2 * services/listen_dnsport.c - listen on port 53 for incoming DNS queries.
4 * Copyright (c) 2007, NLnet Labs. All rights reserved.
6 * This software is open source.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * Redistributions of source code must retain the above copyright notice,
13 * this list of conditions and the following disclaimer.
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.
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.
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.
39 * This file has functions to get queries from clients.
42 #ifdef HAVE_SYS_TYPES_H
43 # include <sys/types.h>
46 #ifdef USE_TCP_FASTOPEN
47 #include <netinet/tcp.h>
49 #include "services/listen_dnsport.h"
50 #include "services/outside_network.h"
51 #include "util/netevent.h"
53 #include "util/config_file.h"
54 #include "util/net_help.h"
55 #include "sldns/sbuffer.h"
56 #include "services/mesh.h"
57 #include "util/fptr_wlist.h"
58 #include "util/locks.h"
70 #include <systemd/sd-daemon.h>
73 /** number of queued TCP connections for listen() */
74 #define TCP_BACKLOG 256
76 /** number of simultaneous requests a client can have */
77 #define TCP_MAX_REQ_SIMULTANEOUS 32
79 #ifndef THREADS_DISABLED
80 /** lock on the counter of stream buffer memory */
81 static lock_basic_type stream_wait_count_lock;
83 /** size (in bytes) of stream wait buffers */
84 static size_t stream_wait_count = 0;
85 /** is the lock initialised for stream wait buffers */
86 static int stream_wait_lock_inited = 0;
89 * Debug print of the getaddrinfo returned address.
90 * @param addr: the address returned.
93 verbose_print_addr(struct addrinfo *addr)
95 if(verbosity >= VERB_ALGO) {
97 void* sinaddr = &((struct sockaddr_in*)addr->ai_addr)->sin_addr;
99 if(addr->ai_family == AF_INET6)
100 sinaddr = &((struct sockaddr_in6*)addr->ai_addr)->
103 if(inet_ntop(addr->ai_family, sinaddr, buf,
104 (socklen_t)sizeof(buf)) == 0) {
105 (void)strlcpy(buf, "(null)", sizeof(buf));
107 buf[sizeof(buf)-1] = 0;
108 verbose(VERB_ALGO, "creating %s%s socket %s %d",
109 addr->ai_socktype==SOCK_DGRAM?"udp":
110 addr->ai_socktype==SOCK_STREAM?"tcp":"otherproto",
111 addr->ai_family==AF_INET?"4":
112 addr->ai_family==AF_INET6?"6":
114 ntohs(((struct sockaddr_in*)addr->ai_addr)->sin_port));
120 systemd_get_activated(int family, int socktype, int listen,
121 struct sockaddr *addr, socklen_t addrlen,
127 const char* listen_pid, *listen_fds;
129 /* We should use "listen" option only for stream protocols. For UDP it should be -1 */
131 if((r = sd_booted()) < 1) {
133 log_warn("systemd is not running");
135 log_err("systemd sd_booted(): %s", strerror(-r));
139 listen_pid = getenv("LISTEN_PID");
140 listen_fds = getenv("LISTEN_FDS");
143 log_warn("Systemd mandatory ENV variable is not defined: LISTEN_PID");
148 log_warn("Systemd mandatory ENV variable is not defined: LISTEN_FDS");
152 if((r = sd_listen_fds(0)) < 1) {
154 log_warn("systemd: did not return socket, check unit configuration");
156 log_err("systemd sd_listen_fds(): %s", strerror(-r));
160 for(i = 0; i < r; i++) {
161 if(sd_is_socket(SD_LISTEN_FDS_START + i, family, socktype, listen)) {
162 s = SD_LISTEN_FDS_START + i;
168 log_err_addr("systemd sd_listen_fds()",
170 (struct sockaddr_storage *)addr, addrlen);
172 log_err("systemd sd_listen_fds(): %s", path);
179 create_udp_sock(int family, int socktype, struct sockaddr* addr,
180 socklen_t addrlen, int v6only, int* inuse, int* noproto,
181 int rcv, int snd, int listen, int* reuseport, int transparent,
182 int freebind, int use_systemd)
185 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_USE_MIN_MTU) || defined(IP_TRANSPARENT) || defined(IP_BINDANY) || defined(IP_FREEBIND) || defined (SO_BINDANY)
189 int mtu = IPV6_MIN_MTU;
191 #if !defined(SO_RCVBUFFORCE) && !defined(SO_RCVBUF)
194 #if !defined(SO_SNDBUFFORCE) && !defined(SO_SNDBUF)
200 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
203 #if !defined(IP_FREEBIND)
207 int got_fd_from_systemd = 0;
211 && (s = systemd_get_activated(family, socktype, -1, addr,
212 addrlen, NULL)) == -1)) {
216 if((s = socket(family, socktype, 0)) == -1) {
219 if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
223 log_err("can't create socket: %s", strerror(errno));
225 if(WSAGetLastError() == WSAEAFNOSUPPORT ||
226 WSAGetLastError() == WSAEPROTONOSUPPORT) {
230 log_err("can't create socket: %s",
231 wsa_strerror(WSAGetLastError()));
238 got_fd_from_systemd = 1;
243 if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
244 (socklen_t)sizeof(on)) < 0) {
246 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
248 if(errno != ENOSYS) {
255 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
256 wsa_strerror(WSAGetLastError()));
263 #endif /* SO_REUSEADDR */
265 # ifdef SO_REUSEPORT_LB
266 /* on FreeBSD 12 we have SO_REUSEPORT_LB that does loadbalance
267 * like SO_REUSEPORT on Linux. This is what the users want
268 * with the config option in unbound.conf; if we actually
269 * need local address and port reuse they'll also need to
270 * have SO_REUSEPORT set for them, assume it was _LB they want.
272 if (reuseport && *reuseport &&
273 setsockopt(s, SOL_SOCKET, SO_REUSEPORT_LB, (void*)&on,
274 (socklen_t)sizeof(on)) < 0) {
276 if(errno != ENOPROTOOPT || verbosity >= 3)
277 log_warn("setsockopt(.. SO_REUSEPORT_LB ..) failed: %s",
280 /* this option is not essential, we can continue */
283 # else /* no SO_REUSEPORT_LB */
285 /* try to set SO_REUSEPORT so that incoming
286 * queries are distributed evenly among the receiving threads.
287 * Each thread must have its own socket bound to the same port,
288 * with SO_REUSEPORT set on each socket.
290 if (reuseport && *reuseport &&
291 setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
292 (socklen_t)sizeof(on)) < 0) {
294 if(errno != ENOPROTOOPT || verbosity >= 3)
295 log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
298 /* this option is not essential, we can continue */
301 # endif /* SO_REUSEPORT_LB */
304 #endif /* defined(SO_REUSEPORT) */
305 #ifdef IP_TRANSPARENT
307 setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
308 (socklen_t)sizeof(on)) < 0) {
309 log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
312 #elif defined(IP_BINDANY)
314 setsockopt(s, (family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
315 (family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
316 (void*)&on, (socklen_t)sizeof(on)) < 0) {
317 log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
318 (family==AF_INET6?"V6":""), strerror(errno));
320 #elif defined(SO_BINDANY)
322 setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on,
323 (socklen_t)sizeof(on)) < 0) {
324 log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
327 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
331 setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
332 (socklen_t)sizeof(on)) < 0) {
333 log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
336 #endif /* IP_FREEBIND */
340 socklen_t slen = (socklen_t)sizeof(got);
341 # ifdef SO_RCVBUFFORCE
342 /* Linux specific: try to use root permission to override
343 * system limits on rcvbuf. The limit is stored in
344 * /proc/sys/net/core/rmem_max or sysctl net.core.rmem_max */
345 if(setsockopt(s, SOL_SOCKET, SO_RCVBUFFORCE, (void*)&rcv,
346 (socklen_t)sizeof(rcv)) < 0) {
349 log_err("setsockopt(..., SO_RCVBUFFORCE, "
350 "...) failed: %s", strerror(errno));
353 log_err("setsockopt(..., SO_RCVBUFFORCE, "
355 wsa_strerror(WSAGetLastError()));
362 # endif /* SO_RCVBUFFORCE */
363 if(setsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&rcv,
364 (socklen_t)sizeof(rcv)) < 0) {
366 log_err("setsockopt(..., SO_RCVBUF, "
367 "...) failed: %s", strerror(errno));
370 log_err("setsockopt(..., SO_RCVBUF, "
372 wsa_strerror(WSAGetLastError()));
379 /* check if we got the right thing or if system
380 * reduced to some system max. Warn if so */
381 if(getsockopt(s, SOL_SOCKET, SO_RCVBUF, (void*)&got,
382 &slen) >= 0 && got < rcv/2) {
383 log_warn("so-rcvbuf %u was not granted. "
384 "Got %u. To fix: start with "
385 "root permissions(linux) or sysctl "
386 "bigger net.core.rmem_max(linux) or "
387 "kern.ipc.maxsockbuf(bsd) values.",
388 (unsigned)rcv, (unsigned)got);
390 # ifdef SO_RCVBUFFORCE
393 #endif /* SO_RCVBUF */
395 /* first do RCVBUF as the receive buffer is more important */
399 socklen_t slen = (socklen_t)sizeof(got);
400 # ifdef SO_SNDBUFFORCE
401 /* Linux specific: try to use root permission to override
402 * system limits on sndbuf. The limit is stored in
403 * /proc/sys/net/core/wmem_max or sysctl net.core.wmem_max */
404 if(setsockopt(s, SOL_SOCKET, SO_SNDBUFFORCE, (void*)&snd,
405 (socklen_t)sizeof(snd)) < 0) {
408 log_err("setsockopt(..., SO_SNDBUFFORCE, "
409 "...) failed: %s", strerror(errno));
412 log_err("setsockopt(..., SO_SNDBUFFORCE, "
414 wsa_strerror(WSAGetLastError()));
421 # endif /* SO_SNDBUFFORCE */
422 if(setsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&snd,
423 (socklen_t)sizeof(snd)) < 0) {
425 log_err("setsockopt(..., SO_SNDBUF, "
426 "...) failed: %s", strerror(errno));
429 log_err("setsockopt(..., SO_SNDBUF, "
431 wsa_strerror(WSAGetLastError()));
438 /* check if we got the right thing or if system
439 * reduced to some system max. Warn if so */
440 if(getsockopt(s, SOL_SOCKET, SO_SNDBUF, (void*)&got,
441 &slen) >= 0 && got < snd/2) {
442 log_warn("so-sndbuf %u was not granted. "
443 "Got %u. To fix: start with "
444 "root permissions(linux) or sysctl "
445 "bigger net.core.wmem_max(linux) or "
446 "kern.ipc.maxsockbuf(bsd) values.",
447 (unsigned)snd, (unsigned)got);
449 # ifdef SO_SNDBUFFORCE
452 #endif /* SO_SNDBUF */
454 if(family == AF_INET6) {
455 # if defined(IPV6_V6ONLY)
457 int val=(v6only==2)?0:1;
458 if (setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
459 (void*)&val, (socklen_t)sizeof(val)) < 0) {
461 log_err("setsockopt(..., IPV6_V6ONLY"
462 ", ...) failed: %s", strerror(errno));
465 log_err("setsockopt(..., IPV6_V6ONLY"
467 wsa_strerror(WSAGetLastError()));
476 # if defined(IPV6_USE_MIN_MTU)
478 * There is no fragmentation of IPv6 datagrams
479 * during forwarding in the network. Therefore
480 * we do not send UDP datagrams larger than
481 * the minimum IPv6 MTU of 1280 octets. The
482 * EDNS0 message length can be larger if the
483 * network stack supports IPV6_USE_MIN_MTU.
485 if (setsockopt(s, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
486 (void*)&on, (socklen_t)sizeof(on)) < 0) {
488 log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
489 "...) failed: %s", strerror(errno));
492 log_err("setsockopt(..., IPV6_USE_MIN_MTU, "
494 wsa_strerror(WSAGetLastError()));
501 # elif defined(IPV6_MTU)
503 * On Linux, to send no larger than 1280, the PMTUD is
504 * disabled by default for datagrams anyway, so we set
507 if (setsockopt(s, IPPROTO_IPV6, IPV6_MTU,
508 (void*)&mtu, (socklen_t)sizeof(mtu)) < 0) {
510 log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
514 log_err("setsockopt(..., IPV6_MTU, ...) failed: %s",
515 wsa_strerror(WSAGetLastError()));
522 # endif /* IPv6 MTU */
523 } else if(family == AF_INET) {
524 # if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
525 /* linux 3.15 has IP_PMTUDISC_OMIT, Hannes Frederic Sowa made it so that
526 * PMTU information is not accepted, but fragmentation is allowed
527 * if and only if the packet size exceeds the outgoing interface MTU
528 * (and also uses the interface mtu to determine the size of the packets).
529 * So there won't be any EMSGSIZE error. Against DNS fragmentation attacks.
530 * FreeBSD already has same semantics without setting the option. */
533 # if defined(IP_PMTUDISC_OMIT)
534 action = IP_PMTUDISC_OMIT;
535 if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
536 &action, (socklen_t)sizeof(action)) < 0) {
538 if (errno != EINVAL) {
539 log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_OMIT...) failed: %s",
558 action = IP_PMTUDISC_DONT;
559 if (setsockopt(s, IPPROTO_IP, IP_MTU_DISCOVER,
560 &action, (socklen_t)sizeof(action)) < 0) {
561 log_err("setsockopt(..., IP_MTU_DISCOVER, IP_PMTUDISC_DONT...) failed: %s",
573 # elif defined(IP_DONTFRAG)
575 if (setsockopt(s, IPPROTO_IP, IP_DONTFRAG,
576 &off, (socklen_t)sizeof(off)) < 0) {
577 log_err("setsockopt(..., IP_DONTFRAG, ...) failed: %s",
588 # endif /* IPv4 MTU */
592 !got_fd_from_systemd &&
594 bind(s, (struct sockaddr*)addr, addrlen) != 0) {
599 *inuse = (errno == EADDRINUSE);
600 /* detect freebsd jail with no ipv6 permission */
601 if(family==AF_INET6 && errno==EINVAL)
603 else if(errno != EADDRINUSE &&
604 !(errno == EACCES && verbosity < 4 && !listen)
606 && !(errno == EADDRNOTAVAIL && verbosity < 4 && !listen)
609 log_err_addr("can't bind socket", strerror(errno),
610 (struct sockaddr_storage*)addr, addrlen);
612 #endif /* EADDRINUSE */
614 #else /* USE_WINSOCK */
615 if(WSAGetLastError() != WSAEADDRINUSE &&
616 WSAGetLastError() != WSAEADDRNOTAVAIL &&
617 !(WSAGetLastError() == WSAEACCES && verbosity < 4 && !listen)) {
618 log_err_addr("can't bind socket",
619 wsa_strerror(WSAGetLastError()),
620 (struct sockaddr_storage*)addr, addrlen);
623 #endif /* USE_WINSOCK */
626 if(!fd_set_nonblock(s)) {
640 create_tcp_accept_sock(struct addrinfo *addr, int v6only, int* noproto,
641 int* reuseport, int transparent, int mss, int freebind, int use_systemd)
644 #if defined(SO_REUSEADDR) || defined(SO_REUSEPORT) || defined(IPV6_V6ONLY) || defined(IP_TRANSPARENT) || defined(IP_BINDANY) || defined(IP_FREEBIND) || defined(SO_BINDANY)
648 int got_fd_from_systemd = 0;
650 #ifdef USE_TCP_FASTOPEN
653 #if !defined(IP_TRANSPARENT) && !defined(IP_BINDANY) && !defined(SO_BINDANY)
656 #if !defined(IP_FREEBIND)
659 verbose_print_addr(addr);
664 && (s = systemd_get_activated(addr->ai_family, addr->ai_socktype, 1,
665 addr->ai_addr, addr->ai_addrlen,
670 if((s = socket(addr->ai_family, addr->ai_socktype, 0)) == -1) {
672 if(errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
676 log_err("can't create socket: %s", strerror(errno));
678 if(WSAGetLastError() == WSAEAFNOSUPPORT ||
679 WSAGetLastError() == WSAEPROTONOSUPPORT) {
683 log_err("can't create socket: %s",
684 wsa_strerror(WSAGetLastError()));
689 #if defined(IPPROTO_TCP) && defined(TCP_MAXSEG)
690 if(setsockopt(s, IPPROTO_TCP, TCP_MAXSEG, (void*)&mss,
691 (socklen_t)sizeof(mss)) < 0) {
693 log_err(" setsockopt(.. TCP_MAXSEG ..) failed: %s",
696 log_err(" setsockopt(.. TCP_MAXSEG ..) failed: %s",
697 wsa_strerror(WSAGetLastError()));
701 " tcp socket mss set to %d", mss);
704 log_warn(" setsockopt(TCP_MAXSEG) unsupported");
705 #endif /* defined(IPPROTO_TCP) && defined(TCP_MAXSEG) */
709 got_fd_from_systemd = 1;
713 if(setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (void*)&on,
714 (socklen_t)sizeof(on)) < 0) {
716 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
720 log_err("setsockopt(.. SO_REUSEADDR ..) failed: %s",
721 wsa_strerror(WSAGetLastError()));
726 #endif /* SO_REUSEADDR */
728 if (freebind && setsockopt(s, IPPROTO_IP, IP_FREEBIND, (void*)&on,
729 (socklen_t)sizeof(on)) < 0) {
730 log_warn("setsockopt(.. IP_FREEBIND ..) failed: %s",
733 #endif /* IP_FREEBIND */
735 /* try to set SO_REUSEPORT so that incoming
736 * connections are distributed evenly among the receiving threads.
737 * Each thread must have its own socket bound to the same port,
738 * with SO_REUSEPORT set on each socket.
740 if (reuseport && *reuseport &&
741 setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void*)&on,
742 (socklen_t)sizeof(on)) < 0) {
744 if(errno != ENOPROTOOPT || verbosity >= 3)
745 log_warn("setsockopt(.. SO_REUSEPORT ..) failed: %s",
748 /* this option is not essential, we can continue */
753 #endif /* defined(SO_REUSEPORT) */
754 #if defined(IPV6_V6ONLY)
755 if(addr->ai_family == AF_INET6 && v6only) {
756 if(setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY,
757 (void*)&on, (socklen_t)sizeof(on)) < 0) {
759 log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
763 log_err("setsockopt(..., IPV6_V6ONLY, ...) failed: %s",
764 wsa_strerror(WSAGetLastError()));
772 #endif /* IPV6_V6ONLY */
773 #ifdef IP_TRANSPARENT
775 setsockopt(s, IPPROTO_IP, IP_TRANSPARENT, (void*)&on,
776 (socklen_t)sizeof(on)) < 0) {
777 log_warn("setsockopt(.. IP_TRANSPARENT ..) failed: %s",
780 #elif defined(IP_BINDANY)
782 setsockopt(s, (addr->ai_family==AF_INET6? IPPROTO_IPV6:IPPROTO_IP),
783 (addr->ai_family == AF_INET6? IPV6_BINDANY:IP_BINDANY),
784 (void*)&on, (socklen_t)sizeof(on)) < 0) {
785 log_warn("setsockopt(.. IP%s_BINDANY ..) failed: %s",
786 (addr->ai_family==AF_INET6?"V6":""), strerror(errno));
788 #elif defined(SO_BINDANY)
790 setsockopt(s, SOL_SOCKET, SO_BINDANY, (void*)&on, (socklen_t)
792 log_warn("setsockopt(.. SO_BINDANY ..) failed: %s",
795 #endif /* IP_TRANSPARENT || IP_BINDANY || SO_BINDANY */
798 !got_fd_from_systemd &&
800 bind(s, addr->ai_addr, addr->ai_addrlen) != 0) {
802 /* detect freebsd jail with no ipv6 permission */
803 if(addr->ai_family==AF_INET6 && errno==EINVAL)
806 log_err_addr("can't bind socket", strerror(errno),
807 (struct sockaddr_storage*)addr->ai_addr,
812 log_err_addr("can't bind socket",
813 wsa_strerror(WSAGetLastError()),
814 (struct sockaddr_storage*)addr->ai_addr,
820 if(!fd_set_nonblock(s)) {
828 if(listen(s, TCP_BACKLOG) == -1) {
830 log_err("can't listen: %s", strerror(errno));
833 log_err("can't listen: %s", wsa_strerror(WSAGetLastError()));
838 #ifdef USE_TCP_FASTOPEN
839 /* qlen specifies how many outstanding TFO requests to allow. Limit is a defense
840 against IP spoofing attacks as suggested in RFC7413 */
842 /* OS X implementation only supports qlen of 1 via this call. Actual
843 value is configured by the net.inet.tcp.fastopen_backlog kernel parm. */
846 /* 5 is recommended on linux */
849 if ((setsockopt(s, IPPROTO_TCP, TCP_FASTOPEN, &qlen,
850 sizeof(qlen))) == -1 ) {
852 /* squelch ENOPROTOOPT: freebsd server mode with kernel support
853 disabled, except when verbosity enabled for debugging */
854 if(errno != ENOPROTOOPT || verbosity >= 3)
857 log_warn("Setting TCP Fast Open as server failed: %s ; this could likely be because sysctl net.inet.tcp.fastopen.enabled, net.inet.tcp.fastopen.server_enable, or net.ipv4.tcp_fastopen is disabled", strerror(errno));
859 log_err("Setting TCP Fast Open as server failed: %s", strerror(errno));
867 create_local_accept_sock(const char *path, int* noproto, int use_systemd)
872 if (use_systemd && (ret = systemd_get_activated(AF_LOCAL, SOCK_STREAM, 1, NULL, 0, path)) != -1)
878 struct sockaddr_un usock;
883 verbose(VERB_ALGO, "creating unix socket %s", path);
884 #ifdef HAVE_STRUCT_SOCKADDR_UN_SUN_LEN
885 /* this member exists on BSDs, not Linux */
886 usock.sun_len = (unsigned)sizeof(usock);
888 usock.sun_family = AF_LOCAL;
889 /* length is 92-108, 104 on FreeBSD */
890 (void)strlcpy(usock.sun_path, path, sizeof(usock.sun_path));
892 if ((s = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) {
893 log_err("Cannot create local socket %s (%s)",
894 path, strerror(errno));
898 if (unlink(path) && errno != ENOENT) {
899 /* The socket already exists and cannot be removed */
900 log_err("Cannot remove old local socket %s (%s)",
901 path, strerror(errno));
905 if (bind(s, (struct sockaddr *)&usock,
906 (socklen_t)sizeof(struct sockaddr_un)) == -1) {
907 log_err("Cannot bind local socket %s (%s)",
908 path, strerror(errno));
912 if (!fd_set_nonblock(s)) {
913 log_err("Cannot set non-blocking mode");
917 if (listen(s, TCP_BACKLOG) == -1) {
918 log_err("can't listen: %s", strerror(errno));
922 (void)noproto; /*unused*/
939 log_err("Local sockets are not supported");
947 * Create socket from getaddrinfo results
950 make_sock(int stype, const char* ifname, const char* port,
951 struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
952 int* reuseport, int transparent, int tcp_mss, int freebind, int use_systemd)
954 struct addrinfo *res = NULL;
955 int r, s, inuse, noproto;
956 hints->ai_socktype = stype;
958 if((r=getaddrinfo(ifname, port, hints, &res)) != 0 || !res) {
960 if(r == EAI_NONAME && hints->ai_family == AF_INET6){
961 *noip6 = 1; /* 'Host not found' for IP6 on winXP */
965 log_err("node %s:%s getaddrinfo: %s %s",
966 ifname?ifname:"default", port, gai_strerror(r),
968 r==EAI_SYSTEM?(char*)strerror(errno):""
975 if(stype == SOCK_DGRAM) {
976 verbose_print_addr(res);
977 s = create_udp_sock(res->ai_family, res->ai_socktype,
978 (struct sockaddr*)res->ai_addr, res->ai_addrlen,
979 v6only, &inuse, &noproto, (int)rcv, (int)snd, 1,
980 reuseport, transparent, freebind, use_systemd);
981 if(s == -1 && inuse) {
982 log_err("bind: address already in use");
983 } else if(s == -1 && noproto && hints->ai_family == AF_INET6){
987 s = create_tcp_accept_sock(res, v6only, &noproto, reuseport,
988 transparent, tcp_mss, freebind, use_systemd);
989 if(s == -1 && noproto && hints->ai_family == AF_INET6){
997 /** make socket and first see if ifname contains port override info */
999 make_sock_port(int stype, const char* ifname, const char* port,
1000 struct addrinfo *hints, int v6only, int* noip6, size_t rcv, size_t snd,
1001 int* reuseport, int transparent, int tcp_mss, int freebind, int use_systemd)
1003 char* s = strchr(ifname, '@');
1005 /* override port with ifspec@port */
1008 if((size_t)(s-ifname) >= sizeof(newif)) {
1009 log_err("ifname too long: %s", ifname);
1013 if(strlen(s+1) >= sizeof(p)) {
1014 log_err("portnumber too long: %s", ifname);
1018 (void)strlcpy(newif, ifname, sizeof(newif));
1019 newif[s-ifname] = 0;
1020 (void)strlcpy(p, s+1, sizeof(p));
1022 return make_sock(stype, newif, p, hints, v6only, noip6,
1023 rcv, snd, reuseport, transparent, tcp_mss, freebind, use_systemd);
1025 return make_sock(stype, ifname, port, hints, v6only, noip6, rcv, snd,
1026 reuseport, transparent, tcp_mss, freebind, use_systemd);
1030 * Add port to open ports list.
1031 * @param list: list head. changed.
1033 * @param ftype: if fd is UDP.
1034 * @return false on failure. list in unchanged then.
1037 port_insert(struct listen_port** list, int s, enum listen_type ftype)
1039 struct listen_port* item = (struct listen_port*)malloc(
1040 sizeof(struct listen_port));
1045 item->ftype = ftype;
1050 /** set fd to receive source address packet info */
1052 set_recvpktinfo(int s, int family)
1054 #if defined(IPV6_RECVPKTINFO) || defined(IPV6_PKTINFO) || (defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)) || defined(IP_PKTINFO)
1059 if(family == AF_INET6) {
1060 # ifdef IPV6_RECVPKTINFO
1061 if(setsockopt(s, IPPROTO_IPV6, IPV6_RECVPKTINFO,
1062 (void*)&on, (socklen_t)sizeof(on)) < 0) {
1063 log_err("setsockopt(..., IPV6_RECVPKTINFO, ...) failed: %s",
1067 # elif defined(IPV6_PKTINFO)
1068 if(setsockopt(s, IPPROTO_IPV6, IPV6_PKTINFO,
1069 (void*)&on, (socklen_t)sizeof(on)) < 0) {
1070 log_err("setsockopt(..., IPV6_PKTINFO, ...) failed: %s",
1075 log_err("no IPV6_RECVPKTINFO and no IPV6_PKTINFO option, please "
1076 "disable interface-automatic or do-ip6 in config");
1078 # endif /* defined IPV6_RECVPKTINFO */
1080 } else if(family == AF_INET) {
1082 if(setsockopt(s, IPPROTO_IP, IP_PKTINFO,
1083 (void*)&on, (socklen_t)sizeof(on)) < 0) {
1084 log_err("setsockopt(..., IP_PKTINFO, ...) failed: %s",
1088 # elif defined(IP_RECVDSTADDR) && defined(IP_SENDSRCADDR)
1089 if(setsockopt(s, IPPROTO_IP, IP_RECVDSTADDR,
1090 (void*)&on, (socklen_t)sizeof(on)) < 0) {
1091 log_err("setsockopt(..., IP_RECVDSTADDR, ...) failed: %s",
1096 log_err("no IP_SENDSRCADDR or IP_PKTINFO option, please disable "
1097 "interface-automatic or do-ip4 in config");
1099 # endif /* IP_PKTINFO */
1105 /** see if interface is ssl, its port number == the ssl port number */
1107 if_is_ssl(const char* ifname, const char* port, int ssl_port,
1108 struct config_strlist* tls_additional_port)
1110 struct config_strlist* s;
1111 char* p = strchr(ifname, '@');
1112 if(!p && atoi(port) == ssl_port)
1114 if(p && atoi(p+1) == ssl_port)
1116 for(s = tls_additional_port; s; s = s->next) {
1117 if(p && atoi(p+1) == atoi(s->str))
1119 if(!p && atoi(port) == atoi(s->str))
1126 * Helper for ports_open. Creates one interface (or NULL for default).
1127 * @param ifname: The interface ip address.
1128 * @param do_auto: use automatic interface detection.
1129 * If enabled, then ifname must be the wildcard name.
1130 * @param do_udp: if udp should be used.
1131 * @param do_tcp: if udp should be used.
1132 * @param hints: for getaddrinfo. family and flags have to be set by caller.
1133 * @param port: Port number to use (as string).
1134 * @param list: list of open ports, appended to, changed to point to list head.
1135 * @param rcv: receive buffer size for UDP
1136 * @param snd: send buffer size for UDP
1137 * @param ssl_port: ssl service port number
1138 * @param tls_additional_port: list of additional ssl service port numbers.
1139 * @param reuseport: try to set SO_REUSEPORT if nonNULL and true.
1140 * set to false on exit if reuseport failed due to no kernel support.
1141 * @param transparent: set IP_TRANSPARENT socket option.
1142 * @param tcp_mss: maximum segment size of tcp socket. default if zero.
1143 * @param freebind: set IP_FREEBIND socket option.
1144 * @param use_systemd: if true, fetch sockets from systemd.
1145 * @param dnscrypt_port: dnscrypt service port number
1146 * @return: returns false on error.
1149 ports_create_if(const char* ifname, int do_auto, int do_udp, int do_tcp,
1150 struct addrinfo *hints, const char* port, struct listen_port** list,
1151 size_t rcv, size_t snd, int ssl_port,
1152 struct config_strlist* tls_additional_port, int* reuseport,
1153 int transparent, int tcp_mss, int freebind, int use_systemd,
1158 int is_dnscrypt = ((strchr(ifname, '@') &&
1159 atoi(strchr(ifname, '@')+1) == dnscrypt_port) ||
1160 (!strchr(ifname, '@') && atoi(port) == dnscrypt_port));
1162 int is_dnscrypt = 0;
1163 (void)dnscrypt_port;
1166 if(!do_udp && !do_tcp)
1169 if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1170 &noip6, rcv, snd, reuseport, transparent,
1171 tcp_mss, freebind, use_systemd)) == -1) {
1173 log_warn("IPv6 protocol not available");
1178 /* getting source addr packet info is highly non-portable */
1179 if(!set_recvpktinfo(s, hints->ai_family)) {
1187 if(!port_insert(list, s,
1188 is_dnscrypt?listen_type_udpancil_dnscrypt:listen_type_udpancil)) {
1197 /* regular udp socket */
1198 if((s = make_sock_port(SOCK_DGRAM, ifname, port, hints, 1,
1199 &noip6, rcv, snd, reuseport, transparent,
1200 tcp_mss, freebind, use_systemd)) == -1) {
1202 log_warn("IPv6 protocol not available");
1207 if(!port_insert(list, s,
1208 is_dnscrypt?listen_type_udp_dnscrypt:listen_type_udp)) {
1218 int is_ssl = if_is_ssl(ifname, port, ssl_port,
1219 tls_additional_port);
1220 if((s = make_sock_port(SOCK_STREAM, ifname, port, hints, 1,
1221 &noip6, 0, 0, reuseport, transparent, tcp_mss,
1222 freebind, use_systemd)) == -1) {
1224 /*log_warn("IPv6 protocol not available");*/
1230 verbose(VERB_ALGO, "setup TCP for SSL service");
1231 if(!port_insert(list, s, is_ssl?listen_type_ssl:
1232 (is_dnscrypt?listen_type_tcp_dnscrypt:listen_type_tcp))) {
1245 * Add items to commpoint list in front.
1246 * @param c: commpoint to add.
1247 * @param front: listen struct.
1248 * @return: false on failure.
1251 listen_cp_insert(struct comm_point* c, struct listen_dnsport* front)
1253 struct listen_list* item = (struct listen_list*)malloc(
1254 sizeof(struct listen_list));
1258 item->next = front->cps;
1263 struct listen_dnsport*
1264 listen_create(struct comm_base* base, struct listen_port* ports,
1265 size_t bufsize, int tcp_accept_count, int tcp_idle_timeout,
1266 struct tcl_list* tcp_conn_limit, void* sslctx,
1267 struct dt_env* dtenv, comm_point_callback_type* cb, void *cb_arg)
1269 struct listen_dnsport* front = (struct listen_dnsport*)
1270 malloc(sizeof(struct listen_dnsport));
1274 front->udp_buff = sldns_buffer_new(bufsize);
1276 front->dnscrypt_udp_buff = NULL;
1278 if(!front->udp_buff) {
1282 if(!stream_wait_lock_inited) {
1283 lock_basic_init(&stream_wait_count_lock);
1284 stream_wait_lock_inited = 1;
1287 /* create comm points as needed */
1289 struct comm_point* cp = NULL;
1290 if(ports->ftype == listen_type_udp ||
1291 ports->ftype == listen_type_udp_dnscrypt)
1292 cp = comm_point_create_udp(base, ports->fd,
1293 front->udp_buff, cb, cb_arg);
1294 else if(ports->ftype == listen_type_tcp ||
1295 ports->ftype == listen_type_tcp_dnscrypt)
1296 cp = comm_point_create_tcp(base, ports->fd,
1297 tcp_accept_count, tcp_idle_timeout,
1298 tcp_conn_limit, bufsize, front->udp_buff,
1300 else if(ports->ftype == listen_type_ssl) {
1301 cp = comm_point_create_tcp(base, ports->fd,
1302 tcp_accept_count, tcp_idle_timeout,
1303 tcp_conn_limit, bufsize, front->udp_buff,
1306 } else if(ports->ftype == listen_type_udpancil ||
1307 ports->ftype == listen_type_udpancil_dnscrypt)
1308 cp = comm_point_create_udp_ancil(base, ports->fd,
1309 front->udp_buff, cb, cb_arg);
1311 log_err("can't create commpoint");
1312 listen_delete(front);
1316 cp->do_not_close = 1;
1318 if (ports->ftype == listen_type_udp_dnscrypt ||
1319 ports->ftype == listen_type_tcp_dnscrypt ||
1320 ports->ftype == listen_type_udpancil_dnscrypt) {
1322 cp->dnscrypt_buffer = sldns_buffer_new(bufsize);
1323 if(!cp->dnscrypt_buffer) {
1324 log_err("can't alloc dnscrypt_buffer");
1325 comm_point_delete(cp);
1326 listen_delete(front);
1329 front->dnscrypt_udp_buff = cp->dnscrypt_buffer;
1332 if(!listen_cp_insert(cp, front)) {
1333 log_err("malloc failed");
1334 comm_point_delete(cp);
1335 listen_delete(front);
1338 ports = ports->next;
1341 log_err("Could not open sockets to accept queries.");
1342 listen_delete(front);
1350 listen_list_delete(struct listen_list* list)
1352 struct listen_list *p = list, *pn;
1355 comm_point_delete(p->com);
1362 listen_delete(struct listen_dnsport* front)
1366 listen_list_delete(front->cps);
1368 if(front->dnscrypt_udp_buff &&
1369 front->udp_buff != front->dnscrypt_udp_buff) {
1370 sldns_buffer_free(front->dnscrypt_udp_buff);
1373 sldns_buffer_free(front->udp_buff);
1375 if(stream_wait_lock_inited) {
1376 stream_wait_lock_inited = 0;
1377 lock_basic_destroy(&stream_wait_count_lock);
1382 listening_ports_open(struct config_file* cfg, int* reuseport)
1384 struct listen_port* list = NULL;
1385 struct addrinfo hints;
1386 int i, do_ip4, do_ip6;
1387 int do_tcp, do_auto;
1389 snprintf(portbuf, sizeof(portbuf), "%d", cfg->port);
1390 do_ip4 = cfg->do_ip4;
1391 do_ip6 = cfg->do_ip6;
1392 do_tcp = cfg->do_tcp;
1393 do_auto = cfg->if_automatic && cfg->do_udp;
1394 if(cfg->incoming_num_tcp == 0)
1398 memset(&hints, 0, sizeof(hints));
1399 hints.ai_flags = AI_PASSIVE;
1400 /* no name lookups on our listening ports */
1401 if(cfg->num_ifs > 0)
1402 hints.ai_flags |= AI_NUMERICHOST;
1403 hints.ai_family = AF_UNSPEC;
1407 if(!do_ip4 && !do_ip6) {
1410 /* create ip4 and ip6 ports so that return addresses are nice. */
1411 if(do_auto || cfg->num_ifs == 0) {
1413 hints.ai_family = AF_INET6;
1414 if(!ports_create_if(do_auto?"::0":"::1",
1415 do_auto, cfg->do_udp, do_tcp,
1416 &hints, portbuf, &list,
1417 cfg->so_rcvbuf, cfg->so_sndbuf,
1418 cfg->ssl_port, cfg->tls_additional_port,
1419 reuseport, cfg->ip_transparent,
1420 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1421 cfg->dnscrypt_port)) {
1422 listening_ports_free(list);
1427 hints.ai_family = AF_INET;
1428 if(!ports_create_if(do_auto?"0.0.0.0":"127.0.0.1",
1429 do_auto, cfg->do_udp, do_tcp,
1430 &hints, portbuf, &list,
1431 cfg->so_rcvbuf, cfg->so_sndbuf,
1432 cfg->ssl_port, cfg->tls_additional_port,
1433 reuseport, cfg->ip_transparent,
1434 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1435 cfg->dnscrypt_port)) {
1436 listening_ports_free(list);
1440 } else for(i = 0; i<cfg->num_ifs; i++) {
1441 if(str_is_ip6(cfg->ifs[i])) {
1444 hints.ai_family = AF_INET6;
1445 if(!ports_create_if(cfg->ifs[i], 0, cfg->do_udp,
1446 do_tcp, &hints, portbuf, &list,
1447 cfg->so_rcvbuf, cfg->so_sndbuf,
1448 cfg->ssl_port, cfg->tls_additional_port,
1449 reuseport, cfg->ip_transparent,
1450 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1451 cfg->dnscrypt_port)) {
1452 listening_ports_free(list);
1458 hints.ai_family = AF_INET;
1459 if(!ports_create_if(cfg->ifs[i], 0, cfg->do_udp,
1460 do_tcp, &hints, portbuf, &list,
1461 cfg->so_rcvbuf, cfg->so_sndbuf,
1462 cfg->ssl_port, cfg->tls_additional_port,
1463 reuseport, cfg->ip_transparent,
1464 cfg->tcp_mss, cfg->ip_freebind, cfg->use_systemd,
1465 cfg->dnscrypt_port)) {
1466 listening_ports_free(list);
1474 void listening_ports_free(struct listen_port* list)
1476 struct listen_port* nx;
1479 if(list->fd != -1) {
1483 closesocket(list->fd);
1491 size_t listen_get_mem(struct listen_dnsport* listen)
1493 struct listen_list* p;
1494 size_t s = sizeof(*listen) + sizeof(*listen->base) +
1495 sizeof(*listen->udp_buff) +
1496 sldns_buffer_capacity(listen->udp_buff);
1498 s += sizeof(*listen->dnscrypt_udp_buff);
1499 if(listen->udp_buff != listen->dnscrypt_udp_buff){
1500 s += sldns_buffer_capacity(listen->dnscrypt_udp_buff);
1503 for(p = listen->cps; p; p = p->next) {
1505 s += comm_point_get_mem(p->com);
1510 void listen_stop_accept(struct listen_dnsport* listen)
1512 /* do not stop the ones that have no tcp_free list
1513 * (they have already stopped listening) */
1514 struct listen_list* p;
1515 for(p=listen->cps; p; p=p->next) {
1516 if(p->com->type == comm_tcp_accept &&
1517 p->com->tcp_free != NULL) {
1518 comm_point_stop_listening(p->com);
1523 void listen_start_accept(struct listen_dnsport* listen)
1525 /* do not start the ones that have no tcp_free list, it is no
1526 * use to listen to them because they have no free tcp handlers */
1527 struct listen_list* p;
1528 for(p=listen->cps; p; p=p->next) {
1529 if(p->com->type == comm_tcp_accept &&
1530 p->com->tcp_free != NULL) {
1531 comm_point_start_listening(p->com, -1, -1);
1536 struct tcp_req_info*
1537 tcp_req_info_create(struct sldns_buffer* spoolbuf)
1539 struct tcp_req_info* req = (struct tcp_req_info*)malloc(sizeof(*req));
1541 log_err("malloc failure for new stream outoforder processing structure");
1544 memset(req, 0, sizeof(*req));
1545 req->spool_buffer = spoolbuf;
1550 tcp_req_info_delete(struct tcp_req_info* req)
1553 tcp_req_info_clear(req);
1554 /* cp is pointer back to commpoint that owns this struct and
1555 * called delete on us */
1556 /* spool_buffer is shared udp buffer, not deleted here */
1560 void tcp_req_info_clear(struct tcp_req_info* req)
1562 struct tcp_req_open_item* open, *nopen;
1563 struct tcp_req_done_item* item, *nitem;
1566 /* free outstanding request mesh reply entries */
1567 open = req->open_req_list;
1570 mesh_state_remove_reply(open->mesh, open->mesh_state, req->cp);
1574 req->open_req_list = NULL;
1575 req->num_open_req = 0;
1577 /* free pending writable result packets */
1578 item = req->done_req_list;
1581 lock_basic_lock(&stream_wait_count_lock);
1582 stream_wait_count -= (sizeof(struct tcp_req_done_item)
1584 lock_basic_unlock(&stream_wait_count_lock);
1589 req->done_req_list = NULL;
1590 req->num_done_req = 0;
1591 req->read_is_closed = 0;
1595 tcp_req_info_remove_mesh_state(struct tcp_req_info* req, struct mesh_state* m)
1597 struct tcp_req_open_item* open, *prev = NULL;
1598 if(!req || !m) return;
1599 open = req->open_req_list;
1601 if(open->mesh_state == m) {
1602 struct tcp_req_open_item* next;
1603 if(prev) prev->next = open->next;
1604 else req->open_req_list = open->next;
1605 /* caller has to manage the mesh state reply entry */
1608 req->num_open_req --;
1619 /** setup listening for read or write */
1621 tcp_req_info_setup_listen(struct tcp_req_info* req)
1626 if(req->cp->tcp_byte_count != 0) {
1627 /* cannot change, halfway through */
1631 if(!req->cp->tcp_is_reading)
1633 if(req->num_open_req + req->num_done_req < TCP_MAX_REQ_SIMULTANEOUS &&
1634 !req->read_is_closed)
1638 req->cp->tcp_is_reading = 0;
1639 comm_point_start_listening(req->cp, -1,
1640 req->cp->tcp_timeout_msec);
1642 req->cp->tcp_is_reading = 1;
1643 comm_point_start_listening(req->cp, -1,
1644 req->cp->tcp_timeout_msec);
1645 /* and also read it (from SSL stack buffers), so
1646 * no event read event is expected since the remainder of
1647 * the TLS frame is sitting in the buffers. */
1648 req->read_again = 1;
1650 comm_point_start_listening(req->cp, -1,
1651 req->cp->tcp_timeout_msec);
1652 comm_point_listen_for_rw(req->cp, 0, 0);
1656 /** remove first item from list of pending results */
1657 static struct tcp_req_done_item*
1658 tcp_req_info_pop_done(struct tcp_req_info* req)
1660 struct tcp_req_done_item* item;
1661 log_assert(req->num_done_req > 0 && req->done_req_list);
1662 item = req->done_req_list;
1663 lock_basic_lock(&stream_wait_count_lock);
1664 stream_wait_count -= (sizeof(struct tcp_req_done_item)+item->len);
1665 lock_basic_unlock(&stream_wait_count_lock);
1666 req->done_req_list = req->done_req_list->next;
1667 req->num_done_req --;
1671 /** Send given buffer and setup to write */
1673 tcp_req_info_start_write_buf(struct tcp_req_info* req, uint8_t* buf,
1676 sldns_buffer_clear(req->cp->buffer);
1677 sldns_buffer_write(req->cp->buffer, buf, len);
1678 sldns_buffer_flip(req->cp->buffer);
1680 req->cp->tcp_is_reading = 0; /* we are now writing */
1683 /** pick up the next result and start writing it to the channel */
1685 tcp_req_pickup_next_result(struct tcp_req_info* req)
1687 if(req->num_done_req > 0) {
1688 /* unlist the done item from the list of pending results */
1689 struct tcp_req_done_item* item = tcp_req_info_pop_done(req);
1690 tcp_req_info_start_write_buf(req, item->buf, item->len);
1696 /** the read channel has closed */
1698 tcp_req_info_handle_read_close(struct tcp_req_info* req)
1700 verbose(VERB_ALGO, "tcp channel read side closed %d", req->cp->fd);
1701 /* reset byte count for (potential) partial read */
1702 req->cp->tcp_byte_count = 0;
1703 /* if we still have results to write, pick up next and write it */
1704 if(req->num_done_req != 0) {
1705 tcp_req_pickup_next_result(req);
1706 tcp_req_info_setup_listen(req);
1709 /* if nothing to do, this closes the connection */
1710 if(req->num_open_req == 0 && req->num_done_req == 0)
1712 /* otherwise, we must be waiting for dns resolve, wait with timeout */
1713 req->read_is_closed = 1;
1714 tcp_req_info_setup_listen(req);
1719 tcp_req_info_handle_writedone(struct tcp_req_info* req)
1721 /* back to reading state, we finished this write event */
1722 sldns_buffer_clear(req->cp->buffer);
1723 if(req->num_done_req == 0 && req->read_is_closed) {
1724 /* no more to write and nothing to read, close it */
1725 comm_point_drop_reply(&req->cp->repinfo);
1728 req->cp->tcp_is_reading = 1;
1729 /* see if another result needs writing */
1730 tcp_req_pickup_next_result(req);
1732 /* see if there is more to write, if not stop_listening for writing */
1733 /* see if new requests are allowed, if so, start_listening
1735 tcp_req_info_setup_listen(req);
1739 tcp_req_info_handle_readdone(struct tcp_req_info* req)
1741 struct comm_point* c = req->cp;
1743 /* we want to read up several requests, unless there are
1744 * pending answers */
1748 req->in_worker_handle = 1;
1749 /* handle the current request */
1750 /* this calls the worker handle request routine that could give
1751 * a cache response, or localdata response, or drop the reply,
1752 * or schedule a mesh entry for later */
1753 fptr_ok(fptr_whitelist_comm_point(c->callback));
1754 if( (*c->callback)(c, c->cb_arg, NETEVENT_NOERROR, &c->repinfo) ) {
1755 req->in_worker_handle = 0;
1756 /* there is an answer, put it up. It is already in the
1757 * c->buffer, just send it. */
1758 /* since we were just reading a query, the channel is
1759 * clear to write to */
1761 c->tcp_is_reading = 0;
1762 comm_point_start_listening(c, -1, c->tcp_timeout_msec);
1765 req->in_worker_handle = 0;
1766 /* it should be waiting in the mesh for recursion.
1767 * If mesh failed to add a new entry and called commpoint_drop_reply.
1768 * Then the mesh state has been cleared. */
1770 /* we can now call drop_reply without recursing into ourselves
1771 * whilst in the callback */
1772 /* we have to close the stream because there is no reply,
1773 * no servfail to send, but the query needs an action, for
1774 * a stream that is close the connection */
1775 sldns_buffer_clear(c->buffer);
1776 comm_point_drop_reply(&c->repinfo);
1779 /* If mesh failed(mallocfail) and called commpoint_send_reply with
1780 * something like servfail then we pick up that reply below. */
1785 sldns_buffer_clear(c->buffer);
1786 /* if pending answers, pick up an answer and start sending it */
1787 tcp_req_pickup_next_result(req);
1789 /* if answers pending, start sending answers */
1790 /* read more requests if we can have more requests */
1791 tcp_req_info_setup_listen(req);
1795 tcp_req_info_add_meshstate(struct tcp_req_info* req,
1796 struct mesh_area* mesh, struct mesh_state* m)
1798 struct tcp_req_open_item* item;
1799 log_assert(req && mesh && m);
1800 item = (struct tcp_req_open_item*)malloc(sizeof(*item));
1802 item->next = req->open_req_list;
1804 item->mesh_state = m;
1805 req->open_req_list = item;
1806 req->num_open_req++;
1810 /** Add a result to the result list. At the end. */
1812 tcp_req_info_add_result(struct tcp_req_info* req, uint8_t* buf, size_t len)
1814 struct tcp_req_done_item* last = NULL;
1815 struct tcp_req_done_item* item;
1818 /* see if we have space */
1819 space = sizeof(struct tcp_req_done_item) + len;
1820 lock_basic_lock(&stream_wait_count_lock);
1821 if(stream_wait_count + space > stream_wait_max) {
1822 lock_basic_unlock(&stream_wait_count_lock);
1823 verbose(VERB_ALGO, "drop stream reply, no space left, in stream-wait-size");
1826 stream_wait_count += space;
1827 lock_basic_unlock(&stream_wait_count_lock);
1829 /* find last element */
1830 last = req->done_req_list;
1831 while(last && last->next)
1834 /* create new element */
1835 item = (struct tcp_req_done_item*)malloc(sizeof(*item));
1837 log_err("malloc failure, for stream result list");
1842 item->buf = memdup(buf, len);
1845 log_err("malloc failure, adding reply to stream result list");
1850 if(last) last->next = item;
1851 else req->done_req_list = item;
1852 req->num_done_req++;
1857 tcp_req_info_send_reply(struct tcp_req_info* req)
1859 if(req->in_worker_handle) {
1860 /* It is in the right buffer to answer straight away */
1864 /* now that the query has been handled, that mesh_reply entry
1865 * should be removed, from the tcp_req_info list,
1866 * the mesh state cleanup removes then with region_cleanup and
1867 * replies_sent true. */
1868 /* see if we can send it straight away (we are not doing
1869 * anything else). If so, copy to buffer and start */
1870 if(req->cp->tcp_is_reading && req->cp->tcp_byte_count == 0) {
1871 /* buffer is free, and was ready to read new query into,
1872 * but we are now going to use it to send this answer */
1873 tcp_req_info_start_write_buf(req,
1874 sldns_buffer_begin(req->spool_buffer),
1875 sldns_buffer_limit(req->spool_buffer));
1876 /* switch to listen to write events */
1877 comm_point_stop_listening(req->cp);
1878 comm_point_start_listening(req->cp, -1,
1879 req->cp->tcp_timeout_msec);
1882 /* queue up the answer behind the others already pending */
1883 if(!tcp_req_info_add_result(req, sldns_buffer_begin(req->spool_buffer),
1884 sldns_buffer_limit(req->spool_buffer))) {
1885 /* drop the connection, we are out of resources */
1886 comm_point_drop_reply(&req->cp->repinfo);
1890 size_t tcp_req_info_get_stream_buffer_size(void)
1893 if(!stream_wait_lock_inited)
1894 return stream_wait_count;
1895 lock_basic_lock(&stream_wait_count_lock);
1896 s = stream_wait_count;
1897 lock_basic_unlock(&stream_wait_count_lock);