xref: /openbsd-src/usr.sbin/unbound/util/netevent.c (revision 4b70baf6e17fc8b27fc1f7fa7929335753fa94c3)
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_OPENSSL_SSL_H
55 #include <openssl/ssl.h>
56 #endif
57 #ifdef HAVE_OPENSSL_ERR_H
58 #include <openssl/err.h>
59 #endif
60 
61 /* -------- Start of local definitions -------- */
62 /** if CMSG_ALIGN is not defined on this platform, a workaround */
63 #ifndef CMSG_ALIGN
64 #  ifdef __CMSG_ALIGN
65 #    define CMSG_ALIGN(n) __CMSG_ALIGN(n)
66 #  elif defined(CMSG_DATA_ALIGN)
67 #    define CMSG_ALIGN _CMSG_DATA_ALIGN
68 #  else
69 #    define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
70 #  endif
71 #endif
72 
73 /** if CMSG_LEN is not defined on this platform, a workaround */
74 #ifndef CMSG_LEN
75 #  define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
76 #endif
77 
78 /** if CMSG_SPACE is not defined on this platform, a workaround */
79 #ifndef CMSG_SPACE
80 #  ifdef _CMSG_HDR_ALIGN
81 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
82 #  else
83 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
84 #  endif
85 #endif
86 
87 /** The TCP writing query timeout in milliseconds */
88 #define TCP_QUERY_TIMEOUT 120000
89 /** The minimum actual TCP timeout to use, regardless of what we advertise,
90  * in msec */
91 #define TCP_QUERY_TIMEOUT_MINIMUM 200
92 
93 #ifndef NONBLOCKING_IS_BROKEN
94 /** number of UDP reads to perform per read indication from select */
95 #define NUM_UDP_PER_SELECT 100
96 #else
97 #define NUM_UDP_PER_SELECT 1
98 #endif
99 
100 /**
101  * The internal event structure for keeping ub_event info for the event.
102  * Possibly other structures (list, tree) this is part of.
103  */
104 struct internal_event {
105 	/** the comm base */
106 	struct comm_base* base;
107 	/** ub_event event type */
108 	struct ub_event* ev;
109 };
110 
111 /**
112  * Internal base structure, so that every thread has its own events.
113  */
114 struct internal_base {
115 	/** ub_event event_base type. */
116 	struct ub_event_base* base;
117 	/** seconds time pointer points here */
118 	time_t secs;
119 	/** timeval with current time */
120 	struct timeval now;
121 	/** the event used for slow_accept timeouts */
122 	struct ub_event* slow_accept;
123 	/** true if slow_accept is enabled */
124 	int slow_accept_enabled;
125 };
126 
127 /**
128  * Internal timer structure, to store timer event in.
129  */
130 struct internal_timer {
131 	/** the super struct from which derived */
132 	struct comm_timer super;
133 	/** the comm base */
134 	struct comm_base* base;
135 	/** ub_event event type */
136 	struct ub_event* ev;
137 	/** is timer enabled */
138 	uint8_t enabled;
139 };
140 
141 /**
142  * Internal signal structure, to store signal event in.
143  */
144 struct internal_signal {
145 	/** ub_event event type */
146 	struct ub_event* ev;
147 	/** next in signal list */
148 	struct internal_signal* next;
149 };
150 
151 /** create a tcp handler with a parent */
152 static struct comm_point* comm_point_create_tcp_handler(
153 	struct comm_base *base, struct comm_point* parent, size_t bufsize,
154 	struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
155 	void* callback_arg);
156 
157 /* -------- End of local definitions -------- */
158 
159 struct comm_base*
160 comm_base_create(int sigs)
161 {
162 	struct comm_base* b = (struct comm_base*)calloc(1,
163 		sizeof(struct comm_base));
164 	const char *evnm="event", *evsys="", *evmethod="";
165 
166 	if(!b)
167 		return NULL;
168 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
169 	if(!b->eb) {
170 		free(b);
171 		return NULL;
172 	}
173 	b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now);
174 	if(!b->eb->base) {
175 		free(b->eb);
176 		free(b);
177 		return NULL;
178 	}
179 	ub_comm_base_now(b);
180 	ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod);
181 	verbose(VERB_ALGO, "%s %s user %s method.", evnm, evsys, evmethod);
182 	return b;
183 }
184 
185 struct comm_base*
186 comm_base_create_event(struct ub_event_base* base)
187 {
188 	struct comm_base* b = (struct comm_base*)calloc(1,
189 		sizeof(struct comm_base));
190 	if(!b)
191 		return NULL;
192 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
193 	if(!b->eb) {
194 		free(b);
195 		return NULL;
196 	}
197 	b->eb->base = base;
198 	ub_comm_base_now(b);
199 	return b;
200 }
201 
202 void
203 comm_base_delete(struct comm_base* b)
204 {
205 	if(!b)
206 		return;
207 	if(b->eb->slow_accept_enabled) {
208 		if(ub_event_del(b->eb->slow_accept) != 0) {
209 			log_err("could not event_del slow_accept");
210 		}
211 		ub_event_free(b->eb->slow_accept);
212 	}
213 	ub_event_base_free(b->eb->base);
214 	b->eb->base = NULL;
215 	free(b->eb);
216 	free(b);
217 }
218 
219 void
220 comm_base_delete_no_base(struct comm_base* b)
221 {
222 	if(!b)
223 		return;
224 	if(b->eb->slow_accept_enabled) {
225 		if(ub_event_del(b->eb->slow_accept) != 0) {
226 			log_err("could not event_del slow_accept");
227 		}
228 		ub_event_free(b->eb->slow_accept);
229 	}
230 	b->eb->base = NULL;
231 	free(b->eb);
232 	free(b);
233 }
234 
235 void
236 comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv)
237 {
238 	*tt = &b->eb->secs;
239 	*tv = &b->eb->now;
240 }
241 
242 void
243 comm_base_dispatch(struct comm_base* b)
244 {
245 	int retval;
246 	retval = ub_event_base_dispatch(b->eb->base);
247 	if(retval < 0) {
248 		fatal_exit("event_dispatch returned error %d, "
249 			"errno is %s", retval, strerror(errno));
250 	}
251 }
252 
253 void comm_base_exit(struct comm_base* b)
254 {
255 	if(ub_event_base_loopexit(b->eb->base) != 0) {
256 		log_err("Could not loopexit");
257 	}
258 }
259 
260 void comm_base_set_slow_accept_handlers(struct comm_base* b,
261 	void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
262 {
263 	b->stop_accept = stop_acc;
264 	b->start_accept = start_acc;
265 	b->cb_arg = arg;
266 }
267 
268 struct ub_event_base* comm_base_internal(struct comm_base* b)
269 {
270 	return b->eb->base;
271 }
272 
273 /** see if errno for udp has to be logged or not uses globals */
274 static int
275 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
276 {
277 	/* do not log transient errors (unless high verbosity) */
278 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
279 	switch(errno) {
280 #  ifdef ENETUNREACH
281 		case ENETUNREACH:
282 #  endif
283 #  ifdef EHOSTDOWN
284 		case EHOSTDOWN:
285 #  endif
286 #  ifdef EHOSTUNREACH
287 		case EHOSTUNREACH:
288 #  endif
289 #  ifdef ENETDOWN
290 		case ENETDOWN:
291 #  endif
292 			if(verbosity < VERB_ALGO)
293 				return 0;
294 		default:
295 			break;
296 	}
297 #endif
298 	/* permission denied is gotten for every send if the
299 	 * network is disconnected (on some OS), squelch it */
300 	if( ((errno == EPERM)
301 #  ifdef EADDRNOTAVAIL
302 		/* 'Cannot assign requested address' also when disconnected */
303 		|| (errno == EADDRNOTAVAIL)
304 #  endif
305 		) && verbosity < VERB_DETAIL)
306 		return 0;
307 #  ifdef EADDRINUSE
308 	/* If SO_REUSEADDR is set, we could try to connect to the same server
309 	 * from the same source port twice. */
310 	if(errno == EADDRINUSE && verbosity < VERB_DETAIL)
311 		return 0;
312 #  endif
313 	/* squelch errors where people deploy AAAA ::ffff:bla for
314 	 * authority servers, which we try for intranets. */
315 	if(errno == EINVAL && addr_is_ip4mapped(
316 		(struct sockaddr_storage*)addr, addrlen) &&
317 		verbosity < VERB_DETAIL)
318 		return 0;
319 	/* SO_BROADCAST sockopt can give access to 255.255.255.255,
320 	 * but a dns cache does not need it. */
321 	if(errno == EACCES && addr_is_broadcast(
322 		(struct sockaddr_storage*)addr, addrlen) &&
323 		verbosity < VERB_DETAIL)
324 		return 0;
325 	return 1;
326 }
327 
328 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
329 {
330 	return udp_send_errno_needs_log(addr, addrlen);
331 }
332 
333 /* send a UDP reply */
334 int
335 comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet,
336 	struct sockaddr* addr, socklen_t addrlen)
337 {
338 	ssize_t sent;
339 	log_assert(c->fd != -1);
340 #ifdef UNBOUND_DEBUG
341 	if(sldns_buffer_remaining(packet) == 0)
342 		log_err("error: send empty UDP packet");
343 #endif
344 	log_assert(addr && addrlen > 0);
345 	sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
346 		sldns_buffer_remaining(packet), 0,
347 		addr, addrlen);
348 	if(sent == -1) {
349 		/* try again and block, waiting for IO to complete,
350 		 * we want to send the answer, and we will wait for
351 		 * the ethernet interface buffer to have space. */
352 #ifndef USE_WINSOCK
353 		if(errno == EAGAIN ||
354 #  ifdef EWOULDBLOCK
355 			errno == EWOULDBLOCK ||
356 #  endif
357 			errno == ENOBUFS) {
358 #else
359 		if(WSAGetLastError() == WSAEINPROGRESS ||
360 			WSAGetLastError() == WSAENOBUFS ||
361 			WSAGetLastError() == WSAEWOULDBLOCK) {
362 #endif
363 			int e;
364 			fd_set_block(c->fd);
365 			sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
366 				sldns_buffer_remaining(packet), 0,
367 				addr, addrlen);
368 			e = errno;
369 			fd_set_nonblock(c->fd);
370 			errno = e;
371 		}
372 	}
373 	if(sent == -1) {
374 		if(!udp_send_errno_needs_log(addr, addrlen))
375 			return 0;
376 #ifndef USE_WINSOCK
377 		verbose(VERB_OPS, "sendto failed: %s", strerror(errno));
378 #else
379 		verbose(VERB_OPS, "sendto failed: %s",
380 			wsa_strerror(WSAGetLastError()));
381 #endif
382 		log_addr(VERB_OPS, "remote address is",
383 			(struct sockaddr_storage*)addr, addrlen);
384 		return 0;
385 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
386 		log_err("sent %d in place of %d bytes",
387 			(int)sent, (int)sldns_buffer_remaining(packet));
388 		return 0;
389 	}
390 	return 1;
391 }
392 
393 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
394 /** print debug ancillary info */
395 static void p_ancil(const char* str, struct comm_reply* r)
396 {
397 	if(r->srctype != 4 && r->srctype != 6) {
398 		log_info("%s: unknown srctype %d", str, r->srctype);
399 		return;
400 	}
401 	if(r->srctype == 6) {
402 		char buf[1024];
403 		if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
404 			buf, (socklen_t)sizeof(buf)) == 0) {
405 			(void)strlcpy(buf, "(inet_ntop error)", sizeof(buf));
406 		}
407 		buf[sizeof(buf)-1]=0;
408 		log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
409 	} else if(r->srctype == 4) {
410 #ifdef IP_PKTINFO
411 		char buf1[1024], buf2[1024];
412 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
413 			buf1, (socklen_t)sizeof(buf1)) == 0) {
414 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
415 		}
416 		buf1[sizeof(buf1)-1]=0;
417 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
418 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
419 			buf2, (socklen_t)sizeof(buf2)) == 0) {
420 			(void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2));
421 		}
422 		buf2[sizeof(buf2)-1]=0;
423 #else
424 		buf2[0]=0;
425 #endif
426 		log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
427 			buf1, buf2);
428 #elif defined(IP_RECVDSTADDR)
429 		char buf1[1024];
430 		if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
431 			buf1, (socklen_t)sizeof(buf1)) == 0) {
432 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
433 		}
434 		buf1[sizeof(buf1)-1]=0;
435 		log_info("%s: %s", str, buf1);
436 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */
437 	}
438 }
439 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */
440 
441 /** send a UDP reply over specified interface*/
442 static int
443 comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet,
444 	struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
445 {
446 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
447 	ssize_t sent;
448 	struct msghdr msg;
449 	struct iovec iov[1];
450 	char control[256];
451 #ifndef S_SPLINT_S
452 	struct cmsghdr *cmsg;
453 #endif /* S_SPLINT_S */
454 
455 	log_assert(c->fd != -1);
456 #ifdef UNBOUND_DEBUG
457 	if(sldns_buffer_remaining(packet) == 0)
458 		log_err("error: send empty UDP packet");
459 #endif
460 	log_assert(addr && addrlen > 0);
461 
462 	msg.msg_name = addr;
463 	msg.msg_namelen = addrlen;
464 	iov[0].iov_base = sldns_buffer_begin(packet);
465 	iov[0].iov_len = sldns_buffer_remaining(packet);
466 	msg.msg_iov = iov;
467 	msg.msg_iovlen = 1;
468 	msg.msg_control = control;
469 #ifndef S_SPLINT_S
470 	msg.msg_controllen = sizeof(control);
471 #endif /* S_SPLINT_S */
472 	msg.msg_flags = 0;
473 
474 #ifndef S_SPLINT_S
475 	cmsg = CMSG_FIRSTHDR(&msg);
476 	if(r->srctype == 4) {
477 #ifdef IP_PKTINFO
478 		void* cmsg_data;
479 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
480 		log_assert(msg.msg_controllen <= sizeof(control));
481 		cmsg->cmsg_level = IPPROTO_IP;
482 		cmsg->cmsg_type = IP_PKTINFO;
483 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
484 			sizeof(struct in_pktinfo));
485 		/* unset the ifindex to not bypass the routing tables */
486 		cmsg_data = CMSG_DATA(cmsg);
487 		((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0;
488 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
489 #elif defined(IP_SENDSRCADDR)
490 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
491 		log_assert(msg.msg_controllen <= sizeof(control));
492 		cmsg->cmsg_level = IPPROTO_IP;
493 		cmsg->cmsg_type = IP_SENDSRCADDR;
494 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
495 			sizeof(struct in_addr));
496 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
497 #else
498 		verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
499 		msg.msg_control = NULL;
500 #endif /* IP_PKTINFO or IP_SENDSRCADDR */
501 	} else if(r->srctype == 6) {
502 		void* cmsg_data;
503 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
504 		log_assert(msg.msg_controllen <= sizeof(control));
505 		cmsg->cmsg_level = IPPROTO_IPV6;
506 		cmsg->cmsg_type = IPV6_PKTINFO;
507 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
508 			sizeof(struct in6_pktinfo));
509 		/* unset the ifindex to not bypass the routing tables */
510 		cmsg_data = CMSG_DATA(cmsg);
511 		((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0;
512 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
513 	} else {
514 		/* try to pass all 0 to use default route */
515 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
516 		log_assert(msg.msg_controllen <= sizeof(control));
517 		cmsg->cmsg_level = IPPROTO_IPV6;
518 		cmsg->cmsg_type = IPV6_PKTINFO;
519 		memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
520 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
521 	}
522 #endif /* S_SPLINT_S */
523 	if(verbosity >= VERB_ALGO)
524 		p_ancil("send_udp over interface", r);
525 	sent = sendmsg(c->fd, &msg, 0);
526 	if(sent == -1) {
527 		/* try again and block, waiting for IO to complete,
528 		 * we want to send the answer, and we will wait for
529 		 * the ethernet interface buffer to have space. */
530 #ifndef USE_WINSOCK
531 		if(errno == EAGAIN ||
532 #  ifdef EWOULDBLOCK
533 			errno == EWOULDBLOCK ||
534 #  endif
535 			errno == ENOBUFS) {
536 #else
537 		if(WSAGetLastError() == WSAEINPROGRESS ||
538 			WSAGetLastError() == WSAENOBUFS ||
539 			WSAGetLastError() == WSAEWOULDBLOCK) {
540 #endif
541 			int e;
542 			fd_set_block(c->fd);
543 			sent = sendmsg(c->fd, &msg, 0);
544 			e = errno;
545 			fd_set_nonblock(c->fd);
546 			errno = e;
547 		}
548 	}
549 	if(sent == -1) {
550 		if(!udp_send_errno_needs_log(addr, addrlen))
551 			return 0;
552 		verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
553 		log_addr(VERB_OPS, "remote address is",
554 			(struct sockaddr_storage*)addr, addrlen);
555 #ifdef __NetBSD__
556 		/* netbsd 7 has IP_PKTINFO for recv but not send */
557 		if(errno == EINVAL && r->srctype == 4)
558 			log_err("sendmsg: No support for sendmsg(IP_PKTINFO). "
559 				"Please disable interface-automatic");
560 #endif
561 		return 0;
562 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
563 		log_err("sent %d in place of %d bytes",
564 			(int)sent, (int)sldns_buffer_remaining(packet));
565 		return 0;
566 	}
567 	return 1;
568 #else
569 	(void)c;
570 	(void)packet;
571 	(void)addr;
572 	(void)addrlen;
573 	(void)r;
574 	log_err("sendmsg: IPV6_PKTINFO not supported");
575 	return 0;
576 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
577 }
578 
579 void
580 comm_point_udp_ancil_callback(int fd, short event, void* arg)
581 {
582 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
583 	struct comm_reply rep;
584 	struct msghdr msg;
585 	struct iovec iov[1];
586 	ssize_t rcv;
587 	char ancil[256];
588 	int i;
589 #ifndef S_SPLINT_S
590 	struct cmsghdr* cmsg;
591 #endif /* S_SPLINT_S */
592 
593 	rep.c = (struct comm_point*)arg;
594 	log_assert(rep.c->type == comm_udp);
595 
596 	if(!(event&UB_EV_READ))
597 		return;
598 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
599 	ub_comm_base_now(rep.c->ev->base);
600 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
601 		sldns_buffer_clear(rep.c->buffer);
602 		rep.addrlen = (socklen_t)sizeof(rep.addr);
603 		log_assert(fd != -1);
604 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
605 		msg.msg_name = &rep.addr;
606 		msg.msg_namelen = (socklen_t)sizeof(rep.addr);
607 		iov[0].iov_base = sldns_buffer_begin(rep.c->buffer);
608 		iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer);
609 		msg.msg_iov = iov;
610 		msg.msg_iovlen = 1;
611 		msg.msg_control = ancil;
612 #ifndef S_SPLINT_S
613 		msg.msg_controllen = sizeof(ancil);
614 #endif /* S_SPLINT_S */
615 		msg.msg_flags = 0;
616 		rcv = recvmsg(fd, &msg, 0);
617 		if(rcv == -1) {
618 			if(errno != EAGAIN && errno != EINTR) {
619 				log_err("recvmsg failed: %s", strerror(errno));
620 			}
621 			return;
622 		}
623 		rep.addrlen = msg.msg_namelen;
624 		sldns_buffer_skip(rep.c->buffer, rcv);
625 		sldns_buffer_flip(rep.c->buffer);
626 		rep.srctype = 0;
627 #ifndef S_SPLINT_S
628 		for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
629 			cmsg = CMSG_NXTHDR(&msg, cmsg)) {
630 			if( cmsg->cmsg_level == IPPROTO_IPV6 &&
631 				cmsg->cmsg_type == IPV6_PKTINFO) {
632 				rep.srctype = 6;
633 				memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
634 					sizeof(struct in6_pktinfo));
635 				break;
636 #ifdef IP_PKTINFO
637 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
638 				cmsg->cmsg_type == IP_PKTINFO) {
639 				rep.srctype = 4;
640 				memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
641 					sizeof(struct in_pktinfo));
642 				break;
643 #elif defined(IP_RECVDSTADDR)
644 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
645 				cmsg->cmsg_type == IP_RECVDSTADDR) {
646 				rep.srctype = 4;
647 				memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
648 					sizeof(struct in_addr));
649 				break;
650 #endif /* IP_PKTINFO or IP_RECVDSTADDR */
651 			}
652 		}
653 		if(verbosity >= VERB_ALGO)
654 			p_ancil("receive_udp on interface", &rep);
655 #endif /* S_SPLINT_S */
656 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
657 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
658 			/* send back immediate reply */
659 			(void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer,
660 				(struct sockaddr*)&rep.addr, rep.addrlen, &rep);
661 		}
662 		if(!rep.c || rep.c->fd == -1) /* commpoint closed */
663 			break;
664 	}
665 #else
666 	(void)fd;
667 	(void)event;
668 	(void)arg;
669 	fatal_exit("recvmsg: No support for IPV6_PKTINFO; IP_PKTINFO or IP_RECVDSTADDR. "
670 		"Please disable interface-automatic");
671 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */
672 }
673 
674 void
675 comm_point_udp_callback(int fd, short event, void* arg)
676 {
677 	struct comm_reply rep;
678 	ssize_t rcv;
679 	int i;
680 	struct sldns_buffer *buffer;
681 
682 	rep.c = (struct comm_point*)arg;
683 	log_assert(rep.c->type == comm_udp);
684 
685 	if(!(event&UB_EV_READ))
686 		return;
687 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
688 	ub_comm_base_now(rep.c->ev->base);
689 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
690 		sldns_buffer_clear(rep.c->buffer);
691 		rep.addrlen = (socklen_t)sizeof(rep.addr);
692 		log_assert(fd != -1);
693 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
694 		rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer),
695 			sldns_buffer_remaining(rep.c->buffer), 0,
696 			(struct sockaddr*)&rep.addr, &rep.addrlen);
697 		if(rcv == -1) {
698 #ifndef USE_WINSOCK
699 			if(errno != EAGAIN && errno != EINTR)
700 				log_err("recvfrom %d failed: %s",
701 					fd, strerror(errno));
702 #else
703 			if(WSAGetLastError() != WSAEINPROGRESS &&
704 				WSAGetLastError() != WSAECONNRESET &&
705 				WSAGetLastError()!= WSAEWOULDBLOCK)
706 				log_err("recvfrom failed: %s",
707 					wsa_strerror(WSAGetLastError()));
708 #endif
709 			return;
710 		}
711 		sldns_buffer_skip(rep.c->buffer, rcv);
712 		sldns_buffer_flip(rep.c->buffer);
713 		rep.srctype = 0;
714 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
715 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
716 			/* send back immediate reply */
717 #ifdef USE_DNSCRYPT
718 			buffer = rep.c->dnscrypt_buffer;
719 #else
720 			buffer = rep.c->buffer;
721 #endif
722 			(void)comm_point_send_udp_msg(rep.c, buffer,
723 				(struct sockaddr*)&rep.addr, rep.addrlen);
724 		}
725 		if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for
726 		another UDP port. Note rep.c cannot be reused with TCP fd. */
727 			break;
728 	}
729 }
730 
731 /** Use a new tcp handler for new query fd, set to read query */
732 static void
733 setup_tcp_handler(struct comm_point* c, int fd, int cur, int max)
734 {
735 	int handler_usage;
736 	log_assert(c->type == comm_tcp);
737 	log_assert(c->fd == -1);
738 	sldns_buffer_clear(c->buffer);
739 #ifdef USE_DNSCRYPT
740 	if (c->dnscrypt)
741 		sldns_buffer_clear(c->dnscrypt_buffer);
742 #endif
743 	c->tcp_is_reading = 1;
744 	c->tcp_byte_count = 0;
745 	/* if more than half the tcp handlers are in use, use a shorter
746 	 * timeout for this TCP connection, we need to make space for
747 	 * other connections to be able to get attention */
748 	/* If > 50% TCP handler structures in use, set timeout to 1/100th
749 	 * 	configured value.
750 	 * If > 65%TCP handler structures in use, set to 1/500th configured
751 	 * 	value.
752 	 * If > 80% TCP handler structures in use, set to 0.
753 	 *
754 	 * If the timeout to use falls below 200 milliseconds, an actual
755 	 * timeout of 200ms is used.
756 	 */
757 	handler_usage = (cur * 100) / max;
758 	if(handler_usage > 50 && handler_usage <= 65)
759 		c->tcp_timeout_msec /= 100;
760 	else if (handler_usage > 65 && handler_usage <= 80)
761 		c->tcp_timeout_msec /= 500;
762 	else if (handler_usage > 80)
763 		c->tcp_timeout_msec = 0;
764 	comm_point_start_listening(c, fd,
765 		c->tcp_timeout_msec < TCP_QUERY_TIMEOUT_MINIMUM
766 			? TCP_QUERY_TIMEOUT_MINIMUM
767 			: c->tcp_timeout_msec);
768 }
769 
770 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
771 	short ATTR_UNUSED(event), void* arg)
772 {
773 	struct comm_base* b = (struct comm_base*)arg;
774 	/* timeout for the slow accept, re-enable accepts again */
775 	if(b->start_accept) {
776 		verbose(VERB_ALGO, "wait is over, slow accept disabled");
777 		fptr_ok(fptr_whitelist_start_accept(b->start_accept));
778 		(*b->start_accept)(b->cb_arg);
779 		b->eb->slow_accept_enabled = 0;
780 	}
781 }
782 
783 int comm_point_perform_accept(struct comm_point* c,
784 	struct sockaddr_storage* addr, socklen_t* addrlen)
785 {
786 	int new_fd;
787 	*addrlen = (socklen_t)sizeof(*addr);
788 #ifndef HAVE_ACCEPT4
789 	new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
790 #else
791 	/* SOCK_NONBLOCK saves extra calls to fcntl for the same result */
792 	new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK);
793 #endif
794 	if(new_fd == -1) {
795 #ifndef USE_WINSOCK
796 		/* EINTR is signal interrupt. others are closed connection. */
797 		if(	errno == EINTR || errno == EAGAIN
798 #ifdef EWOULDBLOCK
799 			|| errno == EWOULDBLOCK
800 #endif
801 #ifdef ECONNABORTED
802 			|| errno == ECONNABORTED
803 #endif
804 #ifdef EPROTO
805 			|| errno == EPROTO
806 #endif /* EPROTO */
807 			)
808 			return -1;
809 #if defined(ENFILE) && defined(EMFILE)
810 		if(errno == ENFILE || errno == EMFILE) {
811 			/* out of file descriptors, likely outside of our
812 			 * control. stop accept() calls for some time */
813 			if(c->ev->base->stop_accept) {
814 				struct comm_base* b = c->ev->base;
815 				struct timeval tv;
816 				verbose(VERB_ALGO, "out of file descriptors: "
817 					"slow accept");
818 				b->eb->slow_accept_enabled = 1;
819 				fptr_ok(fptr_whitelist_stop_accept(
820 					b->stop_accept));
821 				(*b->stop_accept)(b->cb_arg);
822 				/* set timeout, no mallocs */
823 				tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
824 				tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000;
825 				b->eb->slow_accept = ub_event_new(b->eb->base,
826 					-1, UB_EV_TIMEOUT,
827 					comm_base_handle_slow_accept, b);
828 				if(b->eb->slow_accept == NULL) {
829 					/* we do not want to log here, because
830 					 * that would spam the logfiles.
831 					 * error: "event_base_set failed." */
832 				}
833 				else if(ub_event_add(b->eb->slow_accept, &tv)
834 					!= 0) {
835 					/* we do not want to log here,
836 					 * error: "event_add failed." */
837 				}
838 			}
839 			return -1;
840 		}
841 #endif
842 		log_err_addr("accept failed", strerror(errno), addr, *addrlen);
843 #else /* USE_WINSOCK */
844 		if(WSAGetLastError() == WSAEINPROGRESS ||
845 			WSAGetLastError() == WSAECONNRESET)
846 			return -1;
847 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
848 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
849 			return -1;
850 		}
851 		log_err_addr("accept failed", wsa_strerror(WSAGetLastError()),
852 			addr, *addrlen);
853 #endif
854 		return -1;
855 	}
856 	if(c->tcp_conn_limit && c->type == comm_tcp_accept) {
857 		c->tcl_addr = tcl_addr_lookup(c->tcp_conn_limit, addr, *addrlen);
858 		if(!tcl_new_connection(c->tcl_addr)) {
859 			if(verbosity >= 3)
860 				log_err_addr("accept rejected",
861 				"connection limit exceeded", addr, *addrlen);
862 			close(new_fd);
863 			return -1;
864 		}
865 	}
866 #ifndef HAVE_ACCEPT4
867 	fd_set_nonblock(new_fd);
868 #endif
869 	return new_fd;
870 }
871 
872 #ifdef USE_WINSOCK
873 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
874         int ATTR_UNUSED(argi), long argl, long retvalue)
875 {
876 	int wsa_err = WSAGetLastError(); /* store errcode before it is gone */
877 	verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
878 		(oper&BIO_CB_RETURN)?"return":"before",
879 		(oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
880 		wsa_err==WSAEWOULDBLOCK?"wsawb":"");
881 	/* on windows, check if previous operation caused EWOULDBLOCK */
882 	if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
883 		(oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
884 		if(wsa_err == WSAEWOULDBLOCK)
885 			ub_winsock_tcp_wouldblock((struct ub_event*)
886 				BIO_get_callback_arg(b), UB_EV_READ);
887 	}
888 	if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
889 		(oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
890 		if(wsa_err == WSAEWOULDBLOCK)
891 			ub_winsock_tcp_wouldblock((struct ub_event*)
892 				BIO_get_callback_arg(b), UB_EV_WRITE);
893 	}
894 	/* return original return value */
895 	return retvalue;
896 }
897 
898 /** set win bio callbacks for nonblocking operations */
899 void
900 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
901 {
902 	SSL* ssl = (SSL*)thessl;
903 	/* set them both just in case, but usually they are the same BIO */
904 	BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
905 	BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev);
906 	BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
907 	BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev);
908 }
909 #endif
910 
911 void
912 comm_point_tcp_accept_callback(int fd, short event, void* arg)
913 {
914 	struct comm_point* c = (struct comm_point*)arg, *c_hdl;
915 	int new_fd;
916 	log_assert(c->type == comm_tcp_accept);
917 	if(!(event & UB_EV_READ)) {
918 		log_info("ignoring tcp accept event %d", (int)event);
919 		return;
920 	}
921 	ub_comm_base_now(c->ev->base);
922 	/* find free tcp handler. */
923 	if(!c->tcp_free) {
924 		log_warn("accepted too many tcp, connections full");
925 		return;
926 	}
927 	/* accept incoming connection. */
928 	c_hdl = c->tcp_free;
929 	log_assert(fd != -1);
930 	(void)fd;
931 	new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr,
932 		&c_hdl->repinfo.addrlen);
933 	if(new_fd == -1)
934 		return;
935 	if(c->ssl) {
936 		c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
937 		if(!c_hdl->ssl) {
938 			c_hdl->fd = new_fd;
939 			comm_point_close(c_hdl);
940 			return;
941 		}
942 		c_hdl->ssl_shake_state = comm_ssl_shake_read;
943 #ifdef USE_WINSOCK
944 		comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
945 #endif
946 	}
947 
948 	/* grab the tcp handler buffers */
949 	c->cur_tcp_count++;
950 	c->tcp_free = c_hdl->tcp_free;
951 	if(!c->tcp_free) {
952 		/* stop accepting incoming queries for now. */
953 		comm_point_stop_listening(c);
954 	}
955 	setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count);
956 }
957 
958 /** Make tcp handler free for next assignment */
959 static void
960 reclaim_tcp_handler(struct comm_point* c)
961 {
962 	log_assert(c->type == comm_tcp);
963 	if(c->ssl) {
964 #ifdef HAVE_SSL
965 		SSL_shutdown(c->ssl);
966 		SSL_free(c->ssl);
967 		c->ssl = NULL;
968 #endif
969 	}
970 	comm_point_close(c);
971 	if(c->tcp_parent) {
972 		c->tcp_parent->cur_tcp_count--;
973 		c->tcp_free = c->tcp_parent->tcp_free;
974 		c->tcp_parent->tcp_free = c;
975 		if(!c->tcp_free) {
976 			/* re-enable listening on accept socket */
977 			comm_point_start_listening(c->tcp_parent, -1, -1);
978 		}
979 	}
980 }
981 
982 /** do the callback when writing is done */
983 static void
984 tcp_callback_writer(struct comm_point* c)
985 {
986 	log_assert(c->type == comm_tcp);
987 	sldns_buffer_clear(c->buffer);
988 	if(c->tcp_do_toggle_rw)
989 		c->tcp_is_reading = 1;
990 	c->tcp_byte_count = 0;
991 	/* switch from listening(write) to listening(read) */
992 	if(c->tcp_req_info) {
993 		tcp_req_info_handle_writedone(c->tcp_req_info);
994 	} else {
995 		comm_point_stop_listening(c);
996 		comm_point_start_listening(c, -1, -1);
997 	}
998 }
999 
1000 /** do the callback when reading is done */
1001 static void
1002 tcp_callback_reader(struct comm_point* c)
1003 {
1004 	log_assert(c->type == comm_tcp || c->type == comm_local);
1005 	sldns_buffer_flip(c->buffer);
1006 	if(c->tcp_do_toggle_rw)
1007 		c->tcp_is_reading = 0;
1008 	c->tcp_byte_count = 0;
1009 	if(c->tcp_req_info) {
1010 		tcp_req_info_handle_readdone(c->tcp_req_info);
1011 	} else {
1012 		if(c->type == comm_tcp)
1013 			comm_point_stop_listening(c);
1014 		fptr_ok(fptr_whitelist_comm_point(c->callback));
1015 		if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
1016 			comm_point_start_listening(c, -1, c->tcp_timeout_msec);
1017 		}
1018 	}
1019 }
1020 
1021 #ifdef HAVE_SSL
1022 /** log certificate details */
1023 static void
1024 log_cert(unsigned level, const char* str, X509* cert)
1025 {
1026 	BIO* bio;
1027 	char nul = 0;
1028 	char* pp = NULL;
1029 	long len;
1030 	if(verbosity < level) return;
1031 	bio = BIO_new(BIO_s_mem());
1032 	if(!bio) return;
1033 	X509_print_ex(bio, cert, 0, (unsigned long)-1
1034 		^(X509_FLAG_NO_SUBJECT
1035                         |X509_FLAG_NO_ISSUER|X509_FLAG_NO_VALIDITY
1036 			|X509_FLAG_NO_EXTENSIONS|X509_FLAG_NO_AUX
1037 			|X509_FLAG_NO_ATTRIBUTES));
1038 	BIO_write(bio, &nul, (int)sizeof(nul));
1039 	len = BIO_get_mem_data(bio, &pp);
1040 	if(len != 0 && pp) {
1041 		verbose(level, "%s: \n%s", str, pp);
1042 	}
1043 	BIO_free(bio);
1044 }
1045 #endif /* HAVE_SSL */
1046 
1047 /** continue ssl handshake */
1048 #ifdef HAVE_SSL
1049 static int
1050 ssl_handshake(struct comm_point* c)
1051 {
1052 	int r;
1053 	if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
1054 		/* read condition satisfied back to writing */
1055 		comm_point_listen_for_rw(c, 1, 1);
1056 		c->ssl_shake_state = comm_ssl_shake_none;
1057 		return 1;
1058 	}
1059 	if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
1060 		/* write condition satisfied, back to reading */
1061 		comm_point_listen_for_rw(c, 1, 0);
1062 		c->ssl_shake_state = comm_ssl_shake_none;
1063 		return 1;
1064 	}
1065 
1066 	ERR_clear_error();
1067 	r = SSL_do_handshake(c->ssl);
1068 	if(r != 1) {
1069 		int want = SSL_get_error(c->ssl, r);
1070 		if(want == SSL_ERROR_WANT_READ) {
1071 			if(c->ssl_shake_state == comm_ssl_shake_read)
1072 				return 1;
1073 			c->ssl_shake_state = comm_ssl_shake_read;
1074 			comm_point_listen_for_rw(c, 1, 0);
1075 			return 1;
1076 		} else if(want == SSL_ERROR_WANT_WRITE) {
1077 			if(c->ssl_shake_state == comm_ssl_shake_write)
1078 				return 1;
1079 			c->ssl_shake_state = comm_ssl_shake_write;
1080 			comm_point_listen_for_rw(c, 0, 1);
1081 			return 1;
1082 		} else if(r == 0) {
1083 			return 0; /* closed */
1084 		} else if(want == SSL_ERROR_SYSCALL) {
1085 			/* SYSCALL and errno==0 means closed uncleanly */
1086 			if(errno != 0)
1087 				log_err("SSL_handshake syscall: %s",
1088 					strerror(errno));
1089 			return 0;
1090 		} else {
1091 			log_crypto_err("ssl handshake failed");
1092 			log_addr(1, "ssl handshake failed", &c->repinfo.addr,
1093 				c->repinfo.addrlen);
1094 			return 0;
1095 		}
1096 	}
1097 	/* this is where peer verification could take place */
1098 	if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) {
1099 		/* verification */
1100 		if(SSL_get_verify_result(c->ssl) == X509_V_OK) {
1101 			X509* x = SSL_get_peer_certificate(c->ssl);
1102 			if(!x) {
1103 				log_addr(VERB_ALGO, "SSL connection failed: "
1104 					"no certificate",
1105 					&c->repinfo.addr, c->repinfo.addrlen);
1106 				return 0;
1107 			}
1108 			log_cert(VERB_ALGO, "peer certificate", x);
1109 #ifdef HAVE_SSL_GET0_PEERNAME
1110 			if(SSL_get0_peername(c->ssl)) {
1111 				char buf[255];
1112 				snprintf(buf, sizeof(buf), "SSL connection "
1113 					"to %s authenticated",
1114 					SSL_get0_peername(c->ssl));
1115 				log_addr(VERB_ALGO, buf, &c->repinfo.addr,
1116 					c->repinfo.addrlen);
1117 			} else {
1118 #endif
1119 				log_addr(VERB_ALGO, "SSL connection "
1120 					"authenticated", &c->repinfo.addr,
1121 					c->repinfo.addrlen);
1122 #ifdef HAVE_SSL_GET0_PEERNAME
1123 			}
1124 #endif
1125 			X509_free(x);
1126 		} else {
1127 			X509* x = SSL_get_peer_certificate(c->ssl);
1128 			if(x) {
1129 				log_cert(VERB_ALGO, "peer certificate", x);
1130 				X509_free(x);
1131 			}
1132 			log_addr(VERB_ALGO, "SSL connection failed: "
1133 				"failed to authenticate",
1134 				&c->repinfo.addr, c->repinfo.addrlen);
1135 			return 0;
1136 		}
1137 	} else {
1138 		/* unauthenticated, the verify peer flag was not set
1139 		 * in c->ssl when the ssl object was created from ssl_ctx */
1140 		log_addr(VERB_ALGO, "SSL connection", &c->repinfo.addr,
1141 			c->repinfo.addrlen);
1142 	}
1143 
1144 	/* setup listen rw correctly */
1145 	if(c->tcp_is_reading) {
1146 		if(c->ssl_shake_state != comm_ssl_shake_read)
1147 			comm_point_listen_for_rw(c, 1, 0);
1148 	} else {
1149 		comm_point_listen_for_rw(c, 1, 1);
1150 	}
1151 	c->ssl_shake_state = comm_ssl_shake_none;
1152 	return 1;
1153 }
1154 #endif /* HAVE_SSL */
1155 
1156 /** ssl read callback on TCP */
1157 static int
1158 ssl_handle_read(struct comm_point* c)
1159 {
1160 #ifdef HAVE_SSL
1161 	int r;
1162 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1163 		if(!ssl_handshake(c))
1164 			return 0;
1165 		if(c->ssl_shake_state != comm_ssl_shake_none)
1166 			return 1;
1167 	}
1168 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1169 		/* read length bytes */
1170 		ERR_clear_error();
1171 		if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer,
1172 			c->tcp_byte_count), (int)(sizeof(uint16_t) -
1173 			c->tcp_byte_count))) <= 0) {
1174 			int want = SSL_get_error(c->ssl, r);
1175 			if(want == SSL_ERROR_ZERO_RETURN) {
1176 				if(c->tcp_req_info)
1177 					return tcp_req_info_handle_read_close(c->tcp_req_info);
1178 				return 0; /* shutdown, closed */
1179 			} else if(want == SSL_ERROR_WANT_READ) {
1180 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1181 				return 1; /* read more later */
1182 			} else if(want == SSL_ERROR_WANT_WRITE) {
1183 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1184 				comm_point_listen_for_rw(c, 0, 1);
1185 				return 1;
1186 			} else if(want == SSL_ERROR_SYSCALL) {
1187 				if(errno != 0)
1188 					log_err("SSL_read syscall: %s",
1189 						strerror(errno));
1190 				return 0;
1191 			}
1192 			log_crypto_err("could not SSL_read");
1193 			return 0;
1194 		}
1195 		c->tcp_byte_count += r;
1196 		if(c->tcp_byte_count < sizeof(uint16_t))
1197 			return 1;
1198 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1199 			sldns_buffer_capacity(c->buffer)) {
1200 			verbose(VERB_QUERY, "ssl: dropped larger than buffer");
1201 			return 0;
1202 		}
1203 		sldns_buffer_set_limit(c->buffer,
1204 			sldns_buffer_read_u16_at(c->buffer, 0));
1205 		if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1206 			verbose(VERB_QUERY, "ssl: dropped bogus too short.");
1207 			return 0;
1208 		}
1209 		sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t)));
1210 		verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
1211 			(int)sldns_buffer_limit(c->buffer));
1212 	}
1213 	if(sldns_buffer_remaining(c->buffer) > 0) {
1214 		ERR_clear_error();
1215 		r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1216 			(int)sldns_buffer_remaining(c->buffer));
1217 		if(r <= 0) {
1218 			int want = SSL_get_error(c->ssl, r);
1219 			if(want == SSL_ERROR_ZERO_RETURN) {
1220 				if(c->tcp_req_info)
1221 					return tcp_req_info_handle_read_close(c->tcp_req_info);
1222 				return 0; /* shutdown, closed */
1223 			} else if(want == SSL_ERROR_WANT_READ) {
1224 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1225 				return 1; /* read more later */
1226 			} else if(want == SSL_ERROR_WANT_WRITE) {
1227 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1228 				comm_point_listen_for_rw(c, 0, 1);
1229 				return 1;
1230 			} else if(want == SSL_ERROR_SYSCALL) {
1231 				if(errno != 0)
1232 					log_err("SSL_read syscall: %s",
1233 						strerror(errno));
1234 				return 0;
1235 			}
1236 			log_crypto_err("could not SSL_read");
1237 			return 0;
1238 		}
1239 		sldns_buffer_skip(c->buffer, (ssize_t)r);
1240 	}
1241 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1242 		tcp_callback_reader(c);
1243 	}
1244 	return 1;
1245 #else
1246 	(void)c;
1247 	return 0;
1248 #endif /* HAVE_SSL */
1249 }
1250 
1251 /** ssl write callback on TCP */
1252 static int
1253 ssl_handle_write(struct comm_point* c)
1254 {
1255 #ifdef HAVE_SSL
1256 	int r;
1257 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1258 		if(!ssl_handshake(c))
1259 			return 0;
1260 		if(c->ssl_shake_state != comm_ssl_shake_none)
1261 			return 1;
1262 	}
1263 	/* ignore return, if fails we may simply block */
1264 	(void)SSL_set_mode(c->ssl, SSL_MODE_ENABLE_PARTIAL_WRITE);
1265 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1266 		uint16_t len = htons(sldns_buffer_limit(c->buffer));
1267 		ERR_clear_error();
1268 		if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) <
1269 			LDNS_RR_BUF_SIZE) {
1270 			/* combine the tcp length and the query for write,
1271 			 * this emulates writev */
1272 			uint8_t buf[LDNS_RR_BUF_SIZE];
1273 			memmove(buf, &len, sizeof(uint16_t));
1274 			memmove(buf+sizeof(uint16_t),
1275 				sldns_buffer_current(c->buffer),
1276 				sldns_buffer_remaining(c->buffer));
1277 			r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count),
1278 				(int)(sizeof(uint16_t)+
1279 				sldns_buffer_remaining(c->buffer)
1280 				- c->tcp_byte_count));
1281 		} else {
1282 			r = SSL_write(c->ssl,
1283 				(void*)(((uint8_t*)&len)+c->tcp_byte_count),
1284 				(int)(sizeof(uint16_t)-c->tcp_byte_count));
1285 		}
1286 		if(r <= 0) {
1287 			int want = SSL_get_error(c->ssl, r);
1288 			if(want == SSL_ERROR_ZERO_RETURN) {
1289 				return 0; /* closed */
1290 			} else if(want == SSL_ERROR_WANT_READ) {
1291 				c->ssl_shake_state = comm_ssl_shake_read;
1292 				comm_point_listen_for_rw(c, 1, 0);
1293 				return 1; /* wait for read condition */
1294 			} else if(want == SSL_ERROR_WANT_WRITE) {
1295 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1296 				return 1; /* write more later */
1297 			} else if(want == SSL_ERROR_SYSCALL) {
1298 				if(errno != 0)
1299 					log_err("SSL_write syscall: %s",
1300 						strerror(errno));
1301 				return 0;
1302 			}
1303 			log_crypto_err("could not SSL_write");
1304 			return 0;
1305 		}
1306 		c->tcp_byte_count += r;
1307 		if(c->tcp_byte_count < sizeof(uint16_t))
1308 			return 1;
1309 		sldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1310 			sizeof(uint16_t));
1311 		if(sldns_buffer_remaining(c->buffer) == 0) {
1312 			tcp_callback_writer(c);
1313 			return 1;
1314 		}
1315 	}
1316 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1317 	ERR_clear_error();
1318 	r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
1319 		(int)sldns_buffer_remaining(c->buffer));
1320 	if(r <= 0) {
1321 		int want = SSL_get_error(c->ssl, r);
1322 		if(want == SSL_ERROR_ZERO_RETURN) {
1323 			return 0; /* closed */
1324 		} else if(want == SSL_ERROR_WANT_READ) {
1325 			c->ssl_shake_state = comm_ssl_shake_read;
1326 			comm_point_listen_for_rw(c, 1, 0);
1327 			return 1; /* wait for read condition */
1328 		} else if(want == SSL_ERROR_WANT_WRITE) {
1329 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1330 			return 1; /* write more later */
1331 		} else if(want == SSL_ERROR_SYSCALL) {
1332 			if(errno != 0)
1333 				log_err("SSL_write syscall: %s",
1334 					strerror(errno));
1335 			return 0;
1336 		}
1337 		log_crypto_err("could not SSL_write");
1338 		return 0;
1339 	}
1340 	sldns_buffer_skip(c->buffer, (ssize_t)r);
1341 
1342 	if(sldns_buffer_remaining(c->buffer) == 0) {
1343 		tcp_callback_writer(c);
1344 	}
1345 	return 1;
1346 #else
1347 	(void)c;
1348 	return 0;
1349 #endif /* HAVE_SSL */
1350 }
1351 
1352 /** handle ssl tcp connection with dns contents */
1353 static int
1354 ssl_handle_it(struct comm_point* c)
1355 {
1356 	if(c->tcp_is_reading)
1357 		return ssl_handle_read(c);
1358 	return ssl_handle_write(c);
1359 }
1360 
1361 /** Handle tcp reading callback.
1362  * @param fd: file descriptor of socket.
1363  * @param c: comm point to read from into buffer.
1364  * @param short_ok: if true, very short packets are OK (for comm_local).
1365  * @return: 0 on error
1366  */
1367 static int
1368 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
1369 {
1370 	ssize_t r;
1371 	log_assert(c->type == comm_tcp || c->type == comm_local);
1372 	if(c->ssl)
1373 		return ssl_handle_it(c);
1374 	if(!c->tcp_is_reading)
1375 		return 0;
1376 
1377 	log_assert(fd != -1);
1378 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1379 		/* read length bytes */
1380 		r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count),
1381 			sizeof(uint16_t)-c->tcp_byte_count, 0);
1382 		if(r == 0) {
1383 			if(c->tcp_req_info)
1384 				return tcp_req_info_handle_read_close(c->tcp_req_info);
1385 			return 0;
1386 		} else if(r == -1) {
1387 #ifndef USE_WINSOCK
1388 			if(errno == EINTR || errno == EAGAIN)
1389 				return 1;
1390 #ifdef ECONNRESET
1391 			if(errno == ECONNRESET && verbosity < 2)
1392 				return 0; /* silence reset by peer */
1393 #endif
1394 			log_err_addr("read (in tcp s)", strerror(errno),
1395 				&c->repinfo.addr, c->repinfo.addrlen);
1396 #else /* USE_WINSOCK */
1397 			if(WSAGetLastError() == WSAECONNRESET)
1398 				return 0;
1399 			if(WSAGetLastError() == WSAEINPROGRESS)
1400 				return 1;
1401 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
1402 				ub_winsock_tcp_wouldblock(c->ev->ev,
1403 					UB_EV_READ);
1404 				return 1;
1405 			}
1406 			log_err_addr("read (in tcp s)",
1407 				wsa_strerror(WSAGetLastError()),
1408 				&c->repinfo.addr, c->repinfo.addrlen);
1409 #endif
1410 			return 0;
1411 		}
1412 		c->tcp_byte_count += r;
1413 		if(c->tcp_byte_count != sizeof(uint16_t))
1414 			return 1;
1415 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1416 			sldns_buffer_capacity(c->buffer)) {
1417 			verbose(VERB_QUERY, "tcp: dropped larger than buffer");
1418 			return 0;
1419 		}
1420 		sldns_buffer_set_limit(c->buffer,
1421 			sldns_buffer_read_u16_at(c->buffer, 0));
1422 		if(!short_ok &&
1423 			sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1424 			verbose(VERB_QUERY, "tcp: dropped bogus too short.");
1425 			return 0;
1426 		}
1427 		verbose(VERB_ALGO, "Reading tcp query of length %d",
1428 			(int)sldns_buffer_limit(c->buffer));
1429 	}
1430 
1431 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1432 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1433 		sldns_buffer_remaining(c->buffer), 0);
1434 	if(r == 0) {
1435 		if(c->tcp_req_info)
1436 			return tcp_req_info_handle_read_close(c->tcp_req_info);
1437 		return 0;
1438 	} else if(r == -1) {
1439 #ifndef USE_WINSOCK
1440 		if(errno == EINTR || errno == EAGAIN)
1441 			return 1;
1442 		log_err_addr("read (in tcp r)", strerror(errno),
1443 			&c->repinfo.addr, c->repinfo.addrlen);
1444 #else /* USE_WINSOCK */
1445 		if(WSAGetLastError() == WSAECONNRESET)
1446 			return 0;
1447 		if(WSAGetLastError() == WSAEINPROGRESS)
1448 			return 1;
1449 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1450 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1451 			return 1;
1452 		}
1453 		log_err_addr("read (in tcp r)",
1454 			wsa_strerror(WSAGetLastError()),
1455 			&c->repinfo.addr, c->repinfo.addrlen);
1456 #endif
1457 		return 0;
1458 	}
1459 	sldns_buffer_skip(c->buffer, r);
1460 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1461 		tcp_callback_reader(c);
1462 	}
1463 	return 1;
1464 }
1465 
1466 /**
1467  * Handle tcp writing callback.
1468  * @param fd: file descriptor of socket.
1469  * @param c: comm point to write buffer out of.
1470  * @return: 0 on error
1471  */
1472 static int
1473 comm_point_tcp_handle_write(int fd, struct comm_point* c)
1474 {
1475 	ssize_t r;
1476 	struct sldns_buffer *buffer;
1477 	log_assert(c->type == comm_tcp);
1478 #ifdef USE_DNSCRYPT
1479 	buffer = c->dnscrypt_buffer;
1480 #else
1481 	buffer = c->buffer;
1482 #endif
1483 	if(c->tcp_is_reading && !c->ssl)
1484 		return 0;
1485 	log_assert(fd != -1);
1486 	if(c->tcp_byte_count == 0 && c->tcp_check_nb_connect) {
1487 		/* check for pending error from nonblocking connect */
1488 		/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
1489 		int error = 0;
1490 		socklen_t len = (socklen_t)sizeof(error);
1491 		if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
1492 			&len) < 0){
1493 #ifndef USE_WINSOCK
1494 			error = errno; /* on solaris errno is error */
1495 #else /* USE_WINSOCK */
1496 			error = WSAGetLastError();
1497 #endif
1498 		}
1499 #ifndef USE_WINSOCK
1500 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1501 		if(error == EINPROGRESS || error == EWOULDBLOCK)
1502 			return 1; /* try again later */
1503 		else
1504 #endif
1505 		if(error != 0 && verbosity < 2)
1506 			return 0; /* silence lots of chatter in the logs */
1507                 else if(error != 0) {
1508 			log_err_addr("tcp connect", strerror(error),
1509 				&c->repinfo.addr, c->repinfo.addrlen);
1510 #else /* USE_WINSOCK */
1511 		/* examine error */
1512 		if(error == WSAEINPROGRESS)
1513 			return 1;
1514 		else if(error == WSAEWOULDBLOCK) {
1515 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1516 			return 1;
1517 		} else if(error != 0 && verbosity < 2)
1518 			return 0;
1519 		else if(error != 0) {
1520 			log_err_addr("tcp connect", wsa_strerror(error),
1521 				&c->repinfo.addr, c->repinfo.addrlen);
1522 #endif /* USE_WINSOCK */
1523 			return 0;
1524 		}
1525 	}
1526 	if(c->ssl)
1527 		return ssl_handle_it(c);
1528 
1529 #ifdef USE_MSG_FASTOPEN
1530 	/* Only try this on first use of a connection that uses tfo,
1531 	   otherwise fall through to normal write */
1532 	/* Also, TFO support on WINDOWS not implemented at the moment */
1533 	if(c->tcp_do_fastopen == 1) {
1534 		/* this form of sendmsg() does both a connect() and send() so need to
1535 		   look for various flavours of error*/
1536 		uint16_t len = htons(sldns_buffer_limit(buffer));
1537 		struct msghdr msg;
1538 		struct iovec iov[2];
1539 		c->tcp_do_fastopen = 0;
1540 		memset(&msg, 0, sizeof(msg));
1541 		iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1542 		iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1543 		iov[1].iov_base = sldns_buffer_begin(buffer);
1544 		iov[1].iov_len = sldns_buffer_limit(buffer);
1545 		log_assert(iov[0].iov_len > 0);
1546 		log_assert(iov[1].iov_len > 0);
1547 		msg.msg_name = &c->repinfo.addr;
1548 		msg.msg_namelen = c->repinfo.addrlen;
1549 		msg.msg_iov = iov;
1550 		msg.msg_iovlen = 2;
1551 		r = sendmsg(fd, &msg, MSG_FASTOPEN);
1552 		if (r == -1) {
1553 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1554 			/* Handshake is underway, maybe because no TFO cookie available.
1555 			   Come back to write the message*/
1556 			if(errno == EINPROGRESS || errno == EWOULDBLOCK)
1557 				return 1;
1558 #endif
1559 			if(errno == EINTR || errno == EAGAIN)
1560 				return 1;
1561 			/* Not handling EISCONN here as shouldn't ever hit that case.*/
1562 			if(errno != EPIPE && errno != 0 && verbosity < 2)
1563 				return 0; /* silence lots of chatter in the logs */
1564 			if(errno != EPIPE && errno != 0) {
1565 				log_err_addr("tcp sendmsg", strerror(errno),
1566 					&c->repinfo.addr, c->repinfo.addrlen);
1567 				return 0;
1568 			}
1569 			/* fallthrough to nonFASTOPEN
1570 			 * (MSG_FASTOPEN on Linux 3 produces EPIPE)
1571 			 * we need to perform connect() */
1572 			if(connect(fd, (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen) == -1) {
1573 #ifdef EINPROGRESS
1574 				if(errno == EINPROGRESS)
1575 					return 1; /* wait until connect done*/
1576 #endif
1577 #ifdef USE_WINSOCK
1578 				if(WSAGetLastError() == WSAEINPROGRESS ||
1579 					WSAGetLastError() == WSAEWOULDBLOCK)
1580 					return 1; /* wait until connect done*/
1581 #endif
1582 				if(tcp_connect_errno_needs_log(
1583 					(struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen)) {
1584 					log_err_addr("outgoing tcp: connect after EPIPE for fastopen",
1585 						strerror(errno), &c->repinfo.addr, c->repinfo.addrlen);
1586 				}
1587 				return 0;
1588 			}
1589 
1590 		} else {
1591 			c->tcp_byte_count += r;
1592 			if(c->tcp_byte_count < sizeof(uint16_t))
1593 				return 1;
1594 			sldns_buffer_set_position(buffer, c->tcp_byte_count -
1595 				sizeof(uint16_t));
1596 			if(sldns_buffer_remaining(buffer) == 0) {
1597 				tcp_callback_writer(c);
1598 				return 1;
1599 			}
1600 		}
1601 	}
1602 #endif /* USE_MSG_FASTOPEN */
1603 
1604 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1605 		uint16_t len = htons(sldns_buffer_limit(buffer));
1606 #ifdef HAVE_WRITEV
1607 		struct iovec iov[2];
1608 		iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1609 		iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1610 		iov[1].iov_base = sldns_buffer_begin(buffer);
1611 		iov[1].iov_len = sldns_buffer_limit(buffer);
1612 		log_assert(iov[0].iov_len > 0);
1613 		log_assert(iov[1].iov_len > 0);
1614 		r = writev(fd, iov, 2);
1615 #else /* HAVE_WRITEV */
1616 		r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1617 			sizeof(uint16_t)-c->tcp_byte_count, 0);
1618 #endif /* HAVE_WRITEV */
1619 		if(r == -1) {
1620 #ifndef USE_WINSOCK
1621 #  ifdef EPIPE
1622                 	if(errno == EPIPE && verbosity < 2)
1623                         	return 0; /* silence 'broken pipe' */
1624   #endif
1625 			if(errno == EINTR || errno == EAGAIN)
1626 				return 1;
1627 #  ifdef HAVE_WRITEV
1628 			log_err_addr("tcp writev", strerror(errno),
1629 				&c->repinfo.addr, c->repinfo.addrlen);
1630 #  else /* HAVE_WRITEV */
1631 			log_err_addr("tcp send s", strerror(errno),
1632 				&c->repinfo.addr, c->repinfo.addrlen);
1633 #  endif /* HAVE_WRITEV */
1634 #else
1635 			if(WSAGetLastError() == WSAENOTCONN)
1636 				return 1;
1637 			if(WSAGetLastError() == WSAEINPROGRESS)
1638 				return 1;
1639 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
1640 				ub_winsock_tcp_wouldblock(c->ev->ev,
1641 					UB_EV_WRITE);
1642 				return 1;
1643 			}
1644 			log_err_addr("tcp send s",
1645 				wsa_strerror(WSAGetLastError()),
1646 				&c->repinfo.addr, c->repinfo.addrlen);
1647 #endif
1648 			return 0;
1649 		}
1650 		c->tcp_byte_count += r;
1651 		if(c->tcp_byte_count < sizeof(uint16_t))
1652 			return 1;
1653 		sldns_buffer_set_position(buffer, c->tcp_byte_count -
1654 			sizeof(uint16_t));
1655 		if(sldns_buffer_remaining(buffer) == 0) {
1656 			tcp_callback_writer(c);
1657 			return 1;
1658 		}
1659 	}
1660 	log_assert(sldns_buffer_remaining(buffer) > 0);
1661 	r = send(fd, (void*)sldns_buffer_current(buffer),
1662 		sldns_buffer_remaining(buffer), 0);
1663 	if(r == -1) {
1664 #ifndef USE_WINSOCK
1665 		if(errno == EINTR || errno == EAGAIN)
1666 			return 1;
1667 		log_err_addr("tcp send r", strerror(errno),
1668 			&c->repinfo.addr, c->repinfo.addrlen);
1669 #else
1670 		if(WSAGetLastError() == WSAEINPROGRESS)
1671 			return 1;
1672 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1673 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1674 			return 1;
1675 		}
1676 		log_err_addr("tcp send r", wsa_strerror(WSAGetLastError()),
1677 			&c->repinfo.addr, c->repinfo.addrlen);
1678 #endif
1679 		return 0;
1680 	}
1681 	sldns_buffer_skip(buffer, r);
1682 
1683 	if(sldns_buffer_remaining(buffer) == 0) {
1684 		tcp_callback_writer(c);
1685 	}
1686 
1687 	return 1;
1688 }
1689 
1690 /** read again to drain buffers when there could be more to read */
1691 static void
1692 tcp_req_info_read_again(int fd, struct comm_point* c)
1693 {
1694 	while(c->tcp_req_info->read_again) {
1695 		int r;
1696 		c->tcp_req_info->read_again = 0;
1697 		if(c->tcp_is_reading)
1698 			r = comm_point_tcp_handle_read(fd, c, 0);
1699 		else 	r = comm_point_tcp_handle_write(fd, c);
1700 		if(!r) {
1701 			reclaim_tcp_handler(c);
1702 			if(!c->tcp_do_close) {
1703 				fptr_ok(fptr_whitelist_comm_point(
1704 					c->callback));
1705 				(void)(*c->callback)(c, c->cb_arg,
1706 					NETEVENT_CLOSED, NULL);
1707 			}
1708 			return;
1709 		}
1710 	}
1711 }
1712 
1713 void
1714 comm_point_tcp_handle_callback(int fd, short event, void* arg)
1715 {
1716 	struct comm_point* c = (struct comm_point*)arg;
1717 	log_assert(c->type == comm_tcp);
1718 	ub_comm_base_now(c->ev->base);
1719 
1720 #ifdef USE_DNSCRYPT
1721 	/* Initialize if this is a dnscrypt socket */
1722 	if(c->tcp_parent) {
1723 		c->dnscrypt = c->tcp_parent->dnscrypt;
1724 	}
1725 	if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) {
1726 		c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer));
1727 		if(!c->dnscrypt_buffer) {
1728 			log_err("Could not allocate dnscrypt buffer");
1729 			reclaim_tcp_handler(c);
1730 			if(!c->tcp_do_close) {
1731 				fptr_ok(fptr_whitelist_comm_point(
1732 					c->callback));
1733 				(void)(*c->callback)(c, c->cb_arg,
1734 					NETEVENT_CLOSED, NULL);
1735 			}
1736 			return;
1737 		}
1738 	}
1739 #endif
1740 
1741 	if(event&UB_EV_READ) {
1742 		int has_tcpq = (c->tcp_req_info != NULL);
1743 		if(!comm_point_tcp_handle_read(fd, c, 0)) {
1744 			reclaim_tcp_handler(c);
1745 			if(!c->tcp_do_close) {
1746 				fptr_ok(fptr_whitelist_comm_point(
1747 					c->callback));
1748 				(void)(*c->callback)(c, c->cb_arg,
1749 					NETEVENT_CLOSED, NULL);
1750 			}
1751 		}
1752 		if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
1753 			tcp_req_info_read_again(fd, c);
1754 		return;
1755 	}
1756 	if(event&UB_EV_WRITE) {
1757 		int has_tcpq = (c->tcp_req_info != NULL);
1758 		if(!comm_point_tcp_handle_write(fd, c)) {
1759 			reclaim_tcp_handler(c);
1760 			if(!c->tcp_do_close) {
1761 				fptr_ok(fptr_whitelist_comm_point(
1762 					c->callback));
1763 				(void)(*c->callback)(c, c->cb_arg,
1764 					NETEVENT_CLOSED, NULL);
1765 			}
1766 		}
1767 		if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
1768 			tcp_req_info_read_again(fd, c);
1769 		return;
1770 	}
1771 	if(event&UB_EV_TIMEOUT) {
1772 		verbose(VERB_QUERY, "tcp took too long, dropped");
1773 		reclaim_tcp_handler(c);
1774 		if(!c->tcp_do_close) {
1775 			fptr_ok(fptr_whitelist_comm_point(c->callback));
1776 			(void)(*c->callback)(c, c->cb_arg,
1777 				NETEVENT_TIMEOUT, NULL);
1778 		}
1779 		return;
1780 	}
1781 	log_err("Ignored event %d for tcphdl.", event);
1782 }
1783 
1784 /** Make http handler free for next assignment */
1785 static void
1786 reclaim_http_handler(struct comm_point* c)
1787 {
1788 	log_assert(c->type == comm_http);
1789 	if(c->ssl) {
1790 #ifdef HAVE_SSL
1791 		SSL_shutdown(c->ssl);
1792 		SSL_free(c->ssl);
1793 		c->ssl = NULL;
1794 #endif
1795 	}
1796 	comm_point_close(c);
1797 	if(c->tcp_parent) {
1798 		c->tcp_parent->cur_tcp_count--;
1799 		c->tcp_free = c->tcp_parent->tcp_free;
1800 		c->tcp_parent->tcp_free = c;
1801 		if(!c->tcp_free) {
1802 			/* re-enable listening on accept socket */
1803 			comm_point_start_listening(c->tcp_parent, -1, -1);
1804 		}
1805 	}
1806 }
1807 
1808 /** read more data for http (with ssl) */
1809 static int
1810 ssl_http_read_more(struct comm_point* c)
1811 {
1812 #ifdef HAVE_SSL
1813 	int r;
1814 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1815 	ERR_clear_error();
1816 	r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1817 		(int)sldns_buffer_remaining(c->buffer));
1818 	if(r <= 0) {
1819 		int want = SSL_get_error(c->ssl, r);
1820 		if(want == SSL_ERROR_ZERO_RETURN) {
1821 			return 0; /* shutdown, closed */
1822 		} else if(want == SSL_ERROR_WANT_READ) {
1823 			return 1; /* read more later */
1824 		} else if(want == SSL_ERROR_WANT_WRITE) {
1825 			c->ssl_shake_state = comm_ssl_shake_hs_write;
1826 			comm_point_listen_for_rw(c, 0, 1);
1827 			return 1;
1828 		} else if(want == SSL_ERROR_SYSCALL) {
1829 			if(errno != 0)
1830 				log_err("SSL_read syscall: %s",
1831 					strerror(errno));
1832 			return 0;
1833 		}
1834 		log_crypto_err("could not SSL_read");
1835 		return 0;
1836 	}
1837 	sldns_buffer_skip(c->buffer, (ssize_t)r);
1838 	return 1;
1839 #else
1840 	(void)c;
1841 	return 0;
1842 #endif /* HAVE_SSL */
1843 }
1844 
1845 /** read more data for http */
1846 static int
1847 http_read_more(int fd, struct comm_point* c)
1848 {
1849 	ssize_t r;
1850 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1851 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1852 		sldns_buffer_remaining(c->buffer), 0);
1853 	if(r == 0) {
1854 		return 0;
1855 	} else if(r == -1) {
1856 #ifndef USE_WINSOCK
1857 		if(errno == EINTR || errno == EAGAIN)
1858 			return 1;
1859 		log_err_addr("read (in http r)", strerror(errno),
1860 			&c->repinfo.addr, c->repinfo.addrlen);
1861 #else /* USE_WINSOCK */
1862 		if(WSAGetLastError() == WSAECONNRESET)
1863 			return 0;
1864 		if(WSAGetLastError() == WSAEINPROGRESS)
1865 			return 1;
1866 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1867 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1868 			return 1;
1869 		}
1870 		log_err_addr("read (in http r)",
1871 			wsa_strerror(WSAGetLastError()),
1872 			&c->repinfo.addr, c->repinfo.addrlen);
1873 #endif
1874 		return 0;
1875 	}
1876 	sldns_buffer_skip(c->buffer, r);
1877 	return 1;
1878 }
1879 
1880 /** return true if http header has been read (one line complete) */
1881 static int
1882 http_header_done(sldns_buffer* buf)
1883 {
1884 	size_t i;
1885 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
1886 		/* there was a \r before the \n, but we ignore that */
1887 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n')
1888 			return 1;
1889 	}
1890 	return 0;
1891 }
1892 
1893 /** return character string into buffer for header line, moves buffer
1894  * past that line and puts zero terminator into linefeed-newline */
1895 static char*
1896 http_header_line(sldns_buffer* buf)
1897 {
1898 	char* result = (char*)sldns_buffer_current(buf);
1899 	size_t i;
1900 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
1901 		/* terminate the string on the \r */
1902 		if((char)sldns_buffer_read_u8_at(buf, i) == '\r')
1903 			sldns_buffer_write_u8_at(buf, i, 0);
1904 		/* terminate on the \n and skip past the it and done */
1905 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n') {
1906 			sldns_buffer_write_u8_at(buf, i, 0);
1907 			sldns_buffer_set_position(buf, i+1);
1908 			return result;
1909 		}
1910 	}
1911 	return NULL;
1912 }
1913 
1914 /** move unread buffer to start and clear rest for putting the rest into it */
1915 static void
1916 http_moveover_buffer(sldns_buffer* buf)
1917 {
1918 	size_t pos = sldns_buffer_position(buf);
1919 	size_t len = sldns_buffer_remaining(buf);
1920 	sldns_buffer_clear(buf);
1921 	memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len);
1922 	sldns_buffer_set_position(buf, len);
1923 }
1924 
1925 /** a http header is complete, process it */
1926 static int
1927 http_process_initial_header(struct comm_point* c)
1928 {
1929 	char* line = http_header_line(c->buffer);
1930 	if(!line) return 1;
1931 	verbose(VERB_ALGO, "http header: %s", line);
1932 	if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) {
1933 		/* check returncode */
1934 		if(line[9] != '2') {
1935 			verbose(VERB_ALGO, "http bad status %s", line+9);
1936 			return 0;
1937 		}
1938 	} else if(strncasecmp(line, "Content-Length: ", 16) == 0) {
1939 		if(!c->http_is_chunked)
1940 			c->tcp_byte_count = (size_t)atoi(line+16);
1941 	} else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) {
1942 		c->tcp_byte_count = 0;
1943 		c->http_is_chunked = 1;
1944 	} else if(line[0] == 0) {
1945 		/* end of initial headers */
1946 		c->http_in_headers = 0;
1947 		if(c->http_is_chunked)
1948 			c->http_in_chunk_headers = 1;
1949 		/* remove header text from front of buffer
1950 		 * the buffer is going to be used to return the data segment
1951 		 * itself and we don't want the header to get returned
1952 		 * prepended with it */
1953 		http_moveover_buffer(c->buffer);
1954 		sldns_buffer_flip(c->buffer);
1955 		return 1;
1956 	}
1957 	/* ignore other headers */
1958 	return 1;
1959 }
1960 
1961 /** a chunk header is complete, process it, return 0=fail, 1=continue next
1962  * header line, 2=done with chunked transfer*/
1963 static int
1964 http_process_chunk_header(struct comm_point* c)
1965 {
1966 	char* line = http_header_line(c->buffer);
1967 	if(!line) return 1;
1968 	if(c->http_in_chunk_headers == 3) {
1969 		verbose(VERB_ALGO, "http chunk trailer: %s", line);
1970 		/* are we done ? */
1971 		if(line[0] == 0 && c->tcp_byte_count == 0) {
1972 			/* callback of http reader when NETEVENT_DONE,
1973 			 * end of data, with no data in buffer */
1974 			sldns_buffer_set_position(c->buffer, 0);
1975 			sldns_buffer_set_limit(c->buffer, 0);
1976 			fptr_ok(fptr_whitelist_comm_point(c->callback));
1977 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
1978 			/* return that we are done */
1979 			return 2;
1980 		}
1981 		if(line[0] == 0) {
1982 			/* continue with header of the next chunk */
1983 			c->http_in_chunk_headers = 1;
1984 			/* remove header text from front of buffer */
1985 			http_moveover_buffer(c->buffer);
1986 			sldns_buffer_flip(c->buffer);
1987 			return 1;
1988 		}
1989 		/* ignore further trail headers */
1990 		return 1;
1991 	}
1992 	verbose(VERB_ALGO, "http chunk header: %s", line);
1993 	if(c->http_in_chunk_headers == 1) {
1994 		/* read chunked start line */
1995 		char* end = NULL;
1996 		c->tcp_byte_count = (size_t)strtol(line, &end, 16);
1997 		if(end == line)
1998 			return 0;
1999 		c->http_in_chunk_headers = 0;
2000 		/* remove header text from front of buffer */
2001 		http_moveover_buffer(c->buffer);
2002 		sldns_buffer_flip(c->buffer);
2003 		if(c->tcp_byte_count == 0) {
2004 			/* done with chunks, process chunk_trailer lines */
2005 			c->http_in_chunk_headers = 3;
2006 		}
2007 		return 1;
2008 	}
2009 	/* ignore other headers */
2010 	return 1;
2011 }
2012 
2013 /** handle nonchunked data segment */
2014 static int
2015 http_nonchunk_segment(struct comm_point* c)
2016 {
2017 	/* c->buffer at position..limit has new data we read in.
2018 	 * the buffer itself is full of nonchunked data.
2019 	 * we are looking to read tcp_byte_count more data
2020 	 * and then the transfer is done. */
2021 	size_t remainbufferlen;
2022 	size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
2023 	if(c->tcp_byte_count <= got_now) {
2024 		/* done, this is the last data fragment */
2025 		c->http_stored = 0;
2026 		sldns_buffer_set_position(c->buffer, 0);
2027 		fptr_ok(fptr_whitelist_comm_point(c->callback));
2028 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
2029 		return 1;
2030 	}
2031 	c->tcp_byte_count -= got_now;
2032 	/* if we have the buffer space,
2033 	 * read more data collected into the buffer */
2034 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
2035 		sldns_buffer_limit(c->buffer);
2036 	if(remainbufferlen >= c->tcp_byte_count ||
2037 		remainbufferlen >= 2048) {
2038 		size_t total = sldns_buffer_limit(c->buffer);
2039 		sldns_buffer_clear(c->buffer);
2040 		sldns_buffer_set_position(c->buffer, total);
2041 		c->http_stored = total;
2042 		/* return and wait to read more */
2043 		return 1;
2044 	}
2045 	/* call callback with this data amount, then
2046 	 * wait for more */
2047 	c->http_stored = 0;
2048 	sldns_buffer_set_position(c->buffer, 0);
2049 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2050 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2051 	/* c->callback has to buffer_clear(c->buffer). */
2052 	/* return and wait to read more */
2053 	return 1;
2054 }
2055 
2056 /** handle nonchunked data segment, return 0=fail, 1=wait, 2=process more */
2057 static int
2058 http_chunked_segment(struct comm_point* c)
2059 {
2060 	/* the c->buffer has from position..limit new data we read. */
2061 	/* the current chunk has length tcp_byte_count.
2062 	 * once we read that read more chunk headers.
2063 	 */
2064 	size_t remainbufferlen;
2065 	size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
2066 	if(c->tcp_byte_count <= got_now) {
2067 		/* the chunk has completed (with perhaps some extra data
2068 		 * from next chunk header and next chunk) */
2069 		/* save too much info into temp buffer */
2070 		size_t fraglen;
2071 		struct comm_reply repinfo;
2072 		c->http_stored = 0;
2073 		sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count);
2074 		sldns_buffer_clear(c->http_temp);
2075 		sldns_buffer_write(c->http_temp,
2076 			sldns_buffer_current(c->buffer),
2077 			sldns_buffer_remaining(c->buffer));
2078 		sldns_buffer_flip(c->http_temp);
2079 
2080 		/* callback with this fragment */
2081 		fraglen = sldns_buffer_position(c->buffer);
2082 		sldns_buffer_set_position(c->buffer, 0);
2083 		sldns_buffer_set_limit(c->buffer, fraglen);
2084 		repinfo = c->repinfo;
2085 		fptr_ok(fptr_whitelist_comm_point(c->callback));
2086 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo);
2087 		/* c->callback has to buffer_clear(). */
2088 
2089 		/* is commpoint deleted? */
2090 		if(!repinfo.c) {
2091 			return 1;
2092 		}
2093 		/* copy waiting info */
2094 		sldns_buffer_clear(c->buffer);
2095 		sldns_buffer_write(c->buffer,
2096 			sldns_buffer_begin(c->http_temp),
2097 			sldns_buffer_remaining(c->http_temp));
2098 		sldns_buffer_flip(c->buffer);
2099 		/* process end of chunk trailer header lines, until
2100 		 * an empty line */
2101 		c->http_in_chunk_headers = 3;
2102 		/* process more data in buffer (if any) */
2103 		return 2;
2104 	}
2105 	c->tcp_byte_count -= got_now;
2106 
2107 	/* if we have the buffer space,
2108 	 * read more data collected into the buffer */
2109 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
2110 		sldns_buffer_limit(c->buffer);
2111 	if(remainbufferlen >= c->tcp_byte_count ||
2112 		remainbufferlen >= 2048) {
2113 		size_t total = sldns_buffer_limit(c->buffer);
2114 		sldns_buffer_clear(c->buffer);
2115 		sldns_buffer_set_position(c->buffer, total);
2116 		c->http_stored = total;
2117 		/* return and wait to read more */
2118 		return 1;
2119 	}
2120 
2121 	/* callback of http reader for a new part of the data */
2122 	c->http_stored = 0;
2123 	sldns_buffer_set_position(c->buffer, 0);
2124 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2125 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2126 	/* c->callback has to buffer_clear(c->buffer). */
2127 	/* return and wait to read more */
2128 	return 1;
2129 }
2130 
2131 /**
2132  * Handle http reading callback.
2133  * @param fd: file descriptor of socket.
2134  * @param c: comm point to read from into buffer.
2135  * @return: 0 on error
2136  */
2137 static int
2138 comm_point_http_handle_read(int fd, struct comm_point* c)
2139 {
2140 	log_assert(c->type == comm_http);
2141 	log_assert(fd != -1);
2142 
2143 	/* if we are in ssl handshake, handle SSL handshake */
2144 #ifdef HAVE_SSL
2145 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2146 		if(!ssl_handshake(c))
2147 			return 0;
2148 		if(c->ssl_shake_state != comm_ssl_shake_none)
2149 			return 1;
2150 	}
2151 #endif /* HAVE_SSL */
2152 
2153 	if(!c->tcp_is_reading)
2154 		return 1;
2155 	/* read more data */
2156 	if(c->ssl) {
2157 		if(!ssl_http_read_more(c))
2158 			return 0;
2159 	} else {
2160 		if(!http_read_more(fd, c))
2161 			return 0;
2162 	}
2163 
2164 	sldns_buffer_flip(c->buffer);
2165 	while(sldns_buffer_remaining(c->buffer) > 0) {
2166 		/* if we are reading headers, read more headers */
2167 		if(c->http_in_headers || c->http_in_chunk_headers) {
2168 			/* if header is done, process the header */
2169 			if(!http_header_done(c->buffer)) {
2170 				/* copy remaining data to front of buffer
2171 				 * and set rest for writing into it */
2172 				http_moveover_buffer(c->buffer);
2173 				/* return and wait to read more */
2174 				return 1;
2175 			}
2176 			if(!c->http_in_chunk_headers) {
2177 				/* process initial headers */
2178 				if(!http_process_initial_header(c))
2179 					return 0;
2180 			} else {
2181 				/* process chunk headers */
2182 				int r = http_process_chunk_header(c);
2183 				if(r == 0) return 0;
2184 				if(r == 2) return 1; /* done */
2185 				/* r == 1, continue */
2186 			}
2187 			/* see if we have more to process */
2188 			continue;
2189 		}
2190 
2191 		if(!c->http_is_chunked) {
2192 			/* if we are reading nonchunks, process that*/
2193 			return http_nonchunk_segment(c);
2194 		} else {
2195 			/* if we are reading chunks, read the chunk */
2196 			int r = http_chunked_segment(c);
2197 			if(r == 0) return 0;
2198 			if(r == 1) return 1;
2199 			continue;
2200 		}
2201 	}
2202 	/* broke out of the loop; could not process header instead need
2203 	 * to read more */
2204 	/* moveover any remaining data and read more data */
2205 	http_moveover_buffer(c->buffer);
2206 	/* return and wait to read more */
2207 	return 1;
2208 }
2209 
2210 /** check pending connect for http */
2211 static int
2212 http_check_connect(int fd, struct comm_point* c)
2213 {
2214 	/* check for pending error from nonblocking connect */
2215 	/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
2216 	int error = 0;
2217 	socklen_t len = (socklen_t)sizeof(error);
2218 	if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
2219 		&len) < 0){
2220 #ifndef USE_WINSOCK
2221 		error = errno; /* on solaris errno is error */
2222 #else /* USE_WINSOCK */
2223 		error = WSAGetLastError();
2224 #endif
2225 	}
2226 #ifndef USE_WINSOCK
2227 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
2228 	if(error == EINPROGRESS || error == EWOULDBLOCK)
2229 		return 1; /* try again later */
2230 	else
2231 #endif
2232 	if(error != 0 && verbosity < 2)
2233 		return 0; /* silence lots of chatter in the logs */
2234 	else if(error != 0) {
2235 		log_err_addr("http connect", strerror(error),
2236 			&c->repinfo.addr, c->repinfo.addrlen);
2237 #else /* USE_WINSOCK */
2238 	/* examine error */
2239 	if(error == WSAEINPROGRESS)
2240 		return 1;
2241 	else if(error == WSAEWOULDBLOCK) {
2242 		ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2243 		return 1;
2244 	} else if(error != 0 && verbosity < 2)
2245 		return 0;
2246 	else if(error != 0) {
2247 		log_err_addr("http connect", wsa_strerror(error),
2248 			&c->repinfo.addr, c->repinfo.addrlen);
2249 #endif /* USE_WINSOCK */
2250 		return 0;
2251 	}
2252 	/* keep on processing this socket */
2253 	return 2;
2254 }
2255 
2256 /** write more data for http (with ssl) */
2257 static int
2258 ssl_http_write_more(struct comm_point* c)
2259 {
2260 #ifdef HAVE_SSL
2261 	int r;
2262 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2263 	ERR_clear_error();
2264 	r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
2265 		(int)sldns_buffer_remaining(c->buffer));
2266 	if(r <= 0) {
2267 		int want = SSL_get_error(c->ssl, r);
2268 		if(want == SSL_ERROR_ZERO_RETURN) {
2269 			return 0; /* closed */
2270 		} else if(want == SSL_ERROR_WANT_READ) {
2271 			c->ssl_shake_state = comm_ssl_shake_read;
2272 			comm_point_listen_for_rw(c, 1, 0);
2273 			return 1; /* wait for read condition */
2274 		} else if(want == SSL_ERROR_WANT_WRITE) {
2275 			return 1; /* write more later */
2276 		} else if(want == SSL_ERROR_SYSCALL) {
2277 			if(errno != 0)
2278 				log_err("SSL_write syscall: %s",
2279 					strerror(errno));
2280 			return 0;
2281 		}
2282 		log_crypto_err("could not SSL_write");
2283 		return 0;
2284 	}
2285 	sldns_buffer_skip(c->buffer, (ssize_t)r);
2286 	return 1;
2287 #else
2288 	(void)c;
2289 	return 0;
2290 #endif /* HAVE_SSL */
2291 }
2292 
2293 /** write more data for http */
2294 static int
2295 http_write_more(int fd, struct comm_point* c)
2296 {
2297 	ssize_t r;
2298 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2299 	r = send(fd, (void*)sldns_buffer_current(c->buffer),
2300 		sldns_buffer_remaining(c->buffer), 0);
2301 	if(r == -1) {
2302 #ifndef USE_WINSOCK
2303 		if(errno == EINTR || errno == EAGAIN)
2304 			return 1;
2305 		log_err_addr("http send r", strerror(errno),
2306 			&c->repinfo.addr, c->repinfo.addrlen);
2307 #else
2308 		if(WSAGetLastError() == WSAEINPROGRESS)
2309 			return 1;
2310 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
2311 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2312 			return 1;
2313 		}
2314 		log_err_addr("http send r", wsa_strerror(WSAGetLastError()),
2315 			&c->repinfo.addr, c->repinfo.addrlen);
2316 #endif
2317 		return 0;
2318 	}
2319 	sldns_buffer_skip(c->buffer, r);
2320 	return 1;
2321 }
2322 
2323 /**
2324  * Handle http writing callback.
2325  * @param fd: file descriptor of socket.
2326  * @param c: comm point to write buffer out of.
2327  * @return: 0 on error
2328  */
2329 static int
2330 comm_point_http_handle_write(int fd, struct comm_point* c)
2331 {
2332 	log_assert(c->type == comm_http);
2333 	log_assert(fd != -1);
2334 
2335 	/* check pending connect errors, if that fails, we wait for more,
2336 	 * or we can continue to write contents */
2337 	if(c->tcp_check_nb_connect) {
2338 		int r = http_check_connect(fd, c);
2339 		if(r == 0) return 0;
2340 		if(r == 1) return 1;
2341 		c->tcp_check_nb_connect = 0;
2342 	}
2343 	/* if we are in ssl handshake, handle SSL handshake */
2344 #ifdef HAVE_SSL
2345 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2346 		if(!ssl_handshake(c))
2347 			return 0;
2348 		if(c->ssl_shake_state != comm_ssl_shake_none)
2349 			return 1;
2350 	}
2351 #endif /* HAVE_SSL */
2352 	if(c->tcp_is_reading)
2353 		return 1;
2354 	/* if we are writing, write more */
2355 	if(c->ssl) {
2356 		if(!ssl_http_write_more(c))
2357 			return 0;
2358 	} else {
2359 		if(!http_write_more(fd, c))
2360 			return 0;
2361 	}
2362 
2363 	/* we write a single buffer contents, that can contain
2364 	 * the http request, and then flip to read the results */
2365 	/* see if write is done */
2366 	if(sldns_buffer_remaining(c->buffer) == 0) {
2367 		sldns_buffer_clear(c->buffer);
2368 		if(c->tcp_do_toggle_rw)
2369 			c->tcp_is_reading = 1;
2370 		c->tcp_byte_count = 0;
2371 		/* switch from listening(write) to listening(read) */
2372 		comm_point_stop_listening(c);
2373 		comm_point_start_listening(c, -1, -1);
2374 	}
2375 	return 1;
2376 }
2377 
2378 void
2379 comm_point_http_handle_callback(int fd, short event, void* arg)
2380 {
2381 	struct comm_point* c = (struct comm_point*)arg;
2382 	log_assert(c->type == comm_http);
2383 	ub_comm_base_now(c->ev->base);
2384 
2385 	if(event&UB_EV_READ) {
2386 		if(!comm_point_http_handle_read(fd, c)) {
2387 			reclaim_http_handler(c);
2388 			if(!c->tcp_do_close) {
2389 				fptr_ok(fptr_whitelist_comm_point(
2390 					c->callback));
2391 				(void)(*c->callback)(c, c->cb_arg,
2392 					NETEVENT_CLOSED, NULL);
2393 			}
2394 		}
2395 		return;
2396 	}
2397 	if(event&UB_EV_WRITE) {
2398 		if(!comm_point_http_handle_write(fd, c)) {
2399 			reclaim_http_handler(c);
2400 			if(!c->tcp_do_close) {
2401 				fptr_ok(fptr_whitelist_comm_point(
2402 					c->callback));
2403 				(void)(*c->callback)(c, c->cb_arg,
2404 					NETEVENT_CLOSED, NULL);
2405 			}
2406 		}
2407 		return;
2408 	}
2409 	if(event&UB_EV_TIMEOUT) {
2410 		verbose(VERB_QUERY, "http took too long, dropped");
2411 		reclaim_http_handler(c);
2412 		if(!c->tcp_do_close) {
2413 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2414 			(void)(*c->callback)(c, c->cb_arg,
2415 				NETEVENT_TIMEOUT, NULL);
2416 		}
2417 		return;
2418 	}
2419 	log_err("Ignored event %d for httphdl.", event);
2420 }
2421 
2422 void comm_point_local_handle_callback(int fd, short event, void* arg)
2423 {
2424 	struct comm_point* c = (struct comm_point*)arg;
2425 	log_assert(c->type == comm_local);
2426 	ub_comm_base_now(c->ev->base);
2427 
2428 	if(event&UB_EV_READ) {
2429 		if(!comm_point_tcp_handle_read(fd, c, 1)) {
2430 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2431 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
2432 				NULL);
2433 		}
2434 		return;
2435 	}
2436 	log_err("Ignored event %d for localhdl.", event);
2437 }
2438 
2439 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
2440 	short event, void* arg)
2441 {
2442 	struct comm_point* c = (struct comm_point*)arg;
2443 	int err = NETEVENT_NOERROR;
2444 	log_assert(c->type == comm_raw);
2445 	ub_comm_base_now(c->ev->base);
2446 
2447 	if(event&UB_EV_TIMEOUT)
2448 		err = NETEVENT_TIMEOUT;
2449 	fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
2450 	(void)(*c->callback)(c, c->cb_arg, err, NULL);
2451 }
2452 
2453 struct comm_point*
2454 comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer,
2455 	comm_point_callback_type* callback, void* callback_arg)
2456 {
2457 	struct comm_point* c = (struct comm_point*)calloc(1,
2458 		sizeof(struct comm_point));
2459 	short evbits;
2460 	if(!c)
2461 		return NULL;
2462 	c->ev = (struct internal_event*)calloc(1,
2463 		sizeof(struct internal_event));
2464 	if(!c->ev) {
2465 		free(c);
2466 		return NULL;
2467 	}
2468 	c->ev->base = base;
2469 	c->fd = fd;
2470 	c->buffer = buffer;
2471 	c->timeout = NULL;
2472 	c->tcp_is_reading = 0;
2473 	c->tcp_byte_count = 0;
2474 	c->tcp_parent = NULL;
2475 	c->max_tcp_count = 0;
2476 	c->cur_tcp_count = 0;
2477 	c->tcp_handlers = NULL;
2478 	c->tcp_free = NULL;
2479 	c->type = comm_udp;
2480 	c->tcp_do_close = 0;
2481 	c->do_not_close = 0;
2482 	c->tcp_do_toggle_rw = 0;
2483 	c->tcp_check_nb_connect = 0;
2484 #ifdef USE_MSG_FASTOPEN
2485 	c->tcp_do_fastopen = 0;
2486 #endif
2487 #ifdef USE_DNSCRYPT
2488 	c->dnscrypt = 0;
2489 	c->dnscrypt_buffer = buffer;
2490 #endif
2491 	c->inuse = 0;
2492 	c->callback = callback;
2493 	c->cb_arg = callback_arg;
2494 	evbits = UB_EV_READ | UB_EV_PERSIST;
2495 	/* ub_event stuff */
2496 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2497 		comm_point_udp_callback, c);
2498 	if(c->ev->ev == NULL) {
2499 		log_err("could not baseset udp event");
2500 		comm_point_delete(c);
2501 		return NULL;
2502 	}
2503 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
2504 		log_err("could not add udp event");
2505 		comm_point_delete(c);
2506 		return NULL;
2507 	}
2508 	return c;
2509 }
2510 
2511 struct comm_point*
2512 comm_point_create_udp_ancil(struct comm_base *base, int fd,
2513 	sldns_buffer* buffer,
2514 	comm_point_callback_type* callback, void* callback_arg)
2515 {
2516 	struct comm_point* c = (struct comm_point*)calloc(1,
2517 		sizeof(struct comm_point));
2518 	short evbits;
2519 	if(!c)
2520 		return NULL;
2521 	c->ev = (struct internal_event*)calloc(1,
2522 		sizeof(struct internal_event));
2523 	if(!c->ev) {
2524 		free(c);
2525 		return NULL;
2526 	}
2527 	c->ev->base = base;
2528 	c->fd = fd;
2529 	c->buffer = buffer;
2530 	c->timeout = NULL;
2531 	c->tcp_is_reading = 0;
2532 	c->tcp_byte_count = 0;
2533 	c->tcp_parent = NULL;
2534 	c->max_tcp_count = 0;
2535 	c->cur_tcp_count = 0;
2536 	c->tcp_handlers = NULL;
2537 	c->tcp_free = NULL;
2538 	c->type = comm_udp;
2539 	c->tcp_do_close = 0;
2540 	c->do_not_close = 0;
2541 #ifdef USE_DNSCRYPT
2542 	c->dnscrypt = 0;
2543 	c->dnscrypt_buffer = buffer;
2544 #endif
2545 	c->inuse = 0;
2546 	c->tcp_do_toggle_rw = 0;
2547 	c->tcp_check_nb_connect = 0;
2548 #ifdef USE_MSG_FASTOPEN
2549 	c->tcp_do_fastopen = 0;
2550 #endif
2551 	c->callback = callback;
2552 	c->cb_arg = callback_arg;
2553 	evbits = UB_EV_READ | UB_EV_PERSIST;
2554 	/* ub_event stuff */
2555 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2556 		comm_point_udp_ancil_callback, c);
2557 	if(c->ev->ev == NULL) {
2558 		log_err("could not baseset udp event");
2559 		comm_point_delete(c);
2560 		return NULL;
2561 	}
2562 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
2563 		log_err("could not add udp event");
2564 		comm_point_delete(c);
2565 		return NULL;
2566 	}
2567 	return c;
2568 }
2569 
2570 static struct comm_point*
2571 comm_point_create_tcp_handler(struct comm_base *base,
2572 	struct comm_point* parent, size_t bufsize,
2573 	struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
2574 	void* callback_arg)
2575 {
2576 	struct comm_point* c = (struct comm_point*)calloc(1,
2577 		sizeof(struct comm_point));
2578 	short evbits;
2579 	if(!c)
2580 		return NULL;
2581 	c->ev = (struct internal_event*)calloc(1,
2582 		sizeof(struct internal_event));
2583 	if(!c->ev) {
2584 		free(c);
2585 		return NULL;
2586 	}
2587 	c->ev->base = base;
2588 	c->fd = -1;
2589 	c->buffer = sldns_buffer_new(bufsize);
2590 	if(!c->buffer) {
2591 		free(c->ev);
2592 		free(c);
2593 		return NULL;
2594 	}
2595 	c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
2596 	if(!c->timeout) {
2597 		sldns_buffer_free(c->buffer);
2598 		free(c->ev);
2599 		free(c);
2600 		return NULL;
2601 	}
2602 	c->tcp_is_reading = 0;
2603 	c->tcp_byte_count = 0;
2604 	c->tcp_parent = parent;
2605 	c->tcp_timeout_msec = parent->tcp_timeout_msec;
2606 	c->tcp_conn_limit = parent->tcp_conn_limit;
2607 	c->tcl_addr = NULL;
2608 	c->tcp_keepalive = 0;
2609 	c->max_tcp_count = 0;
2610 	c->cur_tcp_count = 0;
2611 	c->tcp_handlers = NULL;
2612 	c->tcp_free = NULL;
2613 	c->type = comm_tcp;
2614 	c->tcp_do_close = 0;
2615 	c->do_not_close = 0;
2616 	c->tcp_do_toggle_rw = 1;
2617 	c->tcp_check_nb_connect = 0;
2618 #ifdef USE_MSG_FASTOPEN
2619 	c->tcp_do_fastopen = 0;
2620 #endif
2621 #ifdef USE_DNSCRYPT
2622 	c->dnscrypt = 0;
2623 	/* We don't know just yet if this is a dnscrypt channel. Allocation
2624 	 * will be done when handling the callback. */
2625 	c->dnscrypt_buffer = c->buffer;
2626 #endif
2627 	c->repinfo.c = c;
2628 	c->callback = callback;
2629 	c->cb_arg = callback_arg;
2630 	if(spoolbuf) {
2631 		c->tcp_req_info = tcp_req_info_create(spoolbuf);
2632 		if(!c->tcp_req_info) {
2633 			log_err("could not create tcp commpoint");
2634 			sldns_buffer_free(c->buffer);
2635 			free(c->timeout);
2636 			free(c->ev);
2637 			free(c);
2638 			return NULL;
2639 		}
2640 		c->tcp_req_info->cp = c;
2641 		c->tcp_do_close = 1;
2642 		c->tcp_do_toggle_rw = 0;
2643 	}
2644 	/* add to parent free list */
2645 	c->tcp_free = parent->tcp_free;
2646 	parent->tcp_free = c;
2647 	/* ub_event stuff */
2648 	evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
2649 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2650 		comm_point_tcp_handle_callback, c);
2651 	if(c->ev->ev == NULL)
2652 	{
2653 		log_err("could not basetset tcphdl event");
2654 		parent->tcp_free = c->tcp_free;
2655 		tcp_req_info_delete(c->tcp_req_info);
2656 		sldns_buffer_free(c->buffer);
2657 		free(c->timeout);
2658 		free(c->ev);
2659 		free(c);
2660 		return NULL;
2661 	}
2662 	return c;
2663 }
2664 
2665 struct comm_point*
2666 comm_point_create_tcp(struct comm_base *base, int fd, int num,
2667 	int idle_timeout, struct tcl_list* tcp_conn_limit, size_t bufsize,
2668 	struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
2669 	void* callback_arg)
2670 {
2671 	struct comm_point* c = (struct comm_point*)calloc(1,
2672 		sizeof(struct comm_point));
2673 	short evbits;
2674 	int i;
2675 	/* first allocate the TCP accept listener */
2676 	if(!c)
2677 		return NULL;
2678 	c->ev = (struct internal_event*)calloc(1,
2679 		sizeof(struct internal_event));
2680 	if(!c->ev) {
2681 		free(c);
2682 		return NULL;
2683 	}
2684 	c->ev->base = base;
2685 	c->fd = fd;
2686 	c->buffer = NULL;
2687 	c->timeout = NULL;
2688 	c->tcp_is_reading = 0;
2689 	c->tcp_byte_count = 0;
2690 	c->tcp_timeout_msec = idle_timeout;
2691 	c->tcp_conn_limit = tcp_conn_limit;
2692 	c->tcl_addr = NULL;
2693 	c->tcp_keepalive = 0;
2694 	c->tcp_parent = NULL;
2695 	c->max_tcp_count = num;
2696 	c->cur_tcp_count = 0;
2697 	c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
2698 		sizeof(struct comm_point*));
2699 	if(!c->tcp_handlers) {
2700 		free(c->ev);
2701 		free(c);
2702 		return NULL;
2703 	}
2704 	c->tcp_free = NULL;
2705 	c->type = comm_tcp_accept;
2706 	c->tcp_do_close = 0;
2707 	c->do_not_close = 0;
2708 	c->tcp_do_toggle_rw = 0;
2709 	c->tcp_check_nb_connect = 0;
2710 #ifdef USE_MSG_FASTOPEN
2711 	c->tcp_do_fastopen = 0;
2712 #endif
2713 #ifdef USE_DNSCRYPT
2714 	c->dnscrypt = 0;
2715 	c->dnscrypt_buffer = NULL;
2716 #endif
2717 	c->callback = NULL;
2718 	c->cb_arg = NULL;
2719 	evbits = UB_EV_READ | UB_EV_PERSIST;
2720 	/* ub_event stuff */
2721 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2722 		comm_point_tcp_accept_callback, c);
2723 	if(c->ev->ev == NULL) {
2724 		log_err("could not baseset tcpacc event");
2725 		comm_point_delete(c);
2726 		return NULL;
2727 	}
2728 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2729 		log_err("could not add tcpacc event");
2730 		comm_point_delete(c);
2731 		return NULL;
2732 	}
2733 	/* now prealloc the tcp handlers */
2734 	for(i=0; i<num; i++) {
2735 		c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
2736 			c, bufsize, spoolbuf, callback, callback_arg);
2737 		if(!c->tcp_handlers[i]) {
2738 			comm_point_delete(c);
2739 			return NULL;
2740 		}
2741 	}
2742 
2743 	return c;
2744 }
2745 
2746 struct comm_point*
2747 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
2748         comm_point_callback_type* callback, void* callback_arg)
2749 {
2750 	struct comm_point* c = (struct comm_point*)calloc(1,
2751 		sizeof(struct comm_point));
2752 	short evbits;
2753 	if(!c)
2754 		return NULL;
2755 	c->ev = (struct internal_event*)calloc(1,
2756 		sizeof(struct internal_event));
2757 	if(!c->ev) {
2758 		free(c);
2759 		return NULL;
2760 	}
2761 	c->ev->base = base;
2762 	c->fd = -1;
2763 	c->buffer = sldns_buffer_new(bufsize);
2764 	if(!c->buffer) {
2765 		free(c->ev);
2766 		free(c);
2767 		return NULL;
2768 	}
2769 	c->timeout = NULL;
2770 	c->tcp_is_reading = 0;
2771 	c->tcp_byte_count = 0;
2772 	c->tcp_timeout_msec = TCP_QUERY_TIMEOUT;
2773 	c->tcp_conn_limit = NULL;
2774 	c->tcl_addr = NULL;
2775 	c->tcp_keepalive = 0;
2776 	c->tcp_parent = NULL;
2777 	c->max_tcp_count = 0;
2778 	c->cur_tcp_count = 0;
2779 	c->tcp_handlers = NULL;
2780 	c->tcp_free = NULL;
2781 	c->type = comm_tcp;
2782 	c->tcp_do_close = 0;
2783 	c->do_not_close = 0;
2784 	c->tcp_do_toggle_rw = 1;
2785 	c->tcp_check_nb_connect = 1;
2786 #ifdef USE_MSG_FASTOPEN
2787 	c->tcp_do_fastopen = 1;
2788 #endif
2789 #ifdef USE_DNSCRYPT
2790 	c->dnscrypt = 0;
2791 	c->dnscrypt_buffer = c->buffer;
2792 #endif
2793 	c->repinfo.c = c;
2794 	c->callback = callback;
2795 	c->cb_arg = callback_arg;
2796 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
2797 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2798 		comm_point_tcp_handle_callback, c);
2799 	if(c->ev->ev == NULL)
2800 	{
2801 		log_err("could not baseset tcpout event");
2802 		sldns_buffer_free(c->buffer);
2803 		free(c->ev);
2804 		free(c);
2805 		return NULL;
2806 	}
2807 
2808 	return c;
2809 }
2810 
2811 struct comm_point*
2812 comm_point_create_http_out(struct comm_base *base, size_t bufsize,
2813         comm_point_callback_type* callback, void* callback_arg,
2814 	sldns_buffer* temp)
2815 {
2816 	struct comm_point* c = (struct comm_point*)calloc(1,
2817 		sizeof(struct comm_point));
2818 	short evbits;
2819 	if(!c)
2820 		return NULL;
2821 	c->ev = (struct internal_event*)calloc(1,
2822 		sizeof(struct internal_event));
2823 	if(!c->ev) {
2824 		free(c);
2825 		return NULL;
2826 	}
2827 	c->ev->base = base;
2828 	c->fd = -1;
2829 	c->buffer = sldns_buffer_new(bufsize);
2830 	if(!c->buffer) {
2831 		free(c->ev);
2832 		free(c);
2833 		return NULL;
2834 	}
2835 	c->timeout = NULL;
2836 	c->tcp_is_reading = 0;
2837 	c->tcp_byte_count = 0;
2838 	c->tcp_parent = NULL;
2839 	c->max_tcp_count = 0;
2840 	c->cur_tcp_count = 0;
2841 	c->tcp_handlers = NULL;
2842 	c->tcp_free = NULL;
2843 	c->type = comm_http;
2844 	c->tcp_do_close = 0;
2845 	c->do_not_close = 0;
2846 	c->tcp_do_toggle_rw = 1;
2847 	c->tcp_check_nb_connect = 1;
2848 	c->http_in_headers = 1;
2849 	c->http_in_chunk_headers = 0;
2850 	c->http_is_chunked = 0;
2851 	c->http_temp = temp;
2852 #ifdef USE_MSG_FASTOPEN
2853 	c->tcp_do_fastopen = 1;
2854 #endif
2855 #ifdef USE_DNSCRYPT
2856 	c->dnscrypt = 0;
2857 	c->dnscrypt_buffer = c->buffer;
2858 #endif
2859 	c->repinfo.c = c;
2860 	c->callback = callback;
2861 	c->cb_arg = callback_arg;
2862 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
2863 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2864 		comm_point_http_handle_callback, c);
2865 	if(c->ev->ev == NULL)
2866 	{
2867 		log_err("could not baseset tcpout event");
2868 #ifdef HAVE_SSL
2869 		SSL_free(c->ssl);
2870 #endif
2871 		sldns_buffer_free(c->buffer);
2872 		free(c->ev);
2873 		free(c);
2874 		return NULL;
2875 	}
2876 
2877 	return c;
2878 }
2879 
2880 struct comm_point*
2881 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
2882         comm_point_callback_type* callback, void* callback_arg)
2883 {
2884 	struct comm_point* c = (struct comm_point*)calloc(1,
2885 		sizeof(struct comm_point));
2886 	short evbits;
2887 	if(!c)
2888 		return NULL;
2889 	c->ev = (struct internal_event*)calloc(1,
2890 		sizeof(struct internal_event));
2891 	if(!c->ev) {
2892 		free(c);
2893 		return NULL;
2894 	}
2895 	c->ev->base = base;
2896 	c->fd = fd;
2897 	c->buffer = sldns_buffer_new(bufsize);
2898 	if(!c->buffer) {
2899 		free(c->ev);
2900 		free(c);
2901 		return NULL;
2902 	}
2903 	c->timeout = NULL;
2904 	c->tcp_is_reading = 1;
2905 	c->tcp_byte_count = 0;
2906 	c->tcp_parent = NULL;
2907 	c->max_tcp_count = 0;
2908 	c->cur_tcp_count = 0;
2909 	c->tcp_handlers = NULL;
2910 	c->tcp_free = NULL;
2911 	c->type = comm_local;
2912 	c->tcp_do_close = 0;
2913 	c->do_not_close = 1;
2914 	c->tcp_do_toggle_rw = 0;
2915 	c->tcp_check_nb_connect = 0;
2916 #ifdef USE_MSG_FASTOPEN
2917 	c->tcp_do_fastopen = 0;
2918 #endif
2919 #ifdef USE_DNSCRYPT
2920 	c->dnscrypt = 0;
2921 	c->dnscrypt_buffer = c->buffer;
2922 #endif
2923 	c->callback = callback;
2924 	c->cb_arg = callback_arg;
2925 	/* ub_event stuff */
2926 	evbits = UB_EV_PERSIST | UB_EV_READ;
2927 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2928 		comm_point_local_handle_callback, c);
2929 	if(c->ev->ev == NULL) {
2930 		log_err("could not baseset localhdl event");
2931 		free(c->ev);
2932 		free(c);
2933 		return NULL;
2934 	}
2935 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2936 		log_err("could not add localhdl event");
2937 		ub_event_free(c->ev->ev);
2938 		free(c->ev);
2939 		free(c);
2940 		return NULL;
2941 	}
2942 	return c;
2943 }
2944 
2945 struct comm_point*
2946 comm_point_create_raw(struct comm_base* base, int fd, int writing,
2947 	comm_point_callback_type* callback, void* callback_arg)
2948 {
2949 	struct comm_point* c = (struct comm_point*)calloc(1,
2950 		sizeof(struct comm_point));
2951 	short evbits;
2952 	if(!c)
2953 		return NULL;
2954 	c->ev = (struct internal_event*)calloc(1,
2955 		sizeof(struct internal_event));
2956 	if(!c->ev) {
2957 		free(c);
2958 		return NULL;
2959 	}
2960 	c->ev->base = base;
2961 	c->fd = fd;
2962 	c->buffer = NULL;
2963 	c->timeout = NULL;
2964 	c->tcp_is_reading = 0;
2965 	c->tcp_byte_count = 0;
2966 	c->tcp_parent = NULL;
2967 	c->max_tcp_count = 0;
2968 	c->cur_tcp_count = 0;
2969 	c->tcp_handlers = NULL;
2970 	c->tcp_free = NULL;
2971 	c->type = comm_raw;
2972 	c->tcp_do_close = 0;
2973 	c->do_not_close = 1;
2974 	c->tcp_do_toggle_rw = 0;
2975 	c->tcp_check_nb_connect = 0;
2976 #ifdef USE_MSG_FASTOPEN
2977 	c->tcp_do_fastopen = 0;
2978 #endif
2979 #ifdef USE_DNSCRYPT
2980 	c->dnscrypt = 0;
2981 	c->dnscrypt_buffer = c->buffer;
2982 #endif
2983 	c->callback = callback;
2984 	c->cb_arg = callback_arg;
2985 	/* ub_event stuff */
2986 	if(writing)
2987 		evbits = UB_EV_PERSIST | UB_EV_WRITE;
2988 	else 	evbits = UB_EV_PERSIST | UB_EV_READ;
2989 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2990 		comm_point_raw_handle_callback, c);
2991 	if(c->ev->ev == NULL) {
2992 		log_err("could not baseset rawhdl event");
2993 		free(c->ev);
2994 		free(c);
2995 		return NULL;
2996 	}
2997 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2998 		log_err("could not add rawhdl event");
2999 		ub_event_free(c->ev->ev);
3000 		free(c->ev);
3001 		free(c);
3002 		return NULL;
3003 	}
3004 	return c;
3005 }
3006 
3007 void
3008 comm_point_close(struct comm_point* c)
3009 {
3010 	if(!c)
3011 		return;
3012 	if(c->fd != -1) {
3013 		if(ub_event_del(c->ev->ev) != 0) {
3014 			log_err("could not event_del on close");
3015 		}
3016 	}
3017 	tcl_close_connection(c->tcl_addr);
3018 	if(c->tcp_req_info)
3019 		tcp_req_info_clear(c->tcp_req_info);
3020 	/* close fd after removing from event lists, or epoll.. is messed up */
3021 	if(c->fd != -1 && !c->do_not_close) {
3022 		if(c->type == comm_tcp || c->type == comm_http) {
3023 			/* delete sticky events for the fd, it gets closed */
3024 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
3025 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
3026 		}
3027 		verbose(VERB_ALGO, "close fd %d", c->fd);
3028 #ifndef USE_WINSOCK
3029 		close(c->fd);
3030 #else
3031 		closesocket(c->fd);
3032 #endif
3033 	}
3034 	c->fd = -1;
3035 }
3036 
3037 void
3038 comm_point_delete(struct comm_point* c)
3039 {
3040 	if(!c)
3041 		return;
3042 	if((c->type == comm_tcp || c->type == comm_http) && c->ssl) {
3043 #ifdef HAVE_SSL
3044 		SSL_shutdown(c->ssl);
3045 		SSL_free(c->ssl);
3046 #endif
3047 	}
3048 	comm_point_close(c);
3049 	if(c->tcp_handlers) {
3050 		int i;
3051 		for(i=0; i<c->max_tcp_count; i++)
3052 			comm_point_delete(c->tcp_handlers[i]);
3053 		free(c->tcp_handlers);
3054 	}
3055 	free(c->timeout);
3056 	if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) {
3057 		sldns_buffer_free(c->buffer);
3058 #ifdef USE_DNSCRYPT
3059 		if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) {
3060 			sldns_buffer_free(c->dnscrypt_buffer);
3061 		}
3062 #endif
3063 		if(c->tcp_req_info) {
3064 			tcp_req_info_delete(c->tcp_req_info);
3065 		}
3066 	}
3067 	ub_event_free(c->ev->ev);
3068 	free(c->ev);
3069 	free(c);
3070 }
3071 
3072 void
3073 comm_point_send_reply(struct comm_reply *repinfo)
3074 {
3075 	struct sldns_buffer* buffer;
3076 	log_assert(repinfo && repinfo->c);
3077 #ifdef USE_DNSCRYPT
3078 	buffer = repinfo->c->dnscrypt_buffer;
3079 	if(!dnsc_handle_uncurved_request(repinfo)) {
3080 		return;
3081 	}
3082 #else
3083 	buffer = repinfo->c->buffer;
3084 #endif
3085 	if(repinfo->c->type == comm_udp) {
3086 		if(repinfo->srctype)
3087 			comm_point_send_udp_msg_if(repinfo->c,
3088 			buffer, (struct sockaddr*)&repinfo->addr,
3089 			repinfo->addrlen, repinfo);
3090 		else
3091 			comm_point_send_udp_msg(repinfo->c, buffer,
3092 			(struct sockaddr*)&repinfo->addr, repinfo->addrlen);
3093 #ifdef USE_DNSTAP
3094 		if(repinfo->c->dtenv != NULL &&
3095 		   repinfo->c->dtenv->log_client_response_messages)
3096 			dt_msg_send_client_response(repinfo->c->dtenv,
3097 			&repinfo->addr, repinfo->c->type, repinfo->c->buffer);
3098 #endif
3099 	} else {
3100 #ifdef USE_DNSTAP
3101 		if(repinfo->c->tcp_parent->dtenv != NULL &&
3102 		   repinfo->c->tcp_parent->dtenv->log_client_response_messages)
3103 			dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv,
3104 			&repinfo->addr, repinfo->c->type, repinfo->c->buffer);
3105 #endif
3106 		if(repinfo->c->tcp_req_info) {
3107 			tcp_req_info_send_reply(repinfo->c->tcp_req_info);
3108 		} else {
3109 			comm_point_start_listening(repinfo->c, -1,
3110 				repinfo->c->tcp_timeout_msec);
3111 		}
3112 	}
3113 }
3114 
3115 void
3116 comm_point_drop_reply(struct comm_reply* repinfo)
3117 {
3118 	if(!repinfo)
3119 		return;
3120 	log_assert(repinfo && repinfo->c);
3121 	log_assert(repinfo->c->type != comm_tcp_accept);
3122 	if(repinfo->c->type == comm_udp)
3123 		return;
3124 	if(repinfo->c->tcp_req_info)
3125 		repinfo->c->tcp_req_info->is_drop = 1;
3126 	reclaim_tcp_handler(repinfo->c);
3127 }
3128 
3129 void
3130 comm_point_stop_listening(struct comm_point* c)
3131 {
3132 	verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
3133 	if(ub_event_del(c->ev->ev) != 0) {
3134 		log_err("event_del error to stoplisten");
3135 	}
3136 }
3137 
3138 void
3139 comm_point_start_listening(struct comm_point* c, int newfd, int msec)
3140 {
3141 	verbose(VERB_ALGO, "comm point start listening %d",
3142 		c->fd==-1?newfd:c->fd);
3143 	if(c->type == comm_tcp_accept && !c->tcp_free) {
3144 		/* no use to start listening no free slots. */
3145 		return;
3146 	}
3147 	if(msec != -1 && msec != 0) {
3148 		if(!c->timeout) {
3149 			c->timeout = (struct timeval*)malloc(sizeof(
3150 				struct timeval));
3151 			if(!c->timeout) {
3152 				log_err("cpsl: malloc failed. No net read.");
3153 				return;
3154 			}
3155 		}
3156 		ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT);
3157 #ifndef S_SPLINT_S /* splint fails on struct timeval. */
3158 		c->timeout->tv_sec = msec/1000;
3159 		c->timeout->tv_usec = (msec%1000)*1000;
3160 #endif /* S_SPLINT_S */
3161 	}
3162 	if(c->type == comm_tcp || c->type == comm_http) {
3163 		ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
3164 		if(c->tcp_is_reading)
3165 			ub_event_add_bits(c->ev->ev, UB_EV_READ);
3166 		else	ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
3167 	}
3168 	if(newfd != -1) {
3169 		if(c->fd != -1) {
3170 #ifndef USE_WINSOCK
3171 			close(c->fd);
3172 #else
3173 			closesocket(c->fd);
3174 #endif
3175 		}
3176 		c->fd = newfd;
3177 		ub_event_set_fd(c->ev->ev, c->fd);
3178 	}
3179 	if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) {
3180 		log_err("event_add failed. in cpsl.");
3181 	}
3182 }
3183 
3184 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
3185 {
3186 	verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
3187 	if(ub_event_del(c->ev->ev) != 0) {
3188 		log_err("event_del error to cplf");
3189 	}
3190 	ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
3191 	if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ);
3192 	if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
3193 	if(ub_event_add(c->ev->ev, c->timeout) != 0) {
3194 		log_err("event_add failed. in cplf.");
3195 	}
3196 }
3197 
3198 size_t comm_point_get_mem(struct comm_point* c)
3199 {
3200 	size_t s;
3201 	if(!c)
3202 		return 0;
3203 	s = sizeof(*c) + sizeof(*c->ev);
3204 	if(c->timeout)
3205 		s += sizeof(*c->timeout);
3206 	if(c->type == comm_tcp || c->type == comm_local) {
3207 		s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer);
3208 #ifdef USE_DNSCRYPT
3209 		s += sizeof(*c->dnscrypt_buffer);
3210 		if(c->buffer != c->dnscrypt_buffer) {
3211 			s += sldns_buffer_capacity(c->dnscrypt_buffer);
3212 		}
3213 #endif
3214 	}
3215 	if(c->type == comm_tcp_accept) {
3216 		int i;
3217 		for(i=0; i<c->max_tcp_count; i++)
3218 			s += comm_point_get_mem(c->tcp_handlers[i]);
3219 	}
3220 	return s;
3221 }
3222 
3223 struct comm_timer*
3224 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
3225 {
3226 	struct internal_timer *tm = (struct internal_timer*)calloc(1,
3227 		sizeof(struct internal_timer));
3228 	if(!tm) {
3229 		log_err("malloc failed");
3230 		return NULL;
3231 	}
3232 	tm->super.ev_timer = tm;
3233 	tm->base = base;
3234 	tm->super.callback = cb;
3235 	tm->super.cb_arg = cb_arg;
3236 	tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT,
3237 		comm_timer_callback, &tm->super);
3238 	if(tm->ev == NULL) {
3239 		log_err("timer_create: event_base_set failed.");
3240 		free(tm);
3241 		return NULL;
3242 	}
3243 	return &tm->super;
3244 }
3245 
3246 void
3247 comm_timer_disable(struct comm_timer* timer)
3248 {
3249 	if(!timer)
3250 		return;
3251 	ub_timer_del(timer->ev_timer->ev);
3252 	timer->ev_timer->enabled = 0;
3253 }
3254 
3255 void
3256 comm_timer_set(struct comm_timer* timer, struct timeval* tv)
3257 {
3258 	log_assert(tv);
3259 	if(timer->ev_timer->enabled)
3260 		comm_timer_disable(timer);
3261 	if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base,
3262 		comm_timer_callback, timer, tv) != 0)
3263 		log_err("comm_timer_set: evtimer_add failed.");
3264 	timer->ev_timer->enabled = 1;
3265 }
3266 
3267 void
3268 comm_timer_delete(struct comm_timer* timer)
3269 {
3270 	if(!timer)
3271 		return;
3272 	comm_timer_disable(timer);
3273 	/* Free the sub struct timer->ev_timer derived from the super struct timer.
3274 	 * i.e. assert(timer == timer->ev_timer)
3275 	 */
3276 	ub_event_free(timer->ev_timer->ev);
3277 	free(timer->ev_timer);
3278 }
3279 
3280 void
3281 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
3282 {
3283 	struct comm_timer* tm = (struct comm_timer*)arg;
3284 	if(!(event&UB_EV_TIMEOUT))
3285 		return;
3286 	ub_comm_base_now(tm->ev_timer->base);
3287 	tm->ev_timer->enabled = 0;
3288 	fptr_ok(fptr_whitelist_comm_timer(tm->callback));
3289 	(*tm->callback)(tm->cb_arg);
3290 }
3291 
3292 int
3293 comm_timer_is_set(struct comm_timer* timer)
3294 {
3295 	return (int)timer->ev_timer->enabled;
3296 }
3297 
3298 size_t
3299 comm_timer_get_mem(struct comm_timer* ATTR_UNUSED(timer))
3300 {
3301 	return sizeof(struct internal_timer);
3302 }
3303 
3304 struct comm_signal*
3305 comm_signal_create(struct comm_base* base,
3306         void (*callback)(int, void*), void* cb_arg)
3307 {
3308 	struct comm_signal* com = (struct comm_signal*)malloc(
3309 		sizeof(struct comm_signal));
3310 	if(!com) {
3311 		log_err("malloc failed");
3312 		return NULL;
3313 	}
3314 	com->base = base;
3315 	com->callback = callback;
3316 	com->cb_arg = cb_arg;
3317 	com->ev_signal = NULL;
3318 	return com;
3319 }
3320 
3321 void
3322 comm_signal_callback(int sig, short event, void* arg)
3323 {
3324 	struct comm_signal* comsig = (struct comm_signal*)arg;
3325 	if(!(event & UB_EV_SIGNAL))
3326 		return;
3327 	ub_comm_base_now(comsig->base);
3328 	fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
3329 	(*comsig->callback)(sig, comsig->cb_arg);
3330 }
3331 
3332 int
3333 comm_signal_bind(struct comm_signal* comsig, int sig)
3334 {
3335 	struct internal_signal* entry = (struct internal_signal*)calloc(1,
3336 		sizeof(struct internal_signal));
3337 	if(!entry) {
3338 		log_err("malloc failed");
3339 		return 0;
3340 	}
3341 	log_assert(comsig);
3342 	/* add signal event */
3343 	entry->ev = ub_signal_new(comsig->base->eb->base, sig,
3344 		comm_signal_callback, comsig);
3345 	if(entry->ev == NULL) {
3346 		log_err("Could not create signal event");
3347 		free(entry);
3348 		return 0;
3349 	}
3350 	if(ub_signal_add(entry->ev, NULL) != 0) {
3351 		log_err("Could not add signal handler");
3352 		ub_event_free(entry->ev);
3353 		free(entry);
3354 		return 0;
3355 	}
3356 	/* link into list */
3357 	entry->next = comsig->ev_signal;
3358 	comsig->ev_signal = entry;
3359 	return 1;
3360 }
3361 
3362 void
3363 comm_signal_delete(struct comm_signal* comsig)
3364 {
3365 	struct internal_signal* p, *np;
3366 	if(!comsig)
3367 		return;
3368 	p=comsig->ev_signal;
3369 	while(p) {
3370 		np = p->next;
3371 		ub_signal_del(p->ev);
3372 		ub_event_free(p->ev);
3373 		free(p);
3374 		p = np;
3375 	}
3376 	free(comsig);
3377 }
3378