xref: /freebsd/contrib/unbound/util/netevent.c (revision 3d32dc633c5e21bf15dd0d968734efe72776afdc)
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/fptr_wlist.h"
47 #include "sldns/pkthdr.h"
48 #include "sldns/sbuffer.h"
49 #include "sldns/str2wire.h"
50 #include "dnstap/dnstap.h"
51 #include "dnscrypt/dnscrypt.h"
52 #ifdef HAVE_OPENSSL_SSL_H
53 #include <openssl/ssl.h>
54 #endif
55 #ifdef HAVE_OPENSSL_ERR_H
56 #include <openssl/err.h>
57 #endif
58 
59 /* -------- Start of local definitions -------- */
60 /** if CMSG_ALIGN is not defined on this platform, a workaround */
61 #ifndef CMSG_ALIGN
62 #  ifdef __CMSG_ALIGN
63 #    define CMSG_ALIGN(n) __CMSG_ALIGN(n)
64 #  elif defined(CMSG_DATA_ALIGN)
65 #    define CMSG_ALIGN _CMSG_DATA_ALIGN
66 #  else
67 #    define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
68 #  endif
69 #endif
70 
71 /** if CMSG_LEN is not defined on this platform, a workaround */
72 #ifndef CMSG_LEN
73 #  define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
74 #endif
75 
76 /** if CMSG_SPACE is not defined on this platform, a workaround */
77 #ifndef CMSG_SPACE
78 #  ifdef _CMSG_HDR_ALIGN
79 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
80 #  else
81 #    define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
82 #  endif
83 #endif
84 
85 /** The TCP reading or writing query timeout in milliseconds */
86 #define TCP_QUERY_TIMEOUT 120000
87 /** The TCP timeout in msec for fast queries, above half are used */
88 #define TCP_QUERY_TIMEOUT_FAST 200
89 
90 #ifndef NONBLOCKING_IS_BROKEN
91 /** number of UDP reads to perform per read indication from select */
92 #define NUM_UDP_PER_SELECT 100
93 #else
94 #define NUM_UDP_PER_SELECT 1
95 #endif
96 
97 /**
98  * The internal event structure for keeping ub_event info for the event.
99  * Possibly other structures (list, tree) this is part of.
100  */
101 struct internal_event {
102 	/** the comm base */
103 	struct comm_base* base;
104 	/** ub_event event type */
105 	struct ub_event* ev;
106 };
107 
108 /**
109  * Internal base structure, so that every thread has its own events.
110  */
111 struct internal_base {
112 	/** ub_event event_base type. */
113 	struct ub_event_base* base;
114 	/** seconds time pointer points here */
115 	time_t secs;
116 	/** timeval with current time */
117 	struct timeval now;
118 	/** the event used for slow_accept timeouts */
119 	struct ub_event* slow_accept;
120 	/** true if slow_accept is enabled */
121 	int slow_accept_enabled;
122 };
123 
124 /**
125  * Internal timer structure, to store timer event in.
126  */
127 struct internal_timer {
128 	/** the super struct from which derived */
129 	struct comm_timer super;
130 	/** the comm base */
131 	struct comm_base* base;
132 	/** ub_event event type */
133 	struct ub_event* ev;
134 	/** is timer enabled */
135 	uint8_t enabled;
136 };
137 
138 /**
139  * Internal signal structure, to store signal event in.
140  */
141 struct internal_signal {
142 	/** ub_event event type */
143 	struct ub_event* ev;
144 	/** next in signal list */
145 	struct internal_signal* next;
146 };
147 
148 /** create a tcp handler with a parent */
149 static struct comm_point* comm_point_create_tcp_handler(
150 	struct comm_base *base, struct comm_point* parent, size_t bufsize,
151         comm_point_callback_type* callback, void* callback_arg);
152 
153 /* -------- End of local definitions -------- */
154 
155 struct comm_base*
156 comm_base_create(int sigs)
157 {
158 	struct comm_base* b = (struct comm_base*)calloc(1,
159 		sizeof(struct comm_base));
160 	const char *evnm="event", *evsys="", *evmethod="";
161 
162 	if(!b)
163 		return NULL;
164 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
165 	if(!b->eb) {
166 		free(b);
167 		return NULL;
168 	}
169 	b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now);
170 	if(!b->eb->base) {
171 		free(b->eb);
172 		free(b);
173 		return NULL;
174 	}
175 	ub_comm_base_now(b);
176 	ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod);
177 	verbose(VERB_ALGO, "%s %s user %s method.", evnm, evsys, evmethod);
178 	return b;
179 }
180 
181 struct comm_base*
182 comm_base_create_event(struct ub_event_base* base)
183 {
184 	struct comm_base* b = (struct comm_base*)calloc(1,
185 		sizeof(struct comm_base));
186 	if(!b)
187 		return NULL;
188 	b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
189 	if(!b->eb) {
190 		free(b);
191 		return NULL;
192 	}
193 	b->eb->base = base;
194 	ub_comm_base_now(b);
195 	return b;
196 }
197 
198 void
199 comm_base_delete(struct comm_base* b)
200 {
201 	if(!b)
202 		return;
203 	if(b->eb->slow_accept_enabled) {
204 		if(ub_event_del(b->eb->slow_accept) != 0) {
205 			log_err("could not event_del slow_accept");
206 		}
207 		ub_event_free(b->eb->slow_accept);
208 	}
209 	ub_event_base_free(b->eb->base);
210 	b->eb->base = NULL;
211 	free(b->eb);
212 	free(b);
213 }
214 
215 void
216 comm_base_delete_no_base(struct comm_base* b)
217 {
218 	if(!b)
219 		return;
220 	if(b->eb->slow_accept_enabled) {
221 		if(ub_event_del(b->eb->slow_accept) != 0) {
222 			log_err("could not event_del slow_accept");
223 		}
224 		ub_event_free(b->eb->slow_accept);
225 	}
226 	b->eb->base = NULL;
227 	free(b->eb);
228 	free(b);
229 }
230 
231 void
232 comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv)
233 {
234 	*tt = &b->eb->secs;
235 	*tv = &b->eb->now;
236 }
237 
238 void
239 comm_base_dispatch(struct comm_base* b)
240 {
241 	int retval;
242 	retval = ub_event_base_dispatch(b->eb->base);
243 	if(retval < 0) {
244 		fatal_exit("event_dispatch returned error %d, "
245 			"errno is %s", retval, strerror(errno));
246 	}
247 }
248 
249 void comm_base_exit(struct comm_base* b)
250 {
251 	if(ub_event_base_loopexit(b->eb->base) != 0) {
252 		log_err("Could not loopexit");
253 	}
254 }
255 
256 void comm_base_set_slow_accept_handlers(struct comm_base* b,
257 	void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
258 {
259 	b->stop_accept = stop_acc;
260 	b->start_accept = start_acc;
261 	b->cb_arg = arg;
262 }
263 
264 struct ub_event_base* comm_base_internal(struct comm_base* b)
265 {
266 	return b->eb->base;
267 }
268 
269 /** see if errno for udp has to be logged or not uses globals */
270 static int
271 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
272 {
273 	/* do not log transient errors (unless high verbosity) */
274 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
275 	switch(errno) {
276 #  ifdef ENETUNREACH
277 		case ENETUNREACH:
278 #  endif
279 #  ifdef EHOSTDOWN
280 		case EHOSTDOWN:
281 #  endif
282 #  ifdef EHOSTUNREACH
283 		case EHOSTUNREACH:
284 #  endif
285 #  ifdef ENETDOWN
286 		case ENETDOWN:
287 #  endif
288 			if(verbosity < VERB_ALGO)
289 				return 0;
290 		default:
291 			break;
292 	}
293 #endif
294 	/* permission denied is gotten for every send if the
295 	 * network is disconnected (on some OS), squelch it */
296 	if( ((errno == EPERM)
297 #  ifdef EADDRNOTAVAIL
298 		/* 'Cannot assign requested address' also when disconnected */
299 		|| (errno == EADDRNOTAVAIL)
300 #  endif
301 		) && verbosity < VERB_DETAIL)
302 		return 0;
303 #  ifdef EADDRINUSE
304 	/* If SO_REUSEADDR is set, we could try to connect to the same server
305 	 * from the same source port twice. */
306 	if(errno == EADDRINUSE && verbosity < VERB_DETAIL)
307 		return 0;
308 #  endif
309 	/* squelch errors where people deploy AAAA ::ffff:bla for
310 	 * authority servers, which we try for intranets. */
311 	if(errno == EINVAL && addr_is_ip4mapped(
312 		(struct sockaddr_storage*)addr, addrlen) &&
313 		verbosity < VERB_DETAIL)
314 		return 0;
315 	/* SO_BROADCAST sockopt can give access to 255.255.255.255,
316 	 * but a dns cache does not need it. */
317 	if(errno == EACCES && addr_is_broadcast(
318 		(struct sockaddr_storage*)addr, addrlen) &&
319 		verbosity < VERB_DETAIL)
320 		return 0;
321 	return 1;
322 }
323 
324 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
325 {
326 	return udp_send_errno_needs_log(addr, addrlen);
327 }
328 
329 /* send a UDP reply */
330 int
331 comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet,
332 	struct sockaddr* addr, socklen_t addrlen)
333 {
334 	ssize_t sent;
335 	log_assert(c->fd != -1);
336 #ifdef UNBOUND_DEBUG
337 	if(sldns_buffer_remaining(packet) == 0)
338 		log_err("error: send empty UDP packet");
339 #endif
340 	log_assert(addr && addrlen > 0);
341 	sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
342 		sldns_buffer_remaining(packet), 0,
343 		addr, addrlen);
344 	if(sent == -1) {
345 		/* try again and block, waiting for IO to complete,
346 		 * we want to send the answer, and we will wait for
347 		 * the ethernet interface buffer to have space. */
348 #ifndef USE_WINSOCK
349 		if(errno == EAGAIN ||
350 #  ifdef EWOULDBLOCK
351 			errno == EWOULDBLOCK ||
352 #  endif
353 			errno == ENOBUFS) {
354 #else
355 		if(WSAGetLastError() == WSAEINPROGRESS ||
356 			WSAGetLastError() == WSAENOBUFS ||
357 			WSAGetLastError() == WSAEWOULDBLOCK) {
358 #endif
359 			int e;
360 			fd_set_block(c->fd);
361 			sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
362 				sldns_buffer_remaining(packet), 0,
363 				addr, addrlen);
364 			e = errno;
365 			fd_set_nonblock(c->fd);
366 			errno = e;
367 		}
368 	}
369 	if(sent == -1) {
370 		if(!udp_send_errno_needs_log(addr, addrlen))
371 			return 0;
372 #ifndef USE_WINSOCK
373 		verbose(VERB_OPS, "sendto failed: %s", strerror(errno));
374 #else
375 		verbose(VERB_OPS, "sendto failed: %s",
376 			wsa_strerror(WSAGetLastError()));
377 #endif
378 		log_addr(VERB_OPS, "remote address is",
379 			(struct sockaddr_storage*)addr, addrlen);
380 		return 0;
381 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
382 		log_err("sent %d in place of %d bytes",
383 			(int)sent, (int)sldns_buffer_remaining(packet));
384 		return 0;
385 	}
386 	return 1;
387 }
388 
389 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
390 /** print debug ancillary info */
391 static void p_ancil(const char* str, struct comm_reply* r)
392 {
393 	if(r->srctype != 4 && r->srctype != 6) {
394 		log_info("%s: unknown srctype %d", str, r->srctype);
395 		return;
396 	}
397 	if(r->srctype == 6) {
398 		char buf[1024];
399 		if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
400 			buf, (socklen_t)sizeof(buf)) == 0) {
401 			(void)strlcpy(buf, "(inet_ntop error)", sizeof(buf));
402 		}
403 		buf[sizeof(buf)-1]=0;
404 		log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
405 	} else if(r->srctype == 4) {
406 #ifdef IP_PKTINFO
407 		char buf1[1024], buf2[1024];
408 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
409 			buf1, (socklen_t)sizeof(buf1)) == 0) {
410 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
411 		}
412 		buf1[sizeof(buf1)-1]=0;
413 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
414 		if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
415 			buf2, (socklen_t)sizeof(buf2)) == 0) {
416 			(void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2));
417 		}
418 		buf2[sizeof(buf2)-1]=0;
419 #else
420 		buf2[0]=0;
421 #endif
422 		log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
423 			buf1, buf2);
424 #elif defined(IP_RECVDSTADDR)
425 		char buf1[1024];
426 		if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
427 			buf1, (socklen_t)sizeof(buf1)) == 0) {
428 			(void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
429 		}
430 		buf1[sizeof(buf1)-1]=0;
431 		log_info("%s: %s", str, buf1);
432 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */
433 	}
434 }
435 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */
436 
437 /** send a UDP reply over specified interface*/
438 static int
439 comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet,
440 	struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
441 {
442 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
443 	ssize_t sent;
444 	struct msghdr msg;
445 	struct iovec iov[1];
446 	char control[256];
447 #ifndef S_SPLINT_S
448 	struct cmsghdr *cmsg;
449 #endif /* S_SPLINT_S */
450 
451 	log_assert(c->fd != -1);
452 #ifdef UNBOUND_DEBUG
453 	if(sldns_buffer_remaining(packet) == 0)
454 		log_err("error: send empty UDP packet");
455 #endif
456 	log_assert(addr && addrlen > 0);
457 
458 	msg.msg_name = addr;
459 	msg.msg_namelen = addrlen;
460 	iov[0].iov_base = sldns_buffer_begin(packet);
461 	iov[0].iov_len = sldns_buffer_remaining(packet);
462 	msg.msg_iov = iov;
463 	msg.msg_iovlen = 1;
464 	msg.msg_control = control;
465 #ifndef S_SPLINT_S
466 	msg.msg_controllen = sizeof(control);
467 #endif /* S_SPLINT_S */
468 	msg.msg_flags = 0;
469 
470 #ifndef S_SPLINT_S
471 	cmsg = CMSG_FIRSTHDR(&msg);
472 	if(r->srctype == 4) {
473 #ifdef IP_PKTINFO
474 		void* cmsg_data;
475 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
476 		log_assert(msg.msg_controllen <= sizeof(control));
477 		cmsg->cmsg_level = IPPROTO_IP;
478 		cmsg->cmsg_type = IP_PKTINFO;
479 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
480 			sizeof(struct in_pktinfo));
481 		/* unset the ifindex to not bypass the routing tables */
482 		cmsg_data = CMSG_DATA(cmsg);
483 		((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0;
484 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
485 #elif defined(IP_SENDSRCADDR)
486 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
487 		log_assert(msg.msg_controllen <= sizeof(control));
488 		cmsg->cmsg_level = IPPROTO_IP;
489 		cmsg->cmsg_type = IP_SENDSRCADDR;
490 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
491 			sizeof(struct in_addr));
492 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
493 #else
494 		verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
495 		msg.msg_control = NULL;
496 #endif /* IP_PKTINFO or IP_SENDSRCADDR */
497 	} else if(r->srctype == 6) {
498 		void* cmsg_data;
499 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
500 		log_assert(msg.msg_controllen <= sizeof(control));
501 		cmsg->cmsg_level = IPPROTO_IPV6;
502 		cmsg->cmsg_type = IPV6_PKTINFO;
503 		memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
504 			sizeof(struct in6_pktinfo));
505 		/* unset the ifindex to not bypass the routing tables */
506 		cmsg_data = CMSG_DATA(cmsg);
507 		((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0;
508 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
509 	} else {
510 		/* try to pass all 0 to use default route */
511 		msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
512 		log_assert(msg.msg_controllen <= sizeof(control));
513 		cmsg->cmsg_level = IPPROTO_IPV6;
514 		cmsg->cmsg_type = IPV6_PKTINFO;
515 		memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
516 		cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
517 	}
518 #endif /* S_SPLINT_S */
519 	if(verbosity >= VERB_ALGO)
520 		p_ancil("send_udp over interface", r);
521 	sent = sendmsg(c->fd, &msg, 0);
522 	if(sent == -1) {
523 		/* try again and block, waiting for IO to complete,
524 		 * we want to send the answer, and we will wait for
525 		 * the ethernet interface buffer to have space. */
526 #ifndef USE_WINSOCK
527 		if(errno == EAGAIN ||
528 #  ifdef EWOULDBLOCK
529 			errno == EWOULDBLOCK ||
530 #  endif
531 			errno == ENOBUFS) {
532 #else
533 		if(WSAGetLastError() == WSAEINPROGRESS ||
534 			WSAGetLastError() == WSAENOBUFS ||
535 			WSAGetLastError() == WSAEWOULDBLOCK) {
536 #endif
537 			int e;
538 			fd_set_block(c->fd);
539 			sent = sendmsg(c->fd, &msg, 0);
540 			e = errno;
541 			fd_set_nonblock(c->fd);
542 			errno = e;
543 		}
544 	}
545 	if(sent == -1) {
546 		if(!udp_send_errno_needs_log(addr, addrlen))
547 			return 0;
548 		verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
549 		log_addr(VERB_OPS, "remote address is",
550 			(struct sockaddr_storage*)addr, addrlen);
551 #ifdef __NetBSD__
552 		/* netbsd 7 has IP_PKTINFO for recv but not send */
553 		if(errno == EINVAL && r->srctype == 4)
554 			log_err("sendmsg: No support for sendmsg(IP_PKTINFO). "
555 				"Please disable interface-automatic");
556 #endif
557 		return 0;
558 	} else if((size_t)sent != sldns_buffer_remaining(packet)) {
559 		log_err("sent %d in place of %d bytes",
560 			(int)sent, (int)sldns_buffer_remaining(packet));
561 		return 0;
562 	}
563 	return 1;
564 #else
565 	(void)c;
566 	(void)packet;
567 	(void)addr;
568 	(void)addrlen;
569 	(void)r;
570 	log_err("sendmsg: IPV6_PKTINFO not supported");
571 	return 0;
572 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
573 }
574 
575 void
576 comm_point_udp_ancil_callback(int fd, short event, void* arg)
577 {
578 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
579 	struct comm_reply rep;
580 	struct msghdr msg;
581 	struct iovec iov[1];
582 	ssize_t rcv;
583 	char ancil[256];
584 	int i;
585 #ifndef S_SPLINT_S
586 	struct cmsghdr* cmsg;
587 #endif /* S_SPLINT_S */
588 
589 	rep.c = (struct comm_point*)arg;
590 	log_assert(rep.c->type == comm_udp);
591 
592 	if(!(event&UB_EV_READ))
593 		return;
594 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
595 	ub_comm_base_now(rep.c->ev->base);
596 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
597 		sldns_buffer_clear(rep.c->buffer);
598 		rep.addrlen = (socklen_t)sizeof(rep.addr);
599 		log_assert(fd != -1);
600 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
601 		msg.msg_name = &rep.addr;
602 		msg.msg_namelen = (socklen_t)sizeof(rep.addr);
603 		iov[0].iov_base = sldns_buffer_begin(rep.c->buffer);
604 		iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer);
605 		msg.msg_iov = iov;
606 		msg.msg_iovlen = 1;
607 		msg.msg_control = ancil;
608 #ifndef S_SPLINT_S
609 		msg.msg_controllen = sizeof(ancil);
610 #endif /* S_SPLINT_S */
611 		msg.msg_flags = 0;
612 		rcv = recvmsg(fd, &msg, 0);
613 		if(rcv == -1) {
614 			if(errno != EAGAIN && errno != EINTR) {
615 				log_err("recvmsg failed: %s", strerror(errno));
616 			}
617 			return;
618 		}
619 		rep.addrlen = msg.msg_namelen;
620 		sldns_buffer_skip(rep.c->buffer, rcv);
621 		sldns_buffer_flip(rep.c->buffer);
622 		rep.srctype = 0;
623 #ifndef S_SPLINT_S
624 		for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
625 			cmsg = CMSG_NXTHDR(&msg, cmsg)) {
626 			if( cmsg->cmsg_level == IPPROTO_IPV6 &&
627 				cmsg->cmsg_type == IPV6_PKTINFO) {
628 				rep.srctype = 6;
629 				memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
630 					sizeof(struct in6_pktinfo));
631 				break;
632 #ifdef IP_PKTINFO
633 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
634 				cmsg->cmsg_type == IP_PKTINFO) {
635 				rep.srctype = 4;
636 				memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
637 					sizeof(struct in_pktinfo));
638 				break;
639 #elif defined(IP_RECVDSTADDR)
640 			} else if( cmsg->cmsg_level == IPPROTO_IP &&
641 				cmsg->cmsg_type == IP_RECVDSTADDR) {
642 				rep.srctype = 4;
643 				memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
644 					sizeof(struct in_addr));
645 				break;
646 #endif /* IP_PKTINFO or IP_RECVDSTADDR */
647 			}
648 		}
649 		if(verbosity >= VERB_ALGO)
650 			p_ancil("receive_udp on interface", &rep);
651 #endif /* S_SPLINT_S */
652 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
653 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
654 			/* send back immediate reply */
655 			(void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer,
656 				(struct sockaddr*)&rep.addr, rep.addrlen, &rep);
657 		}
658 		if(!rep.c || rep.c->fd == -1) /* commpoint closed */
659 			break;
660 	}
661 #else
662 	(void)fd;
663 	(void)event;
664 	(void)arg;
665 	fatal_exit("recvmsg: No support for IPV6_PKTINFO; IP_PKTINFO or IP_RECVDSTADDR. "
666 		"Please disable interface-automatic");
667 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */
668 }
669 
670 void
671 comm_point_udp_callback(int fd, short event, void* arg)
672 {
673 	struct comm_reply rep;
674 	ssize_t rcv;
675 	int i;
676 	struct sldns_buffer *buffer;
677 
678 	rep.c = (struct comm_point*)arg;
679 	log_assert(rep.c->type == comm_udp);
680 
681 	if(!(event&UB_EV_READ))
682 		return;
683 	log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
684 	ub_comm_base_now(rep.c->ev->base);
685 	for(i=0; i<NUM_UDP_PER_SELECT; i++) {
686 		sldns_buffer_clear(rep.c->buffer);
687 		rep.addrlen = (socklen_t)sizeof(rep.addr);
688 		log_assert(fd != -1);
689 		log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
690 		rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer),
691 			sldns_buffer_remaining(rep.c->buffer), 0,
692 			(struct sockaddr*)&rep.addr, &rep.addrlen);
693 		if(rcv == -1) {
694 #ifndef USE_WINSOCK
695 			if(errno != EAGAIN && errno != EINTR)
696 				log_err("recvfrom %d failed: %s",
697 					fd, strerror(errno));
698 #else
699 			if(WSAGetLastError() != WSAEINPROGRESS &&
700 				WSAGetLastError() != WSAECONNRESET &&
701 				WSAGetLastError()!= WSAEWOULDBLOCK)
702 				log_err("recvfrom failed: %s",
703 					wsa_strerror(WSAGetLastError()));
704 #endif
705 			return;
706 		}
707 		sldns_buffer_skip(rep.c->buffer, rcv);
708 		sldns_buffer_flip(rep.c->buffer);
709 		rep.srctype = 0;
710 		fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
711 		if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
712 			/* send back immediate reply */
713 #ifdef USE_DNSCRYPT
714 			buffer = rep.c->dnscrypt_buffer;
715 #else
716 			buffer = rep.c->buffer;
717 #endif
718 			(void)comm_point_send_udp_msg(rep.c, buffer,
719 				(struct sockaddr*)&rep.addr, rep.addrlen);
720 		}
721 		if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for
722 		another UDP port. Note rep.c cannot be reused with TCP fd. */
723 			break;
724 	}
725 }
726 
727 /** Use a new tcp handler for new query fd, set to read query */
728 static void
729 setup_tcp_handler(struct comm_point* c, int fd, int cur, int max)
730 {
731 	log_assert(c->type == comm_tcp);
732 	log_assert(c->fd == -1);
733 	sldns_buffer_clear(c->buffer);
734 #ifdef USE_DNSCRYPT
735 	if (c->dnscrypt)
736 		sldns_buffer_clear(c->dnscrypt_buffer);
737 #endif
738 	c->tcp_is_reading = 1;
739 	c->tcp_byte_count = 0;
740 	c->tcp_timeout_msec = TCP_QUERY_TIMEOUT;
741 	/* if more than half the tcp handlers are in use, use a shorter
742 	 * timeout for this TCP connection, we need to make space for
743 	 * other connections to be able to get attention */
744 	if(cur > max/2)
745 		c->tcp_timeout_msec = TCP_QUERY_TIMEOUT_FAST;
746 	comm_point_start_listening(c, fd, c->tcp_timeout_msec);
747 }
748 
749 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
750 	short ATTR_UNUSED(event), void* arg)
751 {
752 	struct comm_base* b = (struct comm_base*)arg;
753 	/* timeout for the slow accept, re-enable accepts again */
754 	if(b->start_accept) {
755 		verbose(VERB_ALGO, "wait is over, slow accept disabled");
756 		fptr_ok(fptr_whitelist_start_accept(b->start_accept));
757 		(*b->start_accept)(b->cb_arg);
758 		b->eb->slow_accept_enabled = 0;
759 	}
760 }
761 
762 int comm_point_perform_accept(struct comm_point* c,
763 	struct sockaddr_storage* addr, socklen_t* addrlen)
764 {
765 	int new_fd;
766 	*addrlen = (socklen_t)sizeof(*addr);
767 #ifndef HAVE_ACCEPT4
768 	new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
769 #else
770 	/* SOCK_NONBLOCK saves extra calls to fcntl for the same result */
771 	new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK);
772 #endif
773 	if(new_fd == -1) {
774 #ifndef USE_WINSOCK
775 		/* EINTR is signal interrupt. others are closed connection. */
776 		if(	errno == EINTR || errno == EAGAIN
777 #ifdef EWOULDBLOCK
778 			|| errno == EWOULDBLOCK
779 #endif
780 #ifdef ECONNABORTED
781 			|| errno == ECONNABORTED
782 #endif
783 #ifdef EPROTO
784 			|| errno == EPROTO
785 #endif /* EPROTO */
786 			)
787 			return -1;
788 #if defined(ENFILE) && defined(EMFILE)
789 		if(errno == ENFILE || errno == EMFILE) {
790 			/* out of file descriptors, likely outside of our
791 			 * control. stop accept() calls for some time */
792 			if(c->ev->base->stop_accept) {
793 				struct comm_base* b = c->ev->base;
794 				struct timeval tv;
795 				verbose(VERB_ALGO, "out of file descriptors: "
796 					"slow accept");
797 				b->eb->slow_accept_enabled = 1;
798 				fptr_ok(fptr_whitelist_stop_accept(
799 					b->stop_accept));
800 				(*b->stop_accept)(b->cb_arg);
801 				/* set timeout, no mallocs */
802 				tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
803 				tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000;
804 				b->eb->slow_accept = ub_event_new(b->eb->base,
805 					-1, UB_EV_TIMEOUT,
806 					comm_base_handle_slow_accept, b);
807 				if(b->eb->slow_accept == NULL) {
808 					/* we do not want to log here, because
809 					 * that would spam the logfiles.
810 					 * error: "event_base_set failed." */
811 				}
812 				else if(ub_event_add(b->eb->slow_accept, &tv)
813 					!= 0) {
814 					/* we do not want to log here,
815 					 * error: "event_add failed." */
816 				}
817 			}
818 			return -1;
819 		}
820 #endif
821 		log_err_addr("accept failed", strerror(errno), addr, *addrlen);
822 #else /* USE_WINSOCK */
823 		if(WSAGetLastError() == WSAEINPROGRESS ||
824 			WSAGetLastError() == WSAECONNRESET)
825 			return -1;
826 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
827 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
828 			return -1;
829 		}
830 		log_err_addr("accept failed", wsa_strerror(WSAGetLastError()),
831 			addr, *addrlen);
832 #endif
833 		return -1;
834 	}
835 #ifndef HAVE_ACCEPT4
836 	fd_set_nonblock(new_fd);
837 #endif
838 	return new_fd;
839 }
840 
841 #ifdef USE_WINSOCK
842 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
843         int ATTR_UNUSED(argi), long argl, long retvalue)
844 {
845 	int wsa_err = WSAGetLastError(); /* store errcode before it is gone */
846 	verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
847 		(oper&BIO_CB_RETURN)?"return":"before",
848 		(oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
849 		wsa_err==WSAEWOULDBLOCK?"wsawb":"");
850 	/* on windows, check if previous operation caused EWOULDBLOCK */
851 	if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
852 		(oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
853 		if(wsa_err == WSAEWOULDBLOCK)
854 			ub_winsock_tcp_wouldblock((struct ub_event*)
855 				BIO_get_callback_arg(b), UB_EV_READ);
856 	}
857 	if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
858 		(oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
859 		if(wsa_err == WSAEWOULDBLOCK)
860 			ub_winsock_tcp_wouldblock((struct ub_event*)
861 				BIO_get_callback_arg(b), UB_EV_WRITE);
862 	}
863 	/* return original return value */
864 	return retvalue;
865 }
866 
867 /** set win bio callbacks for nonblocking operations */
868 void
869 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
870 {
871 	SSL* ssl = (SSL*)thessl;
872 	/* set them both just in case, but usually they are the same BIO */
873 	BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
874 	BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev);
875 	BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
876 	BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev);
877 }
878 #endif
879 
880 void
881 comm_point_tcp_accept_callback(int fd, short event, void* arg)
882 {
883 	struct comm_point* c = (struct comm_point*)arg, *c_hdl;
884 	int new_fd;
885 	log_assert(c->type == comm_tcp_accept);
886 	if(!(event & UB_EV_READ)) {
887 		log_info("ignoring tcp accept event %d", (int)event);
888 		return;
889 	}
890 	ub_comm_base_now(c->ev->base);
891 	/* find free tcp handler. */
892 	if(!c->tcp_free) {
893 		log_warn("accepted too many tcp, connections full");
894 		return;
895 	}
896 	/* accept incoming connection. */
897 	c_hdl = c->tcp_free;
898 	log_assert(fd != -1);
899 	(void)fd;
900 	new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr,
901 		&c_hdl->repinfo.addrlen);
902 	if(new_fd == -1)
903 		return;
904 	if(c->ssl) {
905 		c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
906 		if(!c_hdl->ssl) {
907 			c_hdl->fd = new_fd;
908 			comm_point_close(c_hdl);
909 			return;
910 		}
911 		c_hdl->ssl_shake_state = comm_ssl_shake_read;
912 #ifdef USE_WINSOCK
913 		comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
914 #endif
915 	}
916 
917 	/* grab the tcp handler buffers */
918 	c->cur_tcp_count++;
919 	c->tcp_free = c_hdl->tcp_free;
920 	if(!c->tcp_free) {
921 		/* stop accepting incoming queries for now. */
922 		comm_point_stop_listening(c);
923 	}
924 	setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count);
925 }
926 
927 /** Make tcp handler free for next assignment */
928 static void
929 reclaim_tcp_handler(struct comm_point* c)
930 {
931 	log_assert(c->type == comm_tcp);
932 	if(c->ssl) {
933 #ifdef HAVE_SSL
934 		SSL_shutdown(c->ssl);
935 		SSL_free(c->ssl);
936 		c->ssl = NULL;
937 #endif
938 	}
939 	comm_point_close(c);
940 	if(c->tcp_parent) {
941 		c->tcp_parent->cur_tcp_count--;
942 		c->tcp_free = c->tcp_parent->tcp_free;
943 		c->tcp_parent->tcp_free = c;
944 		if(!c->tcp_free) {
945 			/* re-enable listening on accept socket */
946 			comm_point_start_listening(c->tcp_parent, -1, -1);
947 		}
948 	}
949 }
950 
951 /** do the callback when writing is done */
952 static void
953 tcp_callback_writer(struct comm_point* c)
954 {
955 	log_assert(c->type == comm_tcp);
956 	sldns_buffer_clear(c->buffer);
957 	if(c->tcp_do_toggle_rw)
958 		c->tcp_is_reading = 1;
959 	c->tcp_byte_count = 0;
960 	/* switch from listening(write) to listening(read) */
961 	comm_point_stop_listening(c);
962 	comm_point_start_listening(c, -1, -1);
963 }
964 
965 /** do the callback when reading is done */
966 static void
967 tcp_callback_reader(struct comm_point* c)
968 {
969 	log_assert(c->type == comm_tcp || c->type == comm_local);
970 	sldns_buffer_flip(c->buffer);
971 	if(c->tcp_do_toggle_rw)
972 		c->tcp_is_reading = 0;
973 	c->tcp_byte_count = 0;
974 	if(c->type == comm_tcp)
975 		comm_point_stop_listening(c);
976 	fptr_ok(fptr_whitelist_comm_point(c->callback));
977 	if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
978 		comm_point_start_listening(c, -1, c->tcp_timeout_msec);
979 	}
980 }
981 
982 #ifdef HAVE_SSL
983 /** log certificate details */
984 static void
985 log_cert(unsigned level, const char* str, X509* cert)
986 {
987 	BIO* bio;
988 	char nul = 0;
989 	char* pp = NULL;
990 	long len;
991 	if(verbosity < level) return;
992 	bio = BIO_new(BIO_s_mem());
993 	if(!bio) return;
994 	X509_print_ex(bio, cert, 0, (unsigned long)-1
995 		^(X509_FLAG_NO_SUBJECT
996                         |X509_FLAG_NO_ISSUER|X509_FLAG_NO_VALIDITY
997 			|X509_FLAG_NO_EXTENSIONS|X509_FLAG_NO_AUX
998 			|X509_FLAG_NO_ATTRIBUTES));
999 	BIO_write(bio, &nul, (int)sizeof(nul));
1000 	len = BIO_get_mem_data(bio, &pp);
1001 	if(len != 0 && pp) {
1002 		verbose(level, "%s: \n%s", str, pp);
1003 	}
1004 	BIO_free(bio);
1005 }
1006 #endif /* HAVE_SSL */
1007 
1008 /** continue ssl handshake */
1009 #ifdef HAVE_SSL
1010 static int
1011 ssl_handshake(struct comm_point* c)
1012 {
1013 	int r;
1014 	if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
1015 		/* read condition satisfied back to writing */
1016 		comm_point_listen_for_rw(c, 1, 1);
1017 		c->ssl_shake_state = comm_ssl_shake_none;
1018 		return 1;
1019 	}
1020 	if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
1021 		/* write condition satisfied, back to reading */
1022 		comm_point_listen_for_rw(c, 1, 0);
1023 		c->ssl_shake_state = comm_ssl_shake_none;
1024 		return 1;
1025 	}
1026 
1027 	ERR_clear_error();
1028 	r = SSL_do_handshake(c->ssl);
1029 	if(r != 1) {
1030 		int want = SSL_get_error(c->ssl, r);
1031 		if(want == SSL_ERROR_WANT_READ) {
1032 			if(c->ssl_shake_state == comm_ssl_shake_read)
1033 				return 1;
1034 			c->ssl_shake_state = comm_ssl_shake_read;
1035 			comm_point_listen_for_rw(c, 1, 0);
1036 			return 1;
1037 		} else if(want == SSL_ERROR_WANT_WRITE) {
1038 			if(c->ssl_shake_state == comm_ssl_shake_write)
1039 				return 1;
1040 			c->ssl_shake_state = comm_ssl_shake_write;
1041 			comm_point_listen_for_rw(c, 0, 1);
1042 			return 1;
1043 		} else if(r == 0) {
1044 			return 0; /* closed */
1045 		} else if(want == SSL_ERROR_SYSCALL) {
1046 			/* SYSCALL and errno==0 means closed uncleanly */
1047 			if(errno != 0)
1048 				log_err("SSL_handshake syscall: %s",
1049 					strerror(errno));
1050 			return 0;
1051 		} else {
1052 			log_crypto_err("ssl handshake failed");
1053 			log_addr(1, "ssl handshake failed", &c->repinfo.addr,
1054 				c->repinfo.addrlen);
1055 			return 0;
1056 		}
1057 	}
1058 	/* this is where peer verification could take place */
1059 	if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) {
1060 		/* verification */
1061 		if(SSL_get_verify_result(c->ssl) == X509_V_OK) {
1062 			X509* x = SSL_get_peer_certificate(c->ssl);
1063 			if(!x) {
1064 				log_addr(VERB_ALGO, "SSL connection failed: "
1065 					"no certificate",
1066 					&c->repinfo.addr, c->repinfo.addrlen);
1067 				return 0;
1068 			}
1069 			log_cert(VERB_ALGO, "peer certificate", x);
1070 #ifdef HAVE_SSL_GET0_PEERNAME
1071 			if(SSL_get0_peername(c->ssl)) {
1072 				char buf[255];
1073 				snprintf(buf, sizeof(buf), "SSL connection "
1074 					"to %s authenticated",
1075 					SSL_get0_peername(c->ssl));
1076 				log_addr(VERB_ALGO, buf, &c->repinfo.addr,
1077 					c->repinfo.addrlen);
1078 			} else {
1079 #endif
1080 				log_addr(VERB_ALGO, "SSL connection "
1081 					"authenticated", &c->repinfo.addr,
1082 					c->repinfo.addrlen);
1083 #ifdef HAVE_SSL_GET0_PEERNAME
1084 			}
1085 #endif
1086 			X509_free(x);
1087 		} else {
1088 			X509* x = SSL_get_peer_certificate(c->ssl);
1089 			if(x) {
1090 				log_cert(VERB_ALGO, "peer certificate", x);
1091 				X509_free(x);
1092 			}
1093 			log_addr(VERB_ALGO, "SSL connection failed: "
1094 				"failed to authenticate",
1095 				&c->repinfo.addr, c->repinfo.addrlen);
1096 			return 0;
1097 		}
1098 	} else {
1099 		/* unauthenticated, the verify peer flag was not set
1100 		 * in c->ssl when the ssl object was created from ssl_ctx */
1101 		log_addr(VERB_ALGO, "SSL connection", &c->repinfo.addr,
1102 			c->repinfo.addrlen);
1103 	}
1104 
1105 	/* setup listen rw correctly */
1106 	if(c->tcp_is_reading) {
1107 		if(c->ssl_shake_state != comm_ssl_shake_read)
1108 			comm_point_listen_for_rw(c, 1, 0);
1109 	} else {
1110 		comm_point_listen_for_rw(c, 1, 1);
1111 	}
1112 	c->ssl_shake_state = comm_ssl_shake_none;
1113 	return 1;
1114 }
1115 #endif /* HAVE_SSL */
1116 
1117 /** ssl read callback on TCP */
1118 static int
1119 ssl_handle_read(struct comm_point* c)
1120 {
1121 #ifdef HAVE_SSL
1122 	int r;
1123 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1124 		if(!ssl_handshake(c))
1125 			return 0;
1126 		if(c->ssl_shake_state != comm_ssl_shake_none)
1127 			return 1;
1128 	}
1129 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1130 		/* read length bytes */
1131 		ERR_clear_error();
1132 		if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer,
1133 			c->tcp_byte_count), (int)(sizeof(uint16_t) -
1134 			c->tcp_byte_count))) <= 0) {
1135 			int want = SSL_get_error(c->ssl, r);
1136 			if(want == SSL_ERROR_ZERO_RETURN) {
1137 				return 0; /* shutdown, closed */
1138 			} else if(want == SSL_ERROR_WANT_READ) {
1139 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1140 				return 1; /* read more later */
1141 			} else if(want == SSL_ERROR_WANT_WRITE) {
1142 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1143 				comm_point_listen_for_rw(c, 0, 1);
1144 				return 1;
1145 			} else if(want == SSL_ERROR_SYSCALL) {
1146 				if(errno != 0)
1147 					log_err("SSL_read syscall: %s",
1148 						strerror(errno));
1149 				return 0;
1150 			}
1151 			log_crypto_err("could not SSL_read");
1152 			return 0;
1153 		}
1154 		c->tcp_byte_count += r;
1155 		if(c->tcp_byte_count < sizeof(uint16_t))
1156 			return 1;
1157 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1158 			sldns_buffer_capacity(c->buffer)) {
1159 			verbose(VERB_QUERY, "ssl: dropped larger than buffer");
1160 			return 0;
1161 		}
1162 		sldns_buffer_set_limit(c->buffer,
1163 			sldns_buffer_read_u16_at(c->buffer, 0));
1164 		if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1165 			verbose(VERB_QUERY, "ssl: dropped bogus too short.");
1166 			return 0;
1167 		}
1168 		sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t)));
1169 		verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
1170 			(int)sldns_buffer_limit(c->buffer));
1171 	}
1172 	if(sldns_buffer_remaining(c->buffer) > 0) {
1173 		ERR_clear_error();
1174 		r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1175 			(int)sldns_buffer_remaining(c->buffer));
1176 		if(r <= 0) {
1177 			int want = SSL_get_error(c->ssl, r);
1178 			if(want == SSL_ERROR_ZERO_RETURN) {
1179 				return 0; /* shutdown, closed */
1180 			} else if(want == SSL_ERROR_WANT_READ) {
1181 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1182 				return 1; /* read more later */
1183 			} else if(want == SSL_ERROR_WANT_WRITE) {
1184 				c->ssl_shake_state = comm_ssl_shake_hs_write;
1185 				comm_point_listen_for_rw(c, 0, 1);
1186 				return 1;
1187 			} else if(want == SSL_ERROR_SYSCALL) {
1188 				if(errno != 0)
1189 					log_err("SSL_read syscall: %s",
1190 						strerror(errno));
1191 				return 0;
1192 			}
1193 			log_crypto_err("could not SSL_read");
1194 			return 0;
1195 		}
1196 		sldns_buffer_skip(c->buffer, (ssize_t)r);
1197 	}
1198 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1199 		tcp_callback_reader(c);
1200 	}
1201 	return 1;
1202 #else
1203 	(void)c;
1204 	return 0;
1205 #endif /* HAVE_SSL */
1206 }
1207 
1208 /** ssl write callback on TCP */
1209 static int
1210 ssl_handle_write(struct comm_point* c)
1211 {
1212 #ifdef HAVE_SSL
1213 	int r;
1214 	if(c->ssl_shake_state != comm_ssl_shake_none) {
1215 		if(!ssl_handshake(c))
1216 			return 0;
1217 		if(c->ssl_shake_state != comm_ssl_shake_none)
1218 			return 1;
1219 	}
1220 	/* ignore return, if fails we may simply block */
1221 	(void)SSL_set_mode(c->ssl, SSL_MODE_ENABLE_PARTIAL_WRITE);
1222 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1223 		uint16_t len = htons(sldns_buffer_limit(c->buffer));
1224 		ERR_clear_error();
1225 		if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) <
1226 			LDNS_RR_BUF_SIZE) {
1227 			/* combine the tcp length and the query for write,
1228 			 * this emulates writev */
1229 			uint8_t buf[LDNS_RR_BUF_SIZE];
1230 			memmove(buf, &len, sizeof(uint16_t));
1231 			memmove(buf+sizeof(uint16_t),
1232 				sldns_buffer_current(c->buffer),
1233 				sldns_buffer_remaining(c->buffer));
1234 			r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count),
1235 				(int)(sizeof(uint16_t)+
1236 				sldns_buffer_remaining(c->buffer)
1237 				- c->tcp_byte_count));
1238 		} else {
1239 			r = SSL_write(c->ssl,
1240 				(void*)(((uint8_t*)&len)+c->tcp_byte_count),
1241 				(int)(sizeof(uint16_t)-c->tcp_byte_count));
1242 		}
1243 		if(r <= 0) {
1244 			int want = SSL_get_error(c->ssl, r);
1245 			if(want == SSL_ERROR_ZERO_RETURN) {
1246 				return 0; /* closed */
1247 			} else if(want == SSL_ERROR_WANT_READ) {
1248 				c->ssl_shake_state = comm_ssl_shake_read;
1249 				comm_point_listen_for_rw(c, 1, 0);
1250 				return 1; /* wait for read condition */
1251 			} else if(want == SSL_ERROR_WANT_WRITE) {
1252 				ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1253 				return 1; /* write more later */
1254 			} else if(want == SSL_ERROR_SYSCALL) {
1255 				if(errno != 0)
1256 					log_err("SSL_write syscall: %s",
1257 						strerror(errno));
1258 				return 0;
1259 			}
1260 			log_crypto_err("could not SSL_write");
1261 			return 0;
1262 		}
1263 		c->tcp_byte_count += r;
1264 		if(c->tcp_byte_count < sizeof(uint16_t))
1265 			return 1;
1266 		sldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1267 			sizeof(uint16_t));
1268 		if(sldns_buffer_remaining(c->buffer) == 0) {
1269 			tcp_callback_writer(c);
1270 			return 1;
1271 		}
1272 	}
1273 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1274 	ERR_clear_error();
1275 	r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
1276 		(int)sldns_buffer_remaining(c->buffer));
1277 	if(r <= 0) {
1278 		int want = SSL_get_error(c->ssl, r);
1279 		if(want == SSL_ERROR_ZERO_RETURN) {
1280 			return 0; /* closed */
1281 		} else if(want == SSL_ERROR_WANT_READ) {
1282 			c->ssl_shake_state = comm_ssl_shake_read;
1283 			comm_point_listen_for_rw(c, 1, 0);
1284 			return 1; /* wait for read condition */
1285 		} else if(want == SSL_ERROR_WANT_WRITE) {
1286 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1287 			return 1; /* write more later */
1288 		} else if(want == SSL_ERROR_SYSCALL) {
1289 			if(errno != 0)
1290 				log_err("SSL_write syscall: %s",
1291 					strerror(errno));
1292 			return 0;
1293 		}
1294 		log_crypto_err("could not SSL_write");
1295 		return 0;
1296 	}
1297 	sldns_buffer_skip(c->buffer, (ssize_t)r);
1298 
1299 	if(sldns_buffer_remaining(c->buffer) == 0) {
1300 		tcp_callback_writer(c);
1301 	}
1302 	return 1;
1303 #else
1304 	(void)c;
1305 	return 0;
1306 #endif /* HAVE_SSL */
1307 }
1308 
1309 /** handle ssl tcp connection with dns contents */
1310 static int
1311 ssl_handle_it(struct comm_point* c)
1312 {
1313 	if(c->tcp_is_reading)
1314 		return ssl_handle_read(c);
1315 	return ssl_handle_write(c);
1316 }
1317 
1318 /** Handle tcp reading callback.
1319  * @param fd: file descriptor of socket.
1320  * @param c: comm point to read from into buffer.
1321  * @param short_ok: if true, very short packets are OK (for comm_local).
1322  * @return: 0 on error
1323  */
1324 static int
1325 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
1326 {
1327 	ssize_t r;
1328 	log_assert(c->type == comm_tcp || c->type == comm_local);
1329 	if(c->ssl)
1330 		return ssl_handle_it(c);
1331 	if(!c->tcp_is_reading)
1332 		return 0;
1333 
1334 	log_assert(fd != -1);
1335 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1336 		/* read length bytes */
1337 		r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count),
1338 			sizeof(uint16_t)-c->tcp_byte_count, 0);
1339 		if(r == 0)
1340 			return 0;
1341 		else if(r == -1) {
1342 #ifndef USE_WINSOCK
1343 			if(errno == EINTR || errno == EAGAIN)
1344 				return 1;
1345 #ifdef ECONNRESET
1346 			if(errno == ECONNRESET && verbosity < 2)
1347 				return 0; /* silence reset by peer */
1348 #endif
1349 			log_err_addr("read (in tcp s)", strerror(errno),
1350 				&c->repinfo.addr, c->repinfo.addrlen);
1351 #else /* USE_WINSOCK */
1352 			if(WSAGetLastError() == WSAECONNRESET)
1353 				return 0;
1354 			if(WSAGetLastError() == WSAEINPROGRESS)
1355 				return 1;
1356 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
1357 				ub_winsock_tcp_wouldblock(c->ev->ev,
1358 					UB_EV_READ);
1359 				return 1;
1360 			}
1361 			log_err_addr("read (in tcp s)",
1362 				wsa_strerror(WSAGetLastError()),
1363 				&c->repinfo.addr, c->repinfo.addrlen);
1364 #endif
1365 			return 0;
1366 		}
1367 		c->tcp_byte_count += r;
1368 		if(c->tcp_byte_count != sizeof(uint16_t))
1369 			return 1;
1370 		if(sldns_buffer_read_u16_at(c->buffer, 0) >
1371 			sldns_buffer_capacity(c->buffer)) {
1372 			verbose(VERB_QUERY, "tcp: dropped larger than buffer");
1373 			return 0;
1374 		}
1375 		sldns_buffer_set_limit(c->buffer,
1376 			sldns_buffer_read_u16_at(c->buffer, 0));
1377 		if(!short_ok &&
1378 			sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1379 			verbose(VERB_QUERY, "tcp: dropped bogus too short.");
1380 			return 0;
1381 		}
1382 		verbose(VERB_ALGO, "Reading tcp query of length %d",
1383 			(int)sldns_buffer_limit(c->buffer));
1384 	}
1385 
1386 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1387 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1388 		sldns_buffer_remaining(c->buffer), 0);
1389 	if(r == 0) {
1390 		return 0;
1391 	} else if(r == -1) {
1392 #ifndef USE_WINSOCK
1393 		if(errno == EINTR || errno == EAGAIN)
1394 			return 1;
1395 		log_err_addr("read (in tcp r)", strerror(errno),
1396 			&c->repinfo.addr, c->repinfo.addrlen);
1397 #else /* USE_WINSOCK */
1398 		if(WSAGetLastError() == WSAECONNRESET)
1399 			return 0;
1400 		if(WSAGetLastError() == WSAEINPROGRESS)
1401 			return 1;
1402 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1403 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1404 			return 1;
1405 		}
1406 		log_err_addr("read (in tcp r)",
1407 			wsa_strerror(WSAGetLastError()),
1408 			&c->repinfo.addr, c->repinfo.addrlen);
1409 #endif
1410 		return 0;
1411 	}
1412 	sldns_buffer_skip(c->buffer, r);
1413 	if(sldns_buffer_remaining(c->buffer) <= 0) {
1414 		tcp_callback_reader(c);
1415 	}
1416 	return 1;
1417 }
1418 
1419 /**
1420  * Handle tcp writing callback.
1421  * @param fd: file descriptor of socket.
1422  * @param c: comm point to write buffer out of.
1423  * @return: 0 on error
1424  */
1425 static int
1426 comm_point_tcp_handle_write(int fd, struct comm_point* c)
1427 {
1428 	ssize_t r;
1429 	struct sldns_buffer *buffer;
1430 	log_assert(c->type == comm_tcp);
1431 #ifdef USE_DNSCRYPT
1432 	buffer = c->dnscrypt_buffer;
1433 #else
1434 	buffer = c->buffer;
1435 #endif
1436 	if(c->tcp_is_reading && !c->ssl)
1437 		return 0;
1438 	log_assert(fd != -1);
1439 	if(c->tcp_byte_count == 0 && c->tcp_check_nb_connect) {
1440 		/* check for pending error from nonblocking connect */
1441 		/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
1442 		int error = 0;
1443 		socklen_t len = (socklen_t)sizeof(error);
1444 		if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
1445 			&len) < 0){
1446 #ifndef USE_WINSOCK
1447 			error = errno; /* on solaris errno is error */
1448 #else /* USE_WINSOCK */
1449 			error = WSAGetLastError();
1450 #endif
1451 		}
1452 #ifndef USE_WINSOCK
1453 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1454 		if(error == EINPROGRESS || error == EWOULDBLOCK)
1455 			return 1; /* try again later */
1456 		else
1457 #endif
1458 		if(error != 0 && verbosity < 2)
1459 			return 0; /* silence lots of chatter in the logs */
1460                 else if(error != 0) {
1461 			log_err_addr("tcp connect", strerror(error),
1462 				&c->repinfo.addr, c->repinfo.addrlen);
1463 #else /* USE_WINSOCK */
1464 		/* examine error */
1465 		if(error == WSAEINPROGRESS)
1466 			return 1;
1467 		else if(error == WSAEWOULDBLOCK) {
1468 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1469 			return 1;
1470 		} else if(error != 0 && verbosity < 2)
1471 			return 0;
1472 		else if(error != 0) {
1473 			log_err_addr("tcp connect", wsa_strerror(error),
1474 				&c->repinfo.addr, c->repinfo.addrlen);
1475 #endif /* USE_WINSOCK */
1476 			return 0;
1477 		}
1478 	}
1479 	if(c->ssl)
1480 		return ssl_handle_it(c);
1481 
1482 #ifdef USE_MSG_FASTOPEN
1483 	/* Only try this on first use of a connection that uses tfo,
1484 	   otherwise fall through to normal write */
1485 	/* Also, TFO support on WINDOWS not implemented at the moment */
1486 	if(c->tcp_do_fastopen == 1) {
1487 		/* this form of sendmsg() does both a connect() and send() so need to
1488 		   look for various flavours of error*/
1489 		uint16_t len = htons(sldns_buffer_limit(buffer));
1490 		struct msghdr msg;
1491 		struct iovec iov[2];
1492 		c->tcp_do_fastopen = 0;
1493 		memset(&msg, 0, sizeof(msg));
1494 		iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1495 		iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1496 		iov[1].iov_base = sldns_buffer_begin(buffer);
1497 		iov[1].iov_len = sldns_buffer_limit(buffer);
1498 		log_assert(iov[0].iov_len > 0);
1499 		log_assert(iov[1].iov_len > 0);
1500 		msg.msg_name = &c->repinfo.addr;
1501 		msg.msg_namelen = c->repinfo.addrlen;
1502 		msg.msg_iov = iov;
1503 		msg.msg_iovlen = 2;
1504 		r = sendmsg(fd, &msg, MSG_FASTOPEN);
1505 		if (r == -1) {
1506 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1507 			/* Handshake is underway, maybe because no TFO cookie available.
1508 			   Come back to write the message*/
1509 			if(errno == EINPROGRESS || errno == EWOULDBLOCK)
1510 				return 1;
1511 #endif
1512 			if(errno == EINTR || errno == EAGAIN)
1513 				return 1;
1514 			/* Not handling EISCONN here as shouldn't ever hit that case.*/
1515 			if(errno != EPIPE && errno != 0 && verbosity < 2)
1516 				return 0; /* silence lots of chatter in the logs */
1517 			if(errno != EPIPE && errno != 0) {
1518 				log_err_addr("tcp sendmsg", strerror(errno),
1519 					&c->repinfo.addr, c->repinfo.addrlen);
1520 				return 0;
1521 			}
1522 			/* fallthrough to nonFASTOPEN
1523 			 * (MSG_FASTOPEN on Linux 3 produces EPIPE)
1524 			 * we need to perform connect() */
1525 			if(connect(fd, (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen) == -1) {
1526 #ifdef EINPROGRESS
1527 				if(errno == EINPROGRESS)
1528 					return 1; /* wait until connect done*/
1529 #endif
1530 #ifdef USE_WINSOCK
1531 				if(WSAGetLastError() == WSAEINPROGRESS ||
1532 					WSAGetLastError() == WSAEWOULDBLOCK)
1533 					return 1; /* wait until connect done*/
1534 #endif
1535 				if(tcp_connect_errno_needs_log(
1536 					(struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen)) {
1537 					log_err_addr("outgoing tcp: connect after EPIPE for fastopen",
1538 						strerror(errno), &c->repinfo.addr, c->repinfo.addrlen);
1539 				}
1540 				return 0;
1541 			}
1542 
1543 		} else {
1544 			c->tcp_byte_count += r;
1545 			if(c->tcp_byte_count < sizeof(uint16_t))
1546 				return 1;
1547 			sldns_buffer_set_position(buffer, c->tcp_byte_count -
1548 				sizeof(uint16_t));
1549 			if(sldns_buffer_remaining(buffer) == 0) {
1550 				tcp_callback_writer(c);
1551 				return 1;
1552 			}
1553 		}
1554 	}
1555 #endif /* USE_MSG_FASTOPEN */
1556 
1557 	if(c->tcp_byte_count < sizeof(uint16_t)) {
1558 		uint16_t len = htons(sldns_buffer_limit(buffer));
1559 #ifdef HAVE_WRITEV
1560 		struct iovec iov[2];
1561 		iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1562 		iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1563 		iov[1].iov_base = sldns_buffer_begin(buffer);
1564 		iov[1].iov_len = sldns_buffer_limit(buffer);
1565 		log_assert(iov[0].iov_len > 0);
1566 		log_assert(iov[1].iov_len > 0);
1567 		r = writev(fd, iov, 2);
1568 #else /* HAVE_WRITEV */
1569 		r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1570 			sizeof(uint16_t)-c->tcp_byte_count, 0);
1571 #endif /* HAVE_WRITEV */
1572 		if(r == -1) {
1573 #ifndef USE_WINSOCK
1574 #  ifdef EPIPE
1575                 	if(errno == EPIPE && verbosity < 2)
1576                         	return 0; /* silence 'broken pipe' */
1577   #endif
1578 			if(errno == EINTR || errno == EAGAIN)
1579 				return 1;
1580 #  ifdef HAVE_WRITEV
1581 			log_err_addr("tcp writev", strerror(errno),
1582 				&c->repinfo.addr, c->repinfo.addrlen);
1583 #  else /* HAVE_WRITEV */
1584 			log_err_addr("tcp send s", strerror(errno),
1585 				&c->repinfo.addr, c->repinfo.addrlen);
1586 #  endif /* HAVE_WRITEV */
1587 #else
1588 			if(WSAGetLastError() == WSAENOTCONN)
1589 				return 1;
1590 			if(WSAGetLastError() == WSAEINPROGRESS)
1591 				return 1;
1592 			if(WSAGetLastError() == WSAEWOULDBLOCK) {
1593 				ub_winsock_tcp_wouldblock(c->ev->ev,
1594 					UB_EV_WRITE);
1595 				return 1;
1596 			}
1597 			log_err_addr("tcp send s",
1598 				wsa_strerror(WSAGetLastError()),
1599 				&c->repinfo.addr, c->repinfo.addrlen);
1600 #endif
1601 			return 0;
1602 		}
1603 		c->tcp_byte_count += r;
1604 		if(c->tcp_byte_count < sizeof(uint16_t))
1605 			return 1;
1606 		sldns_buffer_set_position(buffer, c->tcp_byte_count -
1607 			sizeof(uint16_t));
1608 		if(sldns_buffer_remaining(buffer) == 0) {
1609 			tcp_callback_writer(c);
1610 			return 1;
1611 		}
1612 	}
1613 	log_assert(sldns_buffer_remaining(buffer) > 0);
1614 	r = send(fd, (void*)sldns_buffer_current(buffer),
1615 		sldns_buffer_remaining(buffer), 0);
1616 	if(r == -1) {
1617 #ifndef USE_WINSOCK
1618 		if(errno == EINTR || errno == EAGAIN)
1619 			return 1;
1620 		log_err_addr("tcp send r", strerror(errno),
1621 			&c->repinfo.addr, c->repinfo.addrlen);
1622 #else
1623 		if(WSAGetLastError() == WSAEINPROGRESS)
1624 			return 1;
1625 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1626 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1627 			return 1;
1628 		}
1629 		log_err_addr("tcp send r", wsa_strerror(WSAGetLastError()),
1630 			&c->repinfo.addr, c->repinfo.addrlen);
1631 #endif
1632 		return 0;
1633 	}
1634 	sldns_buffer_skip(buffer, r);
1635 
1636 	if(sldns_buffer_remaining(buffer) == 0) {
1637 		tcp_callback_writer(c);
1638 	}
1639 
1640 	return 1;
1641 }
1642 
1643 void
1644 comm_point_tcp_handle_callback(int fd, short event, void* arg)
1645 {
1646 	struct comm_point* c = (struct comm_point*)arg;
1647 	log_assert(c->type == comm_tcp);
1648 	ub_comm_base_now(c->ev->base);
1649 
1650 #ifdef USE_DNSCRYPT
1651 	/* Initialize if this is a dnscrypt socket */
1652 	if(c->tcp_parent) {
1653 		c->dnscrypt = c->tcp_parent->dnscrypt;
1654 	}
1655 	if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) {
1656 		c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer));
1657 		if(!c->dnscrypt_buffer) {
1658 			log_err("Could not allocate dnscrypt buffer");
1659 			reclaim_tcp_handler(c);
1660 			if(!c->tcp_do_close) {
1661 				fptr_ok(fptr_whitelist_comm_point(
1662 					c->callback));
1663 				(void)(*c->callback)(c, c->cb_arg,
1664 					NETEVENT_CLOSED, NULL);
1665 			}
1666 			return;
1667 		}
1668 	}
1669 #endif
1670 
1671 	if(event&UB_EV_READ) {
1672 		if(!comm_point_tcp_handle_read(fd, c, 0)) {
1673 			reclaim_tcp_handler(c);
1674 			if(!c->tcp_do_close) {
1675 				fptr_ok(fptr_whitelist_comm_point(
1676 					c->callback));
1677 				(void)(*c->callback)(c, c->cb_arg,
1678 					NETEVENT_CLOSED, NULL);
1679 			}
1680 		}
1681 		return;
1682 	}
1683 	if(event&UB_EV_WRITE) {
1684 		if(!comm_point_tcp_handle_write(fd, c)) {
1685 			reclaim_tcp_handler(c);
1686 			if(!c->tcp_do_close) {
1687 				fptr_ok(fptr_whitelist_comm_point(
1688 					c->callback));
1689 				(void)(*c->callback)(c, c->cb_arg,
1690 					NETEVENT_CLOSED, NULL);
1691 			}
1692 		}
1693 		return;
1694 	}
1695 	if(event&UB_EV_TIMEOUT) {
1696 		verbose(VERB_QUERY, "tcp took too long, dropped");
1697 		reclaim_tcp_handler(c);
1698 		if(!c->tcp_do_close) {
1699 			fptr_ok(fptr_whitelist_comm_point(c->callback));
1700 			(void)(*c->callback)(c, c->cb_arg,
1701 				NETEVENT_TIMEOUT, NULL);
1702 		}
1703 		return;
1704 	}
1705 	log_err("Ignored event %d for tcphdl.", event);
1706 }
1707 
1708 /** Make http handler free for next assignment */
1709 static void
1710 reclaim_http_handler(struct comm_point* c)
1711 {
1712 	log_assert(c->type == comm_http);
1713 	if(c->ssl) {
1714 #ifdef HAVE_SSL
1715 		SSL_shutdown(c->ssl);
1716 		SSL_free(c->ssl);
1717 		c->ssl = NULL;
1718 #endif
1719 	}
1720 	comm_point_close(c);
1721 	if(c->tcp_parent) {
1722 		c->tcp_parent->cur_tcp_count--;
1723 		c->tcp_free = c->tcp_parent->tcp_free;
1724 		c->tcp_parent->tcp_free = c;
1725 		if(!c->tcp_free) {
1726 			/* re-enable listening on accept socket */
1727 			comm_point_start_listening(c->tcp_parent, -1, -1);
1728 		}
1729 	}
1730 }
1731 
1732 /** read more data for http (with ssl) */
1733 static int
1734 ssl_http_read_more(struct comm_point* c)
1735 {
1736 #ifdef HAVE_SSL
1737 	int r;
1738 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1739 	ERR_clear_error();
1740 	r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1741 		(int)sldns_buffer_remaining(c->buffer));
1742 	if(r <= 0) {
1743 		int want = SSL_get_error(c->ssl, r);
1744 		if(want == SSL_ERROR_ZERO_RETURN) {
1745 			return 0; /* shutdown, closed */
1746 		} else if(want == SSL_ERROR_WANT_READ) {
1747 			return 1; /* read more later */
1748 		} else if(want == SSL_ERROR_WANT_WRITE) {
1749 			c->ssl_shake_state = comm_ssl_shake_hs_write;
1750 			comm_point_listen_for_rw(c, 0, 1);
1751 			return 1;
1752 		} else if(want == SSL_ERROR_SYSCALL) {
1753 			if(errno != 0)
1754 				log_err("SSL_read syscall: %s",
1755 					strerror(errno));
1756 			return 0;
1757 		}
1758 		log_crypto_err("could not SSL_read");
1759 		return 0;
1760 	}
1761 	sldns_buffer_skip(c->buffer, (ssize_t)r);
1762 	return 1;
1763 #else
1764 	(void)c;
1765 	return 0;
1766 #endif /* HAVE_SSL */
1767 }
1768 
1769 /** read more data for http */
1770 static int
1771 http_read_more(int fd, struct comm_point* c)
1772 {
1773 	ssize_t r;
1774 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
1775 	r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1776 		sldns_buffer_remaining(c->buffer), 0);
1777 	if(r == 0) {
1778 		return 0;
1779 	} else if(r == -1) {
1780 #ifndef USE_WINSOCK
1781 		if(errno == EINTR || errno == EAGAIN)
1782 			return 1;
1783 		log_err_addr("read (in http r)", strerror(errno),
1784 			&c->repinfo.addr, c->repinfo.addrlen);
1785 #else /* USE_WINSOCK */
1786 		if(WSAGetLastError() == WSAECONNRESET)
1787 			return 0;
1788 		if(WSAGetLastError() == WSAEINPROGRESS)
1789 			return 1;
1790 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
1791 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1792 			return 1;
1793 		}
1794 		log_err_addr("read (in http r)",
1795 			wsa_strerror(WSAGetLastError()),
1796 			&c->repinfo.addr, c->repinfo.addrlen);
1797 #endif
1798 		return 0;
1799 	}
1800 	sldns_buffer_skip(c->buffer, r);
1801 	return 1;
1802 }
1803 
1804 /** return true if http header has been read (one line complete) */
1805 static int
1806 http_header_done(sldns_buffer* buf)
1807 {
1808 	size_t i;
1809 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
1810 		/* there was a \r before the \n, but we ignore that */
1811 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n')
1812 			return 1;
1813 	}
1814 	return 0;
1815 }
1816 
1817 /** return character string into buffer for header line, moves buffer
1818  * past that line and puts zero terminator into linefeed-newline */
1819 static char*
1820 http_header_line(sldns_buffer* buf)
1821 {
1822 	char* result = (char*)sldns_buffer_current(buf);
1823 	size_t i;
1824 	for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
1825 		/* terminate the string on the \r */
1826 		if((char)sldns_buffer_read_u8_at(buf, i) == '\r')
1827 			sldns_buffer_write_u8_at(buf, i, 0);
1828 		/* terminate on the \n and skip past the it and done */
1829 		if((char)sldns_buffer_read_u8_at(buf, i) == '\n') {
1830 			sldns_buffer_write_u8_at(buf, i, 0);
1831 			sldns_buffer_set_position(buf, i+1);
1832 			return result;
1833 		}
1834 	}
1835 	return NULL;
1836 }
1837 
1838 /** move unread buffer to start and clear rest for putting the rest into it */
1839 static void
1840 http_moveover_buffer(sldns_buffer* buf)
1841 {
1842 	size_t pos = sldns_buffer_position(buf);
1843 	size_t len = sldns_buffer_remaining(buf);
1844 	sldns_buffer_clear(buf);
1845 	memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len);
1846 	sldns_buffer_set_position(buf, len);
1847 }
1848 
1849 /** a http header is complete, process it */
1850 static int
1851 http_process_initial_header(struct comm_point* c)
1852 {
1853 	char* line = http_header_line(c->buffer);
1854 	if(!line) return 1;
1855 	verbose(VERB_ALGO, "http header: %s", line);
1856 	if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) {
1857 		/* check returncode */
1858 		if(line[9] != '2') {
1859 			verbose(VERB_ALGO, "http bad status %s", line+9);
1860 			return 0;
1861 		}
1862 	} else if(strncasecmp(line, "Content-Length: ", 16) == 0) {
1863 		if(!c->http_is_chunked)
1864 			c->tcp_byte_count = (size_t)atoi(line+16);
1865 	} else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) {
1866 		c->tcp_byte_count = 0;
1867 		c->http_is_chunked = 1;
1868 	} else if(line[0] == 0) {
1869 		/* end of initial headers */
1870 		c->http_in_headers = 0;
1871 		if(c->http_is_chunked)
1872 			c->http_in_chunk_headers = 1;
1873 		/* remove header text from front of buffer
1874 		 * the buffer is going to be used to return the data segment
1875 		 * itself and we don't want the header to get returned
1876 		 * prepended with it */
1877 		http_moveover_buffer(c->buffer);
1878 		sldns_buffer_flip(c->buffer);
1879 		return 1;
1880 	}
1881 	/* ignore other headers */
1882 	return 1;
1883 }
1884 
1885 /** a chunk header is complete, process it, return 0=fail, 1=continue next
1886  * header line, 2=done with chunked transfer*/
1887 static int
1888 http_process_chunk_header(struct comm_point* c)
1889 {
1890 	char* line = http_header_line(c->buffer);
1891 	if(!line) return 1;
1892 	if(c->http_in_chunk_headers == 3) {
1893 		verbose(VERB_ALGO, "http chunk trailer: %s", line);
1894 		/* are we done ? */
1895 		if(line[0] == 0 && c->tcp_byte_count == 0) {
1896 			/* callback of http reader when NETEVENT_DONE,
1897 			 * end of data, with no data in buffer */
1898 			sldns_buffer_set_position(c->buffer, 0);
1899 			sldns_buffer_set_limit(c->buffer, 0);
1900 			fptr_ok(fptr_whitelist_comm_point(c->callback));
1901 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
1902 			/* return that we are done */
1903 			return 2;
1904 		}
1905 		if(line[0] == 0) {
1906 			/* continue with header of the next chunk */
1907 			c->http_in_chunk_headers = 1;
1908 			/* remove header text from front of buffer */
1909 			http_moveover_buffer(c->buffer);
1910 			sldns_buffer_flip(c->buffer);
1911 			return 1;
1912 		}
1913 		/* ignore further trail headers */
1914 		return 1;
1915 	}
1916 	verbose(VERB_ALGO, "http chunk header: %s", line);
1917 	if(c->http_in_chunk_headers == 1) {
1918 		/* read chunked start line */
1919 		char* end = NULL;
1920 		c->tcp_byte_count = (size_t)strtol(line, &end, 16);
1921 		if(end == line)
1922 			return 0;
1923 		c->http_in_chunk_headers = 0;
1924 		/* remove header text from front of buffer */
1925 		http_moveover_buffer(c->buffer);
1926 		sldns_buffer_flip(c->buffer);
1927 		if(c->tcp_byte_count == 0) {
1928 			/* done with chunks, process chunk_trailer lines */
1929 			c->http_in_chunk_headers = 3;
1930 		}
1931 		return 1;
1932 	}
1933 	/* ignore other headers */
1934 	return 1;
1935 }
1936 
1937 /** handle nonchunked data segment */
1938 static int
1939 http_nonchunk_segment(struct comm_point* c)
1940 {
1941 	/* c->buffer at position..limit has new data we read in.
1942 	 * the buffer itself is full of nonchunked data.
1943 	 * we are looking to read tcp_byte_count more data
1944 	 * and then the transfer is done. */
1945 	size_t remainbufferlen;
1946 	size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
1947 	if(c->tcp_byte_count <= got_now) {
1948 		/* done, this is the last data fragment */
1949 		c->http_stored = 0;
1950 		sldns_buffer_set_position(c->buffer, 0);
1951 		fptr_ok(fptr_whitelist_comm_point(c->callback));
1952 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
1953 		return 1;
1954 	}
1955 	c->tcp_byte_count -= got_now;
1956 	/* if we have the buffer space,
1957 	 * read more data collected into the buffer */
1958 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
1959 		sldns_buffer_limit(c->buffer);
1960 	if(remainbufferlen >= c->tcp_byte_count ||
1961 		remainbufferlen >= 2048) {
1962 		size_t total = sldns_buffer_limit(c->buffer);
1963 		sldns_buffer_clear(c->buffer);
1964 		sldns_buffer_set_position(c->buffer, total);
1965 		c->http_stored = total;
1966 		/* return and wait to read more */
1967 		return 1;
1968 	}
1969 	/* call callback with this data amount, then
1970 	 * wait for more */
1971 	c->http_stored = 0;
1972 	sldns_buffer_set_position(c->buffer, 0);
1973 	fptr_ok(fptr_whitelist_comm_point(c->callback));
1974 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
1975 	/* c->callback has to buffer_clear(c->buffer). */
1976 	/* return and wait to read more */
1977 	return 1;
1978 }
1979 
1980 /** handle nonchunked data segment, return 0=fail, 1=wait, 2=process more */
1981 static int
1982 http_chunked_segment(struct comm_point* c)
1983 {
1984 	/* the c->buffer has from position..limit new data we read. */
1985 	/* the current chunk has length tcp_byte_count.
1986 	 * once we read that read more chunk headers.
1987 	 */
1988 	size_t remainbufferlen;
1989 	size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
1990 	if(c->tcp_byte_count <= got_now) {
1991 		/* the chunk has completed (with perhaps some extra data
1992 		 * from next chunk header and next chunk) */
1993 		/* save too much info into temp buffer */
1994 		size_t fraglen;
1995 		struct comm_reply repinfo;
1996 		c->http_stored = 0;
1997 		sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count);
1998 		sldns_buffer_clear(c->http_temp);
1999 		sldns_buffer_write(c->http_temp,
2000 			sldns_buffer_current(c->buffer),
2001 			sldns_buffer_remaining(c->buffer));
2002 		sldns_buffer_flip(c->http_temp);
2003 
2004 		/* callback with this fragment */
2005 		fraglen = sldns_buffer_position(c->buffer);
2006 		sldns_buffer_set_position(c->buffer, 0);
2007 		sldns_buffer_set_limit(c->buffer, fraglen);
2008 		repinfo = c->repinfo;
2009 		fptr_ok(fptr_whitelist_comm_point(c->callback));
2010 		(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo);
2011 		/* c->callback has to buffer_clear(). */
2012 
2013 		/* is commpoint deleted? */
2014 		if(!repinfo.c) {
2015 			return 1;
2016 		}
2017 		/* copy waiting info */
2018 		sldns_buffer_clear(c->buffer);
2019 		sldns_buffer_write(c->buffer,
2020 			sldns_buffer_begin(c->http_temp),
2021 			sldns_buffer_remaining(c->http_temp));
2022 		sldns_buffer_flip(c->buffer);
2023 		/* process end of chunk trailer header lines, until
2024 		 * an empty line */
2025 		c->http_in_chunk_headers = 3;
2026 		/* process more data in buffer (if any) */
2027 		return 2;
2028 	}
2029 	c->tcp_byte_count -= got_now;
2030 
2031 	/* if we have the buffer space,
2032 	 * read more data collected into the buffer */
2033 	remainbufferlen = sldns_buffer_capacity(c->buffer) -
2034 		sldns_buffer_limit(c->buffer);
2035 	if(remainbufferlen >= c->tcp_byte_count ||
2036 		remainbufferlen >= 2048) {
2037 		size_t total = sldns_buffer_limit(c->buffer);
2038 		sldns_buffer_clear(c->buffer);
2039 		sldns_buffer_set_position(c->buffer, total);
2040 		c->http_stored = total;
2041 		/* return and wait to read more */
2042 		return 1;
2043 	}
2044 
2045 	/* callback of http reader for a new part of the data */
2046 	c->http_stored = 0;
2047 	sldns_buffer_set_position(c->buffer, 0);
2048 	fptr_ok(fptr_whitelist_comm_point(c->callback));
2049 	(void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2050 	/* c->callback has to buffer_clear(c->buffer). */
2051 	/* return and wait to read more */
2052 	return 1;
2053 }
2054 
2055 /**
2056  * Handle http reading callback.
2057  * @param fd: file descriptor of socket.
2058  * @param c: comm point to read from into buffer.
2059  * @return: 0 on error
2060  */
2061 static int
2062 comm_point_http_handle_read(int fd, struct comm_point* c)
2063 {
2064 	log_assert(c->type == comm_http);
2065 	log_assert(fd != -1);
2066 
2067 	/* if we are in ssl handshake, handle SSL handshake */
2068 #ifdef HAVE_SSL
2069 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2070 		if(!ssl_handshake(c))
2071 			return 0;
2072 		if(c->ssl_shake_state != comm_ssl_shake_none)
2073 			return 1;
2074 	}
2075 #endif /* HAVE_SSL */
2076 
2077 	if(!c->tcp_is_reading)
2078 		return 1;
2079 	/* read more data */
2080 	if(c->ssl) {
2081 		if(!ssl_http_read_more(c))
2082 			return 0;
2083 	} else {
2084 		if(!http_read_more(fd, c))
2085 			return 0;
2086 	}
2087 
2088 	sldns_buffer_flip(c->buffer);
2089 	while(sldns_buffer_remaining(c->buffer) > 0) {
2090 		/* if we are reading headers, read more headers */
2091 		if(c->http_in_headers || c->http_in_chunk_headers) {
2092 			/* if header is done, process the header */
2093 			if(!http_header_done(c->buffer)) {
2094 				/* copy remaining data to front of buffer
2095 				 * and set rest for writing into it */
2096 				http_moveover_buffer(c->buffer);
2097 				/* return and wait to read more */
2098 				return 1;
2099 			}
2100 			if(!c->http_in_chunk_headers) {
2101 				/* process initial headers */
2102 				if(!http_process_initial_header(c))
2103 					return 0;
2104 			} else {
2105 				/* process chunk headers */
2106 				int r = http_process_chunk_header(c);
2107 				if(r == 0) return 0;
2108 				if(r == 2) return 1; /* done */
2109 				/* r == 1, continue */
2110 			}
2111 			/* see if we have more to process */
2112 			continue;
2113 		}
2114 
2115 		if(!c->http_is_chunked) {
2116 			/* if we are reading nonchunks, process that*/
2117 			return http_nonchunk_segment(c);
2118 		} else {
2119 			/* if we are reading chunks, read the chunk */
2120 			int r = http_chunked_segment(c);
2121 			if(r == 0) return 0;
2122 			if(r == 1) return 1;
2123 			continue;
2124 		}
2125 	}
2126 	/* broke out of the loop; could not process header instead need
2127 	 * to read more */
2128 	/* moveover any remaining data and read more data */
2129 	http_moveover_buffer(c->buffer);
2130 	/* return and wait to read more */
2131 	return 1;
2132 }
2133 
2134 /** check pending connect for http */
2135 static int
2136 http_check_connect(int fd, struct comm_point* c)
2137 {
2138 	/* check for pending error from nonblocking connect */
2139 	/* from Stevens, unix network programming, vol1, 3rd ed, p450*/
2140 	int error = 0;
2141 	socklen_t len = (socklen_t)sizeof(error);
2142 	if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
2143 		&len) < 0){
2144 #ifndef USE_WINSOCK
2145 		error = errno; /* on solaris errno is error */
2146 #else /* USE_WINSOCK */
2147 		error = WSAGetLastError();
2148 #endif
2149 	}
2150 #ifndef USE_WINSOCK
2151 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
2152 	if(error == EINPROGRESS || error == EWOULDBLOCK)
2153 		return 1; /* try again later */
2154 	else
2155 #endif
2156 	if(error != 0 && verbosity < 2)
2157 		return 0; /* silence lots of chatter in the logs */
2158 	else if(error != 0) {
2159 		log_err_addr("http connect", strerror(error),
2160 			&c->repinfo.addr, c->repinfo.addrlen);
2161 #else /* USE_WINSOCK */
2162 	/* examine error */
2163 	if(error == WSAEINPROGRESS)
2164 		return 1;
2165 	else if(error == WSAEWOULDBLOCK) {
2166 		ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2167 		return 1;
2168 	} else if(error != 0 && verbosity < 2)
2169 		return 0;
2170 	else if(error != 0) {
2171 		log_err_addr("http connect", wsa_strerror(error),
2172 			&c->repinfo.addr, c->repinfo.addrlen);
2173 #endif /* USE_WINSOCK */
2174 		return 0;
2175 	}
2176 	/* keep on processing this socket */
2177 	return 2;
2178 }
2179 
2180 /** write more data for http (with ssl) */
2181 static int
2182 ssl_http_write_more(struct comm_point* c)
2183 {
2184 #ifdef HAVE_SSL
2185 	int r;
2186 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2187 	ERR_clear_error();
2188 	r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
2189 		(int)sldns_buffer_remaining(c->buffer));
2190 	if(r <= 0) {
2191 		int want = SSL_get_error(c->ssl, r);
2192 		if(want == SSL_ERROR_ZERO_RETURN) {
2193 			return 0; /* closed */
2194 		} else if(want == SSL_ERROR_WANT_READ) {
2195 			c->ssl_shake_state = comm_ssl_shake_read;
2196 			comm_point_listen_for_rw(c, 1, 0);
2197 			return 1; /* wait for read condition */
2198 		} else if(want == SSL_ERROR_WANT_WRITE) {
2199 			return 1; /* write more later */
2200 		} else if(want == SSL_ERROR_SYSCALL) {
2201 			if(errno != 0)
2202 				log_err("SSL_write syscall: %s",
2203 					strerror(errno));
2204 			return 0;
2205 		}
2206 		log_crypto_err("could not SSL_write");
2207 		return 0;
2208 	}
2209 	sldns_buffer_skip(c->buffer, (ssize_t)r);
2210 	return 1;
2211 #else
2212 	(void)c;
2213 	return 0;
2214 #endif /* HAVE_SSL */
2215 }
2216 
2217 /** write more data for http */
2218 static int
2219 http_write_more(int fd, struct comm_point* c)
2220 {
2221 	ssize_t r;
2222 	log_assert(sldns_buffer_remaining(c->buffer) > 0);
2223 	r = send(fd, (void*)sldns_buffer_current(c->buffer),
2224 		sldns_buffer_remaining(c->buffer), 0);
2225 	if(r == -1) {
2226 #ifndef USE_WINSOCK
2227 		if(errno == EINTR || errno == EAGAIN)
2228 			return 1;
2229 		log_err_addr("http send r", strerror(errno),
2230 			&c->repinfo.addr, c->repinfo.addrlen);
2231 #else
2232 		if(WSAGetLastError() == WSAEINPROGRESS)
2233 			return 1;
2234 		if(WSAGetLastError() == WSAEWOULDBLOCK) {
2235 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2236 			return 1;
2237 		}
2238 		log_err_addr("http send r", wsa_strerror(WSAGetLastError()),
2239 			&c->repinfo.addr, c->repinfo.addrlen);
2240 #endif
2241 		return 0;
2242 	}
2243 	sldns_buffer_skip(c->buffer, r);
2244 	return 1;
2245 }
2246 
2247 /**
2248  * Handle http writing callback.
2249  * @param fd: file descriptor of socket.
2250  * @param c: comm point to write buffer out of.
2251  * @return: 0 on error
2252  */
2253 static int
2254 comm_point_http_handle_write(int fd, struct comm_point* c)
2255 {
2256 	log_assert(c->type == comm_http);
2257 	log_assert(fd != -1);
2258 
2259 	/* check pending connect errors, if that fails, we wait for more,
2260 	 * or we can continue to write contents */
2261 	if(c->tcp_check_nb_connect) {
2262 		int r = http_check_connect(fd, c);
2263 		if(r == 0) return 0;
2264 		if(r == 1) return 1;
2265 		c->tcp_check_nb_connect = 0;
2266 	}
2267 	/* if we are in ssl handshake, handle SSL handshake */
2268 #ifdef HAVE_SSL
2269 	if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2270 		if(!ssl_handshake(c))
2271 			return 0;
2272 		if(c->ssl_shake_state != comm_ssl_shake_none)
2273 			return 1;
2274 	}
2275 #endif /* HAVE_SSL */
2276 	if(c->tcp_is_reading)
2277 		return 1;
2278 	/* if we are writing, write more */
2279 	if(c->ssl) {
2280 		if(!ssl_http_write_more(c))
2281 			return 0;
2282 	} else {
2283 		if(!http_write_more(fd, c))
2284 			return 0;
2285 	}
2286 
2287 	/* we write a single buffer contents, that can contain
2288 	 * the http request, and then flip to read the results */
2289 	/* see if write is done */
2290 	if(sldns_buffer_remaining(c->buffer) == 0) {
2291 		sldns_buffer_clear(c->buffer);
2292 		if(c->tcp_do_toggle_rw)
2293 			c->tcp_is_reading = 1;
2294 		c->tcp_byte_count = 0;
2295 		/* switch from listening(write) to listening(read) */
2296 		comm_point_stop_listening(c);
2297 		comm_point_start_listening(c, -1, -1);
2298 	}
2299 	return 1;
2300 }
2301 
2302 void
2303 comm_point_http_handle_callback(int fd, short event, void* arg)
2304 {
2305 	struct comm_point* c = (struct comm_point*)arg;
2306 	log_assert(c->type == comm_http);
2307 	ub_comm_base_now(c->ev->base);
2308 
2309 	if(event&UB_EV_READ) {
2310 		if(!comm_point_http_handle_read(fd, c)) {
2311 			reclaim_http_handler(c);
2312 			if(!c->tcp_do_close) {
2313 				fptr_ok(fptr_whitelist_comm_point(
2314 					c->callback));
2315 				(void)(*c->callback)(c, c->cb_arg,
2316 					NETEVENT_CLOSED, NULL);
2317 			}
2318 		}
2319 		return;
2320 	}
2321 	if(event&UB_EV_WRITE) {
2322 		if(!comm_point_http_handle_write(fd, c)) {
2323 			reclaim_http_handler(c);
2324 			if(!c->tcp_do_close) {
2325 				fptr_ok(fptr_whitelist_comm_point(
2326 					c->callback));
2327 				(void)(*c->callback)(c, c->cb_arg,
2328 					NETEVENT_CLOSED, NULL);
2329 			}
2330 		}
2331 		return;
2332 	}
2333 	if(event&UB_EV_TIMEOUT) {
2334 		verbose(VERB_QUERY, "http took too long, dropped");
2335 		reclaim_http_handler(c);
2336 		if(!c->tcp_do_close) {
2337 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2338 			(void)(*c->callback)(c, c->cb_arg,
2339 				NETEVENT_TIMEOUT, NULL);
2340 		}
2341 		return;
2342 	}
2343 	log_err("Ignored event %d for httphdl.", event);
2344 }
2345 
2346 void comm_point_local_handle_callback(int fd, short event, void* arg)
2347 {
2348 	struct comm_point* c = (struct comm_point*)arg;
2349 	log_assert(c->type == comm_local);
2350 	ub_comm_base_now(c->ev->base);
2351 
2352 	if(event&UB_EV_READ) {
2353 		if(!comm_point_tcp_handle_read(fd, c, 1)) {
2354 			fptr_ok(fptr_whitelist_comm_point(c->callback));
2355 			(void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
2356 				NULL);
2357 		}
2358 		return;
2359 	}
2360 	log_err("Ignored event %d for localhdl.", event);
2361 }
2362 
2363 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
2364 	short event, void* arg)
2365 {
2366 	struct comm_point* c = (struct comm_point*)arg;
2367 	int err = NETEVENT_NOERROR;
2368 	log_assert(c->type == comm_raw);
2369 	ub_comm_base_now(c->ev->base);
2370 
2371 	if(event&UB_EV_TIMEOUT)
2372 		err = NETEVENT_TIMEOUT;
2373 	fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
2374 	(void)(*c->callback)(c, c->cb_arg, err, NULL);
2375 }
2376 
2377 struct comm_point*
2378 comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer,
2379 	comm_point_callback_type* callback, void* callback_arg)
2380 {
2381 	struct comm_point* c = (struct comm_point*)calloc(1,
2382 		sizeof(struct comm_point));
2383 	short evbits;
2384 	if(!c)
2385 		return NULL;
2386 	c->ev = (struct internal_event*)calloc(1,
2387 		sizeof(struct internal_event));
2388 	if(!c->ev) {
2389 		free(c);
2390 		return NULL;
2391 	}
2392 	c->ev->base = base;
2393 	c->fd = fd;
2394 	c->buffer = buffer;
2395 	c->timeout = NULL;
2396 	c->tcp_is_reading = 0;
2397 	c->tcp_byte_count = 0;
2398 	c->tcp_parent = NULL;
2399 	c->max_tcp_count = 0;
2400 	c->cur_tcp_count = 0;
2401 	c->tcp_handlers = NULL;
2402 	c->tcp_free = NULL;
2403 	c->type = comm_udp;
2404 	c->tcp_do_close = 0;
2405 	c->do_not_close = 0;
2406 	c->tcp_do_toggle_rw = 0;
2407 	c->tcp_check_nb_connect = 0;
2408 #ifdef USE_MSG_FASTOPEN
2409 	c->tcp_do_fastopen = 0;
2410 #endif
2411 #ifdef USE_DNSCRYPT
2412 	c->dnscrypt = 0;
2413 	c->dnscrypt_buffer = buffer;
2414 #endif
2415 	c->inuse = 0;
2416 	c->callback = callback;
2417 	c->cb_arg = callback_arg;
2418 	evbits = UB_EV_READ | UB_EV_PERSIST;
2419 	/* ub_event stuff */
2420 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2421 		comm_point_udp_callback, c);
2422 	if(c->ev->ev == NULL) {
2423 		log_err("could not baseset udp event");
2424 		comm_point_delete(c);
2425 		return NULL;
2426 	}
2427 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
2428 		log_err("could not add udp event");
2429 		comm_point_delete(c);
2430 		return NULL;
2431 	}
2432 	return c;
2433 }
2434 
2435 struct comm_point*
2436 comm_point_create_udp_ancil(struct comm_base *base, int fd,
2437 	sldns_buffer* buffer,
2438 	comm_point_callback_type* callback, void* callback_arg)
2439 {
2440 	struct comm_point* c = (struct comm_point*)calloc(1,
2441 		sizeof(struct comm_point));
2442 	short evbits;
2443 	if(!c)
2444 		return NULL;
2445 	c->ev = (struct internal_event*)calloc(1,
2446 		sizeof(struct internal_event));
2447 	if(!c->ev) {
2448 		free(c);
2449 		return NULL;
2450 	}
2451 	c->ev->base = base;
2452 	c->fd = fd;
2453 	c->buffer = buffer;
2454 	c->timeout = NULL;
2455 	c->tcp_is_reading = 0;
2456 	c->tcp_byte_count = 0;
2457 	c->tcp_parent = NULL;
2458 	c->max_tcp_count = 0;
2459 	c->cur_tcp_count = 0;
2460 	c->tcp_handlers = NULL;
2461 	c->tcp_free = NULL;
2462 	c->type = comm_udp;
2463 	c->tcp_do_close = 0;
2464 	c->do_not_close = 0;
2465 #ifdef USE_DNSCRYPT
2466 	c->dnscrypt = 0;
2467 	c->dnscrypt_buffer = buffer;
2468 #endif
2469 	c->inuse = 0;
2470 	c->tcp_do_toggle_rw = 0;
2471 	c->tcp_check_nb_connect = 0;
2472 #ifdef USE_MSG_FASTOPEN
2473 	c->tcp_do_fastopen = 0;
2474 #endif
2475 	c->callback = callback;
2476 	c->cb_arg = callback_arg;
2477 	evbits = UB_EV_READ | UB_EV_PERSIST;
2478 	/* ub_event stuff */
2479 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2480 		comm_point_udp_ancil_callback, c);
2481 	if(c->ev->ev == NULL) {
2482 		log_err("could not baseset udp event");
2483 		comm_point_delete(c);
2484 		return NULL;
2485 	}
2486 	if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
2487 		log_err("could not add udp event");
2488 		comm_point_delete(c);
2489 		return NULL;
2490 	}
2491 	return c;
2492 }
2493 
2494 static struct comm_point*
2495 comm_point_create_tcp_handler(struct comm_base *base,
2496 	struct comm_point* parent, size_t bufsize,
2497         comm_point_callback_type* callback, void* callback_arg)
2498 {
2499 	struct comm_point* c = (struct comm_point*)calloc(1,
2500 		sizeof(struct comm_point));
2501 	short evbits;
2502 	if(!c)
2503 		return NULL;
2504 	c->ev = (struct internal_event*)calloc(1,
2505 		sizeof(struct internal_event));
2506 	if(!c->ev) {
2507 		free(c);
2508 		return NULL;
2509 	}
2510 	c->ev->base = base;
2511 	c->fd = -1;
2512 	c->buffer = sldns_buffer_new(bufsize);
2513 	if(!c->buffer) {
2514 		free(c->ev);
2515 		free(c);
2516 		return NULL;
2517 	}
2518 	c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
2519 	if(!c->timeout) {
2520 		sldns_buffer_free(c->buffer);
2521 		free(c->ev);
2522 		free(c);
2523 		return NULL;
2524 	}
2525 	c->tcp_is_reading = 0;
2526 	c->tcp_byte_count = 0;
2527 	c->tcp_parent = parent;
2528 	c->max_tcp_count = 0;
2529 	c->cur_tcp_count = 0;
2530 	c->tcp_handlers = NULL;
2531 	c->tcp_free = NULL;
2532 	c->type = comm_tcp;
2533 	c->tcp_do_close = 0;
2534 	c->do_not_close = 0;
2535 	c->tcp_do_toggle_rw = 1;
2536 	c->tcp_check_nb_connect = 0;
2537 #ifdef USE_MSG_FASTOPEN
2538 	c->tcp_do_fastopen = 0;
2539 #endif
2540 #ifdef USE_DNSCRYPT
2541 	c->dnscrypt = 0;
2542 	/* We don't know just yet if this is a dnscrypt channel. Allocation
2543 	 * will be done when handling the callback. */
2544 	c->dnscrypt_buffer = c->buffer;
2545 #endif
2546 	c->repinfo.c = c;
2547 	c->callback = callback;
2548 	c->cb_arg = callback_arg;
2549 	/* add to parent free list */
2550 	c->tcp_free = parent->tcp_free;
2551 	parent->tcp_free = c;
2552 	/* ub_event stuff */
2553 	evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
2554 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2555 		comm_point_tcp_handle_callback, c);
2556 	if(c->ev->ev == NULL)
2557 	{
2558 		log_err("could not basetset tcphdl event");
2559 		parent->tcp_free = c->tcp_free;
2560 		free(c->ev);
2561 		free(c);
2562 		return NULL;
2563 	}
2564 	return c;
2565 }
2566 
2567 struct comm_point*
2568 comm_point_create_tcp(struct comm_base *base, int fd, int num, size_t bufsize,
2569         comm_point_callback_type* callback, void* callback_arg)
2570 {
2571 	struct comm_point* c = (struct comm_point*)calloc(1,
2572 		sizeof(struct comm_point));
2573 	short evbits;
2574 	int i;
2575 	/* first allocate the TCP accept listener */
2576 	if(!c)
2577 		return NULL;
2578 	c->ev = (struct internal_event*)calloc(1,
2579 		sizeof(struct internal_event));
2580 	if(!c->ev) {
2581 		free(c);
2582 		return NULL;
2583 	}
2584 	c->ev->base = base;
2585 	c->fd = fd;
2586 	c->buffer = NULL;
2587 	c->timeout = NULL;
2588 	c->tcp_is_reading = 0;
2589 	c->tcp_byte_count = 0;
2590 	c->tcp_parent = NULL;
2591 	c->max_tcp_count = num;
2592 	c->cur_tcp_count = 0;
2593 	c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
2594 		sizeof(struct comm_point*));
2595 	if(!c->tcp_handlers) {
2596 		free(c->ev);
2597 		free(c);
2598 		return NULL;
2599 	}
2600 	c->tcp_free = NULL;
2601 	c->type = comm_tcp_accept;
2602 	c->tcp_do_close = 0;
2603 	c->do_not_close = 0;
2604 	c->tcp_do_toggle_rw = 0;
2605 	c->tcp_check_nb_connect = 0;
2606 #ifdef USE_MSG_FASTOPEN
2607 	c->tcp_do_fastopen = 0;
2608 #endif
2609 #ifdef USE_DNSCRYPT
2610 	c->dnscrypt = 0;
2611 	c->dnscrypt_buffer = NULL;
2612 #endif
2613 	c->callback = NULL;
2614 	c->cb_arg = NULL;
2615 	evbits = UB_EV_READ | UB_EV_PERSIST;
2616 	/* ub_event stuff */
2617 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2618 		comm_point_tcp_accept_callback, c);
2619 	if(c->ev->ev == NULL) {
2620 		log_err("could not baseset tcpacc event");
2621 		comm_point_delete(c);
2622 		return NULL;
2623 	}
2624 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2625 		log_err("could not add tcpacc event");
2626 		comm_point_delete(c);
2627 		return NULL;
2628 	}
2629 	/* now prealloc the tcp handlers */
2630 	for(i=0; i<num; i++) {
2631 		c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
2632 			c, bufsize, callback, callback_arg);
2633 		if(!c->tcp_handlers[i]) {
2634 			comm_point_delete(c);
2635 			return NULL;
2636 		}
2637 	}
2638 
2639 	return c;
2640 }
2641 
2642 struct comm_point*
2643 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
2644         comm_point_callback_type* callback, void* callback_arg)
2645 {
2646 	struct comm_point* c = (struct comm_point*)calloc(1,
2647 		sizeof(struct comm_point));
2648 	short evbits;
2649 	if(!c)
2650 		return NULL;
2651 	c->ev = (struct internal_event*)calloc(1,
2652 		sizeof(struct internal_event));
2653 	if(!c->ev) {
2654 		free(c);
2655 		return NULL;
2656 	}
2657 	c->ev->base = base;
2658 	c->fd = -1;
2659 	c->buffer = sldns_buffer_new(bufsize);
2660 	if(!c->buffer) {
2661 		free(c->ev);
2662 		free(c);
2663 		return NULL;
2664 	}
2665 	c->timeout = NULL;
2666 	c->tcp_is_reading = 0;
2667 	c->tcp_byte_count = 0;
2668 	c->tcp_parent = NULL;
2669 	c->max_tcp_count = 0;
2670 	c->cur_tcp_count = 0;
2671 	c->tcp_handlers = NULL;
2672 	c->tcp_free = NULL;
2673 	c->type = comm_tcp;
2674 	c->tcp_do_close = 0;
2675 	c->do_not_close = 0;
2676 	c->tcp_do_toggle_rw = 1;
2677 	c->tcp_check_nb_connect = 1;
2678 #ifdef USE_MSG_FASTOPEN
2679 	c->tcp_do_fastopen = 1;
2680 #endif
2681 #ifdef USE_DNSCRYPT
2682 	c->dnscrypt = 0;
2683 	c->dnscrypt_buffer = c->buffer;
2684 #endif
2685 	c->repinfo.c = c;
2686 	c->callback = callback;
2687 	c->cb_arg = callback_arg;
2688 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
2689 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2690 		comm_point_tcp_handle_callback, c);
2691 	if(c->ev->ev == NULL)
2692 	{
2693 		log_err("could not baseset tcpout event");
2694 		sldns_buffer_free(c->buffer);
2695 		free(c->ev);
2696 		free(c);
2697 		return NULL;
2698 	}
2699 
2700 	return c;
2701 }
2702 
2703 struct comm_point*
2704 comm_point_create_http_out(struct comm_base *base, size_t bufsize,
2705         comm_point_callback_type* callback, void* callback_arg,
2706 	sldns_buffer* temp)
2707 {
2708 	struct comm_point* c = (struct comm_point*)calloc(1,
2709 		sizeof(struct comm_point));
2710 	short evbits;
2711 	if(!c)
2712 		return NULL;
2713 	c->ev = (struct internal_event*)calloc(1,
2714 		sizeof(struct internal_event));
2715 	if(!c->ev) {
2716 		free(c);
2717 		return NULL;
2718 	}
2719 	c->ev->base = base;
2720 	c->fd = -1;
2721 	c->buffer = sldns_buffer_new(bufsize);
2722 	if(!c->buffer) {
2723 		free(c->ev);
2724 		free(c);
2725 		return NULL;
2726 	}
2727 	c->timeout = NULL;
2728 	c->tcp_is_reading = 0;
2729 	c->tcp_byte_count = 0;
2730 	c->tcp_parent = NULL;
2731 	c->max_tcp_count = 0;
2732 	c->cur_tcp_count = 0;
2733 	c->tcp_handlers = NULL;
2734 	c->tcp_free = NULL;
2735 	c->type = comm_http;
2736 	c->tcp_do_close = 0;
2737 	c->do_not_close = 0;
2738 	c->tcp_do_toggle_rw = 1;
2739 	c->tcp_check_nb_connect = 1;
2740 	c->http_in_headers = 1;
2741 	c->http_in_chunk_headers = 0;
2742 	c->http_is_chunked = 0;
2743 	c->http_temp = temp;
2744 #ifdef USE_MSG_FASTOPEN
2745 	c->tcp_do_fastopen = 1;
2746 #endif
2747 #ifdef USE_DNSCRYPT
2748 	c->dnscrypt = 0;
2749 	c->dnscrypt_buffer = c->buffer;
2750 #endif
2751 	c->repinfo.c = c;
2752 	c->callback = callback;
2753 	c->cb_arg = callback_arg;
2754 	evbits = UB_EV_PERSIST | UB_EV_WRITE;
2755 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2756 		comm_point_http_handle_callback, c);
2757 	if(c->ev->ev == NULL)
2758 	{
2759 		log_err("could not baseset tcpout event");
2760 #ifdef HAVE_SSL
2761 		SSL_free(c->ssl);
2762 #endif
2763 		sldns_buffer_free(c->buffer);
2764 		free(c->ev);
2765 		free(c);
2766 		return NULL;
2767 	}
2768 
2769 	return c;
2770 }
2771 
2772 struct comm_point*
2773 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
2774         comm_point_callback_type* callback, void* callback_arg)
2775 {
2776 	struct comm_point* c = (struct comm_point*)calloc(1,
2777 		sizeof(struct comm_point));
2778 	short evbits;
2779 	if(!c)
2780 		return NULL;
2781 	c->ev = (struct internal_event*)calloc(1,
2782 		sizeof(struct internal_event));
2783 	if(!c->ev) {
2784 		free(c);
2785 		return NULL;
2786 	}
2787 	c->ev->base = base;
2788 	c->fd = fd;
2789 	c->buffer = sldns_buffer_new(bufsize);
2790 	if(!c->buffer) {
2791 		free(c->ev);
2792 		free(c);
2793 		return NULL;
2794 	}
2795 	c->timeout = NULL;
2796 	c->tcp_is_reading = 1;
2797 	c->tcp_byte_count = 0;
2798 	c->tcp_parent = NULL;
2799 	c->max_tcp_count = 0;
2800 	c->cur_tcp_count = 0;
2801 	c->tcp_handlers = NULL;
2802 	c->tcp_free = NULL;
2803 	c->type = comm_local;
2804 	c->tcp_do_close = 0;
2805 	c->do_not_close = 1;
2806 	c->tcp_do_toggle_rw = 0;
2807 	c->tcp_check_nb_connect = 0;
2808 #ifdef USE_MSG_FASTOPEN
2809 	c->tcp_do_fastopen = 0;
2810 #endif
2811 #ifdef USE_DNSCRYPT
2812 	c->dnscrypt = 0;
2813 	c->dnscrypt_buffer = c->buffer;
2814 #endif
2815 	c->callback = callback;
2816 	c->cb_arg = callback_arg;
2817 	/* ub_event stuff */
2818 	evbits = UB_EV_PERSIST | UB_EV_READ;
2819 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2820 		comm_point_local_handle_callback, c);
2821 	if(c->ev->ev == NULL) {
2822 		log_err("could not baseset localhdl event");
2823 		free(c->ev);
2824 		free(c);
2825 		return NULL;
2826 	}
2827 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2828 		log_err("could not add localhdl event");
2829 		ub_event_free(c->ev->ev);
2830 		free(c->ev);
2831 		free(c);
2832 		return NULL;
2833 	}
2834 	return c;
2835 }
2836 
2837 struct comm_point*
2838 comm_point_create_raw(struct comm_base* base, int fd, int writing,
2839 	comm_point_callback_type* callback, void* callback_arg)
2840 {
2841 	struct comm_point* c = (struct comm_point*)calloc(1,
2842 		sizeof(struct comm_point));
2843 	short evbits;
2844 	if(!c)
2845 		return NULL;
2846 	c->ev = (struct internal_event*)calloc(1,
2847 		sizeof(struct internal_event));
2848 	if(!c->ev) {
2849 		free(c);
2850 		return NULL;
2851 	}
2852 	c->ev->base = base;
2853 	c->fd = fd;
2854 	c->buffer = NULL;
2855 	c->timeout = NULL;
2856 	c->tcp_is_reading = 0;
2857 	c->tcp_byte_count = 0;
2858 	c->tcp_parent = NULL;
2859 	c->max_tcp_count = 0;
2860 	c->cur_tcp_count = 0;
2861 	c->tcp_handlers = NULL;
2862 	c->tcp_free = NULL;
2863 	c->type = comm_raw;
2864 	c->tcp_do_close = 0;
2865 	c->do_not_close = 1;
2866 	c->tcp_do_toggle_rw = 0;
2867 	c->tcp_check_nb_connect = 0;
2868 #ifdef USE_MSG_FASTOPEN
2869 	c->tcp_do_fastopen = 0;
2870 #endif
2871 #ifdef USE_DNSCRYPT
2872 	c->dnscrypt = 0;
2873 	c->dnscrypt_buffer = c->buffer;
2874 #endif
2875 	c->callback = callback;
2876 	c->cb_arg = callback_arg;
2877 	/* ub_event stuff */
2878 	if(writing)
2879 		evbits = UB_EV_PERSIST | UB_EV_WRITE;
2880 	else 	evbits = UB_EV_PERSIST | UB_EV_READ;
2881 	c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
2882 		comm_point_raw_handle_callback, c);
2883 	if(c->ev->ev == NULL) {
2884 		log_err("could not baseset rawhdl event");
2885 		free(c->ev);
2886 		free(c);
2887 		return NULL;
2888 	}
2889 	if (ub_event_add(c->ev->ev, c->timeout) != 0) {
2890 		log_err("could not add rawhdl event");
2891 		ub_event_free(c->ev->ev);
2892 		free(c->ev);
2893 		free(c);
2894 		return NULL;
2895 	}
2896 	return c;
2897 }
2898 
2899 void
2900 comm_point_close(struct comm_point* c)
2901 {
2902 	if(!c)
2903 		return;
2904 	if(c->fd != -1) {
2905 		if(ub_event_del(c->ev->ev) != 0) {
2906 			log_err("could not event_del on close");
2907 		}
2908 	}
2909 	/* close fd after removing from event lists, or epoll.. is messed up */
2910 	if(c->fd != -1 && !c->do_not_close) {
2911 		if(c->type == comm_tcp || c->type == comm_http) {
2912 			/* delete sticky events for the fd, it gets closed */
2913 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
2914 			ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2915 		}
2916 		verbose(VERB_ALGO, "close fd %d", c->fd);
2917 #ifndef USE_WINSOCK
2918 		close(c->fd);
2919 #else
2920 		closesocket(c->fd);
2921 #endif
2922 	}
2923 	c->fd = -1;
2924 }
2925 
2926 void
2927 comm_point_delete(struct comm_point* c)
2928 {
2929 	if(!c)
2930 		return;
2931 	if((c->type == comm_tcp || c->type == comm_http) && c->ssl) {
2932 #ifdef HAVE_SSL
2933 		SSL_shutdown(c->ssl);
2934 		SSL_free(c->ssl);
2935 #endif
2936 	}
2937 	comm_point_close(c);
2938 	if(c->tcp_handlers) {
2939 		int i;
2940 		for(i=0; i<c->max_tcp_count; i++)
2941 			comm_point_delete(c->tcp_handlers[i]);
2942 		free(c->tcp_handlers);
2943 	}
2944 	free(c->timeout);
2945 	if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) {
2946 		sldns_buffer_free(c->buffer);
2947 #ifdef USE_DNSCRYPT
2948 		if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) {
2949 			sldns_buffer_free(c->dnscrypt_buffer);
2950 		}
2951 #endif
2952 	}
2953 	ub_event_free(c->ev->ev);
2954 	free(c->ev);
2955 	free(c);
2956 }
2957 
2958 void
2959 comm_point_send_reply(struct comm_reply *repinfo)
2960 {
2961 	struct sldns_buffer* buffer;
2962 	log_assert(repinfo && repinfo->c);
2963 #ifdef USE_DNSCRYPT
2964 	buffer = repinfo->c->dnscrypt_buffer;
2965 	if(!dnsc_handle_uncurved_request(repinfo)) {
2966 		return;
2967 	}
2968 #else
2969 	buffer = repinfo->c->buffer;
2970 #endif
2971 	if(repinfo->c->type == comm_udp) {
2972 		if(repinfo->srctype)
2973 			comm_point_send_udp_msg_if(repinfo->c,
2974 			buffer, (struct sockaddr*)&repinfo->addr,
2975 			repinfo->addrlen, repinfo);
2976 		else
2977 			comm_point_send_udp_msg(repinfo->c, buffer,
2978 			(struct sockaddr*)&repinfo->addr, repinfo->addrlen);
2979 #ifdef USE_DNSTAP
2980 		if(repinfo->c->dtenv != NULL &&
2981 		   repinfo->c->dtenv->log_client_response_messages)
2982 			dt_msg_send_client_response(repinfo->c->dtenv,
2983 			&repinfo->addr, repinfo->c->type, repinfo->c->buffer);
2984 #endif
2985 	} else {
2986 #ifdef USE_DNSTAP
2987 		if(repinfo->c->tcp_parent->dtenv != NULL &&
2988 		   repinfo->c->tcp_parent->dtenv->log_client_response_messages)
2989 			dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv,
2990 			&repinfo->addr, repinfo->c->type, repinfo->c->buffer);
2991 #endif
2992 		comm_point_start_listening(repinfo->c, -1,
2993 			repinfo->c->tcp_timeout_msec);
2994 	}
2995 }
2996 
2997 void
2998 comm_point_drop_reply(struct comm_reply* repinfo)
2999 {
3000 	if(!repinfo)
3001 		return;
3002 	log_assert(repinfo && repinfo->c);
3003 	log_assert(repinfo->c->type != comm_tcp_accept);
3004 	if(repinfo->c->type == comm_udp)
3005 		return;
3006 	reclaim_tcp_handler(repinfo->c);
3007 }
3008 
3009 void
3010 comm_point_stop_listening(struct comm_point* c)
3011 {
3012 	verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
3013 	if(ub_event_del(c->ev->ev) != 0) {
3014 		log_err("event_del error to stoplisten");
3015 	}
3016 }
3017 
3018 void
3019 comm_point_start_listening(struct comm_point* c, int newfd, int msec)
3020 {
3021 	verbose(VERB_ALGO, "comm point start listening %d",
3022 		c->fd==-1?newfd:c->fd);
3023 	if(c->type == comm_tcp_accept && !c->tcp_free) {
3024 		/* no use to start listening no free slots. */
3025 		return;
3026 	}
3027 	if(msec != -1 && msec != 0) {
3028 		if(!c->timeout) {
3029 			c->timeout = (struct timeval*)malloc(sizeof(
3030 				struct timeval));
3031 			if(!c->timeout) {
3032 				log_err("cpsl: malloc failed. No net read.");
3033 				return;
3034 			}
3035 		}
3036 		ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT);
3037 #ifndef S_SPLINT_S /* splint fails on struct timeval. */
3038 		c->timeout->tv_sec = msec/1000;
3039 		c->timeout->tv_usec = (msec%1000)*1000;
3040 #endif /* S_SPLINT_S */
3041 	}
3042 	if(c->type == comm_tcp || c->type == comm_http) {
3043 		ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
3044 		if(c->tcp_is_reading)
3045 			ub_event_add_bits(c->ev->ev, UB_EV_READ);
3046 		else	ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
3047 	}
3048 	if(newfd != -1) {
3049 		if(c->fd != -1) {
3050 #ifndef USE_WINSOCK
3051 			close(c->fd);
3052 #else
3053 			closesocket(c->fd);
3054 #endif
3055 		}
3056 		c->fd = newfd;
3057 		ub_event_set_fd(c->ev->ev, c->fd);
3058 	}
3059 	if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) {
3060 		log_err("event_add failed. in cpsl.");
3061 	}
3062 }
3063 
3064 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
3065 {
3066 	verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
3067 	if(ub_event_del(c->ev->ev) != 0) {
3068 		log_err("event_del error to cplf");
3069 	}
3070 	ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
3071 	if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ);
3072 	if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
3073 	if(ub_event_add(c->ev->ev, c->timeout) != 0) {
3074 		log_err("event_add failed. in cplf.");
3075 	}
3076 }
3077 
3078 size_t comm_point_get_mem(struct comm_point* c)
3079 {
3080 	size_t s;
3081 	if(!c)
3082 		return 0;
3083 	s = sizeof(*c) + sizeof(*c->ev);
3084 	if(c->timeout)
3085 		s += sizeof(*c->timeout);
3086 	if(c->type == comm_tcp || c->type == comm_local) {
3087 		s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer);
3088 #ifdef USE_DNSCRYPT
3089 		s += sizeof(*c->dnscrypt_buffer);
3090 		if(c->buffer != c->dnscrypt_buffer) {
3091 			s += sldns_buffer_capacity(c->dnscrypt_buffer);
3092 		}
3093 #endif
3094 	}
3095 	if(c->type == comm_tcp_accept) {
3096 		int i;
3097 		for(i=0; i<c->max_tcp_count; i++)
3098 			s += comm_point_get_mem(c->tcp_handlers[i]);
3099 	}
3100 	return s;
3101 }
3102 
3103 struct comm_timer*
3104 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
3105 {
3106 	struct internal_timer *tm = (struct internal_timer*)calloc(1,
3107 		sizeof(struct internal_timer));
3108 	if(!tm) {
3109 		log_err("malloc failed");
3110 		return NULL;
3111 	}
3112 	tm->super.ev_timer = tm;
3113 	tm->base = base;
3114 	tm->super.callback = cb;
3115 	tm->super.cb_arg = cb_arg;
3116 	tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT,
3117 		comm_timer_callback, &tm->super);
3118 	if(tm->ev == NULL) {
3119 		log_err("timer_create: event_base_set failed.");
3120 		free(tm);
3121 		return NULL;
3122 	}
3123 	return &tm->super;
3124 }
3125 
3126 void
3127 comm_timer_disable(struct comm_timer* timer)
3128 {
3129 	if(!timer)
3130 		return;
3131 	ub_timer_del(timer->ev_timer->ev);
3132 	timer->ev_timer->enabled = 0;
3133 }
3134 
3135 void
3136 comm_timer_set(struct comm_timer* timer, struct timeval* tv)
3137 {
3138 	log_assert(tv);
3139 	if(timer->ev_timer->enabled)
3140 		comm_timer_disable(timer);
3141 	if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base,
3142 		comm_timer_callback, timer, tv) != 0)
3143 		log_err("comm_timer_set: evtimer_add failed.");
3144 	timer->ev_timer->enabled = 1;
3145 }
3146 
3147 void
3148 comm_timer_delete(struct comm_timer* timer)
3149 {
3150 	if(!timer)
3151 		return;
3152 	comm_timer_disable(timer);
3153 	/* Free the sub struct timer->ev_timer derived from the super struct timer.
3154 	 * i.e. assert(timer == timer->ev_timer)
3155 	 */
3156 	ub_event_free(timer->ev_timer->ev);
3157 	free(timer->ev_timer);
3158 }
3159 
3160 void
3161 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
3162 {
3163 	struct comm_timer* tm = (struct comm_timer*)arg;
3164 	if(!(event&UB_EV_TIMEOUT))
3165 		return;
3166 	ub_comm_base_now(tm->ev_timer->base);
3167 	tm->ev_timer->enabled = 0;
3168 	fptr_ok(fptr_whitelist_comm_timer(tm->callback));
3169 	(*tm->callback)(tm->cb_arg);
3170 }
3171 
3172 int
3173 comm_timer_is_set(struct comm_timer* timer)
3174 {
3175 	return (int)timer->ev_timer->enabled;
3176 }
3177 
3178 size_t
3179 comm_timer_get_mem(struct comm_timer* ATTR_UNUSED(timer))
3180 {
3181 	return sizeof(struct internal_timer);
3182 }
3183 
3184 struct comm_signal*
3185 comm_signal_create(struct comm_base* base,
3186         void (*callback)(int, void*), void* cb_arg)
3187 {
3188 	struct comm_signal* com = (struct comm_signal*)malloc(
3189 		sizeof(struct comm_signal));
3190 	if(!com) {
3191 		log_err("malloc failed");
3192 		return NULL;
3193 	}
3194 	com->base = base;
3195 	com->callback = callback;
3196 	com->cb_arg = cb_arg;
3197 	com->ev_signal = NULL;
3198 	return com;
3199 }
3200 
3201 void
3202 comm_signal_callback(int sig, short event, void* arg)
3203 {
3204 	struct comm_signal* comsig = (struct comm_signal*)arg;
3205 	if(!(event & UB_EV_SIGNAL))
3206 		return;
3207 	ub_comm_base_now(comsig->base);
3208 	fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
3209 	(*comsig->callback)(sig, comsig->cb_arg);
3210 }
3211 
3212 int
3213 comm_signal_bind(struct comm_signal* comsig, int sig)
3214 {
3215 	struct internal_signal* entry = (struct internal_signal*)calloc(1,
3216 		sizeof(struct internal_signal));
3217 	if(!entry) {
3218 		log_err("malloc failed");
3219 		return 0;
3220 	}
3221 	log_assert(comsig);
3222 	/* add signal event */
3223 	entry->ev = ub_signal_new(comsig->base->eb->base, sig,
3224 		comm_signal_callback, comsig);
3225 	if(entry->ev == NULL) {
3226 		log_err("Could not create signal event");
3227 		free(entry);
3228 		return 0;
3229 	}
3230 	if(ub_signal_add(entry->ev, NULL) != 0) {
3231 		log_err("Could not add signal handler");
3232 		ub_event_free(entry->ev);
3233 		free(entry);
3234 		return 0;
3235 	}
3236 	/* link into list */
3237 	entry->next = comsig->ev_signal;
3238 	comsig->ev_signal = entry;
3239 	return 1;
3240 }
3241 
3242 void
3243 comm_signal_delete(struct comm_signal* comsig)
3244 {
3245 	struct internal_signal* p, *np;
3246 	if(!comsig)
3247 		return;
3248 	p=comsig->ev_signal;
3249 	while(p) {
3250 		np = p->next;
3251 		ub_signal_del(p->ev);
3252 		ub_event_free(p->ev);
3253 		free(p);
3254 		p = np;
3255 	}
3256 	free(comsig);
3257 }
3258