1 /*
2 * util/netevent.c - event notification
3 *
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
5 *
6 * This software is open source.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
14 *
15 * Redistributions in binary form must reproduce the above copyright notice,
16 * this list of conditions and the following disclaimer in the documentation
17 * and/or other materials provided with the distribution.
18 *
19 * Neither the name of the NLNET LABS nor the names of its contributors may
20 * be used to endorse or promote products derived from this software without
21 * specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27 * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
29 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 */
35
36 /**
37 * \file
38 *
39 * This file contains event notification functions.
40 */
41 #include "config.h"
42 #include "util/netevent.h"
43 #include "util/ub_event.h"
44 #include "util/log.h"
45 #include "util/net_help.h"
46 #include "util/tcp_conn_limit.h"
47 #include "util/fptr_wlist.h"
48 #include "sldns/pkthdr.h"
49 #include "sldns/sbuffer.h"
50 #include "sldns/str2wire.h"
51 #include "dnstap/dnstap.h"
52 #include "dnscrypt/dnscrypt.h"
53 #include "services/listen_dnsport.h"
54 #ifdef HAVE_SYS_TYPES_H
55 #include <sys/types.h>
56 #endif
57 #ifdef HAVE_SYS_SOCKET_H
58 #include <sys/socket.h>
59 #endif
60 #ifdef HAVE_NETDB_H
61 #include <netdb.h>
62 #endif
63
64 #ifdef HAVE_OPENSSL_SSL_H
65 #include <openssl/ssl.h>
66 #endif
67 #ifdef HAVE_OPENSSL_ERR_H
68 #include <openssl/err.h>
69 #endif
70
71 /* -------- Start of local definitions -------- */
72 /** if CMSG_ALIGN is not defined on this platform, a workaround */
73 #ifndef CMSG_ALIGN
74 # ifdef __CMSG_ALIGN
75 # define CMSG_ALIGN(n) __CMSG_ALIGN(n)
76 # elif defined(CMSG_DATA_ALIGN)
77 # define CMSG_ALIGN _CMSG_DATA_ALIGN
78 # else
79 # define CMSG_ALIGN(len) (((len)+sizeof(long)-1) & ~(sizeof(long)-1))
80 # endif
81 #endif
82
83 /** if CMSG_LEN is not defined on this platform, a workaround */
84 #ifndef CMSG_LEN
85 # define CMSG_LEN(len) (CMSG_ALIGN(sizeof(struct cmsghdr))+(len))
86 #endif
87
88 /** if CMSG_SPACE is not defined on this platform, a workaround */
89 #ifndef CMSG_SPACE
90 # ifdef _CMSG_HDR_ALIGN
91 # define CMSG_SPACE(l) (CMSG_ALIGN(l)+_CMSG_HDR_ALIGN(sizeof(struct cmsghdr)))
92 # else
93 # define CMSG_SPACE(l) (CMSG_ALIGN(l)+CMSG_ALIGN(sizeof(struct cmsghdr)))
94 # endif
95 #endif
96
97 /** The TCP writing query timeout in milliseconds */
98 #define TCP_QUERY_TIMEOUT 120000
99 /** The minimum actual TCP timeout to use, regardless of what we advertise,
100 * in msec */
101 #define TCP_QUERY_TIMEOUT_MINIMUM 200
102
103 #ifndef NONBLOCKING_IS_BROKEN
104 /** number of UDP reads to perform per read indication from select */
105 #define NUM_UDP_PER_SELECT 100
106 #else
107 #define NUM_UDP_PER_SELECT 1
108 #endif
109
110 /**
111 * The internal event structure for keeping ub_event info for the event.
112 * Possibly other structures (list, tree) this is part of.
113 */
114 struct internal_event {
115 /** the comm base */
116 struct comm_base* base;
117 /** ub_event event type */
118 struct ub_event* ev;
119 };
120
121 /**
122 * Internal base structure, so that every thread has its own events.
123 */
124 struct internal_base {
125 /** ub_event event_base type. */
126 struct ub_event_base* base;
127 /** seconds time pointer points here */
128 time_t secs;
129 /** timeval with current time */
130 struct timeval now;
131 /** the event used for slow_accept timeouts */
132 struct ub_event* slow_accept;
133 /** true if slow_accept is enabled */
134 int slow_accept_enabled;
135 };
136
137 /**
138 * Internal timer structure, to store timer event in.
139 */
140 struct internal_timer {
141 /** the super struct from which derived */
142 struct comm_timer super;
143 /** the comm base */
144 struct comm_base* base;
145 /** ub_event event type */
146 struct ub_event* ev;
147 /** is timer enabled */
148 uint8_t enabled;
149 };
150
151 /**
152 * Internal signal structure, to store signal event in.
153 */
154 struct internal_signal {
155 /** ub_event event type */
156 struct ub_event* ev;
157 /** next in signal list */
158 struct internal_signal* next;
159 };
160
161 /** create a tcp handler with a parent */
162 static struct comm_point* comm_point_create_tcp_handler(
163 struct comm_base *base, struct comm_point* parent, size_t bufsize,
164 struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
165 void* callback_arg, struct unbound_socket* socket);
166
167 /* -------- End of local definitions -------- */
168
169 struct comm_base*
comm_base_create(int sigs)170 comm_base_create(int sigs)
171 {
172 struct comm_base* b = (struct comm_base*)calloc(1,
173 sizeof(struct comm_base));
174 const char *evnm="event", *evsys="", *evmethod="";
175
176 if(!b)
177 return NULL;
178 b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
179 if(!b->eb) {
180 free(b);
181 return NULL;
182 }
183 b->eb->base = ub_default_event_base(sigs, &b->eb->secs, &b->eb->now);
184 if(!b->eb->base) {
185 free(b->eb);
186 free(b);
187 return NULL;
188 }
189 ub_comm_base_now(b);
190 ub_get_event_sys(b->eb->base, &evnm, &evsys, &evmethod);
191 verbose(VERB_ALGO, "%s %s uses %s method.", evnm, evsys, evmethod);
192 return b;
193 }
194
195 struct comm_base*
comm_base_create_event(struct ub_event_base * base)196 comm_base_create_event(struct ub_event_base* base)
197 {
198 struct comm_base* b = (struct comm_base*)calloc(1,
199 sizeof(struct comm_base));
200 if(!b)
201 return NULL;
202 b->eb = (struct internal_base*)calloc(1, sizeof(struct internal_base));
203 if(!b->eb) {
204 free(b);
205 return NULL;
206 }
207 b->eb->base = base;
208 ub_comm_base_now(b);
209 return b;
210 }
211
212 void
comm_base_delete(struct comm_base * b)213 comm_base_delete(struct comm_base* b)
214 {
215 if(!b)
216 return;
217 if(b->eb->slow_accept_enabled) {
218 if(ub_event_del(b->eb->slow_accept) != 0) {
219 log_err("could not event_del slow_accept");
220 }
221 ub_event_free(b->eb->slow_accept);
222 }
223 ub_event_base_free(b->eb->base);
224 b->eb->base = NULL;
225 free(b->eb);
226 free(b);
227 }
228
229 void
comm_base_delete_no_base(struct comm_base * b)230 comm_base_delete_no_base(struct comm_base* b)
231 {
232 if(!b)
233 return;
234 if(b->eb->slow_accept_enabled) {
235 if(ub_event_del(b->eb->slow_accept) != 0) {
236 log_err("could not event_del slow_accept");
237 }
238 ub_event_free(b->eb->slow_accept);
239 }
240 b->eb->base = NULL;
241 free(b->eb);
242 free(b);
243 }
244
245 void
comm_base_timept(struct comm_base * b,time_t ** tt,struct timeval ** tv)246 comm_base_timept(struct comm_base* b, time_t** tt, struct timeval** tv)
247 {
248 *tt = &b->eb->secs;
249 *tv = &b->eb->now;
250 }
251
252 void
comm_base_dispatch(struct comm_base * b)253 comm_base_dispatch(struct comm_base* b)
254 {
255 int retval;
256 retval = ub_event_base_dispatch(b->eb->base);
257 if(retval < 0) {
258 fatal_exit("event_dispatch returned error %d, "
259 "errno is %s", retval, strerror(errno));
260 }
261 }
262
comm_base_exit(struct comm_base * b)263 void comm_base_exit(struct comm_base* b)
264 {
265 if(ub_event_base_loopexit(b->eb->base) != 0) {
266 log_err("Could not loopexit");
267 }
268 }
269
comm_base_set_slow_accept_handlers(struct comm_base * b,void (* stop_acc)(void *),void (* start_acc)(void *),void * arg)270 void comm_base_set_slow_accept_handlers(struct comm_base* b,
271 void (*stop_acc)(void*), void (*start_acc)(void*), void* arg)
272 {
273 b->stop_accept = stop_acc;
274 b->start_accept = start_acc;
275 b->cb_arg = arg;
276 }
277
comm_base_internal(struct comm_base * b)278 struct ub_event_base* comm_base_internal(struct comm_base* b)
279 {
280 return b->eb->base;
281 }
282
283 /** see if errno for udp has to be logged or not uses globals */
284 static int
udp_send_errno_needs_log(struct sockaddr * addr,socklen_t addrlen)285 udp_send_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
286 {
287 /* do not log transient errors (unless high verbosity) */
288 #if defined(ENETUNREACH) || defined(EHOSTDOWN) || defined(EHOSTUNREACH) || defined(ENETDOWN)
289 switch(errno) {
290 # ifdef ENETUNREACH
291 case ENETUNREACH:
292 # endif
293 # ifdef EHOSTDOWN
294 case EHOSTDOWN:
295 # endif
296 # ifdef EHOSTUNREACH
297 case EHOSTUNREACH:
298 # endif
299 # ifdef ENETDOWN
300 case ENETDOWN:
301 # endif
302 case EPERM:
303 case EACCES:
304 if(verbosity < VERB_ALGO)
305 return 0;
306 default:
307 break;
308 }
309 #endif
310 /* permission denied is gotten for every send if the
311 * network is disconnected (on some OS), squelch it */
312 if( ((errno == EPERM)
313 # ifdef EADDRNOTAVAIL
314 /* 'Cannot assign requested address' also when disconnected */
315 || (errno == EADDRNOTAVAIL)
316 # endif
317 ) && verbosity < VERB_ALGO)
318 return 0;
319 # ifdef EADDRINUSE
320 /* If SO_REUSEADDR is set, we could try to connect to the same server
321 * from the same source port twice. */
322 if(errno == EADDRINUSE && verbosity < VERB_DETAIL)
323 return 0;
324 # endif
325 /* squelch errors where people deploy AAAA ::ffff:bla for
326 * authority servers, which we try for intranets. */
327 if(errno == EINVAL && addr_is_ip4mapped(
328 (struct sockaddr_storage*)addr, addrlen) &&
329 verbosity < VERB_DETAIL)
330 return 0;
331 /* SO_BROADCAST sockopt can give access to 255.255.255.255,
332 * but a dns cache does not need it. */
333 if(errno == EACCES && addr_is_broadcast(
334 (struct sockaddr_storage*)addr, addrlen) &&
335 verbosity < VERB_DETAIL)
336 return 0;
337 return 1;
338 }
339
tcp_connect_errno_needs_log(struct sockaddr * addr,socklen_t addrlen)340 int tcp_connect_errno_needs_log(struct sockaddr* addr, socklen_t addrlen)
341 {
342 return udp_send_errno_needs_log(addr, addrlen);
343 }
344
345 /* send a UDP reply */
346 int
comm_point_send_udp_msg(struct comm_point * c,sldns_buffer * packet,struct sockaddr * addr,socklen_t addrlen,int is_connected)347 comm_point_send_udp_msg(struct comm_point *c, sldns_buffer* packet,
348 struct sockaddr* addr, socklen_t addrlen, int is_connected)
349 {
350 ssize_t sent;
351 log_assert(c->fd != -1);
352 #ifdef UNBOUND_DEBUG
353 if(sldns_buffer_remaining(packet) == 0)
354 log_err("error: send empty UDP packet");
355 #endif
356 log_assert(addr && addrlen > 0);
357 if(!is_connected) {
358 sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
359 sldns_buffer_remaining(packet), 0,
360 addr, addrlen);
361 } else {
362 sent = send(c->fd, (void*)sldns_buffer_begin(packet),
363 sldns_buffer_remaining(packet), 0);
364 }
365 if(sent == -1) {
366 /* try again and block, waiting for IO to complete,
367 * we want to send the answer, and we will wait for
368 * the ethernet interface buffer to have space. */
369 #ifndef USE_WINSOCK
370 if(errno == EAGAIN ||
371 # ifdef EWOULDBLOCK
372 errno == EWOULDBLOCK ||
373 # endif
374 errno == ENOBUFS) {
375 #else
376 if(WSAGetLastError() == WSAEINPROGRESS ||
377 WSAGetLastError() == WSAENOBUFS ||
378 WSAGetLastError() == WSAEWOULDBLOCK) {
379 #endif
380 int e;
381 fd_set_block(c->fd);
382 if (!is_connected) {
383 sent = sendto(c->fd, (void*)sldns_buffer_begin(packet),
384 sldns_buffer_remaining(packet), 0,
385 addr, addrlen);
386 } else {
387 sent = send(c->fd, (void*)sldns_buffer_begin(packet),
388 sldns_buffer_remaining(packet), 0);
389 }
390 e = errno;
391 fd_set_nonblock(c->fd);
392 errno = e;
393 }
394 }
395 if(sent == -1) {
396 if(!udp_send_errno_needs_log(addr, addrlen))
397 return 0;
398 if (!is_connected) {
399 verbose(VERB_OPS, "sendto failed: %s", sock_strerror(errno));
400 } else {
401 verbose(VERB_OPS, "send failed: %s", sock_strerror(errno));
402 }
403 if(addr)
404 log_addr(VERB_OPS, "remote address is",
405 (struct sockaddr_storage*)addr, addrlen);
406 return 0;
407 } else if((size_t)sent != sldns_buffer_remaining(packet)) {
408 log_err("sent %d in place of %d bytes",
409 (int)sent, (int)sldns_buffer_remaining(packet));
410 return 0;
411 }
412 return 1;
413 }
414
415 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && (defined(HAVE_RECVMSG) || defined(HAVE_SENDMSG))
416 /** print debug ancillary info */
417 static void p_ancil(const char* str, struct comm_reply* r)
418 {
419 if(r->srctype != 4 && r->srctype != 6) {
420 log_info("%s: unknown srctype %d", str, r->srctype);
421 return;
422 }
423
424 if(r->srctype == 6) {
425 #ifdef IPV6_PKTINFO
426 char buf[1024];
427 if(inet_ntop(AF_INET6, &r->pktinfo.v6info.ipi6_addr,
428 buf, (socklen_t)sizeof(buf)) == 0) {
429 (void)strlcpy(buf, "(inet_ntop error)", sizeof(buf));
430 }
431 buf[sizeof(buf)-1]=0;
432 log_info("%s: %s %d", str, buf, r->pktinfo.v6info.ipi6_ifindex);
433 #endif
434 } else if(r->srctype == 4) {
435 #ifdef IP_PKTINFO
436 char buf1[1024], buf2[1024];
437 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_addr,
438 buf1, (socklen_t)sizeof(buf1)) == 0) {
439 (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
440 }
441 buf1[sizeof(buf1)-1]=0;
442 #ifdef HAVE_STRUCT_IN_PKTINFO_IPI_SPEC_DST
443 if(inet_ntop(AF_INET, &r->pktinfo.v4info.ipi_spec_dst,
444 buf2, (socklen_t)sizeof(buf2)) == 0) {
445 (void)strlcpy(buf2, "(inet_ntop error)", sizeof(buf2));
446 }
447 buf2[sizeof(buf2)-1]=0;
448 #else
449 buf2[0]=0;
450 #endif
451 log_info("%s: %d %s %s", str, r->pktinfo.v4info.ipi_ifindex,
452 buf1, buf2);
453 #elif defined(IP_RECVDSTADDR)
454 char buf1[1024];
455 if(inet_ntop(AF_INET, &r->pktinfo.v4addr,
456 buf1, (socklen_t)sizeof(buf1)) == 0) {
457 (void)strlcpy(buf1, "(inet_ntop error)", sizeof(buf1));
458 }
459 buf1[sizeof(buf1)-1]=0;
460 log_info("%s: %s", str, buf1);
461 #endif /* IP_PKTINFO or PI_RECVDSTDADDR */
462 }
463 }
464 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG||HAVE_SENDMSG */
465
466 /** send a UDP reply over specified interface*/
467 static int
468 comm_point_send_udp_msg_if(struct comm_point *c, sldns_buffer* packet,
469 struct sockaddr* addr, socklen_t addrlen, struct comm_reply* r)
470 {
471 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_SENDMSG)
472 ssize_t sent;
473 struct msghdr msg;
474 struct iovec iov[1];
475 union {
476 struct cmsghdr hdr;
477 char buf[256];
478 } control;
479 #ifndef S_SPLINT_S
480 struct cmsghdr *cmsg;
481 #endif /* S_SPLINT_S */
482
483 log_assert(c->fd != -1);
484 #ifdef UNBOUND_DEBUG
485 if(sldns_buffer_remaining(packet) == 0)
486 log_err("error: send empty UDP packet");
487 #endif
488 log_assert(addr && addrlen > 0);
489
490 msg.msg_name = addr;
491 msg.msg_namelen = addrlen;
492 iov[0].iov_base = sldns_buffer_begin(packet);
493 iov[0].iov_len = sldns_buffer_remaining(packet);
494 msg.msg_iov = iov;
495 msg.msg_iovlen = 1;
496 msg.msg_control = control.buf;
497 #ifndef S_SPLINT_S
498 msg.msg_controllen = sizeof(control.buf);
499 #endif /* S_SPLINT_S */
500 msg.msg_flags = 0;
501
502 #ifndef S_SPLINT_S
503 cmsg = CMSG_FIRSTHDR(&msg);
504 if(r->srctype == 4) {
505 #ifdef IP_PKTINFO
506 void* cmsg_data;
507 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_pktinfo));
508 log_assert(msg.msg_controllen <= sizeof(control.buf));
509 cmsg->cmsg_level = IPPROTO_IP;
510 cmsg->cmsg_type = IP_PKTINFO;
511 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4info,
512 sizeof(struct in_pktinfo));
513 /* unset the ifindex to not bypass the routing tables */
514 cmsg_data = CMSG_DATA(cmsg);
515 ((struct in_pktinfo *) cmsg_data)->ipi_ifindex = 0;
516 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_pktinfo));
517 #elif defined(IP_SENDSRCADDR)
518 msg.msg_controllen = CMSG_SPACE(sizeof(struct in_addr));
519 log_assert(msg.msg_controllen <= sizeof(control.buf));
520 cmsg->cmsg_level = IPPROTO_IP;
521 cmsg->cmsg_type = IP_SENDSRCADDR;
522 memmove(CMSG_DATA(cmsg), &r->pktinfo.v4addr,
523 sizeof(struct in_addr));
524 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in_addr));
525 #else
526 verbose(VERB_ALGO, "no IP_PKTINFO or IP_SENDSRCADDR");
527 msg.msg_control = NULL;
528 #endif /* IP_PKTINFO or IP_SENDSRCADDR */
529 } else if(r->srctype == 6) {
530 void* cmsg_data;
531 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
532 log_assert(msg.msg_controllen <= sizeof(control.buf));
533 cmsg->cmsg_level = IPPROTO_IPV6;
534 cmsg->cmsg_type = IPV6_PKTINFO;
535 memmove(CMSG_DATA(cmsg), &r->pktinfo.v6info,
536 sizeof(struct in6_pktinfo));
537 /* unset the ifindex to not bypass the routing tables */
538 cmsg_data = CMSG_DATA(cmsg);
539 ((struct in6_pktinfo *) cmsg_data)->ipi6_ifindex = 0;
540 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
541 } else {
542 /* try to pass all 0 to use default route */
543 msg.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
544 log_assert(msg.msg_controllen <= sizeof(control.buf));
545 cmsg->cmsg_level = IPPROTO_IPV6;
546 cmsg->cmsg_type = IPV6_PKTINFO;
547 memset(CMSG_DATA(cmsg), 0, sizeof(struct in6_pktinfo));
548 cmsg->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
549 }
550 #endif /* S_SPLINT_S */
551 if(verbosity >= VERB_ALGO)
552 p_ancil("send_udp over interface", r);
553 sent = sendmsg(c->fd, &msg, 0);
554 if(sent == -1) {
555 /* try again and block, waiting for IO to complete,
556 * we want to send the answer, and we will wait for
557 * the ethernet interface buffer to have space. */
558 #ifndef USE_WINSOCK
559 if(errno == EAGAIN ||
560 # ifdef EWOULDBLOCK
561 errno == EWOULDBLOCK ||
562 # endif
563 errno == ENOBUFS) {
564 #else
565 if(WSAGetLastError() == WSAEINPROGRESS ||
566 WSAGetLastError() == WSAENOBUFS ||
567 WSAGetLastError() == WSAEWOULDBLOCK) {
568 #endif
569 int e;
570 fd_set_block(c->fd);
571 sent = sendmsg(c->fd, &msg, 0);
572 e = errno;
573 fd_set_nonblock(c->fd);
574 errno = e;
575 }
576 }
577 if(sent == -1) {
578 if(!udp_send_errno_needs_log(addr, addrlen))
579 return 0;
580 verbose(VERB_OPS, "sendmsg failed: %s", strerror(errno));
581 log_addr(VERB_OPS, "remote address is",
582 (struct sockaddr_storage*)addr, addrlen);
583 #ifdef __NetBSD__
584 /* netbsd 7 has IP_PKTINFO for recv but not send */
585 if(errno == EINVAL && r->srctype == 4)
586 log_err("sendmsg: No support for sendmsg(IP_PKTINFO). "
587 "Please disable interface-automatic");
588 #endif
589 return 0;
590 } else if((size_t)sent != sldns_buffer_remaining(packet)) {
591 log_err("sent %d in place of %d bytes",
592 (int)sent, (int)sldns_buffer_remaining(packet));
593 return 0;
594 }
595 return 1;
596 #else
597 (void)c;
598 (void)packet;
599 (void)addr;
600 (void)addrlen;
601 (void)r;
602 log_err("sendmsg: IPV6_PKTINFO not supported");
603 return 0;
604 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_SENDMSG */
605 }
606
607 /** return true is UDP receive error needs to be logged */
608 static int udp_recv_needs_log(int err)
609 {
610 switch(err) {
611 case EACCES: /* some hosts send ICMP 'Permission Denied' */
612 #ifndef USE_WINSOCK
613 case ECONNREFUSED:
614 # ifdef ENETUNREACH
615 case ENETUNREACH:
616 # endif
617 # ifdef EHOSTDOWN
618 case EHOSTDOWN:
619 # endif
620 # ifdef EHOSTUNREACH
621 case EHOSTUNREACH:
622 # endif
623 # ifdef ENETDOWN
624 case ENETDOWN:
625 # endif
626 #else /* USE_WINSOCK */
627 case WSAECONNREFUSED:
628 case WSAENETUNREACH:
629 case WSAEHOSTDOWN:
630 case WSAEHOSTUNREACH:
631 case WSAENETDOWN:
632 #endif
633 if(verbosity >= VERB_ALGO)
634 return 1;
635 return 0;
636 default:
637 break;
638 }
639 return 1;
640 }
641
642 void
643 comm_point_udp_ancil_callback(int fd, short event, void* arg)
644 {
645 #if defined(AF_INET6) && defined(IPV6_PKTINFO) && defined(HAVE_RECVMSG)
646 struct comm_reply rep;
647 struct msghdr msg;
648 struct iovec iov[1];
649 ssize_t rcv;
650 union {
651 struct cmsghdr hdr;
652 char buf[256];
653 } ancil;
654 int i;
655 #ifndef S_SPLINT_S
656 struct cmsghdr* cmsg;
657 #endif /* S_SPLINT_S */
658
659 rep.c = (struct comm_point*)arg;
660 log_assert(rep.c->type == comm_udp);
661
662 if(!(event&UB_EV_READ))
663 return;
664 log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
665 ub_comm_base_now(rep.c->ev->base);
666 for(i=0; i<NUM_UDP_PER_SELECT; i++) {
667 sldns_buffer_clear(rep.c->buffer);
668 rep.addrlen = (socklen_t)sizeof(rep.addr);
669 log_assert(fd != -1);
670 log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
671 msg.msg_name = &rep.addr;
672 msg.msg_namelen = (socklen_t)sizeof(rep.addr);
673 iov[0].iov_base = sldns_buffer_begin(rep.c->buffer);
674 iov[0].iov_len = sldns_buffer_remaining(rep.c->buffer);
675 msg.msg_iov = iov;
676 msg.msg_iovlen = 1;
677 msg.msg_control = ancil.buf;
678 #ifndef S_SPLINT_S
679 msg.msg_controllen = sizeof(ancil.buf);
680 #endif /* S_SPLINT_S */
681 msg.msg_flags = 0;
682 rcv = recvmsg(fd, &msg, 0);
683 if(rcv == -1) {
684 if(errno != EAGAIN && errno != EINTR
685 && udp_recv_needs_log(errno)) {
686 log_err("recvmsg failed: %s", strerror(errno));
687 }
688 return;
689 }
690 rep.addrlen = msg.msg_namelen;
691 sldns_buffer_skip(rep.c->buffer, rcv);
692 sldns_buffer_flip(rep.c->buffer);
693 rep.srctype = 0;
694 #ifndef S_SPLINT_S
695 for(cmsg = CMSG_FIRSTHDR(&msg); cmsg != NULL;
696 cmsg = CMSG_NXTHDR(&msg, cmsg)) {
697 if( cmsg->cmsg_level == IPPROTO_IPV6 &&
698 cmsg->cmsg_type == IPV6_PKTINFO) {
699 rep.srctype = 6;
700 memmove(&rep.pktinfo.v6info, CMSG_DATA(cmsg),
701 sizeof(struct in6_pktinfo));
702 break;
703 #ifdef IP_PKTINFO
704 } else if( cmsg->cmsg_level == IPPROTO_IP &&
705 cmsg->cmsg_type == IP_PKTINFO) {
706 rep.srctype = 4;
707 memmove(&rep.pktinfo.v4info, CMSG_DATA(cmsg),
708 sizeof(struct in_pktinfo));
709 break;
710 #elif defined(IP_RECVDSTADDR)
711 } else if( cmsg->cmsg_level == IPPROTO_IP &&
712 cmsg->cmsg_type == IP_RECVDSTADDR) {
713 rep.srctype = 4;
714 memmove(&rep.pktinfo.v4addr, CMSG_DATA(cmsg),
715 sizeof(struct in_addr));
716 break;
717 #endif /* IP_PKTINFO or IP_RECVDSTADDR */
718 }
719 }
720 if(verbosity >= VERB_ALGO)
721 p_ancil("receive_udp on interface", &rep);
722 #endif /* S_SPLINT_S */
723 fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
724 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
725 /* send back immediate reply */
726 (void)comm_point_send_udp_msg_if(rep.c, rep.c->buffer,
727 (struct sockaddr*)&rep.addr, rep.addrlen, &rep);
728 }
729 if(!rep.c || rep.c->fd == -1) /* commpoint closed */
730 break;
731 }
732 #else
733 (void)fd;
734 (void)event;
735 (void)arg;
736 fatal_exit("recvmsg: No support for IPV6_PKTINFO; IP_PKTINFO or IP_RECVDSTADDR. "
737 "Please disable interface-automatic");
738 #endif /* AF_INET6 && IPV6_PKTINFO && HAVE_RECVMSG */
739 }
740
741 void
742 comm_point_udp_callback(int fd, short event, void* arg)
743 {
744 struct comm_reply rep;
745 ssize_t rcv;
746 int i;
747 struct sldns_buffer *buffer;
748
749 rep.c = (struct comm_point*)arg;
750 log_assert(rep.c->type == comm_udp);
751
752 if(!(event&UB_EV_READ))
753 return;
754 log_assert(rep.c && rep.c->buffer && rep.c->fd == fd);
755 ub_comm_base_now(rep.c->ev->base);
756 for(i=0; i<NUM_UDP_PER_SELECT; i++) {
757 sldns_buffer_clear(rep.c->buffer);
758 rep.addrlen = (socklen_t)sizeof(rep.addr);
759 log_assert(fd != -1);
760 log_assert(sldns_buffer_remaining(rep.c->buffer) > 0);
761 rcv = recvfrom(fd, (void*)sldns_buffer_begin(rep.c->buffer),
762 sldns_buffer_remaining(rep.c->buffer), 0,
763 (struct sockaddr*)&rep.addr, &rep.addrlen);
764 if(rcv == -1) {
765 #ifndef USE_WINSOCK
766 if(errno != EAGAIN && errno != EINTR
767 && udp_recv_needs_log(errno))
768 log_err("recvfrom %d failed: %s",
769 fd, strerror(errno));
770 #else
771 if(WSAGetLastError() != WSAEINPROGRESS &&
772 WSAGetLastError() != WSAECONNRESET &&
773 WSAGetLastError()!= WSAEWOULDBLOCK &&
774 udp_recv_needs_log(WSAGetLastError()))
775 log_err("recvfrom failed: %s",
776 wsa_strerror(WSAGetLastError()));
777 #endif
778 return;
779 }
780 sldns_buffer_skip(rep.c->buffer, rcv);
781 sldns_buffer_flip(rep.c->buffer);
782 rep.srctype = 0;
783 fptr_ok(fptr_whitelist_comm_point(rep.c->callback));
784 if((*rep.c->callback)(rep.c, rep.c->cb_arg, NETEVENT_NOERROR, &rep)) {
785 /* send back immediate reply */
786 #ifdef USE_DNSCRYPT
787 buffer = rep.c->dnscrypt_buffer;
788 #else
789 buffer = rep.c->buffer;
790 #endif
791 (void)comm_point_send_udp_msg(rep.c, buffer,
792 (struct sockaddr*)&rep.addr, rep.addrlen, 0);
793 }
794 if(!rep.c || rep.c->fd != fd) /* commpoint closed to -1 or reused for
795 another UDP port. Note rep.c cannot be reused with TCP fd. */
796 break;
797 }
798 }
799
800 int adjusted_tcp_timeout(struct comm_point* c)
801 {
802 if(c->tcp_timeout_msec < TCP_QUERY_TIMEOUT_MINIMUM)
803 return TCP_QUERY_TIMEOUT_MINIMUM;
804 return c->tcp_timeout_msec;
805 }
806
807 /** Use a new tcp handler for new query fd, set to read query */
808 static void
809 setup_tcp_handler(struct comm_point* c, int fd, int cur, int max)
810 {
811 int handler_usage;
812 log_assert(c->type == comm_tcp || c->type == comm_http);
813 log_assert(c->fd == -1);
814 sldns_buffer_clear(c->buffer);
815 #ifdef USE_DNSCRYPT
816 if (c->dnscrypt)
817 sldns_buffer_clear(c->dnscrypt_buffer);
818 #endif
819 c->tcp_is_reading = 1;
820 c->tcp_byte_count = 0;
821 c->tcp_keepalive = 0;
822 /* if more than half the tcp handlers are in use, use a shorter
823 * timeout for this TCP connection, we need to make space for
824 * other connections to be able to get attention */
825 /* If > 50% TCP handler structures in use, set timeout to 1/100th
826 * configured value.
827 * If > 65%TCP handler structures in use, set to 1/500th configured
828 * value.
829 * If > 80% TCP handler structures in use, set to 0.
830 *
831 * If the timeout to use falls below 200 milliseconds, an actual
832 * timeout of 200ms is used.
833 */
834 handler_usage = (cur * 100) / max;
835 if(handler_usage > 50 && handler_usage <= 65)
836 c->tcp_timeout_msec /= 100;
837 else if (handler_usage > 65 && handler_usage <= 80)
838 c->tcp_timeout_msec /= 500;
839 else if (handler_usage > 80)
840 c->tcp_timeout_msec = 0;
841 comm_point_start_listening(c, fd, adjusted_tcp_timeout(c));
842 }
843
844 void comm_base_handle_slow_accept(int ATTR_UNUSED(fd),
845 short ATTR_UNUSED(event), void* arg)
846 {
847 struct comm_base* b = (struct comm_base*)arg;
848 /* timeout for the slow accept, re-enable accepts again */
849 if(b->start_accept) {
850 verbose(VERB_ALGO, "wait is over, slow accept disabled");
851 fptr_ok(fptr_whitelist_start_accept(b->start_accept));
852 (*b->start_accept)(b->cb_arg);
853 b->eb->slow_accept_enabled = 0;
854 }
855 }
856
857 int comm_point_perform_accept(struct comm_point* c,
858 struct sockaddr_storage* addr, socklen_t* addrlen)
859 {
860 int new_fd;
861 *addrlen = (socklen_t)sizeof(*addr);
862 #ifndef HAVE_ACCEPT4
863 new_fd = accept(c->fd, (struct sockaddr*)addr, addrlen);
864 #else
865 /* SOCK_NONBLOCK saves extra calls to fcntl for the same result */
866 new_fd = accept4(c->fd, (struct sockaddr*)addr, addrlen, SOCK_NONBLOCK);
867 #endif
868 if(new_fd == -1) {
869 #ifndef USE_WINSOCK
870 /* EINTR is signal interrupt. others are closed connection. */
871 if( errno == EINTR || errno == EAGAIN
872 #ifdef EWOULDBLOCK
873 || errno == EWOULDBLOCK
874 #endif
875 #ifdef ECONNABORTED
876 || errno == ECONNABORTED
877 #endif
878 #ifdef EPROTO
879 || errno == EPROTO
880 #endif /* EPROTO */
881 )
882 return -1;
883 #if defined(ENFILE) && defined(EMFILE)
884 if(errno == ENFILE || errno == EMFILE) {
885 /* out of file descriptors, likely outside of our
886 * control. stop accept() calls for some time */
887 if(c->ev->base->stop_accept) {
888 struct comm_base* b = c->ev->base;
889 struct timeval tv;
890 verbose(VERB_ALGO, "out of file descriptors: "
891 "slow accept");
892 b->eb->slow_accept_enabled = 1;
893 fptr_ok(fptr_whitelist_stop_accept(
894 b->stop_accept));
895 (*b->stop_accept)(b->cb_arg);
896 /* set timeout, no mallocs */
897 tv.tv_sec = NETEVENT_SLOW_ACCEPT_TIME/1000;
898 tv.tv_usec = (NETEVENT_SLOW_ACCEPT_TIME%1000)*1000;
899 b->eb->slow_accept = ub_event_new(b->eb->base,
900 -1, UB_EV_TIMEOUT,
901 comm_base_handle_slow_accept, b);
902 if(b->eb->slow_accept == NULL) {
903 /* we do not want to log here, because
904 * that would spam the logfiles.
905 * error: "event_base_set failed." */
906 }
907 else if(ub_event_add(b->eb->slow_accept, &tv)
908 != 0) {
909 /* we do not want to log here,
910 * error: "event_add failed." */
911 }
912 }
913 return -1;
914 }
915 #endif
916 #else /* USE_WINSOCK */
917 if(WSAGetLastError() == WSAEINPROGRESS ||
918 WSAGetLastError() == WSAECONNRESET)
919 return -1;
920 if(WSAGetLastError() == WSAEWOULDBLOCK) {
921 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
922 return -1;
923 }
924 #endif
925 log_err_addr("accept failed", sock_strerror(errno), addr,
926 *addrlen);
927 return -1;
928 }
929 if(c->tcp_conn_limit && c->type == comm_tcp_accept) {
930 c->tcl_addr = tcl_addr_lookup(c->tcp_conn_limit, addr, *addrlen);
931 if(!tcl_new_connection(c->tcl_addr)) {
932 if(verbosity >= 3)
933 log_err_addr("accept rejected",
934 "connection limit exceeded", addr, *addrlen);
935 close(new_fd);
936 return -1;
937 }
938 }
939 #ifndef HAVE_ACCEPT4
940 fd_set_nonblock(new_fd);
941 #endif
942 return new_fd;
943 }
944
945 #ifdef USE_WINSOCK
946 static long win_bio_cb(BIO *b, int oper, const char* ATTR_UNUSED(argp),
947 #ifdef HAVE_BIO_SET_CALLBACK_EX
948 size_t ATTR_UNUSED(len),
949 #endif
950 int ATTR_UNUSED(argi), long argl,
951 #ifndef HAVE_BIO_SET_CALLBACK_EX
952 long retvalue
953 #else
954 int retvalue, size_t* ATTR_UNUSED(processed)
955 #endif
956 )
957 {
958 int wsa_err = WSAGetLastError(); /* store errcode before it is gone */
959 verbose(VERB_ALGO, "bio_cb %d, %s %s %s", oper,
960 (oper&BIO_CB_RETURN)?"return":"before",
961 (oper&BIO_CB_READ)?"read":((oper&BIO_CB_WRITE)?"write":"other"),
962 wsa_err==WSAEWOULDBLOCK?"wsawb":"");
963 /* on windows, check if previous operation caused EWOULDBLOCK */
964 if( (oper == (BIO_CB_READ|BIO_CB_RETURN) && argl == 0) ||
965 (oper == (BIO_CB_GETS|BIO_CB_RETURN) && argl == 0)) {
966 if(wsa_err == WSAEWOULDBLOCK)
967 ub_winsock_tcp_wouldblock((struct ub_event*)
968 BIO_get_callback_arg(b), UB_EV_READ);
969 }
970 if( (oper == (BIO_CB_WRITE|BIO_CB_RETURN) && argl == 0) ||
971 (oper == (BIO_CB_PUTS|BIO_CB_RETURN) && argl == 0)) {
972 if(wsa_err == WSAEWOULDBLOCK)
973 ub_winsock_tcp_wouldblock((struct ub_event*)
974 BIO_get_callback_arg(b), UB_EV_WRITE);
975 }
976 /* return original return value */
977 return retvalue;
978 }
979
980 /** set win bio callbacks for nonblocking operations */
981 void
982 comm_point_tcp_win_bio_cb(struct comm_point* c, void* thessl)
983 {
984 SSL* ssl = (SSL*)thessl;
985 /* set them both just in case, but usually they are the same BIO */
986 #ifdef HAVE_BIO_SET_CALLBACK_EX
987 BIO_set_callback_ex(SSL_get_rbio(ssl), &win_bio_cb);
988 #else
989 BIO_set_callback(SSL_get_rbio(ssl), &win_bio_cb);
990 #endif
991 BIO_set_callback_arg(SSL_get_rbio(ssl), (char*)c->ev->ev);
992 #ifdef HAVE_BIO_SET_CALLBACK_EX
993 BIO_set_callback_ex(SSL_get_wbio(ssl), &win_bio_cb);
994 #else
995 BIO_set_callback(SSL_get_wbio(ssl), &win_bio_cb);
996 #endif
997 BIO_set_callback_arg(SSL_get_wbio(ssl), (char*)c->ev->ev);
998 }
999 #endif
1000
1001 #ifdef HAVE_NGHTTP2
1002 /** Create http2 session server. Per connection, after TCP accepted.*/
1003 static int http2_session_server_create(struct http2_session* h2_session)
1004 {
1005 log_assert(h2_session->callbacks);
1006 h2_session->is_drop = 0;
1007 if(nghttp2_session_server_new(&h2_session->session,
1008 h2_session->callbacks,
1009 h2_session) == NGHTTP2_ERR_NOMEM) {
1010 log_err("failed to create nghttp2 session server");
1011 return 0;
1012 }
1013
1014 return 1;
1015 }
1016
1017 /** Submit http2 setting to session. Once per session. */
1018 static int http2_submit_settings(struct http2_session* h2_session)
1019 {
1020 int ret;
1021 nghttp2_settings_entry settings[1] = {
1022 {NGHTTP2_SETTINGS_MAX_CONCURRENT_STREAMS,
1023 h2_session->c->http2_max_streams}};
1024
1025 ret = nghttp2_submit_settings(h2_session->session, NGHTTP2_FLAG_NONE,
1026 settings, 1);
1027 if(ret) {
1028 verbose(VERB_QUERY, "http2: submit_settings failed, "
1029 "error: %s", nghttp2_strerror(ret));
1030 return 0;
1031 }
1032 return 1;
1033 }
1034 #endif /* HAVE_NGHTTP2 */
1035
1036
1037 void
1038 comm_point_tcp_accept_callback(int fd, short event, void* arg)
1039 {
1040 struct comm_point* c = (struct comm_point*)arg, *c_hdl;
1041 int new_fd;
1042 log_assert(c->type == comm_tcp_accept);
1043 if(!(event & UB_EV_READ)) {
1044 log_info("ignoring tcp accept event %d", (int)event);
1045 return;
1046 }
1047 ub_comm_base_now(c->ev->base);
1048 /* find free tcp handler. */
1049 if(!c->tcp_free) {
1050 log_warn("accepted too many tcp, connections full");
1051 return;
1052 }
1053 /* accept incoming connection. */
1054 c_hdl = c->tcp_free;
1055 /* clear leftover flags from previous use, and then set the
1056 * correct event base for the event structure for libevent */
1057 ub_event_free(c_hdl->ev->ev);
1058 c_hdl->ev->ev = NULL;
1059 if((c_hdl->type == comm_tcp && c_hdl->tcp_req_info) ||
1060 c_hdl->type == comm_local || c_hdl->type == comm_raw)
1061 c_hdl->tcp_do_toggle_rw = 0;
1062 else c_hdl->tcp_do_toggle_rw = 1;
1063
1064 if(c_hdl->type == comm_http) {
1065 #ifdef HAVE_NGHTTP2
1066 if(!c_hdl->h2_session ||
1067 !http2_session_server_create(c_hdl->h2_session)) {
1068 log_warn("failed to create nghttp2");
1069 return;
1070 }
1071 if(!c_hdl->h2_session ||
1072 !http2_submit_settings(c_hdl->h2_session)) {
1073 log_warn("failed to submit http2 settings");
1074 return;
1075 }
1076 if(!c->ssl) {
1077 c_hdl->tcp_do_toggle_rw = 0;
1078 c_hdl->use_h2 = 1;
1079 }
1080 #endif
1081 c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
1082 UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
1083 comm_point_http_handle_callback, c_hdl);
1084 } else {
1085 c_hdl->ev->ev = ub_event_new(c_hdl->ev->base->eb->base, -1,
1086 UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT,
1087 comm_point_tcp_handle_callback, c_hdl);
1088 }
1089 if(!c_hdl->ev->ev) {
1090 log_warn("could not ub_event_new, dropped tcp");
1091 return;
1092 }
1093 log_assert(fd != -1);
1094 (void)fd;
1095 new_fd = comm_point_perform_accept(c, &c_hdl->repinfo.addr,
1096 &c_hdl->repinfo.addrlen);
1097 if(new_fd == -1)
1098 return;
1099 if(c->ssl) {
1100 c_hdl->ssl = incoming_ssl_fd(c->ssl, new_fd);
1101 if(!c_hdl->ssl) {
1102 c_hdl->fd = new_fd;
1103 comm_point_close(c_hdl);
1104 return;
1105 }
1106 c_hdl->ssl_shake_state = comm_ssl_shake_read;
1107 #ifdef USE_WINSOCK
1108 comm_point_tcp_win_bio_cb(c_hdl, c_hdl->ssl);
1109 #endif
1110 }
1111
1112 /* grab the tcp handler buffers */
1113 c->cur_tcp_count++;
1114 c->tcp_free = c_hdl->tcp_free;
1115 c_hdl->tcp_free = NULL;
1116 if(!c->tcp_free) {
1117 /* stop accepting incoming queries for now. */
1118 comm_point_stop_listening(c);
1119 }
1120 setup_tcp_handler(c_hdl, new_fd, c->cur_tcp_count, c->max_tcp_count);
1121 }
1122
1123 /** Make tcp handler free for next assignment */
1124 static void
1125 reclaim_tcp_handler(struct comm_point* c)
1126 {
1127 log_assert(c->type == comm_tcp);
1128 if(c->ssl) {
1129 #ifdef HAVE_SSL
1130 SSL_shutdown(c->ssl);
1131 SSL_free(c->ssl);
1132 c->ssl = NULL;
1133 #endif
1134 }
1135 comm_point_close(c);
1136 if(c->tcp_parent) {
1137 if(c != c->tcp_parent->tcp_free) {
1138 c->tcp_parent->cur_tcp_count--;
1139 c->tcp_free = c->tcp_parent->tcp_free;
1140 c->tcp_parent->tcp_free = c;
1141 }
1142 if(!c->tcp_free) {
1143 /* re-enable listening on accept socket */
1144 comm_point_start_listening(c->tcp_parent, -1, -1);
1145 }
1146 }
1147 c->tcp_more_read_again = NULL;
1148 c->tcp_more_write_again = NULL;
1149 }
1150
1151 /** do the callback when writing is done */
1152 static void
1153 tcp_callback_writer(struct comm_point* c)
1154 {
1155 log_assert(c->type == comm_tcp);
1156 if(!c->tcp_write_and_read) {
1157 sldns_buffer_clear(c->buffer);
1158 c->tcp_byte_count = 0;
1159 }
1160 if(c->tcp_do_toggle_rw)
1161 c->tcp_is_reading = 1;
1162 /* switch from listening(write) to listening(read) */
1163 if(c->tcp_req_info) {
1164 tcp_req_info_handle_writedone(c->tcp_req_info);
1165 } else {
1166 comm_point_stop_listening(c);
1167 if(c->tcp_write_and_read) {
1168 fptr_ok(fptr_whitelist_comm_point(c->callback));
1169 if( (*c->callback)(c, c->cb_arg, NETEVENT_PKT_WRITTEN,
1170 &c->repinfo) ) {
1171 comm_point_start_listening(c, -1,
1172 adjusted_tcp_timeout(c));
1173 }
1174 } else {
1175 comm_point_start_listening(c, -1,
1176 adjusted_tcp_timeout(c));
1177 }
1178 }
1179 }
1180
1181 /** do the callback when reading is done */
1182 static void
1183 tcp_callback_reader(struct comm_point* c)
1184 {
1185 log_assert(c->type == comm_tcp || c->type == comm_local);
1186 sldns_buffer_flip(c->buffer);
1187 if(c->tcp_do_toggle_rw)
1188 c->tcp_is_reading = 0;
1189 c->tcp_byte_count = 0;
1190 if(c->tcp_req_info) {
1191 tcp_req_info_handle_readdone(c->tcp_req_info);
1192 } else {
1193 if(c->type == comm_tcp)
1194 comm_point_stop_listening(c);
1195 fptr_ok(fptr_whitelist_comm_point(c->callback));
1196 if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
1197 comm_point_start_listening(c, -1,
1198 adjusted_tcp_timeout(c));
1199 }
1200 }
1201 }
1202
1203 #ifdef HAVE_SSL
1204 /** true if the ssl handshake error has to be squelched from the logs */
1205 int
1206 squelch_err_ssl_handshake(unsigned long err)
1207 {
1208 if(verbosity >= VERB_QUERY)
1209 return 0; /* only squelch on low verbosity */
1210 /* this is very specific, we could filter on ERR_GET_REASON()
1211 * (the third element in ERR_PACK) */
1212 if(err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_HTTPS_PROXY_REQUEST) ||
1213 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_HTTP_REQUEST) ||
1214 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_GET_RECORD, SSL_R_WRONG_VERSION_NUMBER) ||
1215 err == ERR_PACK(ERR_LIB_SSL, SSL_F_SSL3_READ_BYTES, SSL_R_SSLV3_ALERT_BAD_CERTIFICATE)
1216 #ifdef SSL_F_TLS_POST_PROCESS_CLIENT_HELLO
1217 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_POST_PROCESS_CLIENT_HELLO, SSL_R_NO_SHARED_CIPHER)
1218 #endif
1219 #ifdef SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO
1220 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_UNKNOWN_PROTOCOL)
1221 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_UNSUPPORTED_PROTOCOL)
1222 # ifdef SSL_R_VERSION_TOO_LOW
1223 || err == ERR_PACK(ERR_LIB_SSL, SSL_F_TLS_EARLY_POST_PROCESS_CLIENT_HELLO, SSL_R_VERSION_TOO_LOW)
1224 # endif
1225 #endif
1226 )
1227 return 1;
1228 return 0;
1229 }
1230 #endif /* HAVE_SSL */
1231
1232 /** continue ssl handshake */
1233 #ifdef HAVE_SSL
1234 static int
1235 ssl_handshake(struct comm_point* c)
1236 {
1237 int r;
1238 if(c->ssl_shake_state == comm_ssl_shake_hs_read) {
1239 /* read condition satisfied back to writing */
1240 comm_point_listen_for_rw(c, 0, 1);
1241 c->ssl_shake_state = comm_ssl_shake_none;
1242 return 1;
1243 }
1244 if(c->ssl_shake_state == comm_ssl_shake_hs_write) {
1245 /* write condition satisfied, back to reading */
1246 comm_point_listen_for_rw(c, 1, 0);
1247 c->ssl_shake_state = comm_ssl_shake_none;
1248 return 1;
1249 }
1250
1251 ERR_clear_error();
1252 r = SSL_do_handshake(c->ssl);
1253 if(r != 1) {
1254 int want = SSL_get_error(c->ssl, r);
1255 if(want == SSL_ERROR_WANT_READ) {
1256 if(c->ssl_shake_state == comm_ssl_shake_read)
1257 return 1;
1258 c->ssl_shake_state = comm_ssl_shake_read;
1259 comm_point_listen_for_rw(c, 1, 0);
1260 return 1;
1261 } else if(want == SSL_ERROR_WANT_WRITE) {
1262 if(c->ssl_shake_state == comm_ssl_shake_write)
1263 return 1;
1264 c->ssl_shake_state = comm_ssl_shake_write;
1265 comm_point_listen_for_rw(c, 0, 1);
1266 return 1;
1267 } else if(r == 0) {
1268 return 0; /* closed */
1269 } else if(want == SSL_ERROR_SYSCALL) {
1270 /* SYSCALL and errno==0 means closed uncleanly */
1271 #ifdef EPIPE
1272 if(errno == EPIPE && verbosity < 2)
1273 return 0; /* silence 'broken pipe' */
1274 #endif
1275 #ifdef ECONNRESET
1276 if(errno == ECONNRESET && verbosity < 2)
1277 return 0; /* silence reset by peer */
1278 #endif
1279 if(errno != 0)
1280 log_err("SSL_handshake syscall: %s",
1281 strerror(errno));
1282 return 0;
1283 } else {
1284 unsigned long err = ERR_get_error();
1285 if(!squelch_err_ssl_handshake(err)) {
1286 log_crypto_err_code("ssl handshake failed", err);
1287 log_addr(VERB_OPS, "ssl handshake failed", &c->repinfo.addr,
1288 c->repinfo.addrlen);
1289 }
1290 return 0;
1291 }
1292 }
1293 /* this is where peer verification could take place */
1294 if((SSL_get_verify_mode(c->ssl)&SSL_VERIFY_PEER)) {
1295 /* verification */
1296 if(SSL_get_verify_result(c->ssl) == X509_V_OK) {
1297 #ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
1298 X509* x = SSL_get1_peer_certificate(c->ssl);
1299 #else
1300 X509* x = SSL_get_peer_certificate(c->ssl);
1301 #endif
1302 if(!x) {
1303 log_addr(VERB_ALGO, "SSL connection failed: "
1304 "no certificate",
1305 &c->repinfo.addr, c->repinfo.addrlen);
1306 return 0;
1307 }
1308 log_cert(VERB_ALGO, "peer certificate", x);
1309 #ifdef HAVE_SSL_GET0_PEERNAME
1310 if(SSL_get0_peername(c->ssl)) {
1311 char buf[255];
1312 snprintf(buf, sizeof(buf), "SSL connection "
1313 "to %s authenticated",
1314 SSL_get0_peername(c->ssl));
1315 log_addr(VERB_ALGO, buf, &c->repinfo.addr,
1316 c->repinfo.addrlen);
1317 } else {
1318 #endif
1319 log_addr(VERB_ALGO, "SSL connection "
1320 "authenticated", &c->repinfo.addr,
1321 c->repinfo.addrlen);
1322 #ifdef HAVE_SSL_GET0_PEERNAME
1323 }
1324 #endif
1325 X509_free(x);
1326 } else {
1327 #ifdef HAVE_SSL_GET1_PEER_CERTIFICATE
1328 X509* x = SSL_get1_peer_certificate(c->ssl);
1329 #else
1330 X509* x = SSL_get_peer_certificate(c->ssl);
1331 #endif
1332 if(x) {
1333 log_cert(VERB_ALGO, "peer certificate", x);
1334 X509_free(x);
1335 }
1336 log_addr(VERB_ALGO, "SSL connection failed: "
1337 "failed to authenticate",
1338 &c->repinfo.addr, c->repinfo.addrlen);
1339 return 0;
1340 }
1341 } else {
1342 /* unauthenticated, the verify peer flag was not set
1343 * in c->ssl when the ssl object was created from ssl_ctx */
1344 log_addr(VERB_ALGO, "SSL connection", &c->repinfo.addr,
1345 c->repinfo.addrlen);
1346 }
1347
1348 #ifdef HAVE_SSL_GET0_ALPN_SELECTED
1349 /* check if http2 use is negotiated */
1350 if(c->type == comm_http && c->h2_session) {
1351 const unsigned char *alpn;
1352 unsigned int alpnlen = 0;
1353 SSL_get0_alpn_selected(c->ssl, &alpn, &alpnlen);
1354 if(alpnlen == 2 && memcmp("h2", alpn, 2) == 0) {
1355 /* connection upgraded to HTTP2 */
1356 c->tcp_do_toggle_rw = 0;
1357 c->use_h2 = 1;
1358 }
1359 }
1360 #endif
1361
1362 /* setup listen rw correctly */
1363 if(c->tcp_is_reading) {
1364 if(c->ssl_shake_state != comm_ssl_shake_read)
1365 comm_point_listen_for_rw(c, 1, 0);
1366 } else {
1367 comm_point_listen_for_rw(c, 0, 1);
1368 }
1369 c->ssl_shake_state = comm_ssl_shake_none;
1370 return 1;
1371 }
1372 #endif /* HAVE_SSL */
1373
1374 /** ssl read callback on TCP */
1375 static int
1376 ssl_handle_read(struct comm_point* c)
1377 {
1378 #ifdef HAVE_SSL
1379 int r;
1380 if(c->ssl_shake_state != comm_ssl_shake_none) {
1381 if(!ssl_handshake(c))
1382 return 0;
1383 if(c->ssl_shake_state != comm_ssl_shake_none)
1384 return 1;
1385 }
1386 if(c->tcp_byte_count < sizeof(uint16_t)) {
1387 /* read length bytes */
1388 ERR_clear_error();
1389 if((r=SSL_read(c->ssl, (void*)sldns_buffer_at(c->buffer,
1390 c->tcp_byte_count), (int)(sizeof(uint16_t) -
1391 c->tcp_byte_count))) <= 0) {
1392 int want = SSL_get_error(c->ssl, r);
1393 if(want == SSL_ERROR_ZERO_RETURN) {
1394 if(c->tcp_req_info)
1395 return tcp_req_info_handle_read_close(c->tcp_req_info);
1396 return 0; /* shutdown, closed */
1397 } else if(want == SSL_ERROR_WANT_READ) {
1398 #ifdef USE_WINSOCK
1399 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1400 #endif
1401 return 1; /* read more later */
1402 } else if(want == SSL_ERROR_WANT_WRITE) {
1403 c->ssl_shake_state = comm_ssl_shake_hs_write;
1404 comm_point_listen_for_rw(c, 0, 1);
1405 return 1;
1406 } else if(want == SSL_ERROR_SYSCALL) {
1407 #ifdef ECONNRESET
1408 if(errno == ECONNRESET && verbosity < 2)
1409 return 0; /* silence reset by peer */
1410 #endif
1411 if(errno != 0)
1412 log_err("SSL_read syscall: %s",
1413 strerror(errno));
1414 return 0;
1415 }
1416 log_crypto_err("could not SSL_read");
1417 return 0;
1418 }
1419 c->tcp_byte_count += r;
1420 if(c->tcp_byte_count < sizeof(uint16_t))
1421 return 1;
1422 if(sldns_buffer_read_u16_at(c->buffer, 0) >
1423 sldns_buffer_capacity(c->buffer)) {
1424 verbose(VERB_QUERY, "ssl: dropped larger than buffer");
1425 return 0;
1426 }
1427 sldns_buffer_set_limit(c->buffer,
1428 sldns_buffer_read_u16_at(c->buffer, 0));
1429 if(sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1430 verbose(VERB_QUERY, "ssl: dropped bogus too short.");
1431 return 0;
1432 }
1433 sldns_buffer_skip(c->buffer, (ssize_t)(c->tcp_byte_count-sizeof(uint16_t)));
1434 verbose(VERB_ALGO, "Reading ssl tcp query of length %d",
1435 (int)sldns_buffer_limit(c->buffer));
1436 }
1437 if(sldns_buffer_remaining(c->buffer) > 0) {
1438 ERR_clear_error();
1439 r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
1440 (int)sldns_buffer_remaining(c->buffer));
1441 if(r <= 0) {
1442 int want = SSL_get_error(c->ssl, r);
1443 if(want == SSL_ERROR_ZERO_RETURN) {
1444 if(c->tcp_req_info)
1445 return tcp_req_info_handle_read_close(c->tcp_req_info);
1446 return 0; /* shutdown, closed */
1447 } else if(want == SSL_ERROR_WANT_READ) {
1448 #ifdef USE_WINSOCK
1449 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1450 #endif
1451 return 1; /* read more later */
1452 } else if(want == SSL_ERROR_WANT_WRITE) {
1453 c->ssl_shake_state = comm_ssl_shake_hs_write;
1454 comm_point_listen_for_rw(c, 0, 1);
1455 return 1;
1456 } else if(want == SSL_ERROR_SYSCALL) {
1457 #ifdef ECONNRESET
1458 if(errno == ECONNRESET && verbosity < 2)
1459 return 0; /* silence reset by peer */
1460 #endif
1461 if(errno != 0)
1462 log_err("SSL_read syscall: %s",
1463 strerror(errno));
1464 return 0;
1465 }
1466 log_crypto_err("could not SSL_read");
1467 return 0;
1468 }
1469 sldns_buffer_skip(c->buffer, (ssize_t)r);
1470 }
1471 if(sldns_buffer_remaining(c->buffer) <= 0) {
1472 tcp_callback_reader(c);
1473 }
1474 return 1;
1475 #else
1476 (void)c;
1477 return 0;
1478 #endif /* HAVE_SSL */
1479 }
1480
1481 /** ssl write callback on TCP */
1482 static int
1483 ssl_handle_write(struct comm_point* c)
1484 {
1485 #ifdef HAVE_SSL
1486 int r;
1487 if(c->ssl_shake_state != comm_ssl_shake_none) {
1488 if(!ssl_handshake(c))
1489 return 0;
1490 if(c->ssl_shake_state != comm_ssl_shake_none)
1491 return 1;
1492 }
1493 /* ignore return, if fails we may simply block */
1494 (void)SSL_set_mode(c->ssl, (long)SSL_MODE_ENABLE_PARTIAL_WRITE);
1495 if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
1496 uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(c->buffer));
1497 ERR_clear_error();
1498 if(c->tcp_write_and_read) {
1499 if(c->tcp_write_pkt_len + 2 < LDNS_RR_BUF_SIZE) {
1500 /* combine the tcp length and the query for
1501 * write, this emulates writev */
1502 uint8_t buf[LDNS_RR_BUF_SIZE];
1503 memmove(buf, &len, sizeof(uint16_t));
1504 memmove(buf+sizeof(uint16_t),
1505 c->tcp_write_pkt,
1506 c->tcp_write_pkt_len);
1507 r = SSL_write(c->ssl,
1508 (void*)(buf+c->tcp_write_byte_count),
1509 c->tcp_write_pkt_len + 2 -
1510 c->tcp_write_byte_count);
1511 } else {
1512 r = SSL_write(c->ssl,
1513 (void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
1514 (int)(sizeof(uint16_t)-c->tcp_write_byte_count));
1515 }
1516 } else if(sizeof(uint16_t)+sldns_buffer_remaining(c->buffer) <
1517 LDNS_RR_BUF_SIZE) {
1518 /* combine the tcp length and the query for write,
1519 * this emulates writev */
1520 uint8_t buf[LDNS_RR_BUF_SIZE];
1521 memmove(buf, &len, sizeof(uint16_t));
1522 memmove(buf+sizeof(uint16_t),
1523 sldns_buffer_current(c->buffer),
1524 sldns_buffer_remaining(c->buffer));
1525 r = SSL_write(c->ssl, (void*)(buf+c->tcp_byte_count),
1526 (int)(sizeof(uint16_t)+
1527 sldns_buffer_remaining(c->buffer)
1528 - c->tcp_byte_count));
1529 } else {
1530 r = SSL_write(c->ssl,
1531 (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1532 (int)(sizeof(uint16_t)-c->tcp_byte_count));
1533 }
1534 if(r <= 0) {
1535 int want = SSL_get_error(c->ssl, r);
1536 if(want == SSL_ERROR_ZERO_RETURN) {
1537 return 0; /* closed */
1538 } else if(want == SSL_ERROR_WANT_READ) {
1539 c->ssl_shake_state = comm_ssl_shake_hs_read;
1540 comm_point_listen_for_rw(c, 1, 0);
1541 return 1; /* wait for read condition */
1542 } else if(want == SSL_ERROR_WANT_WRITE) {
1543 #ifdef USE_WINSOCK
1544 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1545 #endif
1546 return 1; /* write more later */
1547 } else if(want == SSL_ERROR_SYSCALL) {
1548 #ifdef EPIPE
1549 if(errno == EPIPE && verbosity < 2)
1550 return 0; /* silence 'broken pipe' */
1551 #endif
1552 if(errno != 0)
1553 log_err("SSL_write syscall: %s",
1554 strerror(errno));
1555 return 0;
1556 }
1557 log_crypto_err("could not SSL_write");
1558 return 0;
1559 }
1560 if(c->tcp_write_and_read) {
1561 c->tcp_write_byte_count += r;
1562 if(c->tcp_write_byte_count < sizeof(uint16_t))
1563 return 1;
1564 } else {
1565 c->tcp_byte_count += r;
1566 if(c->tcp_byte_count < sizeof(uint16_t))
1567 return 1;
1568 sldns_buffer_set_position(c->buffer, c->tcp_byte_count -
1569 sizeof(uint16_t));
1570 }
1571 if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1572 tcp_callback_writer(c);
1573 return 1;
1574 }
1575 }
1576 log_assert(c->tcp_write_and_read || sldns_buffer_remaining(c->buffer) > 0);
1577 log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
1578 ERR_clear_error();
1579 if(c->tcp_write_and_read) {
1580 r = SSL_write(c->ssl, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
1581 (int)(c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count));
1582 } else {
1583 r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
1584 (int)sldns_buffer_remaining(c->buffer));
1585 }
1586 if(r <= 0) {
1587 int want = SSL_get_error(c->ssl, r);
1588 if(want == SSL_ERROR_ZERO_RETURN) {
1589 return 0; /* closed */
1590 } else if(want == SSL_ERROR_WANT_READ) {
1591 c->ssl_shake_state = comm_ssl_shake_hs_read;
1592 comm_point_listen_for_rw(c, 1, 0);
1593 return 1; /* wait for read condition */
1594 } else if(want == SSL_ERROR_WANT_WRITE) {
1595 #ifdef USE_WINSOCK
1596 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1597 #endif
1598 return 1; /* write more later */
1599 } else if(want == SSL_ERROR_SYSCALL) {
1600 #ifdef EPIPE
1601 if(errno == EPIPE && verbosity < 2)
1602 return 0; /* silence 'broken pipe' */
1603 #endif
1604 if(errno != 0)
1605 log_err("SSL_write syscall: %s",
1606 strerror(errno));
1607 return 0;
1608 }
1609 log_crypto_err("could not SSL_write");
1610 return 0;
1611 }
1612 if(c->tcp_write_and_read) {
1613 c->tcp_write_byte_count += r;
1614 } else {
1615 sldns_buffer_skip(c->buffer, (ssize_t)r);
1616 }
1617
1618 if((!c->tcp_write_and_read && sldns_buffer_remaining(c->buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1619 tcp_callback_writer(c);
1620 }
1621 return 1;
1622 #else
1623 (void)c;
1624 return 0;
1625 #endif /* HAVE_SSL */
1626 }
1627
1628 /** handle ssl tcp connection with dns contents */
1629 static int
1630 ssl_handle_it(struct comm_point* c, int is_write)
1631 {
1632 /* handle case where renegotiation wants read during write call
1633 * or write during read calls */
1634 if(is_write && c->ssl_shake_state == comm_ssl_shake_hs_write)
1635 return ssl_handle_read(c);
1636 else if(!is_write && c->ssl_shake_state == comm_ssl_shake_hs_read)
1637 return ssl_handle_write(c);
1638 /* handle read events for read operation and write events for a
1639 * write operation */
1640 else if(!is_write)
1641 return ssl_handle_read(c);
1642 return ssl_handle_write(c);
1643 }
1644
1645 /** Handle tcp reading callback.
1646 * @param fd: file descriptor of socket.
1647 * @param c: comm point to read from into buffer.
1648 * @param short_ok: if true, very short packets are OK (for comm_local).
1649 * @return: 0 on error
1650 */
1651 static int
1652 comm_point_tcp_handle_read(int fd, struct comm_point* c, int short_ok)
1653 {
1654 ssize_t r;
1655 log_assert(c->type == comm_tcp || c->type == comm_local);
1656 if(c->ssl)
1657 return ssl_handle_it(c, 0);
1658 if(!c->tcp_is_reading && !c->tcp_write_and_read)
1659 return 0;
1660
1661 log_assert(fd != -1);
1662 if(c->tcp_byte_count < sizeof(uint16_t)) {
1663 /* read length bytes */
1664 r = recv(fd,(void*)sldns_buffer_at(c->buffer,c->tcp_byte_count),
1665 sizeof(uint16_t)-c->tcp_byte_count, 0);
1666 if(r == 0) {
1667 if(c->tcp_req_info)
1668 return tcp_req_info_handle_read_close(c->tcp_req_info);
1669 return 0;
1670 } else if(r == -1) {
1671 #ifndef USE_WINSOCK
1672 if(errno == EINTR || errno == EAGAIN)
1673 return 1;
1674 #ifdef ECONNRESET
1675 if(errno == ECONNRESET && verbosity < 2)
1676 return 0; /* silence reset by peer */
1677 #endif
1678 #ifdef ECONNREFUSED
1679 if(errno == ECONNREFUSED && verbosity < 2)
1680 return 0; /* silence reset by peer */
1681 #endif
1682 #ifdef ENETUNREACH
1683 if(errno == ENETUNREACH && verbosity < 2)
1684 return 0; /* silence it */
1685 #endif
1686 #ifdef EHOSTDOWN
1687 if(errno == EHOSTDOWN && verbosity < 2)
1688 return 0; /* silence it */
1689 #endif
1690 #ifdef EHOSTUNREACH
1691 if(errno == EHOSTUNREACH && verbosity < 2)
1692 return 0; /* silence it */
1693 #endif
1694 #ifdef ENETDOWN
1695 if(errno == ENETDOWN && verbosity < 2)
1696 return 0; /* silence it */
1697 #endif
1698 #ifdef EACCES
1699 if(errno == EACCES && verbosity < 2)
1700 return 0; /* silence it */
1701 #endif
1702 #ifdef ENOTCONN
1703 if(errno == ENOTCONN) {
1704 log_err_addr("read (in tcp s) failed and this could be because TCP Fast Open is enabled [--disable-tfo-client --disable-tfo-server] but does not work", sock_strerror(errno),
1705 &c->repinfo.addr, c->repinfo.addrlen);
1706 return 0;
1707 }
1708 #endif
1709 #else /* USE_WINSOCK */
1710 if(WSAGetLastError() == WSAECONNREFUSED && verbosity < 2)
1711 return 0;
1712 if(WSAGetLastError() == WSAEHOSTDOWN && verbosity < 2)
1713 return 0;
1714 if(WSAGetLastError() == WSAEHOSTUNREACH && verbosity < 2)
1715 return 0;
1716 if(WSAGetLastError() == WSAENETDOWN && verbosity < 2)
1717 return 0;
1718 if(WSAGetLastError() == WSAENETUNREACH && verbosity < 2)
1719 return 0;
1720 if(WSAGetLastError() == WSAECONNRESET)
1721 return 0;
1722 if(WSAGetLastError() == WSAEINPROGRESS)
1723 return 1;
1724 if(WSAGetLastError() == WSAEWOULDBLOCK) {
1725 ub_winsock_tcp_wouldblock(c->ev->ev,
1726 UB_EV_READ);
1727 return 1;
1728 }
1729 #endif
1730 log_err_addr("read (in tcp s)", sock_strerror(errno),
1731 &c->repinfo.addr, c->repinfo.addrlen);
1732 return 0;
1733 }
1734 c->tcp_byte_count += r;
1735 if(c->tcp_byte_count != sizeof(uint16_t))
1736 return 1;
1737 if(sldns_buffer_read_u16_at(c->buffer, 0) >
1738 sldns_buffer_capacity(c->buffer)) {
1739 verbose(VERB_QUERY, "tcp: dropped larger than buffer");
1740 return 0;
1741 }
1742 sldns_buffer_set_limit(c->buffer,
1743 sldns_buffer_read_u16_at(c->buffer, 0));
1744 if(!short_ok &&
1745 sldns_buffer_limit(c->buffer) < LDNS_HEADER_SIZE) {
1746 verbose(VERB_QUERY, "tcp: dropped bogus too short.");
1747 return 0;
1748 }
1749 verbose(VERB_ALGO, "Reading tcp query of length %d",
1750 (int)sldns_buffer_limit(c->buffer));
1751 }
1752
1753 if(sldns_buffer_remaining(c->buffer) == 0)
1754 log_err("in comm_point_tcp_handle_read buffer_remaining is not > 0 as expected, continuing with (harmless) 0 length recv");
1755 r = recv(fd, (void*)sldns_buffer_current(c->buffer),
1756 sldns_buffer_remaining(c->buffer), 0);
1757 if(r == 0) {
1758 if(c->tcp_req_info)
1759 return tcp_req_info_handle_read_close(c->tcp_req_info);
1760 return 0;
1761 } else if(r == -1) {
1762 #ifndef USE_WINSOCK
1763 if(errno == EINTR || errno == EAGAIN)
1764 return 1;
1765 #else /* USE_WINSOCK */
1766 if(WSAGetLastError() == WSAECONNRESET)
1767 return 0;
1768 if(WSAGetLastError() == WSAEINPROGRESS)
1769 return 1;
1770 if(WSAGetLastError() == WSAEWOULDBLOCK) {
1771 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
1772 return 1;
1773 }
1774 #endif
1775 log_err_addr("read (in tcp r)", sock_strerror(errno),
1776 &c->repinfo.addr, c->repinfo.addrlen);
1777 return 0;
1778 }
1779 sldns_buffer_skip(c->buffer, r);
1780 if(sldns_buffer_remaining(c->buffer) <= 0) {
1781 tcp_callback_reader(c);
1782 }
1783 return 1;
1784 }
1785
1786 /**
1787 * Handle tcp writing callback.
1788 * @param fd: file descriptor of socket.
1789 * @param c: comm point to write buffer out of.
1790 * @return: 0 on error
1791 */
1792 static int
1793 comm_point_tcp_handle_write(int fd, struct comm_point* c)
1794 {
1795 ssize_t r;
1796 struct sldns_buffer *buffer;
1797 log_assert(c->type == comm_tcp);
1798 #ifdef USE_DNSCRYPT
1799 buffer = c->dnscrypt_buffer;
1800 #else
1801 buffer = c->buffer;
1802 #endif
1803 if(c->tcp_is_reading && !c->ssl && !c->tcp_write_and_read)
1804 return 0;
1805 log_assert(fd != -1);
1806 if(((!c->tcp_write_and_read && c->tcp_byte_count == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == 0)) && c->tcp_check_nb_connect) {
1807 /* check for pending error from nonblocking connect */
1808 /* from Stevens, unix network programming, vol1, 3rd ed, p450*/
1809 int error = 0;
1810 socklen_t len = (socklen_t)sizeof(error);
1811 if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
1812 &len) < 0){
1813 #ifndef USE_WINSOCK
1814 error = errno; /* on solaris errno is error */
1815 #else /* USE_WINSOCK */
1816 error = WSAGetLastError();
1817 #endif
1818 }
1819 #ifndef USE_WINSOCK
1820 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1821 if(error == EINPROGRESS || error == EWOULDBLOCK)
1822 return 1; /* try again later */
1823 else
1824 #endif
1825 if(error != 0 && verbosity < 2)
1826 return 0; /* silence lots of chatter in the logs */
1827 else if(error != 0) {
1828 log_err_addr("tcp connect", strerror(error),
1829 &c->repinfo.addr, c->repinfo.addrlen);
1830 #else /* USE_WINSOCK */
1831 /* examine error */
1832 if(error == WSAEINPROGRESS)
1833 return 1;
1834 else if(error == WSAEWOULDBLOCK) {
1835 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
1836 return 1;
1837 } else if(error != 0 && verbosity < 2)
1838 return 0;
1839 else if(error != 0) {
1840 log_err_addr("tcp connect", wsa_strerror(error),
1841 &c->repinfo.addr, c->repinfo.addrlen);
1842 #endif /* USE_WINSOCK */
1843 return 0;
1844 }
1845 }
1846 if(c->ssl)
1847 return ssl_handle_it(c, 1);
1848
1849 #ifdef USE_MSG_FASTOPEN
1850 /* Only try this on first use of a connection that uses tfo,
1851 otherwise fall through to normal write */
1852 /* Also, TFO support on WINDOWS not implemented at the moment */
1853 if(c->tcp_do_fastopen == 1) {
1854 /* this form of sendmsg() does both a connect() and send() so need to
1855 look for various flavours of error*/
1856 uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
1857 struct msghdr msg;
1858 struct iovec iov[2];
1859 c->tcp_do_fastopen = 0;
1860 memset(&msg, 0, sizeof(msg));
1861 if(c->tcp_write_and_read) {
1862 iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
1863 iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
1864 iov[1].iov_base = c->tcp_write_pkt;
1865 iov[1].iov_len = c->tcp_write_pkt_len;
1866 } else {
1867 iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1868 iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1869 iov[1].iov_base = sldns_buffer_begin(buffer);
1870 iov[1].iov_len = sldns_buffer_limit(buffer);
1871 }
1872 log_assert(iov[0].iov_len > 0);
1873 msg.msg_name = &c->repinfo.addr;
1874 msg.msg_namelen = c->repinfo.addrlen;
1875 msg.msg_iov = iov;
1876 msg.msg_iovlen = 2;
1877 r = sendmsg(fd, &msg, MSG_FASTOPEN);
1878 if (r == -1) {
1879 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
1880 /* Handshake is underway, maybe because no TFO cookie available.
1881 Come back to write the message*/
1882 if(errno == EINPROGRESS || errno == EWOULDBLOCK)
1883 return 1;
1884 #endif
1885 if(errno == EINTR || errno == EAGAIN)
1886 return 1;
1887 /* Not handling EISCONN here as shouldn't ever hit that case.*/
1888 if(errno != EPIPE
1889 #ifdef EOPNOTSUPP
1890 /* if /proc/sys/net/ipv4/tcp_fastopen is
1891 * disabled on Linux, sendmsg may return
1892 * 'Operation not supported', if so
1893 * fallthrough to ordinary connect. */
1894 && errno != EOPNOTSUPP
1895 #endif
1896 && errno != 0) {
1897 if(verbosity < 2)
1898 return 0; /* silence lots of chatter in the logs */
1899 log_err_addr("tcp sendmsg", strerror(errno),
1900 &c->repinfo.addr, c->repinfo.addrlen);
1901 return 0;
1902 }
1903 verbose(VERB_ALGO, "tcp sendmsg for fastopen failed (with %s), try normal connect", strerror(errno));
1904 /* fallthrough to nonFASTOPEN
1905 * (MSG_FASTOPEN on Linux 3 produces EPIPE)
1906 * we need to perform connect() */
1907 if(connect(fd, (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen) == -1) {
1908 #ifdef EINPROGRESS
1909 if(errno == EINPROGRESS)
1910 return 1; /* wait until connect done*/
1911 #endif
1912 #ifdef USE_WINSOCK
1913 if(WSAGetLastError() == WSAEINPROGRESS ||
1914 WSAGetLastError() == WSAEWOULDBLOCK)
1915 return 1; /* wait until connect done*/
1916 #endif
1917 if(tcp_connect_errno_needs_log(
1918 (struct sockaddr *)&c->repinfo.addr, c->repinfo.addrlen)) {
1919 log_err_addr("outgoing tcp: connect after EPIPE for fastopen",
1920 strerror(errno), &c->repinfo.addr, c->repinfo.addrlen);
1921 }
1922 return 0;
1923 }
1924
1925 } else {
1926 if(c->tcp_write_and_read) {
1927 c->tcp_write_byte_count += r;
1928 if(c->tcp_write_byte_count < sizeof(uint16_t))
1929 return 1;
1930 } else {
1931 c->tcp_byte_count += r;
1932 if(c->tcp_byte_count < sizeof(uint16_t))
1933 return 1;
1934 sldns_buffer_set_position(buffer, c->tcp_byte_count -
1935 sizeof(uint16_t));
1936 }
1937 if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
1938 tcp_callback_writer(c);
1939 return 1;
1940 }
1941 }
1942 }
1943 #endif /* USE_MSG_FASTOPEN */
1944
1945 if((c->tcp_write_and_read?c->tcp_write_byte_count:c->tcp_byte_count) < sizeof(uint16_t)) {
1946 uint16_t len = htons(c->tcp_write_and_read?c->tcp_write_pkt_len:sldns_buffer_limit(buffer));
1947 #ifdef HAVE_WRITEV
1948 struct iovec iov[2];
1949 if(c->tcp_write_and_read) {
1950 iov[0].iov_base = (uint8_t*)&len + c->tcp_write_byte_count;
1951 iov[0].iov_len = sizeof(uint16_t) - c->tcp_write_byte_count;
1952 iov[1].iov_base = c->tcp_write_pkt;
1953 iov[1].iov_len = c->tcp_write_pkt_len;
1954 } else {
1955 iov[0].iov_base = (uint8_t*)&len + c->tcp_byte_count;
1956 iov[0].iov_len = sizeof(uint16_t) - c->tcp_byte_count;
1957 iov[1].iov_base = sldns_buffer_begin(buffer);
1958 iov[1].iov_len = sldns_buffer_limit(buffer);
1959 }
1960 log_assert(iov[0].iov_len > 0);
1961 r = writev(fd, iov, 2);
1962 #else /* HAVE_WRITEV */
1963 if(c->tcp_write_and_read) {
1964 r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_write_byte_count),
1965 sizeof(uint16_t)-c->tcp_write_byte_count, 0);
1966 } else {
1967 r = send(fd, (void*)(((uint8_t*)&len)+c->tcp_byte_count),
1968 sizeof(uint16_t)-c->tcp_byte_count, 0);
1969 }
1970 #endif /* HAVE_WRITEV */
1971 if(r == -1) {
1972 #ifndef USE_WINSOCK
1973 # ifdef EPIPE
1974 if(errno == EPIPE && verbosity < 2)
1975 return 0; /* silence 'broken pipe' */
1976 #endif
1977 if(errno == EINTR || errno == EAGAIN)
1978 return 1;
1979 #ifdef ECONNRESET
1980 if(errno == ECONNRESET && verbosity < 2)
1981 return 0; /* silence reset by peer */
1982 #endif
1983 # ifdef HAVE_WRITEV
1984 log_err_addr("tcp writev", strerror(errno),
1985 &c->repinfo.addr, c->repinfo.addrlen);
1986 # else /* HAVE_WRITEV */
1987 log_err_addr("tcp send s", strerror(errno),
1988 &c->repinfo.addr, c->repinfo.addrlen);
1989 # endif /* HAVE_WRITEV */
1990 #else
1991 if(WSAGetLastError() == WSAENOTCONN)
1992 return 1;
1993 if(WSAGetLastError() == WSAEINPROGRESS)
1994 return 1;
1995 if(WSAGetLastError() == WSAEWOULDBLOCK) {
1996 ub_winsock_tcp_wouldblock(c->ev->ev,
1997 UB_EV_WRITE);
1998 return 1;
1999 }
2000 if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
2001 return 0; /* silence reset by peer */
2002 log_err_addr("tcp send s",
2003 wsa_strerror(WSAGetLastError()),
2004 &c->repinfo.addr, c->repinfo.addrlen);
2005 #endif
2006 return 0;
2007 }
2008 if(c->tcp_write_and_read) {
2009 c->tcp_write_byte_count += r;
2010 if(c->tcp_write_byte_count < sizeof(uint16_t))
2011 return 1;
2012 } else {
2013 c->tcp_byte_count += r;
2014 if(c->tcp_byte_count < sizeof(uint16_t))
2015 return 1;
2016 sldns_buffer_set_position(buffer, c->tcp_byte_count -
2017 sizeof(uint16_t));
2018 }
2019 if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
2020 tcp_callback_writer(c);
2021 return 1;
2022 }
2023 }
2024 log_assert(c->tcp_write_and_read || sldns_buffer_remaining(buffer) > 0);
2025 log_assert(!c->tcp_write_and_read || c->tcp_write_byte_count < c->tcp_write_pkt_len + 2);
2026 if(c->tcp_write_and_read) {
2027 r = send(fd, (void*)(c->tcp_write_pkt + c->tcp_write_byte_count - 2),
2028 c->tcp_write_pkt_len + 2 - c->tcp_write_byte_count, 0);
2029 } else {
2030 r = send(fd, (void*)sldns_buffer_current(buffer),
2031 sldns_buffer_remaining(buffer), 0);
2032 }
2033 if(r == -1) {
2034 #ifndef USE_WINSOCK
2035 if(errno == EINTR || errno == EAGAIN)
2036 return 1;
2037 #ifdef ECONNRESET
2038 if(errno == ECONNRESET && verbosity < 2)
2039 return 0; /* silence reset by peer */
2040 #endif
2041 #else
2042 if(WSAGetLastError() == WSAEINPROGRESS)
2043 return 1;
2044 if(WSAGetLastError() == WSAEWOULDBLOCK) {
2045 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2046 return 1;
2047 }
2048 if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
2049 return 0; /* silence reset by peer */
2050 #endif
2051 log_err_addr("tcp send r", sock_strerror(errno),
2052 &c->repinfo.addr, c->repinfo.addrlen);
2053 return 0;
2054 }
2055 if(c->tcp_write_and_read) {
2056 c->tcp_write_byte_count += r;
2057 } else {
2058 sldns_buffer_skip(buffer, r);
2059 }
2060
2061 if((!c->tcp_write_and_read && sldns_buffer_remaining(buffer) == 0) || (c->tcp_write_and_read && c->tcp_write_byte_count == c->tcp_write_pkt_len + 2)) {
2062 tcp_callback_writer(c);
2063 }
2064
2065 return 1;
2066 }
2067
2068 /** read again to drain buffers when there could be more to read */
2069 static void
2070 tcp_req_info_read_again(int fd, struct comm_point* c)
2071 {
2072 while(c->tcp_req_info->read_again) {
2073 int r;
2074 c->tcp_req_info->read_again = 0;
2075 if(c->tcp_is_reading)
2076 r = comm_point_tcp_handle_read(fd, c, 0);
2077 else r = comm_point_tcp_handle_write(fd, c);
2078 if(!r) {
2079 reclaim_tcp_handler(c);
2080 if(!c->tcp_do_close) {
2081 fptr_ok(fptr_whitelist_comm_point(
2082 c->callback));
2083 (void)(*c->callback)(c, c->cb_arg,
2084 NETEVENT_CLOSED, NULL);
2085 }
2086 return;
2087 }
2088 }
2089 }
2090
2091 /** read again to drain buffers when there could be more to read */
2092 static void
2093 tcp_more_read_again(int fd, struct comm_point* c)
2094 {
2095 /* if the packet is done, but another one could be waiting on
2096 * the connection, the callback signals this, and we try again */
2097 /* this continues until the read routines get EAGAIN or so,
2098 * and thus does not call the callback, and the bool is 0 */
2099 int* moreread = c->tcp_more_read_again;
2100 while(moreread && *moreread) {
2101 *moreread = 0;
2102 if(!comm_point_tcp_handle_read(fd, c, 0)) {
2103 reclaim_tcp_handler(c);
2104 if(!c->tcp_do_close) {
2105 fptr_ok(fptr_whitelist_comm_point(
2106 c->callback));
2107 (void)(*c->callback)(c, c->cb_arg,
2108 NETEVENT_CLOSED, NULL);
2109 }
2110 return;
2111 }
2112 }
2113 }
2114
2115 /** write again to fill up when there could be more to write */
2116 static void
2117 tcp_more_write_again(int fd, struct comm_point* c)
2118 {
2119 /* if the packet is done, but another is waiting to be written,
2120 * the callback signals it and we try again. */
2121 /* this continues until the write routines get EAGAIN or so,
2122 * and thus does not call the callback, and the bool is 0 */
2123 int* morewrite = c->tcp_more_write_again;
2124 while(morewrite && *morewrite) {
2125 *morewrite = 0;
2126 if(!comm_point_tcp_handle_write(fd, c)) {
2127 reclaim_tcp_handler(c);
2128 if(!c->tcp_do_close) {
2129 fptr_ok(fptr_whitelist_comm_point(
2130 c->callback));
2131 (void)(*c->callback)(c, c->cb_arg,
2132 NETEVENT_CLOSED, NULL);
2133 }
2134 return;
2135 }
2136 }
2137 }
2138
2139 void
2140 comm_point_tcp_handle_callback(int fd, short event, void* arg)
2141 {
2142 struct comm_point* c = (struct comm_point*)arg;
2143 log_assert(c->type == comm_tcp);
2144 ub_comm_base_now(c->ev->base);
2145
2146 #ifdef USE_DNSCRYPT
2147 /* Initialize if this is a dnscrypt socket */
2148 if(c->tcp_parent) {
2149 c->dnscrypt = c->tcp_parent->dnscrypt;
2150 }
2151 if(c->dnscrypt && c->dnscrypt_buffer == c->buffer) {
2152 c->dnscrypt_buffer = sldns_buffer_new(sldns_buffer_capacity(c->buffer));
2153 if(!c->dnscrypt_buffer) {
2154 log_err("Could not allocate dnscrypt buffer");
2155 reclaim_tcp_handler(c);
2156 if(!c->tcp_do_close) {
2157 fptr_ok(fptr_whitelist_comm_point(
2158 c->callback));
2159 (void)(*c->callback)(c, c->cb_arg,
2160 NETEVENT_CLOSED, NULL);
2161 }
2162 return;
2163 }
2164 }
2165 #endif
2166
2167 if(event&UB_EV_TIMEOUT) {
2168 verbose(VERB_QUERY, "tcp took too long, dropped");
2169 reclaim_tcp_handler(c);
2170 if(!c->tcp_do_close) {
2171 fptr_ok(fptr_whitelist_comm_point(c->callback));
2172 (void)(*c->callback)(c, c->cb_arg,
2173 NETEVENT_TIMEOUT, NULL);
2174 }
2175 return;
2176 }
2177 if(event&UB_EV_READ
2178 #ifdef USE_MSG_FASTOPEN
2179 && !(c->tcp_do_fastopen && (event&UB_EV_WRITE))
2180 #endif
2181 ) {
2182 int has_tcpq = (c->tcp_req_info != NULL);
2183 int* moreread = c->tcp_more_read_again;
2184 if(!comm_point_tcp_handle_read(fd, c, 0)) {
2185 reclaim_tcp_handler(c);
2186 if(!c->tcp_do_close) {
2187 fptr_ok(fptr_whitelist_comm_point(
2188 c->callback));
2189 (void)(*c->callback)(c, c->cb_arg,
2190 NETEVENT_CLOSED, NULL);
2191 }
2192 return;
2193 }
2194 if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
2195 tcp_req_info_read_again(fd, c);
2196 if(moreread && *moreread)
2197 tcp_more_read_again(fd, c);
2198 return;
2199 }
2200 if(event&UB_EV_WRITE) {
2201 int has_tcpq = (c->tcp_req_info != NULL);
2202 int* morewrite = c->tcp_more_write_again;
2203 if(!comm_point_tcp_handle_write(fd, c)) {
2204 reclaim_tcp_handler(c);
2205 if(!c->tcp_do_close) {
2206 fptr_ok(fptr_whitelist_comm_point(
2207 c->callback));
2208 (void)(*c->callback)(c, c->cb_arg,
2209 NETEVENT_CLOSED, NULL);
2210 }
2211 return;
2212 }
2213 if(has_tcpq && c->tcp_req_info && c->tcp_req_info->read_again)
2214 tcp_req_info_read_again(fd, c);
2215 if(morewrite && *morewrite)
2216 tcp_more_write_again(fd, c);
2217 return;
2218 }
2219 log_err("Ignored event %d for tcphdl.", event);
2220 }
2221
2222 /** Make http handler free for next assignment */
2223 static void
2224 reclaim_http_handler(struct comm_point* c)
2225 {
2226 log_assert(c->type == comm_http);
2227 if(c->ssl) {
2228 #ifdef HAVE_SSL
2229 SSL_shutdown(c->ssl);
2230 SSL_free(c->ssl);
2231 c->ssl = NULL;
2232 #endif
2233 }
2234 comm_point_close(c);
2235 if(c->tcp_parent) {
2236 if(c != c->tcp_parent->tcp_free) {
2237 c->tcp_parent->cur_tcp_count--;
2238 c->tcp_free = c->tcp_parent->tcp_free;
2239 c->tcp_parent->tcp_free = c;
2240 }
2241 if(!c->tcp_free) {
2242 /* re-enable listening on accept socket */
2243 comm_point_start_listening(c->tcp_parent, -1, -1);
2244 }
2245 }
2246 }
2247
2248 /** read more data for http (with ssl) */
2249 static int
2250 ssl_http_read_more(struct comm_point* c)
2251 {
2252 #ifdef HAVE_SSL
2253 int r;
2254 log_assert(sldns_buffer_remaining(c->buffer) > 0);
2255 ERR_clear_error();
2256 r = SSL_read(c->ssl, (void*)sldns_buffer_current(c->buffer),
2257 (int)sldns_buffer_remaining(c->buffer));
2258 if(r <= 0) {
2259 int want = SSL_get_error(c->ssl, r);
2260 if(want == SSL_ERROR_ZERO_RETURN) {
2261 return 0; /* shutdown, closed */
2262 } else if(want == SSL_ERROR_WANT_READ) {
2263 return 1; /* read more later */
2264 } else if(want == SSL_ERROR_WANT_WRITE) {
2265 c->ssl_shake_state = comm_ssl_shake_hs_write;
2266 comm_point_listen_for_rw(c, 0, 1);
2267 return 1;
2268 } else if(want == SSL_ERROR_SYSCALL) {
2269 #ifdef ECONNRESET
2270 if(errno == ECONNRESET && verbosity < 2)
2271 return 0; /* silence reset by peer */
2272 #endif
2273 if(errno != 0)
2274 log_err("SSL_read syscall: %s",
2275 strerror(errno));
2276 return 0;
2277 }
2278 log_crypto_err("could not SSL_read");
2279 return 0;
2280 }
2281 verbose(VERB_ALGO, "ssl http read more skip to %d + %d",
2282 (int)sldns_buffer_position(c->buffer), (int)r);
2283 sldns_buffer_skip(c->buffer, (ssize_t)r);
2284 return 1;
2285 #else
2286 (void)c;
2287 return 0;
2288 #endif /* HAVE_SSL */
2289 }
2290
2291 /** read more data for http */
2292 static int
2293 http_read_more(int fd, struct comm_point* c)
2294 {
2295 ssize_t r;
2296 log_assert(sldns_buffer_remaining(c->buffer) > 0);
2297 r = recv(fd, (void*)sldns_buffer_current(c->buffer),
2298 sldns_buffer_remaining(c->buffer), 0);
2299 if(r == 0) {
2300 return 0;
2301 } else if(r == -1) {
2302 #ifndef USE_WINSOCK
2303 if(errno == EINTR || errno == EAGAIN)
2304 return 1;
2305 #else /* USE_WINSOCK */
2306 if(WSAGetLastError() == WSAECONNRESET)
2307 return 0;
2308 if(WSAGetLastError() == WSAEINPROGRESS)
2309 return 1;
2310 if(WSAGetLastError() == WSAEWOULDBLOCK) {
2311 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
2312 return 1;
2313 }
2314 #endif
2315 log_err_addr("read (in http r)", sock_strerror(errno),
2316 &c->repinfo.addr, c->repinfo.addrlen);
2317 return 0;
2318 }
2319 verbose(VERB_ALGO, "http read more skip to %d + %d",
2320 (int)sldns_buffer_position(c->buffer), (int)r);
2321 sldns_buffer_skip(c->buffer, r);
2322 return 1;
2323 }
2324
2325 /** return true if http header has been read (one line complete) */
2326 static int
2327 http_header_done(sldns_buffer* buf)
2328 {
2329 size_t i;
2330 for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
2331 /* there was a \r before the \n, but we ignore that */
2332 if((char)sldns_buffer_read_u8_at(buf, i) == '\n')
2333 return 1;
2334 }
2335 return 0;
2336 }
2337
2338 /** return character string into buffer for header line, moves buffer
2339 * past that line and puts zero terminator into linefeed-newline */
2340 static char*
2341 http_header_line(sldns_buffer* buf)
2342 {
2343 char* result = (char*)sldns_buffer_current(buf);
2344 size_t i;
2345 for(i=sldns_buffer_position(buf); i<sldns_buffer_limit(buf); i++) {
2346 /* terminate the string on the \r */
2347 if((char)sldns_buffer_read_u8_at(buf, i) == '\r')
2348 sldns_buffer_write_u8_at(buf, i, 0);
2349 /* terminate on the \n and skip past the it and done */
2350 if((char)sldns_buffer_read_u8_at(buf, i) == '\n') {
2351 sldns_buffer_write_u8_at(buf, i, 0);
2352 sldns_buffer_set_position(buf, i+1);
2353 return result;
2354 }
2355 }
2356 return NULL;
2357 }
2358
2359 /** move unread buffer to start and clear rest for putting the rest into it */
2360 static void
2361 http_moveover_buffer(sldns_buffer* buf)
2362 {
2363 size_t pos = sldns_buffer_position(buf);
2364 size_t len = sldns_buffer_remaining(buf);
2365 sldns_buffer_clear(buf);
2366 memmove(sldns_buffer_begin(buf), sldns_buffer_at(buf, pos), len);
2367 sldns_buffer_set_position(buf, len);
2368 }
2369
2370 /** a http header is complete, process it */
2371 static int
2372 http_process_initial_header(struct comm_point* c)
2373 {
2374 char* line = http_header_line(c->buffer);
2375 if(!line) return 1;
2376 verbose(VERB_ALGO, "http header: %s", line);
2377 if(strncasecmp(line, "HTTP/1.1 ", 9) == 0) {
2378 /* check returncode */
2379 if(line[9] != '2') {
2380 verbose(VERB_ALGO, "http bad status %s", line+9);
2381 return 0;
2382 }
2383 } else if(strncasecmp(line, "Content-Length: ", 16) == 0) {
2384 if(!c->http_is_chunked)
2385 c->tcp_byte_count = (size_t)atoi(line+16);
2386 } else if(strncasecmp(line, "Transfer-Encoding: chunked", 19+7) == 0) {
2387 c->tcp_byte_count = 0;
2388 c->http_is_chunked = 1;
2389 } else if(line[0] == 0) {
2390 /* end of initial headers */
2391 c->http_in_headers = 0;
2392 if(c->http_is_chunked)
2393 c->http_in_chunk_headers = 1;
2394 /* remove header text from front of buffer
2395 * the buffer is going to be used to return the data segment
2396 * itself and we don't want the header to get returned
2397 * prepended with it */
2398 http_moveover_buffer(c->buffer);
2399 sldns_buffer_flip(c->buffer);
2400 return 1;
2401 }
2402 /* ignore other headers */
2403 return 1;
2404 }
2405
2406 /** a chunk header is complete, process it, return 0=fail, 1=continue next
2407 * header line, 2=done with chunked transfer*/
2408 static int
2409 http_process_chunk_header(struct comm_point* c)
2410 {
2411 char* line = http_header_line(c->buffer);
2412 if(!line) return 1;
2413 if(c->http_in_chunk_headers == 3) {
2414 verbose(VERB_ALGO, "http chunk trailer: %s", line);
2415 /* are we done ? */
2416 if(line[0] == 0 && c->tcp_byte_count == 0) {
2417 /* callback of http reader when NETEVENT_DONE,
2418 * end of data, with no data in buffer */
2419 sldns_buffer_set_position(c->buffer, 0);
2420 sldns_buffer_set_limit(c->buffer, 0);
2421 fptr_ok(fptr_whitelist_comm_point(c->callback));
2422 (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
2423 /* return that we are done */
2424 return 2;
2425 }
2426 if(line[0] == 0) {
2427 /* continue with header of the next chunk */
2428 c->http_in_chunk_headers = 1;
2429 /* remove header text from front of buffer */
2430 http_moveover_buffer(c->buffer);
2431 sldns_buffer_flip(c->buffer);
2432 return 1;
2433 }
2434 /* ignore further trail headers */
2435 return 1;
2436 }
2437 verbose(VERB_ALGO, "http chunk header: %s", line);
2438 if(c->http_in_chunk_headers == 1) {
2439 /* read chunked start line */
2440 char* end = NULL;
2441 c->tcp_byte_count = (size_t)strtol(line, &end, 16);
2442 if(end == line)
2443 return 0;
2444 c->http_in_chunk_headers = 0;
2445 /* remove header text from front of buffer */
2446 http_moveover_buffer(c->buffer);
2447 sldns_buffer_flip(c->buffer);
2448 if(c->tcp_byte_count == 0) {
2449 /* done with chunks, process chunk_trailer lines */
2450 c->http_in_chunk_headers = 3;
2451 }
2452 return 1;
2453 }
2454 /* ignore other headers */
2455 return 1;
2456 }
2457
2458 /** handle nonchunked data segment, 0=fail, 1=wait */
2459 static int
2460 http_nonchunk_segment(struct comm_point* c)
2461 {
2462 /* c->buffer at position..limit has new data we read in.
2463 * the buffer itself is full of nonchunked data.
2464 * we are looking to read tcp_byte_count more data
2465 * and then the transfer is done. */
2466 size_t remainbufferlen;
2467 size_t got_now = sldns_buffer_limit(c->buffer);
2468 if(c->tcp_byte_count <= got_now) {
2469 /* done, this is the last data fragment */
2470 c->http_stored = 0;
2471 sldns_buffer_set_position(c->buffer, 0);
2472 fptr_ok(fptr_whitelist_comm_point(c->callback));
2473 (void)(*c->callback)(c, c->cb_arg, NETEVENT_DONE, NULL);
2474 return 1;
2475 }
2476 /* if we have the buffer space,
2477 * read more data collected into the buffer */
2478 remainbufferlen = sldns_buffer_capacity(c->buffer) -
2479 sldns_buffer_limit(c->buffer);
2480 if(remainbufferlen+got_now >= c->tcp_byte_count ||
2481 remainbufferlen >= (c->ssl?16384:2048)) {
2482 size_t total = sldns_buffer_limit(c->buffer);
2483 sldns_buffer_clear(c->buffer);
2484 sldns_buffer_set_position(c->buffer, total);
2485 c->http_stored = total;
2486 /* return and wait to read more */
2487 return 1;
2488 }
2489 /* call callback with this data amount, then
2490 * wait for more */
2491 c->tcp_byte_count -= got_now;
2492 c->http_stored = 0;
2493 sldns_buffer_set_position(c->buffer, 0);
2494 fptr_ok(fptr_whitelist_comm_point(c->callback));
2495 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2496 /* c->callback has to buffer_clear(c->buffer). */
2497 /* return and wait to read more */
2498 return 1;
2499 }
2500
2501 /** handle chunked data segment, return 0=fail, 1=wait, 2=process more */
2502 static int
2503 http_chunked_segment(struct comm_point* c)
2504 {
2505 /* the c->buffer has from position..limit new data we read. */
2506 /* the current chunk has length tcp_byte_count.
2507 * once we read that read more chunk headers.
2508 */
2509 size_t remainbufferlen;
2510 size_t got_now = sldns_buffer_limit(c->buffer) - c->http_stored;
2511 verbose(VERB_ALGO, "http_chunked_segment: got now %d, tcpbytcount %d, http_stored %d, buffer pos %d, buffer limit %d", (int)got_now, (int)c->tcp_byte_count, (int)c->http_stored, (int)sldns_buffer_position(c->buffer), (int)sldns_buffer_limit(c->buffer));
2512 if(c->tcp_byte_count <= got_now) {
2513 /* the chunk has completed (with perhaps some extra data
2514 * from next chunk header and next chunk) */
2515 /* save too much info into temp buffer */
2516 size_t fraglen;
2517 struct comm_reply repinfo;
2518 c->http_stored = 0;
2519 sldns_buffer_skip(c->buffer, (ssize_t)c->tcp_byte_count);
2520 sldns_buffer_clear(c->http_temp);
2521 sldns_buffer_write(c->http_temp,
2522 sldns_buffer_current(c->buffer),
2523 sldns_buffer_remaining(c->buffer));
2524 sldns_buffer_flip(c->http_temp);
2525
2526 /* callback with this fragment */
2527 fraglen = sldns_buffer_position(c->buffer);
2528 sldns_buffer_set_position(c->buffer, 0);
2529 sldns_buffer_set_limit(c->buffer, fraglen);
2530 repinfo = c->repinfo;
2531 fptr_ok(fptr_whitelist_comm_point(c->callback));
2532 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &repinfo);
2533 /* c->callback has to buffer_clear(). */
2534
2535 /* is commpoint deleted? */
2536 if(!repinfo.c) {
2537 return 1;
2538 }
2539 /* copy waiting info */
2540 sldns_buffer_clear(c->buffer);
2541 sldns_buffer_write(c->buffer,
2542 sldns_buffer_begin(c->http_temp),
2543 sldns_buffer_remaining(c->http_temp));
2544 sldns_buffer_flip(c->buffer);
2545 /* process end of chunk trailer header lines, until
2546 * an empty line */
2547 c->http_in_chunk_headers = 3;
2548 /* process more data in buffer (if any) */
2549 return 2;
2550 }
2551 c->tcp_byte_count -= got_now;
2552
2553 /* if we have the buffer space,
2554 * read more data collected into the buffer */
2555 remainbufferlen = sldns_buffer_capacity(c->buffer) -
2556 sldns_buffer_limit(c->buffer);
2557 if(remainbufferlen >= c->tcp_byte_count ||
2558 remainbufferlen >= 2048) {
2559 size_t total = sldns_buffer_limit(c->buffer);
2560 sldns_buffer_clear(c->buffer);
2561 sldns_buffer_set_position(c->buffer, total);
2562 c->http_stored = total;
2563 /* return and wait to read more */
2564 return 1;
2565 }
2566
2567 /* callback of http reader for a new part of the data */
2568 c->http_stored = 0;
2569 sldns_buffer_set_position(c->buffer, 0);
2570 fptr_ok(fptr_whitelist_comm_point(c->callback));
2571 (void)(*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, NULL);
2572 /* c->callback has to buffer_clear(c->buffer). */
2573 /* return and wait to read more */
2574 return 1;
2575 }
2576
2577 #ifdef HAVE_NGHTTP2
2578 /** Create new http2 session. Called when creating handling comm point. */
2579 static struct http2_session* http2_session_create(struct comm_point* c)
2580 {
2581 struct http2_session* session = calloc(1, sizeof(*session));
2582 if(!session) {
2583 log_err("malloc failure while creating http2 session");
2584 return NULL;
2585 }
2586 session->c = c;
2587
2588 return session;
2589 }
2590 #endif
2591
2592 /** Delete http2 session. After closing connection or on error */
2593 static void http2_session_delete(struct http2_session* h2_session)
2594 {
2595 #ifdef HAVE_NGHTTP2
2596 if(h2_session->callbacks)
2597 nghttp2_session_callbacks_del(h2_session->callbacks);
2598 free(h2_session);
2599 #else
2600 (void)h2_session;
2601 #endif
2602 }
2603
2604 #ifdef HAVE_NGHTTP2
2605 struct http2_stream* http2_stream_create(int32_t stream_id)
2606 {
2607 struct http2_stream* h2_stream = calloc(1, sizeof(*h2_stream));
2608 if(!h2_stream) {
2609 log_err("malloc failure while creating http2 stream");
2610 return NULL;
2611 }
2612 h2_stream->stream_id = stream_id;
2613 return h2_stream;
2614 }
2615
2616 /** Delete http2 stream. After session delete or stream close callback */
2617 static void http2_stream_delete(struct http2_session* h2_session,
2618 struct http2_stream* h2_stream)
2619 {
2620 if(h2_stream->mesh_state) {
2621 mesh_state_remove_reply(h2_stream->mesh, h2_stream->mesh_state,
2622 h2_session->c);
2623 h2_stream->mesh_state = NULL;
2624 }
2625 http2_req_stream_clear(h2_stream);
2626 free(h2_stream);
2627 }
2628 #endif
2629
2630 void http2_stream_add_meshstate(struct http2_stream* h2_stream,
2631 struct mesh_area* mesh, struct mesh_state* m)
2632 {
2633 h2_stream->mesh = mesh;
2634 h2_stream->mesh_state = m;
2635 }
2636
2637 /** delete http2 session server. After closing connection. */
2638 static void http2_session_server_delete(struct http2_session* h2_session)
2639 {
2640 #ifdef HAVE_NGHTTP2
2641 struct http2_stream* h2_stream, *next;
2642 nghttp2_session_del(h2_session->session); /* NULL input is fine */
2643 h2_session->session = NULL;
2644 for(h2_stream = h2_session->first_stream; h2_stream;) {
2645 next = h2_stream->next;
2646 http2_stream_delete(h2_session, h2_stream);
2647 h2_stream = next;
2648 }
2649 h2_session->first_stream = NULL;
2650 h2_session->is_drop = 0;
2651 h2_session->postpone_drop = 0;
2652 h2_session->c->h2_stream = NULL;
2653 #endif
2654 (void)h2_session;
2655 }
2656
2657 #ifdef HAVE_NGHTTP2
2658 void http2_session_add_stream(struct http2_session* h2_session,
2659 struct http2_stream* h2_stream)
2660 {
2661 if(h2_session->first_stream)
2662 h2_session->first_stream->prev = h2_stream;
2663 h2_stream->next = h2_session->first_stream;
2664 h2_session->first_stream = h2_stream;
2665 }
2666
2667 /** remove stream from session linked list. After stream close callback or
2668 * closing connection */
2669 static void http2_session_remove_stream(struct http2_session* h2_session,
2670 struct http2_stream* h2_stream)
2671 {
2672 if(h2_stream->prev)
2673 h2_stream->prev->next = h2_stream->next;
2674 else
2675 h2_session->first_stream = h2_stream->next;
2676 if(h2_stream->next)
2677 h2_stream->next->prev = h2_stream->prev;
2678
2679 }
2680
2681 int http2_stream_close_cb(nghttp2_session* ATTR_UNUSED(session),
2682 int32_t stream_id, uint32_t ATTR_UNUSED(error_code), void* cb_arg)
2683 {
2684 struct http2_stream* h2_stream;
2685 struct http2_session* h2_session = (struct http2_session*)cb_arg;
2686 if(!(h2_stream = nghttp2_session_get_stream_user_data(
2687 h2_session->session, stream_id))) {
2688 return 0;
2689 }
2690 http2_session_remove_stream(h2_session, h2_stream);
2691 http2_stream_delete(h2_session, h2_stream);
2692 return 0;
2693 }
2694
2695 ssize_t http2_recv_cb(nghttp2_session* ATTR_UNUSED(session), uint8_t* buf,
2696 size_t len, int ATTR_UNUSED(flags), void* cb_arg)
2697 {
2698 struct http2_session* h2_session = (struct http2_session*)cb_arg;
2699 ssize_t ret;
2700
2701 log_assert(h2_session->c->type == comm_http);
2702 log_assert(h2_session->c->h2_session);
2703
2704 #ifdef HAVE_SSL
2705 if(h2_session->c->ssl) {
2706 int r;
2707 ERR_clear_error();
2708 r = SSL_read(h2_session->c->ssl, buf, len);
2709 if(r <= 0) {
2710 int want = SSL_get_error(h2_session->c->ssl, r);
2711 if(want == SSL_ERROR_ZERO_RETURN) {
2712 return NGHTTP2_ERR_EOF;
2713 } else if(want == SSL_ERROR_WANT_READ) {
2714 return NGHTTP2_ERR_WOULDBLOCK;
2715 } else if(want == SSL_ERROR_WANT_WRITE) {
2716 h2_session->c->ssl_shake_state = comm_ssl_shake_hs_write;
2717 comm_point_listen_for_rw(h2_session->c, 0, 1);
2718 return NGHTTP2_ERR_WOULDBLOCK;
2719 } else if(want == SSL_ERROR_SYSCALL) {
2720 #ifdef ECONNRESET
2721 if(errno == ECONNRESET && verbosity < 2)
2722 return NGHTTP2_ERR_CALLBACK_FAILURE;
2723 #endif
2724 if(errno != 0)
2725 log_err("SSL_read syscall: %s",
2726 strerror(errno));
2727 return NGHTTP2_ERR_CALLBACK_FAILURE;
2728 }
2729 log_crypto_err("could not SSL_read");
2730 return NGHTTP2_ERR_CALLBACK_FAILURE;
2731 }
2732 return r;
2733 }
2734 #endif /* HAVE_SSL */
2735
2736 ret = recv(h2_session->c->fd, buf, len, 0);
2737 if(ret == 0) {
2738 return NGHTTP2_ERR_EOF;
2739 } else if(ret < 0) {
2740 #ifndef USE_WINSOCK
2741 if(errno == EINTR || errno == EAGAIN)
2742 return NGHTTP2_ERR_WOULDBLOCK;
2743 #ifdef ECONNRESET
2744 if(errno == ECONNRESET && verbosity < 2)
2745 return NGHTTP2_ERR_CALLBACK_FAILURE;
2746 #endif
2747 log_err_addr("could not http2 recv: %s", strerror(errno),
2748 &h2_session->c->repinfo.addr,
2749 h2_session->c->repinfo.addrlen);
2750 #else /* USE_WINSOCK */
2751 if(WSAGetLastError() == WSAECONNRESET)
2752 return NGHTTP2_ERR_CALLBACK_FAILURE;
2753 if(WSAGetLastError() == WSAEINPROGRESS)
2754 return NGHTTP2_ERR_WOULDBLOCK;
2755 if(WSAGetLastError() == WSAEWOULDBLOCK) {
2756 ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
2757 UB_EV_READ);
2758 return NGHTTP2_ERR_WOULDBLOCK;
2759 }
2760 log_err_addr("could not http2 recv: %s",
2761 wsa_strerror(WSAGetLastError()),
2762 &h2_session->c->repinfo.addr,
2763 h2_session->c->repinfo.addrlen);
2764 #endif
2765 return NGHTTP2_ERR_CALLBACK_FAILURE;
2766 }
2767 return ret;
2768 }
2769 #endif /* HAVE_NGHTTP2 */
2770
2771 /** Handle http2 read */
2772 static int
2773 comm_point_http2_handle_read(int ATTR_UNUSED(fd), struct comm_point* c)
2774 {
2775 #ifdef HAVE_NGHTTP2
2776 int ret;
2777 log_assert(c->h2_session);
2778
2779 /* reading until recv cb returns NGHTTP2_ERR_WOULDBLOCK */
2780 ret = nghttp2_session_recv(c->h2_session->session);
2781 if(ret) {
2782 if(ret != NGHTTP2_ERR_EOF &&
2783 ret != NGHTTP2_ERR_CALLBACK_FAILURE) {
2784 char a[256];
2785 addr_to_str(&c->repinfo.addr, c->repinfo.addrlen,
2786 a, sizeof(a));
2787 verbose(VERB_QUERY, "http2: session_recv from %s failed, "
2788 "error: %s", a, nghttp2_strerror(ret));
2789 }
2790 return 0;
2791 }
2792 if(nghttp2_session_want_write(c->h2_session->session)) {
2793 c->tcp_is_reading = 0;
2794 comm_point_stop_listening(c);
2795 comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
2796 } else if(!nghttp2_session_want_read(c->h2_session->session))
2797 return 0; /* connection can be closed */
2798 return 1;
2799 #else
2800 (void)c;
2801 return 0;
2802 #endif
2803 }
2804
2805 /**
2806 * Handle http reading callback.
2807 * @param fd: file descriptor of socket.
2808 * @param c: comm point to read from into buffer.
2809 * @return: 0 on error
2810 */
2811 static int
2812 comm_point_http_handle_read(int fd, struct comm_point* c)
2813 {
2814 log_assert(c->type == comm_http);
2815 log_assert(fd != -1);
2816
2817 /* if we are in ssl handshake, handle SSL handshake */
2818 #ifdef HAVE_SSL
2819 if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
2820 if(!ssl_handshake(c))
2821 return 0;
2822 if(c->ssl_shake_state != comm_ssl_shake_none)
2823 return 1;
2824 }
2825 #endif /* HAVE_SSL */
2826
2827 if(!c->tcp_is_reading)
2828 return 1;
2829
2830 if(c->use_h2) {
2831 return comm_point_http2_handle_read(fd, c);
2832 }
2833
2834 /* http version is <= http/1.1 */
2835
2836 if(c->http_min_version >= http_version_2) {
2837 /* HTTP/2 failed, not allowed to use lower version. */
2838 return 0;
2839 }
2840
2841 /* read more data */
2842 if(c->ssl) {
2843 if(!ssl_http_read_more(c))
2844 return 0;
2845 } else {
2846 if(!http_read_more(fd, c))
2847 return 0;
2848 }
2849
2850 if(c->http_stored >= sldns_buffer_position(c->buffer)) {
2851 /* read did not work but we wanted more data, there is
2852 * no bytes to process now. */
2853 return 1;
2854 }
2855 sldns_buffer_flip(c->buffer);
2856 /* if we are partway in a segment of data, position us at the point
2857 * where we left off previously */
2858 if(c->http_stored < sldns_buffer_limit(c->buffer))
2859 sldns_buffer_set_position(c->buffer, c->http_stored);
2860 else sldns_buffer_set_position(c->buffer, sldns_buffer_limit(c->buffer));
2861
2862 while(sldns_buffer_remaining(c->buffer) > 0) {
2863 /* Handle HTTP/1.x data */
2864 /* if we are reading headers, read more headers */
2865 if(c->http_in_headers || c->http_in_chunk_headers) {
2866 /* if header is done, process the header */
2867 if(!http_header_done(c->buffer)) {
2868 /* copy remaining data to front of buffer
2869 * and set rest for writing into it */
2870 http_moveover_buffer(c->buffer);
2871 /* return and wait to read more */
2872 return 1;
2873 }
2874 if(!c->http_in_chunk_headers) {
2875 /* process initial headers */
2876 if(!http_process_initial_header(c))
2877 return 0;
2878 } else {
2879 /* process chunk headers */
2880 int r = http_process_chunk_header(c);
2881 if(r == 0) return 0;
2882 if(r == 2) return 1; /* done */
2883 /* r == 1, continue */
2884 }
2885 /* see if we have more to process */
2886 continue;
2887 }
2888
2889 if(!c->http_is_chunked) {
2890 /* if we are reading nonchunks, process that*/
2891 return http_nonchunk_segment(c);
2892 } else {
2893 /* if we are reading chunks, read the chunk */
2894 int r = http_chunked_segment(c);
2895 if(r == 0) return 0;
2896 if(r == 1) return 1;
2897 continue;
2898 }
2899 }
2900 /* broke out of the loop; could not process header instead need
2901 * to read more */
2902 /* moveover any remaining data and read more data */
2903 http_moveover_buffer(c->buffer);
2904 /* return and wait to read more */
2905 return 1;
2906 }
2907
2908 /** check pending connect for http */
2909 static int
2910 http_check_connect(int fd, struct comm_point* c)
2911 {
2912 /* check for pending error from nonblocking connect */
2913 /* from Stevens, unix network programming, vol1, 3rd ed, p450*/
2914 int error = 0;
2915 socklen_t len = (socklen_t)sizeof(error);
2916 if(getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&error,
2917 &len) < 0){
2918 #ifndef USE_WINSOCK
2919 error = errno; /* on solaris errno is error */
2920 #else /* USE_WINSOCK */
2921 error = WSAGetLastError();
2922 #endif
2923 }
2924 #ifndef USE_WINSOCK
2925 #if defined(EINPROGRESS) && defined(EWOULDBLOCK)
2926 if(error == EINPROGRESS || error == EWOULDBLOCK)
2927 return 1; /* try again later */
2928 else
2929 #endif
2930 if(error != 0 && verbosity < 2)
2931 return 0; /* silence lots of chatter in the logs */
2932 else if(error != 0) {
2933 log_err_addr("http connect", strerror(error),
2934 &c->repinfo.addr, c->repinfo.addrlen);
2935 #else /* USE_WINSOCK */
2936 /* examine error */
2937 if(error == WSAEINPROGRESS)
2938 return 1;
2939 else if(error == WSAEWOULDBLOCK) {
2940 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
2941 return 1;
2942 } else if(error != 0 && verbosity < 2)
2943 return 0;
2944 else if(error != 0) {
2945 log_err_addr("http connect", wsa_strerror(error),
2946 &c->repinfo.addr, c->repinfo.addrlen);
2947 #endif /* USE_WINSOCK */
2948 return 0;
2949 }
2950 /* keep on processing this socket */
2951 return 2;
2952 }
2953
2954 /** write more data for http (with ssl) */
2955 static int
2956 ssl_http_write_more(struct comm_point* c)
2957 {
2958 #ifdef HAVE_SSL
2959 int r;
2960 log_assert(sldns_buffer_remaining(c->buffer) > 0);
2961 ERR_clear_error();
2962 r = SSL_write(c->ssl, (void*)sldns_buffer_current(c->buffer),
2963 (int)sldns_buffer_remaining(c->buffer));
2964 if(r <= 0) {
2965 int want = SSL_get_error(c->ssl, r);
2966 if(want == SSL_ERROR_ZERO_RETURN) {
2967 return 0; /* closed */
2968 } else if(want == SSL_ERROR_WANT_READ) {
2969 c->ssl_shake_state = comm_ssl_shake_hs_read;
2970 comm_point_listen_for_rw(c, 1, 0);
2971 return 1; /* wait for read condition */
2972 } else if(want == SSL_ERROR_WANT_WRITE) {
2973 return 1; /* write more later */
2974 } else if(want == SSL_ERROR_SYSCALL) {
2975 #ifdef EPIPE
2976 if(errno == EPIPE && verbosity < 2)
2977 return 0; /* silence 'broken pipe' */
2978 #endif
2979 if(errno != 0)
2980 log_err("SSL_write syscall: %s",
2981 strerror(errno));
2982 return 0;
2983 }
2984 log_crypto_err("could not SSL_write");
2985 return 0;
2986 }
2987 sldns_buffer_skip(c->buffer, (ssize_t)r);
2988 return 1;
2989 #else
2990 (void)c;
2991 return 0;
2992 #endif /* HAVE_SSL */
2993 }
2994
2995 /** write more data for http */
2996 static int
2997 http_write_more(int fd, struct comm_point* c)
2998 {
2999 ssize_t r;
3000 log_assert(sldns_buffer_remaining(c->buffer) > 0);
3001 r = send(fd, (void*)sldns_buffer_current(c->buffer),
3002 sldns_buffer_remaining(c->buffer), 0);
3003 if(r == -1) {
3004 #ifndef USE_WINSOCK
3005 if(errno == EINTR || errno == EAGAIN)
3006 return 1;
3007 #else
3008 if(WSAGetLastError() == WSAEINPROGRESS)
3009 return 1;
3010 if(WSAGetLastError() == WSAEWOULDBLOCK) {
3011 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
3012 return 1;
3013 }
3014 #endif
3015 log_err_addr("http send r", sock_strerror(errno),
3016 &c->repinfo.addr, c->repinfo.addrlen);
3017 return 0;
3018 }
3019 sldns_buffer_skip(c->buffer, r);
3020 return 1;
3021 }
3022
3023 #ifdef HAVE_NGHTTP2
3024 ssize_t http2_send_cb(nghttp2_session* ATTR_UNUSED(session), const uint8_t* buf,
3025 size_t len, int ATTR_UNUSED(flags), void* cb_arg)
3026 {
3027 ssize_t ret;
3028 struct http2_session* h2_session = (struct http2_session*)cb_arg;
3029 log_assert(h2_session->c->type == comm_http);
3030 log_assert(h2_session->c->h2_session);
3031
3032 #ifdef HAVE_SSL
3033 if(h2_session->c->ssl) {
3034 int r;
3035 ERR_clear_error();
3036 r = SSL_write(h2_session->c->ssl, buf, len);
3037 if(r <= 0) {
3038 int want = SSL_get_error(h2_session->c->ssl, r);
3039 if(want == SSL_ERROR_ZERO_RETURN) {
3040 return NGHTTP2_ERR_CALLBACK_FAILURE;
3041 } else if(want == SSL_ERROR_WANT_READ) {
3042 h2_session->c->ssl_shake_state = comm_ssl_shake_hs_read;
3043 comm_point_listen_for_rw(h2_session->c, 1, 0);
3044 return NGHTTP2_ERR_WOULDBLOCK;
3045 } else if(want == SSL_ERROR_WANT_WRITE) {
3046 return NGHTTP2_ERR_WOULDBLOCK;
3047 } else if(want == SSL_ERROR_SYSCALL) {
3048 #ifdef EPIPE
3049 if(errno == EPIPE && verbosity < 2)
3050 return NGHTTP2_ERR_CALLBACK_FAILURE;
3051 #endif
3052 if(errno != 0)
3053 log_err("SSL_write syscall: %s",
3054 strerror(errno));
3055 return NGHTTP2_ERR_CALLBACK_FAILURE;
3056 }
3057 log_crypto_err("could not SSL_write");
3058 return NGHTTP2_ERR_CALLBACK_FAILURE;
3059 }
3060 return r;
3061 }
3062 #endif /* HAVE_SSL */
3063
3064 ret = send(h2_session->c->fd, buf, len, 0);
3065 if(ret == 0) {
3066 return NGHTTP2_ERR_CALLBACK_FAILURE;
3067 } else if(ret < 0) {
3068 #ifndef USE_WINSOCK
3069 if(errno == EINTR || errno == EAGAIN)
3070 return NGHTTP2_ERR_WOULDBLOCK;
3071 #ifdef EPIPE
3072 if(errno == EPIPE && verbosity < 2)
3073 return NGHTTP2_ERR_CALLBACK_FAILURE;
3074 #endif
3075 #ifdef ECONNRESET
3076 if(errno == ECONNRESET && verbosity < 2)
3077 return NGHTTP2_ERR_CALLBACK_FAILURE;
3078 #endif
3079 log_err_addr("could not http2 write: %s", strerror(errno),
3080 &h2_session->c->repinfo.addr,
3081 h2_session->c->repinfo.addrlen);
3082 #else /* USE_WINSOCK */
3083 if(WSAGetLastError() == WSAENOTCONN)
3084 return NGHTTP2_ERR_WOULDBLOCK;
3085 if(WSAGetLastError() == WSAEINPROGRESS)
3086 return NGHTTP2_ERR_WOULDBLOCK;
3087 if(WSAGetLastError() == WSAEWOULDBLOCK) {
3088 ub_winsock_tcp_wouldblock(h2_session->c->ev->ev,
3089 UB_EV_WRITE);
3090 return NGHTTP2_ERR_WOULDBLOCK;
3091 }
3092 if(WSAGetLastError() == WSAECONNRESET && verbosity < 2)
3093 return NGHTTP2_ERR_CALLBACK_FAILURE;
3094 log_err_addr("could not http2 write: %s",
3095 wsa_strerror(WSAGetLastError()),
3096 &h2_session->c->repinfo.addr,
3097 h2_session->c->repinfo.addrlen);
3098 #endif
3099 return NGHTTP2_ERR_CALLBACK_FAILURE;
3100 }
3101 return ret;
3102 }
3103 #endif /* HAVE_NGHTTP2 */
3104
3105 /** Handle http2 writing */
3106 static int
3107 comm_point_http2_handle_write(int ATTR_UNUSED(fd), struct comm_point* c)
3108 {
3109 #ifdef HAVE_NGHTTP2
3110 int ret;
3111 log_assert(c->h2_session);
3112
3113 ret = nghttp2_session_send(c->h2_session->session);
3114 if(ret) {
3115 verbose(VERB_QUERY, "http2: session_send failed, "
3116 "error: %s", nghttp2_strerror(ret));
3117 return 0;
3118 }
3119
3120 if(nghttp2_session_want_read(c->h2_session->session)) {
3121 c->tcp_is_reading = 1;
3122 comm_point_stop_listening(c);
3123 comm_point_start_listening(c, -1, adjusted_tcp_timeout(c));
3124 } else if(!nghttp2_session_want_write(c->h2_session->session))
3125 return 0; /* connection can be closed */
3126 return 1;
3127 #else
3128 (void)c;
3129 return 0;
3130 #endif
3131 }
3132
3133 /**
3134 * Handle http writing callback.
3135 * @param fd: file descriptor of socket.
3136 * @param c: comm point to write buffer out of.
3137 * @return: 0 on error
3138 */
3139 static int
3140 comm_point_http_handle_write(int fd, struct comm_point* c)
3141 {
3142 log_assert(c->type == comm_http);
3143 log_assert(fd != -1);
3144
3145 /* check pending connect errors, if that fails, we wait for more,
3146 * or we can continue to write contents */
3147 if(c->tcp_check_nb_connect) {
3148 int r = http_check_connect(fd, c);
3149 if(r == 0) return 0;
3150 if(r == 1) return 1;
3151 c->tcp_check_nb_connect = 0;
3152 }
3153 /* if we are in ssl handshake, handle SSL handshake */
3154 #ifdef HAVE_SSL
3155 if(c->ssl && c->ssl_shake_state != comm_ssl_shake_none) {
3156 if(!ssl_handshake(c))
3157 return 0;
3158 if(c->ssl_shake_state != comm_ssl_shake_none)
3159 return 1;
3160 }
3161 #endif /* HAVE_SSL */
3162 if(c->tcp_is_reading)
3163 return 1;
3164
3165 if(c->use_h2) {
3166 return comm_point_http2_handle_write(fd, c);
3167 }
3168
3169 /* http version is <= http/1.1 */
3170
3171 if(c->http_min_version >= http_version_2) {
3172 /* HTTP/2 failed, not allowed to use lower version. */
3173 return 0;
3174 }
3175
3176 /* if we are writing, write more */
3177 if(c->ssl) {
3178 if(!ssl_http_write_more(c))
3179 return 0;
3180 } else {
3181 if(!http_write_more(fd, c))
3182 return 0;
3183 }
3184
3185 /* we write a single buffer contents, that can contain
3186 * the http request, and then flip to read the results */
3187 /* see if write is done */
3188 if(sldns_buffer_remaining(c->buffer) == 0) {
3189 sldns_buffer_clear(c->buffer);
3190 if(c->tcp_do_toggle_rw)
3191 c->tcp_is_reading = 1;
3192 c->tcp_byte_count = 0;
3193 /* switch from listening(write) to listening(read) */
3194 comm_point_stop_listening(c);
3195 comm_point_start_listening(c, -1, -1);
3196 }
3197 return 1;
3198 }
3199
3200 void
3201 comm_point_http_handle_callback(int fd, short event, void* arg)
3202 {
3203 struct comm_point* c = (struct comm_point*)arg;
3204 log_assert(c->type == comm_http);
3205 ub_comm_base_now(c->ev->base);
3206
3207 if(event&UB_EV_TIMEOUT) {
3208 verbose(VERB_QUERY, "http took too long, dropped");
3209 reclaim_http_handler(c);
3210 if(!c->tcp_do_close) {
3211 fptr_ok(fptr_whitelist_comm_point(c->callback));
3212 (void)(*c->callback)(c, c->cb_arg,
3213 NETEVENT_TIMEOUT, NULL);
3214 }
3215 return;
3216 }
3217 if(event&UB_EV_READ) {
3218 if(!comm_point_http_handle_read(fd, c)) {
3219 reclaim_http_handler(c);
3220 if(!c->tcp_do_close) {
3221 fptr_ok(fptr_whitelist_comm_point(
3222 c->callback));
3223 (void)(*c->callback)(c, c->cb_arg,
3224 NETEVENT_CLOSED, NULL);
3225 }
3226 }
3227 return;
3228 }
3229 if(event&UB_EV_WRITE) {
3230 if(!comm_point_http_handle_write(fd, c)) {
3231 reclaim_http_handler(c);
3232 if(!c->tcp_do_close) {
3233 fptr_ok(fptr_whitelist_comm_point(
3234 c->callback));
3235 (void)(*c->callback)(c, c->cb_arg,
3236 NETEVENT_CLOSED, NULL);
3237 }
3238 }
3239 return;
3240 }
3241 log_err("Ignored event %d for httphdl.", event);
3242 }
3243
3244 void comm_point_local_handle_callback(int fd, short event, void* arg)
3245 {
3246 struct comm_point* c = (struct comm_point*)arg;
3247 log_assert(c->type == comm_local);
3248 ub_comm_base_now(c->ev->base);
3249
3250 if(event&UB_EV_READ) {
3251 if(!comm_point_tcp_handle_read(fd, c, 1)) {
3252 fptr_ok(fptr_whitelist_comm_point(c->callback));
3253 (void)(*c->callback)(c, c->cb_arg, NETEVENT_CLOSED,
3254 NULL);
3255 }
3256 return;
3257 }
3258 log_err("Ignored event %d for localhdl.", event);
3259 }
3260
3261 void comm_point_raw_handle_callback(int ATTR_UNUSED(fd),
3262 short event, void* arg)
3263 {
3264 struct comm_point* c = (struct comm_point*)arg;
3265 int err = NETEVENT_NOERROR;
3266 log_assert(c->type == comm_raw);
3267 ub_comm_base_now(c->ev->base);
3268
3269 if(event&UB_EV_TIMEOUT)
3270 err = NETEVENT_TIMEOUT;
3271 fptr_ok(fptr_whitelist_comm_point_raw(c->callback));
3272 (void)(*c->callback)(c, c->cb_arg, err, NULL);
3273 }
3274
3275 struct comm_point*
3276 comm_point_create_udp(struct comm_base *base, int fd, sldns_buffer* buffer,
3277 comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3278 {
3279 struct comm_point* c = (struct comm_point*)calloc(1,
3280 sizeof(struct comm_point));
3281 short evbits;
3282 if(!c)
3283 return NULL;
3284 c->ev = (struct internal_event*)calloc(1,
3285 sizeof(struct internal_event));
3286 if(!c->ev) {
3287 free(c);
3288 return NULL;
3289 }
3290 c->ev->base = base;
3291 c->fd = fd;
3292 c->buffer = buffer;
3293 c->timeout = NULL;
3294 c->tcp_is_reading = 0;
3295 c->tcp_byte_count = 0;
3296 c->tcp_parent = NULL;
3297 c->max_tcp_count = 0;
3298 c->cur_tcp_count = 0;
3299 c->tcp_handlers = NULL;
3300 c->tcp_free = NULL;
3301 c->type = comm_udp;
3302 c->tcp_do_close = 0;
3303 c->do_not_close = 0;
3304 c->tcp_do_toggle_rw = 0;
3305 c->tcp_check_nb_connect = 0;
3306 #ifdef USE_MSG_FASTOPEN
3307 c->tcp_do_fastopen = 0;
3308 #endif
3309 #ifdef USE_DNSCRYPT
3310 c->dnscrypt = 0;
3311 c->dnscrypt_buffer = buffer;
3312 #endif
3313 c->inuse = 0;
3314 c->callback = callback;
3315 c->cb_arg = callback_arg;
3316 c->socket = socket;
3317 evbits = UB_EV_READ | UB_EV_PERSIST;
3318 /* ub_event stuff */
3319 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3320 comm_point_udp_callback, c);
3321 if(c->ev->ev == NULL) {
3322 log_err("could not baseset udp event");
3323 comm_point_delete(c);
3324 return NULL;
3325 }
3326 if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
3327 log_err("could not add udp event");
3328 comm_point_delete(c);
3329 return NULL;
3330 }
3331 c->event_added = 1;
3332 return c;
3333 }
3334
3335 struct comm_point*
3336 comm_point_create_udp_ancil(struct comm_base *base, int fd,
3337 sldns_buffer* buffer,
3338 comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3339 {
3340 struct comm_point* c = (struct comm_point*)calloc(1,
3341 sizeof(struct comm_point));
3342 short evbits;
3343 if(!c)
3344 return NULL;
3345 c->ev = (struct internal_event*)calloc(1,
3346 sizeof(struct internal_event));
3347 if(!c->ev) {
3348 free(c);
3349 return NULL;
3350 }
3351 c->ev->base = base;
3352 c->fd = fd;
3353 c->buffer = buffer;
3354 c->timeout = NULL;
3355 c->tcp_is_reading = 0;
3356 c->tcp_byte_count = 0;
3357 c->tcp_parent = NULL;
3358 c->max_tcp_count = 0;
3359 c->cur_tcp_count = 0;
3360 c->tcp_handlers = NULL;
3361 c->tcp_free = NULL;
3362 c->type = comm_udp;
3363 c->tcp_do_close = 0;
3364 c->do_not_close = 0;
3365 #ifdef USE_DNSCRYPT
3366 c->dnscrypt = 0;
3367 c->dnscrypt_buffer = buffer;
3368 #endif
3369 c->inuse = 0;
3370 c->tcp_do_toggle_rw = 0;
3371 c->tcp_check_nb_connect = 0;
3372 #ifdef USE_MSG_FASTOPEN
3373 c->tcp_do_fastopen = 0;
3374 #endif
3375 c->callback = callback;
3376 c->cb_arg = callback_arg;
3377 c->socket = socket;
3378 evbits = UB_EV_READ | UB_EV_PERSIST;
3379 /* ub_event stuff */
3380 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3381 comm_point_udp_ancil_callback, c);
3382 if(c->ev->ev == NULL) {
3383 log_err("could not baseset udp event");
3384 comm_point_delete(c);
3385 return NULL;
3386 }
3387 if(fd!=-1 && ub_event_add(c->ev->ev, c->timeout) != 0 ) {
3388 log_err("could not add udp event");
3389 comm_point_delete(c);
3390 return NULL;
3391 }
3392 c->event_added = 1;
3393 return c;
3394 }
3395
3396 static struct comm_point*
3397 comm_point_create_tcp_handler(struct comm_base *base,
3398 struct comm_point* parent, size_t bufsize,
3399 struct sldns_buffer* spoolbuf, comm_point_callback_type* callback,
3400 void* callback_arg, struct unbound_socket* socket)
3401 {
3402 struct comm_point* c = (struct comm_point*)calloc(1,
3403 sizeof(struct comm_point));
3404 short evbits;
3405 if(!c)
3406 return NULL;
3407 c->ev = (struct internal_event*)calloc(1,
3408 sizeof(struct internal_event));
3409 if(!c->ev) {
3410 free(c);
3411 return NULL;
3412 }
3413 c->ev->base = base;
3414 c->fd = -1;
3415 c->buffer = sldns_buffer_new(bufsize);
3416 if(!c->buffer) {
3417 free(c->ev);
3418 free(c);
3419 return NULL;
3420 }
3421 c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
3422 if(!c->timeout) {
3423 sldns_buffer_free(c->buffer);
3424 free(c->ev);
3425 free(c);
3426 return NULL;
3427 }
3428 c->tcp_is_reading = 0;
3429 c->tcp_byte_count = 0;
3430 c->tcp_parent = parent;
3431 c->tcp_timeout_msec = parent->tcp_timeout_msec;
3432 c->tcp_conn_limit = parent->tcp_conn_limit;
3433 c->tcl_addr = NULL;
3434 c->tcp_keepalive = 0;
3435 c->max_tcp_count = 0;
3436 c->cur_tcp_count = 0;
3437 c->tcp_handlers = NULL;
3438 c->tcp_free = NULL;
3439 c->type = comm_tcp;
3440 c->tcp_do_close = 0;
3441 c->do_not_close = 0;
3442 c->tcp_do_toggle_rw = 1;
3443 c->tcp_check_nb_connect = 0;
3444 #ifdef USE_MSG_FASTOPEN
3445 c->tcp_do_fastopen = 0;
3446 #endif
3447 #ifdef USE_DNSCRYPT
3448 c->dnscrypt = 0;
3449 /* We don't know just yet if this is a dnscrypt channel. Allocation
3450 * will be done when handling the callback. */
3451 c->dnscrypt_buffer = c->buffer;
3452 #endif
3453 c->repinfo.c = c;
3454 c->callback = callback;
3455 c->cb_arg = callback_arg;
3456 c->socket = socket;
3457 if(spoolbuf) {
3458 c->tcp_req_info = tcp_req_info_create(spoolbuf);
3459 if(!c->tcp_req_info) {
3460 log_err("could not create tcp commpoint");
3461 sldns_buffer_free(c->buffer);
3462 free(c->timeout);
3463 free(c->ev);
3464 free(c);
3465 return NULL;
3466 }
3467 c->tcp_req_info->cp = c;
3468 c->tcp_do_close = 1;
3469 c->tcp_do_toggle_rw = 0;
3470 }
3471 /* add to parent free list */
3472 c->tcp_free = parent->tcp_free;
3473 parent->tcp_free = c;
3474 /* ub_event stuff */
3475 evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
3476 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3477 comm_point_tcp_handle_callback, c);
3478 if(c->ev->ev == NULL)
3479 {
3480 log_err("could not basetset tcphdl event");
3481 parent->tcp_free = c->tcp_free;
3482 tcp_req_info_delete(c->tcp_req_info);
3483 sldns_buffer_free(c->buffer);
3484 free(c->timeout);
3485 free(c->ev);
3486 free(c);
3487 return NULL;
3488 }
3489 return c;
3490 }
3491
3492 static struct comm_point*
3493 comm_point_create_http_handler(struct comm_base *base,
3494 struct comm_point* parent, size_t bufsize, int harden_large_queries,
3495 uint32_t http_max_streams, char* http_endpoint,
3496 comm_point_callback_type* callback, void* callback_arg,
3497 struct unbound_socket* socket)
3498 {
3499 struct comm_point* c = (struct comm_point*)calloc(1,
3500 sizeof(struct comm_point));
3501 short evbits;
3502 if(!c)
3503 return NULL;
3504 c->ev = (struct internal_event*)calloc(1,
3505 sizeof(struct internal_event));
3506 if(!c->ev) {
3507 free(c);
3508 return NULL;
3509 }
3510 c->ev->base = base;
3511 c->fd = -1;
3512 c->buffer = sldns_buffer_new(bufsize);
3513 if(!c->buffer) {
3514 free(c->ev);
3515 free(c);
3516 return NULL;
3517 }
3518 c->timeout = (struct timeval*)malloc(sizeof(struct timeval));
3519 if(!c->timeout) {
3520 sldns_buffer_free(c->buffer);
3521 free(c->ev);
3522 free(c);
3523 return NULL;
3524 }
3525 c->tcp_is_reading = 0;
3526 c->tcp_byte_count = 0;
3527 c->tcp_parent = parent;
3528 c->tcp_timeout_msec = parent->tcp_timeout_msec;
3529 c->tcp_conn_limit = parent->tcp_conn_limit;
3530 c->tcl_addr = NULL;
3531 c->tcp_keepalive = 0;
3532 c->max_tcp_count = 0;
3533 c->cur_tcp_count = 0;
3534 c->tcp_handlers = NULL;
3535 c->tcp_free = NULL;
3536 c->type = comm_http;
3537 c->tcp_do_close = 1;
3538 c->do_not_close = 0;
3539 c->tcp_do_toggle_rw = 1; /* will be set to 0 after http2 upgrade */
3540 c->tcp_check_nb_connect = 0;
3541 #ifdef USE_MSG_FASTOPEN
3542 c->tcp_do_fastopen = 0;
3543 #endif
3544 #ifdef USE_DNSCRYPT
3545 c->dnscrypt = 0;
3546 c->dnscrypt_buffer = NULL;
3547 #endif
3548 c->repinfo.c = c;
3549 c->callback = callback;
3550 c->cb_arg = callback_arg;
3551 c->socket = socket;
3552
3553 c->http_min_version = http_version_2;
3554 c->http2_stream_max_qbuffer_size = bufsize;
3555 if(harden_large_queries && bufsize > 512)
3556 c->http2_stream_max_qbuffer_size = 512;
3557 c->http2_max_streams = http_max_streams;
3558 if(!(c->http_endpoint = strdup(http_endpoint))) {
3559 log_err("could not strdup http_endpoint");
3560 sldns_buffer_free(c->buffer);
3561 free(c->timeout);
3562 free(c->ev);
3563 free(c);
3564 return NULL;
3565 }
3566 c->use_h2 = 0;
3567 #ifdef HAVE_NGHTTP2
3568 if(!(c->h2_session = http2_session_create(c))) {
3569 log_err("could not create http2 session");
3570 free(c->http_endpoint);
3571 sldns_buffer_free(c->buffer);
3572 free(c->timeout);
3573 free(c->ev);
3574 free(c);
3575 return NULL;
3576 }
3577 if(!(c->h2_session->callbacks = http2_req_callbacks_create())) {
3578 log_err("could not create http2 callbacks");
3579 http2_session_delete(c->h2_session);
3580 free(c->http_endpoint);
3581 sldns_buffer_free(c->buffer);
3582 free(c->timeout);
3583 free(c->ev);
3584 free(c);
3585 return NULL;
3586 }
3587 #endif
3588
3589 /* add to parent free list */
3590 c->tcp_free = parent->tcp_free;
3591 parent->tcp_free = c;
3592 /* ub_event stuff */
3593 evbits = UB_EV_PERSIST | UB_EV_READ | UB_EV_TIMEOUT;
3594 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3595 comm_point_http_handle_callback, c);
3596 if(c->ev->ev == NULL)
3597 {
3598 log_err("could not set http handler event");
3599 parent->tcp_free = c->tcp_free;
3600 http2_session_delete(c->h2_session);
3601 sldns_buffer_free(c->buffer);
3602 free(c->timeout);
3603 free(c->ev);
3604 free(c);
3605 return NULL;
3606 }
3607 return c;
3608 }
3609
3610 struct comm_point*
3611 comm_point_create_tcp(struct comm_base *base, int fd, int num,
3612 int idle_timeout, int harden_large_queries,
3613 uint32_t http_max_streams, char* http_endpoint,
3614 struct tcl_list* tcp_conn_limit, size_t bufsize,
3615 struct sldns_buffer* spoolbuf, enum listen_type port_type,
3616 comm_point_callback_type* callback, void* callback_arg, struct unbound_socket* socket)
3617 {
3618 struct comm_point* c = (struct comm_point*)calloc(1,
3619 sizeof(struct comm_point));
3620 short evbits;
3621 int i;
3622 /* first allocate the TCP accept listener */
3623 if(!c)
3624 return NULL;
3625 c->ev = (struct internal_event*)calloc(1,
3626 sizeof(struct internal_event));
3627 if(!c->ev) {
3628 free(c);
3629 return NULL;
3630 }
3631 c->ev->base = base;
3632 c->fd = fd;
3633 c->buffer = NULL;
3634 c->timeout = NULL;
3635 c->tcp_is_reading = 0;
3636 c->tcp_byte_count = 0;
3637 c->tcp_timeout_msec = idle_timeout;
3638 c->tcp_conn_limit = tcp_conn_limit;
3639 c->tcl_addr = NULL;
3640 c->tcp_keepalive = 0;
3641 c->tcp_parent = NULL;
3642 c->max_tcp_count = num;
3643 c->cur_tcp_count = 0;
3644 c->tcp_handlers = (struct comm_point**)calloc((size_t)num,
3645 sizeof(struct comm_point*));
3646 if(!c->tcp_handlers) {
3647 free(c->ev);
3648 free(c);
3649 return NULL;
3650 }
3651 c->tcp_free = NULL;
3652 c->type = comm_tcp_accept;
3653 c->tcp_do_close = 0;
3654 c->do_not_close = 0;
3655 c->tcp_do_toggle_rw = 0;
3656 c->tcp_check_nb_connect = 0;
3657 #ifdef USE_MSG_FASTOPEN
3658 c->tcp_do_fastopen = 0;
3659 #endif
3660 #ifdef USE_DNSCRYPT
3661 c->dnscrypt = 0;
3662 c->dnscrypt_buffer = NULL;
3663 #endif
3664 c->callback = NULL;
3665 c->cb_arg = NULL;
3666 c->socket = socket;
3667 evbits = UB_EV_READ | UB_EV_PERSIST;
3668 /* ub_event stuff */
3669 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3670 comm_point_tcp_accept_callback, c);
3671 if(c->ev->ev == NULL) {
3672 log_err("could not baseset tcpacc event");
3673 comm_point_delete(c);
3674 return NULL;
3675 }
3676 if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3677 log_err("could not add tcpacc event");
3678 comm_point_delete(c);
3679 return NULL;
3680 }
3681 c->event_added = 1;
3682 /* now prealloc the handlers */
3683 for(i=0; i<num; i++) {
3684 if(port_type == listen_type_tcp ||
3685 port_type == listen_type_ssl ||
3686 port_type == listen_type_tcp_dnscrypt) {
3687 c->tcp_handlers[i] = comm_point_create_tcp_handler(base,
3688 c, bufsize, spoolbuf, callback, callback_arg, socket);
3689 } else if(port_type == listen_type_http) {
3690 c->tcp_handlers[i] = comm_point_create_http_handler(
3691 base, c, bufsize, harden_large_queries,
3692 http_max_streams, http_endpoint,
3693 callback, callback_arg, socket);
3694 }
3695 else {
3696 log_err("could not create tcp handler, unknown listen "
3697 "type");
3698 return NULL;
3699 }
3700 if(!c->tcp_handlers[i]) {
3701 comm_point_delete(c);
3702 return NULL;
3703 }
3704 }
3705
3706 return c;
3707 }
3708
3709 struct comm_point*
3710 comm_point_create_tcp_out(struct comm_base *base, size_t bufsize,
3711 comm_point_callback_type* callback, void* callback_arg)
3712 {
3713 struct comm_point* c = (struct comm_point*)calloc(1,
3714 sizeof(struct comm_point));
3715 short evbits;
3716 if(!c)
3717 return NULL;
3718 c->ev = (struct internal_event*)calloc(1,
3719 sizeof(struct internal_event));
3720 if(!c->ev) {
3721 free(c);
3722 return NULL;
3723 }
3724 c->ev->base = base;
3725 c->fd = -1;
3726 c->buffer = sldns_buffer_new(bufsize);
3727 if(!c->buffer) {
3728 free(c->ev);
3729 free(c);
3730 return NULL;
3731 }
3732 c->timeout = NULL;
3733 c->tcp_is_reading = 0;
3734 c->tcp_byte_count = 0;
3735 c->tcp_timeout_msec = TCP_QUERY_TIMEOUT;
3736 c->tcp_conn_limit = NULL;
3737 c->tcl_addr = NULL;
3738 c->tcp_keepalive = 0;
3739 c->tcp_parent = NULL;
3740 c->max_tcp_count = 0;
3741 c->cur_tcp_count = 0;
3742 c->tcp_handlers = NULL;
3743 c->tcp_free = NULL;
3744 c->type = comm_tcp;
3745 c->tcp_do_close = 0;
3746 c->do_not_close = 0;
3747 c->tcp_do_toggle_rw = 1;
3748 c->tcp_check_nb_connect = 1;
3749 #ifdef USE_MSG_FASTOPEN
3750 c->tcp_do_fastopen = 1;
3751 #endif
3752 #ifdef USE_DNSCRYPT
3753 c->dnscrypt = 0;
3754 c->dnscrypt_buffer = c->buffer;
3755 #endif
3756 c->repinfo.c = c;
3757 c->callback = callback;
3758 c->cb_arg = callback_arg;
3759 evbits = UB_EV_PERSIST | UB_EV_WRITE;
3760 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3761 comm_point_tcp_handle_callback, c);
3762 if(c->ev->ev == NULL)
3763 {
3764 log_err("could not baseset tcpout event");
3765 sldns_buffer_free(c->buffer);
3766 free(c->ev);
3767 free(c);
3768 return NULL;
3769 }
3770
3771 return c;
3772 }
3773
3774 struct comm_point*
3775 comm_point_create_http_out(struct comm_base *base, size_t bufsize,
3776 comm_point_callback_type* callback, void* callback_arg,
3777 sldns_buffer* temp)
3778 {
3779 struct comm_point* c = (struct comm_point*)calloc(1,
3780 sizeof(struct comm_point));
3781 short evbits;
3782 if(!c)
3783 return NULL;
3784 c->ev = (struct internal_event*)calloc(1,
3785 sizeof(struct internal_event));
3786 if(!c->ev) {
3787 free(c);
3788 return NULL;
3789 }
3790 c->ev->base = base;
3791 c->fd = -1;
3792 c->buffer = sldns_buffer_new(bufsize);
3793 if(!c->buffer) {
3794 free(c->ev);
3795 free(c);
3796 return NULL;
3797 }
3798 c->timeout = NULL;
3799 c->tcp_is_reading = 0;
3800 c->tcp_byte_count = 0;
3801 c->tcp_parent = NULL;
3802 c->max_tcp_count = 0;
3803 c->cur_tcp_count = 0;
3804 c->tcp_handlers = NULL;
3805 c->tcp_free = NULL;
3806 c->type = comm_http;
3807 c->tcp_do_close = 0;
3808 c->do_not_close = 0;
3809 c->tcp_do_toggle_rw = 1;
3810 c->tcp_check_nb_connect = 1;
3811 c->http_in_headers = 1;
3812 c->http_in_chunk_headers = 0;
3813 c->http_is_chunked = 0;
3814 c->http_temp = temp;
3815 #ifdef USE_MSG_FASTOPEN
3816 c->tcp_do_fastopen = 1;
3817 #endif
3818 #ifdef USE_DNSCRYPT
3819 c->dnscrypt = 0;
3820 c->dnscrypt_buffer = c->buffer;
3821 #endif
3822 c->repinfo.c = c;
3823 c->callback = callback;
3824 c->cb_arg = callback_arg;
3825 evbits = UB_EV_PERSIST | UB_EV_WRITE;
3826 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3827 comm_point_http_handle_callback, c);
3828 if(c->ev->ev == NULL)
3829 {
3830 log_err("could not baseset tcpout event");
3831 #ifdef HAVE_SSL
3832 SSL_free(c->ssl);
3833 #endif
3834 sldns_buffer_free(c->buffer);
3835 free(c->ev);
3836 free(c);
3837 return NULL;
3838 }
3839
3840 return c;
3841 }
3842
3843 struct comm_point*
3844 comm_point_create_local(struct comm_base *base, int fd, size_t bufsize,
3845 comm_point_callback_type* callback, void* callback_arg)
3846 {
3847 struct comm_point* c = (struct comm_point*)calloc(1,
3848 sizeof(struct comm_point));
3849 short evbits;
3850 if(!c)
3851 return NULL;
3852 c->ev = (struct internal_event*)calloc(1,
3853 sizeof(struct internal_event));
3854 if(!c->ev) {
3855 free(c);
3856 return NULL;
3857 }
3858 c->ev->base = base;
3859 c->fd = fd;
3860 c->buffer = sldns_buffer_new(bufsize);
3861 if(!c->buffer) {
3862 free(c->ev);
3863 free(c);
3864 return NULL;
3865 }
3866 c->timeout = NULL;
3867 c->tcp_is_reading = 1;
3868 c->tcp_byte_count = 0;
3869 c->tcp_parent = NULL;
3870 c->max_tcp_count = 0;
3871 c->cur_tcp_count = 0;
3872 c->tcp_handlers = NULL;
3873 c->tcp_free = NULL;
3874 c->type = comm_local;
3875 c->tcp_do_close = 0;
3876 c->do_not_close = 1;
3877 c->tcp_do_toggle_rw = 0;
3878 c->tcp_check_nb_connect = 0;
3879 #ifdef USE_MSG_FASTOPEN
3880 c->tcp_do_fastopen = 0;
3881 #endif
3882 #ifdef USE_DNSCRYPT
3883 c->dnscrypt = 0;
3884 c->dnscrypt_buffer = c->buffer;
3885 #endif
3886 c->callback = callback;
3887 c->cb_arg = callback_arg;
3888 /* ub_event stuff */
3889 evbits = UB_EV_PERSIST | UB_EV_READ;
3890 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3891 comm_point_local_handle_callback, c);
3892 if(c->ev->ev == NULL) {
3893 log_err("could not baseset localhdl event");
3894 free(c->ev);
3895 free(c);
3896 return NULL;
3897 }
3898 if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3899 log_err("could not add localhdl event");
3900 ub_event_free(c->ev->ev);
3901 free(c->ev);
3902 free(c);
3903 return NULL;
3904 }
3905 c->event_added = 1;
3906 return c;
3907 }
3908
3909 struct comm_point*
3910 comm_point_create_raw(struct comm_base* base, int fd, int writing,
3911 comm_point_callback_type* callback, void* callback_arg)
3912 {
3913 struct comm_point* c = (struct comm_point*)calloc(1,
3914 sizeof(struct comm_point));
3915 short evbits;
3916 if(!c)
3917 return NULL;
3918 c->ev = (struct internal_event*)calloc(1,
3919 sizeof(struct internal_event));
3920 if(!c->ev) {
3921 free(c);
3922 return NULL;
3923 }
3924 c->ev->base = base;
3925 c->fd = fd;
3926 c->buffer = NULL;
3927 c->timeout = NULL;
3928 c->tcp_is_reading = 0;
3929 c->tcp_byte_count = 0;
3930 c->tcp_parent = NULL;
3931 c->max_tcp_count = 0;
3932 c->cur_tcp_count = 0;
3933 c->tcp_handlers = NULL;
3934 c->tcp_free = NULL;
3935 c->type = comm_raw;
3936 c->tcp_do_close = 0;
3937 c->do_not_close = 1;
3938 c->tcp_do_toggle_rw = 0;
3939 c->tcp_check_nb_connect = 0;
3940 #ifdef USE_MSG_FASTOPEN
3941 c->tcp_do_fastopen = 0;
3942 #endif
3943 #ifdef USE_DNSCRYPT
3944 c->dnscrypt = 0;
3945 c->dnscrypt_buffer = c->buffer;
3946 #endif
3947 c->callback = callback;
3948 c->cb_arg = callback_arg;
3949 /* ub_event stuff */
3950 if(writing)
3951 evbits = UB_EV_PERSIST | UB_EV_WRITE;
3952 else evbits = UB_EV_PERSIST | UB_EV_READ;
3953 c->ev->ev = ub_event_new(base->eb->base, c->fd, evbits,
3954 comm_point_raw_handle_callback, c);
3955 if(c->ev->ev == NULL) {
3956 log_err("could not baseset rawhdl event");
3957 free(c->ev);
3958 free(c);
3959 return NULL;
3960 }
3961 if (ub_event_add(c->ev->ev, c->timeout) != 0) {
3962 log_err("could not add rawhdl event");
3963 ub_event_free(c->ev->ev);
3964 free(c->ev);
3965 free(c);
3966 return NULL;
3967 }
3968 c->event_added = 1;
3969 return c;
3970 }
3971
3972 void
3973 comm_point_close(struct comm_point* c)
3974 {
3975 if(!c)
3976 return;
3977 if(c->fd != -1) {
3978 verbose(5, "comm_point_close of %d: event_del", c->fd);
3979 if(c->event_added) {
3980 if(ub_event_del(c->ev->ev) != 0) {
3981 log_err("could not event_del on close");
3982 }
3983 c->event_added = 0;
3984 }
3985 }
3986 tcl_close_connection(c->tcl_addr);
3987 if(c->tcp_req_info)
3988 tcp_req_info_clear(c->tcp_req_info);
3989 if(c->h2_session)
3990 http2_session_server_delete(c->h2_session);
3991
3992 /* close fd after removing from event lists, or epoll.. is messed up */
3993 if(c->fd != -1 && !c->do_not_close) {
3994 #ifdef USE_WINSOCK
3995 if(c->type == comm_tcp || c->type == comm_http) {
3996 /* delete sticky events for the fd, it gets closed */
3997 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_READ);
3998 ub_winsock_tcp_wouldblock(c->ev->ev, UB_EV_WRITE);
3999 }
4000 #endif
4001 verbose(VERB_ALGO, "close fd %d", c->fd);
4002 sock_close(c->fd);
4003 }
4004 c->fd = -1;
4005 }
4006
4007 void
4008 comm_point_delete(struct comm_point* c)
4009 {
4010 if(!c)
4011 return;
4012 if((c->type == comm_tcp || c->type == comm_http) && c->ssl) {
4013 #ifdef HAVE_SSL
4014 SSL_shutdown(c->ssl);
4015 SSL_free(c->ssl);
4016 #endif
4017 }
4018 if(c->type == comm_http && c->http_endpoint) {
4019 free(c->http_endpoint);
4020 c->http_endpoint = NULL;
4021 }
4022 comm_point_close(c);
4023 if(c->tcp_handlers) {
4024 int i;
4025 for(i=0; i<c->max_tcp_count; i++)
4026 comm_point_delete(c->tcp_handlers[i]);
4027 free(c->tcp_handlers);
4028 }
4029 free(c->timeout);
4030 if(c->type == comm_tcp || c->type == comm_local || c->type == comm_http) {
4031 sldns_buffer_free(c->buffer);
4032 #ifdef USE_DNSCRYPT
4033 if(c->dnscrypt && c->dnscrypt_buffer != c->buffer) {
4034 sldns_buffer_free(c->dnscrypt_buffer);
4035 }
4036 #endif
4037 if(c->tcp_req_info) {
4038 tcp_req_info_delete(c->tcp_req_info);
4039 }
4040 if(c->h2_session) {
4041 http2_session_delete(c->h2_session);
4042 }
4043 }
4044 ub_event_free(c->ev->ev);
4045 free(c->ev);
4046 free(c);
4047 }
4048
4049 void
4050 comm_point_send_reply(struct comm_reply *repinfo)
4051 {
4052 struct sldns_buffer* buffer;
4053 log_assert(repinfo && repinfo->c);
4054 #ifdef USE_DNSCRYPT
4055 buffer = repinfo->c->dnscrypt_buffer;
4056 if(!dnsc_handle_uncurved_request(repinfo)) {
4057 return;
4058 }
4059 #else
4060 buffer = repinfo->c->buffer;
4061 #endif
4062 if(repinfo->c->type == comm_udp) {
4063 if(repinfo->srctype)
4064 comm_point_send_udp_msg_if(repinfo->c,
4065 buffer, (struct sockaddr*)&repinfo->addr,
4066 repinfo->addrlen, repinfo);
4067 else
4068 comm_point_send_udp_msg(repinfo->c, buffer,
4069 (struct sockaddr*)&repinfo->addr, repinfo->addrlen, 0);
4070 #ifdef USE_DNSTAP
4071 /*
4072 * sending src (client)/dst (local service) addresses over DNSTAP from udp callback
4073 */
4074 if(repinfo->c->dtenv != NULL && repinfo->c->dtenv->log_client_response_messages) {
4075 log_addr(VERB_ALGO, "from local addr", (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->socket->addr->ai_addrlen);
4076 log_addr(VERB_ALGO, "response to client", &repinfo->addr, repinfo->addrlen);
4077 dt_msg_send_client_response(repinfo->c->dtenv, &repinfo->addr, (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->type, repinfo->c->buffer);
4078 }
4079 #endif
4080 } else {
4081 #ifdef USE_DNSTAP
4082 /*
4083 * sending src (client)/dst (local service) addresses over DNSTAP from TCP callback
4084 */
4085 if(repinfo->c->tcp_parent->dtenv != NULL && repinfo->c->tcp_parent->dtenv->log_client_response_messages) {
4086 log_addr(VERB_ALGO, "from local addr", (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->socket->addr->ai_addrlen);
4087 log_addr(VERB_ALGO, "response to client", &repinfo->addr, repinfo->addrlen);
4088 dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv, &repinfo->addr, (void*)repinfo->c->socket->addr->ai_addr, repinfo->c->type,
4089 ( repinfo->c->tcp_req_info? repinfo->c->tcp_req_info->spool_buffer: repinfo->c->buffer ));
4090 }
4091 #endif
4092 if(repinfo->c->tcp_req_info) {
4093 tcp_req_info_send_reply(repinfo->c->tcp_req_info);
4094 } else if(repinfo->c->use_h2) {
4095 if(!http2_submit_dns_response(repinfo->c->h2_session)) {
4096 comm_point_drop_reply(repinfo);
4097 return;
4098 }
4099 repinfo->c->h2_stream = NULL;
4100 repinfo->c->tcp_is_reading = 0;
4101 comm_point_stop_listening(repinfo->c);
4102 comm_point_start_listening(repinfo->c, -1,
4103 adjusted_tcp_timeout(repinfo->c));
4104 return;
4105 } else {
4106 comm_point_start_listening(repinfo->c, -1,
4107 adjusted_tcp_timeout(repinfo->c));
4108 }
4109 }
4110 }
4111
4112 void
4113 comm_point_drop_reply(struct comm_reply* repinfo)
4114 {
4115 if(!repinfo)
4116 return;
4117 log_assert(repinfo->c);
4118 log_assert(repinfo->c->type != comm_tcp_accept);
4119 if(repinfo->c->type == comm_udp)
4120 return;
4121 if(repinfo->c->tcp_req_info)
4122 repinfo->c->tcp_req_info->is_drop = 1;
4123 if(repinfo->c->type == comm_http) {
4124 if(repinfo->c->h2_session) {
4125 repinfo->c->h2_session->is_drop = 1;
4126 if(!repinfo->c->h2_session->postpone_drop)
4127 reclaim_http_handler(repinfo->c);
4128 return;
4129 }
4130 reclaim_http_handler(repinfo->c);
4131 return;
4132 }
4133 reclaim_tcp_handler(repinfo->c);
4134 }
4135
4136 void
4137 comm_point_stop_listening(struct comm_point* c)
4138 {
4139 verbose(VERB_ALGO, "comm point stop listening %d", c->fd);
4140 if(c->event_added) {
4141 if(ub_event_del(c->ev->ev) != 0) {
4142 log_err("event_del error to stoplisten");
4143 }
4144 c->event_added = 0;
4145 }
4146 }
4147
4148 void
4149 comm_point_start_listening(struct comm_point* c, int newfd, int msec)
4150 {
4151 verbose(VERB_ALGO, "comm point start listening %d (%d msec)",
4152 c->fd==-1?newfd:c->fd, msec);
4153 if(c->type == comm_tcp_accept && !c->tcp_free) {
4154 /* no use to start listening no free slots. */
4155 return;
4156 }
4157 if(c->event_added) {
4158 if(ub_event_del(c->ev->ev) != 0) {
4159 log_err("event_del error to startlisten");
4160 }
4161 c->event_added = 0;
4162 }
4163 if(msec != -1 && msec != 0) {
4164 if(!c->timeout) {
4165 c->timeout = (struct timeval*)malloc(sizeof(
4166 struct timeval));
4167 if(!c->timeout) {
4168 log_err("cpsl: malloc failed. No net read.");
4169 return;
4170 }
4171 }
4172 ub_event_add_bits(c->ev->ev, UB_EV_TIMEOUT);
4173 #ifndef S_SPLINT_S /* splint fails on struct timeval. */
4174 c->timeout->tv_sec = msec/1000;
4175 c->timeout->tv_usec = (msec%1000)*1000;
4176 #endif /* S_SPLINT_S */
4177 } else {
4178 if(msec == 0 || !c->timeout) {
4179 ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
4180 }
4181 }
4182 if(c->type == comm_tcp || c->type == comm_http) {
4183 ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4184 if(c->tcp_write_and_read) {
4185 verbose(5, "startlistening %d mode rw", (newfd==-1?c->fd:newfd));
4186 ub_event_add_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4187 } else if(c->tcp_is_reading) {
4188 verbose(5, "startlistening %d mode r", (newfd==-1?c->fd:newfd));
4189 ub_event_add_bits(c->ev->ev, UB_EV_READ);
4190 } else {
4191 verbose(5, "startlistening %d mode w", (newfd==-1?c->fd:newfd));
4192 ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
4193 }
4194 }
4195 if(newfd != -1) {
4196 if(c->fd != -1 && c->fd != newfd) {
4197 verbose(5, "cpsl close of fd %d for %d", c->fd, newfd);
4198 sock_close(c->fd);
4199 }
4200 c->fd = newfd;
4201 ub_event_set_fd(c->ev->ev, c->fd);
4202 }
4203 if(ub_event_add(c->ev->ev, msec==0?NULL:c->timeout) != 0) {
4204 log_err("event_add failed. in cpsl.");
4205 return;
4206 }
4207 c->event_added = 1;
4208 }
4209
4210 void comm_point_listen_for_rw(struct comm_point* c, int rd, int wr)
4211 {
4212 verbose(VERB_ALGO, "comm point listen_for_rw %d %d", c->fd, wr);
4213 if(c->event_added) {
4214 if(ub_event_del(c->ev->ev) != 0) {
4215 log_err("event_del error to cplf");
4216 }
4217 c->event_added = 0;
4218 }
4219 if(!c->timeout) {
4220 ub_event_del_bits(c->ev->ev, UB_EV_TIMEOUT);
4221 }
4222 ub_event_del_bits(c->ev->ev, UB_EV_READ|UB_EV_WRITE);
4223 if(rd) ub_event_add_bits(c->ev->ev, UB_EV_READ);
4224 if(wr) ub_event_add_bits(c->ev->ev, UB_EV_WRITE);
4225 if(ub_event_add(c->ev->ev, c->timeout) != 0) {
4226 log_err("event_add failed. in cplf.");
4227 return;
4228 }
4229 c->event_added = 1;
4230 }
4231
4232 size_t comm_point_get_mem(struct comm_point* c)
4233 {
4234 size_t s;
4235 if(!c)
4236 return 0;
4237 s = sizeof(*c) + sizeof(*c->ev);
4238 if(c->timeout)
4239 s += sizeof(*c->timeout);
4240 if(c->type == comm_tcp || c->type == comm_local) {
4241 s += sizeof(*c->buffer) + sldns_buffer_capacity(c->buffer);
4242 #ifdef USE_DNSCRYPT
4243 s += sizeof(*c->dnscrypt_buffer);
4244 if(c->buffer != c->dnscrypt_buffer) {
4245 s += sldns_buffer_capacity(c->dnscrypt_buffer);
4246 }
4247 #endif
4248 }
4249 if(c->type == comm_tcp_accept) {
4250 int i;
4251 for(i=0; i<c->max_tcp_count; i++)
4252 s += comm_point_get_mem(c->tcp_handlers[i]);
4253 }
4254 return s;
4255 }
4256
4257 struct comm_timer*
4258 comm_timer_create(struct comm_base* base, void (*cb)(void*), void* cb_arg)
4259 {
4260 struct internal_timer *tm = (struct internal_timer*)calloc(1,
4261 sizeof(struct internal_timer));
4262 if(!tm) {
4263 log_err("malloc failed");
4264 return NULL;
4265 }
4266 tm->super.ev_timer = tm;
4267 tm->base = base;
4268 tm->super.callback = cb;
4269 tm->super.cb_arg = cb_arg;
4270 tm->ev = ub_event_new(base->eb->base, -1, UB_EV_TIMEOUT,
4271 comm_timer_callback, &tm->super);
4272 if(tm->ev == NULL) {
4273 log_err("timer_create: event_base_set failed.");
4274 free(tm);
4275 return NULL;
4276 }
4277 return &tm->super;
4278 }
4279
4280 void
4281 comm_timer_disable(struct comm_timer* timer)
4282 {
4283 if(!timer)
4284 return;
4285 ub_timer_del(timer->ev_timer->ev);
4286 timer->ev_timer->enabled = 0;
4287 }
4288
4289 void
4290 comm_timer_set(struct comm_timer* timer, struct timeval* tv)
4291 {
4292 log_assert(tv);
4293 if(timer->ev_timer->enabled)
4294 comm_timer_disable(timer);
4295 if(ub_timer_add(timer->ev_timer->ev, timer->ev_timer->base->eb->base,
4296 comm_timer_callback, timer, tv) != 0)
4297 log_err("comm_timer_set: evtimer_add failed.");
4298 timer->ev_timer->enabled = 1;
4299 }
4300
4301 void
4302 comm_timer_delete(struct comm_timer* timer)
4303 {
4304 if(!timer)
4305 return;
4306 comm_timer_disable(timer);
4307 /* Free the sub struct timer->ev_timer derived from the super struct timer.
4308 * i.e. assert(timer == timer->ev_timer)
4309 */
4310 ub_event_free(timer->ev_timer->ev);
4311 free(timer->ev_timer);
4312 }
4313
4314 void
4315 comm_timer_callback(int ATTR_UNUSED(fd), short event, void* arg)
4316 {
4317 struct comm_timer* tm = (struct comm_timer*)arg;
4318 if(!(event&UB_EV_TIMEOUT))
4319 return;
4320 ub_comm_base_now(tm->ev_timer->base);
4321 tm->ev_timer->enabled = 0;
4322 fptr_ok(fptr_whitelist_comm_timer(tm->callback));
4323 (*tm->callback)(tm->cb_arg);
4324 }
4325
4326 int
4327 comm_timer_is_set(struct comm_timer* timer)
4328 {
4329 return (int)timer->ev_timer->enabled;
4330 }
4331
4332 size_t
4333 comm_timer_get_mem(struct comm_timer* ATTR_UNUSED(timer))
4334 {
4335 return sizeof(struct internal_timer);
4336 }
4337
4338 struct comm_signal*
4339 comm_signal_create(struct comm_base* base,
4340 void (*callback)(int, void*), void* cb_arg)
4341 {
4342 struct comm_signal* com = (struct comm_signal*)malloc(
4343 sizeof(struct comm_signal));
4344 if(!com) {
4345 log_err("malloc failed");
4346 return NULL;
4347 }
4348 com->base = base;
4349 com->callback = callback;
4350 com->cb_arg = cb_arg;
4351 com->ev_signal = NULL;
4352 return com;
4353 }
4354
4355 void
4356 comm_signal_callback(int sig, short event, void* arg)
4357 {
4358 struct comm_signal* comsig = (struct comm_signal*)arg;
4359 if(!(event & UB_EV_SIGNAL))
4360 return;
4361 ub_comm_base_now(comsig->base);
4362 fptr_ok(fptr_whitelist_comm_signal(comsig->callback));
4363 (*comsig->callback)(sig, comsig->cb_arg);
4364 }
4365
4366 int
4367 comm_signal_bind(struct comm_signal* comsig, int sig)
4368 {
4369 struct internal_signal* entry = (struct internal_signal*)calloc(1,
4370 sizeof(struct internal_signal));
4371 if(!entry) {
4372 log_err("malloc failed");
4373 return 0;
4374 }
4375 log_assert(comsig);
4376 /* add signal event */
4377 entry->ev = ub_signal_new(comsig->base->eb->base, sig,
4378 comm_signal_callback, comsig);
4379 if(entry->ev == NULL) {
4380 log_err("Could not create signal event");
4381 free(entry);
4382 return 0;
4383 }
4384 if(ub_signal_add(entry->ev, NULL) != 0) {
4385 log_err("Could not add signal handler");
4386 ub_event_free(entry->ev);
4387 free(entry);
4388 return 0;
4389 }
4390 /* link into list */
4391 entry->next = comsig->ev_signal;
4392 comsig->ev_signal = entry;
4393 return 1;
4394 }
4395
4396 void
4397 comm_signal_delete(struct comm_signal* comsig)
4398 {
4399 struct internal_signal* p, *np;
4400 if(!comsig)
4401 return;
4402 p=comsig->ev_signal;
4403 while(p) {
4404 np = p->next;
4405 ub_signal_del(p->ev);
4406 ub_event_free(p->ev);
4407 free(p);
4408 p = np;
4409 }
4410 free(comsig);
4411 }
4412