4 * - TCP connection management
5 * - user-visible check/wait and event-loop-related functions
8 * This file is part of adns, which is
9 * Copyright (C) 1997-2000,2003,2006,2014 Ian Jackson
10 * Copyright (C) 2014 Mark Wooding
11 * Copyright (C) 1999-2000,2003,2006 Tony Finch
12 * Copyright (C) 1991 Massachusetts Institute of Technology
13 * (See the file INSTALL for full details.)
15 * This program is free software; you can redistribute it and/or modify
16 * it under the terms of the GNU General Public License as published by
17 * the Free Software Foundation; either version 3, or (at your option)
20 * This program is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 * GNU General Public License for more details.
25 * You should have received a copy of the GNU General Public License
26 * along with this program; if not, write to the Free Software Foundation.
33 #include <sys/types.h>
36 #include <sys/socket.h>
37 #include <netinet/in.h>
38 #include <arpa/inet.h>
43 /* TCP connection management. */
45 static void tcp_close(adns_state ads) {
46 close(ads->tcpsocket);
48 ads->tcprecv.used= ads->tcprecv_skip= ads->tcpsend.used= 0;
51 void adns__tcp_broken(adns_state ads, const char *what, const char *why) {
55 assert(ads->tcpstate == server_connecting || ads->tcpstate == server_ok);
57 if (what) adns__warn(ads,serv,0,"TCP connection failed: %s: %s",what,why);
59 if (ads->tcpstate == server_connecting) {
60 /* Counts as a retry for all the queries waiting for TCP. */
61 for (qu= ads->tcpw.head; qu; qu= qu->next)
66 ads->tcpstate= server_broken;
67 ads->tcpserver= (serv+1)%ads->nservers;
70 static void tcp_connected(adns_state ads, struct timeval now) {
73 adns__debug(ads,ads->tcpserver,0,"TCP connected");
74 ads->tcpstate= server_ok;
75 for (qu= ads->tcpw.head; qu && ads->tcpstate == server_ok; qu= nqu) {
77 assert(qu->state == query_tcpw);
78 adns__querysend_tcp(qu,now);
82 static void tcp_broken_events(adns_state ads) {
85 assert(ads->tcpstate == server_broken);
86 for (qu= ads->tcpw.head; qu; qu= nqu) {
88 assert(qu->state == query_tcpw);
89 if (qu->retries > ads->nservers) {
90 LIST_UNLINK(ads->tcpw,qu);
91 adns__query_fail(qu,adns_s_allservfail);
94 ads->tcpstate= server_disconnected;
97 void adns__tcp_tryconnect(adns_state ads, struct timeval now) {
100 struct protoent *proto;
102 for (tries=0; tries<ads->nservers; tries++) {
103 switch (ads->tcpstate) {
104 case server_connecting:
108 case server_disconnected:
114 assert(!ads->tcpsend.used);
115 assert(!ads->tcprecv.used);
116 assert(!ads->tcprecv_skip);
118 proto= getprotobyname("tcp");
120 adns__diag(ads,-1,0,"unable to find protocol no. for TCP !");
123 addr = &ads->servers[ads->tcpserver];
124 fd= socket(addr->addr.sa.sa_family, SOCK_STREAM, proto->p_proto);
126 adns__diag(ads,-1,0,"cannot create TCP socket: %s",strerror(errno));
129 r= adns__setnonblock(ads,fd);
131 adns__diag(ads,-1,0,"cannot make TCP socket nonblocking:"
136 r= connect(fd,&addr->addr.sa,addr->len);
138 ads->tcpstate= server_connecting;
139 if (r==0) { tcp_connected(ads,now); return; }
140 if (errno == EWOULDBLOCK || errno == EINPROGRESS) {
141 ads->tcptimeout= now;
142 timevaladd(&ads->tcptimeout,TCPCONNMS);
145 adns__tcp_broken(ads,"connect",strerror(errno));
146 tcp_broken_events(ads);
150 /* Timeout handling functions. */
152 void adns__must_gettimeofday(adns_state ads, const struct timeval **now_io,
153 struct timeval *tv_buf) {
154 const struct timeval *now;
159 r= gettimeofday(tv_buf,0); if (!r) { *now_io= tv_buf; return; }
160 adns__diag(ads,-1,0,"gettimeofday failed: %s",strerror(errno));
161 adns_globalsystemfailure(ads);
165 static void inter_immed(struct timeval **tv_io, struct timeval *tvbuf) {
166 struct timeval *rbuf;
171 if (!rbuf) { *tv_io= rbuf= tvbuf; }
176 static void inter_maxto(struct timeval **tv_io, struct timeval *tvbuf,
177 struct timeval maxto) {
178 struct timeval *rbuf;
183 *tvbuf= maxto; *tv_io= tvbuf;
185 if (timercmp(rbuf,&maxto,>)) *rbuf= maxto;
187 /*fprintf(stderr,"inter_maxto maxto=%ld.%06ld result=%ld.%06ld\n",
188 maxto.tv_sec,maxto.tv_usec,(**tv_io).tv_sec,(**tv_io).tv_usec);*/
191 static void inter_maxtoabs(struct timeval **tv_io, struct timeval *tvbuf,
192 struct timeval now, struct timeval maxtime) {
196 /*fprintf(stderr,"inter_maxtoabs now=%ld.%06ld maxtime=%ld.%06ld\n",
197 now.tv_sec,now.tv_usec,maxtime.tv_sec,maxtime.tv_usec);*/
199 maxtime.tv_sec -= (now.tv_sec+2);
200 maxtime.tv_usec -= (now.tv_usec-2000000);
201 dr= ldiv(maxtime.tv_usec,1000000);
202 maxtime.tv_sec += dr.quot;
203 maxtime.tv_usec -= dr.quot*1000000;
204 if (maxtime.tv_sec<0) timerclear(&maxtime);
205 inter_maxto(tv_io,tvbuf,maxtime);
208 static void timeouts_queue(adns_state ads, int act,
209 struct timeval **tv_io, struct timeval *tvbuf,
210 struct timeval now, struct query_queue *queue) {
213 for (qu= queue->head; qu; qu= nqu) {
215 if (!timercmp(&now,&qu->timeout,>)) {
216 inter_maxtoabs(tv_io,tvbuf,now,qu->timeout);
218 if (!act) { inter_immed(tv_io,tvbuf); return; }
219 LIST_UNLINK(*queue,qu);
220 if (qu->state != query_tosend) {
221 adns__query_fail(qu,adns_s_timeout);
223 adns__query_send(qu,now);
230 static void tcp_events(adns_state ads, int act,
231 struct timeval **tv_io, struct timeval *tvbuf,
232 struct timeval now) {
234 switch (ads->tcpstate) {
236 if (!act) { inter_immed(tv_io,tvbuf); return; }
237 tcp_broken_events(ads);
238 case server_disconnected: /* fall through */
239 if (!ads->tcpw.head) return;
240 if (!act) { inter_immed(tv_io,tvbuf); return; }
241 adns__tcp_tryconnect(ads,now);
244 if (ads->tcpw.head) return;
245 if (!ads->tcptimeout.tv_sec) {
246 assert(!ads->tcptimeout.tv_usec);
247 ads->tcptimeout= now;
248 timevaladd(&ads->tcptimeout,TCPIDLEMS);
250 case server_connecting: /* fall through */
251 if (!act || !timercmp(&now,&ads->tcptimeout,>)) {
252 inter_maxtoabs(tv_io,tvbuf,now,ads->tcptimeout);
255 /* TCP timeout has happened */
256 switch (ads->tcpstate) {
257 case server_connecting: /* failed to connect */
258 adns__tcp_broken(ads,"unable to make connection","timed out");
260 case server_ok: /* idle timeout */
262 ads->tcpstate= server_disconnected;
276 void adns__timeouts(adns_state ads, int act,
277 struct timeval **tv_io, struct timeval *tvbuf,
278 struct timeval now) {
279 timeouts_queue(ads,act,tv_io,tvbuf,now, &ads->udpw);
280 timeouts_queue(ads,act,tv_io,tvbuf,now, &ads->tcpw);
281 tcp_events(ads,act,tv_io,tvbuf,now);
284 void adns_firsttimeout(adns_state ads,
285 struct timeval **tv_io, struct timeval *tvbuf,
286 struct timeval now) {
287 adns__consistency(ads,0,cc_entex);
288 adns__timeouts(ads, 0, tv_io,tvbuf, now);
289 adns__returning(ads,0);
292 void adns_processtimeouts(adns_state ads, const struct timeval *now) {
293 struct timeval tv_buf;
295 adns__consistency(ads,0,cc_entex);
296 adns__must_gettimeofday(ads,&now,&tv_buf);
297 if (now) adns__timeouts(ads, 1, 0,0, *now);
298 adns__returning(ads,0);
301 /* fd handling functions. These are the top-level of the real work of
302 * reception and often transmission.
305 int adns__pollfds(adns_state ads, struct pollfd pollfds_buf[MAX_POLLFDS]) {
306 /* Returns the number of entries filled in. Always zeroes revents. */
308 #define ADD_POLLFD(wantfd, wantevents) do{ \
309 pollfds_buf[nwanted].fd= (wantfd); \
310 pollfds_buf[nwanted].events= (wantevents); \
311 pollfds_buf[nwanted].revents= 0; \
317 assert(MAX_POLLFDS == MAXUDP + 1);
319 for (i=0; i<ads->nudpsockets; i++)
320 ADD_POLLFD(ads->udpsockets[i].fd, POLLIN);
322 switch (ads->tcpstate) {
323 case server_disconnected:
326 case server_connecting:
327 ADD_POLLFD(ads->tcpsocket, POLLOUT);
330 ADD_POLLFD(ads->tcpsocket,
331 ads->tcpsend.used ? POLLIN|POLLOUT|POLLPRI : POLLIN|POLLPRI);
336 assert(nwanted<=MAX_POLLFDS);
341 int adns_processreadable(adns_state ads, int fd, const struct timeval *now) {
342 int want, dgramlen, r, i, serv, old_skip;
343 socklen_t udpaddrlen;
344 byte udpbuf[DNS_MAXUDP];
345 char addrbuf[ADNS_ADDR2TEXT_BUFLEN];
346 struct udpsocket *udp;
347 adns_sockaddr udpaddr;
349 adns__consistency(ads,0,cc_entex);
351 switch (ads->tcpstate) {
352 case server_disconnected:
354 case server_connecting:
357 if (fd != ads->tcpsocket) break;
358 assert(!ads->tcprecv_skip);
360 if (ads->tcprecv.used >= ads->tcprecv_skip+2) {
361 dgramlen= ((ads->tcprecv.buf[ads->tcprecv_skip]<<8) |
362 ads->tcprecv.buf[ads->tcprecv_skip+1]);
363 if (ads->tcprecv.used >= ads->tcprecv_skip+2+dgramlen) {
364 old_skip= ads->tcprecv_skip;
365 ads->tcprecv_skip += 2+dgramlen;
366 adns__procdgram(ads, ads->tcprecv.buf+old_skip+2,
367 dgramlen, ads->tcpserver, 1,*now);
375 ads->tcprecv.used -= ads->tcprecv_skip;
376 memmove(ads->tcprecv.buf, ads->tcprecv.buf+ads->tcprecv_skip,
378 ads->tcprecv_skip= 0;
379 if (!adns__vbuf_ensure(&ads->tcprecv,want)) { r= ENOMEM; goto xit; }
380 assert(ads->tcprecv.used <= ads->tcprecv.avail);
381 if (ads->tcprecv.used == ads->tcprecv.avail) continue;
382 r= read(ads->tcpsocket,
383 ads->tcprecv.buf+ads->tcprecv.used,
384 ads->tcprecv.avail-ads->tcprecv.used);
386 ads->tcprecv.used+= r;
389 if (errno==EAGAIN || errno==EWOULDBLOCK) { r= 0; goto xit; }
390 if (errno==EINTR) continue;
391 if (errno_resources(errno)) { r= errno; goto xit; }
393 adns__tcp_broken(ads,"read",r?strerror(errno):"closed");
395 } while (ads->tcpstate == server_ok);
400 for (i=0; i<ads->nudpsockets; i++) {
401 udp= &ads->udpsockets[i];
402 if (fd != udp->fd) continue;
404 udpaddrlen= sizeof(udpaddr);
405 r= recvfrom(fd,udpbuf,sizeof(udpbuf),0, &udpaddr.sa,&udpaddrlen);
407 if (errno == EAGAIN || errno == EWOULDBLOCK) { r= 0; goto xit; }
408 if (errno == EINTR) continue;
409 if (errno_resources(errno)) { r= errno; goto xit; }
410 adns__warn(ads,-1,0,"datagram receive error: %s",strerror(errno));
414 serv < ads->nservers &&
415 !adns__sockaddrs_equal(&udpaddr.sa,
416 &ads->servers[serv].addr.sa);
418 if (serv >= ads->nservers) {
419 adns__warn(ads,-1,0,"datagram received from unknown nameserver %s",
420 adns__sockaddr_ntoa(&udpaddr.sa, addrbuf));
423 adns__procdgram(ads,udpbuf,r,serv,0,*now);
429 adns__returning(ads,0);
433 int adns_processwriteable(adns_state ads, int fd, const struct timeval *now) {
436 adns__consistency(ads,0,cc_entex);
438 switch (ads->tcpstate) {
439 case server_disconnected:
442 case server_connecting:
443 if (fd != ads->tcpsocket) break;
444 assert(ads->tcprecv.used==0);
445 assert(ads->tcprecv_skip==0);
447 /* This function can be called even if the fd wasn't actually
448 * flagged as writeable. For asynch tcp connect we have to
449 * actually use the writeability to tell us the connect has
450 * completed (or failed), so we need to double check. */
452 struct timeval timeout = { 0,0 };
454 FD_SET(ads->tcpsocket,&writeable);
455 r= select(ads->tcpsocket+1,0,&writeable,0,&timeout);
458 if (errno==EINTR) continue;
459 adns__tcp_broken(ads,"select","failed connecting writeability check");
462 assert(FD_ISSET(ads->tcpsocket,&writeable));
463 if (!adns__vbuf_ensure(&ads->tcprecv,1)) { r= ENOMEM; goto xit; }
464 r= read(ads->tcpsocket,&ads->tcprecv.buf,1);
465 if (r==0 || (r<0 && (errno==EAGAIN || errno==EWOULDBLOCK))) {
466 tcp_connected(ads,*now);
470 adns__tcp_broken(ads,"connect/read","sent data before first request");
473 if (errno==EINTR) continue;
474 if (errno_resources(errno)) { r= errno; goto xit; }
475 adns__tcp_broken(ads,"connect/read",strerror(errno));
479 if (fd != ads->tcpsocket) break;
480 while (ads->tcpsend.used) {
481 adns__sigpipe_protect(ads);
482 r= write(ads->tcpsocket,ads->tcpsend.buf,ads->tcpsend.used);
483 adns__sigpipe_unprotect(ads);
485 if (errno==EINTR) continue;
486 if (errno==EAGAIN || errno==EWOULDBLOCK) { r= 0; goto xit; }
487 if (errno_resources(errno)) { r= errno; goto xit; }
488 adns__tcp_broken(ads,"write",strerror(errno));
491 ads->tcpsend.used -= r;
492 memmove(ads->tcpsend.buf,ads->tcpsend.buf+r,ads->tcpsend.used);
502 adns__returning(ads,0);
506 int adns_processexceptional(adns_state ads, int fd,
507 const struct timeval *now) {
508 adns__consistency(ads,0,cc_entex);
509 switch (ads->tcpstate) {
510 case server_disconnected:
513 case server_connecting:
515 if (fd != ads->tcpsocket) break;
516 adns__tcp_broken(ads,"poll/select","exceptional condition detected");
521 adns__returning(ads,0);
525 static void fd_event(adns_state ads, int fd,
526 int revent, int pollflag,
527 int maxfd, const fd_set *fds,
528 int (*func)(adns_state, int fd,
529 const struct timeval *now),
530 struct timeval now, int *r_r) {
533 if (!(revent & pollflag)) return;
534 if (fds && !(fd<maxfd && FD_ISSET(fd,fds))) return;
535 r= func(ads,fd,&now);
540 adns__diag(ads,-1,0,"process fd failed after select:"
541 " %s",strerror(errno));
542 adns_globalsystemfailure(ads);
547 void adns__fdevents(adns_state ads,
548 const struct pollfd *pollfds, int npollfds,
549 int maxfd, const fd_set *readfds,
550 const fd_set *writefds, const fd_set *exceptfds,
551 struct timeval now, int *r_r) {
554 for (i=0; i<npollfds; i++) {
556 if (fd >= maxfd) maxfd= fd+1;
557 revents= pollfds[i].revents;
558 #define EV(pollfl,fds,how) \
559 fd_event(ads,fd, revents,pollfl, maxfd,fds, adns_process##how,now,r_r)
560 EV( POLLIN, readfds, readable );
561 EV( POLLOUT, writefds, writeable );
562 EV( POLLPRI, exceptfds, exceptional );
567 /* Wrappers for select(2). */
569 void adns_beforeselect(adns_state ads, int *maxfd_io, fd_set *readfds_io,
570 fd_set *writefds_io, fd_set *exceptfds_io,
571 struct timeval **tv_mod, struct timeval *tv_tobuf,
572 const struct timeval *now) {
573 struct timeval tv_nowbuf;
574 struct pollfd pollfds[MAX_POLLFDS];
575 int i, fd, maxfd, npollfds;
577 adns__consistency(ads,0,cc_entex);
579 if (tv_mod && (!*tv_mod || (*tv_mod)->tv_sec || (*tv_mod)->tv_usec)) {
580 /* The caller is planning to sleep. */
581 adns__must_gettimeofday(ads,&now,&tv_nowbuf);
582 if (!now) { inter_immed(tv_mod,tv_tobuf); goto xit; }
583 adns__timeouts(ads, 0, tv_mod,tv_tobuf, *now);
586 npollfds= adns__pollfds(ads,pollfds);
588 for (i=0; i<npollfds; i++) {
590 if (fd >= maxfd) maxfd= fd+1;
591 if (pollfds[i].events & POLLIN) FD_SET(fd,readfds_io);
592 if (pollfds[i].events & POLLOUT) FD_SET(fd,writefds_io);
593 if (pollfds[i].events & POLLPRI) FD_SET(fd,exceptfds_io);
598 adns__returning(ads,0);
601 void adns_afterselect(adns_state ads, int maxfd, const fd_set *readfds,
602 const fd_set *writefds, const fd_set *exceptfds,
603 const struct timeval *now) {
604 struct timeval tv_buf;
605 struct pollfd pollfds[MAX_POLLFDS];
608 adns__consistency(ads,0,cc_entex);
609 adns__must_gettimeofday(ads,&now,&tv_buf);
611 adns_processtimeouts(ads,now);
613 npollfds= adns__pollfds(ads,pollfds);
614 for (i=0; i<npollfds; i++) pollfds[i].revents= POLLIN|POLLOUT|POLLPRI;
617 maxfd,readfds,writefds,exceptfds,
620 adns__returning(ads,0);
623 /* General helpful functions. */
625 void adns_globalsystemfailure(adns_state ads) {
626 adns__consistency(ads,0,cc_entex);
628 while (ads->udpw.head) adns__query_fail(ads->udpw.head, adns_s_systemfail);
629 while (ads->tcpw.head) adns__query_fail(ads->tcpw.head, adns_s_systemfail);
631 switch (ads->tcpstate) {
632 case server_connecting:
634 adns__tcp_broken(ads,0,0);
636 case server_disconnected:
642 adns__returning(ads,0);
645 int adns_processany(adns_state ads) {
648 struct pollfd pollfds[MAX_POLLFDS];
651 adns__consistency(ads,0,cc_entex);
653 r= gettimeofday(&now,0);
654 if (!r) adns_processtimeouts(ads,&now);
656 /* We just use adns__fdevents to loop over the fd's trying them.
657 * This seems more sensible than calling select, since we're most
658 * likely just to want to do a read on one or two fds anyway.
660 npollfds= adns__pollfds(ads,pollfds);
661 for (i=0; i<npollfds; i++) pollfds[i].revents= pollfds[i].events & ~POLLPRI;
667 adns__returning(ads,0);
671 void adns__autosys(adns_state ads, struct timeval now) {
672 if (ads->iflags & adns_if_noautosys) return;
673 adns_processany(ads);
676 int adns__internal_check(adns_state ads,
677 adns_query *query_io,
678 adns_answer **answer,
684 if (ads->output.head) {
685 qu= ads->output.head;
686 } else if (ads->udpw.head || ads->tcpw.head) {
692 if (qu->id>=0) return EAGAIN;
694 LIST_UNLINK(ads->output,qu);
696 if (context_r) *context_r= qu->ctx.ext;
702 int adns_wait(adns_state ads,
703 adns_query *query_io,
704 adns_answer **answer_r,
707 fd_set readfds, writefds, exceptfds;
708 struct timeval tvbuf, *tvp;
710 adns__consistency(ads,*query_io,cc_entex);
712 r= adns__internal_check(ads,query_io,answer_r,context_r);
713 if (r != EAGAIN) break;
715 FD_ZERO(&readfds); FD_ZERO(&writefds); FD_ZERO(&exceptfds);
716 adns_beforeselect(ads,&maxfd,&readfds,&writefds,&exceptfds,&tvp,&tvbuf,0);
718 rsel= select(maxfd,&readfds,&writefds,&exceptfds,tvp);
720 if (errno == EINTR) {
721 if (ads->iflags & adns_if_eintr) { r= EINTR; break; }
723 adns__diag(ads,-1,0,"select failed in wait: %s",strerror(errno));
724 adns_globalsystemfailure(ads);
728 adns_afterselect(ads,maxfd,&readfds,&writefds,&exceptfds,0);
731 adns__returning(ads,0);
735 int adns_check(adns_state ads,
736 adns_query *query_io,
737 adns_answer **answer_r,
742 adns__consistency(ads,*query_io,cc_entex);
743 r= gettimeofday(&now,0);
744 if (!r) adns__autosys(ads,now);
746 r= adns__internal_check(ads,query_io,answer_r,context_r);
747 adns__returning(ads,0);