xref: /freebsd/contrib/unbound/util/netevent.c (revision d30a1689f5b37e78ea189232a8b94a7011dc0dc8)
1 /*
2  * util/netevent.c - event notification
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 contains event notification functions.
40  */
41 #include "config.h"
42 #include "util/netevent.h"
43 #include "util/ub_event.h"
44 #include "util/log.h"
45 #include "util/net_help.h"
46 #include "util/tcp_conn_limit.h"
47 #include "util/fptr_wlist.h"
48 #include "sldns/pkthdr.h"
49 #include "sldns/sbuffer.h"
50 #include "sldns/str2wire.h"
51 #include "dnstap/dnstap.h"
52 #include "dnscrypt/dnscrypt.h"
53 #include "services/listen_dnsport.h"
54 #ifdef HAVE_SYS_TYPES_H
55 #include <sys/types.h>
56 #endif
57 #ifdef HAVE_SYS_SOCKET_H
58 #include <sys/socket.h>
59 #endif
60 #ifdef HAVE_NETDB_H
61 #include <netdb.h>
62 #endif
63 
64 #ifdef HAVE_OPENSSL_SSL_H
65 #include <openssl/ssl.h>
66 #endif
67 #ifdef HAVE_OPENSSL_ERR_H
68 #include <openssl/err.h>
69 #endif
70 
71 /* -------- Start of local definitions -------- */
72 /** if CMSG_ALIGN is not defined on this platform, a workaround */
73 #ifndef CMSG_ALIGN
74 #  ifdef __CMSG_ALIGN
75 #    define CMSG_ALIGN(n) __CMSG_ALIGN(n)
76 #  elif defined(CMSG_DATA_ALIGN)
77 #    define CMSG_ALIGN _CMSG_DATA_ALIGN
78 #  else
79 #    define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
80 #  endif
81 #endif
82 
83 /** if CMSG_LEN is not defined on this platform, a workaround */
84 #ifndef CMSG_LEN
85 #  define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
86 #endif
87 
88 /** if CMSG_SPACE is not defined on this platform, a workaround */
89 #ifndef CMSG_SPACE
90 #  ifdef _CMSG_HDR_ALIGN
91 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
92 #  else
93 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
94 #  endif
95 #endif
96 
97 /** The TCP writing query timeout in milliseconds */
98 #define TCP_QUERY_TIMEOUT 120000
99 /** The minimum actual TCP timeout to use, regardless of what we advertise,
100  * in msec */
101 #define TCP_QUERY_TIMEOUT_MINIMUM 200
102 
103 #ifndef NONBLOCKING_IS_BROKEN
104 /** number of UDP reads to perform per read indication from select */
105 #define NUM_UDP_PER_SELECT 100
106 #else
107 #define NUM_UDP_PER_SELECT 1
108 #endif
109 
110 /**
111  * The internal event structure for keeping ub_event info for the event.
112  * Possibly other structures (list, tree) this is part of.
113  */
114 struct internal_event {
115 	/** the comm base */
116 	struct comm_base* base;
117 	/** ub_event event type */
118 	struct ub_event* ev;
119 };
120 
121 /**
122  * Internal base structure, so that every thread has its own events.
123  */
124 struct internal_base {
125 	/** ub_event event_base type. */
126 	struct ub_event_base* base;
127 	/** seconds time pointer points here */
128 	time_t secs;
129 	/** timeval with current time */
130 	struct timeval now;
131 	/** the event used for slow_accept timeouts */
132 	struct ub_event* slow_accept;
133 	/** true if slow_accept is enabled */
134 	int slow_accept_enabled;
135 };
136 
137 /**
138  * Internal timer structure, to store timer event in.
139  */
140 struct internal_timer {
141 	/** the super struct from which derived */
142 	struct comm_timer super;
143 	/** the comm base */
144 	struct comm_base* base;
145 	/** ub_event event type */
146 	struct ub_event* ev;
147 	/** is timer enabled */
148 	uint8_t enabled;
149 };
150 
151 /**
152  * Internal signal structure, to store signal event in.
153  */
154 struct internal_signal {
155 	/** ub_event event type */
156 	struct ub_event* ev;
157 	/** next in signal list */
158 	struct internal_signal* next;
159 };
160 
161 /** create a tcp handler with a parent */
162 static struct comm_point* comm_point_create_tcp_handler(
163 	struct comm_base *base, struct comm_point* parent, size_t bufsize,
164 	struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
165 	void* callback_arg, struct unbound_socket* socket);
166 
167 /* -------- End of local definitions -------- */
168 
169 struct comm_base*
170 comm_base_create(int sigs)
171 {
172 	struct comm_base* b = (struct comm_base*)calloc(1,
173 		sizeof(struct comm_base));
174 	const char *evnm="event", *evsys="", *evmethod="";
175 
176 	if(!b)
177 		return NULL;
178 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
179 	if(!b->eb) {
180 		free(b);
181 		return NULL;
182 	}
183 	b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now);
184 	if(!b->eb->base) {
185 		free(b->eb);
186 		free(b);
187 		return NULL;
188 	}
189 	ub_comm_base_now(b);
190 	ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod);
191 	verbose(VERB_ALGO, "%s %s uses %s method.", evnm, evsys, evmethod);
192 	return b;
193 }
194 
195 struct comm_base*
196 comm_base_create_event(struct ub_event_base* base)
197 {
198 	struct comm_base* b = (struct comm_base*)calloc(1,
199 		sizeof(struct comm_base));
200 	if(!b)
201 		return NULL;
202 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
203 	if(!b->eb) {
204 		free(b);
205 		return NULL;
206 	}
207 	b->eb->base = base;
208 	ub_comm_base_now(b);
209 	return b;
210 }
211 
212 void
213 comm_base_delete(struct comm_base* b)
214 {
215 	if(!b)
216 		return;
217 	if(b->eb->slow_accept_enabled) {
218 		if(ub_event_del(b->eb->slow_accept) != 0) {
219 			log_err("could not event_del slow_accept");
220 		}
221 		ub_event_free(b->eb->slow_accept);
222 	}
223 	ub_event_base_free(b->eb->base);
224 	b->eb->base = NULL;
225 	free(b->eb);
226 	free(b);
227 }
228 
229 void
230 comm_base_delete_no_base(struct comm_base* b)
231 {
232 	if(!b)
233 		return;
234 	if(b->eb->slow_accept_enabled) {
235 		if(ub_event_del(b->eb->slow_accept) != 0) {
236 			log_err("could not event_del slow_accept");
237 		}
238 		ub_event_free(b->eb->slow_accept);
239 	}
240 	b->eb->base = NULL;
241 	free(b->eb);
242 	free(b);
243 }
244 
245 void
246 comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv)
247 {
248 	*tt = &b->eb->secs;
249 	*tv = &b->eb->now;
250 }
251 
252 void
253 comm_base_dispatch(struct comm_base* b)
254 {
255 	int retval;
256 	retval = ub_event_base_dispatch(b->eb->base);
257 	if(retval < 0) {
258 		fatal_exit("event_dispatch returned error %d, "
259 			"errno is %s", retval, strerror(errno));
260 	}
261 }
262 
263 void comm_base_exit(struct comm_base* b)
264 {
265 	if(ub_event_base_loopexit(b->eb->base) != 0) {
266 		log_err("Could not loopexit");
267 	}
268 }
269 
270 void comm_base_set_slow_accept_handlers(struct comm_base* b,
271 	void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
272 {
273 	b->stop_accept = stop_acc;
274 	b->start_accept = start_acc;
275 	b->cb_arg = arg;
276 }
277 
278 struct ub_event_base* comm_base_internal(struct comm_base* b)
279 {
280 	return b->eb->base;
281 }
282 
283 /** see if errno for udp has to be logged or not uses globals */
284 static int
285 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
286 {
287 	/* do not log transient errors (unless high verbosity) */
288 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
289 	switch(errno) {
290 #  ifdef ENETUNREACH
291 		case ENETUNREACH:
292 #  endif
293 #  ifdef EHOSTDOWN
294 		case EHOSTDOWN:
295 #  endif
296 #  ifdef EHOSTUNREACH
297 		case EHOSTUNREACH:
298 #  endif
299 #  ifdef ENETDOWN
300 		case ENETDOWN:
301 #  endif
302 		case EPERM:
303 		case EACCES:
304 			if(verbosity < VERB_ALGO)
305 				return 0;
306 		default:
307 			break;
308 	}
309 #endif
310 	/* permission denied is gotten for every send if the
311 	 * network is disconnected (on some OS), squelch it */
312 	if( ((errno == EPERM)
313 #  ifdef EADDRNOTAVAIL
314 		/* 'Cannot assign requested address' also when disconnected */
315 		|| (errno == EADDRNOTAVAIL)
316 #  endif
317 		) && verbosity < VERB_ALGO)
318 		return 0;
319 #  ifdef EADDRINUSE
320 	/* If SO_REUSEADDR is set, we could try to connect to the same server
321 	 * from the same source port twice. */
322 	if(errno == EADDRINUSE && verbosity < VERB_DETAIL)
323 		return 0;
324 #  endif
325 	/* squelch errors where people deploy AAAA ::ffff:bla for
326 	 * authority servers, which we try for intranets. */
327 	if(errno == EINVAL && addr_is_ip4mapped(
328 		(struct sockaddr_storage*)addr, addrlen) &&
329 		verbosity < VERB_DETAIL)
330 		return 0;
331 	/* SO_BROADCAST sockopt can give access to 255.255.255.255,
332 	 * but a dns cache does not need it. */
333 	if(errno == EACCES && addr_is_broadcast(
334 		(struct sockaddr_storage*)addr, addrlen) &&
335 		verbosity < VERB_DETAIL)
336 		return 0;
337 	return 1;
338 }
339 
340 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
341 {
342 	return udp_send_errno_needs_log(addr, addrlen);
343 }
344 
345 /* send a UDP reply */
346 int
347 comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet,
348 	struct sockaddr* addr, socklen_t addrlen, int is_connected)
349 {
350 	ssize_t sent;
351 	log_assert(c->fd != -1);
352 #ifdef UNBOUND_DEBUG
353 	if(sldns_buffer_remaining(packet) == 0)
354 		log_err("error: send empty UDP packet");
355 #endif
356 	log_assert(addr && addrlen > 0);
357 	if(!is_connected) {
358 		sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
359 			sldns_buffer_remaining(packet), 0,
360 			addr, addrlen);
361 	} else {
362 		sent = send(c->fd, (void*)sldns_buffer_begin(packet),
363 			sldns_buffer_remaining(packet), 0);
364 	}
365 	if(sent == -1) {
366 		/* try again and block, waiting for IO to complete,
367 		 * we want to send the answer, and we will wait for
368 		 * the ethernet interface buffer to have space. */
369 #ifndef USE_WINSOCK
370 		if(errno == EAGAIN ||
371 #  ifdef EWOULDBLOCK
372 			errno == EWOULDBLOCK ||
373 #  endif
374 			errno == ENOBUFS) {
375 #else
376 		if(WSAGetLastError() == WSAEINPROGRESS ||
377 			WSAGetLastError() == WSAENOBUFS ||
378 			WSAGetLastError() == WSAEWOULDBLOCK) {
379 #endif
380 			int e;
381 			fd_set_block(c->fd);
382 			if (!is_connected) {
383 				sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
384 					sldns_buffer_remaining(packet), 0,
385 					addr, addrlen);
386 			} else {
387 				sent = send(c->fd, (void*)sldns_buffer_begin(packet),
388 					sldns_buffer_remaining(packet), 0);
389 			}
390 			e = errno;
391 			fd_set_nonblock(c->fd);
392 			errno = e;
393 		}
394 	}
395 	if(sent == -1) {
396 		if(!udp_send_errno_needs_log(addr, addrlen))
397 			return 0;
398 		if (!is_connected) {
399 			verbose(VERB_OPS, "sendto failed: %s", sock_strerror(errno));
400 		} else {
401 			verbose(VERB_OPS, "send failed: %s", sock_strerror(errno));
402 		}
403 		if(addr)
404 			log_addr(VERB_OPS, "remote address is",
405 				(struct sockaddr_storage*)addr, addrlen);
406 		return 0;
407 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
408 		log_err("sent %d in place of %d bytes",
409 			(int)sent, (int)sldns_buffer_remaining(packet));
410 		return 0;
411 	}
412 	return 1;
413 }
414 
415 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
416 /** print debug ancillary info */
417 static void p_ancil(const char* str, struct comm_reply* r)
418 {
419 	if(r->srctype != 4 && r->srctype != 6) {
420 		log_info("%s: unknown srctype %d", str, r->srctype);
421 		return;
422 	}
423 
424 	if(r->srctype == 6) {
425 #ifdef IPV6_PKTINFO
426 		char buf[1024];
427 		if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
428 			buf, (socklen_t)sizeof(buf)) == 0) {
429 			(void)strlcpy(buf, "(inet_ntop error)", sizeof(buf));
430 		}
431 		buf[sizeof(buf)-1]=0;
432 		log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
433 #endif
434 	} else if(r->srctype == 4) {
435 #ifdef IP_PKTINFO
436 		char buf1[1024], buf2[1024];
437 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
438 			buf1, (socklen_t)sizeof(buf1)) == 0) {
439 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
440 		}
441 		buf1[sizeof(buf1)-1]=0;
442 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
443 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
444 			buf2, (socklen_t)sizeof(buf2)) == 0) {
445 			(void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2));
446 		}
447 		buf2[sizeof(buf2)-1]=0;
448 #else
449 		buf2[0]=0;
450 #endif
451 		log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
452 			buf1, buf2);
453 #elif defined(IP_RECVDSTADDR)
454 		char buf1[1024];
455 		if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
456 			buf1, (socklen_t)sizeof(buf1)) == 0) {
457 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
458 		}
459 		buf1[sizeof(buf1)-1]=0;
460 		log_info("%s: %s", str, buf1);
461 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */
462 	}
463 }
464 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */
465 
466 /** send a UDP reply over specified interface*/
467 static int
468 comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet,
469 	struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
470 {
471 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
472 	ssize_t sent;
473 	struct msghdr msg;
474 	struct iovec iov[1];
475 	union {
476 		struct cmsghdr hdr;
477 		char buf[256];
478 	} control;
479 #ifndef S_SPLINT_S
480 	struct cmsghdr *cmsg;
481 #endif /* S_SPLINT_S */
482 
483 	log_assert(c->fd != -1);
484 #ifdef UNBOUND_DEBUG
485 	if(sldns_buffer_remaining(packet) == 0)
486 		log_err("error: send empty UDP packet");
487 #endif
488 	log_assert(addr && addrlen > 0);
489 
490 	msg.msg_name = addr;
491 	msg.msg_namelen = addrlen;
492 	iov[0].iov_base = sldns_buffer_begin(packet);
493 	iov[0].iov_len = sldns_buffer_remaining(packet);
494 	msg.msg_iov = iov;
495 	msg.msg_iovlen = 1;
496 	msg.msg_control = control.buf;
497 #ifndef S_SPLINT_S
498 	msg.msg_controllen = sizeof(control.buf);
499 #endif /* S_SPLINT_S */
500 	msg.msg_flags = 0;
501 
502 #ifndef S_SPLINT_S
503 	cmsg = CMSG_FIRSTHDR(&msg);
504 	if(r->srctype == 4) {
505 #ifdef IP_PKTINFO
506 		void* cmsg_data;
507 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
508 		log_assert(msg.msg_controllen <= sizeof(control.buf));
509 		cmsg->cmsg_level = IPPROTO_IP;
510 		cmsg->cmsg_type = IP_PKTINFO;
511 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
512 			sizeof(struct in_pktinfo));
513 		/* unset the ifindex to not bypass the routing tables */
514 		cmsg_data = CMSG_DATA(cmsg);
515 		((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0;
516 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
517 #elif defined(IP_SENDSRCADDR)
518 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
519 		log_assert(msg.msg_controllen <= sizeof(control.buf));
520 		cmsg->cmsg_level = IPPROTO_IP;
521 		cmsg->cmsg_type = IP_SENDSRCADDR;
522 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
523 			sizeof(struct in_addr));
524 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
525 #else
526 		verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
527 		msg.msg_control = NULL;
528 #endif /* IP_PKTINFO or IP_SENDSRCADDR */
529 	} else if(r->srctype == 6) {
530 		void* cmsg_data;
531 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
532 		log_assert(msg.msg_controllen <= sizeof(control.buf));
533 		cmsg->cmsg_level = IPPROTO_IPV6;
534 		cmsg->cmsg_type = IPV6_PKTINFO;
535 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
536 			sizeof(struct in6_pktinfo));
537 		/* unset the ifindex to not bypass the routing tables */
538 		cmsg_data = CMSG_DATA(cmsg);
539 		((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0;
540 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
541 	} else {
542 		/* try to pass all 0 to use default route */
543 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
544 		log_assert(msg.msg_controllen <= sizeof(control.buf));
545 		cmsg->cmsg_level = IPPROTO_IPV6;
546 		cmsg->cmsg_type = IPV6_PKTINFO;
547 		memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
548 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
549 	}
550 #endif /* S_SPLINT_S */
551 	if(verbosity >= VERB_ALGO)
552 		p_ancil("send_udp over interface", r);
553 	sent = sendmsg(c->fd, &msg, 0);
554 	if(sent == -1) {
555 		/* try again and block, waiting for IO to complete,
556 		 * we want to send the answer, and we will wait for
557 		 * the ethernet interface buffer to have space. */
558 #ifndef USE_WINSOCK
559 		if(errno == EAGAIN ||
560 #  ifdef EWOULDBLOCK
561 			errno == EWOULDBLOCK ||
562 #  endif
563 			errno == ENOBUFS) {
564 #else
565 		if(WSAGetLastError() == WSAEINPROGRESS ||
566 			WSAGetLastError() == WSAENOBUFS ||
567 			WSAGetLastError() == WSAEWOULDBLOCK) {
568 #endif
569 			int e;
570 			fd_set_block(c->fd);
571 			sent = sendmsg(c->fd, &msg, 0);
572 			e = errno;
573 			fd_set_nonblock(c->fd);
574 			errno = e;
575 		}
576 	}
577 	if(sent == -1) {
578 		if(!udp_send_errno_needs_log(addr, addrlen))
579 			return 0;
580 		verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
581 		log_addr(VERB_OPS, "remote address is",
582 			(struct sockaddr_storage*)addr, addrlen);
583 #ifdef __NetBSD__
584 		/* netbsd 7 has IP_PKTINFO for recv but not send */
585 		if(errno == EINVAL && r->srctype == 4)
586 			log_err("sendmsg: No support for sendmsg(IP_PKTINFO). "
587 				"Please disable interface-automatic");
588 #endif
589 		return 0;
590 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
591 		log_err("sent %d in place of %d bytes",
592 			(int)sent, (int)sldns_buffer_remaining(packet));
593 		return 0;
594 	}
595 	return 1;
596 #else
597 	(void)c;
598 	(void)packet;
599 	(void)addr;
600 	(void)addrlen;
601 	(void)r;
602 	log_err("sendmsg: IPV6_PKTINFO not supported");
603 	return 0;
604 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
605 }
606 
607 /** return true is UDP receive error needs to be logged */
608 static int udp_recv_needs_log(int err)
609 {
610 	switch(err) {
611 	case EACCES: /* some hosts send ICMP 'Permission Denied' */
612 #ifndef USE_WINSOCK
613 	case ECONNREFUSED:
614 #  ifdef ENETUNREACH
615 	case ENETUNREACH:
616 #  endif
617 #  ifdef EHOSTDOWN
618 	case EHOSTDOWN:
619 #  endif
620 #  ifdef EHOSTUNREACH
621 	case EHOSTUNREACH:
622 #  endif
623 #  ifdef ENETDOWN
624 	case ENETDOWN:
625 #  endif
626 #else /* USE_WINSOCK */
627 	case WSAECONNREFUSED:
628 	case WSAENETUNREACH:
629 	case WSAEHOSTDOWN:
630 	case WSAEHOSTUNREACH:
631 	case WSAENETDOWN:
632 #endif
633 		if(verbosity >= VERB_ALGO)
634 			return 1;
635 		return 0;
636 	default:
637 		break;
638 	}
639 	return 1;
640 }
641 
642 void
643 comm_point_udp_ancil_callback(int fd, short event, void* arg)
644 {
645 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
646 	struct comm_reply rep;
647 	struct msghdr msg;
648 	struct iovec iov[1];
649 	ssize_t rcv;
650 	union {
651 		struct cmsghdr hdr;
652 		char buf[256];
653 	} ancil;
654 	int i;
655 #ifndef S_SPLINT_S
656 	struct cmsghdr* cmsg;
657 #endif /* S_SPLINT_S */
658 
659 	rep.c = (struct comm_point*)arg;
660 	log_assert(rep.c->type == comm_udp);
661 
662 	if(!(event&UB_EV_READ))
663 		return;
664 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
665 	ub_comm_base_now(rep.c->ev->base);
666 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
667 		sldns_buffer_clear(rep.c->buffer);
668 		rep.addrlen = (socklen_t)sizeof(rep.addr);
669 		log_assert(fd != -1);
670 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
671 		msg.msg_name = &rep.addr;
672 		msg.msg_namelen = (socklen_t)sizeof(rep.addr);
673 		iov[0].iov_base = sldns_buffer_begin(rep.c->buffer);
674 		iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer);
675 		msg.msg_iov = iov;
676 		msg.msg_iovlen = 1;
677 		msg.msg_control = ancil.buf;
678 #ifndef S_SPLINT_S
679 		msg.msg_controllen = sizeof(ancil.buf);
680 #endif /* S_SPLINT_S */
681 		msg.msg_flags = 0;
682 		rcv = recvmsg(fd, &msg, 0);
683 		if(rcv == -1) {
684 			if(errno != EAGAIN && errno != EINTR
685 				&& udp_recv_needs_log(errno)) {
686 				log_err("recvmsg failed: %s", strerror(errno));
687 			}
688 			return;
689 		}
690 		rep.addrlen = msg.msg_namelen;
691 		sldns_buffer_skip(rep.c->buffer, rcv);
692 		sldns_buffer_flip(rep.c->buffer);
693 		rep.srctype = 0;
694 #ifndef S_SPLINT_S
695 		for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
696 			cmsg = CMSG_NXTHDR(&msg, cmsg)) {
697 			if( cmsg->cmsg_level == IPPROTO_IPV6 &&
698 				cmsg->cmsg_type == IPV6_PKTINFO) {
699 				rep.srctype = 6;
700 				memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
701 					sizeof(struct in6_pktinfo));
702 				break;
703 #ifdef IP_PKTINFO
704 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
705 				cmsg->cmsg_type == IP_PKTINFO) {
706 				rep.srctype = 4;
707 				memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
708 					sizeof(struct in_pktinfo));
709 				break;
710 #elif defined(IP_RECVDSTADDR)
711 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
712 				cmsg->cmsg_type == IP_RECVDSTADDR) {
713 				rep.srctype = 4;
714 				memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
715 					sizeof(struct in_addr));
716 				break;
717 #endif /* IP_PKTINFO or IP_RECVDSTADDR */
718 			}
719 		}
720 		if(verbosity >= VERB_ALGO)
721 			p_ancil("receive_udp on interface", &rep);
722 #endif /* S_SPLINT_S */
723 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
724 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
725 			/* send back immediate reply */
726 			(void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer,
727 				(struct sockaddr*)&rep.addr, rep.addrlen, &rep);
728 		}
729 		if(!rep.c || rep.c->fd == -1) /* commpoint closed */
730 			break;
731 	}
732 #else
733 	(void)fd;
734 	(void)event;
735 	(void)arg;
736 	fatal_exit("recvmsg: No support for IPV6_PKTINFO; IP_PKTINFO or IP_RECVDSTADDR. "
737 		"Please disable interface-automatic");
738 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */
739 }
740 
741 void
742 comm_point_udp_callback(int fd, short event, void* arg)
743 {
744 	struct comm_reply rep;
745 	ssize_t rcv;
746 	int i;
747 	struct sldns_buffer *buffer;
748 
749 	rep.c = (struct comm_point*)arg;
750 	log_assert(rep.c->type == comm_udp);
751 
752 	if(!(event&UB_EV_READ))
753 		return;
754 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
755 	ub_comm_base_now(rep.c->ev->base);
756 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
757 		sldns_buffer_clear(rep.c->buffer);
758 		rep.addrlen = (socklen_t)sizeof(rep.addr);
759 		log_assert(fd != -1);
760 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
761 		rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer),
762 			sldns_buffer_remaining(rep.c->buffer), 0,
763 			(struct sockaddr*)&rep.addr, &rep.addrlen);
764 		if(rcv == -1) {
765 #ifndef USE_WINSOCK
766 			if(errno != EAGAIN && errno != EINTR
767 				&& udp_recv_needs_log(errno))
768 				log_err("recvfrom %d failed: %s",
769 					fd, strerror(errno));
770 #else
771 			if(WSAGetLastError() != WSAEINPROGRESS &&
772 				WSAGetLastError() != WSAECONNRESET &&
773 				WSAGetLastError()!= WSAEWOULDBLOCK &&
774 				udp_recv_needs_log(WSAGetLastError()))
775 				log_err("recvfrom failed: %s",
776 					wsa_strerror(WSAGetLastError()));
777 #endif
778 			return;
779 		}
780 		sldns_buffer_skip(rep.c->buffer, rcv);
781 		sldns_buffer_flip(rep.c->buffer);
782 		rep.srctype = 0;
783 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
784 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
785 			/* send back immediate reply */
786 #ifdef USE_DNSCRYPT
787 			buffer = rep.c->dnscrypt_buffer;
788 #else
789 			buffer = rep.c->buffer;
790 #endif
791 			(void)comm_point_send_udp_msg(rep.c, buffer,
792 				(struct sockaddr*)&rep.addr, rep.addrlen, 0);
793 		}
794 		if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for
795 		another UDP port. Note rep.c cannot be reused with TCP fd. */
796 			break;
797 	}
798 }
799 
800 int adjusted_tcp_timeout(struct comm_point* c)
801 {
802 	if(c->tcp_timeout_msec < TCP_QUERY_TIMEOUT_MINIMUM)
803 		return TCP_QUERY_TIMEOUT_MINIMUM;
804 	return c->tcp_timeout_msec;
805 }
806 
807 /** Use a new tcp handler for new query fd, set to read query */
808 static void
809 setup_tcp_handler(struct comm_point* c, int fd, int cur, int max)
810 {
811 	int handler_usage;
812 	log_assert(c->type == comm_tcp || c->type == comm_http);
813 	log_assert(c->fd == -1);
814 	sldns_buffer_clear(c->buffer);
815 #ifdef USE_DNSCRYPT
816 	if (c->dnscrypt)
817 		sldns_buffer_clear(c->dnscrypt_buffer);
818 #endif
819 	c->tcp_is_reading = 1;
820 	c->tcp_byte_count = 0;
821 	c->tcp_keepalive = 0;
822 	/* if more than half the tcp handlers are in use, use a shorter
823 	 * timeout for this TCP connection, we need to make space for
824 	 * other connections to be able to get attention */
825 	/* If > 50% TCP handler structures in use, set timeout to 1/100th
826 	 * 	configured value.
827 	 * If > 65%TCP handler structures in use, set to 1/500th configured
828 	 * 	value.
829 	 * If > 80% TCP handler structures in use, set to 0.
830 	 *
831 	 * If the timeout to use falls below 200 milliseconds, an actual
832 	 * timeout of 200ms is used.
833 	 */
834 	handler_usage = (cur * 100) / max;
835 	if(handler_usage > 50 && handler_usage <= 65)
836 		c->tcp_timeout_msec /= 100;
837 	else if (handler_usage > 65 && handler_usage <= 80)
838 		c->tcp_timeout_msec /= 500;
839 	else if (handler_usage > 80)
840 		c->tcp_timeout_msec = 0;
841 	comm_point_start_listening(c, fd, adjusted_tcp_timeout(c));
842 }
843 
844 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
845 	short ATTR_UNUSED(event), void* arg)
846 {
847 	struct comm_base* b = (struct comm_base*)arg;
848 	/* timeout for the slow accept, re-enable accepts again */
849 	if(b->start_accept) {
850 		verbose(VERB_ALGO, "wait is over, slow accept disabled");
851 		fptr_ok(fptr_whitelist_start_accept(b->start_accept));
852 		(*b->start_accept)(b->cb_arg);
853 		b->eb->slow_accept_enabled = 0;
854 	}
855 }
856 
857 int comm_point_perform_accept(struct comm_point* c,
858 	struct sockaddr_storage* addr, socklen_t* addrlen)
859 {
860 	int new_fd;
861 	*addrlen = (socklen_t)sizeof(*addr);
862 #ifndef HAVE_ACCEPT4
863 	new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
864 #else
865 	/* SOCK_NONBLOCK saves extra calls to fcntl for the same result */
866 	new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK);
867 #endif
868 	if(new_fd == -1) {
869 #ifndef USE_WINSOCK
870 		/* EINTR is signal interrupt. others are closed connection. */
871 		if(	errno == EINTR || errno == EAGAIN
872 #ifdef EWOULDBLOCK
873 			|| errno == EWOULDBLOCK
874 #endif
875 #ifdef ECONNABORTED
876 			|| errno == ECONNABORTED
877 #endif
878 #ifdef EPROTO
879 			|| errno == EPROTO
880 #endif /* EPROTO */
881 			)
882 			return -1;
883 #if defined(ENFILE) && defined(EMFILE)
884 		if(errno == ENFILE || errno == EMFILE) {
885 			/* out of file descriptors, likely outside of our
886 			 * control. stop accept() calls for some time */
887 			if(c->ev->base->stop_accept) {
888 				struct comm_base* b = c->ev->base;
889 				struct timeval tv;
890 				verbose(VERB_ALGO, "out of file descriptors: "
891 					"slow accept");
892 				b->eb->slow_accept_enabled = 1;
893 				fptr_ok(fptr_whitelist_stop_accept(
894 					b->stop_accept));
895 				(*b->stop_accept)(b->cb_arg);
896 				/* set timeout, no mallocs */
897 				tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
898 				tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000;
899 				b->eb->slow_accept = ub_event_new(b->eb->base,
900 					-1, UB_EV_TIMEOUT,
901 					comm_base_handle_slow_accept, b);
902 				if(b->eb->slow_accept == NULL) {
903 					/* we do not want to log here, because
904 					 * that would spam the logfiles.
905 					 * error: "event_base_set failed." */
906 				}
907 				else if(ub_event_add(b->eb->slow_accept, &tv)
908 					!= 0) {
909 					/* we do not want to log here,
910 					 * error: "event_add failed." */
911 				}
912 			}
913 			return -1;
914 		}
915 #endif
916 #else /* USE_WINSOCK */
917 		if(WSAGetLastError() == WSAEINPROGRESS ||
918 			WSAGetLastError() == WSAECONNRESET)
919 			return -1;
920 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
921 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
922 			return -1;
923 		}
924 #endif
925 		log_err_addr("accept failed", sock_strerror(errno), addr,
926 			*addrlen);
927 		return -1;
928 	}
929 	if(c->tcp_conn_limit && c->type == comm_tcp_accept) {
930 		c->tcl_addr = tcl_addr_lookup(c->tcp_conn_limit, addr, *addrlen);
931 		if(!tcl_new_connection(c->tcl_addr)) {
932 			if(verbosity >= 3)
933 				log_err_addr("accept rejected",
934 				"connection limit exceeded", addr, *addrlen);
935 			close(new_fd);
936 			return -1;
937 		}
938 	}
939 #ifndef HAVE_ACCEPT4
940 	fd_set_nonblock(new_fd);
941 #endif
942 	return new_fd;
943 }
944 
945 #ifdef USE_WINSOCK
946 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
947 #ifdef HAVE_BIO_SET_CALLBACK_EX
948 	size_t ATTR_UNUSED(len),
949 #endif
950         int ATTR_UNUSED(argi), long argl,
951 #ifndef HAVE_BIO_SET_CALLBACK_EX
952 	long retvalue
953 #else
954 	int retvalue, size_t* ATTR_UNUSED(processed)
955 #endif
956 	)
957 {
958 	int wsa_err = WSAGetLastError(); /* store errcode before it is gone */
959 	verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
960 		(oper&BIO_CB_RETURN)?"return":"before",
961 		(oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
962 		wsa_err==WSAEWOULDBLOCK?"wsawb":"");
963 	/* on windows, check if previous operation caused EWOULDBLOCK */
964 	if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
965 		(oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
966 		if(wsa_err == WSAEWOULDBLOCK)
967 			ub_winsock_tcp_wouldblock((struct ub_event*)
968 				BIO_get_callback_arg(b), UB_EV_READ);
969 	}
970 	if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
971 		(oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
972 		if(wsa_err == WSAEWOULDBLOCK)
973 			ub_winsock_tcp_wouldblock((struct ub_event*)
974 				BIO_get_callback_arg(b), UB_EV_WRITE);
975 	}
976 	/* return original return value */
977 	return retvalue;
978 }
979 
980 /** set win bio callbacks for nonblocking operations */
981 void
982 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
983 {
984 	SSL* ssl = (SSL*)thessl;
985 	/* set them both just in case, but usually they are the same BIO */
986 #ifdef HAVE_BIO_SET_CALLBACK_EX
987 	BIO_set_callback_ex(SSL_get_rbio(ssl), &win_bio_cb);
988 #else
989 	BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
990 #endif
991 	BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev);
992 #ifdef HAVE_BIO_SET_CALLBACK_EX
993 	BIO_set_callback_ex(SSL_get_wbio(ssl), &win_bio_cb);
994 #else
995 	BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
996 #endif
997 	BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev);
998 }
999 #endif
1000 
1001 #ifdef HAVE_NGHTTP2
1002 /** Create http2 session server.  Per connection, after TCP accepted.*/
1003 static int http2_session_server_create(struct http2_session* h2_session)
1004 {
1005 	log_assert(h2_session->callbacks);
1006 	h2_session->is_drop = 0;
1007 	if(nghttp2_session_server_new(&h2_session->session,
1008 			h2_session->callbacks,
1009 		h2_session) == NGHTTP2_ERR_NOMEM) {
1010 		log_err("failed to create nghttp2 session server");
1011 		return 0;
1012 	}
1013 
1014 	return 1;
1015 }
1016 
1017 /** Submit http2 setting to session. Once per session. */
1018 static int http2_submit_settings(struct http2_session* h2_session)
1019 {
1020 	int ret;
1021 	nghttp2_settings_entry settings[1] = {
1022 		{NGHTTP2_SETTINGS_MAX_CONCURRENT_STREAMS,
1023 		 h2_session->c->http2_max_streams}};
1024 
1025 	ret = nghttp2_submit_settings(h2_session->session, NGHTTP2_FLAG_NONE,
1026 		settings, 1);
1027 	if(ret) {
1028 		verbose(VERB_QUERY, "http2: submit_settings failed, "
1029 			"error: %s", nghttp2_strerror(ret));
1030 		return 0;
1031 	}
1032 	return 1;
1033 }
1034 #endif /* HAVE_NGHTTP2 */
1035 
1036 
1037 void
1038 comm_point_tcp_accept_callback(int fd, short event, void* arg)
1039 {
1040 	struct comm_point* c = (struct comm_point*)arg, *c_hdl;
1041 	int new_fd;
1042 	log_assert(c->type == comm_tcp_accept);
1043 	if(!(event & UB_EV_READ)) {
1044 		log_info("ignoring tcp accept event %d", (int)event);
1045 		return;
1046 	}
1047 	ub_comm_base_now(c->ev->base);
1048 	/* find free tcp handler. */
1049 	if(!c->tcp_free) {
1050 		log_warn("accepted too many tcp, connections full");
1051 		return;
1052 	}
1053 	/* accept incoming connection. */
1054 	c_hdl = c->tcp_free;
1055 	/* clear leftover flags from previous use, and then set the
1056 	 * correct event base for the event structure for libevent */
1057 	ub_event_free(c_hdl->ev->ev);
1058 	c_hdl->ev->ev = NULL;
1059 	if((c_hdl->type == comm_tcp && c_hdl->tcp_req_info) ||
1060 		c_hdl->type == comm_local || c_hdl->type == comm_raw)
1061 		c_hdl->tcp_do_toggle_rw = 0;
1062 	else	c_hdl->tcp_do_toggle_rw = 1;
1063 
1064 	if(c_hdl->type == comm_http) {
1065 #ifdef HAVE_NGHTTP2
1066 		if(!c_hdl->h2_session ||
1067 			!http2_session_server_create(c_hdl->h2_session)) {
1068 			log_warn("failed to create nghttp2");
1069 			return;
1070 		}
1071 		if(!c_hdl->h2_session ||
1072 			!http2_submit_settings(c_hdl->h2_session)) {
1073 			log_warn("failed to submit http2 settings");
1074 			return;
1075 		}
1076 		if(!c->ssl) {
1077 			c_hdl->tcp_do_toggle_rw = 0;
1078 			c_hdl->use_h2 = 1;
1079 		}
1080 #endif
1081 		c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
1082 			UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
1083 			comm_point_http_handle_callback, c_hdl);
1084 	} else {
1085 		c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
1086 			UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
1087 			comm_point_tcp_handle_callback, c_hdl);
1088 	}
1089 	if(!c_hdl->ev->ev) {
1090 		log_warn("could not ub_event_new, dropped tcp");
1091 		return;
1092 	}
1093 	log_assert(fd != -1);
1094 	(void)fd;
1095 	new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr,
1096 		&c_hdl->repinfo.addrlen);
1097 	if(new_fd == -1)
1098 		return;
1099 	if(c->ssl) {
1100 		c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
1101 		if(!c_hdl->ssl) {
1102 			c_hdl->fd = new_fd;
1103 			comm_point_close(c_hdl);
1104 			return;
1105 		}
1106 		c_hdl->ssl_shake_state = comm_ssl_shake_read;
1107 #ifdef USE_WINSOCK
1108 		comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
1109 #endif
1110 	}
1111 
1112 	/* grab the tcp handler buffers */
1113 	c->cur_tcp_count++;
1114 	c->tcp_free = c_hdl->tcp_free;
1115 	c_hdl->tcp_free = NULL;
1116 	if(!c->tcp_free) {
1117 		/* stop accepting incoming queries for now. */
1118 		comm_point_stop_listening(c);
1119 	}
1120 	setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count);
1121 }
1122 
1123 /** Make tcp handler free for next assignment */
1124 static void
1125 reclaim_tcp_handler(struct comm_point* c)
1126 {
1127 	log_assert(c->type == comm_tcp);
1128 	if(c->ssl) {
1129 #ifdef HAVE_SSL
1130 		SSL_shutdown(c->ssl);
1131 		SSL_free(c->ssl);
1132 		c->ssl = NULL;
1133 #endif
1134 	}
1135 	comm_point_close(c);
1136 	if(c->tcp_parent) {
1137 		if(c != c->tcp_parent->tcp_free) {
1138 			c->tcp_parent->cur_tcp_count--;
1139 			c->tcp_free = c->tcp_parent->tcp_free;
1140 			c->tcp_parent->tcp_free = c;
1141 		}
1142 		if(!c->tcp_free) {
1143 			/* re-enable listening on accept socket */
1144 			comm_point_start_listening(c->tcp_parent, -1, -1);
1145 		}
1146 	}
1147 	c->tcp_more_read_again = NULL;
1148 	c->tcp_more_write_again = NULL;
1149 	c->tcp_byte_count = 0;
1150 	sldns_buffer_clear(c->buffer);
1151 }
1152 
1153 /** do the callback when writing is done */
1154 static void
1155 tcp_callback_writer(struct comm_point* c)
1156 {
1157 	log_assert(c->type == comm_tcp);
1158 	if(!c->tcp_write_and_read) {
1159 		sldns_buffer_clear(c->buffer);
1160 		c->tcp_byte_count = 0;
1161 	}
1162 	if(c->tcp_do_toggle_rw)
1163 		c->tcp_is_reading = 1;
1164 	/* switch from listening(write) to listening(read) */
1165 	if(c->tcp_req_info) {
1166 		tcp_req_info_handle_writedone(c->tcp_req_info);
1167 	} else {
1168 		comm_point_stop_listening(c);
1169 		if(c->tcp_write_and_read) {
1170 			fptr_ok(fptr_whitelist_comm_point(c->callback));
1171 			if( (*c->callback)(c, c->cb_arg, NETEVENT_PKT_WRITTEN,
1172 				&c->repinfo) ) {
1173 				comm_point_start_listening(c, -1,
1174 					adjusted_tcp_timeout(c));
1175 			}
1176 		} else {
1177 			comm_point_start_listening(c, -1,
1178 					adjusted_tcp_timeout(c));
1179 		}
1180 	}
1181 }
1182 
1183 /** do the callback when reading is done */
1184 static void
1185 tcp_callback_reader(struct comm_point* c)
1186 {
1187 	log_assert(c->type == comm_tcp || c->type == comm_local);
1188 	sldns_buffer_flip(c->buffer);
1189 	if(c->tcp_do_toggle_rw)
1190 		c->tcp_is_reading = 0;
1191 	c->tcp_byte_count = 0;
1192 	if(c->tcp_req_info) {
1193 		tcp_req_info_handle_readdone(c->tcp_req_info);
1194 	} else {
1195 		if(c->type == comm_tcp)
1196 			comm_point_stop_listening(c);
1197 		fptr_ok(fptr_whitelist_comm_point(c->callback));
1198 		if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
1199 			comm_point_start_listening(c, -1,
1200 					adjusted_tcp_timeout(c));
1201 		}
1202 	}
1203 }
1204 
1205 #ifdef HAVE_SSL
1206 /** true if the ssl handshake error has to be squelched from the logs */
1207 int
1208 squelch_err_ssl_handshake(unsigned long err)
1209 {
1210 	if(verbosity >= VERB_QUERY)
1211 		return 0; /* only squelch on low verbosity */
1212 	if(ERR_GET_LIB(err) == ERR_LIB_SSL &&
1213 		(ERR_GET_REASON(err) == SSL_R_HTTPS_PROXY_REQUEST ||
1214 		 ERR_GET_REASON(err) == SSL_R_HTTP_REQUEST ||
1215 		 ERR_GET_REASON(err) == SSL_R_WRONG_VERSION_NUMBER ||
1216 		 ERR_GET_REASON(err) == SSL_R_SSLV3_ALERT_BAD_CERTIFICATE
1217 #ifdef SSL_F_TLS_POST_PROCESS_CLIENT_HELLO
1218 		 || ERR_GET_REASON(err) == SSL_R_NO_SHARED_CIPHER
1219 #endif
1220 #ifdef SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO
1221 		 || ERR_GET_REASON(err) == SSL_R_UNKNOWN_PROTOCOL
1222 		 || ERR_GET_REASON(err) == SSL_R_UNSUPPORTED_PROTOCOL
1223 #  ifdef SSL_R_VERSION_TOO_LOW
1224 		 || ERR_GET_REASON(err) == SSL_R_VERSION_TOO_LOW
1225 #  endif
1226 #endif
1227 		))
1228 		return 1;
1229 	return 0;
1230 }
1231 #endif /* HAVE_SSL */
1232 
1233 /** continue ssl handshake */
1234 #ifdef HAVE_SSL
1235 static int
1236 ssl_handshake(struct comm_point* c)
1237 {
1238 	int r;
1239 	if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
1240 		/* read condition satisfied back to writing */
1241 		comm_point_listen_for_rw(c, 0, 1);
1242 		c->ssl_shake_state = comm_ssl_shake_none;
1243 		return 1;
1244 	}
1245 	if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
1246 		/* write condition satisfied, back to reading */
1247 		comm_point_listen_for_rw(c, 1, 0);
1248 		c->ssl_shake_state = comm_ssl_shake_none;
1249 		return 1;
1250 	}
1251 
1252 	ERR_clear_error();
1253 	r = SSL_do_handshake(c->ssl);
1254 	if(r != 1) {
1255 		int want = SSL_get_error(c->ssl, r);
1256 		if(want == SSL_ERROR_WANT_READ) {
1257 			if(c->ssl_shake_state == comm_ssl_shake_read)
1258 				return 1;
1259 			c->ssl_shake_state = comm_ssl_shake_read;
1260 			comm_point_listen_for_rw(c, 1, 0);
1261 			return 1;
1262 		} else if(want == SSL_ERROR_WANT_WRITE) {
1263 			if(c->ssl_shake_state == comm_ssl_shake_write)
1264 				return 1;
1265 			c->ssl_shake_state = comm_ssl_shake_write;
1266 			comm_point_listen_for_rw(c, 0, 1);
1267 			return 1;
1268 		} else if(r == 0) {
1269 			return 0; /* closed */
1270 		} else if(want == SSL_ERROR_SYSCALL) {
1271 			/* SYSCALL and errno==0 means closed uncleanly */
1272 #ifdef EPIPE
1273 			if(errno == EPIPE && verbosity < 2)
1274 				return 0; /* silence 'broken pipe' */
1275 #endif
1276 #ifdef ECONNRESET
1277 			if(errno == ECONNRESET && verbosity < 2)
1278 				return 0; /* silence reset by peer */
1279 #endif
1280 			if(!tcp_connect_errno_needs_log(
1281 				(struct sockaddr*)&c->repinfo.addr,
1282 				c->repinfo.addrlen))
1283 				return 0; /* silence connect failures that
1284 				show up because after connect this is the
1285 				first system call that accesses the socket */
1286 			if(errno != 0)
1287 				log_err("SSL_handshake syscall: %s",
1288 					strerror(errno));
1289 			return 0;
1290 		} else {
1291 			unsigned long err = ERR_get_error();
1292 			if(!squelch_err_ssl_handshake(err)) {
1293 				log_crypto_err_code("ssl handshake failed", err);
1294 				log_addr(VERB_OPS, "ssl handshake failed", &c->repinfo.addr,
1295 					c->repinfo.addrlen);
1296 			}
1297 			return 0;
1298 		}
1299 	}
1300 	/* this is where peer verification could take place */
1301 	if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) {
1302 		/* verification */
1303 		if(SSL_get_verify_result(c->ssl) == X509_V_OK) {
1304 #ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
1305 			X509* x = SSL_get1_peer_certificate(c->ssl);
1306 #else
1307 			X509* x = SSL_get_peer_certificate(c->ssl);
1308 #endif
1309 			if(!x) {
1310 				log_addr(VERB_ALGO, "SSL connection failed: "
1311 					"no certificate",
1312 					&c->repinfo.addr, c->repinfo.addrlen);
1313 				return 0;
1314 			}
1315 			log_cert(VERB_ALGO, "peer certificate", x);
1316 #ifdef HAVE_SSL_GET0_PEERNAME
1317 			if(SSL_get0_peername(c->ssl)) {
1318 				char buf[255];
1319 				snprintf(buf, sizeof(buf), "SSL connection "
1320 					"to %s authenticated",
1321 					SSL_get0_peername(c->ssl));
1322 				log_addr(VERB_ALGO, buf, &c->repinfo.addr,
1323 					c->repinfo.addrlen);
1324 			} else {
1325 #endif
1326 				log_addr(VERB_ALGO, "SSL connection "
1327 					"authenticated", &c->repinfo.addr,
1328 					c->repinfo.addrlen);
1329 #ifdef HAVE_SSL_GET0_PEERNAME
1330 			}
1331 #endif
1332 			X509_free(x);
1333 		} else {
1334 #ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
1335 			X509* x = SSL_get1_peer_certificate(c->ssl);
1336 #else
1337 			X509* x = SSL_get_peer_certificate(c->ssl);
1338 #endif
1339 			if(x) {
1340 				log_cert(VERB_ALGO, "peer certificate", x);
1341 				X509_free(x);
1342 			}
1343 			log_addr(VERB_ALGO, "SSL connection failed: "
1344 				"failed to authenticate",
1345 				&c->repinfo.addr, c->repinfo.addrlen);
1346 			return 0;
1347 		}
1348 	} else {
1349 		/* unauthenticated, the verify peer flag was not set
1350 		 * in c->ssl when the ssl object was created from ssl_ctx */
1351 		log_addr(VERB_ALGO, "SSL connection", &c->repinfo.addr,
1352 			c->repinfo.addrlen);
1353 	}
1354 
1355 #ifdef HAVE_SSL_GET0_ALPN_SELECTED
1356 	/* check if http2 use is negotiated */
1357 	if(c->type == comm_http && c->h2_session) {
1358 		const unsigned char *alpn;
1359 		unsigned int alpnlen = 0;
1360 		SSL_get0_alpn_selected(c->ssl, &alpn, &alpnlen);
1361 		if(alpnlen == 2 && memcmp("h2", alpn, 2) == 0) {
1362 			/* connection upgraded to HTTP2 */
1363 			c->tcp_do_toggle_rw = 0;
1364 			c->use_h2 = 1;
1365 		}
1366 	}
1367 #endif
1368 
1369 	/* setup listen rw correctly */
1370 	if(c->tcp_is_reading) {
1371 		if(c->ssl_shake_state != comm_ssl_shake_read)
1372 			comm_point_listen_for_rw(c, 1, 0);
1373 	} else {
1374 		comm_point_listen_for_rw(c, 0, 1);
1375 	}
1376 	c->ssl_shake_state = comm_ssl_shake_none;
1377 	return 1;
1378 }
1379 #endif /* HAVE_SSL */
1380 
1381 /** ssl read callback on TCP */
1382 static int
1383 ssl_handle_read(struct comm_point* c)
1384 {
1385 #ifdef HAVE_SSL
1386 	int r;
1387 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1388 		if(!ssl_handshake(c))
1389 			return 0;
1390 		if(c->ssl_shake_state != comm_ssl_shake_none)
1391 			return 1;
1392 	}
1393 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1394 		/* read length bytes */
1395 		ERR_clear_error();
1396 		if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer,
1397 			c->tcp_byte_count), (int)(sizeof(uint16_t) -
1398 			c->tcp_byte_count))) <= 0) {
1399 			int want = SSL_get_error(c->ssl, r);
1400 			if(want == SSL_ERROR_ZERO_RETURN) {
1401 				if(c->tcp_req_info)
1402 					return tcp_req_info_handle_read_close(c->tcp_req_info);
1403 				return 0; /* shutdown, closed */
1404 			} else if(want == SSL_ERROR_WANT_READ) {
1405 #ifdef USE_WINSOCK
1406 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1407 #endif
1408 				return 1; /* read more later */
1409 			} else if(want == SSL_ERROR_WANT_WRITE) {
1410 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1411 				comm_point_listen_for_rw(c, 0, 1);
1412 				return 1;
1413 			} else if(want == SSL_ERROR_SYSCALL) {
1414 #ifdef ECONNRESET
1415 				if(errno == ECONNRESET && verbosity < 2)
1416 					return 0; /* silence reset by peer */
1417 #endif
1418 				if(errno != 0)
1419 					log_err("SSL_read syscall: %s",
1420 						strerror(errno));
1421 				return 0;
1422 			}
1423 			log_crypto_err("could not SSL_read");
1424 			return 0;
1425 		}
1426 		c->tcp_byte_count += r;
1427 		if(c->tcp_byte_count < sizeof(uint16_t))
1428 			return 1;
1429 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1430 			sldns_buffer_capacity(c->buffer)) {
1431 			verbose(VERB_QUERY, "ssl: dropped larger than buffer");
1432 			return 0;
1433 		}
1434 		sldns_buffer_set_limit(c->buffer,
1435 			sldns_buffer_read_u16_at(c->buffer, 0));
1436 		if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1437 			verbose(VERB_QUERY, "ssl: dropped bogus too short.");
1438 			return 0;
1439 		}
1440 		sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t)));
1441 		verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
1442 			(int)sldns_buffer_limit(c->buffer));
1443 	}
1444 	if(sldns_buffer_remaining(c->buffer) > 0) {
1445 		ERR_clear_error();
1446 		r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1447 			(int)sldns_buffer_remaining(c->buffer));
1448 		if(r <= 0) {
1449 			int want = SSL_get_error(c->ssl, r);
1450 			if(want == SSL_ERROR_ZERO_RETURN) {
1451 				if(c->tcp_req_info)
1452 					return tcp_req_info_handle_read_close(c->tcp_req_info);
1453 				return 0; /* shutdown, closed */
1454 			} else if(want == SSL_ERROR_WANT_READ) {
1455 #ifdef USE_WINSOCK
1456 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1457 #endif
1458 				return 1; /* read more later */
1459 			} else if(want == SSL_ERROR_WANT_WRITE) {
1460 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1461 				comm_point_listen_for_rw(c, 0, 1);
1462 				return 1;
1463 			} else if(want == SSL_ERROR_SYSCALL) {
1464 #ifdef ECONNRESET
1465 				if(errno == ECONNRESET && verbosity < 2)
1466 					return 0; /* silence reset by peer */
1467 #endif
1468 				if(errno != 0)
1469 					log_err("SSL_read syscall: %s",
1470 						strerror(errno));
1471 				return 0;
1472 			}
1473 			log_crypto_err("could not SSL_read");
1474 			return 0;
1475 		}
1476 		sldns_buffer_skip(c->buffer, (ssize_t)r);
1477 	}
1478 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1479 		tcp_callback_reader(c);
1480 	}
1481 	return 1;
1482 #else
1483 	(void)c;
1484 	return 0;
1485 #endif /* HAVE_SSL */
1486 }
1487 
1488 /** ssl write callback on TCP */
1489 static int
1490 ssl_handle_write(struct comm_point* c)
1491 {
1492 #ifdef HAVE_SSL
1493 	int r;
1494 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1495 		if(!ssl_handshake(c))
1496 			return 0;
1497 		if(c->ssl_shake_state != comm_ssl_shake_none)
1498 			return 1;
1499 	}
1500 	/* ignore return, if fails we may simply block */
1501 	(void)SSL_set_mode(c->ssl, (long)SSL_MODE_ENABLE_PARTIAL_WRITE);
1502 	if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
1503 		uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(c->buffer));
1504 		ERR_clear_error();
1505 		if(c->tcp_write_and_read) {
1506 			if(c->tcp_write_pkt_len + 2 < LDNS_RR_BUF_SIZE) {
1507 				/* combine the tcp length and the query for
1508 				 * write, this emulates writev */
1509 				uint8_t buf[LDNS_RR_BUF_SIZE];
1510 				memmove(buf, &len, sizeof(uint16_t));
1511 				memmove(buf+sizeof(uint16_t),
1512 					c->tcp_write_pkt,
1513 					c->tcp_write_pkt_len);
1514 				r = SSL_write(c->ssl,
1515 					(void*)(buf+c->tcp_write_byte_count),
1516 					c->tcp_write_pkt_len + 2 -
1517 					c->tcp_write_byte_count);
1518 			} else {
1519 				r = SSL_write(c->ssl,
1520 					(void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
1521 					(int)(sizeof(uint16_t)-c->tcp_write_byte_count));
1522 			}
1523 		} else if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) <
1524 			LDNS_RR_BUF_SIZE) {
1525 			/* combine the tcp length and the query for write,
1526 			 * this emulates writev */
1527 			uint8_t buf[LDNS_RR_BUF_SIZE];
1528 			memmove(buf, &len, sizeof(uint16_t));
1529 			memmove(buf+sizeof(uint16_t),
1530 				sldns_buffer_current(c->buffer),
1531 				sldns_buffer_remaining(c->buffer));
1532 			r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count),
1533 				(int)(sizeof(uint16_t)+
1534 				sldns_buffer_remaining(c->buffer)
1535 				- c->tcp_byte_count));
1536 		} else {
1537 			r = SSL_write(c->ssl,
1538 				(void*)(((uint8_t*)&len)+c->tcp_byte_count),
1539 				(int)(sizeof(uint16_t)-c->tcp_byte_count));
1540 		}
1541 		if(r <= 0) {
1542 			int want = SSL_get_error(c->ssl, r);
1543 			if(want == SSL_ERROR_ZERO_RETURN) {
1544 				return 0; /* closed */
1545 			} else if(want == SSL_ERROR_WANT_READ) {
1546 				c->ssl_shake_state = comm_ssl_shake_hs_read;
1547 				comm_point_listen_for_rw(c, 1, 0);
1548 				return 1; /* wait for read condition */
1549 			} else if(want == SSL_ERROR_WANT_WRITE) {
1550 #ifdef USE_WINSOCK
1551 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1552 #endif
1553 				return 1; /* write more later */
1554 			} else if(want == SSL_ERROR_SYSCALL) {
1555 #ifdef EPIPE
1556 				if(errno == EPIPE && verbosity < 2)
1557 					return 0; /* silence 'broken pipe' */
1558 #endif
1559 				if(errno != 0)
1560 					log_err("SSL_write syscall: %s",
1561 						strerror(errno));
1562 				return 0;
1563 			}
1564 			log_crypto_err("could not SSL_write");
1565 			return 0;
1566 		}
1567 		if(c->tcp_write_and_read) {
1568 			c->tcp_write_byte_count += r;
1569 			if(c->tcp_write_byte_count < sizeof(uint16_t))
1570 				return 1;
1571 		} else {
1572 			c->tcp_byte_count += r;
1573 			if(c->tcp_byte_count < sizeof(uint16_t))
1574 				return 1;
1575 			sldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1576 				sizeof(uint16_t));
1577 		}
1578 		if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1579 			tcp_callback_writer(c);
1580 			return 1;
1581 		}
1582 	}
1583 	log_assert(c->tcp_write_and_read || sldns_buffer_remaining(c->buffer) > 0);
1584 	log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
1585 	ERR_clear_error();
1586 	if(c->tcp_write_and_read) {
1587 		r = SSL_write(c->ssl, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
1588 			(int)(c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count));
1589 	} else {
1590 		r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
1591 			(int)sldns_buffer_remaining(c->buffer));
1592 	}
1593 	if(r <= 0) {
1594 		int want = SSL_get_error(c->ssl, r);
1595 		if(want == SSL_ERROR_ZERO_RETURN) {
1596 			return 0; /* closed */
1597 		} else if(want == SSL_ERROR_WANT_READ) {
1598 			c->ssl_shake_state = comm_ssl_shake_hs_read;
1599 			comm_point_listen_for_rw(c, 1, 0);
1600 			return 1; /* wait for read condition */
1601 		} else if(want == SSL_ERROR_WANT_WRITE) {
1602 #ifdef USE_WINSOCK
1603 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1604 #endif
1605 			return 1; /* write more later */
1606 		} else if(want == SSL_ERROR_SYSCALL) {
1607 #ifdef EPIPE
1608 			if(errno == EPIPE && verbosity < 2)
1609 				return 0; /* silence 'broken pipe' */
1610 #endif
1611 			if(errno != 0)
1612 				log_err("SSL_write syscall: %s",
1613 					strerror(errno));
1614 			return 0;
1615 		}
1616 		log_crypto_err("could not SSL_write");
1617 		return 0;
1618 	}
1619 	if(c->tcp_write_and_read) {
1620 		c->tcp_write_byte_count += r;
1621 	} else {
1622 		sldns_buffer_skip(c->buffer, (ssize_t)r);
1623 	}
1624 
1625 	if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1626 		tcp_callback_writer(c);
1627 	}
1628 	return 1;
1629 #else
1630 	(void)c;
1631 	return 0;
1632 #endif /* HAVE_SSL */
1633 }
1634 
1635 /** handle ssl tcp connection with dns contents */
1636 static int
1637 ssl_handle_it(struct comm_point* c, int is_write)
1638 {
1639 	/* handle case where renegotiation wants read during write call
1640 	 * or write during read calls */
1641 	if(is_write && c->ssl_shake_state == comm_ssl_shake_hs_write)
1642 		return ssl_handle_read(c);
1643 	else if(!is_write && c->ssl_shake_state == comm_ssl_shake_hs_read)
1644 		return ssl_handle_write(c);
1645 	/* handle read events for read operation and write events for a
1646 	 * write operation */
1647 	else if(!is_write)
1648 		return ssl_handle_read(c);
1649 	return ssl_handle_write(c);
1650 }
1651 
1652 /** Handle tcp reading callback.
1653  * @param fd: file descriptor of socket.
1654  * @param c: comm point to read from into buffer.
1655  * @param short_ok: if true, very short packets are OK (for comm_local).
1656  * @return: 0 on error
1657  */
1658 static int
1659 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
1660 {
1661 	ssize_t r;
1662 	log_assert(c->type == comm_tcp || c->type == comm_local);
1663 	if(c->ssl)
1664 		return ssl_handle_it(c, 0);
1665 	if(!c->tcp_is_reading && !c->tcp_write_and_read)
1666 		return 0;
1667 
1668 	log_assert(fd != -1);
1669 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1670 		/* read length bytes */
1671 		r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count),
1672 			sizeof(uint16_t)-c->tcp_byte_count, 0);
1673 		if(r == 0) {
1674 			if(c->tcp_req_info)
1675 				return tcp_req_info_handle_read_close(c->tcp_req_info);
1676 			return 0;
1677 		} else if(r == -1) {
1678 #ifndef USE_WINSOCK
1679 			if(errno == EINTR || errno == EAGAIN)
1680 				return 1;
1681 #ifdef ECONNRESET
1682 			if(errno == ECONNRESET && verbosity < 2)
1683 				return 0; /* silence reset by peer */
1684 #endif
1685 #ifdef ECONNREFUSED
1686 			if(errno == ECONNREFUSED && verbosity < 2)
1687 				return 0; /* silence reset by peer */
1688 #endif
1689 #ifdef ENETUNREACH
1690 			if(errno == ENETUNREACH && verbosity < 2)
1691 				return 0; /* silence it */
1692 #endif
1693 #ifdef EHOSTDOWN
1694 			if(errno == EHOSTDOWN && verbosity < 2)
1695 				return 0; /* silence it */
1696 #endif
1697 #ifdef EHOSTUNREACH
1698 			if(errno == EHOSTUNREACH && verbosity < 2)
1699 				return 0; /* silence it */
1700 #endif
1701 #ifdef ENETDOWN
1702 			if(errno == ENETDOWN && verbosity < 2)
1703 				return 0; /* silence it */
1704 #endif
1705 #ifdef EACCES
1706 			if(errno == EACCES && verbosity < 2)
1707 				return 0; /* silence it */
1708 #endif
1709 #ifdef ENOTCONN
1710 			if(errno == ENOTCONN) {
1711 				log_err_addr("read (in tcp s) failed and this could be because TCP Fast Open is enabled [--disable-tfo-client --disable-tfo-server] but does not work", sock_strerror(errno),
1712 					&c->repinfo.addr, c->repinfo.addrlen);
1713 				return 0;
1714 			}
1715 #endif
1716 #else /* USE_WINSOCK */
1717 			if(WSAGetLastError() == WSAECONNREFUSED && verbosity < 2)
1718 				return 0;
1719 			if(WSAGetLastError() == WSAEHOSTDOWN && verbosity < 2)
1720 				return 0;
1721 			if(WSAGetLastError() == WSAEHOSTUNREACH && verbosity < 2)
1722 				return 0;
1723 			if(WSAGetLastError() == WSAENETDOWN && verbosity < 2)
1724 				return 0;
1725 			if(WSAGetLastError() == WSAENETUNREACH && verbosity < 2)
1726 				return 0;
1727 			if(WSAGetLastError() == WSAECONNRESET)
1728 				return 0;
1729 			if(WSAGetLastError() == WSAEINPROGRESS)
1730 				return 1;
1731 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
1732 				ub_winsock_tcp_wouldblock(c->ev->ev,
1733 					UB_EV_READ);
1734 				return 1;
1735 			}
1736 #endif
1737 			log_err_addr("read (in tcp s)", sock_strerror(errno),
1738 				&c->repinfo.addr, c->repinfo.addrlen);
1739 			return 0;
1740 		}
1741 		c->tcp_byte_count += r;
1742 		if(c->tcp_byte_count != sizeof(uint16_t))
1743 			return 1;
1744 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1745 			sldns_buffer_capacity(c->buffer)) {
1746 			verbose(VERB_QUERY, "tcp: dropped larger than buffer");
1747 			return 0;
1748 		}
1749 		sldns_buffer_set_limit(c->buffer,
1750 			sldns_buffer_read_u16_at(c->buffer, 0));
1751 		if(!short_ok &&
1752 			sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1753 			verbose(VERB_QUERY, "tcp: dropped bogus too short.");
1754 			return 0;
1755 		}
1756 		verbose(VERB_ALGO, "Reading tcp query of length %d",
1757 			(int)sldns_buffer_limit(c->buffer));
1758 	}
1759 
1760 	if(sldns_buffer_remaining(c->buffer) == 0)
1761 		log_err("in comm_point_tcp_handle_read buffer_remaining is not > 0 as expected, continuing with (harmless) 0 length recv");
1762 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1763 		sldns_buffer_remaining(c->buffer), 0);
1764 	if(r == 0) {
1765 		if(c->tcp_req_info)
1766 			return tcp_req_info_handle_read_close(c->tcp_req_info);
1767 		return 0;
1768 	} else if(r == -1) {
1769 #ifndef USE_WINSOCK
1770 		if(errno == EINTR || errno == EAGAIN)
1771 			return 1;
1772 #else /* USE_WINSOCK */
1773 		if(WSAGetLastError() == WSAECONNRESET)
1774 			return 0;
1775 		if(WSAGetLastError() == WSAEINPROGRESS)
1776 			return 1;
1777 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1778 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1779 			return 1;
1780 		}
1781 #endif
1782 		log_err_addr("read (in tcp r)", sock_strerror(errno),
1783 			&c->repinfo.addr, c->repinfo.addrlen);
1784 		return 0;
1785 	}
1786 	sldns_buffer_skip(c->buffer, r);
1787 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1788 		tcp_callback_reader(c);
1789 	}
1790 	return 1;
1791 }
1792 
1793 /**
1794  * Handle tcp writing callback.
1795  * @param fd: file descriptor of socket.
1796  * @param c: comm point to write buffer out of.
1797  * @return: 0 on error
1798  */
1799 static int
1800 comm_point_tcp_handle_write(int fd, struct comm_point* c)
1801 {
1802 	ssize_t r;
1803 	struct sldns_buffer *buffer;
1804 	log_assert(c->type == comm_tcp);
1805 #ifdef USE_DNSCRYPT
1806 	buffer = c->dnscrypt_buffer;
1807 #else
1808 	buffer = c->buffer;
1809 #endif
1810 	if(c->tcp_is_reading && !c->ssl && !c->tcp_write_and_read)
1811 		return 0;
1812 	log_assert(fd != -1);
1813 	if(((!c->tcp_write_and_read && c->tcp_byte_count == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == 0)) && c->tcp_check_nb_connect) {
1814 		/* check for pending error from nonblocking connect */
1815 		/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
1816 		int error = 0;
1817 		socklen_t len = (socklen_t)sizeof(error);
1818 		if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
1819 			&len) < 0){
1820 #ifndef USE_WINSOCK
1821 			error = errno; /* on solaris errno is error */
1822 #else /* USE_WINSOCK */
1823 			error = WSAGetLastError();
1824 #endif
1825 		}
1826 #ifndef USE_WINSOCK
1827 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1828 		if(error == EINPROGRESS || error == EWOULDBLOCK)
1829 			return 1; /* try again later */
1830 		else
1831 #endif
1832 		if(error != 0 && verbosity < 2)
1833 			return 0; /* silence lots of chatter in the logs */
1834                 else if(error != 0) {
1835 			log_err_addr("tcp connect", strerror(error),
1836 				&c->repinfo.addr, c->repinfo.addrlen);
1837 #else /* USE_WINSOCK */
1838 		/* examine error */
1839 		if(error == WSAEINPROGRESS)
1840 			return 1;
1841 		else if(error == WSAEWOULDBLOCK) {
1842 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1843 			return 1;
1844 		} else if(error != 0 && verbosity < 2)
1845 			return 0;
1846 		else if(error != 0) {
1847 			log_err_addr("tcp connect", wsa_strerror(error),
1848 				&c->repinfo.addr, c->repinfo.addrlen);
1849 #endif /* USE_WINSOCK */
1850 			return 0;
1851 		}
1852 	}
1853 	if(c->ssl)
1854 		return ssl_handle_it(c, 1);
1855 
1856 #ifdef USE_MSG_FASTOPEN
1857 	/* Only try this on first use of a connection that uses tfo,
1858 	   otherwise fall through to normal write */
1859 	/* Also, TFO support on WINDOWS not implemented at the moment */
1860 	if(c->tcp_do_fastopen == 1) {
1861 		/* this form of sendmsg() does both a connect() and send() so need to
1862 		   look for various flavours of error*/
1863 		uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
1864 		struct msghdr msg;
1865 		struct iovec iov[2];
1866 		c->tcp_do_fastopen = 0;
1867 		memset(&msg, 0, sizeof(msg));
1868 		if(c->tcp_write_and_read) {
1869 			iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
1870 			iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
1871 			iov[1].iov_base = c->tcp_write_pkt;
1872 			iov[1].iov_len = c->tcp_write_pkt_len;
1873 		} else {
1874 			iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1875 			iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1876 			iov[1].iov_base = sldns_buffer_begin(buffer);
1877 			iov[1].iov_len = sldns_buffer_limit(buffer);
1878 		}
1879 		log_assert(iov[0].iov_len > 0);
1880 		msg.msg_name = &c->repinfo.addr;
1881 		msg.msg_namelen = c->repinfo.addrlen;
1882 		msg.msg_iov = iov;
1883 		msg.msg_iovlen = 2;
1884 		r = sendmsg(fd, &msg, MSG_FASTOPEN);
1885 		if (r == -1) {
1886 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1887 			/* Handshake is underway, maybe because no TFO cookie available.
1888 			   Come back to write the message*/
1889 			if(errno == EINPROGRESS || errno == EWOULDBLOCK)
1890 				return 1;
1891 #endif
1892 			if(errno == EINTR || errno == EAGAIN)
1893 				return 1;
1894 			/* Not handling EISCONN here as shouldn't ever hit that case.*/
1895 			if(errno != EPIPE
1896 #ifdef EOPNOTSUPP
1897 				/* if /proc/sys/net/ipv4/tcp_fastopen is
1898 				 * disabled on Linux, sendmsg may return
1899 				 * 'Operation not supported', if so
1900 				 * fallthrough to ordinary connect. */
1901 				&& errno != EOPNOTSUPP
1902 #endif
1903 				&& errno != 0) {
1904 				if(verbosity < 2)
1905 					return 0; /* silence lots of chatter in the logs */
1906 				log_err_addr("tcp sendmsg", strerror(errno),
1907 					&c->repinfo.addr, c->repinfo.addrlen);
1908 				return 0;
1909 			}
1910 			verbose(VERB_ALGO, "tcp sendmsg for fastopen failed (with %s), try normal connect", strerror(errno));
1911 			/* fallthrough to nonFASTOPEN
1912 			 * (MSG_FASTOPEN on Linux 3 produces EPIPE)
1913 			 * we need to perform connect() */
1914 			if(connect(fd, (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen) == -1) {
1915 #ifdef EINPROGRESS
1916 				if(errno == EINPROGRESS)
1917 					return 1; /* wait until connect done*/
1918 #endif
1919 #ifdef USE_WINSOCK
1920 				if(WSAGetLastError() == WSAEINPROGRESS ||
1921 					WSAGetLastError() == WSAEWOULDBLOCK)
1922 					return 1; /* wait until connect done*/
1923 #endif
1924 				if(tcp_connect_errno_needs_log(
1925 					(struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen)) {
1926 					log_err_addr("outgoing tcp: connect after EPIPE for fastopen",
1927 						strerror(errno), &c->repinfo.addr, c->repinfo.addrlen);
1928 				}
1929 				return 0;
1930 			}
1931 
1932 		} else {
1933 			if(c->tcp_write_and_read) {
1934 				c->tcp_write_byte_count += r;
1935 				if(c->tcp_write_byte_count < sizeof(uint16_t))
1936 					return 1;
1937 			} else {
1938 				c->tcp_byte_count += r;
1939 				if(c->tcp_byte_count < sizeof(uint16_t))
1940 					return 1;
1941 				sldns_buffer_set_position(buffer, c->tcp_byte_count -
1942 					sizeof(uint16_t));
1943 			}
1944 			if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1945 				tcp_callback_writer(c);
1946 				return 1;
1947 			}
1948 		}
1949 	}
1950 #endif /* USE_MSG_FASTOPEN */
1951 
1952 	if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
1953 		uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
1954 #ifdef HAVE_WRITEV
1955 		struct iovec iov[2];
1956 		if(c->tcp_write_and_read) {
1957 			iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
1958 			iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
1959 			iov[1].iov_base = c->tcp_write_pkt;
1960 			iov[1].iov_len = c->tcp_write_pkt_len;
1961 		} else {
1962 			iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1963 			iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1964 			iov[1].iov_base = sldns_buffer_begin(buffer);
1965 			iov[1].iov_len = sldns_buffer_limit(buffer);
1966 		}
1967 		log_assert(iov[0].iov_len > 0);
1968 		r = writev(fd, iov, 2);
1969 #else /* HAVE_WRITEV */
1970 		if(c->tcp_write_and_read) {
1971 			r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
1972 				sizeof(uint16_t)-c->tcp_write_byte_count, 0);
1973 		} else {
1974 			r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1975 				sizeof(uint16_t)-c->tcp_byte_count, 0);
1976 		}
1977 #endif /* HAVE_WRITEV */
1978 		if(r == -1) {
1979 #ifndef USE_WINSOCK
1980 #  ifdef EPIPE
1981                 	if(errno == EPIPE && verbosity < 2)
1982                         	return 0; /* silence 'broken pipe' */
1983   #endif
1984 			if(errno == EINTR || errno == EAGAIN)
1985 				return 1;
1986 #ifdef ECONNRESET
1987 			if(errno == ECONNRESET && verbosity < 2)
1988 				return 0; /* silence reset by peer */
1989 #endif
1990 #  ifdef HAVE_WRITEV
1991 			log_err_addr("tcp writev", strerror(errno),
1992 				&c->repinfo.addr, c->repinfo.addrlen);
1993 #  else /* HAVE_WRITEV */
1994 			log_err_addr("tcp send s", strerror(errno),
1995 				&c->repinfo.addr, c->repinfo.addrlen);
1996 #  endif /* HAVE_WRITEV */
1997 #else
1998 			if(WSAGetLastError() == WSAENOTCONN)
1999 				return 1;
2000 			if(WSAGetLastError() == WSAEINPROGRESS)
2001 				return 1;
2002 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
2003 				ub_winsock_tcp_wouldblock(c->ev->ev,
2004 					UB_EV_WRITE);
2005 				return 1;
2006 			}
2007 			if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
2008 				return 0; /* silence reset by peer */
2009 			log_err_addr("tcp send s",
2010 				wsa_strerror(WSAGetLastError()),
2011 				&c->repinfo.addr, c->repinfo.addrlen);
2012 #endif
2013 			return 0;
2014 		}
2015 		if(c->tcp_write_and_read) {
2016 			c->tcp_write_byte_count += r;
2017 			if(c->tcp_write_byte_count < sizeof(uint16_t))
2018 				return 1;
2019 		} else {
2020 			c->tcp_byte_count += r;
2021 			if(c->tcp_byte_count < sizeof(uint16_t))
2022 				return 1;
2023 			sldns_buffer_set_position(buffer, c->tcp_byte_count -
2024 				sizeof(uint16_t));
2025 		}
2026 		if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
2027 			tcp_callback_writer(c);
2028 			return 1;
2029 		}
2030 	}
2031 	log_assert(c->tcp_write_and_read || sldns_buffer_remaining(buffer) > 0);
2032 	log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
2033 	if(c->tcp_write_and_read) {
2034 		r = send(fd, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
2035 			c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count, 0);
2036 	} else {
2037 		r = send(fd, (void*)sldns_buffer_current(buffer),
2038 			sldns_buffer_remaining(buffer), 0);
2039 	}
2040 	if(r == -1) {
2041 #ifndef USE_WINSOCK
2042 		if(errno == EINTR || errno == EAGAIN)
2043 			return 1;
2044 #ifdef ECONNRESET
2045 		if(errno == ECONNRESET && verbosity < 2)
2046 			return 0; /* silence reset by peer */
2047 #endif
2048 #else
2049 		if(WSAGetLastError() == WSAEINPROGRESS)
2050 			return 1;
2051 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
2052 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2053 			return 1;
2054 		}
2055 		if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
2056 			return 0; /* silence reset by peer */
2057 #endif
2058 		log_err_addr("tcp send r", sock_strerror(errno),
2059 			&c->repinfo.addr, c->repinfo.addrlen);
2060 		return 0;
2061 	}
2062 	if(c->tcp_write_and_read) {
2063 		c->tcp_write_byte_count += r;
2064 	} else {
2065 		sldns_buffer_skip(buffer, r);
2066 	}
2067 
2068 	if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
2069 		tcp_callback_writer(c);
2070 	}
2071 
2072 	return 1;
2073 }
2074 
2075 /** read again to drain buffers when there could be more to read */
2076 static void
2077 tcp_req_info_read_again(int fd, struct comm_point* c)
2078 {
2079 	while(c->tcp_req_info->read_again) {
2080 		int r;
2081 		c->tcp_req_info->read_again = 0;
2082 		if(c->tcp_is_reading)
2083 			r = comm_point_tcp_handle_read(fd, c, 0);
2084 		else 	r = comm_point_tcp_handle_write(fd, c);
2085 		if(!r) {
2086 			reclaim_tcp_handler(c);
2087 			if(!c->tcp_do_close) {
2088 				fptr_ok(fptr_whitelist_comm_point(
2089 					c->callback));
2090 				(void)(*c->callback)(c, c->cb_arg,
2091 					NETEVENT_CLOSED, NULL);
2092 			}
2093 			return;
2094 		}
2095 	}
2096 }
2097 
2098 /** read again to drain buffers when there could be more to read */
2099 static void
2100 tcp_more_read_again(int fd, struct comm_point* c)
2101 {
2102 	/* if the packet is done, but another one could be waiting on
2103 	 * the connection, the callback signals this, and we try again */
2104 	/* this continues until the read routines get EAGAIN or so,
2105 	 * and thus does not call the callback, and the bool is 0 */
2106 	int* moreread = c->tcp_more_read_again;
2107 	while(moreread && *moreread) {
2108 		*moreread = 0;
2109 		if(!comm_point_tcp_handle_read(fd, c, 0)) {
2110 			reclaim_tcp_handler(c);
2111 			if(!c->tcp_do_close) {
2112 				fptr_ok(fptr_whitelist_comm_point(
2113 					c->callback));
2114 				(void)(*c->callback)(c, c->cb_arg,
2115 					NETEVENT_CLOSED, NULL);
2116 			}
2117 			return;
2118 		}
2119 	}
2120 }
2121 
2122 /** write again to fill up when there could be more to write */
2123 static void
2124 tcp_more_write_again(int fd, struct comm_point* c)
2125 {
2126 	/* if the packet is done, but another is waiting to be written,
2127 	 * the callback signals it and we try again. */
2128 	/* this continues until the write routines get EAGAIN or so,
2129 	 * and thus does not call the callback, and the bool is 0 */
2130 	int* morewrite = c->tcp_more_write_again;
2131 	while(morewrite && *morewrite) {
2132 		*morewrite = 0;
2133 		if(!comm_point_tcp_handle_write(fd, c)) {
2134 			reclaim_tcp_handler(c);
2135 			if(!c->tcp_do_close) {
2136 				fptr_ok(fptr_whitelist_comm_point(
2137 					c->callback));
2138 				(void)(*c->callback)(c, c->cb_arg,
2139 					NETEVENT_CLOSED, NULL);
2140 			}
2141 			return;
2142 		}
2143 	}
2144 }
2145 
2146 void
2147 comm_point_tcp_handle_callback(int fd, short event, void* arg)
2148 {
2149 	struct comm_point* c = (struct comm_point*)arg;
2150 	log_assert(c->type == comm_tcp);
2151 	ub_comm_base_now(c->ev->base);
2152 
2153 #ifdef USE_DNSCRYPT
2154 	/* Initialize if this is a dnscrypt socket */
2155 	if(c->tcp_parent) {
2156 		c->dnscrypt = c->tcp_parent->dnscrypt;
2157 	}
2158 	if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) {
2159 		c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer));
2160 		if(!c->dnscrypt_buffer) {
2161 			log_err("Could not allocate dnscrypt buffer");
2162 			reclaim_tcp_handler(c);
2163 			if(!c->tcp_do_close) {
2164 				fptr_ok(fptr_whitelist_comm_point(
2165 					c->callback));
2166 				(void)(*c->callback)(c, c->cb_arg,
2167 					NETEVENT_CLOSED, NULL);
2168 			}
2169 			return;
2170 		}
2171 	}
2172 #endif
2173 
2174 	if(event&UB_EV_TIMEOUT) {
2175 		verbose(VERB_QUERY, "tcp took too long, dropped");
2176 		reclaim_tcp_handler(c);
2177 		if(!c->tcp_do_close) {
2178 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2179 			(void)(*c->callback)(c, c->cb_arg,
2180 				NETEVENT_TIMEOUT, NULL);
2181 		}
2182 		return;
2183 	}
2184 	if(event&UB_EV_READ
2185 #ifdef USE_MSG_FASTOPEN
2186 		&& !(c->tcp_do_fastopen && (event&UB_EV_WRITE))
2187 #endif
2188 		) {
2189 		int has_tcpq = (c->tcp_req_info != NULL);
2190 		int* moreread = c->tcp_more_read_again;
2191 		if(!comm_point_tcp_handle_read(fd, c, 0)) {
2192 			reclaim_tcp_handler(c);
2193 			if(!c->tcp_do_close) {
2194 				fptr_ok(fptr_whitelist_comm_point(
2195 					c->callback));
2196 				(void)(*c->callback)(c, c->cb_arg,
2197 					NETEVENT_CLOSED, NULL);
2198 			}
2199 			return;
2200 		}
2201 		if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
2202 			tcp_req_info_read_again(fd, c);
2203 		if(moreread && *moreread)
2204 			tcp_more_read_again(fd, c);
2205 		return;
2206 	}
2207 	if(event&UB_EV_WRITE) {
2208 		int has_tcpq = (c->tcp_req_info != NULL);
2209 		int* morewrite = c->tcp_more_write_again;
2210 		if(!comm_point_tcp_handle_write(fd, c)) {
2211 			reclaim_tcp_handler(c);
2212 			if(!c->tcp_do_close) {
2213 				fptr_ok(fptr_whitelist_comm_point(
2214 					c->callback));
2215 				(void)(*c->callback)(c, c->cb_arg,
2216 					NETEVENT_CLOSED, NULL);
2217 			}
2218 			return;
2219 		}
2220 		if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
2221 			tcp_req_info_read_again(fd, c);
2222 		if(morewrite && *morewrite)
2223 			tcp_more_write_again(fd, c);
2224 		return;
2225 	}
2226 	log_err("Ignored event %d for tcphdl.", event);
2227 }
2228 
2229 /** Make http handler free for next assignment */
2230 static void
2231 reclaim_http_handler(struct comm_point* c)
2232 {
2233 	log_assert(c->type == comm_http);
2234 	if(c->ssl) {
2235 #ifdef HAVE_SSL
2236 		SSL_shutdown(c->ssl);
2237 		SSL_free(c->ssl);
2238 		c->ssl = NULL;
2239 #endif
2240 	}
2241 	comm_point_close(c);
2242 	if(c->tcp_parent) {
2243 		if(c != c->tcp_parent->tcp_free) {
2244 			c->tcp_parent->cur_tcp_count--;
2245 			c->tcp_free = c->tcp_parent->tcp_free;
2246 			c->tcp_parent->tcp_free = c;
2247 		}
2248 		if(!c->tcp_free) {
2249 			/* re-enable listening on accept socket */
2250 			comm_point_start_listening(c->tcp_parent, -1, -1);
2251 		}
2252 	}
2253 }
2254 
2255 /** read more data for http (with ssl) */
2256 static int
2257 ssl_http_read_more(struct comm_point* c)
2258 {
2259 #ifdef HAVE_SSL
2260 	int r;
2261 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2262 	ERR_clear_error();
2263 	r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
2264 		(int)sldns_buffer_remaining(c->buffer));
2265 	if(r <= 0) {
2266 		int want = SSL_get_error(c->ssl, r);
2267 		if(want == SSL_ERROR_ZERO_RETURN) {
2268 			return 0; /* shutdown, closed */
2269 		} else if(want == SSL_ERROR_WANT_READ) {
2270 			return 1; /* read more later */
2271 		} else if(want == SSL_ERROR_WANT_WRITE) {
2272 			c->ssl_shake_state = comm_ssl_shake_hs_write;
2273 			comm_point_listen_for_rw(c, 0, 1);
2274 			return 1;
2275 		} else if(want == SSL_ERROR_SYSCALL) {
2276 #ifdef ECONNRESET
2277 			if(errno == ECONNRESET && verbosity < 2)
2278 				return 0; /* silence reset by peer */
2279 #endif
2280 			if(errno != 0)
2281 				log_err("SSL_read syscall: %s",
2282 					strerror(errno));
2283 			return 0;
2284 		}
2285 		log_crypto_err("could not SSL_read");
2286 		return 0;
2287 	}
2288 	verbose(VERB_ALGO, "ssl http read more skip to %d + %d",
2289 		(int)sldns_buffer_position(c->buffer), (int)r);
2290 	sldns_buffer_skip(c->buffer, (ssize_t)r);
2291 	return 1;
2292 #else
2293 	(void)c;
2294 	return 0;
2295 #endif /* HAVE_SSL */
2296 }
2297 
2298 /** read more data for http */
2299 static int
2300 http_read_more(int fd, struct comm_point* c)
2301 {
2302 	ssize_t r;
2303 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2304 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
2305 		sldns_buffer_remaining(c->buffer), 0);
2306 	if(r == 0) {
2307 		return 0;
2308 	} else if(r == -1) {
2309 #ifndef USE_WINSOCK
2310 		if(errno == EINTR || errno == EAGAIN)
2311 			return 1;
2312 #else /* USE_WINSOCK */
2313 		if(WSAGetLastError() == WSAECONNRESET)
2314 			return 0;
2315 		if(WSAGetLastError() == WSAEINPROGRESS)
2316 			return 1;
2317 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
2318 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
2319 			return 1;
2320 		}
2321 #endif
2322 		log_err_addr("read (in http r)", sock_strerror(errno),
2323 			&c->repinfo.addr, c->repinfo.addrlen);
2324 		return 0;
2325 	}
2326 	verbose(VERB_ALGO, "http read more skip to %d + %d",
2327 		(int)sldns_buffer_position(c->buffer), (int)r);
2328 	sldns_buffer_skip(c->buffer, r);
2329 	return 1;
2330 }
2331 
2332 /** return true if http header has been read (one line complete) */
2333 static int
2334 http_header_done(sldns_buffer* buf)
2335 {
2336 	size_t i;
2337 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
2338 		/* there was a \r before the \n, but we ignore that */
2339 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n')
2340 			return 1;
2341 	}
2342 	return 0;
2343 }
2344 
2345 /** return character string into buffer for header line, moves buffer
2346  * past that line and puts zero terminator into linefeed-newline */
2347 static char*
2348 http_header_line(sldns_buffer* buf)
2349 {
2350 	char* result = (char*)sldns_buffer_current(buf);
2351 	size_t i;
2352 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
2353 		/* terminate the string on the \r */
2354 		if((char)sldns_buffer_read_u8_at(buf, i) == '\r')
2355 			sldns_buffer_write_u8_at(buf, i, 0);
2356 		/* terminate on the \n and skip past the it and done */
2357 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n') {
2358 			sldns_buffer_write_u8_at(buf, i, 0);
2359 			sldns_buffer_set_position(buf, i+1);
2360 			return result;
2361 		}
2362 	}
2363 	return NULL;
2364 }
2365 
2366 /** move unread buffer to start and clear rest for putting the rest into it */
2367 static void
2368 http_moveover_buffer(sldns_buffer* buf)
2369 {
2370 	size_t pos = sldns_buffer_position(buf);
2371 	size_t len = sldns_buffer_remaining(buf);
2372 	sldns_buffer_clear(buf);
2373 	memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len);
2374 	sldns_buffer_set_position(buf, len);
2375 }
2376 
2377 /** a http header is complete, process it */
2378 static int
2379 http_process_initial_header(struct comm_point* c)
2380 {
2381 	char* line = http_header_line(c->buffer);
2382 	if(!line) return 1;
2383 	verbose(VERB_ALGO, "http header: %s", line);
2384 	if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) {
2385 		/* check returncode */
2386 		if(line[9] != '2') {
2387 			verbose(VERB_ALGO, "http bad status %s", line+9);
2388 			return 0;
2389 		}
2390 	} else if(strncasecmp(line, "Content-Length: ", 16) == 0) {
2391 		if(!c->http_is_chunked)
2392 			c->tcp_byte_count = (size_t)atoi(line+16);
2393 	} else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) {
2394 		c->tcp_byte_count = 0;
2395 		c->http_is_chunked = 1;
2396 	} else if(line[0] == 0) {
2397 		/* end of initial headers */
2398 		c->http_in_headers = 0;
2399 		if(c->http_is_chunked)
2400 			c->http_in_chunk_headers = 1;
2401 		/* remove header text from front of buffer
2402 		 * the buffer is going to be used to return the data segment
2403 		 * itself and we don't want the header to get returned
2404 		 * prepended with it */
2405 		http_moveover_buffer(c->buffer);
2406 		sldns_buffer_flip(c->buffer);
2407 		return 1;
2408 	}
2409 	/* ignore other headers */
2410 	return 1;
2411 }
2412 
2413 /** a chunk header is complete, process it, return 0=fail, 1=continue next
2414  * header line, 2=done with chunked transfer*/
2415 static int
2416 http_process_chunk_header(struct comm_point* c)
2417 {
2418 	char* line = http_header_line(c->buffer);
2419 	if(!line) return 1;
2420 	if(c->http_in_chunk_headers == 3) {
2421 		verbose(VERB_ALGO, "http chunk trailer: %s", line);
2422 		/* are we done ? */
2423 		if(line[0] == 0 && c->tcp_byte_count == 0) {
2424 			/* callback of http reader when NETEVENT_DONE,
2425 			 * end of data, with no data in buffer */
2426 			sldns_buffer_set_position(c->buffer, 0);
2427 			sldns_buffer_set_limit(c->buffer, 0);
2428 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2429 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
2430 			/* return that we are done */
2431 			return 2;
2432 		}
2433 		if(line[0] == 0) {
2434 			/* continue with header of the next chunk */
2435 			c->http_in_chunk_headers = 1;
2436 			/* remove header text from front of buffer */
2437 			http_moveover_buffer(c->buffer);
2438 			sldns_buffer_flip(c->buffer);
2439 			return 1;
2440 		}
2441 		/* ignore further trail headers */
2442 		return 1;
2443 	}
2444 	verbose(VERB_ALGO, "http chunk header: %s", line);
2445 	if(c->http_in_chunk_headers == 1) {
2446 		/* read chunked start line */
2447 		char* end = NULL;
2448 		c->tcp_byte_count = (size_t)strtol(line, &end, 16);
2449 		if(end == line)
2450 			return 0;
2451 		c->http_in_chunk_headers = 0;
2452 		/* remove header text from front of buffer */
2453 		http_moveover_buffer(c->buffer);
2454 		sldns_buffer_flip(c->buffer);
2455 		if(c->tcp_byte_count == 0) {
2456 			/* done with chunks, process chunk_trailer lines */
2457 			c->http_in_chunk_headers = 3;
2458 		}
2459 		return 1;
2460 	}
2461 	/* ignore other headers */
2462 	return 1;
2463 }
2464 
2465 /** handle nonchunked data segment, 0=fail, 1=wait */
2466 static int
2467 http_nonchunk_segment(struct comm_point* c)
2468 {
2469 	/* c->buffer at position..limit has new data we read in.
2470 	 * the buffer itself is full of nonchunked data.
2471 	 * we are looking to read tcp_byte_count more data
2472 	 * and then the transfer is done. */
2473 	size_t remainbufferlen;
2474 	size_t got_now = sldns_buffer_limit(c->buffer);
2475 	if(c->tcp_byte_count <= got_now) {
2476 		/* done, this is the last data fragment */
2477 		c->http_stored = 0;
2478 		sldns_buffer_set_position(c->buffer, 0);
2479 		fptr_ok(fptr_whitelist_comm_point(c->callback));
2480 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
2481 		return 1;
2482 	}
2483 	/* if we have the buffer space,
2484 	 * read more data collected into the buffer */
2485 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
2486 		sldns_buffer_limit(c->buffer);
2487 	if(remainbufferlen+got_now >= c->tcp_byte_count ||
2488 		remainbufferlen >= (size_t)(c->ssl?16384:2048)) {
2489 		size_t total = sldns_buffer_limit(c->buffer);
2490 		sldns_buffer_clear(c->buffer);
2491 		sldns_buffer_set_position(c->buffer, total);
2492 		c->http_stored = total;
2493 		/* return and wait to read more */
2494 		return 1;
2495 	}
2496 	/* call callback with this data amount, then
2497 	 * wait for more */
2498 	c->tcp_byte_count -= got_now;
2499 	c->http_stored = 0;
2500 	sldns_buffer_set_position(c->buffer, 0);
2501 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2502 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2503 	/* c->callback has to buffer_clear(c->buffer). */
2504 	/* return and wait to read more */
2505 	return 1;
2506 }
2507 
2508 /** handle chunked data segment, return 0=fail, 1=wait, 2=process more */
2509 static int
2510 http_chunked_segment(struct comm_point* c)
2511 {
2512 	/* the c->buffer has from position..limit new data we read. */
2513 	/* the current chunk has length tcp_byte_count.
2514 	 * once we read that read more chunk headers.
2515 	 */
2516 	size_t remainbufferlen;
2517 	size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
2518 	verbose(VERB_ALGO, "http_chunked_segment: got now %d, tcpbytcount %d, http_stored %d, buffer pos %d, buffer limit %d", (int)got_now, (int)c->tcp_byte_count, (int)c->http_stored, (int)sldns_buffer_position(c->buffer), (int)sldns_buffer_limit(c->buffer));
2519 	if(c->tcp_byte_count <= got_now) {
2520 		/* the chunk has completed (with perhaps some extra data
2521 		 * from next chunk header and next chunk) */
2522 		/* save too much info into temp buffer */
2523 		size_t fraglen;
2524 		struct comm_reply repinfo;
2525 		c->http_stored = 0;
2526 		sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count);
2527 		sldns_buffer_clear(c->http_temp);
2528 		sldns_buffer_write(c->http_temp,
2529 			sldns_buffer_current(c->buffer),
2530 			sldns_buffer_remaining(c->buffer));
2531 		sldns_buffer_flip(c->http_temp);
2532 
2533 		/* callback with this fragment */
2534 		fraglen = sldns_buffer_position(c->buffer);
2535 		sldns_buffer_set_position(c->buffer, 0);
2536 		sldns_buffer_set_limit(c->buffer, fraglen);
2537 		repinfo = c->repinfo;
2538 		fptr_ok(fptr_whitelist_comm_point(c->callback));
2539 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo);
2540 		/* c->callback has to buffer_clear(). */
2541 
2542 		/* is commpoint deleted? */
2543 		if(!repinfo.c) {
2544 			return 1;
2545 		}
2546 		/* copy waiting info */
2547 		sldns_buffer_clear(c->buffer);
2548 		sldns_buffer_write(c->buffer,
2549 			sldns_buffer_begin(c->http_temp),
2550 			sldns_buffer_remaining(c->http_temp));
2551 		sldns_buffer_flip(c->buffer);
2552 		/* process end of chunk trailer header lines, until
2553 		 * an empty line */
2554 		c->http_in_chunk_headers = 3;
2555 		/* process more data in buffer (if any) */
2556 		return 2;
2557 	}
2558 	c->tcp_byte_count -= got_now;
2559 
2560 	/* if we have the buffer space,
2561 	 * read more data collected into the buffer */
2562 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
2563 		sldns_buffer_limit(c->buffer);
2564 	if(remainbufferlen >= c->tcp_byte_count ||
2565 		remainbufferlen >= 2048) {
2566 		size_t total = sldns_buffer_limit(c->buffer);
2567 		sldns_buffer_clear(c->buffer);
2568 		sldns_buffer_set_position(c->buffer, total);
2569 		c->http_stored = total;
2570 		/* return and wait to read more */
2571 		return 1;
2572 	}
2573 
2574 	/* callback of http reader for a new part of the data */
2575 	c->http_stored = 0;
2576 	sldns_buffer_set_position(c->buffer, 0);
2577 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2578 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2579 	/* c->callback has to buffer_clear(c->buffer). */
2580 	/* return and wait to read more */
2581 	return 1;
2582 }
2583 
2584 #ifdef HAVE_NGHTTP2
2585 /** Create new http2 session. Called when creating handling comm point. */
2586 static struct http2_session* http2_session_create(struct comm_point* c)
2587 {
2588 	struct http2_session* session = calloc(1, sizeof(*session));
2589 	if(!session) {
2590 		log_err("malloc failure while creating http2 session");
2591 		return NULL;
2592 	}
2593 	session->c = c;
2594 
2595 	return session;
2596 }
2597 #endif
2598 
2599 /** Delete http2 session. After closing connection or on error */
2600 static void http2_session_delete(struct http2_session* h2_session)
2601 {
2602 #ifdef HAVE_NGHTTP2
2603 	if(h2_session->callbacks)
2604 		nghttp2_session_callbacks_del(h2_session->callbacks);
2605 	free(h2_session);
2606 #else
2607 	(void)h2_session;
2608 #endif
2609 }
2610 
2611 #ifdef HAVE_NGHTTP2
2612 struct http2_stream* http2_stream_create(int32_t stream_id)
2613 {
2614 	struct http2_stream* h2_stream = calloc(1, sizeof(*h2_stream));
2615 	if(!h2_stream) {
2616 		log_err("malloc failure while creating http2 stream");
2617 		return NULL;
2618 	}
2619 	h2_stream->stream_id = stream_id;
2620 	return h2_stream;
2621 }
2622 
2623 /** Delete http2 stream. After session delete or stream close callback */
2624 static void http2_stream_delete(struct http2_session* h2_session,
2625 	struct http2_stream* h2_stream)
2626 {
2627 	if(h2_stream->mesh_state) {
2628 		mesh_state_remove_reply(h2_stream->mesh, h2_stream->mesh_state,
2629 			h2_session->c);
2630 		h2_stream->mesh_state = NULL;
2631 	}
2632 	http2_req_stream_clear(h2_stream);
2633 	free(h2_stream);
2634 }
2635 #endif
2636 
2637 void http2_stream_add_meshstate(struct http2_stream* h2_stream,
2638 	struct mesh_area* mesh, struct mesh_state* m)
2639 {
2640 	h2_stream->mesh = mesh;
2641 	h2_stream->mesh_state = m;
2642 }
2643 
2644 /** delete http2 session server. After closing connection. */
2645 static void http2_session_server_delete(struct http2_session* h2_session)
2646 {
2647 #ifdef HAVE_NGHTTP2
2648 	struct http2_stream* h2_stream, *next;
2649 	nghttp2_session_del(h2_session->session); /* NULL input is fine */
2650 	h2_session->session = NULL;
2651 	for(h2_stream = h2_session->first_stream; h2_stream;) {
2652 		next = h2_stream->next;
2653 		http2_stream_delete(h2_session, h2_stream);
2654 		h2_stream = next;
2655 	}
2656 	h2_session->first_stream = NULL;
2657 	h2_session->is_drop = 0;
2658 	h2_session->postpone_drop = 0;
2659 	h2_session->c->h2_stream = NULL;
2660 #endif
2661 	(void)h2_session;
2662 }
2663 
2664 #ifdef HAVE_NGHTTP2
2665 void http2_session_add_stream(struct http2_session* h2_session,
2666 	struct http2_stream* h2_stream)
2667 {
2668 	if(h2_session->first_stream)
2669 		h2_session->first_stream->prev = h2_stream;
2670 	h2_stream->next = h2_session->first_stream;
2671 	h2_session->first_stream = h2_stream;
2672 }
2673 
2674 /** remove stream from session linked list. After stream close callback or
2675  * closing connection */
2676 static void http2_session_remove_stream(struct http2_session* h2_session,
2677 	struct http2_stream* h2_stream)
2678 {
2679 	if(h2_stream->prev)
2680 		h2_stream->prev->next = h2_stream->next;
2681 	else
2682 		h2_session->first_stream = h2_stream->next;
2683 	if(h2_stream->next)
2684 		h2_stream->next->prev = h2_stream->prev;
2685 
2686 }
2687 
2688 int http2_stream_close_cb(nghttp2_session* ATTR_UNUSED(session),
2689 	int32_t stream_id, uint32_t ATTR_UNUSED(error_code), void* cb_arg)
2690 {
2691 	struct http2_stream* h2_stream;
2692 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
2693 	if(!(h2_stream = nghttp2_session_get_stream_user_data(
2694 		h2_session->session, stream_id))) {
2695 		return 0;
2696 	}
2697 	http2_session_remove_stream(h2_session, h2_stream);
2698 	http2_stream_delete(h2_session, h2_stream);
2699 	return 0;
2700 }
2701 
2702 ssize_t http2_recv_cb(nghttp2_session* ATTR_UNUSED(session), uint8_t* buf,
2703 	size_t len, int ATTR_UNUSED(flags), void* cb_arg)
2704 {
2705 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
2706 	ssize_t ret;
2707 
2708 	log_assert(h2_session->c->type == comm_http);
2709 	log_assert(h2_session->c->h2_session);
2710 
2711 #ifdef HAVE_SSL
2712 	if(h2_session->c->ssl) {
2713 		int r;
2714 		ERR_clear_error();
2715 		r = SSL_read(h2_session->c->ssl, buf, len);
2716 		if(r <= 0) {
2717 			int want = SSL_get_error(h2_session->c->ssl, r);
2718 			if(want == SSL_ERROR_ZERO_RETURN) {
2719 				return NGHTTP2_ERR_EOF;
2720 			} else if(want == SSL_ERROR_WANT_READ) {
2721 				return NGHTTP2_ERR_WOULDBLOCK;
2722 			} else if(want == SSL_ERROR_WANT_WRITE) {
2723 				h2_session->c->ssl_shake_state = comm_ssl_shake_hs_write;
2724 				comm_point_listen_for_rw(h2_session->c, 0, 1);
2725 				return NGHTTP2_ERR_WOULDBLOCK;
2726 			} else if(want == SSL_ERROR_SYSCALL) {
2727 #ifdef ECONNRESET
2728 				if(errno == ECONNRESET && verbosity < 2)
2729 					return NGHTTP2_ERR_CALLBACK_FAILURE;
2730 #endif
2731 				if(errno != 0)
2732 					log_err("SSL_read syscall: %s",
2733 						strerror(errno));
2734 				return NGHTTP2_ERR_CALLBACK_FAILURE;
2735 			}
2736 			log_crypto_err("could not SSL_read");
2737 			return NGHTTP2_ERR_CALLBACK_FAILURE;
2738 		}
2739 		return r;
2740 	}
2741 #endif /* HAVE_SSL */
2742 
2743 	ret = recv(h2_session->c->fd, buf, len, 0);
2744 	if(ret == 0) {
2745 		return NGHTTP2_ERR_EOF;
2746 	} else if(ret < 0) {
2747 #ifndef USE_WINSOCK
2748 		if(errno == EINTR || errno == EAGAIN)
2749 			return NGHTTP2_ERR_WOULDBLOCK;
2750 #ifdef ECONNRESET
2751 		if(errno == ECONNRESET && verbosity < 2)
2752 			return NGHTTP2_ERR_CALLBACK_FAILURE;
2753 #endif
2754 		log_err_addr("could not http2 recv: %s", strerror(errno),
2755 			&h2_session->c->repinfo.addr,
2756 			h2_session->c->repinfo.addrlen);
2757 #else /* USE_WINSOCK */
2758 		if(WSAGetLastError() == WSAECONNRESET)
2759 			return NGHTTP2_ERR_CALLBACK_FAILURE;
2760 		if(WSAGetLastError() == WSAEINPROGRESS)
2761 			return NGHTTP2_ERR_WOULDBLOCK;
2762 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
2763 			ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
2764 				UB_EV_READ);
2765 			return NGHTTP2_ERR_WOULDBLOCK;
2766 		}
2767 		log_err_addr("could not http2 recv: %s",
2768 			wsa_strerror(WSAGetLastError()),
2769 			&h2_session->c->repinfo.addr,
2770 			h2_session->c->repinfo.addrlen);
2771 #endif
2772 		return NGHTTP2_ERR_CALLBACK_FAILURE;
2773 	}
2774 	return ret;
2775 }
2776 #endif /* HAVE_NGHTTP2 */
2777 
2778 /** Handle http2 read */
2779 static int
2780 comm_point_http2_handle_read(int ATTR_UNUSED(fd), struct comm_point* c)
2781 {
2782 #ifdef HAVE_NGHTTP2
2783 	int ret;
2784 	log_assert(c->h2_session);
2785 
2786 	/* reading until recv cb returns NGHTTP2_ERR_WOULDBLOCK */
2787 	ret = nghttp2_session_recv(c->h2_session->session);
2788 	if(ret) {
2789 		if(ret != NGHTTP2_ERR_EOF &&
2790 			ret != NGHTTP2_ERR_CALLBACK_FAILURE) {
2791 			char a[256];
2792 			addr_to_str(&c->repinfo.addr, c->repinfo.addrlen,
2793 				a, sizeof(a));
2794 			verbose(VERB_QUERY, "http2: session_recv from %s failed, "
2795 				"error: %s", a, nghttp2_strerror(ret));
2796 		}
2797 		return 0;
2798 	}
2799 	if(nghttp2_session_want_write(c->h2_session->session)) {
2800 		c->tcp_is_reading = 0;
2801 		comm_point_stop_listening(c);
2802 		comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
2803 	} else if(!nghttp2_session_want_read(c->h2_session->session))
2804 		return 0; /* connection can be closed */
2805 	return 1;
2806 #else
2807 	(void)c;
2808 	return 0;
2809 #endif
2810 }
2811 
2812 /**
2813  * Handle http reading callback.
2814  * @param fd: file descriptor of socket.
2815  * @param c: comm point to read from into buffer.
2816  * @return: 0 on error
2817  */
2818 static int
2819 comm_point_http_handle_read(int fd, struct comm_point* c)
2820 {
2821 	log_assert(c->type == comm_http);
2822 	log_assert(fd != -1);
2823 
2824 	/* if we are in ssl handshake, handle SSL handshake */
2825 #ifdef HAVE_SSL
2826 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2827 		if(!ssl_handshake(c))
2828 			return 0;
2829 		if(c->ssl_shake_state != comm_ssl_shake_none)
2830 			return 1;
2831 	}
2832 #endif /* HAVE_SSL */
2833 
2834 	if(!c->tcp_is_reading)
2835 		return 1;
2836 
2837 	if(c->use_h2) {
2838 		return comm_point_http2_handle_read(fd, c);
2839 	}
2840 
2841 	/* http version is <= http/1.1 */
2842 
2843 	if(c->http_min_version >= http_version_2) {
2844 		/* HTTP/2 failed, not allowed to use lower version. */
2845 		return 0;
2846 	}
2847 
2848 	/* read more data */
2849 	if(c->ssl) {
2850 		if(!ssl_http_read_more(c))
2851 			return 0;
2852 	} else {
2853 		if(!http_read_more(fd, c))
2854 			return 0;
2855 	}
2856 
2857 	if(c->http_stored >= sldns_buffer_position(c->buffer)) {
2858 		/* read did not work but we wanted more data, there is
2859 		 * no bytes to process now. */
2860 		return 1;
2861 	}
2862 	sldns_buffer_flip(c->buffer);
2863 	/* if we are partway in a segment of data, position us at the point
2864 	 * where we left off previously */
2865 	if(c->http_stored < sldns_buffer_limit(c->buffer))
2866 		sldns_buffer_set_position(c->buffer, c->http_stored);
2867 	else	sldns_buffer_set_position(c->buffer, sldns_buffer_limit(c->buffer));
2868 
2869 	while(sldns_buffer_remaining(c->buffer) > 0) {
2870 		/* Handle HTTP/1.x data */
2871 		/* if we are reading headers, read more headers */
2872 		if(c->http_in_headers || c->http_in_chunk_headers) {
2873 			/* if header is done, process the header */
2874 			if(!http_header_done(c->buffer)) {
2875 				/* copy remaining data to front of buffer
2876 				 * and set rest for writing into it */
2877 				http_moveover_buffer(c->buffer);
2878 				/* return and wait to read more */
2879 				return 1;
2880 			}
2881 			if(!c->http_in_chunk_headers) {
2882 				/* process initial headers */
2883 				if(!http_process_initial_header(c))
2884 					return 0;
2885 			} else {
2886 				/* process chunk headers */
2887 				int r = http_process_chunk_header(c);
2888 				if(r == 0) return 0;
2889 				if(r == 2) return 1; /* done */
2890 				/* r == 1, continue */
2891 			}
2892 			/* see if we have more to process */
2893 			continue;
2894 		}
2895 
2896 		if(!c->http_is_chunked) {
2897 			/* if we are reading nonchunks, process that*/
2898 			return http_nonchunk_segment(c);
2899 		} else {
2900 			/* if we are reading chunks, read the chunk */
2901 			int r = http_chunked_segment(c);
2902 			if(r == 0) return 0;
2903 			if(r == 1) return 1;
2904 			continue;
2905 		}
2906 	}
2907 	/* broke out of the loop; could not process header instead need
2908 	 * to read more */
2909 	/* moveover any remaining data and read more data */
2910 	http_moveover_buffer(c->buffer);
2911 	/* return and wait to read more */
2912 	return 1;
2913 }
2914 
2915 /** check pending connect for http */
2916 static int
2917 http_check_connect(int fd, struct comm_point* c)
2918 {
2919 	/* check for pending error from nonblocking connect */
2920 	/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
2921 	int error = 0;
2922 	socklen_t len = (socklen_t)sizeof(error);
2923 	if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
2924 		&len) < 0){
2925 #ifndef USE_WINSOCK
2926 		error = errno; /* on solaris errno is error */
2927 #else /* USE_WINSOCK */
2928 		error = WSAGetLastError();
2929 #endif
2930 	}
2931 #ifndef USE_WINSOCK
2932 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
2933 	if(error == EINPROGRESS || error == EWOULDBLOCK)
2934 		return 1; /* try again later */
2935 	else
2936 #endif
2937 	if(error != 0 && verbosity < 2)
2938 		return 0; /* silence lots of chatter in the logs */
2939 	else if(error != 0) {
2940 		log_err_addr("http connect", strerror(error),
2941 			&c->repinfo.addr, c->repinfo.addrlen);
2942 #else /* USE_WINSOCK */
2943 	/* examine error */
2944 	if(error == WSAEINPROGRESS)
2945 		return 1;
2946 	else if(error == WSAEWOULDBLOCK) {
2947 		ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2948 		return 1;
2949 	} else if(error != 0 && verbosity < 2)
2950 		return 0;
2951 	else if(error != 0) {
2952 		log_err_addr("http connect", wsa_strerror(error),
2953 			&c->repinfo.addr, c->repinfo.addrlen);
2954 #endif /* USE_WINSOCK */
2955 		return 0;
2956 	}
2957 	/* keep on processing this socket */
2958 	return 2;
2959 }
2960 
2961 /** write more data for http (with ssl) */
2962 static int
2963 ssl_http_write_more(struct comm_point* c)
2964 {
2965 #ifdef HAVE_SSL
2966 	int r;
2967 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2968 	ERR_clear_error();
2969 	r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
2970 		(int)sldns_buffer_remaining(c->buffer));
2971 	if(r <= 0) {
2972 		int want = SSL_get_error(c->ssl, r);
2973 		if(want == SSL_ERROR_ZERO_RETURN) {
2974 			return 0; /* closed */
2975 		} else if(want == SSL_ERROR_WANT_READ) {
2976 			c->ssl_shake_state = comm_ssl_shake_hs_read;
2977 			comm_point_listen_for_rw(c, 1, 0);
2978 			return 1; /* wait for read condition */
2979 		} else if(want == SSL_ERROR_WANT_WRITE) {
2980 			return 1; /* write more later */
2981 		} else if(want == SSL_ERROR_SYSCALL) {
2982 #ifdef EPIPE
2983 			if(errno == EPIPE && verbosity < 2)
2984 				return 0; /* silence 'broken pipe' */
2985 #endif
2986 			if(errno != 0)
2987 				log_err("SSL_write syscall: %s",
2988 					strerror(errno));
2989 			return 0;
2990 		}
2991 		log_crypto_err("could not SSL_write");
2992 		return 0;
2993 	}
2994 	sldns_buffer_skip(c->buffer, (ssize_t)r);
2995 	return 1;
2996 #else
2997 	(void)c;
2998 	return 0;
2999 #endif /* HAVE_SSL */
3000 }
3001 
3002 /** write more data for http */
3003 static int
3004 http_write_more(int fd, struct comm_point* c)
3005 {
3006 	ssize_t r;
3007 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
3008 	r = send(fd, (void*)sldns_buffer_current(c->buffer),
3009 		sldns_buffer_remaining(c->buffer), 0);
3010 	if(r == -1) {
3011 #ifndef USE_WINSOCK
3012 		if(errno == EINTR || errno == EAGAIN)
3013 			return 1;
3014 #else
3015 		if(WSAGetLastError() == WSAEINPROGRESS)
3016 			return 1;
3017 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
3018 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
3019 			return 1;
3020 		}
3021 #endif
3022 		log_err_addr("http send r", sock_strerror(errno),
3023 			&c->repinfo.addr, c->repinfo.addrlen);
3024 		return 0;
3025 	}
3026 	sldns_buffer_skip(c->buffer, r);
3027 	return 1;
3028 }
3029 
3030 #ifdef HAVE_NGHTTP2
3031 ssize_t http2_send_cb(nghttp2_session* ATTR_UNUSED(session), const uint8_t* buf,
3032 	size_t len, int ATTR_UNUSED(flags), void* cb_arg)
3033 {
3034 	ssize_t ret;
3035 	struct http2_session* h2_session = (struct http2_session*)cb_arg;
3036 	log_assert(h2_session->c->type == comm_http);
3037 	log_assert(h2_session->c->h2_session);
3038 
3039 #ifdef HAVE_SSL
3040 	if(h2_session->c->ssl) {
3041 		int r;
3042 		ERR_clear_error();
3043 		r = SSL_write(h2_session->c->ssl, buf, len);
3044 		if(r <= 0) {
3045 			int want = SSL_get_error(h2_session->c->ssl, r);
3046 			if(want == SSL_ERROR_ZERO_RETURN) {
3047 				return NGHTTP2_ERR_CALLBACK_FAILURE;
3048 			} else if(want == SSL_ERROR_WANT_READ) {
3049 				h2_session->c->ssl_shake_state = comm_ssl_shake_hs_read;
3050 				comm_point_listen_for_rw(h2_session->c, 1, 0);
3051 				return NGHTTP2_ERR_WOULDBLOCK;
3052 			} else if(want == SSL_ERROR_WANT_WRITE) {
3053 				return NGHTTP2_ERR_WOULDBLOCK;
3054 			} else if(want == SSL_ERROR_SYSCALL) {
3055 #ifdef EPIPE
3056 				if(errno == EPIPE && verbosity < 2)
3057 					return NGHTTP2_ERR_CALLBACK_FAILURE;
3058 #endif
3059 				if(errno != 0)
3060 					log_err("SSL_write syscall: %s",
3061 						strerror(errno));
3062 				return NGHTTP2_ERR_CALLBACK_FAILURE;
3063 			}
3064 			log_crypto_err("could not SSL_write");
3065 			return NGHTTP2_ERR_CALLBACK_FAILURE;
3066 		}
3067 		return r;
3068 	}
3069 #endif /* HAVE_SSL */
3070 
3071 	ret = send(h2_session->c->fd, buf, len, 0);
3072 	if(ret == 0) {
3073 		return NGHTTP2_ERR_CALLBACK_FAILURE;
3074 	} else if(ret < 0) {
3075 #ifndef USE_WINSOCK
3076 		if(errno == EINTR || errno == EAGAIN)
3077 			return NGHTTP2_ERR_WOULDBLOCK;
3078 #ifdef EPIPE
3079 		if(errno == EPIPE && verbosity < 2)
3080 			return NGHTTP2_ERR_CALLBACK_FAILURE;
3081 #endif
3082 #ifdef ECONNRESET
3083 		if(errno == ECONNRESET && verbosity < 2)
3084 			return NGHTTP2_ERR_CALLBACK_FAILURE;
3085 #endif
3086 		log_err_addr("could not http2 write: %s", strerror(errno),
3087 			&h2_session->c->repinfo.addr,
3088 			h2_session->c->repinfo.addrlen);
3089 #else /* USE_WINSOCK */
3090 		if(WSAGetLastError() == WSAENOTCONN)
3091 			return NGHTTP2_ERR_WOULDBLOCK;
3092 		if(WSAGetLastError() == WSAEINPROGRESS)
3093 			return NGHTTP2_ERR_WOULDBLOCK;
3094 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
3095 			ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
3096 				UB_EV_WRITE);
3097 			return NGHTTP2_ERR_WOULDBLOCK;
3098 		}
3099 		if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
3100 			return NGHTTP2_ERR_CALLBACK_FAILURE;
3101 		log_err_addr("could not http2 write: %s",
3102 			wsa_strerror(WSAGetLastError()),
3103 			&h2_session->c->repinfo.addr,
3104 			h2_session->c->repinfo.addrlen);
3105 #endif
3106 		return NGHTTP2_ERR_CALLBACK_FAILURE;
3107 	}
3108 	return ret;
3109 }
3110 #endif /* HAVE_NGHTTP2 */
3111 
3112 /** Handle http2 writing */
3113 static int
3114 comm_point_http2_handle_write(int ATTR_UNUSED(fd), struct comm_point* c)
3115 {
3116 #ifdef HAVE_NGHTTP2
3117 	int ret;
3118 	log_assert(c->h2_session);
3119 
3120 	ret = nghttp2_session_send(c->h2_session->session);
3121 	if(ret) {
3122 		verbose(VERB_QUERY, "http2: session_send failed, "
3123 			"error: %s", nghttp2_strerror(ret));
3124 		return 0;
3125 	}
3126 
3127 	if(nghttp2_session_want_read(c->h2_session->session)) {
3128 		c->tcp_is_reading = 1;
3129 		comm_point_stop_listening(c);
3130 		comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
3131 	} else if(!nghttp2_session_want_write(c->h2_session->session))
3132 		return 0; /* connection can be closed */
3133 	return 1;
3134 #else
3135 	(void)c;
3136 	return 0;
3137 #endif
3138 }
3139 
3140 /**
3141  * Handle http writing callback.
3142  * @param fd: file descriptor of socket.
3143  * @param c: comm point to write buffer out of.
3144  * @return: 0 on error
3145  */
3146 static int
3147 comm_point_http_handle_write(int fd, struct comm_point* c)
3148 {
3149 	log_assert(c->type == comm_http);
3150 	log_assert(fd != -1);
3151 
3152 	/* check pending connect errors, if that fails, we wait for more,
3153 	 * or we can continue to write contents */
3154 	if(c->tcp_check_nb_connect) {
3155 		int r = http_check_connect(fd, c);
3156 		if(r == 0) return 0;
3157 		if(r == 1) return 1;
3158 		c->tcp_check_nb_connect = 0;
3159 	}
3160 	/* if we are in ssl handshake, handle SSL handshake */
3161 #ifdef HAVE_SSL
3162 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
3163 		if(!ssl_handshake(c))
3164 			return 0;
3165 		if(c->ssl_shake_state != comm_ssl_shake_none)
3166 			return 1;
3167 	}
3168 #endif /* HAVE_SSL */
3169 	if(c->tcp_is_reading)
3170 		return 1;
3171 
3172 	if(c->use_h2) {
3173 		return comm_point_http2_handle_write(fd, c);
3174 	}
3175 
3176 	/* http version is <= http/1.1 */
3177 
3178 	if(c->http_min_version >= http_version_2) {
3179 		/* HTTP/2 failed, not allowed to use lower version. */
3180 		return 0;
3181 	}
3182 
3183 	/* if we are writing, write more */
3184 	if(c->ssl) {
3185 		if(!ssl_http_write_more(c))
3186 			return 0;
3187 	} else {
3188 		if(!http_write_more(fd, c))
3189 			return 0;
3190 	}
3191 
3192 	/* we write a single buffer contents, that can contain
3193 	 * the http request, and then flip to read the results */
3194 	/* see if write is done */
3195 	if(sldns_buffer_remaining(c->buffer) == 0) {
3196 		sldns_buffer_clear(c->buffer);
3197 		if(c->tcp_do_toggle_rw)
3198 			c->tcp_is_reading = 1;
3199 		c->tcp_byte_count = 0;
3200 		/* switch from listening(write) to listening(read) */
3201 		comm_point_stop_listening(c);
3202 		comm_point_start_listening(c, -1, -1);
3203 	}
3204 	return 1;
3205 }
3206 
3207 void
3208 comm_point_http_handle_callback(int fd, short event, void* arg)
3209 {
3210 	struct comm_point* c = (struct comm_point*)arg;
3211 	log_assert(c->type == comm_http);
3212 	ub_comm_base_now(c->ev->base);
3213 
3214 	if(event&UB_EV_TIMEOUT) {
3215 		verbose(VERB_QUERY, "http took too long, dropped");
3216 		reclaim_http_handler(c);
3217 		if(!c->tcp_do_close) {
3218 			fptr_ok(fptr_whitelist_comm_point(c->callback));
3219 			(void)(*c->callback)(c, c->cb_arg,
3220 				NETEVENT_TIMEOUT, NULL);
3221 		}
3222 		return;
3223 	}
3224 	if(event&UB_EV_READ) {
3225 		if(!comm_point_http_handle_read(fd, c)) {
3226 			reclaim_http_handler(c);
3227 			if(!c->tcp_do_close) {
3228 				fptr_ok(fptr_whitelist_comm_point(
3229 					c->callback));
3230 				(void)(*c->callback)(c, c->cb_arg,
3231 					NETEVENT_CLOSED, NULL);
3232 			}
3233 		}
3234 		return;
3235 	}
3236 	if(event&UB_EV_WRITE) {
3237 		if(!comm_point_http_handle_write(fd, c)) {
3238 			reclaim_http_handler(c);
3239 			if(!c->tcp_do_close) {
3240 				fptr_ok(fptr_whitelist_comm_point(
3241 					c->callback));
3242 				(void)(*c->callback)(c, c->cb_arg,
3243 					NETEVENT_CLOSED, NULL);
3244 			}
3245 		}
3246 		return;
3247 	}
3248 	log_err("Ignored event %d for httphdl.", event);
3249 }
3250 
3251 void comm_point_local_handle_callback(int fd, short event, void* arg)
3252 {
3253 	struct comm_point* c = (struct comm_point*)arg;
3254 	log_assert(c->type == comm_local);
3255 	ub_comm_base_now(c->ev->base);
3256 
3257 	if(event&UB_EV_READ) {
3258 		if(!comm_point_tcp_handle_read(fd, c, 1)) {
3259 			fptr_ok(fptr_whitelist_comm_point(c->callback));
3260 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
3261 				NULL);
3262 		}
3263 		return;
3264 	}
3265 	log_err("Ignored event %d for localhdl.", event);
3266 }
3267 
3268 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
3269 	short event, void* arg)
3270 {
3271 	struct comm_point* c = (struct comm_point*)arg;
3272 	int err = NETEVENT_NOERROR;
3273 	log_assert(c->type == comm_raw);
3274 	ub_comm_base_now(c->ev->base);
3275 
3276 	if(event&UB_EV_TIMEOUT)
3277 		err = NETEVENT_TIMEOUT;
3278 	fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
3279 	(void)(*c->callback)(c, c->cb_arg, err, NULL);
3280 }
3281 
3282 struct comm_point*
3283 comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer,
3284 	comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3285 {
3286 	struct comm_point* c = (struct comm_point*)calloc(1,
3287 		sizeof(struct comm_point));
3288 	short evbits;
3289 	if(!c)
3290 		return NULL;
3291 	c->ev = (struct internal_event*)calloc(1,
3292 		sizeof(struct internal_event));
3293 	if(!c->ev) {
3294 		free(c);
3295 		return NULL;
3296 	}
3297 	c->ev->base = base;
3298 	c->fd = fd;
3299 	c->buffer = buffer;
3300 	c->timeout = NULL;
3301 	c->tcp_is_reading = 0;
3302 	c->tcp_byte_count = 0;
3303 	c->tcp_parent = NULL;
3304 	c->max_tcp_count = 0;
3305 	c->cur_tcp_count = 0;
3306 	c->tcp_handlers = NULL;
3307 	c->tcp_free = NULL;
3308 	c->type = comm_udp;
3309 	c->tcp_do_close = 0;
3310 	c->do_not_close = 0;
3311 	c->tcp_do_toggle_rw = 0;
3312 	c->tcp_check_nb_connect = 0;
3313 #ifdef USE_MSG_FASTOPEN
3314 	c->tcp_do_fastopen = 0;
3315 #endif
3316 #ifdef USE_DNSCRYPT
3317 	c->dnscrypt = 0;
3318 	c->dnscrypt_buffer = buffer;
3319 #endif
3320 	c->inuse = 0;
3321 	c->callback = callback;
3322 	c->cb_arg = callback_arg;
3323 	c->socket = socket;
3324 	evbits = UB_EV_READ | UB_EV_PERSIST;
3325 	/* ub_event stuff */
3326 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3327 		comm_point_udp_callback, c);
3328 	if(c->ev->ev == NULL) {
3329 		log_err("could not baseset udp event");
3330 		comm_point_delete(c);
3331 		return NULL;
3332 	}
3333 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
3334 		log_err("could not add udp event");
3335 		comm_point_delete(c);
3336 		return NULL;
3337 	}
3338 	c->event_added = 1;
3339 	return c;
3340 }
3341 
3342 struct comm_point*
3343 comm_point_create_udp_ancil(struct comm_base *base, int fd,
3344 	sldns_buffer* buffer,
3345 	comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3346 {
3347 	struct comm_point* c = (struct comm_point*)calloc(1,
3348 		sizeof(struct comm_point));
3349 	short evbits;
3350 	if(!c)
3351 		return NULL;
3352 	c->ev = (struct internal_event*)calloc(1,
3353 		sizeof(struct internal_event));
3354 	if(!c->ev) {
3355 		free(c);
3356 		return NULL;
3357 	}
3358 	c->ev->base = base;
3359 	c->fd = fd;
3360 	c->buffer = buffer;
3361 	c->timeout = NULL;
3362 	c->tcp_is_reading = 0;
3363 	c->tcp_byte_count = 0;
3364 	c->tcp_parent = NULL;
3365 	c->max_tcp_count = 0;
3366 	c->cur_tcp_count = 0;
3367 	c->tcp_handlers = NULL;
3368 	c->tcp_free = NULL;
3369 	c->type = comm_udp;
3370 	c->tcp_do_close = 0;
3371 	c->do_not_close = 0;
3372 #ifdef USE_DNSCRYPT
3373 	c->dnscrypt = 0;
3374 	c->dnscrypt_buffer = buffer;
3375 #endif
3376 	c->inuse = 0;
3377 	c->tcp_do_toggle_rw = 0;
3378 	c->tcp_check_nb_connect = 0;
3379 #ifdef USE_MSG_FASTOPEN
3380 	c->tcp_do_fastopen = 0;
3381 #endif
3382 	c->callback = callback;
3383 	c->cb_arg = callback_arg;
3384 	c->socket = socket;
3385 	evbits = UB_EV_READ | UB_EV_PERSIST;
3386 	/* ub_event stuff */
3387 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3388 		comm_point_udp_ancil_callback, c);
3389 	if(c->ev->ev == NULL) {
3390 		log_err("could not baseset udp event");
3391 		comm_point_delete(c);
3392 		return NULL;
3393 	}
3394 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
3395 		log_err("could not add udp event");
3396 		comm_point_delete(c);
3397 		return NULL;
3398 	}
3399 	c->event_added = 1;
3400 	return c;
3401 }
3402 
3403 static struct comm_point*
3404 comm_point_create_tcp_handler(struct comm_base *base,
3405 	struct comm_point* parent, size_t bufsize,
3406 	struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
3407 	void* callback_arg, struct unbound_socket* socket)
3408 {
3409 	struct comm_point* c = (struct comm_point*)calloc(1,
3410 		sizeof(struct comm_point));
3411 	short evbits;
3412 	if(!c)
3413 		return NULL;
3414 	c->ev = (struct internal_event*)calloc(1,
3415 		sizeof(struct internal_event));
3416 	if(!c->ev) {
3417 		free(c);
3418 		return NULL;
3419 	}
3420 	c->ev->base = base;
3421 	c->fd = -1;
3422 	c->buffer = sldns_buffer_new(bufsize);
3423 	if(!c->buffer) {
3424 		free(c->ev);
3425 		free(c);
3426 		return NULL;
3427 	}
3428 	c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
3429 	if(!c->timeout) {
3430 		sldns_buffer_free(c->buffer);
3431 		free(c->ev);
3432 		free(c);
3433 		return NULL;
3434 	}
3435 	c->tcp_is_reading = 0;
3436 	c->tcp_byte_count = 0;
3437 	c->tcp_parent = parent;
3438 	c->tcp_timeout_msec = parent->tcp_timeout_msec;
3439 	c->tcp_conn_limit = parent->tcp_conn_limit;
3440 	c->tcl_addr = NULL;
3441 	c->tcp_keepalive = 0;
3442 	c->max_tcp_count = 0;
3443 	c->cur_tcp_count = 0;
3444 	c->tcp_handlers = NULL;
3445 	c->tcp_free = NULL;
3446 	c->type = comm_tcp;
3447 	c->tcp_do_close = 0;
3448 	c->do_not_close = 0;
3449 	c->tcp_do_toggle_rw = 1;
3450 	c->tcp_check_nb_connect = 0;
3451 #ifdef USE_MSG_FASTOPEN
3452 	c->tcp_do_fastopen = 0;
3453 #endif
3454 #ifdef USE_DNSCRYPT
3455 	c->dnscrypt = 0;
3456 	/* We don't know just yet if this is a dnscrypt channel. Allocation
3457 	 * will be done when handling the callback. */
3458 	c->dnscrypt_buffer = c->buffer;
3459 #endif
3460 	c->repinfo.c = c;
3461 	c->callback = callback;
3462 	c->cb_arg = callback_arg;
3463 	c->socket = socket;
3464 	if(spoolbuf) {
3465 		c->tcp_req_info = tcp_req_info_create(spoolbuf);
3466 		if(!c->tcp_req_info) {
3467 			log_err("could not create tcp commpoint");
3468 			sldns_buffer_free(c->buffer);
3469 			free(c->timeout);
3470 			free(c->ev);
3471 			free(c);
3472 			return NULL;
3473 		}
3474 		c->tcp_req_info->cp = c;
3475 		c->tcp_do_close = 1;
3476 		c->tcp_do_toggle_rw = 0;
3477 	}
3478 	/* add to parent free list */
3479 	c->tcp_free = parent->tcp_free;
3480 	parent->tcp_free = c;
3481 	/* ub_event stuff */
3482 	evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
3483 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3484 		comm_point_tcp_handle_callback, c);
3485 	if(c->ev->ev == NULL)
3486 	{
3487 		log_err("could not basetset tcphdl event");
3488 		parent->tcp_free = c->tcp_free;
3489 		tcp_req_info_delete(c->tcp_req_info);
3490 		sldns_buffer_free(c->buffer);
3491 		free(c->timeout);
3492 		free(c->ev);
3493 		free(c);
3494 		return NULL;
3495 	}
3496 	return c;
3497 }
3498 
3499 static struct comm_point*
3500 comm_point_create_http_handler(struct comm_base *base,
3501 	struct comm_point* parent, size_t bufsize, int harden_large_queries,
3502 	uint32_t http_max_streams, char* http_endpoint,
3503 	comm_point_callback_type* callback, void* callback_arg,
3504 	struct unbound_socket* socket)
3505 {
3506 	struct comm_point* c = (struct comm_point*)calloc(1,
3507 		sizeof(struct comm_point));
3508 	short evbits;
3509 	if(!c)
3510 		return NULL;
3511 	c->ev = (struct internal_event*)calloc(1,
3512 		sizeof(struct internal_event));
3513 	if(!c->ev) {
3514 		free(c);
3515 		return NULL;
3516 	}
3517 	c->ev->base = base;
3518 	c->fd = -1;
3519 	c->buffer = sldns_buffer_new(bufsize);
3520 	if(!c->buffer) {
3521 		free(c->ev);
3522 		free(c);
3523 		return NULL;
3524 	}
3525 	c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
3526 	if(!c->timeout) {
3527 		sldns_buffer_free(c->buffer);
3528 		free(c->ev);
3529 		free(c);
3530 		return NULL;
3531 	}
3532 	c->tcp_is_reading = 0;
3533 	c->tcp_byte_count = 0;
3534 	c->tcp_parent = parent;
3535 	c->tcp_timeout_msec = parent->tcp_timeout_msec;
3536 	c->tcp_conn_limit = parent->tcp_conn_limit;
3537 	c->tcl_addr = NULL;
3538 	c->tcp_keepalive = 0;
3539 	c->max_tcp_count = 0;
3540 	c->cur_tcp_count = 0;
3541 	c->tcp_handlers = NULL;
3542 	c->tcp_free = NULL;
3543 	c->type = comm_http;
3544 	c->tcp_do_close = 1;
3545 	c->do_not_close = 0;
3546 	c->tcp_do_toggle_rw = 1; /* will be set to 0 after http2 upgrade */
3547 	c->tcp_check_nb_connect = 0;
3548 #ifdef USE_MSG_FASTOPEN
3549 	c->tcp_do_fastopen = 0;
3550 #endif
3551 #ifdef USE_DNSCRYPT
3552 	c->dnscrypt = 0;
3553 	c->dnscrypt_buffer = NULL;
3554 #endif
3555 	c->repinfo.c = c;
3556 	c->callback = callback;
3557 	c->cb_arg = callback_arg;
3558 	c->socket = socket;
3559 
3560 	c->http_min_version = http_version_2;
3561 	c->http2_stream_max_qbuffer_size = bufsize;
3562 	if(harden_large_queries && bufsize > 512)
3563 		c->http2_stream_max_qbuffer_size = 512;
3564 	c->http2_max_streams = http_max_streams;
3565 	if(!(c->http_endpoint = strdup(http_endpoint))) {
3566 		log_err("could not strdup http_endpoint");
3567 		sldns_buffer_free(c->buffer);
3568 		free(c->timeout);
3569 		free(c->ev);
3570 		free(c);
3571 		return NULL;
3572 	}
3573 	c->use_h2 = 0;
3574 #ifdef HAVE_NGHTTP2
3575 	if(!(c->h2_session = http2_session_create(c))) {
3576 		log_err("could not create http2 session");
3577 		free(c->http_endpoint);
3578 		sldns_buffer_free(c->buffer);
3579 		free(c->timeout);
3580 		free(c->ev);
3581 		free(c);
3582 		return NULL;
3583 	}
3584 	if(!(c->h2_session->callbacks = http2_req_callbacks_create())) {
3585 		log_err("could not create http2 callbacks");
3586 		http2_session_delete(c->h2_session);
3587 		free(c->http_endpoint);
3588 		sldns_buffer_free(c->buffer);
3589 		free(c->timeout);
3590 		free(c->ev);
3591 		free(c);
3592 		return NULL;
3593 	}
3594 #endif
3595 
3596 	/* add to parent free list */
3597 	c->tcp_free = parent->tcp_free;
3598 	parent->tcp_free = c;
3599 	/* ub_event stuff */
3600 	evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
3601 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3602 		comm_point_http_handle_callback, c);
3603 	if(c->ev->ev == NULL)
3604 	{
3605 		log_err("could not set http handler event");
3606 		parent->tcp_free = c->tcp_free;
3607 		http2_session_delete(c->h2_session);
3608 		sldns_buffer_free(c->buffer);
3609 		free(c->timeout);
3610 		free(c->ev);
3611 		free(c);
3612 		return NULL;
3613 	}
3614 	return c;
3615 }
3616 
3617 struct comm_point*
3618 comm_point_create_tcp(struct comm_base *base, int fd, int num,
3619 	int idle_timeout, int harden_large_queries,
3620 	uint32_t http_max_streams, char* http_endpoint,
3621 	struct tcl_list* tcp_conn_limit, size_t bufsize,
3622 	struct sldns_buffer* spoolbuf, enum listen_type port_type,
3623 	comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3624 {
3625 	struct comm_point* c = (struct comm_point*)calloc(1,
3626 		sizeof(struct comm_point));
3627 	short evbits;
3628 	int i;
3629 	/* first allocate the TCP accept listener */
3630 	if(!c)
3631 		return NULL;
3632 	c->ev = (struct internal_event*)calloc(1,
3633 		sizeof(struct internal_event));
3634 	if(!c->ev) {
3635 		free(c);
3636 		return NULL;
3637 	}
3638 	c->ev->base = base;
3639 	c->fd = fd;
3640 	c->buffer = NULL;
3641 	c->timeout = NULL;
3642 	c->tcp_is_reading = 0;
3643 	c->tcp_byte_count = 0;
3644 	c->tcp_timeout_msec = idle_timeout;
3645 	c->tcp_conn_limit = tcp_conn_limit;
3646 	c->tcl_addr = NULL;
3647 	c->tcp_keepalive = 0;
3648 	c->tcp_parent = NULL;
3649 	c->max_tcp_count = num;
3650 	c->cur_tcp_count = 0;
3651 	c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
3652 		sizeof(struct comm_point*));
3653 	if(!c->tcp_handlers) {
3654 		free(c->ev);
3655 		free(c);
3656 		return NULL;
3657 	}
3658 	c->tcp_free = NULL;
3659 	c->type = comm_tcp_accept;
3660 	c->tcp_do_close = 0;
3661 	c->do_not_close = 0;
3662 	c->tcp_do_toggle_rw = 0;
3663 	c->tcp_check_nb_connect = 0;
3664 #ifdef USE_MSG_FASTOPEN
3665 	c->tcp_do_fastopen = 0;
3666 #endif
3667 #ifdef USE_DNSCRYPT
3668 	c->dnscrypt = 0;
3669 	c->dnscrypt_buffer = NULL;
3670 #endif
3671 	c->callback = NULL;
3672 	c->cb_arg = NULL;
3673 	c->socket = socket;
3674 	evbits = UB_EV_READ | UB_EV_PERSIST;
3675 	/* ub_event stuff */
3676 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3677 		comm_point_tcp_accept_callback, c);
3678 	if(c->ev->ev == NULL) {
3679 		log_err("could not baseset tcpacc event");
3680 		comm_point_delete(c);
3681 		return NULL;
3682 	}
3683 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3684 		log_err("could not add tcpacc event");
3685 		comm_point_delete(c);
3686 		return NULL;
3687 	}
3688 	c->event_added = 1;
3689 	/* now prealloc the handlers */
3690 	for(i=0; i<num; i++) {
3691 		if(port_type == listen_type_tcp ||
3692 			port_type == listen_type_ssl ||
3693 			port_type == listen_type_tcp_dnscrypt) {
3694 			c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
3695 				c, bufsize, spoolbuf, callback, callback_arg, socket);
3696 		} else if(port_type == listen_type_http) {
3697 			c->tcp_handlers[i] = comm_point_create_http_handler(
3698 				base, c, bufsize, harden_large_queries,
3699 				http_max_streams, http_endpoint,
3700 				callback, callback_arg, socket);
3701 		}
3702 		else {
3703 			log_err("could not create tcp handler, unknown listen "
3704 				"type");
3705 			return NULL;
3706 		}
3707 		if(!c->tcp_handlers[i]) {
3708 			comm_point_delete(c);
3709 			return NULL;
3710 		}
3711 	}
3712 
3713 	return c;
3714 }
3715 
3716 struct comm_point*
3717 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
3718         comm_point_callback_type* callback, void* callback_arg)
3719 {
3720 	struct comm_point* c = (struct comm_point*)calloc(1,
3721 		sizeof(struct comm_point));
3722 	short evbits;
3723 	if(!c)
3724 		return NULL;
3725 	c->ev = (struct internal_event*)calloc(1,
3726 		sizeof(struct internal_event));
3727 	if(!c->ev) {
3728 		free(c);
3729 		return NULL;
3730 	}
3731 	c->ev->base = base;
3732 	c->fd = -1;
3733 	c->buffer = sldns_buffer_new(bufsize);
3734 	if(!c->buffer) {
3735 		free(c->ev);
3736 		free(c);
3737 		return NULL;
3738 	}
3739 	c->timeout = NULL;
3740 	c->tcp_is_reading = 0;
3741 	c->tcp_byte_count = 0;
3742 	c->tcp_timeout_msec = TCP_QUERY_TIMEOUT;
3743 	c->tcp_conn_limit = NULL;
3744 	c->tcl_addr = NULL;
3745 	c->tcp_keepalive = 0;
3746 	c->tcp_parent = NULL;
3747 	c->max_tcp_count = 0;
3748 	c->cur_tcp_count = 0;
3749 	c->tcp_handlers = NULL;
3750 	c->tcp_free = NULL;
3751 	c->type = comm_tcp;
3752 	c->tcp_do_close = 0;
3753 	c->do_not_close = 0;
3754 	c->tcp_do_toggle_rw = 1;
3755 	c->tcp_check_nb_connect = 1;
3756 #ifdef USE_MSG_FASTOPEN
3757 	c->tcp_do_fastopen = 1;
3758 #endif
3759 #ifdef USE_DNSCRYPT
3760 	c->dnscrypt = 0;
3761 	c->dnscrypt_buffer = c->buffer;
3762 #endif
3763 	c->repinfo.c = c;
3764 	c->callback = callback;
3765 	c->cb_arg = callback_arg;
3766 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
3767 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3768 		comm_point_tcp_handle_callback, c);
3769 	if(c->ev->ev == NULL)
3770 	{
3771 		log_err("could not baseset tcpout event");
3772 		sldns_buffer_free(c->buffer);
3773 		free(c->ev);
3774 		free(c);
3775 		return NULL;
3776 	}
3777 
3778 	return c;
3779 }
3780 
3781 struct comm_point*
3782 comm_point_create_http_out(struct comm_base *base, size_t bufsize,
3783         comm_point_callback_type* callback, void* callback_arg,
3784 	sldns_buffer* temp)
3785 {
3786 	struct comm_point* c = (struct comm_point*)calloc(1,
3787 		sizeof(struct comm_point));
3788 	short evbits;
3789 	if(!c)
3790 		return NULL;
3791 	c->ev = (struct internal_event*)calloc(1,
3792 		sizeof(struct internal_event));
3793 	if(!c->ev) {
3794 		free(c);
3795 		return NULL;
3796 	}
3797 	c->ev->base = base;
3798 	c->fd = -1;
3799 	c->buffer = sldns_buffer_new(bufsize);
3800 	if(!c->buffer) {
3801 		free(c->ev);
3802 		free(c);
3803 		return NULL;
3804 	}
3805 	c->timeout = NULL;
3806 	c->tcp_is_reading = 0;
3807 	c->tcp_byte_count = 0;
3808 	c->tcp_parent = NULL;
3809 	c->max_tcp_count = 0;
3810 	c->cur_tcp_count = 0;
3811 	c->tcp_handlers = NULL;
3812 	c->tcp_free = NULL;
3813 	c->type = comm_http;
3814 	c->tcp_do_close = 0;
3815 	c->do_not_close = 0;
3816 	c->tcp_do_toggle_rw = 1;
3817 	c->tcp_check_nb_connect = 1;
3818 	c->http_in_headers = 1;
3819 	c->http_in_chunk_headers = 0;
3820 	c->http_is_chunked = 0;
3821 	c->http_temp = temp;
3822 #ifdef USE_MSG_FASTOPEN
3823 	c->tcp_do_fastopen = 1;
3824 #endif
3825 #ifdef USE_DNSCRYPT
3826 	c->dnscrypt = 0;
3827 	c->dnscrypt_buffer = c->buffer;
3828 #endif
3829 	c->repinfo.c = c;
3830 	c->callback = callback;
3831 	c->cb_arg = callback_arg;
3832 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
3833 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3834 		comm_point_http_handle_callback, c);
3835 	if(c->ev->ev == NULL)
3836 	{
3837 		log_err("could not baseset tcpout event");
3838 #ifdef HAVE_SSL
3839 		SSL_free(c->ssl);
3840 #endif
3841 		sldns_buffer_free(c->buffer);
3842 		free(c->ev);
3843 		free(c);
3844 		return NULL;
3845 	}
3846 
3847 	return c;
3848 }
3849 
3850 struct comm_point*
3851 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
3852         comm_point_callback_type* callback, void* callback_arg)
3853 {
3854 	struct comm_point* c = (struct comm_point*)calloc(1,
3855 		sizeof(struct comm_point));
3856 	short evbits;
3857 	if(!c)
3858 		return NULL;
3859 	c->ev = (struct internal_event*)calloc(1,
3860 		sizeof(struct internal_event));
3861 	if(!c->ev) {
3862 		free(c);
3863 		return NULL;
3864 	}
3865 	c->ev->base = base;
3866 	c->fd = fd;
3867 	c->buffer = sldns_buffer_new(bufsize);
3868 	if(!c->buffer) {
3869 		free(c->ev);
3870 		free(c);
3871 		return NULL;
3872 	}
3873 	c->timeout = NULL;
3874 	c->tcp_is_reading = 1;
3875 	c->tcp_byte_count = 0;
3876 	c->tcp_parent = NULL;
3877 	c->max_tcp_count = 0;
3878 	c->cur_tcp_count = 0;
3879 	c->tcp_handlers = NULL;
3880 	c->tcp_free = NULL;
3881 	c->type = comm_local;
3882 	c->tcp_do_close = 0;
3883 	c->do_not_close = 1;
3884 	c->tcp_do_toggle_rw = 0;
3885 	c->tcp_check_nb_connect = 0;
3886 #ifdef USE_MSG_FASTOPEN
3887 	c->tcp_do_fastopen = 0;
3888 #endif
3889 #ifdef USE_DNSCRYPT
3890 	c->dnscrypt = 0;
3891 	c->dnscrypt_buffer = c->buffer;
3892 #endif
3893 	c->callback = callback;
3894 	c->cb_arg = callback_arg;
3895 	/* ub_event stuff */
3896 	evbits = UB_EV_PERSIST | UB_EV_READ;
3897 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3898 		comm_point_local_handle_callback, c);
3899 	if(c->ev->ev == NULL) {
3900 		log_err("could not baseset localhdl event");
3901 		free(c->ev);
3902 		free(c);
3903 		return NULL;
3904 	}
3905 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3906 		log_err("could not add localhdl event");
3907 		ub_event_free(c->ev->ev);
3908 		free(c->ev);
3909 		free(c);
3910 		return NULL;
3911 	}
3912 	c->event_added = 1;
3913 	return c;
3914 }
3915 
3916 struct comm_point*
3917 comm_point_create_raw(struct comm_base* base, int fd, int writing,
3918 	comm_point_callback_type* callback, void* callback_arg)
3919 {
3920 	struct comm_point* c = (struct comm_point*)calloc(1,
3921 		sizeof(struct comm_point));
3922 	short evbits;
3923 	if(!c)
3924 		return NULL;
3925 	c->ev = (struct internal_event*)calloc(1,
3926 		sizeof(struct internal_event));
3927 	if(!c->ev) {
3928 		free(c);
3929 		return NULL;
3930 	}
3931 	c->ev->base = base;
3932 	c->fd = fd;
3933 	c->buffer = NULL;
3934 	c->timeout = NULL;
3935 	c->tcp_is_reading = 0;
3936 	c->tcp_byte_count = 0;
3937 	c->tcp_parent = NULL;
3938 	c->max_tcp_count = 0;
3939 	c->cur_tcp_count = 0;
3940 	c->tcp_handlers = NULL;
3941 	c->tcp_free = NULL;
3942 	c->type = comm_raw;
3943 	c->tcp_do_close = 0;
3944 	c->do_not_close = 1;
3945 	c->tcp_do_toggle_rw = 0;
3946 	c->tcp_check_nb_connect = 0;
3947 #ifdef USE_MSG_FASTOPEN
3948 	c->tcp_do_fastopen = 0;
3949 #endif
3950 #ifdef USE_DNSCRYPT
3951 	c->dnscrypt = 0;
3952 	c->dnscrypt_buffer = c->buffer;
3953 #endif
3954 	c->callback = callback;
3955 	c->cb_arg = callback_arg;
3956 	/* ub_event stuff */
3957 	if(writing)
3958 		evbits = UB_EV_PERSIST | UB_EV_WRITE;
3959 	else 	evbits = UB_EV_PERSIST | UB_EV_READ;
3960 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3961 		comm_point_raw_handle_callback, c);
3962 	if(c->ev->ev == NULL) {
3963 		log_err("could not baseset rawhdl event");
3964 		free(c->ev);
3965 		free(c);
3966 		return NULL;
3967 	}
3968 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3969 		log_err("could not add rawhdl event");
3970 		ub_event_free(c->ev->ev);
3971 		free(c->ev);
3972 		free(c);
3973 		return NULL;
3974 	}
3975 	c->event_added = 1;
3976 	return c;
3977 }
3978 
3979 void
3980 comm_point_close(struct comm_point* c)
3981 {
3982 	if(!c)
3983 		return;
3984 	if(c->fd != -1) {
3985 		verbose(5, "comm_point_close of %d: event_del", c->fd);
3986 		if(c->event_added) {
3987 			if(ub_event_del(c->ev->ev) != 0) {
3988 				log_err("could not event_del on close");
3989 			}
3990 			c->event_added = 0;
3991 		}
3992 	}
3993 	tcl_close_connection(c->tcl_addr);
3994 	if(c->tcp_req_info)
3995 		tcp_req_info_clear(c->tcp_req_info);
3996 	if(c->h2_session)
3997 		http2_session_server_delete(c->h2_session);
3998 
3999 	/* close fd after removing from event lists, or epoll.. is messed up */
4000 	if(c->fd != -1 && !c->do_not_close) {
4001 #ifdef USE_WINSOCK
4002 		if(c->type == comm_tcp || c->type == comm_http) {
4003 			/* delete sticky events for the fd, it gets closed */
4004 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
4005 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
4006 		}
4007 #endif
4008 		verbose(VERB_ALGO, "close fd %d", c->fd);
4009 		sock_close(c->fd);
4010 	}
4011 	c->fd = -1;
4012 }
4013 
4014 void
4015 comm_point_delete(struct comm_point* c)
4016 {
4017 	if(!c)
4018 		return;
4019 	if((c->type == comm_tcp || c->type == comm_http) && c->ssl) {
4020 #ifdef HAVE_SSL
4021 		SSL_shutdown(c->ssl);
4022 		SSL_free(c->ssl);
4023 #endif
4024 	}
4025 	if(c->type == comm_http && c->http_endpoint) {
4026 		free(c->http_endpoint);
4027 		c->http_endpoint = NULL;
4028 	}
4029 	comm_point_close(c);
4030 	if(c->tcp_handlers) {
4031 		int i;
4032 		for(i=0; i<c->max_tcp_count; i++)
4033 			comm_point_delete(c->tcp_handlers[i]);
4034 		free(c->tcp_handlers);
4035 	}
4036 	free(c->timeout);
4037 	if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) {
4038 		sldns_buffer_free(c->buffer);
4039 #ifdef USE_DNSCRYPT
4040 		if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) {
4041 			sldns_buffer_free(c->dnscrypt_buffer);
4042 		}
4043 #endif
4044 		if(c->tcp_req_info) {
4045 			tcp_req_info_delete(c->tcp_req_info);
4046 		}
4047 		if(c->h2_session) {
4048 			http2_session_delete(c->h2_session);
4049 		}
4050 	}
4051 	ub_event_free(c->ev->ev);
4052 	free(c->ev);
4053 	free(c);
4054 }
4055 
4056 void
4057 comm_point_send_reply(struct comm_reply *repinfo)
4058 {
4059 	struct sldns_buffer* buffer;
4060 	log_assert(repinfo && repinfo->c);
4061 #ifdef USE_DNSCRYPT
4062 	buffer = repinfo->c->dnscrypt_buffer;
4063 	if(!dnsc_handle_uncurved_request(repinfo)) {
4064 		return;
4065 	}
4066 #else
4067 	buffer = repinfo->c->buffer;
4068 #endif
4069 	if(repinfo->c->type == comm_udp) {
4070 		if(repinfo->srctype)
4071 			comm_point_send_udp_msg_if(repinfo->c,
4072 			buffer, (struct sockaddr*)&repinfo->addr,
4073 			repinfo->addrlen, repinfo);
4074 		else
4075 			comm_point_send_udp_msg(repinfo->c, buffer,
4076 			(struct sockaddr*)&repinfo->addr, repinfo->addrlen, 0);
4077 #ifdef USE_DNSTAP
4078 		/*
4079 		 * sending src (client)/dst (local service) addresses over DNSTAP from udp callback
4080 		 */
4081 		if(repinfo->c->dtenv != NULL && repinfo->c->dtenv->log_client_response_messages) {
4082 			log_addr(VERB_ALGO, "from local addr", (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->socket->addr->ai_addrlen);
4083 			log_addr(VERB_ALGO, "response to client", &repinfo->addr, repinfo->addrlen);
4084 			dt_msg_send_client_response(repinfo->c->dtenv, &repinfo->addr, (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->type, repinfo->c->buffer);
4085 		}
4086 #endif
4087 	} else {
4088 #ifdef USE_DNSTAP
4089 		/*
4090 		 * sending src (client)/dst (local service) addresses over DNSTAP from TCP callback
4091 		 */
4092 		if(repinfo->c->tcp_parent->dtenv != NULL && repinfo->c->tcp_parent->dtenv->log_client_response_messages) {
4093 			log_addr(VERB_ALGO, "from local addr", (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->socket->addr->ai_addrlen);
4094 			log_addr(VERB_ALGO, "response to client", &repinfo->addr, repinfo->addrlen);
4095 			dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv, &repinfo->addr, (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->type,
4096 				( repinfo->c->tcp_req_info? repinfo->c->tcp_req_info->spool_buffer: repinfo->c->buffer ));
4097 		}
4098 #endif
4099 		if(repinfo->c->tcp_req_info) {
4100 			tcp_req_info_send_reply(repinfo->c->tcp_req_info);
4101 		} else if(repinfo->c->use_h2) {
4102 			if(!http2_submit_dns_response(repinfo->c->h2_session)) {
4103 				comm_point_drop_reply(repinfo);
4104 				return;
4105 			}
4106 			repinfo->c->h2_stream = NULL;
4107 			repinfo->c->tcp_is_reading = 0;
4108 			comm_point_stop_listening(repinfo->c);
4109 			comm_point_start_listening(repinfo->c, -1,
4110 				adjusted_tcp_timeout(repinfo->c));
4111 			return;
4112 		} else {
4113 			comm_point_start_listening(repinfo->c, -1,
4114 				adjusted_tcp_timeout(repinfo->c));
4115 		}
4116 	}
4117 }
4118 
4119 void
4120 comm_point_drop_reply(struct comm_reply* repinfo)
4121 {
4122 	if(!repinfo)
4123 		return;
4124 	log_assert(repinfo->c);
4125 	log_assert(repinfo->c->type != comm_tcp_accept);
4126 	if(repinfo->c->type == comm_udp)
4127 		return;
4128 	if(repinfo->c->tcp_req_info)
4129 		repinfo->c->tcp_req_info->is_drop = 1;
4130 	if(repinfo->c->type == comm_http) {
4131 		if(repinfo->c->h2_session) {
4132 			repinfo->c->h2_session->is_drop = 1;
4133 			if(!repinfo->c->h2_session->postpone_drop)
4134 				reclaim_http_handler(repinfo->c);
4135 			return;
4136 		}
4137 		reclaim_http_handler(repinfo->c);
4138 		return;
4139 	}
4140 	reclaim_tcp_handler(repinfo->c);
4141 }
4142 
4143 void
4144 comm_point_stop_listening(struct comm_point* c)
4145 {
4146 	verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
4147 	if(c->event_added) {
4148 		if(ub_event_del(c->ev->ev) != 0) {
4149 			log_err("event_del error to stoplisten");
4150 		}
4151 		c->event_added = 0;
4152 	}
4153 }
4154 
4155 void
4156 comm_point_start_listening(struct comm_point* c, int newfd, int msec)
4157 {
4158 	verbose(VERB_ALGO, "comm point start listening %d (%d msec)",
4159 		c->fd==-1?newfd:c->fd, msec);
4160 	if(c->type == comm_tcp_accept && !c->tcp_free) {
4161 		/* no use to start listening no free slots. */
4162 		return;
4163 	}
4164 	if(c->event_added) {
4165 		if(ub_event_del(c->ev->ev) != 0) {
4166 			log_err("event_del error to startlisten");
4167 		}
4168 		c->event_added = 0;
4169 	}
4170 	if(msec != -1 && msec != 0) {
4171 		if(!c->timeout) {
4172 			c->timeout = (struct timeval*)malloc(sizeof(
4173 				struct timeval));
4174 			if(!c->timeout) {
4175 				log_err("cpsl: malloc failed. No net read.");
4176 				return;
4177 			}
4178 		}
4179 		ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT);
4180 #ifndef S_SPLINT_S /* splint fails on struct timeval. */
4181 		c->timeout->tv_sec = msec/1000;
4182 		c->timeout->tv_usec = (msec%1000)*1000;
4183 #endif /* S_SPLINT_S */
4184 	} else {
4185 		if(msec == 0 || !c->timeout) {
4186 			ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
4187 		}
4188 	}
4189 	if(c->type == comm_tcp || c->type == comm_http) {
4190 		ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4191 		if(c->tcp_write_and_read) {
4192 			verbose(5, "startlistening %d mode rw", (newfd==-1?c->fd:newfd));
4193 			ub_event_add_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4194 		} else if(c->tcp_is_reading) {
4195 			verbose(5, "startlistening %d mode r", (newfd==-1?c->fd:newfd));
4196 			ub_event_add_bits(c->ev->ev, UB_EV_READ);
4197 		} else	{
4198 			verbose(5, "startlistening %d mode w", (newfd==-1?c->fd:newfd));
4199 			ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
4200 		}
4201 	}
4202 	if(newfd != -1) {
4203 		if(c->fd != -1 && c->fd != newfd) {
4204 			verbose(5, "cpsl close of fd %d for %d", c->fd, newfd);
4205 			sock_close(c->fd);
4206 		}
4207 		c->fd = newfd;
4208 		ub_event_set_fd(c->ev->ev, c->fd);
4209 	}
4210 	if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) {
4211 		log_err("event_add failed. in cpsl.");
4212 		return;
4213 	}
4214 	c->event_added = 1;
4215 }
4216 
4217 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
4218 {
4219 	verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
4220 	if(c->event_added) {
4221 		if(ub_event_del(c->ev->ev) != 0) {
4222 			log_err("event_del error to cplf");
4223 		}
4224 		c->event_added = 0;
4225 	}
4226 	if(!c->timeout) {
4227 		ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
4228 	}
4229 	ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4230 	if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ);
4231 	if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
4232 	if(ub_event_add(c->ev->ev, c->timeout) != 0) {
4233 		log_err("event_add failed. in cplf.");
4234 		return;
4235 	}
4236 	c->event_added = 1;
4237 }
4238 
4239 size_t comm_point_get_mem(struct comm_point* c)
4240 {
4241 	size_t s;
4242 	if(!c)
4243 		return 0;
4244 	s = sizeof(*c) + sizeof(*c->ev);
4245 	if(c->timeout)
4246 		s += sizeof(*c->timeout);
4247 	if(c->type == comm_tcp || c->type == comm_local) {
4248 		s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer);
4249 #ifdef USE_DNSCRYPT
4250 		s += sizeof(*c->dnscrypt_buffer);
4251 		if(c->buffer != c->dnscrypt_buffer) {
4252 			s += sldns_buffer_capacity(c->dnscrypt_buffer);
4253 		}
4254 #endif
4255 	}
4256 	if(c->type == comm_tcp_accept) {
4257 		int i;
4258 		for(i=0; i<c->max_tcp_count; i++)
4259 			s += comm_point_get_mem(c->tcp_handlers[i]);
4260 	}
4261 	return s;
4262 }
4263 
4264 struct comm_timer*
4265 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
4266 {
4267 	struct internal_timer *tm = (struct internal_timer*)calloc(1,
4268 		sizeof(struct internal_timer));
4269 	if(!tm) {
4270 		log_err("malloc failed");
4271 		return NULL;
4272 	}
4273 	tm->super.ev_timer = tm;
4274 	tm->base = base;
4275 	tm->super.callback = cb;
4276 	tm->super.cb_arg = cb_arg;
4277 	tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT,
4278 		comm_timer_callback, &tm->super);
4279 	if(tm->ev == NULL) {
4280 		log_err("timer_create: event_base_set failed.");
4281 		free(tm);
4282 		return NULL;
4283 	}
4284 	return &tm->super;
4285 }
4286 
4287 void
4288 comm_timer_disable(struct comm_timer* timer)
4289 {
4290 	if(!timer)
4291 		return;
4292 	ub_timer_del(timer->ev_timer->ev);
4293 	timer->ev_timer->enabled = 0;
4294 }
4295 
4296 void
4297 comm_timer_set(struct comm_timer* timer, struct timeval* tv)
4298 {
4299 	log_assert(tv);
4300 	if(timer->ev_timer->enabled)
4301 		comm_timer_disable(timer);
4302 	if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base,
4303 		comm_timer_callback, timer, tv) != 0)
4304 		log_err("comm_timer_set: evtimer_add failed.");
4305 	timer->ev_timer->enabled = 1;
4306 }
4307 
4308 void
4309 comm_timer_delete(struct comm_timer* timer)
4310 {
4311 	if(!timer)
4312 		return;
4313 	comm_timer_disable(timer);
4314 	/* Free the sub struct timer->ev_timer derived from the super struct timer.
4315 	 * i.e. assert(timer == timer->ev_timer)
4316 	 */
4317 	ub_event_free(timer->ev_timer->ev);
4318 	free(timer->ev_timer);
4319 }
4320 
4321 void
4322 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
4323 {
4324 	struct comm_timer* tm = (struct comm_timer*)arg;
4325 	if(!(event&UB_EV_TIMEOUT))
4326 		return;
4327 	ub_comm_base_now(tm->ev_timer->base);
4328 	tm->ev_timer->enabled = 0;
4329 	fptr_ok(fptr_whitelist_comm_timer(tm->callback));
4330 	(*tm->callback)(tm->cb_arg);
4331 }
4332 
4333 int
4334 comm_timer_is_set(struct comm_timer* timer)
4335 {
4336 	return (int)timer->ev_timer->enabled;
4337 }
4338 
4339 size_t
4340 comm_timer_get_mem(struct comm_timer* ATTR_UNUSED(timer))
4341 {
4342 	return sizeof(struct internal_timer);
4343 }
4344 
4345 struct comm_signal*
4346 comm_signal_create(struct comm_base* base,
4347         void (*callback)(int, void*), void* cb_arg)
4348 {
4349 	struct comm_signal* com = (struct comm_signal*)malloc(
4350 		sizeof(struct comm_signal));
4351 	if(!com) {
4352 		log_err("malloc failed");
4353 		return NULL;
4354 	}
4355 	com->base = base;
4356 	com->callback = callback;
4357 	com->cb_arg = cb_arg;
4358 	com->ev_signal = NULL;
4359 	return com;
4360 }
4361 
4362 void
4363 comm_signal_callback(int sig, short event, void* arg)
4364 {
4365 	struct comm_signal* comsig = (struct comm_signal*)arg;
4366 	if(!(event & UB_EV_SIGNAL))
4367 		return;
4368 	ub_comm_base_now(comsig->base);
4369 	fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
4370 	(*comsig->callback)(sig, comsig->cb_arg);
4371 }
4372 
4373 int
4374 comm_signal_bind(struct comm_signal* comsig, int sig)
4375 {
4376 	struct internal_signal* entry = (struct internal_signal*)calloc(1,
4377 		sizeof(struct internal_signal));
4378 	if(!entry) {
4379 		log_err("malloc failed");
4380 		return 0;
4381 	}
4382 	log_assert(comsig);
4383 	/* add signal event */
4384 	entry->ev = ub_signal_new(comsig->base->eb->base, sig,
4385 		comm_signal_callback, comsig);
4386 	if(entry->ev == NULL) {
4387 		log_err("Could not create signal event");
4388 		free(entry);
4389 		return 0;
4390 	}
4391 	if(ub_signal_add(entry->ev, NULL) != 0) {
4392 		log_err("Could not add signal handler");
4393 		ub_event_free(entry->ev);
4394 		free(entry);
4395 		return 0;
4396 	}
4397 	/* link into list */
4398 	entry->next = comsig->ev_signal;
4399 	comsig->ev_signal = entry;
4400 	return 1;
4401 }
4402 
4403 void
4404 comm_signal_delete(struct comm_signal* comsig)
4405 {
4406 	struct internal_signal* p, *np;
4407 	if(!comsig)
4408 		return;
4409 	p=comsig->ev_signal;
4410 	while(p) {
4411 		np = p->next;
4412 		ub_signal_del(p->ev);
4413 		ub_event_free(p->ev);
4414 		free(p);
4415 		p = np;
4416 	}
4417 	free(comsig);
4418 }
4419