X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=adns.git;a=blobdiff_plain;f=src%2Fevent.c;h=04ba26bdcb1a10797af4ecd2b7ec4f7b2992865e;hp=3007a95fdacd582fc290d040a6fda8624db65af7;hb=13d889da08a24f8999f0f98fff913bec6c9b71d2;hpb=8402e34c7df0adad223cf1fa1328bb524a15d99f

diff --git a/src/event.c b/src/event.c
index 3007a95..04ba26b 100644
--- a/src/event.c
+++ b/src/event.c
@@ -1,285 +1,692 @@
-/**/
+/*
+ * event.c
+ * - event loop core
+ * - TCP connection management
+ * - user-visible check/wait and event-loop-related functions
+ */
+/*
+ *  This file is part of adns, which is
+ *    Copyright (C) 1997-2000,2003,2006  Ian Jackson
+ *    Copyright (C) 1999-2000,2003,2006  Tony Finch
+ *    Copyright (C) 1991 Massachusetts Institute of Technology
+ *  (See the file INSTALL for full details.)
+ *  
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2, or (at your option)
+ *  any later version.
+ *  
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *  
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software Foundation,
+ *  Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 
+ */
 
-#include "adns-internal.h"
+#include <errno.h>
+#include <stdlib.h>
+#include <unistd.h>
 
-static void autosys(adns_state ads, struct timeval now) {
-  if (ads->iflags & adns_if_noautosys) return;
-  adns_callback(ads,-1,0,0,0);
-}
+#include <sys/types.h>
+#include <sys/time.h>
+#include <netdb.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+
+#include "internal.h"
+#include "tvarith.h"
+
+/* TCP connection management. */
 
-static int callb_checkfd(int maxfd, const fd_set *fds, int fd) {
-  return maxfd<0 || !fds ? 1 :
-         fd<maxfd && FD_ISSET(fd,fds);
+static void tcp_close(adns_state ads) {
+  close(ads->tcpsocket);
+  ads->tcpsocket= -1;
+  ads->tcprecv.used= ads->tcprecv_skip= ads->tcpsend.used= 0;
 }
 
 void adns__tcp_broken(adns_state ads, const char *what, const char *why) {
   int serv;
+  adns_query qu;
   
   assert(ads->tcpstate == server_connecting || ads->tcpstate == server_ok);
-  warn("nameserver %s TCP connection lost: %s: %s",
-       inet_ntoa(ads->servers[tcpserver].addr,what,why));
-  close(ads->tcpsocket);
-  ads->tcpstate= server_disconnected;
   serv= ads->tcpserver;
+  if (what) adns__warn(ads,serv,0,"TCP connection failed: %s: %s",what,why);
+
+  if (ads->tcpstate == server_connecting) {
+    /* Counts as a retry for all the queries waiting for TCP. */
+    for (qu= ads->tcpw.head; qu; qu= qu->next)
+      qu->retries++;
+  }
+
+  tcp_close(ads);
+  ads->tcpstate= server_broken;
+  ads->tcpserver= (serv+1)%ads->nservers;
+}
+
+static void tcp_connected(adns_state ads, struct timeval now) {
+  adns_query qu, nqu;
   
-  for (qu= ads->timew; qu; qu= nqu) {
+  adns__debug(ads,ads->tcpserver,0,"TCP connected");
+  ads->tcpstate= server_ok;
+  for (qu= ads->tcpw.head; qu && ads->tcpstate == server_ok; qu= nqu) {
     nqu= qu->next;
-    if (qu->senttcpserver == -1) continue;
-    assert(qu->senttcpserver == serv);
-    DLIST_UNLINK(ads->timew,qu);
-    adns__query_fail(ads,qu,adns_s_connlost); /* wishlist: send to other servers ? */
+    assert(qu->state == query_tcpw);
+    adns__querysend_tcp(qu,now);
   }
+}
 
-  ads->tcpbuf.used= 0;
-  ads->tcpserver= (serv+1)%ads->nservers;
+static void tcp_broken_events(adns_state ads) {
+  adns_query qu, nqu;
+  
+  assert(ads->tcpstate == server_broken);
+  for (qu= ads->tcpw.head; qu; qu= nqu) {
+    nqu= qu->next;
+    assert(qu->state == query_tcpw);
+    if (qu->retries > ads->nservers) {
+      LIST_UNLINK(ads->tcpw,qu);
+      adns__query_fail(qu,adns_s_allservfail);
+    }
+  }
+  ads->tcpstate= server_disconnected;
 }
 
-void adns__tcp_tryconnect(adns_state ads) {
+void adns__tcp_tryconnect(adns_state ads, struct timeval now) {
   int r, fd, tries;
-  sockaddr_in addr;
+  adns_rr_addr *addr;
+  struct protoent *proto;
 
   for (tries=0; tries<ads->nservers; tries++) {
-    if (ads->tcpstate == server_connecting || ads->tcpstate == server_ok) return;
-    assert(ads->tcpstate == server_disconnected);
-    assert(!ads->tcpbuf.used);
+    switch (ads->tcpstate) {
+    case server_connecting:
+    case server_ok:
+    case server_broken:
+      return;
+    case server_disconnected:
+      break;
+    default:
+      abort();
+    }
+    
+    assert(!ads->tcpsend.used);
+    assert(!ads->tcprecv.used);
+    assert(!ads->tcprecv_skip);
 
     proto= getprotobyname("tcp");
-    if (!proto) { diag(ads,"unable to find protocol number for TCP !",-1); return; }
-    fd= socket(AF_INET,SOCK_STREAM,proto->p_proto);
-    if (fd<0) { diag(ads,"cannot create TCP socket: %s",-1,strerror(errno)); return; }
-    if (!adns__setnonblock(fd)) return;
-    memset(&addr,0,sizeof(addr));
-    addr.sin_family= AF_INET;
-    addr.sin_port= htons(NSPORT);
-    addr.sin_addr= ads->servers[ads->tcpserver].addr;
-    r= connect(fd,&addr,sizeof(addr));
+    if (!proto) {
+      adns__diag(ads,-1,0,"unable to find protocol no. for TCP !");
+      return;
+    }
+    addr = &ads->servers[ads->tcpserver];
+    fd= socket(addr->addr.sa.sa_family, SOCK_STREAM, proto->p_proto);
+    if (fd<0) {
+      adns__diag(ads,-1,0,"cannot create TCP socket: %s",strerror(errno));
+      return;
+    }
+    r= adns__setnonblock(ads,fd);
+    if (r) {
+      adns__diag(ads,-1,0,"cannot make TCP socket nonblocking:"
+		 " %s",strerror(r));
+      close(fd);
+      return;
+    }
+    r= connect(fd,&addr->addr.sa,addr->len);
     ads->tcpsocket= fd;
     ads->tcpstate= server_connecting;
-    if (r==0) { ads->tcpstate= server_ok; return; }
-    if (errno == EWOULDBLOCK || errno == EINPROGRESS) return;
-    tcpserver_broken(ads,"connect",strerror(errno));
+    if (r==0) { tcp_connected(ads,now); return; }
+    if (errno == EWOULDBLOCK || errno == EINPROGRESS) {
+      ads->tcptimeout= now;
+      timevaladd(&ads->tcptimeout,TCPCONNMS);
+      return;
+    }
+    adns__tcp_broken(ads,"connect",strerror(errno));
+    tcp_broken_events(ads);
   }
 }
 
-int adns_callback(adns_state ads, int maxfd,
-		  const fd_set *readfds, const fd_set *writefds,
-		  const fd_set *exceptfds) {
-  int skip, dgramlen, count, udpaddrlen;
-  enum adns__tcpstate oldtcpstate;
-  unsigned char udpbuf[UDPMAXDGRAM];
-  struct sockaddr_in udpaddr;
+/* Timeout handling functions. */
+
+void adns__must_gettimeofday(adns_state ads, const struct timeval **now_io,
+			     struct timeval *tv_buf) {
+  const struct timeval *now;
+  int r;
+
+  now= *now_io;
+  if (now) return;
+  r= gettimeofday(tv_buf,0); if (!r) { *now_io= tv_buf; return; }
+  adns__diag(ads,-1,0,"gettimeofday failed: %s",strerror(errno));
+  adns_globalsystemfailure(ads);
+  return;
+}
+
+static void inter_immed(struct timeval **tv_io, struct timeval *tvbuf) {
+  struct timeval *rbuf;
+
+  if (!tv_io) return;
+
+  rbuf= *tv_io;
+  if (!rbuf) { *tv_io= rbuf= tvbuf; }
+
+  timerclear(rbuf);
+}
+    
+static void inter_maxto(struct timeval **tv_io, struct timeval *tvbuf,
+			struct timeval maxto) {
+  struct timeval *rbuf;
+
+  if (!tv_io) return;
+  rbuf= *tv_io;
+  if (!rbuf) {
+    *tvbuf= maxto; *tv_io= tvbuf;
+  } else {
+    if (timercmp(rbuf,&maxto,>)) *rbuf= maxto;
+  }
+/*fprintf(stderr,"inter_maxto maxto=%ld.%06ld result=%ld.%06ld\n",
+	maxto.tv_sec,maxto.tv_usec,(**tv_io).tv_sec,(**tv_io).tv_usec);*/
+}
+
+static void inter_maxtoabs(struct timeval **tv_io, struct timeval *tvbuf,
+			   struct timeval now, struct timeval maxtime) {
+  /* tv_io may be 0 */
+  ldiv_t dr;
 
-  count= 0;
-  oldtcpstate= ads->tcpstate;
+/*fprintf(stderr,"inter_maxtoabs now=%ld.%06ld maxtime=%ld.%06ld\n",
+	now.tv_sec,now.tv_usec,maxtime.tv_sec,maxtime.tv_usec);*/
+  if (!tv_io) return;
+  maxtime.tv_sec -= (now.tv_sec+2);
+  maxtime.tv_usec -= (now.tv_usec-2000000);
+  dr= ldiv(maxtime.tv_usec,1000000);
+  maxtime.tv_sec += dr.quot;
+  maxtime.tv_usec -= dr.quot*1000000;
+  if (maxtime.tv_sec<0) timerclear(&maxtime);
+  inter_maxto(tv_io,tvbuf,maxtime);
+}
+
+static void timeouts_queue(adns_state ads, int act,
+			   struct timeval **tv_io, struct timeval *tvbuf,
+			   struct timeval now, struct query_queue *queue) {
+  adns_query qu, nqu;
   
-  if (ads->tcpstate == server_connecting) {
-    if (callb_checkfd(maxfd,writefds,ads->tcpsocket)) {
-      count++;
-      assert(ads->tcprecv.used==0);
-      vbuf_ensure(&ads->tcprecv,1);
-      if (ads->tcprecv.buf) {
-	r= read(ads->tcpsocket,&ads->tcprecv.buf,1);
-	if (r==0 || (r<0 && (errno==EAGAIN || errno==EWOULDBLOCK))) {
-	  debug("nameserver %s TCP connected",
-		inet_ntoa(ads->servers[ads->tcpserver].addr));
-	  ads->tcpstate= server_connected;
-	} else if (r>0) {
-	  tcpserver_broken(ads,"connect/read","sent data before first request");
-	} else if (errno!=EINTR) {
-	  tcpserver_broken(ads,"connect/read",strerror(errno));
-	}
+  for (qu= queue->head; qu; qu= nqu) {
+    nqu= qu->next;
+    if (!timercmp(&now,&qu->timeout,>)) {
+      inter_maxtoabs(tv_io,tvbuf,now,qu->timeout);
+    } else {
+      if (!act) { inter_immed(tv_io,tvbuf); return; }
+      LIST_UNLINK(*queue,qu);
+      if (qu->state != query_tosend) {
+	adns__query_fail(qu,adns_s_timeout);
+      } else {
+	adns__query_send(qu,now);
       }
+      nqu= queue->head;
     }
   }
-  if (ads->tcpstate == server_connected) {
-    if (oldtcpstate == server_connected)
-      count+= callb_checkfd(maxfd,readfds,ads->tcpsocket) +
-	      callb_checkfd(maxfd,exceptfds,ads->tcpsocket) +
-	(ads->tcpsend.used && callb_checkfd(maxfd,writefds,ads->tcpsocket));
-    if (oldtcpstate != server_connected || callb_checkfd(maxfd,readfds,ads->tcpsocket)) {
-      skip= 0;
-      for (;;) {
-	if (ads->tcprecv.used<skip+2) {
-	  want= 2;
-	} else {
-	  dgramlen= (ads->tcprecv.buf[skip]<<8) | ads->tcprecv.buf[skip+1];
-	  if (ads->tcprecv.used<skip+2+dgramlen) {
-	    want= 2+dgramlen;
-	  } else {
-	    procdgram(ads,ads->tcprecv.buf+skip+2,dgramlen,ads->tcpserver);
-	    skip+= 2+dgramlen; continue;
-	  }
+}
+
+static void tcp_events(adns_state ads, int act,
+		       struct timeval **tv_io, struct timeval *tvbuf,
+		       struct timeval now) {
+  for (;;) {
+    switch (ads->tcpstate) {
+    case server_broken:
+      if (!act) { inter_immed(tv_io,tvbuf); return; }
+      tcp_broken_events(ads);
+    case server_disconnected: /* fall through */
+      if (!ads->tcpw.head) return;
+      if (!act) { inter_immed(tv_io,tvbuf); return; }
+      adns__tcp_tryconnect(ads,now);
+      break;
+    case server_ok:
+      if (ads->tcpw.head) return;
+      if (!ads->tcptimeout.tv_sec) {
+	assert(!ads->tcptimeout.tv_usec);
+	ads->tcptimeout= now;
+	timevaladd(&ads->tcptimeout,TCPIDLEMS);
+      }
+    case server_connecting: /* fall through */
+      if (!act || !timercmp(&now,&ads->tcptimeout,>)) {
+	inter_maxtoabs(tv_io,tvbuf,now,ads->tcptimeout);
+	return;
+      } {
+	/* TCP timeout has happened */
+	switch (ads->tcpstate) {
+	case server_connecting: /* failed to connect */
+	  adns__tcp_broken(ads,"unable to make connection","timed out");
+	  break;
+	case server_ok: /* idle timeout */
+	  tcp_close(ads);
+	  ads->tcpstate= server_disconnected;
+	  return;
+	default:
+	  abort();
 	}
-	ads->tcprecv.used -= skip;
-	memmove(ads->tcprecv.buf,ads->tcprecv.buf+skip,ads->tcprecv.used);
-	vbuf_ensure(&ads->tcprecv,want);
-	if (ads->tcprecv.used >= ads->tcprecv.avail) break;
-	r= read(ads->tcpsocket,
-		ads->tcprecv.buf+ads->tcprecv.used,
-		ads->tcprecv.avail-ads->tcprecv.used);
-	if (r>0) {
-	  ads->tcprecv.used+= r;
+      }
+      break;
+    default:
+      abort();
+    }
+  }
+  return;
+}
+
+void adns__timeouts(adns_state ads, int act,
+		    struct timeval **tv_io, struct timeval *tvbuf,
+		    struct timeval now) {
+  timeouts_queue(ads,act,tv_io,tvbuf,now, &ads->udpw);
+  timeouts_queue(ads,act,tv_io,tvbuf,now, &ads->tcpw);
+  tcp_events(ads,act,tv_io,tvbuf,now);
+}
+
+void adns_firsttimeout(adns_state ads,
+		       struct timeval **tv_io, struct timeval *tvbuf,
+		       struct timeval now) {
+  adns__consistency(ads,0,cc_entex);
+  adns__timeouts(ads, 0, tv_io,tvbuf, now);
+  adns__consistency(ads,0,cc_entex);
+}
+
+void adns_processtimeouts(adns_state ads, const struct timeval *now) {
+  struct timeval tv_buf;
+
+  adns__consistency(ads,0,cc_entex);
+  adns__must_gettimeofday(ads,&now,&tv_buf);
+  if (now) adns__timeouts(ads, 1, 0,0, *now);
+  adns__consistency(ads,0,cc_entex);
+}
+
+/* fd handling functions.  These are the top-level of the real work of
+ * reception and often transmission.
+ */
+
+int adns__pollfds(adns_state ads, struct pollfd pollfds_buf[MAX_POLLFDS]) {
+  /* Returns the number of entries filled in.  Always zeroes revents. */
+  int nwanted=0;
+#define ADD_POLLFD(wantfd, wantevents) do{	\
+    pollfds_buf[nwanted].fd= (wantfd);		\
+    pollfds_buf[nwanted].events= (wantevents);	\
+    pollfds_buf[nwanted].revents= 0;		\
+    nwanted++;					\
+  }while(0)
+
+  assert(MAX_POLLFDS==2);
+
+  ADD_POLLFD(ads->udpsocket, POLLIN);
+
+  switch (ads->tcpstate) {
+  case server_disconnected:
+  case server_broken:
+    break;
+  case server_connecting:
+    ADD_POLLFD(ads->tcpsocket, POLLOUT);
+    break;
+  case server_ok:
+    ADD_POLLFD(ads->tcpsocket,
+	       ads->tcpsend.used ? POLLIN|POLLOUT|POLLPRI : POLLIN|POLLPRI);
+    break;
+  default:
+    abort();
+  }
+  assert(nwanted<=MAX_POLLFDS);
+#undef ADD_POLLFD
+  return nwanted;
+}
+
+int adns_processreadable(adns_state ads, int fd, const struct timeval *now) {
+  int want, dgramlen, r, udpaddrlen, serv, old_skip;
+  byte udpbuf[DNS_MAXUDP];
+  struct sockaddr_in udpaddr;
+  
+  adns__consistency(ads,0,cc_entex);
+
+  switch (ads->tcpstate) {
+  case server_disconnected:
+  case server_broken:
+  case server_connecting:
+    break;
+  case server_ok:
+    if (fd != ads->tcpsocket) break;
+    assert(!ads->tcprecv_skip);
+    do {
+      if (ads->tcprecv.used >= ads->tcprecv_skip+2) {
+	dgramlen= ((ads->tcprecv.buf[ads->tcprecv_skip]<<8) |
+	           ads->tcprecv.buf[ads->tcprecv_skip+1]);
+	if (ads->tcprecv.used >= ads->tcprecv_skip+2+dgramlen) {
+	  old_skip= ads->tcprecv_skip;
+	  ads->tcprecv_skip += 2+dgramlen;
+	  adns__procdgram(ads, ads->tcprecv.buf+old_skip+2,
+			  dgramlen, ads->tcpserver, 1,*now);
+	  continue;
 	} else {
-	  if (r<0) {
-	    if (errno==EAGAIN || errno==EWOULDBLOCK || errno==ENOMEM) break;
-	    if (errno==EINTR) continue;
-	  }
-	  tcpserver_broken(ads->tcpserver,"read",r?strerror(errno):"closed");
-	  break;
+	  want= 2+dgramlen;
 	}
+      } else {
+	want= 2;
       }
-    } else if (callb_checkfd(maxfd,exceptfds,ads->tcpsocket)) {
-      tcpserver_broken(ads->tcpserver,"select","exceptional condition detected");
-    } else if (ads->tcpsend.used && callb_checkfd(maxfd,writefds,ads->tcpsocket)) {
-      r= write(ads->tcpsocket,ads->tcpsend.buf,ads->tcpsend.used);
-      if (r<0) {
-	if (errno!=EAGAIN && errno!=EWOULDBLOCK && errno!=ENOMEM && errno!=EINTR) {
-	  tcpserver_broken(ads->tcpserver,"write",strerror(errno));
+      ads->tcprecv.used -= ads->tcprecv_skip;
+      memmove(ads->tcprecv.buf, ads->tcprecv.buf+ads->tcprecv_skip,
+	      ads->tcprecv.used);
+      ads->tcprecv_skip= 0;
+      if (!adns__vbuf_ensure(&ads->tcprecv,want)) { r= ENOMEM; goto xit; }
+      assert(ads->tcprecv.used <= ads->tcprecv.avail);
+      if (ads->tcprecv.used == ads->tcprecv.avail) continue;
+      r= read(ads->tcpsocket,
+	      ads->tcprecv.buf+ads->tcprecv.used,
+	      ads->tcprecv.avail-ads->tcprecv.used);
+      if (r>0) {
+	ads->tcprecv.used+= r;
+      } else {
+	if (r) {
+	  if (errno==EAGAIN || errno==EWOULDBLOCK) { r= 0; goto xit; }
+	  if (errno==EINTR) continue;
+	  if (errno_resources(errno)) { r= errno; goto xit; }
 	}
-      } else if (r>0) {
-	ads->tcpsend.used -= r;
-	memmove(ads->tcpsend.buf,ads->tcpsend.buf+r,ads->tcpsend.used);
+	adns__tcp_broken(ads,"read",r?strerror(errno):"closed");
       }
-    }
+    } while (ads->tcpstate == server_ok);
+    r= 0; goto xit;
+  default:
+    abort();
   }
-
-  if (callb_checkfd(maxfd,readfds,ads->udpsocket)) {
-    count++;
+  if (fd == ads->udpsocket) {
     for (;;) {
       udpaddrlen= sizeof(udpaddr);
-      r= recvfrom(ads->udpsocket,udpbuf,sizeof(udpbuf),0,&udpaddr,&udpaddrlen);
+      r= recvfrom(ads->udpsocket,udpbuf,sizeof(udpbuf),0,
+		  (struct sockaddr*)&udpaddr,&udpaddrlen);
       if (r<0) {
-	if (!(errno == EAGAIN || errno == EWOULDBLOCK ||
-	      errno == EINTR || errno == ENOMEM || errno == ENOBUFS))
-	  warn("datagram receive error: %s",strerror(errno));
-	break;
+	if (errno == EAGAIN || errno == EWOULDBLOCK) { r= 0; goto xit; }
+	if (errno == EINTR) continue;
+	if (errno_resources(errno)) { r= errno; goto xit; }
+	adns__warn(ads,-1,0,"datagram receive error: %s",strerror(errno));
+	r= 0; goto xit;
       }
       if (udpaddrlen != sizeof(udpaddr)) {
-	diag("datagram received with wrong address length %d (expected %d)",
-	     udpaddrlen,sizeof(udpaddr));
+	adns__diag(ads,-1,0,"datagram received with wrong address length %d"
+		   " (expected %lu)", udpaddrlen,
+		   (unsigned long)sizeof(udpaddr));
 	continue;
       }
       if (udpaddr.sin_family != AF_INET) {
-	diag("datagram received with wrong protocol family %u (expected %u)",
-	     udpaddr.sin_family,AF_INET);
+	adns__diag(ads,-1,0,"datagram received with wrong protocol family"
+		   " %u (expected %u)",udpaddr.sin_family,AF_INET);
 	continue;
       }
-      if (ntohs(udpaddr.sin_port) != NSPORT) {
-	diag("datagram received from wrong port %u (expected %u)",
-	     ntohs(udpaddr.sin_port),NSPORT);
+      if (ntohs(udpaddr.sin_port) != DNS_PORT) {
+	adns__diag(ads,-1,0,"datagram received from wrong port"
+		   " %u (expected %u)", ntohs(udpaddr.sin_port),DNS_PORT);
 	continue;
       }
       for (serv= 0;
 	   serv < ads->nservers &&
-	     ads->servers[serv].addr.s_addr != udpaddr.sin_addr.s_addr;
+	     !adns__sockaddr_equal_p(&ads->servers[serv].addr.sa,
+				     (const struct sockaddr *)&udpaddr);
 	   serv++);
       if (serv >= ads->nservers) {
-	warn("datagram received from unknown nameserver %s",inet_ntoa(udpaddr.sin_addr));
+	adns__warn(ads,-1,0,"datagram received from unknown nameserver %s",
+		   inet_ntoa(udpaddr.sin_addr));
 	continue;
       }
-      procdgram(ads,udpbuf,r,serv);
+      adns__procdgram(ads,udpbuf,r,serv,0,*now);
     }
   }
+  r= 0;
+xit:
+  adns__consistency(ads,0,cc_entex);
+  return r;
 }
 
-static void inter_maxto(struct timeval **tv_io, struct timeval *tvbuf,
-			struct timeval maxto) {
-  struct timeval rbuf;
+int adns_processwriteable(adns_state ads, int fd, const struct timeval *now) {
+  int r;
+  
+  adns__consistency(ads,0,cc_entex);
 
-  rbuf= *tv_io;
-  if (!rbuf) { *tvbuf= maxto; *tv_io= tvbuf; return; }
-  if (timercmp(rbuf,&maxto,>)) *rbuf= maxto;
+  switch (ads->tcpstate) {
+  case server_disconnected:
+  case server_broken:
+    break;
+  case server_connecting:
+    if (fd != ads->tcpsocket) break;
+    assert(ads->tcprecv.used==0);
+    assert(ads->tcprecv_skip==0);
+    for (;;) {
+      if (!adns__vbuf_ensure(&ads->tcprecv,1)) { r= ENOMEM; goto xit; }
+      r= read(ads->tcpsocket,&ads->tcprecv.buf,1);
+      if (r==0 || (r<0 && (errno==EAGAIN || errno==EWOULDBLOCK))) {
+	tcp_connected(ads,*now);
+	r= 0; goto xit;
+      }
+      if (r>0) {
+	adns__tcp_broken(ads,"connect/read","sent data before first request");
+	r= 0; goto xit;
+      }
+      if (errno==EINTR) continue;
+      if (errno_resources(errno)) { r= errno; goto xit; }
+      adns__tcp_broken(ads,"connect/read",strerror(errno));
+      r= 0; goto xit;
+    } /* not reached */
+  case server_ok:
+    if (fd != ads->tcpsocket) break;
+    while (ads->tcpsend.used) {
+      adns__sigpipe_protect(ads);
+      r= write(ads->tcpsocket,ads->tcpsend.buf,ads->tcpsend.used);
+      adns__sigpipe_unprotect(ads);
+      if (r<0) {
+	if (errno==EINTR) continue;
+	if (errno==EAGAIN || errno==EWOULDBLOCK) { r= 0; goto xit; }
+	if (errno_resources(errno)) { r= errno; goto xit; }
+	adns__tcp_broken(ads,"write",strerror(errno));
+	r= 0; goto xit;
+      } else if (r>0) {
+	ads->tcpsend.used -= r;
+	memmove(ads->tcpsend.buf,ads->tcpsend.buf+r,ads->tcpsend.used);
+      }
+    }
+    r= 0;
+    goto xit;
+  default:
+    abort();
+  }
+  r= 0;
+xit:
+  adns__consistency(ads,0,cc_entex);
+  return r;
 }
-
-static void inter_maxtoabs(struct timeval **tv_io, struct timeval *tvbuf,
-			   struct timeval now, struct timeval maxtime) {
-  ldiv_t dr;
   
-  maxtime.tv_sec -= (now.tv_sec-1);
-  maxtime.tv_usec += (1000-now.tv_usec);
-  dr= ldiv(maxtime.tv_usec,1000);
-  maxtime.tv_sec += dr.quot;
-  maxtime.tv_usec -= dr.rem;
-  inter_maxto(tv_io,tvbuf,maxtime);
+int adns_processexceptional(adns_state ads, int fd,
+			    const struct timeval *now) {
+  adns__consistency(ads,0,cc_entex);
+  switch (ads->tcpstate) {
+  case server_disconnected:
+  case server_broken:
+    break;
+  case server_connecting:
+  case server_ok:
+    if (fd != ads->tcpsocket) break;
+    adns__tcp_broken(ads,"poll/select","exceptional condition detected");
+    break;
+  default:
+    abort();
+  }
+  adns__consistency(ads,0,cc_entex);
+  return 0;
 }
 
-static void localresourcerr(struct timeval **tv_io, struct timeval *tvbuf,
-			    const char *syscall) {
-  struct timeval tvto_lr;
+static void fd_event(adns_state ads, int fd,
+		     int revent, int pollflag,
+		     int maxfd, const fd_set *fds,
+		     int (*func)(adns_state, int fd,
+				 const struct timeval *now),
+		     struct timeval now, int *r_r) {
+  int r;
   
-  warn(ads,"local system resources scarce (during %s): %s",syscall,strerror(errno));
-  timerclear(&tvto_lr); timevaladd(&tvto_lr,LOCALRESOURCEMS);
-  inter_maxto(tv_io, tvbuf, tvto_lr);
-  return;
+  if (!(revent & pollflag)) return;
+  if (fds && !(fd<maxfd && FD_ISSET(fd,fds))) return;
+  r= func(ads,fd,&now);
+  if (r) {
+    if (r_r) {
+      *r_r= r;
+    } else {
+      adns__diag(ads,-1,0,"process fd failed after select:"
+		 " %s",strerror(errno));
+      adns_globalsystemfailure(ads);
+    }
+  }
 }
 
-static void inter_addfd(int *maxfd, fd_set *fds, int fd) {
-  if (fd>=*maxfd) *maxfd= fd+1;
-  FD_SET(fd,fds);
+void adns__fdevents(adns_state ads,
+		    const struct pollfd *pollfds, int npollfds,
+		    int maxfd, const fd_set *readfds,
+		    const fd_set *writefds, const fd_set *exceptfds,
+		    struct timeval now, int *r_r) {
+  int i, fd, revents;
+
+  for (i=0; i<npollfds; i++) {
+    fd= pollfds[i].fd;
+    if (fd >= maxfd) maxfd= fd+1;
+    revents= pollfds[i].revents;
+#define EV(pollfl,fds,how)  \
+    fd_event(ads,fd, revents,pollfl, maxfd,fds, adns_process##how,now,r_r)
+    EV( POLLIN,  readfds,   readable    );
+    EV( POLLOUT, writefds,  writeable   );
+    EV( POLLPRI, exceptfds, exceptional );
+#undef EV
+  }
 }
 
-void adns_interest(adns_state ads, int *maxfd,
-		   fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
-		   struct timeval **tv_io, struct timeval *tvbuf) {
-  struct timeval now;
-  adns_query qu;
-  int r;
+/* Wrappers for select(2). */
+
+void adns_beforeselect(adns_state ads, int *maxfd_io, fd_set *readfds_io,
+		       fd_set *writefds_io, fd_set *exceptfds_io,
+		       struct timeval **tv_mod, struct timeval *tv_tobuf,
+		       const struct timeval *now) {
+  struct timeval tv_nowbuf;
+  struct pollfd pollfds[MAX_POLLFDS];
+  int i, fd, maxfd, npollfds;
   
-  r= gettimeofday(&now,0);
-  if (r) { localresourcerr(tv_io,tvbuf,"gettimeofday"); return; }
+  adns__consistency(ads,0,cc_entex);
 
-  for (qu= ads->timew; qu; qu= nqu) {
-    nqu= qu->next;
-    if (timercmp(&now,qu->timeout,>)) {
-      DLIST_UNLINK(ads->timew,qu);
-      if (qu->nextudpserver == -1) {
-	query_fail(ads,qu,adns_s_notresponding);
-      } else {
-	DLIST_LINKTAIL(ads->tosend,qu);
-      }
-    } else {
-      inter_maxtoabs(tv_io,tvbuf,now,qu->timeout);
-    }
+  if (tv_mod && (!*tv_mod || (*tv_mod)->tv_sec || (*tv_mod)->tv_usec)) {
+    /* The caller is planning to sleep. */
+    adns__must_gettimeofday(ads,&now,&tv_nowbuf);
+    if (!now) { inter_immed(tv_mod,tv_tobuf); goto xit; }
+    adns__timeouts(ads, 0, tv_mod,tv_tobuf, *now);
   }
-  
-  for (qu= ads->tosend; qu; qu= nqu) {
-    nqu= qu->next;
-    quproc_tosend(ads,qu,now);
+
+  npollfds= adns__pollfds(ads,pollfds);
+  maxfd= *maxfd_io;
+  for (i=0; i<npollfds; i++) {
+    fd= pollfds[i].fd;
+    if (fd >= maxfd) maxfd= fd+1;
+    if (pollfds[i].events & POLLIN) FD_SET(fd,readfds_io);
+    if (pollfds[i].events & POLLOUT) FD_SET(fd,writefds_io);
+    if (pollfds[i].events & POLLPRI) FD_SET(fd,exceptfds_io);
   }
+  *maxfd_io= maxfd;
+
+xit:
+  adns__consistency(ads,0,cc_entex);
+}
 
-  inter_addfd(maxfd,readfds,ads->udpsocket);
+void adns_afterselect(adns_state ads, int maxfd, const fd_set *readfds,
+		      const fd_set *writefds, const fd_set *exceptfds,
+		      const struct timeval *now) {
+  struct timeval tv_buf;
+  struct pollfd pollfds[MAX_POLLFDS];
+  int npollfds, i;
 
+  adns__consistency(ads,0,cc_entex);
+  adns__must_gettimeofday(ads,&now,&tv_buf);
+  if (!now) goto xit;
+  adns_processtimeouts(ads,now);
+
+  npollfds= adns__pollfds(ads,pollfds);
+  for (i=0; i<npollfds; i++) pollfds[i].revents= POLLIN|POLLOUT|POLLPRI;
+  adns__fdevents(ads,
+		 pollfds,npollfds,
+		 maxfd,readfds,writefds,exceptfds,
+		 *now, 0);
+xit:
+  adns__consistency(ads,0,cc_entex);
+}
+
+/* General helpful functions. */
+
+void adns_globalsystemfailure(adns_state ads) {
+  adns__consistency(ads,0,cc_entex);
+
+  while (ads->udpw.head) adns__query_fail(ads->udpw.head, adns_s_systemfail);
+  while (ads->tcpw.head) adns__query_fail(ads->tcpw.head, adns_s_systemfail);
+  
   switch (ads->tcpstate) {
-  case server_disc:
-    break;
   case server_connecting:
-    inter_addfd(maxfd,writefds,ads->tcpsocket);
+  case server_ok:
+    adns__tcp_broken(ads,0,0);
+    break;
+  case server_disconnected:
+  case server_broken:
     break;
-  case server_connected:
-    inter_addfd(maxfd,readfds,ads->tcpsocket);
-    inter_addfd(maxfd,exceptfds,ads->tcpsocket);
-    if (ads->opbufused) inter_addfd(maxfd,writefds,ads->tcpsocket);
   default:
     abort();
   }
+  adns__consistency(ads,0,cc_entex);
 }
 
-static int internal_check(adns_state ads,
-			  adns_query *query_io,
-			  adns_answer **answer,
-			  void **context_r) {
+int adns_processany(adns_state ads) {
+  int r, i;
+  struct timeval now;
+  struct pollfd pollfds[MAX_POLLFDS];
+  int npollfds;
+
+  adns__consistency(ads,0,cc_entex);
+
+  r= gettimeofday(&now,0);
+  if (!r) adns_processtimeouts(ads,&now);
+
+  /* We just use adns__fdevents to loop over the fd's trying them.
+   * This seems more sensible than calling select, since we're most
+   * likely just to want to do a read on one or two fds anyway.
+   */
+  npollfds= adns__pollfds(ads,pollfds);
+  for (i=0; i<npollfds; i++) pollfds[i].revents= pollfds[i].events & ~POLLPRI;
+  adns__fdevents(ads,
+		 pollfds,npollfds,
+		 0,0,0,0,
+		 now,&r);
+
+  adns__consistency(ads,0,cc_entex);
+  return 0;
+}
+
+void adns__autosys(adns_state ads, struct timeval now) {
+  if (ads->iflags & adns_if_noautosys) return;
+  adns_processany(ads);
+}
+
+int adns__internal_check(adns_state ads,
+			 adns_query *query_io,
+			 adns_answer **answer,
+			 void **context_r) {
   adns_query qu;
 
   qu= *query_io;
   if (!qu) {
-    if (!ads->output.head) return EWOULDBLOCK;
-    qu= ads->output.head;
+    if (ads->output.head) {
+      qu= ads->output.head;
+    } else if (ads->udpw.head || ads->tcpw.head) {
+      return EAGAIN;
+    } else {
+      return ESRCH;
+    }
   } else {
-    if (qu->id>=0) return EWOULDBLOCK;
+    if (qu->id>=0) return EAGAIN;
   }
   LIST_UNLINK(ads->output,qu);
   *answer= qu->answer;
-  if (context_r) *context_r= qu->context;
+  if (context_r) *context_r= qu->ctx.ext;
+  *query_io= qu;
   free(qu);
   return 0;
 }
@@ -288,27 +695,47 @@ int adns_wait(adns_state ads,
 	      adns_query *query_io,
 	      adns_answer **answer_r,
 	      void **context_r) {
-  int r, maxfd, rsel, rcb;
+  int r, maxfd, rsel;
   fd_set readfds, writefds, exceptfds;
   struct timeval tvbuf, *tvp;
   
+  adns__consistency(ads,*query_io,cc_entex);
   for (;;) {
-    r= internal_check(ads,query_io,answer_r,context_r);
-    if (r && r != EWOULDBLOCK) return r;
+    r= adns__internal_check(ads,query_io,answer_r,context_r);
+    if (r != EAGAIN) break;
     maxfd= 0; tvp= 0;
     FD_ZERO(&readfds); FD_ZERO(&writefds); FD_ZERO(&exceptfds);
-    adns_interest(ads,&maxfd,&readfds,&writefds,&exceptfds,&tvp,&tvbuf);
+    adns_beforeselect(ads,&maxfd,&readfds,&writefds,&exceptfds,&tvp,&tvbuf,0);
+    assert(tvp);
     rsel= select(maxfd,&readfds,&writefds,&exceptfds,tvp);
-    if (rsel==-1) return r;
-    rcb= adns_callback(ads,maxfd,&readfds,&writefds,&exceptfds);
-    assert(rcb==rsel);
+    if (rsel==-1) {
+      if (errno == EINTR) {
+	if (ads->iflags & adns_if_eintr) { r= EINTR; break; }
+      } else {
+	adns__diag(ads,-1,0,"select failed in wait: %s",strerror(errno));
+	adns_globalsystemfailure(ads);
+      }
+    } else {
+      assert(rsel >= 0);
+      adns_afterselect(ads,maxfd,&readfds,&writefds,&exceptfds,0);
+    }
   }
+  adns__consistency(ads,0,cc_entex);
+  return r;
 }
 
 int adns_check(adns_state ads,
 	       adns_query *query_io,
 	       adns_answer **answer_r,
 	       void **context_r) {
-  autosys(ads);
-  return internal_check(ads,query_io,answer_r,context_r);
+  struct timeval now;
+  int r;
+  
+  adns__consistency(ads,*query_io,cc_entex);
+  r= gettimeofday(&now,0);
+  if (!r) adns__autosys(ads,now);
+
+  r= adns__internal_check(ads,query_io,answer_r,context_r);
+  adns__consistency(ads,0,cc_entex);
+  return r;
 }