint_drop_reply.
2369 	 * Then the mesh state has been cleared. */
2370 	if(req->is_drop) {
2371 		/* the reply has been dropped, stream has been closed. */
2372 		return;
2373 	}
2374 	/* If mesh failed(mallocfail) and called commpoint_send_reply with
2375 	 * something like servfail then we pick up that reply below. */
2376 	if(req->is_reply) {
2377 		goto send_it;
2378 	}
2379 
2380 	sldns_buffer_clear(c->buffer);
2381 	/* if pending answers, pick up an answer and start sending it */
2382 	tcp_req_pickup_next_result(req);
2383 
2384 	/* if answers pending, start sending answers */
2385 	/* read more requests if we can have more requests */
2386 	tcp_req_info_setup_listen(req);
2387 }
2388 
2389 int
2390 tcp_req_info_add_meshstate(struct tcp_req_info* req,
2391 	struct mesh_area* mesh, struct mesh_state* m)
2392 {
2393 	struct tcp_req_open_item* item;
2394 	log_assert(req && mesh && m);
2395 	item = (struct tcp_req_open_item*)malloc(sizeof(*item));
2396 	if(!item) return 0;
2397 	item->next = req->open_req_list;
2398 	item->mesh = mesh;
2399 	item->mesh_state = m;
2400 	req->open_req_list = item;
2401 	req->num_open_req++;
2402 	return 1;
2403 }
2404 
2405 /** Add a result to the result list.  At the end. */
2406 static int
2407 tcp_req_info_add_result(struct tcp_req_info* req, uint8_t* buf, size_t len)
2408 {
2409 	struct tcp_req_done_item* last = NULL;
2410 	struct tcp_req_done_item* item;
2411 	size_t space;
2412 
2413 	/* see if we have space */
2414 	space = sizeof(struct tcp_req_done_item) + len;
2415 	lock_basic_lock(&stream_wait_count_lock);
2416 	if(stream_wait_count + space > stream_wait_max) {
2417 		lock_basic_unlock(&stream_wait_count_lock);
2418 		verbose(VERB_ALGO, "drop stream reply, no space left, in stream-wait-size");
2419 		return 0;
2420 	}
2421 	stream_wait_count += space;
2422 	lock_basic_unlock(&stream_wait_count_lock);
2423 
2424 	/* find last element */
2425 	last = req->done_req_list;
2426 	while(last && last->next)
2427 		last = last->next;
2428 
2429 	/* create new element */
2430 	item = (struct tcp_req_done_item*)malloc(sizeof(*item));
2431 	if(!item) {
2432 		log_err("malloc failure, for stream result list");
2433 		return 0;
2434 	}
2435 	item->next = NULL;
2436 	item->len = len;
2437 	item->buf = memdup(buf, len);
2438 	if(!item->buf) {
2439 		free(item);
2440 		log_err("malloc failure, adding reply to stream result list");
2441 		return 0;
2442 	}
2443 
2444 	/* link in */
2445 	if(last) last->next = item;
2446 	else req->done_req_list = item;
2447 	req->num_done_req++;
2448 	return 1;
2449 }
2450 
2451 void
2452 tcp_req_info_send_reply(struct tcp_req_info* req)
2453 {
2454 	if(req->in_worker_handle) {
2455 		/* reply from mesh is in the spool_buffer */
2456 		/* copy now, so that the spool buffer is free for other tasks
2457 		 * before the callback is done */
2458 		sldns_buffer_clear(req->cp->buffer);
2459 		sldns_buffer_write(req->cp->buffer,
2460 			sldns_buffer_begin(req->spool_buffer),
2461 			sldns_buffer_limit(req->spool_buffer));
2462 		sldns_buffer_flip(req->cp->buffer);
2463 		req->is_reply = 1;
2464 		return;
2465 	}
2466 	/* now that the query has been handled, that mesh_reply entry
2467 	 * should be removed, from the tcp_req_info list,
2468 	 * the mesh state cleanup removes then with region_cleanup and
2469 	 * replies_sent true. */
2470 	/* see if we can send it straight away (we are not doing
2471 	 * anything else).  If so, copy to buffer and start */
2472 	if(req->cp->tcp_is_reading && req->cp->tcp_byte_count == 0) {
2473 		/* buffer is free, and was ready to read new query into,
2474 		 * but we are now going to use it to send this answer */
2475 		tcp_req_info_start_write_buf(req,
2476 			sldns_buffer_begin(req->spool_buffer),
2477 			sldns_buffer_limit(req->spool_buffer));
2478 		/* switch to listen to write events */
2479 		comm_point_stop_listening(req->cp);
2480 		comm_point_start_listening(req->cp, -1,
2481 			adjusted_tcp_timeout(req->cp));
2482 		return;
2483 	}
2484 	/* queue up the answer behind the others already pending */
2485 	if(!tcp_req_info_add_result(req, sldns_buffer_begin(req->spool_buffer),
2486 		sldns_buffer_limit(req->spool_buffer))) {
2487 		/* drop the connection, we are out of resources */
2488 		comm_point_drop_reply(&req->cp->repinfo);
2489 	}
2490 }
2491 
2492 size_t tcp_req_info_get_stream_buffer_size(void)
2493 {
2494 	size_t s;
2495 	if(!stream_wait_lock_inited)
2496 		return stream_wait_count;
2497 	lock_basic_lock(&stream_wait_count_lock);
2498 	s = stream_wait_count;
2499 	lock_basic_unlock(&stream_wait_count_lock);
2500 	return s;
2501 }
2502 
2503 size_t http2_get_query_buffer_size(void)
2504 {
2505 	size_t s;
2506 	if(!http2_query_buffer_lock_inited)
2507 		return http2_query_buffer_count;
2508 	lock_basic_lock(&http2_query_buffer_count_lock);
2509 	s = http2_query_buffer_count;
2510 	lock_basic_unlock(&http2_query_buffer_count_lock);
2511 	return s;
2512 }
2513 
2514 size_t http2_get_response_buffer_size(void)
2515 {
2516 	size_t s;
2517 	if(!http2_response_buffer_lock_inited)
2518 		return http2_response_buffer_count;
2519 	lock_basic_lock(&http2_response_buffer_count_lock);
2520 	s = http2_response_buffer_count;
2521 	lock_basic_unlock(&http2_response_buffer_count_lock);
2522 	return s;
2523 }
2524 
2525 #ifdef HAVE_NGHTTP2
2526 /** nghttp2 callback. Used to copy response from rbuffer to nghttp2 session */
2527 static ssize_t http2_submit_response_read_callback(
2528 	nghttp2_session* ATTR_UNUSED(session),
2529 	int32_t stream_id, uint8_t* buf, size_t length, uint32_t* data_flags,
2530 	nghttp2_data_source* source, void* ATTR_UNUSED(cb_arg))
2531 {
2532 	struct http2_stream* h2_stream;
2533 	struct http2_session* h2_session = source->ptr;
2534 	size_t copylen = length;
2535 	if(!(h2_stream = nghttp2_session_get_stream_user_data(
2536 		h2_session->session, stream_id))) {
2537 		verbose(VERB_QUERY, "http2: cannot get stream data, closing "
2538 			"stream");
2539 		return NGHTTP2_ERR_TEMPORAL_CALLBACK_FAILURE;
2540 	}
2541 	if(!h2_stream->rbuffer ||
2542 		sldns_buffer_remaining(h2_stream->rbuffer) == 0) {
2543 		verbose(VERB_QUERY, "http2: cannot submit buffer. No data "
2544 			"available in rbuffer");
2545 		/* rbuffer will be free'd in frame close cb */
2546 		return NGHTTP2_ERR_TEMPORAL_CALLBACK_FAILURE;
2547 	}
2548 
2549 	if(copylen > sldns_buffer_remaining(h2_stream->rbuffer))
2550 		copylen = sldns_buffer_remaining(h2_stream->rbuffer);
2551 	if(copylen > SSIZE_MAX)
2552 		copylen = SSIZE_MAX; /* will probably never happen */
2553 
2554 	memcpy(buf, sldns_buffer_current(h2_stream->rbuffer), copylen);
2555 	sldns_buffer_skip(h2_stream->rbuffer, copylen);
2556 
2557 	if(sldns_buffer_remaining(h2_stream->rbuffer) == 0) {
2558 		*data_flags |= NGHTTP2_DATA_FLAG_EOF;
2559 		lock_basic_lock(&http2_response_buffer_count_lock);
2560 		http2_response_buffer_count -=
2561 			sldns_buffer_capacity(h2_stream->rbuffer);
2562 		lock_basic_unlock(&http2_response_buffer_count_lock);
2563 		sldns_buffer_free(h2_stream->rbuffer);
2564 		h2_stream->rbuffer = NULL;
2565 	}
2566 
2567 	return copylen;
2568 }
2569 
2570 /**
2571  * Send RST_STREAM frame for stream.
2572  * @param h2_session: http2 session to submit frame to
2573  * @param h2_stream: http2 stream containing frame ID to use in RST_STREAM
2574  * @return 0 on error, 1 otherwise
2575  */
2576 static int http2_submit_rst_stream(struct http2_session* h2_session,
2577 		struct http2_stream* h2_stream)
2578 {
2579 	int ret = nghttp2_submit_rst_stream(h2_session->session,
2580 		NGHTTP2_FLAG_NONE, h2_stream->stream_id,
2581 		NGHTTP2_INTERNAL_ERROR);
2582 	if(ret) {
2583 		verbose(VERB_QUERY, "http2: nghttp2_submit_rst_stream failed, "
2584 			"error: %s", nghttp2_strerror(ret));
2585 		return 0;
2586 	}
2587 	return 1;
2588 }
2589 
2590 /**
2591  * DNS response ready to be submitted to nghttp2, to be prepared for sending
2592  * out. Response is stored in c->buffer. Copy to rbuffer because the c->buffer
2593  * might be used before this will be sent out.
2594  * @param h2_session: http2 session, containing c->buffer which contains answer
2595  * @return 0 on error, 1 otherwise
2596  */
2597 int http2_submit_dns_response(struct http2_session* h2_session)
2598 {
2599 	int ret;
2600 	nghttp2_data_provider data_prd;
2601 	char status[4];
2602 	nghttp2_nv headers[3];
2603 	struct http2_stream* h2_stream = h2_session->c->h2_stream;
2604 	size_t rlen;
2605 	char rlen_str[32];
2606 
2607 	if(h2_stream->rbuffer) {
2608 		log_err("http2 submit response error: rbuffer already "
2609 			"exists");
2610 		return 0;
2611 	}
2612 	if(sldns_buffer_remaining(h2_session->c->buffer) == 0) {
2613 		log_err("http2 submit response error: c->buffer not complete");
2614 		return 0;
2615 	}
2616 
2617 	if(snprintf(status, 4, "%d", h2_stream->status) != 3) {
2618 		verbose(VERB_QUERY, "http2: submit response error: "
2619 			"invalid status");
2620 		return 0;
2621 	}
2622 
2623 	rlen = sldns_buffer_remaining(h2_session->c->buffer);
2624 	snprintf(rlen_str, sizeof(rlen_str), "%u", (unsigned)rlen);
2625 
2626 	lock_basic_lock(&http2_response_buffer_count_lock);
2627 	if(http2_response_buffer_count + rlen > http2_response_buffer_max) {
2628 		lock_basic_unlock(&http2_response_buffer_count_lock);
2629 		verbose(VERB_ALGO, "reset HTTP2 stream, no space left, "
2630 			"in https-response-buffer-size");
2631 		return http2_submit_rst_stream(h2_session, h2_stream);
2632 	}
2633 	http2_response_buffer_count += rlen;
2634 	lock_basic_unlock(&http2_response_buffer_count_lock);
2635 
2636 	if(!(h2_stream->rbuffer = sldns_buffer_new(rlen))) {
2637 		lock_basic_lock(&http2_response_buffer_count_lock);
2638 		http2_response_buffer_count -= rlen;
2639 		lock_basic_unlock(&http2_response_buffer_count_lock);
2640 		log_err("http2 submit response error: malloc failure");
2641 		return 0;
2642 	}
2643 
2644 	headers[0].name = (uint8_t*)":status";
2645 	headers[0].namelen = 7;
2646 	headers[0].value = (uint8_t*)status;
2647 	headers[0].valuelen = 3;
2648 	headers[0].flags = NGHTTP2_NV_FLAG_NONE;
2649 
2650 	headers[1].name = (uint8_t*)"content-type";
2651 	headers[1].namelen = 12;
2652 	headers[1].value = (uint8_t*)"application/dns-message";
2653 	headers[1].valuelen = 23;
2654 	headers[1].flags = NGHTTP2_NV_FLAG_NONE;
2655 
2656 	headers[2].name = (uint8_t*)"content-length";
2657 	headers[2].namelen = 14;
2658 	headers[2].value = (uint8_t*)rlen_str;
2659 	headers[2].valuelen = strlen(rlen_str);
2660 	headers[2].flags = NGHTTP2_NV_FLAG_NONE;
2661 
2662 	sldns_buffer_write(h2_stream->rbuffer,
2663 		sldns_buffer_current(h2_session->c->buffer),
2664 		sldns_buffer_remaining(h2_session->c->buffer));
2665 	sldns_buffer_flip(h2_stream->rbuffer);
2666 
2667 	data_prd.source.ptr = h2_session;
2668 	data_prd.read_callback = http2_submit_response_read_callback;
2669 	ret = nghttp2_submit_response(h2_session->session, h2_stream->stream_id,
2670 		headers, 3, &data_prd);
2671 	if(ret) {
2672 		verbose(VERB_QUERY, "http2: set_stream_user_data failed, "
2673 			"error: %s", nghttp2_strerror(ret));
2674 		return 0;
2675 	}
2676 	return 1;
2677 }
2678 #else
2679 int http2_submit_dns_response(void* ATTR_UNUSED(v))
2680 {
2681 	return 0;
2682 }
2683 #endif
2684 
2685 #ifdef HAVE_NGHTTP2
2686 /** HTTP status to descriptive string */
2687 static char* http_status_to_str(enum http_status s)
2688 {
2689 	switch(s) {
2690 		case HTTP_STATUS_OK:
2691 			return "OK";
2692 		case HTTP_STATUS_BAD_REQUEST:
2693 			return "Bad Request";
2694 		case HTTP_STATUS_NOT_FOUND:
2695 			return "Not Found";
2696 		case HTTP_STATUS_PAYLOAD_TOO_LARGE:
2697 			return "Payload Too Large";
2698 		case HTTP_STATUS_URI_TOO_LONG:
2699 			return "URI Too Long";
2700 		case HTTP_STATUS_UNSUPPORTED_MEDIA_TYPE:
2701 			return "Unsupported Media Type";
2702 		case HTTP_STATUS_NOT_IMPLEMENTED:
2703 			return "Not Implemented";
2704 	}
2705 	return "Status Unknown";
2706 }
2707 
2708 /** nghttp2 callback. Used to copy error message to nghttp2 session */
2709 static ssize_t http2_submit_error_read_callback(
2710 	nghttp2_session* ATTR_UNUSED(session),
2711 	int32_t stream_id, uint8_t* buf, size_t length, uint32_t* data_flags,
2712 	nghttp2_data_source* source, void* ATTR_UNUSED(cb_arg))
2713 {
2714 	struct http2_stream* h2_stream;
2715 	struct http2_session* h2_session = source->ptr;
2716 	char* msg;
2717 	if(!(h2_stream = nghttp2_session_get_stream_user_data(
2718 		h2_session->session, stream_id))) {
2719 		verbose(VERB_QUERY, "http2: cannot get stream data, closing "
2720 			"stream");
2721 		return NGHTTP2_ERR_TEMPORAL_CALLBACK_FAILURE;
2722 	}
2723 	*data_flags |= NGHTTP2_DATA_FLAG_EOF;
2724 	msg = http_status_to_str(h2_stream->status);
2725 	if(length < strlen(msg))
2726 		return 0; /* not worth trying over multiple frames */
2727 	memcpy(buf, msg, strlen(msg));
2728 	return strlen(msg);
2729 
2730 }
2731 
2732 /**
2733  * HTTP error response ready to be submitted to nghttp2, to be prepared for
2734  * sending out. Message body will contain descriptive string for HTTP status.
2735  * @param h2_session: http2 session to submit to
2736  * @param h2_stream: http2 stream containing HTTP status to use for error
2737  * @return 0 on error, 1 otherwise
2738  */
2739 static int http2_submit_error(struct http2_session* h2_session,
2740 	struct http2_stream* h2_stream)
2741 {
2742 	int ret;
2743 	char status[4];
2744 	nghttp2_data_provider data_prd;
2745 	nghttp2_nv headers[1]; /* will be copied by nghttp */
2746 	if(snprintf(status, 4, "%d", h2_stream->status) != 3) {
2747 		verbose(VERB_QUERY, "http2: submit error failed, "
2748 			"invalid status");
2749 		return 0;
2750 	}
2751 	headers[0].name = (uint8_t*)":status";
2752 	headers[0].namelen = 7;
2753 	headers[0].value = (uint8_t*)status;
2754 	headers[0].valuelen = 3;
2755 	headers[0].flags = NGHTTP2_NV_FLAG_NONE;
2756 
2757 	data_prd.source.ptr = h2_session;
2758 	data_prd.read_callback = http2_submit_error_read_callback;
2759 
2760 	ret = nghttp2_submit_response(h2_session->session, h2_stream->stream_id,
2761 		headers, 1, &data_prd);
2762 	if(ret) {
2763 		verbose(VERB_QUERY, "http2: submit error failed, "
2764 			"error: %s", nghttp2_strerror(ret));
2765 		return 0;
2766 	}
2767 	return 1;
2768 }
2769 
2770 /**
2771  * Start query handling. Query is stored in the stream, and will be free'd here.
2772  * @param h2_session: http2 session, containing comm point
2773  * @param h2_stream: stream containing buffered query
2774  * @return: -1 on error, 1 if answer is stored in c->buffer, 0 if there is no
2775  * reply available (yet).
2776  */
2777 static int http2_query_read_done(struct http2_session* h2_session,
2778 	struct http2_stream* h2_stream)
2779 {
2780 	log_assert(h2_stream->qbuffer);
2781 
2782 	if(h2_session->c->h2_stream) {
2783 		verbose(VERB_ALGO, "http2_query_read_done failure: shared "
2784 			"buffer already assigned to stream");
2785 		return -1;
2786 	}
2787 
2788     /* the c->buffer might be used by mesh_send_reply and no be cleard
2789 	 * need to be cleared before use */
2790 	sldns_buffer_clear(h2_session->c->buffer);
2791 	if(sldns_buffer_remaining(h2_session->c->buffer) <
2792 		sldns_buffer_remaining(h2_stream->qbuffer)) {
2793 		/* qbuffer will be free'd in frame close cb */
2794 		sldns_buffer_clear(h2_session->c->buffer);
2795 		verbose(VERB_ALGO, "http2_query_read_done failure: can't fit "
2796 			"qbuffer in c->buffer");
2797 		return -1;
2798 	}
2799 
2800 	sldns_buffer_write(h2_session->c->buffer,
2801 		sldns_buffer_current(h2_stream->qbuffer),
2802 		sldns_buffer_remaining(h2_stream->qbuffer));
2803 
2804 	lock_basic_lock(&http2_query_buffer_count_lock);
2805 	http2_query_buffer_count -= sldns_buffer_capacity(h2_stream->qbuffer);
2806 	lock_basic_unlock(&http2_query_buffer_count_lock);
2807 	sldns_buffer_free(h2_stream->qbuffer);
2808 	h2_stream->qbuffer = NULL;
2809 
2810 	sldns_buffer_flip(h2_session->c->buffer);
2811 	h2_session->c->h2_stream = h2_stream;
2812 	fptr_ok(fptr_whitelist_comm_point(h2_session->c->callback));
2813 	if((*h2_session->c->callback)(h2_session->c, h2_session->c->cb_arg,
2814 		NETEVENT_NOERROR, &h2_session->c->repinfo)) {
2815 		return 1; /* answer in c->buffer */
2816 	}
2817 	sldns_buffer_clear(h2_session->c->buffer);
2818 	h2_session->c->h2_stream = NULL;
2819 	return 0; /* mesh state added, or dropped */
2820 }
2821 
2822 /** nghttp2 callback. Used to check if the received frame indicates the end of a
2823  * stream. Gather collected request data and start query handling. */
2824 static int http2_req_frame_recv_cb(nghttp2_session* session,
2825 	const nghttp2_frame* frame, void* cb_arg)
2826 {
2827 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
2828 	struct http2_stream* h2_stream;
2829 	int query_read_done;
2830 
2831 	if((frame->hd.type != NGHTTP2_DATA &&
2832 		frame->hd.type != NGHTTP2_HEADERS) ||
2833 		!(frame->hd.flags & NGHTTP2_FLAG_END_STREAM)) {
2834 			return 0;
2835 	}
2836 
2837 	if(!(h2_stream = nghttp2_session_get_stream_user_data(
2838 		session, frame->hd.stream_id)))
2839 		return 0;
2840 
2841 	if(h2_stream->invalid_endpoint) {
2842 		h2_stream->status = HTTP_STATUS_NOT_FOUND;
2843 		goto submit_http_error;
2844 	}
2845 
2846 	if(h2_stream->invalid_content_type) {
2847 		h2_stream->status = HTTP_STATUS_UNSUPPORTED_MEDIA_TYPE;
2848 		goto submit_http_error;
2849 	}
2850 
2851 	if(h2_stream->http_method != HTTP_METHOD_GET &&
2852 		h2_stream->http_method != HTTP_METHOD_POST) {
2853 		h2_stream->status = HTTP_STATUS_NOT_IMPLEMENTED;
2854 		goto submit_http_error;
2855 	}
2856 
2857 	if(h2_stream->query_too_large) {
2858 		if(h2_stream->http_method == HTTP_METHOD_POST)
2859 			h2_stream->status = HTTP_STATUS_PAYLOAD_TOO_LARGE;
2860 		else
2861 			h2_stream->status = HTTP_STATUS_URI_TOO_LONG;
2862 		goto submit_http_error;
2863 	}
2864 
2865 	if(!h2_stream->qbuffer) {
2866 		h2_stream->status = HTTP_STATUS_BAD_REQUEST;
2867 		goto submit_http_error;
2868 	}
2869 
2870 	if(h2_stream->status) {
2871 submit_http_error:
2872 		verbose(VERB_QUERY, "http2 request invalid, returning :status="
2873 			"%d", h2_stream->status);
2874 		if(!http2_submit_error(h2_session, h2_stream)) {
2875 			return NGHTTP2_ERR_CALLBACK_FAILURE;
2876 		}
2877 		return 0;
2878 	}
2879 	h2_stream->status = HTTP_STATUS_OK;
2880 
2881 	sldns_buffer_flip(h2_stream->qbuffer);
2882 	h2_session->postpone_drop = 1;
2883 	query_read_done = http2_query_read_done(h2_session, h2_stream);
2884 	h2_session->postpone_drop = 0;
2885 	if(query_read_done < 0)
2886 		return NGHTTP2_ERR_CALLBACK_FAILURE;
2887 	else if(!query_read_done) {
2888 		if(h2_session->is_drop) {
2889 			/* connection needs to be closed. Return failure to make
2890 			 * sure no other action are taken anymore on comm point.
2891 			 * failure will result in reclaiming (and closing)
2892 			 * of comm point. */
2893 			verbose(VERB_QUERY, "http2 query dropped in worker cb");
2894 			return NGHTTP2_ERR_CALLBACK_FAILURE;
2895 		}
2896 		/* nothing to submit right now, query added to mesh. */
2897 		return 0;
2898 	}
2899 	if(!http2_submit_dns_response(h2_session)) {
2900 		sldns_buffer_clear(h2_session->c->buffer);
2901 		h2_session->c->h2_stream = NULL;
2902 		return NGHTTP2_ERR_CALLBACK_FAILURE;
2903 	}
2904 	verbose(VERB_QUERY, "http2 query submitted to session");
2905 	sldns_buffer_clear(h2_session->c->buffer);
2906 	h2_session->c->h2_stream = NULL;
2907 	return 0;
2908 }
2909 
2910 /** nghttp2 callback. Used to detect start of new streams. */
2911 static int http2_req_begin_headers_cb(nghttp2_session* session,
2912 	const nghttp2_frame* frame, void* cb_arg)
2913 {
2914 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
2915 	struct http2_stream* h2_stream;
2916 	int ret;
2917 	if(frame->hd.type != NGHTTP2_HEADERS ||
2918 		frame->headers.cat != NGHTTP2_HCAT_REQUEST) {
2919 		/* only interested in request headers */
2920 		return 0;
2921 	}
2922 	if(!(h2_stream = http2_stream_create(frame->hd.stream_id))) {
2923 		log_err("malloc failure while creating http2 stream");
2924 		return NGHTTP2_ERR_CALLBACK_FAILURE;
2925 	}
2926 	http2_session_add_stream(h2_session, h2_stream);
2927 	ret = nghttp2_session_set_stream_user_data(session,
2928 		frame->hd.stream_id, h2_stream);
2929 	if(ret) {
2930 		/* stream does not exist */
2931 		verbose(VERB_QUERY, "http2: set_stream_user_data failed, "
2932 			"error: %s", nghttp2_strerror(ret));
2933 		return NGHTTP2_ERR_CALLBACK_FAILURE;
2934 	}
2935 
2936 	return 0;
2937 }
2938 
2939 /**
2940  * base64url decode, store in qbuffer
2941  * @param h2_session: http2 session
2942  * @param h2_stream: http2 stream
2943  * @param start: start of the base64 string
2944  * @param length: length of the base64 string
2945  * @return: 0 on error, 1 otherwise. query will be stored in h2_stream->qbuffer,
2946  * buffer will be NULL is unparseble.
2947  */
2948 static int http2_buffer_uri_query(struct http2_session* h2_session,
2949 	struct http2_stream* h2_stream, const uint8_t* start, size_t length)
2950 {
2951 	size_t expectb64len;
2952 	int b64len;
2953 	if(h2_stream->http_method == HTTP_METHOD_POST)
2954 		return 1;
2955 	if(length == 0)
2956 		return 1;
2957 	if(h2_stream->qbuffer) {
2958 		verbose(VERB_ALGO, "http2_req_header fail, "
2959 			"qbuffer already set");
2960 		return 0;
2961 	}
2962 
2963 	/* calculate size, might be a bit bigger than the real
2964 	 * decoded buffer size */
2965 	expectb64len = sldns_b64_pton_calculate_size(length);
2966 	log_assert(expectb64len > 0);
2967 	if(expectb64len >
2968 		h2_session->c->http2_stream_max_qbuffer_size) {
2969 		h2_stream->query_too_large = 1;
2970 		return 1;
2971 	}
2972 
2973 	lock_basic_lock(&http2_query_buffer_count_lock);
2974 	if(http2_query_buffer_count + expectb64len > http2_query_buffer_max) {
2975 		lock_basic_unlock(&http2_query_buffer_count_lock);
2976 		verbose(VERB_ALGO, "reset HTTP2 stream, no space left, "
2977 			"in http2-query-buffer-size");
2978 		return http2_submit_rst_stream(h2_session, h2_stream);
2979 	}
2980 	http2_query_buffer_count += expectb64len;
2981 	lock_basic_unlock(&http2_query_buffer_count_lock);
2982 	if(!(h2_stream->qbuffer = sldns_buffer_new(expectb64len))) {
2983 		lock_basic_lock(&http2_query_buffer_count_lock);
2984 		http2_query_buffer_count -= expectb64len;
2985 		lock_basic_unlock(&http2_query_buffer_count_lock);
2986 		log_err("http2_req_header fail, qbuffer "
2987 			"malloc failure");
2988 		return 0;
2989 	}
2990 
2991 	if(sldns_b64_contains_nonurl((char const*)start, length)) {
2992 		char buf[65536+4];
2993 		verbose(VERB_ALGO, "HTTP2 stream contains wrong b64 encoding");
2994 		/* copy to the scratch buffer temporarily to terminate the
2995 		 * string with a zero */
2996 		if(length+1 > sizeof(buf)) {
2997 			/* too long */
2998 			lock_basic_lock(&http2_query_buffer_count_lock);
2999 			http2_query_buffer_count -= expectb64len;
3000 			lock_basic_unlock(&http2_query_buffer_count_lock);
3001 			sldns_buffer_free(h2_stream->qbuffer);
3002 			h2_stream->qbuffer = NULL;
3003 			return 1;
3004 		}
3005 		memmove(buf, start, length);
3006 		buf[length] = 0;
3007 		if(!(b64len = sldns_b64_pton(buf, sldns_buffer_current(
3008 			h2_stream->qbuffer), expectb64len)) || b64len < 0) {
3009 			lock_basic_lock(&http2_query_buffer_count_lock);
3010 			http2_query_buffer_count -= expectb64len;
3011 			lock_basic_unlock(&http2_query_buffer_count_lock);
3012 			sldns_buffer_free(h2_stream->qbuffer);
3013 			h2_stream->qbuffer = NULL;
3014 			return 1;
3015 		}
3016 	} else {
3017 		if(!(b64len = sldns_b64url_pton(
3018 			(char const *)start, length,
3019 			sldns_buffer_current(h2_stream->qbuffer),
3020 			expectb64len)) || b64len < 0) {
3021 			lock_basic_lock(&http2_query_buffer_count_lock);
3022 			http2_query_buffer_count -= expectb64len;
3023 			lock_basic_unlock(&http2_query_buffer_count_lock);
3024 			sldns_buffer_free(h2_stream->qbuffer);
3025 			h2_stream->qbuffer = NULL;
3026 			/* return without error, method can be an
3027 			 * unknown POST */
3028 			return 1;
3029 		}
3030 	}
3031 	sldns_buffer_skip(h2_stream->qbuffer, (size_t)b64len);
3032 	return 1;
3033 }
3034 
3035 /** nghttp2 callback. Used to parse headers from HEADER frames. */
3036 static int http2_req_header_cb(nghttp2_session* session,
3037 	const nghttp2_frame* frame, const uint8_t* name, size_t namelen,
3038 	const uint8_t* value, size_t valuelen, uint8_t ATTR_UNUSED(flags),
3039 	void* cb_arg)
3040 {
3041 	struct http2_stream* h2_stream = NULL;
3042 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
3043 	/* nghttp2 deals with CONTINUATION frames and provides them as part of
3044 	 * the HEADER */
3045 	if(frame->hd.type != NGHTTP2_HEADERS ||
3046 		frame->headers.cat != NGHTTP2_HCAT_REQUEST) {
3047 		/* only interested in request headers */
3048 		return 0;
3049 	}
3050 	if(!(h2_stream = nghttp2_session_get_stream_user_data(session,
3051 		frame->hd.stream_id)))
3052 		return 0;
3053 
3054 	/* earlier checks already indicate we can stop handling this query */
3055 	if(h2_stream->http_method == HTTP_METHOD_UNSUPPORTED ||
3056 		h2_stream->invalid_content_type ||
3057 		h2_stream->invalid_endpoint)
3058 		return 0;
3059 
3060 
3061 	/* nghttp2 performs some sanity checks in the headers, including:
3062 	 * name and value are guaranteed to be null terminated
3063 	 * name is guaranteed to be lowercase
3064 	 * content-length value is guaranteed to contain digits
3065 	 */
3066 
3067 	if(!h2_stream->http_method && namelen == 7 &&
3068 		memcmp(":method", name, namelen) == 0) {
3069 		/* Case insensitive check on :method value to be on the safe
3070 		 * side. I failed to find text about case sensitivity in specs.
3071 		 */
3072 		if(valuelen == 3 && strcasecmp("GET", (const char*)value) == 0)
3073 			h2_stream->http_method = HTTP_METHOD_GET;
3074 		else if(valuelen == 4 &&
3075 			strcasecmp("POST", (const char*)value) == 0) {
3076 			h2_stream->http_method = HTTP_METHOD_POST;
3077 			if(h2_stream->qbuffer) {
3078 				/* POST method uses query from DATA frames */
3079 				lock_basic_lock(&http2_query_buffer_count_lock);
3080 				http2_query_buffer_count -=
3081 					sldns_buffer_capacity(h2_stream->qbuffer);
3082 				lock_basic_unlock(&http2_query_buffer_count_lock);
3083 				sldns_buffer_free(h2_stream->qbuffer);
3084 				h2_stream->qbuffer = NULL;
3085 			}
3086 		} else
3087 			h2_stream->http_method = HTTP_METHOD_UNSUPPORTED;
3088 		return 0;
3089 	}
3090 	if(namelen == 5 && memcmp(":path", name, namelen) == 0) {
3091 		/* :path may contain DNS query, depending on method. Method might
3092 		 * not be known yet here, so check after finishing receiving
3093 		 * stream. */
3094 #define	HTTP_QUERY_PARAM "?dns="
3095 		size_t el = strlen(h2_session->c->http_endpoint);
3096 		size_t qpl = strlen(HTTP_QUERY_PARAM);
3097 
3098 		if(valuelen < el || memcmp(h2_session->c->http_endpoint,
3099 			value, el) != 0) {
3100 			h2_stream->invalid_endpoint = 1;
3101 			return 0;
3102 		}
3103 		/* larger than endpoint only allowed if it is for the query
3104 		 * parameter */
3105 		if(valuelen <= el+qpl ||
3106 			memcmp(HTTP_QUERY_PARAM, value+el, qpl) != 0) {
3107 			if(valuelen != el)
3108 				h2_stream->invalid_endpoint = 1;
3109 			return 0;
3110 		}
3111 
3112 		if(!http2_buffer_uri_query(h2_session, h2_stream,
3113 			value+(el+qpl), valuelen-(el+qpl))) {
3114 			return NGHTTP2_ERR_CALLBACK_FAILURE;
3115 		}
3116 		return 0;
3117 	}
3118 	/* Content type is a SHOULD (rfc7231#section-3.1.1.5) when using POST,
3119 	 * and not needed when using GET. Don't enforce.
3120 	 * If set only allow lowercase "application/dns-message".
3121 	 *
3122 	 * Clients SHOULD (rfc8484#section-4.1) set an accept header, but MUST
3123 	 * be able to handle "application/dns-message". Since that is the only
3124 	 * content-type supported we can ignore the accept header.
3125 	 */
3126 	if((namelen == 12 && memcmp("content-type", name, namelen) == 0)) {
3127 		if(valuelen != 23 || memcmp("application/dns-message", value,
3128 			valuelen) != 0) {
3129 			h2_stream->invalid_content_type = 1;
3130 		}
3131 	}
3132 
3133 	/* Only interested in content-lentg for POST (on not yet known) method.
3134 	 */
3135 	if((!h2_stream->http_method ||
3136 		h2_stream->http_method == HTTP_METHOD_POST) &&
3137 		!h2_stream->content_length && namelen  == 14 &&
3138 		memcmp("content-length", name, namelen) == 0) {
3139 		if(valuelen > 5) {
3140 			h2_stream->query_too_large = 1;
3141 			return 0;
3142 		}
3143 		/* guaranteed to only contain digits and be null terminated */
3144 		h2_stream->content_length = atoi((const char*)value);
3145 		if(h2_stream->content_length >
3146 			h2_session->c->http2_stream_max_qbuffer_size) {
3147 			h2_stream->query_too_large = 1;
3148 			return 0;
3149 		}
3150 	}
3151 	return 0;
3152 }
3153 
3154 /** nghttp2 callback. Used to get data from DATA frames, which can contain
3155  * queries in POST requests. */
3156 static int http2_req_data_chunk_recv_cb(nghttp2_session* ATTR_UNUSED(session),
3157 	uint8_t ATTR_UNUSED(flags), int32_t stream_id, const uint8_t* data,
3158 	size_t len, void* cb_arg)
3159 {
3160 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
3161 	struct http2_stream* h2_stream;
3162 	size_t qlen = 0;
3163 
3164 	if(!(h2_stream = nghttp2_session_get_stream_user_data(
3165 		h2_session->session, stream_id))) {
3166 		return 0;
3167 	}
3168 
3169 	if(h2_stream->query_too_large)
3170 		return 0;
3171 
3172 	if(!h2_stream->qbuffer) {
3173 		if(h2_stream->content_length) {
3174 			if(h2_stream->content_length < len)
3175 				/* getting more data in DATA frame than
3176 				 * advertised in content-length header. */
3177 				return NGHTTP2_ERR_CALLBACK_FAILURE;
3178 			qlen = h2_stream->content_length;
3179 		} else if(len <= h2_session->c->http2_stream_max_qbuffer_size) {
3180 			/* setting this to msg-buffer-size can result in a lot
3181 			 * of memory consumption. Most queries should fit in a
3182 			 * single DATA frame, and most POST queries will
3183 			 * contain content-length which does not impose this
3184 			 * limit. */
3185 			qlen = len;
3186 		}
3187 	}
3188 	if(!h2_stream->qbuffer && qlen) {
3189 		lock_basic_lock(&http2_query_buffer_count_lock);
3190 		if(http2_query_buffer_count + qlen > http2_query_buffer_max) {
3191 			lock_basic_unlock(&http2_query_buffer_count_lock);
3192 			verbose(VERB_ALGO, "reset HTTP2 stream, no space left, "
3193 				"in http2-query-buffer-size");
3194 			return http2_submit_rst_stream(h2_session, h2_stream);
3195 		}
3196 		http2_query_buffer_count += qlen;
3197 		lock_basic_unlock(&http2_query_buffer_count_lock);
3198 		if(!(h2_stream->qbuffer = sldns_buffer_new(qlen))) {
3199 			lock_basic_lock(&http2_query_buffer_count_lock);
3200 			http2_query_buffer_count -= qlen;
3201 			lock_basic_unlock(&http2_query_buffer_count_lock);
3202 		}
3203 	}
3204 
3205 	if(!h2_stream->qbuffer ||
3206 		sldns_buffer_remaining(h2_stream->qbuffer) < len) {
3207 		verbose(VERB_ALGO, "http2 data_chunk_recv failed. Not enough "
3208 			"buffer space for POST query. Can happen on multi "
3209 			"frame requests without content-length header");
3210 		h2_stream->query_too_large = 1;
3211 		return 0;
3212 	}
3213 
3214 	sldns_buffer_write(h2_stream->qbuffer, data, len);
3215 
3216 	return 0;
3217 }
3218 
3219 void http2_req_stream_clear(struct http2_stream* h2_stream)
3220 {
3221 	if(h2_stream->qbuffer) {
3222 		lock_basic_lock(&http2_query_buffer_count_lock);
3223 		http2_query_buffer_count -=
3224 			sldns_buffer_capacity(h2_stream->qbuffer);
3225 		lock_basic_unlock(&http2_query_buffer_count_lock);
3226 		sldns_buffer_free(h2_stream->qbuffer);
3227 		h2_stream->qbuffer = NULL;
3228 	}
3229 	if(h2_stream->rbuffer) {
3230 		lock_basic_lock(&http2_response_buffer_count_lock);
3231 		http2_response_buffer_count -=
3232 			sldns_buffer_capacity(h2_stream->rbuffer);
3233 		lock_basic_unlock(&http2_response_buffer_count_lock);
3234 		sldns_buffer_free(h2_stream->rbuffer);
3235 		h2_stream->rbuffer = NULL;
3236 	}
3237 }
3238 
3239 nghttp2_session_callbacks* http2_req_callbacks_create(void)
3240 {
3241 	nghttp2_session_callbacks *callbacks;
3242 	if(nghttp2_session_callbacks_new(&callbacks) == NGHTTP2_ERR_NOMEM) {
3243 		log_err("failed to initialize nghttp2 callback");
3244 		return NULL;
3245 	}
3246 	/* reception of header block started, used to create h2_stream */
3247 	nghttp2_session_callbacks_set_on_begin_headers_callback(callbacks,
3248 		http2_req_begin_headers_cb);
3249 	/* complete frame received, used to get data from stream if frame
3250 	 * has end stream flag, and start processing query */
3251 	nghttp2_session_callbacks_set_on_frame_recv_callback(callbacks,
3252 		http2_req_frame_recv_cb);
3253 	/* get request info from headers */
3254 	nghttp2_session_callbacks_set_on_header_callback(callbacks,
3255 		http2_req_header_cb);
3256 	/* get data from DATA frames, containing POST query */
3257 	nghttp2_session_callbacks_set_on_data_chunk_recv_callback(callbacks,
3258 		http2_req_data_chunk_recv_cb);
3259 
3260 	/* generic HTTP2 callbacks */
3261 	nghttp2_session_callbacks_set_recv_callback(callbacks, http2_recv_cb);
3262 	nghttp2_session_callbacks_set_send_callback(callbacks, http2_send_cb);
3263 	nghttp2_session_callbacks_set_on_stream_close_callback(callbacks,
3264 		http2_stream_close_cb);
3265 
3266 	return callbacks;
3267 }
3268 #endif /* HAVE_NGHTTP2 */
3269 
3270 #ifdef HAVE_NGTCP2
3271 struct doq_table*
3272 doq_table_create(struct config_file* cfg, struct ub_randstate* rnd)
3273 {
3274 	struct doq_table* table;
3275 
3276 	if (!cfg->quic_port)
3277 		return NULL;
3278 	table = calloc(1, sizeof(*table));
3279 	if(!table)
3280 		return NULL;
3281 #ifdef USE_NGTCP2_CRYPTO_OSSL
3282 	/* Initialize the ossl crypto, it is harmless to call twice,
3283 	 * and this is before use of doq connections. */
3284 	if(ngtcp2_crypto_ossl_init() != 0) {
3285 		log_err("ngtcp2_crypto_ossl_init failed");
3286 		free(table);
3287 		return NULL;
3288 	}
3289 #elif defined(HAVE_NGTCP2_CRYPTO_QUICTLS_INIT)
3290 	if(ngtcp2_crypto_quictls_init() != 0) {
3291 		log_err("ngtcp2_crypto_quictls_init failed");
3292 		free(table);
3293 		return NULL;
3294 	}
3295 #endif
3296 	table->idle_timeout = ((uint64_t)cfg->tcp_idle_timeout)*
3297 		NGTCP2_MILLISECONDS;
3298 	table->sv_scidlen = 16;
3299 	table->static_secret_len = 16;
3300 	table->static_secret = malloc(table->static_secret_len);
3301 	if(!table->static_secret) {
3302 		free(table);
3303 		return NULL;
3304 	}
3305 	doq_fill_rand(rnd, table->static_secret, table->static_secret_len);
3306 	table->conn_tree = rbtree_create(doq_conn_cmp);
3307 	if(!table->conn_tree) {
3308 		free(table->static_secret);
3309 		free(table);
3310 		return NULL;
3311 	}
3312 	table->conid_tree = rbtree_create(doq_conid_cmp);
3313 	if(!table->conid_tree) {
3314 		free(table->static_secret);
3315 		free(table->conn_tree);
3316 		free(table);
3317 		return NULL;
3318 	}
3319 	table->timer_tree = rbtree_create(doq_timer_cmp);
3320 	if(!table->timer_tree) {
3321 		free(table->static_secret);
3322 		free(table->conn_tree);
3323 		free(table->conid_tree);
3324 		free(table);
3325 		return NULL;
3326 	}
3327 	lock_rw_init(&table->lock);
3328 	lock_rw_init(&table->conid_lock);
3329 	lock_basic_init(&table->size_lock);
3330 	lock_protect(&table->lock, &table->static_secret,
3331 		sizeof(table->static_secret));
3332 	lock_protect(&table->lock, &table->static_secret_len,
3333 		sizeof(table->static_secret_len));
3334 	lock_protect(&table->lock, table->static_secret,
3335 		table->static_secret_len);
3336 	lock_protect(&table->lock, &table->sv_scidlen,
3337 		sizeof(table->sv_scidlen));
3338 	lock_protect(&table->lock, &table->idle_timeout,
3339 		sizeof(table->idle_timeout));
3340 	lock_protect(&table->lock, &table->conn_tree, sizeof(table->conn_tree));
3341 	lock_protect(&table->lock, table->conn_tree, sizeof(*table->conn_tree));
3342 	lock_protect(&table->conid_lock, table->conid_tree,
3343 		sizeof(*table->conid_tree));
3344 	lock_protect(&table->lock, table->timer_tree,
3345 		sizeof(*table->timer_tree));
3346 	lock_protect(&table->size_lock, &table->current_size,
3347 		sizeof(table->current_size));
3348 	return table;
3349 }
3350 
3351 /** delete elements from the connection tree */
3352 static void
3353 conn_tree_del(rbnode_type* node, void* arg)
3354 {
3355 	struct doq_table* table = (struct doq_table*)arg;
3356 	struct doq_conn* conn;
3357 	if(!node || !table)
3358 		return;
3359 	conn = (struct doq_conn*)node->key;
3360 	if(conn->timer.timer_in_list) {
3361 		/* Remove timer from list first, because finding the rbnode
3362 		 * element of the setlist of same timeouts needs tree lookup.
3363 		 * Edit the tree structure after that lookup. */
3364 		doq_timer_list_remove(conn->table, &conn->timer);
3365 	}
3366 	if(conn->timer.timer_in_tree)
3367 		doq_timer_tree_remove(conn->table, &conn->timer);
3368 	doq_table_quic_size_subtract(table, sizeof(*conn)+conn->key.dcidlen);
3369 	doq_conn_delete(conn, table);
3370 }
3371 
3372 /** delete elements from the connection id tree */
3373 static void
3374 conid_tree_del(rbnode_type* node, void* ATTR_UNUSED(arg))
3375 {
3376 	if(!node)
3377 		return;
3378 	doq_conid_delete((struct doq_conid*)node->key);
3379 }
3380 
3381 void
3382 doq_table_delete(struct doq_table* table)
3383 {
3384 	if(!table)
3385 		return;
3386 	lock_rw_destroy(&table->lock);
3387 	free(table->static_secret);
3388 	if(table->conn_tree) {
3389 		traverse_postorder(table->conn_tree, conn_tree_del, table);
3390 		free(table->conn_tree);
3391 	}
3392 	lock_rw_destroy(&table->conid_lock);
3393 	if(table->conid_tree) {
3394 		/* The tree should be empty, because the doq_conn_delete calls
3395 		 * above should have also removed their conid elements. */
3396 		traverse_postorder(table->conid_tree, conid_tree_del, NULL);
3397 		free(table->conid_tree);
3398 	}
3399 	lock_basic_destroy(&table->size_lock);
3400 	if(table->timer_tree) {
3401 		/* The tree should be empty, because the conn_tree_del calls
3402 		 * above should also have removed them. Also the doq_timer
3403 		 * is part of the doq_conn struct, so is already freed. */
3404 		free(table->timer_tree);
3405 	}
3406 	table->write_list_first = NULL;
3407 	table->write_list_last = NULL;
3408 	free(table);
3409 }
3410 
3411 struct doq_timer*
3412 doq_timer_find_time(struct doq_table* table, ngtcp2_tstamp ts)
3413 {
3414 	struct doq_timer key;
3415 	struct rbnode_type* node;
3416 	log_assert(table != NULL);
3417 	memset(&key, 0, sizeof(key));
3418 	key.time_mono = ts;
3419 	node = rbtree_search(table->timer_tree, &key);
3420 	if(node)
3421 		return (struct doq_timer*)node->key;
3422 	return NULL;
3423 }
3424 
3425 void
3426 doq_timer_tree_remove(struct doq_table* table, struct doq_timer* timer)
3427 {
3428 	if(!timer->timer_in_tree)
3429 		return;
3430 	rbtree_delete(table->timer_tree, timer);
3431 	timer->timer_in_tree = 0;
3432 	/* This item could have more timers in the same set. */
3433 	if(timer->setlist_first) {
3434 		struct doq_timer* rb_timer = timer->setlist_first;
3435 		/* del first element from setlist */
3436 		if(rb_timer->setlist_next)
3437 			rb_timer->setlist_next->setlist_prev = NULL;
3438 		else
3439 			timer->setlist_last = NULL;
3440 		timer->setlist_first = rb_timer->setlist_next;
3441 		rb_timer->setlist_prev = NULL;
3442 		rb_timer->setlist_next = NULL;
3443 		rb_timer->timer_in_list = 0;
3444 		/* insert it into the tree as new rb element */
3445 		memset(&rb_timer->node, 0, sizeof(rb_timer->node));
3446 		rb_timer->node.key = rb_timer;
3447 		rbtree_insert(table->timer_tree, &rb_timer->node);
3448 		rb_timer->timer_in_tree = 1;
3449 		/* the setlist, if any remainder, moves to the rb element */
3450 		rb_timer->setlist_first = timer->setlist_first;
3451 		rb_timer->setlist_last = timer->setlist_last;
3452 		timer->setlist_first = NULL;
3453 		timer->setlist_last = NULL;
3454 		rb_timer->worker_doq_socket = timer->worker_doq_socket;
3455 	}
3456 	timer->worker_doq_socket = NULL;
3457 }
3458 
3459 void
3460 doq_timer_list_remove(struct doq_table* table, struct doq_timer* timer)
3461 {
3462 	struct doq_timer* rb_timer;
3463 	if(!timer->timer_in_list)
3464 		return;
3465 	/* The item in the rbtree has the list start and end. */
3466 	rb_timer = doq_timer_find_time(table, timer->time_mono);
3467 	if(rb_timer) {
3468 		if(timer->setlist_prev)
3469 			timer->setlist_prev->setlist_next = timer->setlist_next;
3470 		else
3471 			rb_timer->setlist_first = timer->setlist_next;
3472 		if(timer->setlist_next)
3473 			timer->setlist_next->setlist_prev = timer->setlist_prev;
3474 		else
3475 			rb_timer->setlist_last = timer->setlist_prev;
3476 		timer->setlist_prev = NULL;
3477 		timer->setlist_next = NULL;
3478 	}
3479 	timer->timer_in_list = 0;
3480 }
3481 
3482 /** doq append timer to setlist */
3483 static void
3484 doq_timer_list_append(struct doq_timer* rb_timer, struct doq_timer* timer)
3485 {
3486 	log_assert(timer->timer_in_list == 0);
3487 	timer->timer_in_list = 1;
3488 	timer->setlist_next = NULL;
3489 	timer->setlist_prev = rb_timer->setlist_last;
3490 	if(rb_timer->setlist_last)
3491 		rb_timer->setlist_last->setlist_next = timer;
3492 	else
3493 		rb_timer->setlist_first = timer;
3494 	rb_timer->setlist_last = timer;
3495 }
3496 
3497 void
3498 doq_timer_unset(struct doq_table* table, struct doq_timer* timer)
3499 {
3500 	if(timer->timer_in_list) {
3501 		/* Remove timer from list first, because finding the rbnode
3502 		 * element of the setlist of same timeouts needs tree lookup.
3503 		 * Edit the tree structure after that lookup. */
3504 		doq_timer_list_remove(table, timer);
3505 	}
3506 	if(timer->timer_in_tree)
3507 		doq_timer_tree_remove(table, timer);
3508 	timer->worker_doq_socket = NULL;
3509 }
3510 
3511 void doq_timer_set(struct doq_table* table, struct doq_timer* timer,
3512 	struct doq_server_socket* worker_doq_socket, struct timeval* tv,
3513 	ngtcp2_tstamp ts)
3514 {
3515 	struct doq_timer* rb_timer;
3516 	if(verbosity >= VERB_ALGO && timer->conn) {
3517 		char a[256];
3518 		struct timeval rel;
3519 		addr_to_str((void*)&timer->conn->key.paddr.addr,
3520 			timer->conn->key.paddr.addrlen, a, sizeof(a));
3521 		timeval_subtract(&rel, tv, worker_doq_socket->now_tv);
3522 		verbose(VERB_ALGO, "doq %s timer set %d.%6.6d in %d.%6.6d",
3523 			a, (int)tv->tv_sec, (int)tv->tv_usec,
3524 			(int)rel.tv_sec, (int)rel.tv_usec);
3525 	}
3526 	if(timer->timer_in_tree || timer->timer_in_list) {
3527 		if(timer->time_mono == ts)
3528 			return; /* already set on that time */
3529 		doq_timer_unset(table, timer);
3530 	}
3531 	timer->time_real.tv_sec = tv->tv_sec;
3532 	timer->time_real.tv_usec = tv->tv_usec;
3533 	timer->time_mono = ts;
3534 	rb_timer = doq_timer_find_time(table, ts);
3535 	if(rb_timer) {
3536 		/* There is a timeout already with this value. Timer is
3537 		 * added to the setlist. */
3538 		doq_timer_list_append(rb_timer, timer);
3539 	} else {
3540 		/* There is no timeout with this value. Make timer a new
3541 		 * tree element. */
3542 		memset(&timer->node, 0, sizeof(timer->node));
3543 		timer->node.key = timer;
3544 		rbtree_insert(table->timer_tree, &timer->node);
3545 		timer->timer_in_tree = 1;
3546 		timer->setlist_first = NULL;
3547 		timer->setlist_last = NULL;
3548 		timer->worker_doq_socket = worker_doq_socket;
3549 	}
3550 }
3551 
3552 struct doq_conn*
3553 doq_conn_create(struct comm_point* c, struct doq_pkt_addr* paddr,
3554 	const uint8_t* dcid, size_t dcidlen, uint32_t version)
3555 {
3556 	struct doq_conn* conn = calloc(1, sizeof(*conn));
3557 	if(!conn)
3558 		return NULL;
3559 	conn->node.key = conn;
3560 	conn->doq_socket = c->doq_socket;
3561 	conn->table = c->doq_socket->table;
3562 	memmove(&conn->key.paddr.addr, &paddr->addr, paddr->addrlen);
3563 	conn->key.paddr.addrlen = paddr->addrlen;
3564 	memmove(&conn->key.paddr.localaddr, &paddr->localaddr,
3565 		paddr->localaddrlen);
3566 	conn->key.paddr.localaddrlen = paddr->localaddrlen;
3567 	conn->key.paddr.ifindex = paddr->ifindex;
3568 	conn->key.dcid = memdup((void*)dcid, dcidlen);
3569 	if(!conn->key.dcid) {
3570 		free(conn);
3571 		return NULL;
3572 	}
3573 	conn->key.dcidlen = dcidlen;
3574 	conn->version = version;
3575 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
3576 	ngtcp2_ccerr_default(&conn->ccerr);
3577 #else
3578 	ngtcp2_connection_close_error_default(&conn->last_error);
3579 #endif
3580 	rbtree_init(&conn->stream_tree, &doq_stream_cmp);
3581 	conn->timer.conn = conn;
3582 	lock_basic_init(&conn->lock);
3583 	lock_protect(&conn->lock, &conn->key, sizeof(conn->key));
3584 	lock_protect(&conn->lock, &conn->doq_socket, sizeof(conn->doq_socket));
3585 	lock_protect(&conn->lock, &conn->table, sizeof(conn->table));
3586 	lock_protect(&conn->lock, &conn->is_deleted, sizeof(conn->is_deleted));
3587 	lock_protect(&conn->lock, &conn->version, sizeof(conn->version));
3588 	lock_protect(&conn->lock, &conn->conn, sizeof(conn->conn));
3589 	lock_protect(&conn->lock, &conn->conid_list, sizeof(conn->conid_list));
3590 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
3591 	lock_protect(&conn->lock, &conn->ccerr, sizeof(conn->ccerr));
3592 #else
3593 	lock_protect(&conn->lock, &conn->last_error, sizeof(conn->last_error));
3594 #endif
3595 	lock_protect(&conn->lock, &conn->tls_alert, sizeof(conn->tls_alert));
3596 	lock_protect(&conn->lock, &conn->ssl, sizeof(conn->ssl));
3597 	lock_protect(&conn->lock, &conn->close_pkt, sizeof(conn->close_pkt));
3598 	lock_protect(&conn->lock, &conn->close_pkt_len, sizeof(conn->close_pkt_len));
3599 	lock_protect(&conn->lock, &conn->close_ecn, sizeof(conn->close_ecn));
3600 	lock_protect(&conn->lock, &conn->stream_tree, sizeof(conn->stream_tree));
3601 	lock_protect(&conn->lock, &conn->stream_write_first, sizeof(conn->stream_write_first));
3602 	lock_protect(&conn->lock, &conn->stream_write_last, sizeof(conn->stream_write_last));
3603 	lock_protect(&conn->lock, &conn->write_interest, sizeof(conn->write_interest));
3604 	lock_protect(&conn->lock, &conn->on_write_list, sizeof(conn->on_write_list));
3605 	lock_protect(&conn->lock, &conn->write_prev, sizeof(conn->write_prev));
3606 	lock_protect(&conn->lock, &conn->write_next, sizeof(conn->write_next));
3607 	return conn;
3608 }
3609 
3610 /** The arguments for doq stream tree del. */
3611 struct doq_stream_tree_del_args {
3612 	/** The doq table. */
3613 	struct doq_table* table;
3614 	/** The doq connection for the stream. */
3615 	struct doq_conn* conn;
3616 };
3617 
3618 /** delete stream tree node */
3619 static void
3620 stream_tree_del(rbnode_type* node, void* arg)
3621 {
3622 	struct doq_stream_tree_del_args* args = (struct doq_stream_tree_del_args*)arg;
3623 	struct doq_table* table = args->table;
3624 	struct doq_stream* stream;
3625 	if(!node)
3626 		return;
3627 	stream = (struct doq_stream*)node;
3628 	if(stream->mesh_state) {
3629 		mesh_state_remove_reply(stream->mesh, stream->mesh_state,
3630 			args->conn->doq_socket->cp, NULL, stream);
3631 		stream->mesh_state = NULL;
3632 	}
3633 	if(stream->in)
3634 		doq_table_quic_size_subtract(table, stream->inlen);
3635 	if(stream->out)
3636 		doq_table_quic_size_subtract(table, stream->outlen);
3637 	doq_table_quic_size_subtract(table, sizeof(*stream));
3638 	doq_stream_delete(stream);
3639 }
3640 
3641 void
3642 doq_conn_delete(struct doq_conn* conn, struct doq_table* table)
3643 {
3644 	if(!conn)
3645 		return;
3646 	lock_basic_destroy(&conn->lock);
3647 	lock_rw_wrlock(&conn->table->conid_lock);
3648 	doq_conn_clear_conids(conn);
3649 	lock_rw_unlock(&conn->table->conid_lock);
3650 	/* Remove the app data from ngtcp2 before SSL_free of conn->ssl,
3651 	 * because the ngtcp2 conn is deleted. */
3652 	if(conn->ssl)
3653 		SSL_set_app_data(conn->ssl, NULL);
3654 	if(conn->stream_tree.count != 0) {
3655 		struct doq_stream_tree_del_args args;
3656 		memset(&args, 0, sizeof(args));
3657 		args.table = table;
3658 		args.conn = conn;
3659 		traverse_postorder(&conn->stream_tree, stream_tree_del, &args);
3660 	}
3661 	free(conn->key.dcid);
3662 	SSL_free(conn->ssl);
3663 #ifdef USE_NGTCP2_CRYPTO_OSSL
3664 	ngtcp2_crypto_ossl_ctx_del(conn->ossl_ctx);
3665 #endif
3666 	ngtcp2_conn_del(conn->conn);
3667 	free(conn->close_pkt);
3668 	free(conn);
3669 }
3670 
3671 int
3672 doq_conn_cmp(const void* key1, const void* key2)
3673 {
3674 	struct doq_conn* c = (struct doq_conn*)key1;
3675 	struct doq_conn* d = (struct doq_conn*)key2;
3676 	int r;
3677 	/* Compared in the order destination address, then
3678 	 * local address, ifindex and then dcid.
3679 	 * So that for a search for findlessorequal for the destination
3680 	 * address will find connections to that address, with different
3681 	 * dcids.
3682 	 * Also a printout in sorted order prints the connections by IP
3683 	 * address of destination, and then a number of them depending on the
3684 	 * dcids. */
3685 	if(c->key.paddr.addrlen != d->key.paddr.addrlen) {
3686 		if(c->key.paddr.addrlen < d->key.paddr.addrlen)
3687 			return -1;
3688 		return 1;
3689 	}
3690 	if((r=memcmp(&c->key.paddr.addr, &d->key.paddr.addr,
3691 		c->key.paddr.addrlen))!=0)
3692 		return r;
3693 	if(c->key.paddr.localaddrlen != d->key.paddr.localaddrlen) {
3694 		if(c->key.paddr.localaddrlen < d->key.paddr.localaddrlen)
3695 			return -1;
3696 		return 1;
3697 	}
3698 	if((r=memcmp(&c->key.paddr.localaddr, &d->key.paddr.localaddr,
3699 		c->key.paddr.localaddrlen))!=0)
3700 		return r;
3701 	if(c->key.paddr.ifindex != d->key.paddr.ifindex) {
3702 		if(c->key.paddr.ifindex < d->key.paddr.ifindex)
3703 			return -1;
3704 		return 1;
3705 	}
3706 	if(c->key.dcidlen != d->key.dcidlen) {
3707 		if(c->key.dcidlen < d->key.dcidlen)
3708 			return -1;
3709 		return 1;
3710 	}
3711 	if((r=memcmp(c->key.dcid, d->key.dcid, c->key.dcidlen))!=0)
3712 		return r;
3713 	return 0;
3714 }
3715 
3716 int doq_conid_cmp(const void* key1, const void* key2)
3717 {
3718 	struct doq_conid* c = (struct doq_conid*)key1;
3719 	struct doq_conid* d = (struct doq_conid*)key2;
3720 	if(c->cidlen != d->cidlen) {
3721 		if(c->cidlen < d->cidlen)
3722 			return -1;
3723 		return 1;
3724 	}
3725 	return memcmp(c->cid, d->cid, c->cidlen);
3726 }
3727 
3728 int doq_timer_cmp(const void* key1, const void* key2)
3729 {
3730 	struct doq_timer* e = (struct doq_timer*)key1;
3731 	struct doq_timer* f = (struct doq_timer*)key2;
3732 	if(e->time_mono < f->time_mono)
3733 		return -1;
3734 	if(e->time_mono > f->time_mono)
3735 		return 1;
3736 	return 0;
3737 }
3738 
3739 int doq_stream_cmp(const void* key1, const void* key2)
3740 {
3741 	struct doq_stream* c = (struct doq_stream*)key1;
3742 	struct doq_stream* d = (struct doq_stream*)key2;
3743 	if(c->stream_id != d->stream_id) {
3744 		if(c->stream_id < d->stream_id)
3745 			return -1;
3746 		return 1;
3747 	}
3748 	return 0;
3749 }
3750 
3751 /** doq store a local address in repinfo */
3752 static void
3753 doq_repinfo_store_localaddr(struct comm_reply* repinfo,
3754 	struct doq_addr_storage* localaddr, socklen_t localaddrlen)
3755 {
3756 	/* use the pktinfo that we have for ancillary udp data otherwise,
3757 	 * this saves space for a sockaddr */
3758 	memset(&repinfo->pktinfo, 0, sizeof(repinfo->pktinfo));
3759 	if(addr_is_ip6((void*)localaddr, localaddrlen)) {
3760 #ifdef IPV6_PKTINFO
3761 		struct sockaddr_in6* sa6 = (struct sockaddr_in6*)localaddr;
3762 		memmove(&repinfo->pktinfo.v6info.ipi6_addr,
3763 			&sa6->sin6_addr, sizeof(struct in6_addr));
3764 		repinfo->doq_srcport = sa6->sin6_port;
3765 #endif
3766 		repinfo->srctype = 6;
3767 	} else {
3768 #ifdef IP_PKTINFO
3769 		struct sockaddr_in* sa = (struct sockaddr_in*)localaddr;
3770 		memmove(&repinfo->pktinfo.v4info.ipi_addr,
3771 			&sa->sin_addr, sizeof(struct in_addr));
3772 		repinfo->doq_srcport = sa->sin_port;
3773 #elif defined(IP_RECVDSTADDR)
3774 		struct sockaddr_in* sa = (struct sockaddr_in*)localaddr;
3775 		memmove(&repinfo->pktinfo.v4addr, &sa->sin_addr,
3776 			sizeof(struct in_addr));
3777 		repinfo->doq_srcport = sa->sin_port;
3778 #endif
3779 		repinfo->srctype = 4;
3780 	}
3781 }
3782 
3783 /** doq retrieve localaddr from repinfo */
3784 static void
3785 doq_repinfo_retrieve_localaddr(struct comm_reply* repinfo,
3786 	struct doq_addr_storage* localaddr, socklen_t* localaddrlen)
3787 {
3788 	if(repinfo->srctype == 6) {
3789 #ifdef IPV6_PKTINFO
3790 		struct sockaddr_in6* sa6 = (struct sockaddr_in6*)localaddr;
3791 		*localaddrlen = (socklen_t)sizeof(struct sockaddr_in6);
3792 		memset(sa6, 0, *localaddrlen);
3793 		sa6->sin6_family = AF_INET6;
3794 		memmove(&sa6->sin6_addr, &repinfo->pktinfo.v6info.ipi6_addr,
3795 			sizeof(struct in6_addr));
3796 		sa6->sin6_port = repinfo->doq_srcport;
3797 #endif
3798 	} else {
3799 #ifdef IP_PKTINFO
3800 		struct sockaddr_in* sa = (struct sockaddr_in*)localaddr;
3801 		*localaddrlen = (socklen_t)sizeof(struct sockaddr_in);
3802 		memset(sa, 0, *localaddrlen);
3803 		sa->sin_family = AF_INET;
3804 		memmove(&sa->sin_addr, &repinfo->pktinfo.v4info.ipi_addr,
3805 			sizeof(struct in_addr));
3806 		sa->sin_port = repinfo->doq_srcport;
3807 #elif defined(IP_RECVDSTADDR)
3808 		struct sockaddr_in* sa = (struct sockaddr_in*)localaddr;
3809 		*localaddrlen = (socklen_t)sizeof(struct sockaddr_in);
3810 		memset(sa, 0, *localaddrlen);
3811 		sa->sin_family = AF_INET;
3812 		memmove(&sa->sin_addr, &repinfo->pktinfo.v4addr,
3813 			sizeof(struct in_addr));
3814 		sa->sin_port = repinfo->doq_srcport;
3815 #endif
3816 	}
3817 }
3818 
3819 /** doq write a connection key into repinfo, false if it does not fit */
3820 static int
3821 doq_conn_key_store_repinfo(struct doq_conn_key* key,
3822 	struct comm_reply* repinfo)
3823 {
3824 	repinfo->is_proxied = 0;
3825 	repinfo->doq_ifindex = key->paddr.ifindex;
3826 	repinfo->remote_addrlen = key->paddr.addrlen;
3827 	memmove(&repinfo->remote_addr, &key->paddr.addr,
3828 		repinfo->remote_addrlen);
3829 	repinfo->client_addrlen = key->paddr.addrlen;
3830 	memmove(&repinfo->client_addr, &key->paddr.addr,
3831 		repinfo->client_addrlen);
3832 	doq_repinfo_store_localaddr(repinfo, &key->paddr.localaddr,
3833 		key->paddr.localaddrlen);
3834 	if(key->dcidlen > sizeof(repinfo->doq_dcid))
3835 		return 0;
3836 	repinfo->doq_dcidlen = key->dcidlen;
3837 	memmove(repinfo->doq_dcid, key->dcid, key->dcidlen);
3838 	return 1;
3839 }
3840 
3841 void
3842 doq_conn_key_from_repinfo(struct doq_conn_key* key, struct comm_reply* repinfo)
3843 {
3844 	key->paddr.ifindex = repinfo->doq_ifindex;
3845 	key->paddr.addrlen = repinfo->remote_addrlen;
3846 	memmove(&key->paddr.addr, &repinfo->remote_addr,
3847 		repinfo->remote_addrlen);
3848 	doq_repinfo_retrieve_localaddr(repinfo, &key->paddr.localaddr,
3849 		&key->paddr.localaddrlen);
3850 	key->dcidlen = repinfo->doq_dcidlen;
3851 	key->dcid = repinfo->doq_dcid;
3852 }
3853 
3854 /** doq add a stream to the connection */
3855 static void
3856 doq_conn_add_stream(struct doq_conn* conn, struct doq_stream* stream)
3857 {
3858 	(void)rbtree_insert(&conn->stream_tree, &stream->node);
3859 }
3860 
3861 /** doq delete a stream from the connection */
3862 static void
3863 doq_conn_del_stream(struct doq_conn* conn, struct doq_stream* stream)
3864 {
3865 	(void)rbtree_delete(&conn->stream_tree, &stream->node);
3866 }
3867 
3868 /** doq create new stream */
3869 static struct doq_stream*
3870 doq_stream_create(int64_t stream_id)
3871 {
3872 	struct doq_stream* stream = calloc(1, sizeof(*stream));
3873 	if(!stream)
3874 		return NULL;
3875 	stream->node.key = stream;
3876 	stream->stream_id = stream_id;
3877 	return stream;
3878 }
3879 
3880 void doq_stream_delete(struct doq_stream* stream)
3881 {
3882 	if(!stream)
3883 		return;
3884 	free(stream->in);
3885 	free(stream->out);
3886 	free(stream);
3887 }
3888 
3889 struct doq_stream*
3890 doq_stream_find(struct doq_conn* conn, int64_t stream_id)
3891 {
3892 	rbnode_type* node;
3893 	struct doq_stream key;
3894 	key.node.key = &key;
3895 	key.stream_id = stream_id;
3896 	node = rbtree_search(&conn->stream_tree, &key);
3897 	if(node)
3898 		return (struct doq_stream*)node->key;
3899 	return NULL;
3900 }
3901 
3902 /** doq put stream on the conn write list */
3903 static void
3904 doq_stream_on_write_list(struct doq_conn* conn, struct doq_stream* stream)
3905 {
3906 	if(stream->on_write_list)
3907 		return;
3908 	stream->write_prev = conn->stream_write_last;
3909 	if(conn->stream_write_last)
3910 		conn->stream_write_last->write_next = stream;
3911 	else
3912 		conn->stream_write_first = stream;
3913 	conn->stream_write_last = stream;
3914 	stream->write_next = NULL;
3915 	stream->on_write_list = 1;
3916 }
3917 
3918 /** doq remove stream from the conn write list */
3919 static void
3920 doq_stream_off_write_list(struct doq_conn* conn, struct doq_stream* stream)
3921 {
3922 	if(!stream->on_write_list)
3923 		return;
3924 	if(stream->write_next)
3925 		stream->write_next->write_prev = stream->write_prev;
3926 	else conn->stream_write_last = stream->write_prev;
3927 	if(stream->write_prev)
3928 		stream->write_prev->write_next = stream->write_next;
3929 	else conn->stream_write_first = stream->write_next;
3930 	stream->write_prev = NULL;
3931 	stream->write_next = NULL;
3932 	stream->on_write_list = 0;
3933 }
3934 
3935 /** doq stream remove in buffer */
3936 static void
3937 doq_stream_remove_in_buffer(struct doq_stream* stream, struct doq_table* table)
3938 {
3939 	if(stream->in) {
3940 		doq_table_quic_size_subtract(table, stream->inlen);
3941 		free(stream->in);
3942 		stream->in = NULL;
3943 		stream->inlen = 0;
3944 	}
3945 }
3946 
3947 /** doq stream remove out buffer */
3948 static void
3949 doq_stream_remove_out_buffer(struct doq_stream* stream,
3950 	struct doq_table* table)
3951 {
3952 	if(stream->out) {
3953 		doq_table_quic_size_subtract(table, stream->outlen);
3954 		free(stream->out);
3955 		stream->out = NULL;
3956 		stream->outlen = 0;
3957 	}
3958 }
3959 
3960 int
3961 doq_stream_close(struct doq_conn* conn, struct doq_stream* stream,
3962 	int send_shutdown)
3963 {
3964 	int ret;
3965 	if(stream->is_closed)
3966 		return 1;
3967 	stream->is_closed = 1;
3968 	if(stream->mesh_state) {
3969 		mesh_state_remove_reply(stream->mesh, stream->mesh_state,
3970 			conn->doq_socket->cp, NULL, stream);
3971 		stream->mesh_state = NULL;
3972 	}
3973 	doq_stream_off_write_list(conn, stream);
3974 	if(send_shutdown) {
3975 		verbose(VERB_ALGO, "doq: shutdown stream_id %d with app_error_code %d",
3976 			(int)stream->stream_id, (int)DOQ_APP_ERROR_CODE);
3977 		ret = ngtcp2_conn_shutdown_stream(conn->conn,
3978 #ifdef HAVE_NGTCP2_CONN_SHUTDOWN_STREAM4
3979 			0,
3980 #endif
3981 			stream->stream_id, DOQ_APP_ERROR_CODE);
3982 		if(ret != 0) {
3983 			log_err("doq ngtcp2_conn_shutdown_stream %d failed: %s",
3984 				(int)stream->stream_id, ngtcp2_strerror(ret));
3985 			return 0;
3986 		}
3987 		doq_conn_write_enable(conn);
3988 	}
3989 	verbose(VERB_ALGO, "doq: conn extend max streams bidi by 1");
3990 	ngtcp2_conn_extend_max_streams_bidi(conn->conn, 1);
3991 	doq_conn_write_enable(conn);
3992 	doq_stream_remove_in_buffer(stream, conn->doq_socket->table);
3993 	doq_stream_remove_out_buffer(stream, conn->doq_socket->table);
3994 	doq_table_quic_size_subtract(conn->doq_socket->table, sizeof(*stream));
3995 	doq_conn_del_stream(conn, stream);
3996 	doq_stream_delete(stream);
3997 	return 1;
3998 }
3999 
4000 /** doq stream pick up answer data from buffer */
4001 static int
4002 doq_stream_pickup_answer(struct doq_conn* conn, struct doq_stream* stream,
4003 	struct sldns_buffer* buf)
4004 {
4005 	stream->is_answer_available = 1;
4006 	if(stream->out) {
4007 		free(stream->out);
4008 		stream->out = NULL;
4009 		stream->outlen = 0;
4010 	}
4011 	stream->nwrite = 0;
4012 	stream->outlen = sldns_buffer_limit(buf);
4013 	if(!doq_table_quic_size_available(conn->doq_socket->table,
4014 		conn->doq_socket->cfg, stream->outlen)) {
4015 		verbose(VERB_ALGO, "doq stream: no space for reply length");
4016 		return 0;
4017 	}
4018 	/* For quic the output bytes have to stay allocated and available,
4019 	 * for potential resends, until the remote end has acknowledged them.
4020 	 * This includes the tcplen start uint16_t, in outlen_wire. */
4021 	stream->outlen_wire = htons(stream->outlen);
4022 	stream->out = memdup(sldns_buffer_begin(buf), sldns_buffer_limit(buf));
4023 	if(!stream->out) {
4024 		log_err("doq could not send answer: out of memory");
4025 		return 0;
4026 	}
4027 	return 1;
4028 }
4029 
4030 int
4031 doq_stream_send_reply(struct doq_conn* conn, struct doq_stream* stream,
4032 	struct sldns_buffer* buf)
4033 {
4034 	if(verbosity >= VERB_ALGO) {
4035 		char* s = sldns_wire2str_pkt(sldns_buffer_begin(buf),
4036 			sldns_buffer_limit(buf));
4037 		verbose(VERB_ALGO, "doq stream %d response\n%s",
4038 			(int)stream->stream_id, (s?s:"null"));
4039 		free(s);
4040 	}
4041 	if(stream->out)
4042 		doq_table_quic_size_subtract(conn->doq_socket->table,
4043 			stream->outlen);
4044 	if(!doq_stream_pickup_answer(conn, stream, buf))
4045 		return 0;
4046 	doq_table_quic_size_add(conn->doq_socket->table, stream->outlen);
4047 	doq_stream_on_write_list(conn, stream);
4048 	doq_conn_write_enable(conn);
4049 	return 1;
4050 }
4051 #endif /* HAVE_NGTCP2 */
4052 
4053 void
4054 doq_stream_add_meshstate(struct doq_stream* stream,
4055 	struct mesh_area* mesh, struct mesh_state* m)
4056 {
4057 #ifdef HAVE_NGTCP2
4058 	stream->mesh = mesh;
4059 	stream->mesh_state = m;
4060 #else
4061 	(void)stream; (void)mesh; (void)m;
4062 #endif
4063 }
4064 
4065 void
4066 doq_stream_remove_mesh_state(struct doq_stream* stream)
4067 {
4068 #ifdef HAVE_NGTCP2
4069 	if(!stream)
4070 		return;
4071 	stream->mesh_state = NULL;
4072 #else
4073 	(void)stream;
4074 #endif
4075 }
4076 
4077 #ifdef HAVE_NGTCP2
4078 /** doq stream data length has completed, allocations can be done. False on
4079  * allocation failure. */
4080 static int
4081 doq_stream_datalen_complete(struct doq_conn* conn, struct doq_stream* stream,
4082 	struct doq_table* table)
4083 {
4084 	if(stream->inlen > 1024*1024) {
4085 		log_err("doq stream in length too large %d",
4086 			(int)stream->inlen);
4087 		return 0;
4088 	}
4089 	if(!doq_table_quic_size_available(table, conn->doq_socket->cfg,
4090 		stream->inlen)) {
4091 		verbose(VERB_ALGO, "doq stream: no space for query length");
4092 		return 0;
4093 	}
4094 	stream->in = calloc(1, stream->inlen);
4095 	if(!stream->in) {
4096 		log_err("doq could not read stream, calloc failed: "
4097 			"out of memory");
4098 		return 0;
4099 	}
4100 	doq_table_quic_size_add(table, stream->inlen);
4101 	return 1;
4102 }
4103 
4104 /** doq stream data is complete, the input data has been received. */
4105 static int
4106 doq_stream_data_complete(struct doq_conn* conn, struct doq_stream* stream)
4107 {
4108 	struct comm_point* c;
4109 	if(verbosity >= VERB_ALGO) {
4110 		char* s = sldns_wire2str_pkt(stream->in, stream->inlen);
4111 		char a[128];
4112 		addr_to_str((void*)&conn->key.paddr.addr,
4113 			conn->key.paddr.addrlen, a, sizeof(a));
4114 		verbose(VERB_ALGO, "doq %s stream %d incoming query\n%s",
4115 			a, (int)stream->stream_id, (s?s:"null"));
4116 		free(s);
4117 	}
4118 	stream->is_query_complete = 1;
4119 	c = conn->doq_socket->cp;
4120 	if(!stream->in) {
4121 		verbose(VERB_ALGO, "doq_stream_data_complete: no in buffer");
4122 		return 0;
4123 	}
4124 	if(stream->inlen > sldns_buffer_capacity(c->buffer)) {
4125 		verbose(VERB_ALGO, "doq_stream_data_complete: query too long");
4126 		return 0;
4127 	}
4128 	sldns_buffer_clear(c->buffer);
4129 	sldns_buffer_write(c->buffer, stream->in, stream->inlen);
4130 	sldns_buffer_flip(c->buffer);
4131 	c->repinfo.c = c;
4132 	if(!doq_conn_key_store_repinfo(&conn->key, &c->repinfo)) {
4133 		verbose(VERB_ALGO, "doq_stream_data_complete: connection "
4134 			"DCID too long");
4135 		return 0;
4136 	}
4137 	c->repinfo.doq_streamid = stream->stream_id;
4138 	c->repinfo.doq_stream = stream;
4139 	conn->doq_socket->current_conn = conn;
4140 	fptr_ok(fptr_whitelist_comm_point(c->callback));
4141 	if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo)) {
4142 		conn->doq_socket->current_conn = NULL;
4143 		if(!doq_stream_send_reply(conn, stream, c->buffer)) {
4144 			verbose(VERB_ALGO, "doq: failed to send_reply");
4145 			return 0;
4146 		}
4147 		return 1;
4148 	}
4149 	conn->doq_socket->current_conn = NULL;
4150 	return 1;
4151 }
4152 
4153 /** doq receive data for a stream, more bytes of the incoming data */
4154 static int
4155 doq_stream_recv_data(struct doq_conn* conn, struct doq_stream* stream,
4156 	const uint8_t* data, size_t datalen, int* recv_done,
4157 	struct doq_table* table)
4158 {
4159 	int got_data = 0;
4160 	/* read the tcplength uint16_t at the start */
4161 	if(stream->nread < 2) {
4162 		uint16_t tcplen = 0;
4163 		size_t todolen = 2 - stream->nread;
4164 
4165 		if(stream->nread > 0) {
4166 			/* put in the already read byte if there is one */
4167 			tcplen = stream->inlen;
4168 		}
4169 		if(datalen < todolen)
4170 			todolen = datalen;
4171 		memmove(((uint8_t*)&tcplen)+stream->nread, data, todolen);
4172 		stream->nread += todolen;
4173 		data += todolen;
4174 		datalen -= todolen;
4175 		if(stream->nread == 2) {
4176 			/* the initial length value is completed */
4177 			stream->inlen = ntohs(tcplen);
4178 			if(!doq_stream_datalen_complete(conn, stream, table))
4179 				return 0;
4180 		} else {
4181 			/* store for later */
4182 			stream->inlen = tcplen;
4183 			return 1;
4184 		}
4185 	}
4186 	/* if there are more data bytes */
4187 	if(datalen > 0) {
4188 		size_t to_write = datalen;
4189 		if(stream->nread-2 > stream->inlen) {
4190 			verbose(VERB_ALGO, "doq stream buffer too small");
4191 			return 0;
4192 		}
4193 		if(datalen > stream->inlen - (stream->nread-2))
4194 			to_write = stream->inlen - (stream->nread-2);
4195 		if(to_write > 0) {
4196 			if(!stream->in) {
4197 				verbose(VERB_ALGO, "doq: stream has "
4198 					"no buffer");
4199 				return 0;
4200 			}
4201 			memmove(stream->in+(stream->nread-2), data, to_write);
4202 			stream->nread += to_write;
4203 			data += to_write;
4204 			datalen -= to_write;
4205 			got_data = 1;
4206 		}
4207 	}
4208 	/* Are there extra bytes received after the end? If so, log them. */
4209 	if(datalen > 0) {
4210 		if(verbosity >= VERB_ALGO)
4211 			log_hex("doq stream has extra bytes received after end",
4212 				(void*)data, datalen);
4213 	}
4214 	/* Is the input data complete? */
4215 	if(got_data && stream->nread >= stream->inlen+2) {
4216 		if(!stream->in) {
4217 			verbose(VERB_ALGO, "doq: completed stream has "
4218 				"no buffer");
4219 			return 0;
4220 		}
4221 		*recv_done = 1;
4222 	}
4223 	return 1;
4224 }
4225 
4226 /** doq receive FIN for a stream. No more bytes are going to arrive. */
4227 static int
4228 doq_stream_recv_fin(struct doq_conn* conn, struct doq_stream* stream, int
4229 	recv_done)
4230 {
4231 	if(!stream->is_query_complete && !recv_done) {
4232 		verbose(VERB_ALGO, "doq: stream recv FIN, but is "
4233 			"not complete, have %d of %d bytes",
4234 			((int)stream->nread)-2, (int)stream->inlen);
4235 		if(!doq_stream_close(conn, stream, 1))
4236 			return 0;
4237 	}
4238 	return 1;
4239 }
4240 
4241 void doq_fill_rand(struct ub_randstate* rnd, uint8_t* buf, size_t len)
4242 {
4243 	size_t i;
4244 	for(i=0; i<len; i++)
4245 		buf[i] = ub_random(rnd)&0xff;
4246 }
4247 
4248 /** generate new connection id, checks for duplicates.
4249  * caller must hold lock on conid tree. */
4250 static int
4251 doq_conn_generate_new_conid(struct doq_conn* conn, uint8_t* data,
4252 	size_t datalen)
4253 {
4254 	int max_try = 100;
4255 	int i;
4256 	for(i=0; i<max_try; i++) {
4257 		doq_fill_rand(conn->doq_socket->rnd, data, datalen);
4258 		if(!doq_conid_find(conn->table, data, datalen)) {
4259 			/* Found an unused connection id. */
4260 			return 1;
4261 		}
4262 	}
4263 	verbose(VERB_ALGO, "doq_conn_generate_new_conid failed: could not "
4264 		"generate random unused connection id value in %d attempts.",
4265 		max_try);
4266 	return 0;
4267 }
4268 
4269 /** ngtcp2 rand callback function */
4270 static void
4271 doq_rand_cb(uint8_t* dest, size_t destlen, const ngtcp2_rand_ctx* rand_ctx)
4272 {
4273 	struct ub_randstate* rnd = (struct ub_randstate*)
4274 		rand_ctx->native_handle;
4275 	doq_fill_rand(rnd, dest, destlen);
4276 }
4277 
4278 /** ngtcp2 get_new_connection_id callback function */
4279 static int
4280 doq_get_new_connection_id_cb(ngtcp2_conn* ATTR_UNUSED(conn), ngtcp2_cid* cid,
4281 	uint8_t* token, size_t cidlen, void* user_data)
4282 {
4283 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4284 	/* Lock the conid tree, so we can check for duplicates while
4285 	 * generating the id, and then insert it, whilst keeping the tree
4286 	 * locked against other modifications, guaranteeing uniqueness. */
4287 	lock_rw_wrlock(&doq_conn->table->conid_lock);
4288 	if(!doq_conn_generate_new_conid(doq_conn, cid->data, cidlen)) {
4289 		lock_rw_unlock(&doq_conn->table->conid_lock);
4290 		return NGTCP2_ERR_CALLBACK_FAILURE;
4291 	}
4292 	cid->datalen = cidlen;
4293 	if(ngtcp2_crypto_generate_stateless_reset_token(token,
4294 		doq_conn->doq_socket->static_secret,
4295 		doq_conn->doq_socket->static_secret_len, cid) != 0) {
4296 		lock_rw_unlock(&doq_conn->table->conid_lock);
4297 		return NGTCP2_ERR_CALLBACK_FAILURE;
4298 	}
4299 	if(!doq_conn_associate_conid(doq_conn, cid->data, cid->datalen)) {
4300 		lock_rw_unlock(&doq_conn->table->conid_lock);
4301 		return NGTCP2_ERR_CALLBACK_FAILURE;
4302 	}
4303 	lock_rw_unlock(&doq_conn->table->conid_lock);
4304 	return 0;
4305 }
4306 
4307 /** ngtcp2 remove_connection_id callback function */
4308 static int
4309 doq_remove_connection_id_cb(ngtcp2_conn* ATTR_UNUSED(conn),
4310 	const ngtcp2_cid* cid, void* user_data)
4311 {
4312 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4313 	lock_rw_wrlock(&doq_conn->table->conid_lock);
4314 	doq_conn_dissociate_conid(doq_conn, cid->data, cid->datalen);
4315 	lock_rw_unlock(&doq_conn->table->conid_lock);
4316 	return 0;
4317 }
4318 
4319 /** doq submit a new token */
4320 static int
4321 doq_submit_new_token(struct doq_conn* conn)
4322 {
4323 	uint8_t token[NGTCP2_CRYPTO_MAX_REGULAR_TOKENLEN];
4324 	ngtcp2_ssize tokenlen;
4325 	int ret;
4326 	const ngtcp2_path* path = ngtcp2_conn_get_path(conn->conn);
4327 
4328 	tokenlen = ngtcp2_crypto_generate_regular_token(token,
4329 		conn->doq_socket->static_secret,
4330 		conn->doq_socket->static_secret_len, path->remote.addr,
4331 		path->remote.addrlen, doq_get_timestamp_nanosec());
4332 	if(tokenlen < 0) {
4333 		log_err("doq ngtcp2_crypto_generate_regular_token failed");
4334 		return 1;
4335 	}
4336 
4337 	verbose(VERB_ALGO, "doq submit new token");
4338 	ret = ngtcp2_conn_submit_new_token(conn->conn, token, tokenlen);
4339 	if(ret != 0) {
4340 		log_err("doq ngtcp2_conn_submit_new_token failed: %s",
4341 			ngtcp2_strerror(ret));
4342 		return 0;
4343 	}
4344 	return 1;
4345 }
4346 
4347 /** ngtcp2 handshake_completed callback function */
4348 static int
4349 doq_handshake_completed_cb(ngtcp2_conn* ATTR_UNUSED(conn), void* user_data)
4350 {
4351 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4352 	verbose(VERB_ALGO, "doq handshake_completed callback");
4353 	verbose(VERB_ALGO, "ngtcp2_conn_get_max_data_left is %d",
4354 		(int)ngtcp2_conn_get_max_data_left(doq_conn->conn));
4355 #ifdef HAVE_NGTCP2_CONN_GET_MAX_LOCAL_STREAMS_UNI
4356 	verbose(VERB_ALGO, "ngtcp2_conn_get_max_local_streams_uni is %d",
4357 		(int)ngtcp2_conn_get_max_local_streams_uni(doq_conn->conn));
4358 #endif
4359 	verbose(VERB_ALGO, "ngtcp2_conn_get_streams_uni_left is %d",
4360 		(int)ngtcp2_conn_get_streams_uni_left(doq_conn->conn));
4361 	verbose(VERB_ALGO, "ngtcp2_conn_get_streams_bidi_left is %d",
4362 		(int)ngtcp2_conn_get_streams_bidi_left(doq_conn->conn));
4363 	verbose(VERB_ALGO, "negotiated cipher name is %s",
4364 		SSL_get_cipher_name(doq_conn->ssl));
4365 	if(verbosity > VERB_ALGO) {
4366 		const unsigned char* alpn = NULL;
4367 		unsigned int alpnlen = 0;
4368 		char alpnstr[128];
4369 		SSL_get0_alpn_selected(doq_conn->ssl, &alpn, &alpnlen);
4370 		if(alpnlen > sizeof(alpnstr)-1)
4371 			alpnlen = sizeof(alpnstr)-1;
4372 		memmove(alpnstr, alpn, alpnlen);
4373 		alpnstr[alpnlen]=0;
4374 		verbose(VERB_ALGO, "negotiated ALPN is '%s'", alpnstr);
4375 	}
4376 
4377 	if(!doq_submit_new_token(doq_conn))
4378 		return -1;
4379 	return 0;
4380 }
4381 
4382 /** ngtcp2 stream_open callback function */
4383 static int
4384 doq_stream_open_cb(ngtcp2_conn* ATTR_UNUSED(conn), int64_t stream_id,
4385 	void* user_data)
4386 {
4387 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4388 	struct doq_stream* stream;
4389 	verbose(VERB_ALGO, "doq new stream %x", (int)stream_id);
4390 	if(doq_stream_find(doq_conn, stream_id)) {
4391 		verbose(VERB_ALGO, "doq: stream with this id already exists");
4392 		return 0;
4393 	}
4394 	if(!doq_table_quic_size_available(doq_conn->doq_socket->table,
4395 		doq_conn->doq_socket->cfg, sizeof(*stream)
4396 		+ 100 /* estimated query in */
4397 		+ 512 /* estimated response out */
4398 		)) {
4399 		int rv;
4400 		verbose(VERB_ALGO, "doq: no mem for new stream");
4401 		rv = ngtcp2_conn_shutdown_stream(doq_conn->conn,
4402 #ifdef HAVE_NGTCP2_CONN_SHUTDOWN_STREAM4
4403 			0,
4404 #endif
4405 			stream_id, NGTCP2_CONNECTION_REFUSED);
4406 		if(rv != 0) {
4407 			log_err("ngtcp2_conn_shutdown_stream failed: %s",
4408 				ngtcp2_strerror(rv));
4409 			return NGTCP2_ERR_CALLBACK_FAILURE;
4410 		}
4411 		return 0;
4412 	}
4413 	stream = doq_stream_create(stream_id);
4414 	if(!stream) {
4415 		log_err("doq: could not doq_stream_create: out of memory");
4416 		return NGTCP2_ERR_CALLBACK_FAILURE;
4417 	}
4418 	doq_table_quic_size_add(doq_conn->doq_socket->table, sizeof(*stream));
4419 	doq_conn_add_stream(doq_conn, stream);
4420 	return 0;
4421 }
4422 
4423 /** ngtcp2 recv_stream_data callback function */
4424 static int
4425 doq_recv_stream_data_cb(ngtcp2_conn* ATTR_UNUSED(conn), uint32_t flags,
4426 	int64_t stream_id, uint64_t offset, const uint8_t* data,
4427 	size_t datalen, void* user_data, void* ATTR_UNUSED(stream_user_data))
4428 {
4429 	int recv_done = 0;
4430 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4431 	struct doq_stream* stream;
4432 	verbose(VERB_ALGO, "doq recv stream data stream id %d offset %d "
4433 		"datalen %d%s%s", (int)stream_id, (int)offset, (int)datalen,
4434 		((flags&NGTCP2_STREAM_DATA_FLAG_FIN)!=0?" FIN":""),
4435 #ifdef NGTCP2_STREAM_DATA_FLAG_0RTT
4436 		((flags&NGTCP2_STREAM_DATA_FLAG_0RTT)!=0?" 0RTT":"")
4437 #else
4438 		((flags&NGTCP2_STREAM_DATA_FLAG_EARLY)!=0?" EARLY":"")
4439 #endif
4440 		);
4441 	stream = doq_stream_find(doq_conn, stream_id);
4442 	if(!stream) {
4443 		verbose(VERB_ALGO, "doq: received stream data for "
4444 			"unknown stream %d", (int)stream_id);
4445 		return 0;
4446 	}
4447 	if(stream->is_closed) {
4448 		verbose(VERB_ALGO, "doq: stream is closed, ignore recv data");
4449 		return 0;
4450 	}
4451 	if(datalen != 0) {
4452 		if(!doq_stream_recv_data(doq_conn, stream, data, datalen,
4453 			&recv_done, doq_conn->doq_socket->table))
4454 			return NGTCP2_ERR_CALLBACK_FAILURE;
4455 	}
4456 	if((flags&NGTCP2_STREAM_DATA_FLAG_FIN)!=0) {
4457 		if(!doq_stream_recv_fin(doq_conn, stream, recv_done))
4458 			return NGTCP2_ERR_CALLBACK_FAILURE;
4459 	}
4460 	ngtcp2_conn_extend_max_stream_offset(doq_conn->conn, stream_id,
4461 		datalen);
4462 	ngtcp2_conn_extend_max_offset(doq_conn->conn, datalen);
4463 	if(recv_done) {
4464 		if(!doq_stream_data_complete(doq_conn, stream))
4465 			return NGTCP2_ERR_CALLBACK_FAILURE;
4466 	}
4467 	return 0;
4468 }
4469 
4470 /** ngtcp2 stream_close callback function */
4471 static int
4472 doq_stream_close_cb(ngtcp2_conn* ATTR_UNUSED(conn), uint32_t flags,
4473 	int64_t stream_id, uint64_t app_error_code, void* user_data,
4474 	void* ATTR_UNUSED(stream_user_data))
4475 {
4476 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4477 	struct doq_stream* stream;
4478 	if((flags&NGTCP2_STREAM_CLOSE_FLAG_APP_ERROR_CODE_SET)!=0)
4479 		verbose(VERB_ALGO, "doq stream close for stream id %d %sapp_error_code %d",
4480 		(int)stream_id,
4481 		(((flags&NGTCP2_STREAM_CLOSE_FLAG_APP_ERROR_CODE_SET)!=0)?
4482 		"APP_ERROR_CODE_SET ":""),
4483 		(int)app_error_code);
4484 	else
4485 		verbose(VERB_ALGO, "doq stream close for stream id %d",
4486 			(int)stream_id);
4487 
4488 	stream = doq_stream_find(doq_conn, stream_id);
4489 	if(!stream) {
4490 		verbose(VERB_ALGO, "doq: stream close for "
4491 			"unknown stream %d", (int)stream_id);
4492 		return 0;
4493 	}
4494 	if(!doq_stream_close(doq_conn, stream, 0))
4495 		return NGTCP2_ERR_CALLBACK_FAILURE;
4496 	return 0;
4497 }
4498 
4499 /** ngtcp2 stream_reset callback function */
4500 static int
4501 doq_stream_reset_cb(ngtcp2_conn* ATTR_UNUSED(conn), int64_t stream_id,
4502 	uint64_t final_size, uint64_t app_error_code, void* user_data,
4503 	void* ATTR_UNUSED(stream_user_data))
4504 {
4505 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4506 	struct doq_stream* stream;
4507 	verbose(VERB_ALGO, "doq stream reset for stream id %d final_size %d "
4508 		"app_error_code %d", (int)stream_id, (int)final_size,
4509 		(int)app_error_code);
4510 
4511 	stream = doq_stream_find(doq_conn, stream_id);
4512 	if(!stream) {
4513 		verbose(VERB_ALGO, "doq: stream reset for "
4514 			"unknown stream %d", (int)stream_id);
4515 		return 0;
4516 	}
4517 	if(!doq_stream_close(doq_conn, stream, 1))
4518 		return NGTCP2_ERR_CALLBACK_FAILURE;
4519 	return 0;
4520 }
4521 
4522 /** ngtcp2 extend_max_stream_data function */
4523 int doq_extend_max_stream_data_cb(ngtcp2_conn* ATTR_UNUSED(conn),
4524 	int64_t stream_id, uint64_t max_data, void* user_data,
4525 	void* ATTR_UNUSED(stream_user_data))
4526 {
4527 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4528 	struct doq_stream* stream;
4529 	verbose(VERB_ALGO, "doq extend_max_stream_data stream id %d "
4530 		"max_data %d ", (int)stream_id, (int)max_data);
4531 	if(max_data == 0)
4532 		return 0;
4533 	stream = doq_stream_find(doq_conn, stream_id);
4534 	if(!stream) {
4535 		verbose(VERB_ALGO, "doq: unknown stream %d", (int)stream_id);
4536 		return 0;
4537 	}
4538 	if(!stream->is_answer_available)
4539 		return 0;
4540 	doq_stream_on_write_list(doq_conn, stream);
4541 	doq_conn_write_enable(doq_conn);
4542 	return 0;
4543 }
4544 
4545 /** ngtcp2 acked_stream_data_offset callback function */
4546 static int
4547 doq_acked_stream_data_offset_cb(ngtcp2_conn* ATTR_UNUSED(conn),
4548 	int64_t stream_id, uint64_t offset, uint64_t datalen, void* user_data,
4549 	void* ATTR_UNUSED(stream_user_data))
4550 {
4551 	struct doq_conn* doq_conn = (struct doq_conn*)user_data;
4552 	struct doq_stream* stream;
4553 	verbose(VERB_ALGO, "doq stream acked data for stream id %d offset %d "
4554 		"datalen %d", (int)stream_id, (int)offset, (int)datalen);
4555 
4556 	stream = doq_stream_find(doq_conn, stream_id);
4557 	if(!stream) {
4558 		verbose(VERB_ALGO, "doq: stream acked data for "
4559 			"unknown stream %d", (int)stream_id);
4560 		return 0;
4561 	}
4562 	/* Acked the data from [offset .. offset+datalen). */
4563 	if(stream->is_closed)
4564 		return 0;
4565 	if(offset+datalen >= stream->outlen) {
4566 		doq_stream_remove_in_buffer(stream,
4567 			doq_conn->doq_socket->table);
4568 		doq_stream_remove_out_buffer(stream,
4569 			doq_conn->doq_socket->table);
4570 	}
4571 	return 0;
4572 }
4573 
4574 /** ngtc2p log_printf callback function */
4575 static void
4576 doq_log_printf_cb(void* ATTR_UNUSED(user_data), const char* fmt, ...)
4577 {
4578 	char buf[1024];
4579 	va_list ap;
4580 	va_start(ap, fmt);
4581 	vsnprintf(buf, sizeof(buf), fmt, ap);
4582 	verbose(VERB_ALGO, "libngtcp2: %s", buf);
4583 	va_end(ap);
4584 }
4585 
4586 #ifdef MAKE_QUIC_METHOD
4587 /** the doq application tx key callback, false on failure */
4588 static int
4589 doq_application_tx_key_cb(struct doq_conn* conn)
4590 {
4591 	verbose(VERB_ALGO, "doq application tx key cb");
4592 	/* The server does not want to open streams to the client,
4593 	 * the client instead initiates by opening bidi streams. */
4594 	verbose(VERB_ALGO, "doq ngtcp2_conn_get_max_data_left is %d",
4595 		(int)ngtcp2_conn_get_max_data_left(conn->conn));
4596 #ifdef HAVE_NGTCP2_CONN_GET_MAX_LOCAL_STREAMS_UNI
4597 	verbose(VERB_ALGO, "doq ngtcp2_conn_get_max_local_streams_uni is %d",
4598 		(int)ngtcp2_conn_get_max_local_streams_uni(conn->conn));
4599 #endif
4600 	verbose(VERB_ALGO, "doq ngtcp2_conn_get_streams_uni_left is %d",
4601 		(int)ngtcp2_conn_get_streams_uni_left(conn->conn));
4602 	verbose(VERB_ALGO, "doq ngtcp2_conn_get_streams_bidi_left is %d",
4603 		(int)ngtcp2_conn_get_streams_bidi_left(conn->conn));
4604 	return 1;
4605 }
4606 
4607 /** quic_method set_encryption_secrets function */
4608 static int
4609 doq_set_encryption_secrets(SSL *ssl, OSSL_ENCRYPTION_LEVEL ossl_level,
4610 	const uint8_t *read_secret, const uint8_t *write_secret,
4611 	size_t secret_len)
4612 {
4613 	struct doq_conn* doq_conn = (struct doq_conn*)SSL_get_app_data(ssl);
4614 #ifdef HAVE_NGTCP2_ENCRYPTION_LEVEL
4615 	ngtcp2_encryption_level
4616 #else
4617 	ngtcp2_crypto_level
4618 #endif
4619 		level =
4620 #ifdef USE_NGTCP2_CRYPTO_OSSL
4621 		ngtcp2_crypto_ossl_from_ossl_encryption_level(ossl_level);
4622 #elif defined(HAVE_NGTCP2_CRYPTO_QUICTLS_FROM_OSSL_ENCRYPTION_LEVEL)
4623 		ngtcp2_crypto_quictls_from_ossl_encryption_level(ossl_level);
4624 #else
4625 		ngtcp2_crypto_openssl_from_ossl_encryption_level(ossl_level);
4626 #endif
4627 
4628 	if(read_secret) {
4629 		verbose(VERB_ALGO, "doq: ngtcp2_crypto_derive_and_install_rx_key for level %d ossl %d", (int)level, (int)ossl_level);
4630 		if(ngtcp2_crypto_derive_and_install_rx_key(doq_conn->conn,
4631 			NULL, NULL, NULL, level, read_secret, secret_len)
4632 			!= 0) {
4633 			log_err("ngtcp2_crypto_derive_and_install_rx_key "
4634 				"failed");
4635 			return 0;
4636 		}
4637 	}
4638 
4639 	if(write_secret) {
4640 		verbose(VERB_ALGO, "doq: ngtcp2_crypto_derive_and_install_tx_key for level %d ossl %d", (int)level, (int)ossl_level);
4641 		if(ngtcp2_crypto_derive_and_install_tx_key(doq_conn->conn,
4642 			NULL, NULL, NULL, level, write_secret, secret_len)
4643 			!= 0) {
4644 			log_err("ngtcp2_crypto_derive_and_install_tx_key "
4645 				"failed");
4646 			return 0;
4647 		}
4648 		if(level == NGTCP2_CRYPTO_LEVEL_APPLICATION) {
4649 			if(!doq_application_tx_key_cb(doq_conn))
4650 				return 0;
4651 		}
4652 	}
4653 	return 1;
4654 }
4655 
4656 /** quic_method add_handshake_data function */
4657 static int
4658 doq_add_handshake_data(SSL *ssl, OSSL_ENCRYPTION_LEVEL ossl_level,
4659 	const uint8_t *data, size_t len)
4660 {
4661 	struct doq_conn* doq_conn = (struct doq_conn*)SSL_get_app_data(ssl);
4662 #ifdef HAVE_NGTCP2_ENCRYPTION_LEVEL
4663 	ngtcp2_encryption_level
4664 #else
4665 	ngtcp2_crypto_level
4666 #endif
4667 		level =
4668 #ifdef USE_NGTCP2_CRYPTO_OSSL
4669 		ngtcp2_crypto_ossl_from_ossl_encryption_level(ossl_level);
4670 #elif defined(HAVE_NGTCP2_CRYPTO_QUICTLS_FROM_OSSL_ENCRYPTION_LEVEL)
4671 		ngtcp2_crypto_quictls_from_ossl_encryption_level(ossl_level);
4672 #else
4673 		ngtcp2_crypto_openssl_from_ossl_encryption_level(ossl_level);
4674 #endif
4675 	int rv;
4676 
4677 	verbose(VERB_ALGO, "doq_add_handshake_data: "
4678 		"ngtcp2_con_submit_crypto_data level %d", (int)level);
4679 	rv = ngtcp2_conn_submit_crypto_data(doq_conn->conn, level, data, len);
4680 	if(rv != 0) {
4681 		log_err("ngtcp2_conn_submit_crypto_data failed: %s",
4682 			ngtcp2_strerror(rv));
4683 		ngtcp2_conn_set_tls_error(doq_conn->conn, rv);
4684 		return 0;
4685 	}
4686 	return 1;
4687 }
4688 
4689 /** quic_method flush_flight function */
4690 static int
4691 doq_flush_flight(SSL* ATTR_UNUSED(ssl))
4692 {
4693 	return 1;
4694 }
4695 
4696 /** quic_method send_alert function */
4697 static int
4698 doq_send_alert(SSL *ssl, enum ssl_encryption_level_t ATTR_UNUSED(level),
4699 	uint8_t alert)
4700 {
4701 	struct doq_conn* doq_conn = (struct doq_conn*)SSL_get_app_data(ssl);
4702 	doq_conn->tls_alert = alert;
4703 	return 1;
4704 }
4705 #endif /* MAKE_QUIC_METHOD */
4706 
4707 /** ALPN select callback for the doq SSL context */
4708 static int
4709 doq_alpn_select_cb(SSL* ATTR_UNUSED(ssl), const unsigned char** out,
4710 	unsigned char* outlen, const unsigned char* in, unsigned int inlen,
4711 	void* ATTR_UNUSED(arg))
4712 {
4713 	/* select "doq" */
4714 	int ret = SSL_select_next_proto((void*)out, outlen,
4715 		(const unsigned char*)"\x03""doq", 4, in, inlen);
4716 	if(ret == OPENSSL_NPN_NEGOTIATED)
4717 		return SSL_TLSEXT_ERR_OK;
4718 	verbose(VERB_ALGO, "doq alpn_select_cb: ALPN from client does "
4719 		"not have 'doq'");
4720 	return SSL_TLSEXT_ERR_ALERT_FATAL;
4721 }
4722 
4723 void* quic_sslctx_create(char* key, char* pem, char* verifypem)
4724 {
4725 #ifdef HAVE_NGTCP2
4726 	char* sid_ctx = "unbound server";
4727 #ifdef MAKE_QUIC_METHOD
4728 	SSL_QUIC_METHOD* quic_method;
4729 #endif
4730 	SSL_CTX* ctx = SSL_CTX_new(TLS_server_method());
4731 	if(!ctx) {
4732 		log_crypto_err("Could not SSL_CTX_new");
4733 		return NULL;
4734 	}
4735 	if(!key || key[0] == 0) {
4736 		log_err("doq: error, no tls-service-key file specified");
4737 		SSL_CTX_free(ctx);
4738 		return NULL;
4739 	}
4740 	if(!pem || pem[0] == 0) {
4741 		log_err("doq: error, no tls-service-pem file specified");
4742 		SSL_CTX_free(ctx);
4743 		return NULL;
4744 	}
4745 	SSL_CTX_set_options(ctx,
4746 		(SSL_OP_ALL & ~SSL_OP_DONT_INSERT_EMPTY_FRAGMENTS) |
4747 		SSL_OP_SINGLE_ECDH_USE |
4748 		SSL_OP_CIPHER_SERVER_PREFERENCE |
4749 		SSL_OP_NO_ANTI_REPLAY);
4750 	SSL_CTX_set_mode(ctx, SSL_MODE_RELEASE_BUFFERS);
4751 	SSL_CTX_set_min_proto_version(ctx, TLS1_3_VERSION);
4752 	SSL_CTX_set_max_proto_version(ctx, TLS1_3_VERSION);
4753 #ifdef HAVE_SSL_CTX_SET_ALPN_SELECT_CB
4754 	SSL_CTX_set_alpn_select_cb(ctx, doq_alpn_select_cb, NULL);
4755 #endif
4756 	SSL_CTX_set_default_verify_paths(ctx);
4757 	if(!SSL_CTX_use_certificate_chain_file(ctx, pem)) {
4758 		log_err("doq: error for cert file: %s", pem);
4759 		log_crypto_err("doq: error in "
4760 			"SSL_CTX_use_certificate_chain_file");
4761 		SSL_CTX_free(ctx);
4762 		return NULL;
4763 	}
4764 	if(!SSL_CTX_use_PrivateKey_file(ctx, key, SSL_FILETYPE_PEM)) {
4765 		log_err("doq: error for private key file: %s", key);
4766 		log_crypto_err("doq: error in SSL_CTX_use_PrivateKey_file");
4767 		SSL_CTX_free(ctx);
4768 		return NULL;
4769 	}
4770 	if(!SSL_CTX_check_private_key(ctx)) {
4771 		log_err("doq: error for key file: %s", key);
4772 		log_crypto_err("doq: error in SSL_CTX_check_private_key");
4773 		SSL_CTX_free(ctx);
4774 		return NULL;
4775 	}
4776 	SSL_CTX_set_session_id_context(ctx, (void*)sid_ctx, strlen(sid_ctx));
4777 	if(verifypem && verifypem[0]) {
4778 		if(!SSL_CTX_load_verify_locations(ctx, verifypem, NULL)) {
4779 			log_err("doq: error for verify pem file: %s",
4780 				verifypem);
4781 			log_crypto_err("doq: error in "
4782 				"SSL_CTX_load_verify_locations");
4783 			SSL_CTX_free(ctx);
4784 			return NULL;
4785 		}
4786 		SSL_CTX_set_client_CA_list(ctx, SSL_load_client_CA_file(
4787 			verifypem));
4788 		SSL_CTX_set_verify(ctx, SSL_VERIFY_PEER|
4789 			SSL_VERIFY_CLIENT_ONCE|
4790 			SSL_VERIFY_FAIL_IF_NO_PEER_CERT, NULL);
4791 	}
4792 
4793 	SSL_CTX_set_max_early_data(ctx, 0xffffffff);
4794 #ifdef HAVE_NGTCP2_CRYPTO_QUICTLS_CONFIGURE_SERVER_CONTEXT
4795 	if(ngtcp2_crypto_quictls_configure_server_context(ctx) != 0) {
4796 		log_err("ngtcp2_crypto_quictls_configure_server_context failed");
4797 		SSL_CTX_free(ctx);
4798 		return NULL;
4799 	}
4800 #elif defined(MAKE_QUIC_METHOD)
4801 	/* The quic_method needs to remain valid during the SSL_CTX
4802 	 * lifetime, so we allocate it. It is freed with the
4803 	 * doq_server_socket. */
4804 	quic_method = calloc(1, sizeof(SSL_QUIC_METHOD));
4805 	if(!quic_method) {
4806 		log_err("calloc failed: out of memory");
4807 		SSL_CTX_free(ctx);
4808 		return NULL;
4809 	}
4810 	doq_socket->quic_method = quic_method;
4811 	quic_method->set_encryption_secrets = doq_set_encryption_secrets;
4812 	quic_method->add_handshake_data = doq_add_handshake_data;
4813 	quic_method->flush_flight = doq_flush_flight;
4814 	quic_method->send_alert = doq_send_alert;
4815 	SSL_CTX_set_quic_method(ctx, doq_socket->quic_method);
4816 #endif
4817 	return ctx;
4818 #else /* HAVE_NGTCP2 */
4819 	(void)key; (void)pem; (void)verifypem;
4820 	return NULL;
4821 #endif /* HAVE_NGTCP2 */
4822 }
4823 
4824 /** Get the ngtcp2_conn from ssl userdata of type ngtcp2_conn_ref */
4825 static ngtcp2_conn* doq_conn_ref_get_conn(ngtcp2_crypto_conn_ref* conn_ref)
4826 {
4827 	struct doq_conn* conn = (struct doq_conn*)conn_ref->user_data;
4828 	return conn->conn;
4829 }
4830 
4831 /** create new SSL session for server connection */
4832 static SSL*
4833 doq_ssl_server_setup(SSL_CTX* ctx, struct doq_conn* conn)
4834 {
4835 #ifdef USE_NGTCP2_CRYPTO_OSSL
4836 	int ret;
4837 #endif
4838 	SSL* ssl = SSL_new(ctx);
4839 	if(!ssl) {
4840 		log_crypto_err("doq: SSL_new failed");
4841 		return NULL;
4842 	}
4843 #ifdef USE_NGTCP2_CRYPTO_OSSL
4844 	if((ret=ngtcp2_crypto_ossl_ctx_new(&conn->ossl_ctx, NULL)) != 0) {
4845 		log_err("doq: ngtcp2_crypto_ossl_ctx_new failed: %s",
4846 			ngtcp2_strerror(ret));
4847 		SSL_free(ssl);
4848 		return NULL;
4849 	}
4850 	ngtcp2_crypto_ossl_ctx_set_ssl(conn->ossl_ctx, ssl);
4851 	if(ngtcp2_crypto_ossl_configure_server_session(ssl) != 0) {
4852 		log_err("doq: ngtcp2_crypto_ossl_configure_server_session failed");
4853 		SSL_free(ssl);
4854 		return NULL;
4855 	}
4856 #endif
4857 #if defined(USE_NGTCP2_CRYPTO_OSSL) || defined(HAVE_NGTCP2_CRYPTO_QUICTLS_CONFIGURE_SERVER_CONTEXT)
4858 	conn->conn_ref.get_conn = &doq_conn_ref_get_conn;
4859 	conn->conn_ref.user_data = conn;
4860 	SSL_set_app_data(ssl, &conn->conn_ref);
4861 #else
4862 	SSL_set_app_data(ssl, conn);
4863 #endif
4864 	SSL_set_accept_state(ssl);
4865 #ifdef HAVE_SSL_SET_QUIC_TLS_EARLY_DATA_ENABLED
4866 	SSL_set_quic_tls_early_data_enabled(ssl, 1);
4867 #else
4868 	SSL_set_quic_early_data_enabled(ssl, 1);
4869 #endif
4870 	return ssl;
4871 }
4872 
4873 int
4874 doq_conn_setup(struct doq_conn* conn, uint8_t* scid, size_t scidlen,
4875 	uint8_t* ocid, size_t ocidlen, const uint8_t* token, size_t tokenlen)
4876 {
4877 	int rv;
4878 	struct ngtcp2_cid dcid, sv_scid, scid_cid;
4879 	struct ngtcp2_path path;
4880 	struct ngtcp2_callbacks callbacks;
4881 	struct ngtcp2_settings settings;
4882 	struct ngtcp2_transport_params params;
4883 	memset(&dcid, 0, sizeof(dcid));
4884 	memset(&sv_scid, 0, sizeof(sv_scid));
4885 	memset(&scid_cid, 0, sizeof(scid_cid));
4886 	memset(&path, 0, sizeof(path));
4887 	memset(&callbacks, 0, sizeof(callbacks));
4888 	memset(&settings, 0, sizeof(settings));
4889 	memset(&params, 0, sizeof(params));
4890 
4891 	ngtcp2_cid_init(&scid_cid, scid, scidlen);
4892 	ngtcp2_cid_init(&dcid, conn->key.dcid, conn->key.dcidlen);
4893 
4894 	path.remote.addr = (struct sockaddr*)&conn->key.paddr.addr;
4895 	path.remote.addrlen = conn->key.paddr.addrlen;
4896 	path.local.addr = (struct sockaddr*)&conn->key.paddr.localaddr;
4897 	path.local.addrlen = conn->key.paddr.localaddrlen;
4898 
4899 	callbacks.recv_client_initial = ngtcp2_crypto_recv_client_initial_cb;
4900 	callbacks.recv_crypto_data = ngtcp2_crypto_recv_crypto_data_cb;
4901 	callbacks.encrypt = ngtcp2_crypto_encrypt_cb;
4902 	callbacks.decrypt = ngtcp2_crypto_decrypt_cb;
4903 	callbacks.hp_mask = ngtcp2_crypto_hp_mask;
4904 	callbacks.update_key = ngtcp2_crypto_update_key_cb;
4905 	callbacks.delete_crypto_aead_ctx =
4906 		ngtcp2_crypto_delete_crypto_aead_ctx_cb;
4907 	callbacks.delete_crypto_cipher_ctx =
4908 		ngtcp2_crypto_delete_crypto_cipher_ctx_cb;
4909 	callbacks.get_path_challenge_data =
4910 		ngtcp2_crypto_get_path_challenge_data_cb;
4911 	callbacks.version_negotiation = ngtcp2_crypto_version_negotiation_cb;
4912 	callbacks.rand = doq_rand_cb;
4913 	callbacks.get_new_connection_id = doq_get_new_connection_id_cb;
4914 	callbacks.remove_connection_id = doq_remove_connection_id_cb;
4915 	callbacks.handshake_completed = doq_handshake_completed_cb;
4916 	callbacks.stream_open = doq_stream_open_cb;
4917 	callbacks.stream_close = doq_stream_close_cb;
4918 	callbacks.stream_reset = doq_stream_reset_cb;
4919 	callbacks.extend_max_stream_data = doq_extend_max_stream_data_cb;
4920 	callbacks.acked_stream_data_offset = doq_acked_stream_data_offset_cb;
4921 	callbacks.recv_stream_data = doq_recv_stream_data_cb;
4922 
4923 	ngtcp2_settings_default(&settings);
4924 	if(verbosity >= VERB_ALGO) {
4925 		settings.log_printf = doq_log_printf_cb;
4926 	}
4927 	settings.rand_ctx.native_handle = conn->doq_socket->rnd;
4928 	settings.initial_ts = doq_get_timestamp_nanosec();
4929 	settings.max_stream_window = 6*1024*1024;
4930 	settings.max_window = 6*1024*1024;
4931 #ifdef HAVE_STRUCT_NGTCP2_SETTINGS_TOKENLEN
4932 	settings.token = (void*)token;
4933 	settings.tokenlen = tokenlen;
4934 #else
4935 	settings.token.base = (void*)token;
4936 	settings.token.len = tokenlen;
4937 #endif
4938 
4939 	ngtcp2_transport_params_default(&params);
4940 	params.max_idle_timeout = conn->doq_socket->idle_timeout;
4941 	params.active_connection_id_limit = 7;
4942 	params.initial_max_stream_data_bidi_local = 256*1024;
4943 	params.initial_max_stream_data_bidi_remote = 256*1024;
4944 	params.initial_max_data = 1024*1024;
4945 	/* DoQ uses bidi streams, so we allow 0 uni streams. */
4946 	params.initial_max_streams_uni = 0;
4947 	/* Initial max on number of bidi streams the remote end can open.
4948 	 * That is the number of queries it can make, at first. */
4949 	params.initial_max_streams_bidi = 10;
4950 	if(ocid) {
4951 		ngtcp2_cid_init(&params.original_dcid, ocid, ocidlen);
4952 		ngtcp2_cid_init(&params.retry_scid, conn->key.dcid,
4953 			conn->key.dcidlen);
4954 		params.retry_scid_present = 1;
4955 	} else {
4956 		ngtcp2_cid_init(&params.original_dcid, conn->key.dcid,
4957 			conn->key.dcidlen);
4958 	}
4959 #ifdef HAVE_STRUCT_NGTCP2_TRANSPORT_PARAMS_ORIGINAL_DCID_PRESENT
4960 	params.original_dcid_present = 1;
4961 #endif
4962 	doq_fill_rand(conn->doq_socket->rnd, params.stateless_reset_token,
4963 		sizeof(params.stateless_reset_token));
4964 	sv_scid.datalen = conn->doq_socket->sv_scidlen;
4965 	lock_rw_wrlock(&conn->table->conid_lock);
4966 	if(!doq_conn_generate_new_conid(conn, sv_scid.data, sv_scid.datalen)) {
4967 		lock_rw_unlock(&conn->table->conid_lock);
4968 		return 0;
4969 	}
4970 
4971 	rv = ngtcp2_conn_server_new(&conn->conn, &scid_cid, &sv_scid, &path,
4972 		conn->version, &callbacks, &settings, &params, NULL, conn);
4973 	if(rv != 0) {
4974 		conn->conn = NULL;
4975 		lock_rw_unlock(&conn->table->conid_lock);
4976 		log_err("ngtcp2_conn_server_new failed: %s",
4977 			ngtcp2_strerror(rv));
4978 		return 0;
4979 	}
4980 	if(!doq_conn_setup_conids(conn)) {
4981 		lock_rw_unlock(&conn->table->conid_lock);
4982 		log_err("doq_conn_setup_conids failed: out of memory");
4983 		return 0;
4984 	}
4985 	lock_rw_unlock(&conn->table->conid_lock);
4986 	conn->ssl = doq_ssl_server_setup((SSL_CTX*)conn->doq_socket->ctx,
4987 		conn);
4988 	if(!conn->ssl) {
4989 		log_err("doq_ssl_server_setup failed");
4990 		return 0;
4991 	}
4992 #ifdef USE_NGTCP2_CRYPTO_OSSL
4993 	ngtcp2_conn_set_tls_native_handle(conn->conn, conn->ossl_ctx);
4994 #else
4995 	ngtcp2_conn_set_tls_native_handle(conn->conn, conn->ssl);
4996 #endif
4997 	doq_conn_write_enable(conn);
4998 	return 1;
4999 }
5000 
5001 struct doq_conid*
5002 doq_conid_find(struct doq_table* table, const uint8_t* data, size_t datalen)
5003 {
5004 	struct rbnode_type* node;
5005 	struct doq_conid key;
5006 	key.node.key = &key;
5007 	key.cid = (void*)data;
5008 	key.cidlen = datalen;
5009 	log_assert(table != NULL);
5010 	node = rbtree_search(table->conid_tree, &key);
5011 	if(node)
5012 		return (struct doq_conid*)node->key;
5013 	return NULL;
5014 }
5015 
5016 /** insert conid in the conid list */
5017 static void
5018 doq_conid_list_insert(struct doq_conn* conn, struct doq_conid* conid)
5019 {
5020 	conid->prev = NULL;
5021 	conid->next = conn->conid_list;
5022 	if(conn->conid_list)
5023 		conn->conid_list->prev = conid;
5024 	conn->conid_list = conid;
5025 }
5026 
5027 /** remove conid from the conid list */
5028 static void
5029 doq_conid_list_remove(struct doq_conn* conn, struct doq_conid* conid)
5030 {
5031 	if(conid->prev)
5032 		conid->prev->next = conid->next;
5033 	else	conn->conid_list = conid->next;
5034 	if(conid->next)
5035 		conid->next->prev = conid->prev;
5036 }
5037 
5038 /** create a doq_conid */
5039 static struct doq_conid*
5040 doq_conid_create(uint8_t* data, size_t datalen, struct doq_conn_key* key)
5041 {
5042 	struct doq_conid* conid;
5043 	conid = calloc(1, sizeof(*conid));
5044 	if(!conid)
5045 		return NULL;
5046 	conid->cid = memdup(data, datalen);
5047 	if(!conid->cid) {
5048 		free(conid);
5049 		return NULL;
5050 	}
5051 	conid->cidlen = datalen;
5052 	conid->node.key = conid;
5053 	conid->key = *key;
5054 	conid->key.dcid = memdup(key->dcid, key->dcidlen);
5055 	if(!conid->key.dcid) {
5056 		free(conid->cid);
5057 		free(conid);
5058 		return NULL;
5059 	}
5060 	return conid;
5061 }
5062 
5063 void
5064 doq_conid_delete(struct doq_conid* conid)
5065 {
5066 	if(!conid)
5067 		return;
5068 	free(conid->key.dcid);
5069 	free(conid->cid);
5070 	free(conid);
5071 }
5072 
5073 /** return true if the conid is for the conn. */
5074 static int
5075 conid_is_for_conn(struct doq_conn* conn, struct doq_conid* conid)
5076 {
5077 	if(conid->key.dcidlen == conn->key.dcidlen &&
5078 		memcmp(conid->key.dcid, conn->key.dcid, conid->key.dcidlen)==0
5079 		&& conid->key.paddr.addrlen == conn->key.paddr.addrlen &&
5080 		memcmp(&conid->key.paddr.addr, &conn->key.paddr.addr,
5081 			conid->key.paddr.addrlen) == 0 &&
5082 		conid->key.paddr.localaddrlen == conn->key.paddr.localaddrlen &&
5083 		memcmp(&conid->key.paddr.localaddr, &conn->key.paddr.localaddr,
5084 			conid->key.paddr.localaddrlen) == 0 &&
5085 		conid->key.paddr.ifindex == conn->key.paddr.ifindex)
5086 		return 1;
5087 	return 0;
5088 }
5089 
5090 int
5091 doq_conn_associate_conid(struct doq_conn* conn, uint8_t* data, size_t datalen)
5092 {
5093 	struct doq_conid* conid;
5094 	conid = doq_conid_find(conn->table, data, datalen);
5095 	if(conid && !conid_is_for_conn(conn, conid)) {
5096 		verbose(VERB_ALGO, "doq connection id already exists for "
5097 			"another doq_conn. Ignoring second connection id.");
5098 		/* Already exists to another conn, ignore it.
5099 		 * This works, in that the conid is listed in the doq_conn
5100 		 * conid_list element, and removed from there. So our conid
5101 		 * tree and list are fine, when created and removed.
5102 		 * The tree now does not have the lookup element pointing
5103 		 * to this connection. */
5104 		return 1;
5105 	}
5106 	if(conid)
5107 		return 1; /* already inserted */
5108 	conid = doq_conid_create(data, datalen, &conn->key);
5109 	if(!conid)
5110 		return 0;
5111 	doq_conid_list_insert(conn, conid);
5112 	(void)rbtree_insert(conn->table->conid_tree, &conid->node);
5113 	return 1;
5114 }
5115 
5116 void
5117 doq_conn_dissociate_conid(struct doq_conn* conn, const uint8_t* data,
5118 	size_t datalen)
5119 {
5120 	struct doq_conid* conid;
5121 	conid = doq_conid_find(conn->table, data, datalen);
5122 	if(conid && !conid_is_for_conn(conn, conid))
5123 		return;
5124 	if(conid) {
5125 		(void)rbtree_delete(conn->table->conid_tree,
5126 			conid->node.key);
5127 		doq_conid_list_remove(conn, conid);
5128 		doq_conid_delete(conid);
5129 	}
5130 }
5131 
5132 /** associate the scid array and also the dcid.
5133  * caller must hold the locks on conn and doq_table.conid_lock. */
5134 static int
5135 doq_conn_setup_id_array_and_dcid(struct doq_conn* conn,
5136 	struct ngtcp2_cid* scids, size_t num_scid)
5137 {
5138 	size_t i;
5139 	for(i=0; i<num_scid; i++) {
5140 		if(!doq_conn_associate_conid(conn, scids[i].data,
5141 			scids[i].datalen))
5142 			return 0;
5143 	}
5144 	if(!doq_conn_associate_conid(conn, conn->key.dcid, conn->key.dcidlen))
5145 		return 0;
5146 	return 1;
5147 }
5148 
5149 int
5150 doq_conn_setup_conids(struct doq_conn* conn)
5151 {
5152 	size_t num_scid =
5153 #ifndef HAVE_NGTCP2_CONN_GET_NUM_SCID
5154 		ngtcp2_conn_get_scid(conn->conn, NULL);
5155 #else
5156 		ngtcp2_conn_get_num_scid(conn->conn);
5157 #endif
5158 	if(num_scid <= 4) {
5159 		struct ngtcp2_cid ids[4];
5160 		/* Usually there are not that many scids when just accepted,
5161 		 * like only 2. */
5162 		ngtcp2_conn_get_scid(conn->conn, ids);
5163 		return doq_conn_setup_id_array_and_dcid(conn, ids, num_scid);
5164 	} else {
5165 		struct ngtcp2_cid *scids = calloc(num_scid,
5166 			sizeof(struct ngtcp2_cid));
5167 		if(!scids)
5168 			return 0;
5169 		ngtcp2_conn_get_scid(conn->conn, scids);
5170 		if(!doq_conn_setup_id_array_and_dcid(conn, scids, num_scid)) {
5171 			free(scids);
5172 			return 0;
5173 		}
5174 		free(scids);
5175 	}
5176 	return 1;
5177 }
5178 
5179 void
5180 doq_conn_clear_conids(struct doq_conn* conn)
5181 {
5182 	struct doq_conid* p, *next;
5183 	if(!conn)
5184 		return;
5185 	p = conn->conid_list;
5186 	while(p) {
5187 		next = p->next;
5188 		(void)rbtree_delete(conn->table->conid_tree, p->node.key);
5189 		doq_conid_delete(p);
5190 		p = next;
5191 	}
5192 	conn->conid_list = NULL;
5193 }
5194 
5195 ngtcp2_tstamp doq_get_timestamp_nanosec(void)
5196 {
5197 	struct timespec tp;
5198 	memset(&tp, 0, sizeof(tp));
5199 #ifdef CLOCK_BOOTTIME
5200 	if(clock_gettime(CLOCK_BOOTTIME, &tp) == -1) {
5201 #endif
5202 		if(clock_gettime(CLOCK_MONOTONIC, &tp) == -1) {
5203 			log_err("clock_gettime failed: %s", strerror(errno));
5204 		}
5205 #ifdef CLOCK_BOOTTIME
5206 	}
5207 #endif
5208 	return ((uint64_t)tp.tv_sec)*((uint64_t)1000000000) +
5209 		((uint64_t)tp.tv_nsec);
5210 }
5211 
5212 static struct timeval doq_get_timevalue(void)
5213 {
5214 	struct timeval tv;
5215 	memset(&tv, 0, sizeof(tv));
5216 	if(gettimeofday(&tv, NULL) < 0) {
5217 		log_err("gettimeofday failed: %s", strerror(errno));
5218 		memset(&tv, 0, sizeof(tv));
5219 	}
5220 	return tv;
5221 }
5222 
5223 /** doq start the closing period for the connection. */
5224 static int
5225 doq_conn_start_closing_period(struct comm_point* c, struct doq_conn* conn)
5226 {
5227 	struct ngtcp2_path_storage ps;
5228 	struct ngtcp2_pkt_info pi;
5229 	ngtcp2_ssize ret;
5230 	if(!conn)
5231 		return 1;
5232 	if(
5233 #ifdef HAVE_NGTCP2_CONN_IN_CLOSING_PERIOD
5234 		ngtcp2_conn_in_closing_period(conn->conn)
5235 #else
5236 		ngtcp2_conn_is_in_closing_period(conn->conn)
5237 #endif
5238 		)
5239 		return 1;
5240 	if(
5241 #ifdef HAVE_NGTCP2_CONN_IN_DRAINING_PERIOD
5242 		ngtcp2_conn_in_draining_period(conn->conn)
5243 #else
5244 		ngtcp2_conn_is_in_draining_period(conn->conn)
5245 #endif
5246 		) {
5247 		doq_conn_write_disable(conn);
5248 		return 1;
5249 	}
5250 	ngtcp2_path_storage_zero(&ps);
5251 	sldns_buffer_clear(c->doq_socket->pkt_buf);
5252 	/* the call to ngtcp2_conn_write_connection_close causes the
5253 	 * conn to be closed. It is now in the closing period. */
5254 	ret = ngtcp2_conn_write_connection_close(conn->conn, &ps.path,
5255 		&pi, sldns_buffer_begin(c->doq_socket->pkt_buf),
5256 		sldns_buffer_remaining(c->doq_socket->pkt_buf),
5257 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5258 		&conn->ccerr
5259 #else
5260 		&conn->last_error
5261 #endif
5262 		, doq_get_timestamp_nanosec());
5263 	if(ret < 0) {
5264 		log_err("doq ngtcp2_conn_write_connection_close failed: %s",
5265 			ngtcp2_strerror(ret));
5266 		return 0;
5267 	}
5268 	if(ret == 0) {
5269 		return 0;
5270 	}
5271 	sldns_buffer_set_position(c->doq_socket->pkt_buf, ret);
5272 	sldns_buffer_flip(c->doq_socket->pkt_buf);
5273 
5274 	/* The close packet is allocated, because it may have to be repeated.
5275 	 * When incoming packets have this connection dcid. */
5276 	conn->close_pkt = memdup(sldns_buffer_begin(c->doq_socket->pkt_buf),
5277 		sldns_buffer_limit(c->doq_socket->pkt_buf));
5278 	if(!conn->close_pkt) {
5279 		log_err("doq: could not allocate close packet: out of memory");
5280 		return 0;
5281 	}
5282 	conn->close_pkt_len = sldns_buffer_limit(c->doq_socket->pkt_buf);
5283 	conn->close_ecn = pi.ecn;
5284 	return 1;
5285 }
5286 
5287 /** doq send the close packet for the connection, perhaps again. */
5288 int
5289 doq_conn_send_close(struct comm_point* c, struct doq_conn* conn)
5290 {
5291 	if(!conn)
5292 		return 0;
5293 	if(!conn->close_pkt)
5294 		return 0;
5295 	if(conn->close_pkt_len > sldns_buffer_capacity(c->doq_socket->pkt_buf))
5296 		return 0;
5297 	sldns_buffer_clear(c->doq_socket->pkt_buf);
5298 	sldns_buffer_write(c->doq_socket->pkt_buf, conn->close_pkt, conn->close_pkt_len);
5299 	sldns_buffer_flip(c->doq_socket->pkt_buf);
5300 	verbose(VERB_ALGO, "doq send connection close");
5301 	doq_send_pkt(c, &conn->key.paddr, conn->close_ecn);
5302 	doq_conn_write_disable(conn);
5303 	return 1;
5304 }
5305 
5306 /** doq close the connection on error. If it returns a failure, it
5307  * does not wait to send a close, and the connection can be dropped. */
5308 static int
5309 doq_conn_close_error(struct comm_point* c, struct doq_conn* conn)
5310 {
5311 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5312 	if(conn->ccerr.type == NGTCP2_CCERR_TYPE_IDLE_CLOSE)
5313 		return 0;
5314 #else
5315 	if(conn->last_error.type ==
5316 		NGTCP2_CONNECTION_CLOSE_ERROR_CODE_TYPE_TRANSPORT_IDLE_CLOSE)
5317 		return 0;
5318 #endif
5319 	if(!doq_conn_start_closing_period(c, conn))
5320 		return 0;
5321 	if(
5322 #ifdef HAVE_NGTCP2_CONN_IN_DRAINING_PERIOD
5323 		ngtcp2_conn_in_draining_period(conn->conn)
5324 #else
5325 		ngtcp2_conn_is_in_draining_period(conn->conn)
5326 #endif
5327 		) {
5328 		doq_conn_write_disable(conn);
5329 		return 1;
5330 	}
5331 	doq_conn_write_enable(conn);
5332 	if(!doq_conn_send_close(c, conn))
5333 		return 0;
5334 	return 1;
5335 }
5336 
5337 int
5338 doq_conn_recv(struct comm_point* c, struct doq_pkt_addr* paddr,
5339 	struct doq_conn* conn, struct ngtcp2_pkt_info* pi, int* err_retry,
5340 	int* err_drop)
5341 {
5342 	int ret;
5343 	struct ngtcp2_path path;
5344 	memset(&path, 0, sizeof(path));
5345 	path.remote.addr = (struct sockaddr*)&paddr->addr;
5346 	path.remote.addrlen = paddr->addrlen;
5347 	path.local.addr = (struct sockaddr*)&paddr->localaddr;
5348 	path.local.addrlen = paddr->localaddrlen;
5349 
5350 	ret = ngtcp2_conn_read_pkt(conn->conn, &path, pi,
5351 		sldns_buffer_begin(c->doq_socket->pkt_buf),
5352 		sldns_buffer_limit(c->doq_socket->pkt_buf),
5353 		doq_get_timestamp_nanosec());
5354 	if(ret != 0) {
5355 		if(err_retry)
5356 			*err_retry = 0;
5357 		if(err_drop)
5358 			*err_drop = 0;
5359 		if(ret == NGTCP2_ERR_DRAINING) {
5360 			verbose(VERB_ALGO, "ngtcp2_conn_read_pkt returned %s",
5361 				ngtcp2_strerror(ret));
5362 			doq_conn_write_disable(conn);
5363 			return 0;
5364 		} else if(ret == NGTCP2_ERR_DROP_CONN) {
5365 			verbose(VERB_ALGO, "ngtcp2_conn_read_pkt returned %s",
5366 				ngtcp2_strerror(ret));
5367 			if(err_drop)
5368 				*err_drop = 1;
5369 			return 0;
5370 		} else if(ret == NGTCP2_ERR_RETRY) {
5371 			verbose(VERB_ALGO, "ngtcp2_conn_read_pkt returned %s",
5372 				ngtcp2_strerror(ret));
5373 			if(err_retry)
5374 				*err_retry = 1;
5375 			if(err_drop)
5376 				*err_drop = 1;
5377 			return 0;
5378 		} else if(ret == NGTCP2_ERR_CRYPTO) {
5379 			if(
5380 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5381 				!conn->ccerr.error_code
5382 #else
5383 				!conn->last_error.error_code
5384 #endif
5385 				) {
5386 				/* in picotls the tls alert may need to be
5387 				 * copied, but this is with openssl. And there
5388 				 * is conn->tls_alert. */
5389 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5390 				ngtcp2_ccerr_set_tls_alert(&conn->ccerr,
5391 					conn->tls_alert, NULL, 0);
5392 #else
5393 				ngtcp2_connection_close_error_set_transport_error_tls_alert(
5394 					&conn->last_error, conn->tls_alert,
5395 					NULL, 0);
5396 #endif
5397 			}
5398 		} else {
5399 			if(
5400 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5401 				!conn->ccerr.error_code
5402 #else
5403 				!conn->last_error.error_code
5404 #endif
5405 				) {
5406 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5407 				ngtcp2_ccerr_set_liberr(&conn->ccerr, ret,
5408 					NULL, 0);
5409 #else
5410 				ngtcp2_connection_close_error_set_transport_error_liberr(
5411 					&conn->last_error, ret, NULL, 0);
5412 #endif
5413 			}
5414 		}
5415 		log_err("ngtcp2_conn_read_pkt failed: %s",
5416 			ngtcp2_strerror(ret));
5417 		if(!doq_conn_close_error(c, conn)) {
5418 			if(err_drop)
5419 				*err_drop = 1;
5420 		}
5421 		return 0;
5422 	}
5423 	doq_conn_write_enable(conn);
5424 	return 1;
5425 }
5426 
5427 /** doq stream write is done */
5428 static void
5429 doq_stream_write_is_done(struct doq_conn* conn, struct doq_stream* stream)
5430 {
5431 	/* Cannot deallocate, the buffer may be needed for resends. */
5432 	doq_stream_off_write_list(conn, stream);
5433 }
5434 
5435 int
5436 doq_conn_write_streams(struct comm_point* c, struct doq_conn* conn,
5437 	int* err_drop)
5438 {
5439 	struct doq_stream* stream = conn->stream_write_first;
5440 	ngtcp2_path_storage ps;
5441 	size_t num_packets = 0, max_packets = 65535;
5442 	ngtcp2_path_storage_zero(&ps);
5443 
5444 	for(;;) {
5445 		int64_t stream_id;
5446 		uint32_t flags = 0;
5447 		ngtcp2_pkt_info pi;
5448 		ngtcp2_vec datav[2];
5449 		size_t datav_count = 0;
5450 		ngtcp2_ssize ret, ndatalen = 0;
5451 		int fin;
5452 
5453 		if(stream) {
5454 			/* data to send */
5455 			verbose(VERB_ALGO, "doq: doq_conn write stream %d",
5456 				(int)stream->stream_id);
5457 			stream_id = stream->stream_id;
5458 			fin = 1;
5459 			if(stream->nwrite < 2) {
5460 				datav[0].base = ((uint8_t*)&stream->
5461 					outlen_wire) + stream->nwrite;
5462 				datav[0].len = 2 - stream->nwrite;
5463 				datav[1].base = stream->out;
5464 				datav[1].len = stream->outlen;
5465 				datav_count = 2;
5466 			} else {
5467 				datav[0].base = stream->out +
5468 					(stream->nwrite-2);
5469 				datav[0].len = stream->outlen -
5470 					(stream->nwrite-2);
5471 				datav_count = 1;
5472 			}
5473 		} else {
5474 			/* no data to send */
5475 			verbose(VERB_ALGO, "doq: doq_conn write stream -1");
5476 			stream_id = -1;
5477 			fin = 0;
5478 			datav[0].base = NULL;
5479 			datav[0].len = 0;
5480 			datav_count = 1;
5481 		}
5482 
5483 		/* if more streams, set it to write more */
5484 		if(stream && stream->write_next)
5485 			flags |= NGTCP2_WRITE_STREAM_FLAG_MORE;
5486 		if(fin)
5487 			flags |= NGTCP2_WRITE_STREAM_FLAG_FIN;
5488 
5489 		sldns_buffer_clear(c->doq_socket->pkt_buf);
5490 		ret = ngtcp2_conn_writev_stream(conn->conn, &ps.path, &pi,
5491 			sldns_buffer_begin(c->doq_socket->pkt_buf),
5492 			sldns_buffer_remaining(c->doq_socket->pkt_buf),
5493 			&ndatalen, flags, stream_id, datav, datav_count,
5494 			doq_get_timestamp_nanosec());
5495 		if(ret < 0) {
5496 			if(ret == NGTCP2_ERR_WRITE_MORE) {
5497 				verbose(VERB_ALGO, "doq: write more, ndatalen %d", (int)ndatalen);
5498 				if(stream) {
5499 					if(ndatalen >= 0)
5500 						stream->nwrite += ndatalen;
5501 					if(stream->nwrite >= stream->outlen+2)
5502 						doq_stream_write_is_done(
5503 							conn, stream);
5504 					stream = stream->write_next;
5505 				}
5506 				continue;
5507 			} else if(ret == NGTCP2_ERR_STREAM_DATA_BLOCKED) {
5508 				verbose(VERB_ALGO, "doq: ngtcp2_conn_writev_stream returned NGTCP2_ERR_STREAM_DATA_BLOCKED");
5509 				if(stream) {
5510 					doq_stream_off_write_list(conn, stream);
5511 					stream = stream->write_next;
5512 					continue;
5513 				} else {
5514 					break;
5515 				}
5516 			} else if(ret == NGTCP2_ERR_STREAM_SHUT_WR) {
5517 				verbose(VERB_ALGO, "doq: ngtcp2_conn_writev_stream returned NGTCP2_ERR_STREAM_SHUT_WR");
5518 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5519 				ngtcp2_ccerr_set_application_error(
5520 					&conn->ccerr, DOQ_APP_ERROR_CODE, NULL, 0);
5521 #else
5522 				ngtcp2_connection_close_error_set_application_error(&conn->last_error, DOQ_APP_ERROR_CODE, NULL, 0);
5523 #endif
5524 				if(err_drop)
5525 					*err_drop = 0;
5526 				if(!doq_conn_close_error(c, conn)) {
5527 					if(err_drop)
5528 						*err_drop = 1;
5529 				}
5530 				return 0;
5531 			}
5532 
5533 			log_err("doq: ngtcp2_conn_writev_stream failed: %s",
5534 				ngtcp2_strerror(ret));
5535 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5536 			ngtcp2_ccerr_set_liberr(&conn->ccerr, ret, NULL, 0);
5537 #else
5538 			ngtcp2_connection_close_error_set_transport_error_liberr(
5539 				&conn->last_error, ret, NULL, 0);
5540 #endif
5541 			if(err_drop)
5542 				*err_drop = 0;
5543 			if(!doq_conn_close_error(c, conn)) {
5544 				if(err_drop)
5545 					*err_drop = 1;
5546 			}
5547 			return 0;
5548 		}
5549 		verbose(VERB_ALGO, "doq: writev_stream pkt size %d ndatawritten %d",
5550 			(int)ret, (int)ndatalen);
5551 
5552 		if(ndatalen >= 0 && stream) {
5553 			stream->nwrite += ndatalen;
5554 			if(stream->nwrite >= stream->outlen+2)
5555 				doq_stream_write_is_done(conn, stream);
5556 		}
5557 		if(ret == 0) {
5558 			/* congestion limited */
5559 			doq_conn_write_disable(conn);
5560 			ngtcp2_conn_update_pkt_tx_time(conn->conn,
5561 				doq_get_timestamp_nanosec());
5562 			return 1;
5563 		}
5564 		sldns_buffer_set_position(c->doq_socket->pkt_buf, ret);
5565 		sldns_buffer_flip(c->doq_socket->pkt_buf);
5566 		doq_send_pkt(c, &conn->key.paddr, pi.ecn);
5567 
5568 		if(c->doq_socket->have_blocked_pkt)
5569 			break;
5570 		if(++num_packets == max_packets)
5571 			break;
5572 		if(stream)
5573 			stream = stream->write_next;
5574 	}
5575 	ngtcp2_conn_update_pkt_tx_time(conn->conn, doq_get_timestamp_nanosec());
5576 	return 1;
5577 }
5578 
5579 void
5580 doq_conn_write_enable(struct doq_conn* conn)
5581 {
5582 	conn->write_interest = 1;
5583 }
5584 
5585 void
5586 doq_conn_write_disable(struct doq_conn* conn)
5587 {
5588 	conn->write_interest = 0;
5589 }
5590 
5591 /** doq append the connection to the write list */
5592 static void
5593 doq_conn_write_list_append(struct doq_table* table, struct doq_conn* conn)
5594 {
5595 	if(conn->on_write_list)
5596 		return;
5597 	conn->write_prev = table->write_list_last;
5598 	if(table->write_list_last)
5599 		table->write_list_last->write_next = conn;
5600 	else table->write_list_first = conn;
5601 	conn->write_next = NULL;
5602 	table->write_list_last = conn;
5603 	conn->on_write_list = 1;
5604 }
5605 
5606 void
5607 doq_conn_write_list_remove(struct doq_table* table, struct doq_conn* conn)
5608 {
5609 	if(!conn->on_write_list)
5610 		return;
5611 	if(conn->write_next)
5612 		conn->write_next->write_prev = conn->write_prev;
5613 	else table->write_list_last = conn->write_prev;
5614 	if(conn->write_prev)
5615 		conn->write_prev->write_next = conn->write_next;
5616 	else table->write_list_first = conn->write_next;
5617 	conn->write_prev = NULL;
5618 	conn->write_next = NULL;
5619 	conn->on_write_list = 0;
5620 }
5621 
5622 void
5623 doq_conn_set_write_list(struct doq_table* table, struct doq_conn* conn)
5624 {
5625 	if(conn->write_interest && conn->on_write_list)
5626 		return;
5627 	if(!conn->write_interest && !conn->on_write_list)
5628 		return;
5629 	if(conn->write_interest)
5630 		doq_conn_write_list_append(table, conn);
5631 	else doq_conn_write_list_remove(table, conn);
5632 }
5633 
5634 struct doq_conn*
5635 doq_table_pop_first(struct doq_table* table)
5636 {
5637 	struct doq_conn* conn = table->write_list_first;
5638 	if(!conn)
5639 		return NULL;
5640 	lock_basic_lock(&conn->lock);
5641 	table->write_list_first = conn->write_next;
5642 	if(conn->write_next)
5643 		conn->write_next->write_prev = NULL;
5644 	else table->write_list_last = NULL;
5645 	conn->write_next = NULL;
5646 	conn->write_prev = NULL;
5647 	conn->on_write_list = 0;
5648 	return conn;
5649 }
5650 
5651 int
5652 doq_conn_check_timer(struct doq_conn* conn, struct timeval* tv, ngtcp2_tstamp* ts)
5653 {
5654 	ngtcp2_tstamp doq_expiry = ngtcp2_conn_get_expiry(conn->conn);
5655 	ngtcp2_tstamp doq_now = doq_get_timestamp_nanosec();
5656 	ngtcp2_tstamp t;
5657 	struct timeval now = doq_get_timevalue();
5658 
5659 	if(doq_expiry <= doq_now || doq_expiry == UINT64_MAX) {
5660 		/* UINT64_MAX means there is no next expiry. */
5661 		/* The timer has already expired, add with zero timeout.
5662 		 * This should call the callback straight away. Calling it
5663 		 * from the event callbacks is cleaner than calling it here,
5664 		 * because then it is always called with the same locks and
5665 		 * so on. This routine only has the conn.lock. */
5666 		t = doq_now;
5667 		memcpy(tv, &now, sizeof(*tv));
5668 	} else {
5669 		t = doq_expiry;
5670 		memset(tv, 0, sizeof(*tv));
5671 		tv->tv_sec = (doq_expiry - doq_now) / NGTCP2_SECONDS;
5672 		tv->tv_usec = ((doq_expiry - doq_now) / NGTCP2_MICROSECONDS)%1000000;
5673 		timeval_add(tv, &now);
5674 	}
5675 
5676 	*ts = t;
5677 
5678 	/* If we already have a timer, is it the right value? */
5679 	if(conn->timer.timer_in_tree || conn->timer.timer_in_list) {
5680 		if(conn->timer.time_mono == *ts)
5681 			return 0;
5682 	}
5683 	return 1;
5684 }
5685 
5686 /* doq print connection log */
5687 static void
5688 doq_conn_log_line(struct doq_conn* conn, char* s)
5689 {
5690 	char remotestr[256], localstr[256];
5691 	addr_to_str((void*)&conn->key.paddr.addr, conn->key.paddr.addrlen,
5692 		remotestr, sizeof(remotestr));
5693 	addr_to_str((void*)&conn->key.paddr.localaddr,
5694 		conn->key.paddr.localaddrlen, localstr, sizeof(localstr));
5695 	log_info("doq conn %s %s %s", remotestr, localstr, s);
5696 }
5697 
5698 int
5699 doq_conn_handle_timeout(struct doq_conn* conn)
5700 {
5701 	int rv;
5702 
5703 	if(verbosity >= VERB_ALGO)
5704 		doq_conn_log_line(conn, "timeout");
5705 
5706 	rv = ngtcp2_conn_handle_expiry(conn->conn, doq_get_timestamp_nanosec());
5707 	if(rv != 0) {
5708 		verbose(VERB_ALGO, "ngtcp2_conn_handle_expiry failed: %s",
5709 			ngtcp2_strerror(rv));
5710 #ifdef HAVE_NGTCP2_CCERR_DEFAULT
5711 		ngtcp2_ccerr_set_liberr(&conn->ccerr, rv, NULL, 0);
5712 #else
5713 		ngtcp2_connection_close_error_set_transport_error_liberr(
5714 			&conn->last_error, rv, NULL, 0);
5715 #endif
5716 		if(!doq_conn_close_error(conn->doq_socket->cp, conn)) {
5717 			/* failed, return for deletion */
5718 			return 0;
5719 		}
5720 		return 1;
5721 	}
5722 	doq_conn_write_enable(conn);
5723 	if(!doq_conn_write_streams(conn->doq_socket->cp, conn, NULL)) {
5724 		/* failed, return for deletion. */
5725 		return 0;
5726 	}
5727 	return 1;
5728 }
5729 
5730 void
5731 doq_table_quic_size_add(struct doq_table* table, size_t add)
5732 {
5733 	lock_basic_lock(&table->size_lock);
5734 	table->current_size += add;
5735 	lock_basic_unlock(&table->size_lock);
5736 }
5737 
5738 void
5739 doq_table_quic_size_subtract(struct doq_table* table, size_t subtract)
5740 {
5741 	lock_basic_lock(&table->size_lock);
5742 	if(table->current_size < subtract)
5743 		table->current_size = 0;
5744 	else	table->current_size -= subtract;
5745 	lock_basic_unlock(&table->size_lock);
5746 }
5747 
5748 int
5749 doq_table_quic_size_available(struct doq_table* table,
5750 	struct config_file* cfg, size_t mem)
5751 {
5752 	size_t cur;
5753 	if (!table)
5754 		return 0;
5755 	lock_basic_lock(&table->size_lock);
5756 	cur = table->current_size;
5757 	lock_basic_unlock(&table->size_lock);
5758 
5759 	if(cur + mem > cfg->quic_size)
5760 		return 0;
5761 	return 1;
5762 }
5763 
5764 size_t doq_table_quic_size_get(struct doq_table* table)
5765 {
5766 	size_t sz;
5767 	if(!table)
5768 		return 0;
5769 	lock_basic_lock(&table->size_lock);
5770 	sz = table->current_size;
5771 	lock_basic_unlock(&table->size_lock);
5772 	return sz;
5773 }
5774 #endif /* HAVE_NGTCP2 */
5775