X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=userv-utils.git;a=blobdiff_plain;f=ipif%2Fservice.c;h=b9e40e29fcdb3a2f26ccabf02122d49ab230d388;hp=0050dd2020078a7f189506ce11283143b5175e3b;hb=44a77f4851d3c819f9b364018a9695f332758a71;hpb=2ed30784e2e8c13fc2ef14c6c94e1e1f3899ddb7

diff --git a/ipif/service.c b/ipif/service.c
index 0050dd2..b9e40e2 100644
--- a/ipif/service.c
+++ b/ipif/service.c
@@ -1,35 +1,114 @@
 /*
- * userv service (or standalone program)
- * for per-user IP subranges.
+ * userv service (or standalone program) for per-user IP subranges.
+ *
+ * When invoked appropriately, it creates a point-to-point network
+ * interface with specified parameters.  It arranges for packets sent out
+ * via that interface by the kernel to appear on its own stdout in SLIP or
+ * CSLIP encoding, and packets injected into its own stdin to be given to
+ * the kernel as if received on that interface.  Optionally, additional
+ * routes can be set up to arrange for traffic for other address ranges to
+ * be routed through the new interface.
+ *
+ * This is the service program, which is invoked as root from userv (or may
+ * be invoked firectly).
+ *
+ * Its arguments are supposed to be, in order, as follows:
+ *
+ *  The first two arguments are usually supplied by the userv
+ *  configuration.  See the file `ipif/ipif' in the source tree, which
+ *  is installed in /etc/userv/services.d/ipif by `make install':
  *
- * This is invoked as root, directly from userv.
- * Its arguments are supposed to be, in order:
  *  <config>
- *      Specifies address ranges and gids which own them.
- *  --  Indicates that the remaining arguments are user-supplied
- *      and therefore untrusted.
+ *
+ *      Specifies address ranges and gids which own them.  The default
+ *      configuration supplies /etc/userv/ipif-networks, which is then read
+ *      for a list of entries, one per line.
+ *
+ *  --
+ *      Serves to separate the user-supplied and therefore untrusted
+ *      arguments from the trusted first argument.
+ *
+ *  The remaining arguments are supplied by the (untrusted) caller:
+ *
  *  <local-addr>,<peer-addr>,<mtu>,<proto>
- *      As for slattach.  Supported protocols are slip, cslip, and
- *      adaptive.  Alternatively, set to `debug' to print debugging
- *      info.  <local-addr> is address of the interface on chiark;
- *      <peer-addr> is the address of the point-to-point peer.
+ *
+ *      As for slattach.  The only supported protocol is slip.
+ *      Alternatively, set to `debug' to print debugging info and
+ *      exit.  <local-addr> is address of the interface to be created
+ *      on the local system; <peer-addr> is the address of the
+ *      point-to-point peer.  They must be actual addresses (not
+ *      hostnames).
+ *
  *  <prefix>/<mask>,<prefix>/<mask>,...
- *      List of additional routes to add for this interface.
- *      May be the empty argument, or `-' if this is problematic.
  *
- * <config> is either
- *    <gid>,<prefix>/<len>[,<junk>]
- *      indicating that that gid may allocate addresses in
- *      the relevant subspace (<junk> is ignored)
- * or #...
- *      which is a comment
- * or /<config-file-name> or ./<config-file-name> or ../<config-file-name>
- *      which refers to a file which contains lines which
- *      are each <config>
- * or *
- *      which means that anything is permitted
- * 
- * Should be run from userv with no-disconnect-hup.
+ *      List of additional routes to add for this interface.  routes will
+ *      be set up on the local system arranging for packets for those
+ *      networks to be sent via the created interface.  <prefix> must be an
+ *      IPv4 address, and mask must be an integer (dotted-quad masks are
+ *      not supported).  If no additional routes are to be set up, use `-'
+ *      or supply an empty argument.
+ *
+ * Each <config> item - whether a line in a file such as
+ * /etc/userv/ipif-networks, or the single trusted argument supplied
+ * on the service program command line - is one of:
+ *
+ *   /<config-file-name>
+ *   ./<config-file-name>
+ *   ../<config-file-name>
+ *
+ *      Reads a file which contains lines which are each <config>
+ *      items.
+ *
+ *   <gid>,[=][-|+]<prefix>/<len>(-|+<prefix>/<len>...)[,<junk>]
+ *
+ *      Indicates that <gid> may allocate addresses in the relevant address
+ *      range (<junk> is ignored).  <gid> must be numeric.  To specify a
+ *      single host address, you must specify a mask of /32.  If `=' is
+ *      specified then the specific subrange is only allowed for the local
+ *      endpoint address, but not for remote addresses.
+ *
+ *      More than one range may be given, with each range prefixed
+ *      by + or -.  In this case each address range in the rule will
+ *      scanned in order, and the first range in the rule that matches
+ *      any desired rule will count: if that first matching range is
+ *      prefixed by `+' (or nothing) then the rule applies, if it
+ *      is prefixed by `-' (or nothing matches), the rule does not.
+ *
+ *   *
+ *      Means that anything is to be permitted.  This should not appear in
+ *      /etc/userv/ipif-networks, as that would permit any user on the
+ *      system to create any interfaces with any addresses and routes
+ *      attached.  It is provided so that root can usefully invoke the ipif
+ *      service program directly (not via userv), without needing to set up
+ *      permissions in /etc/userv/ipif-networks.
+ *
+ *   #...
+ *
+ *      Comment.  Blank lines are also ignored.
+ *
+ *   NB: Permission is granted if _any_ config entry matches the request.
+ *
+ * The service program should be run from userv with no-disconnect-hup.
+ */
+/*
+ * Copyright (C) 1999-2000,2003 Ian Jackson
+ * This file is part of ipif, part of userv-utils
+ *
+ * This 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 of the License, 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 userv-utils; if not, write to the Free Software
+ * Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * $Id$
  */
 
 #include <stdio.h>
@@ -42,26 +121,39 @@
 #include <limits.h>
 #include <signal.h>
 #include <unistd.h>
+#include <stdint.h>
+#include <poll.h>
 
 #include <sys/types.h>
 #include <sys/wait.h>
 #include <sys/stat.h>
 
+#include <sys/types.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+
+#include <sys/stat.h>
+#include <fcntl.h>
+
+#include <linux/if.h>
+#include <linux/if_tun.h>
+
 #define NARGS 4
-#define MAXEXROUTES 5
-#define ATXTLEN 12
+#define MAXEXROUTES 50
+#define ATXTLEN 16
 
 static const unsigned long gidmaxval= (unsigned long)((gid_t)-2);
-static const char *const protos_ok[]= { "slip", "cslip", "adaptive", 0 };
+static const char *const protos_ok[]= { "slip", 0 };
 static const int signals[]= { SIGHUP, SIGINT, SIGTERM, 0 };
 
 static const char *configstr, *proto;
 static unsigned long localaddr, peeraddr, mtu;
+static int localpming, peerpming;
 static int localallow, peerallow, allallow;
 static int nexroutes;
 static struct exroute {
   unsigned long prefix, mask;
-  int allow;
+  int allow, pming;
   char prefixtxt[ATXTLEN], masktxt[ATXTLEN];
 } exroutes[MAXEXROUTES];
 
@@ -75,53 +167,11 @@ static struct pplace {
 } *cpplace;
 
 
-static int slpipe[2], ptmaster, undoslattach;
-static const char *ifname;
-static const char *ptyname;
+static int tunfd;
+static char *ifname;
 
-#define NPIDS 4
-
-static union {
-  struct { pid_t sl, cout, cin, task; } byname;
-  pid_t bynumber[NPIDS];
-} pids;
-sigset_t emptyset, fullset;
-
-
-static int cleantask(void) {
-  pid_t pid;
-
-  pid= fork();
-  if (!pid) return 1;
-  if (pid == (pid_t)-1)
-    perror("userv-ipif: fork for undo slattach failed - cannot clean up properly");
-  return 0;
-}
 
 static void terminate(int estatus) {
-  int i, status;
-  pid_t pid;
-  
-  for (i=0; i<NPIDS; i++)
-    if (pids.bynumber[i]) kill(pids.bynumber[i], SIGTERM);
-
-  if (undoslattach) {
-    if (cleantask()) {
-      execlp("slattach", "slattach", "-p", "tty", ptyname, (char*)0);
-      perror("userv-ipif: exec slattach for undo slattach failed");
-      exit(-1);
-    }
-    if (ifname && cleantask()) {
-      execlp("ifconfig", "ifconfig", ifname, "down", (char*)0);
-      perror("userv-ipif: exec ifconfig for undo ifconfig failed");
-      exit(-1);
-    }
-  }
-
-  for (;;) {
-    pid= waitpid(-1,&status,0);
-    if (pid == (pid_t)-1) break;
-  }
   exit(estatus);
 }
 
@@ -219,12 +269,17 @@ static unsigned long eat_number(const char **argp, const char *what,
   return rv;
 }
 
-static void addrnet_mustdiffer(const char *w1, unsigned long p1, unsigned long m1,
-			       const char *w2, unsigned long p2, unsigned long m2) {
+static int addrnet_overlap(unsigned long p1, unsigned long m1,
+			   unsigned long p2, unsigned long m2) {
   unsigned long mask;
 
   mask= m1&m2;
-  if ((p1 & mask) != (p2 & mask)) return;
+  return (p1 & mask) == (p2 & mask);
+}
+
+static void addrnet_mustdiffer(const char *w1, unsigned long p1, unsigned long m1,
+			       const char *w2, unsigned long p2, unsigned long m2) {
+  if (!addrnet_overlap(p1,m1,p2,m2)) return;
   badusage("%s %08lx/%08lx overlaps/clashes with %s %08lx/%08lx",
 	   w1,p1,m1, w2,p2,m2);
 }
@@ -258,7 +313,7 @@ static void eat_prefixmask(const char **argp, const char *what,
   sprintf(whatbuf,"%s length",what);
   len= eat_number(argp,whatbuf, 0,32, endchars,endchar_r);
 
-  mask= (~0UL << (32-len));
+  mask= len ? (~0UL << (32-len)) : 0UL;
   if (prefix & ~mask) badusage("%s prefix %08lx not fully contained in mask %08lx",
 			       what,prefix,mask);
   *prefix_r= prefix;
@@ -270,32 +325,66 @@ static int addrnet_isin(unsigned long prefix, unsigned long mask,
 			unsigned long mprefix, unsigned long mmask) {
   return  !(~mask & mmask)  &&  (prefix & mmask) == mprefix;
 }
-  
 
-static void permit(unsigned long pprefix, unsigned long pmask) {
-  int i, any;
+/* Totally hideous algorithm for parsing the config file lines.
+ * For each config file line, we first see if its gid applies.  If not
+ * we skip it.  Otherwise, we do
+ *  permit_begin
+ *     which sets <foo>pming to 1
+ * for each range.  <foo>pming may be 0 if we've determined that
+ * this line does not apply to <foo>.
+ *  permit_range
+ *     which calls permit_range_thing for each <foo>
+ *        which checks to see if <foo> is inside the relevant
+ *        range (for +) or overlaps it (for -) and updates
+ *        <foo>allow and <foo>pming.
+ */
+
+static void permit_begin(void) {
+  int i;
   
-  assert(!(pprefix & ~pmask));
+  localpming= peerpming= 1;
+  for (i=0; i<nexroutes; i++) exroutes[i].pming= 1;
+}
 
-  if (!proto) fputs("permits",stdout);
-  if (addrnet_isin(localaddr,~0UL, pprefix,pmask)) {
-    if (!proto) fputs(" local-addr",stdout);
-    any= localallow= 1;
-  }
-  if (addrnet_isin(peeraddr,~0UL, pprefix,pmask)) {
-    if (!proto) fputs(" peer-addr",stdout);
-    any= peerallow= 1;
+static void permit_range_thing(unsigned long tprefix, unsigned long tmask,
+			       const char *what, int *tallow, int *tpming,
+			       unsigned long pprefix, unsigned long pmask,
+			       int plus, int *any) {
+  if (plus) {
+    if (!addrnet_isin(tprefix,tmask, pprefix,pmask)) return;
+    if (*tpming) *tallow= 1;
+  } else {
+    if (!addrnet_overlap(tprefix,tmask, pprefix,pmask)) return;
+    *tpming= 0;
   }
-  for (i=0; i<nexroutes; i++) {
-    if (addrnet_isin(exroutes[i].prefix,exroutes[i].mask, pprefix,pmask)) {
-      if (!proto) printf(" route#%d",i);
-      any= exroutes[i].allow= 1;
+  if (!proto) printf(" %c%s", plus?'+':'-', what);
+  *any= 1;
+}
+
+static void permit_range(unsigned long prefix, unsigned long mask,
+			 int plus, int localonly) {
+  int i, any;
+  char idbuf[40];
+  
+  assert(!(prefix & ~mask));
+  any= 0;
+
+  permit_range_thing(localaddr,~0UL,"local", &localallow,&localpming,
+		     prefix,mask, plus,&any);
+
+  if (!localonly) {
+    permit_range_thing(peeraddr,~0UL, "peer-addr", &peerallow,&peerpming,
+		       prefix,mask, plus,&any);
+    for (i=0; i<nexroutes; i++) {
+      sprintf(idbuf,"route#%d",i);
+      permit_range_thing(exroutes[i].prefix,exroutes[i].mask, idbuf,
+			 &exroutes[i].allow,&exroutes[i].pming,
+			 prefix,mask, plus,&any);
     }
   }
-  if (!proto) {
-    if (!any) fputs(" nothing!",stderr);
-    putchar('\n');
-  }
+  if (!proto)
+    if (!any) fputs(" nothing",stdout);
 }
 
 static void pconfig(const char *configstr, int truncated);
@@ -347,14 +436,15 @@ static void pfile(const char *filename) {
 
 static void pconfig(const char *configstr, int truncated) {
   unsigned long fgid, tgid, pprefix, pmask;
-  int plen;
+  int plen, localonly, plus, rangeix, delim;
   char ptxt[ATXTLEN];
+  char whattxt[100];
   const char *gidlist;
   
   switch (configstr[0]) {
   case '*':
-    if (strcmp(configstr,"*")) badusage("`*' directive must be only thing on line");
-    permit(0UL,0UL);
+    permit_begin();
+    permit_range(0UL,0UL,1,0);
     return;
     
   case '#':
@@ -370,22 +460,57 @@ static void pconfig(const char *configstr, int truncated) {
       badusage("unknown configuration directive");
     
     fgid= eat_number(&configstr,"gid", 0,gidmaxval, ",",0);
-    eat_prefixmask(&configstr,"permitted-prefix", ",",0, &pprefix,&pmask,&plen);
-    if (!configstr && truncated) badusage("gid,prefix/len,... spec too long");
 
-    if (!proto) printf(" %5lu,%s/%d: ", fgid, ip2txt(pprefix,ptxt), plen);
+    if (!proto) printf(" %5lu", fgid);
 
     gidlist= getenv("USERV_GID");
     if (!gidlist) fatal("USERV_GID not set");
     for (;;) {
       if (!gidlist) {
-	if (!proto) printf("no matching gid\n");
+	if (!proto) printf(" no matching gid\n");
 	return;
       }
       tgid= eat_number(&gidlist,"userv-gid", 0,gidmaxval, " ",0);
       if (tgid == fgid) break;
     }
-    permit(pprefix,pmask);
+
+    if (configstr[0] == '=') {
+      localonly= 1;
+      configstr++;
+    } else {
+      localonly= 0;
+    }
+
+    permit_begin();
+
+    rangeix= 0;
+    plus= 1;
+    switch (configstr[0]) {
+    case '-': plus= 0;     /* fall through */
+    case '+': configstr++;
+    default:;
+    }
+
+    for (;;) {
+      sprintf(whattxt, "%s-prefix#%d",
+	      plus ? "permitted" : "notpermitted",
+	      rangeix);
+      eat_prefixmask(&configstr,whattxt, ",+-",&delim,
+		     &pprefix,&pmask,&plen);
+      if (!configstr && truncated)
+	badusage("gid,prefix/len,... spec too long");
+
+      if (!proto)
+	printf("  %c%s/%d:", plus?'+':'-',ip2txt(pprefix,ptxt), plen);
+
+      permit_range(pprefix,pmask,plus,localonly);
+      if (delim==',') break;
+
+      plus= delim=='-' ? 0 : 1;
+      rangeix++;
+    }
+
+    putchar('\n');
     return;
   }
 }
@@ -501,205 +626,82 @@ static void dumpdebug(void) {
 }
 
 
-static void setsigmask(const sigset_t *ss) {
-  int r;
-  
-  r= sigprocmask(SIG_SETMASK, ss, 0);
-  if (r) sysfatal("[un]block signals");
-}  
-
-static void setsignals(void (*handler)(int), struct sigaction *sa, int chldflags) {
-  const int *signalp;
-  int r, sig;
-  
-  sa->sa_handler= handler;
-  sa->sa_flags= 0; 
-  for (signalp=signals; (sig=*signalp); signalp++) {
-    r= sigaction(sig, sa, 0);  if (r) sysfatal("uncatch signal");
-  }
-  sa->sa_flags= chldflags;
-  r= sigaction(SIGCHLD, sa, 0);  if (r) sysfatal("uncatch children");
-}
-
-static void infork(void) {
-  struct sigaction sa;
-
-  memset(&pids,0,sizeof(pids));
-  sigemptyset(&sa.sa_mask);
-  setsignals(SIG_DFL,&sa,0);
-  setsigmask(&emptyset);
-  undoslattach= 0;
-}
-
-static pid_t makesubproc(void (*entry)(void)) {
-  pid_t pid;
-
-  pid= fork();  if (pid == (pid_t)-1) sysfatal("fork for subprocess");
-  if (pid) return pid;
-
-  infork();
-  entry();
-  abort();
-}
-
-static int task(void) {
-  pid_t pid;
+static int task(const char *desc) {
+  pid_t pid, pidr;
+  int status;
 
   pid= fork();
   if (pid == (pid_t)-1) sysfatal("fork for task");
-  if (!pid) { infork(); return 1; }
-
-  pids.byname.task= pid;
-  while (pids.byname.task) sigsuspend(&emptyset);
-  return 0;
-}
-
-static void mdup2(int fd1, int fd2, const char *what) {
-  int r;
+  if (!pid) return 1;
 
   for (;;) {
-    r= dup2(fd1,fd2); if (r==fd2) return;
-    if (r!=-1) fatal("dup2 in %s gave wrong answer %d instead of %d",what,r,fd2);
-    if (errno != EINTR) sysfatal("dup2 failed in %s",what);
+    pidr= waitpid(pid,&status,0);
+    if (pidr!=(pid_t)-1) break;
+    if (errno==EINTR) continue;
+    sysfatal("waitpid for task");
   }
-}
-
-static void sl_entry(void) {
-  mdup2(slpipe[1],1,"slattach child");
-  execlp("slattach", "slattach", "-v", "-L", "-p",proto, ptyname, (char*)0);
-  sysfatal("cannot exec slattach");
-}
-
-static void cin_entry(void) {
-  mdup2(ptmaster,1,"cat input child");
-  execlp("cat", "cat", (char*)0);
-  sysfatal("cannot exec cat input");
-}
-
-static void cout_entry(void) {
-  mdup2(ptmaster,0,"cat output child");
-  execlp("cat", "cat", (char*)0);
-  sysfatal("cannot exec cat output");
-}
-
-static void sighandler(int signum) {
-  pid_t pid;
-  int estatus, status;
-  const char *taskfail;
-
-  estatus= 4;
-  
-  if (signum == SIGCHLD) {
-    for (;;) {
-      pid= waitpid(-1,&status,WNOHANG);
-      if (!pid || pid == (pid_t)-1) return;
-
-      if (pid == pids.byname.task) {
-	pids.byname.task= 0;
-	if (!status) return;
-	taskfail= "task";
-      } else if (pid == pids.byname.cin) {
-	pids.byname.cin= 0;
-	if (status) {
-	  taskfail= "input cat";
-	} else {
-	  taskfail= 0;
-	  estatus= 0;
-	}
-      } else if (pid == pids.byname.cout) {
-	pids.byname.cout= 0;
-	taskfail= "output cat";
-      } else if (pid == pids.byname.sl) {
-	pids.byname.sl= 0;
-	taskfail= "slattach";
-      } else {
-	continue;
-      }
-      break;
-    }
-    if (taskfail) {
-      if (WIFEXITED(status)) {
-	fprintf(stderr,
-		"userv-ipif service: %s unexpectedly exited with exit status %d\n",
-		taskfail, WEXITSTATUS(status));
-      } else if (WIFSIGNALED(status)) {
-	fprintf(stderr,
-		"userv-ipif service: %s unexpectedly killed by signal %s%s\n",
-		taskfail, strsignal(WTERMSIG(status)),
-		WCOREDUMP(status) ? " (core dumped)" : "");
-      } else {
-	fprintf(stderr, "userv-ipif service: %s unexpectedly terminated"
-		" with unknown status code %d\n", taskfail, status);
-      }
-    }
+  assert(pidr==pid);
+
+  if (WIFEXITED(status)) {
+    if (WEXITSTATUS(status))
+      fatal("userv-ipif service: %s exited with error exit status %d\n",
+	    desc, WEXITSTATUS(status));
+  } else if (WIFSIGNALED(status)) {
+    fatal("userv-ipif service: %s died due to signal %s%s\n",
+	  desc, strsignal(WTERMSIG(status)),
+	  WCOREDUMP(status) ? " (core dumped)" : "");
   } else {
-    fprintf(stderr,
-	    "userv-ipif service: received signal %d, terminating\n",
-	    signum);
+    fatal("userv-ipif service: %s unexpectedly terminated"
+	  " with unknown status code %d\n", desc, status);
   }
 
-  terminate(estatus);
+  return 0;
 }
 
-static void startup(void) {
+static void createif(void) {
+  static const char ifnamepat[]= "userv%d";
+  struct ifreq ifr;
   int r;
-  struct sigaction sa;
-  
-  sigfillset(&fullset);
-  sigemptyset(&emptyset);
-
-  ptmaster= getpt();  if (ptmaster==-1) sysfatal("allocate pty master");
-  r= grantpt(ptmaster);  if (r) sysfatal("grab/grant pty slave");
-  ptyname= ptsname(ptmaster);  if (!ptyname) sysfatal("get pty slave name");
-  r= chmod(ptyname,0600);  if (r) sysfatal("chmod pty slave");
-  r= unlockpt(ptmaster);  if (r) sysfatal("unlock pty");
-
-  sigfillset(&sa.sa_mask);
-  setsignals(sighandler,&sa,SA_NOCLDSTOP);
-  setsigmask(&fullset);
-}
 
-static void startslattach(void) {
-  static char ifnbuf[200];
+  memset(&ifr,0,sizeof(ifr));
+  ifr.ifr_flags= IFF_TUN | IFF_NO_PI;
 
-  FILE *piper;
-  int r, l, k;
+  assert(sizeof(ifr.ifr_name) >= sizeof(ifnamepat));
+  strcpy(ifr.ifr_name, ifnamepat);
 
-  r= pipe(slpipe);  if (r) sysfatal("create pipe");
-  piper= fdopen(slpipe[0],"r");  if (!piper) sysfatal("fdopen pipe");
+  tunfd= open("/dev/net/tun", O_RDWR);
+  if (!tunfd) sysfatal("open /dev/net/tun");
 
-  undoslattach= 1;
-  pids.byname.sl= makesubproc(sl_entry);
+  r= fcntl(tunfd, F_GETFD);
+  if (r==-1) sysfatal("fcntl(tunfd,F_GETFD)");
+  r= fcntl(tunfd, F_SETFD, r|FD_CLOEXEC);
+  if (r==-1) sysfatal("fcntl(tunfd,F_SETFD,|FD_CLOEXEC)");
 
-  close(slpipe[1]);
-  setsigmask(&emptyset);
-  if (!fgets(ifnbuf,sizeof(ifnbuf),piper)) {
-    if (ferror(piper)) sysfatal("cannot read ifname from slattach");
-    else fatal("cannot read ifname from slattach");
-  }
-  setsigmask(&fullset);
-  l= strlen(ifnbuf);
-  if (l<0 || ifnbuf[l-1] != '\n') fatal("slattach gave strange output `%s'",ifnbuf);
-  ifnbuf[l-1]= 0;
-  for (k=l; k>0 && ifnbuf[k-1]!=' '; k--);
-  ifname= ifnbuf+k;
+  r= ioctl(tunfd, TUNSETIFF, (void*)&ifr);
+  if (r) sysfatal("ioctl TUNSETIFF");
+
+  /* ifr.ifr_name might not be null-terminated.  crazy abi. */
+  ifname= malloc(sizeof(ifr.ifr_name)+1);
+  if (!ifname) sysfatal("malloc for interface name");
+  memcpy(ifname, ifr.ifr_name, sizeof(ifr.ifr_name));
+  ifname[sizeof(ifr.ifr_name)]= 0;
 }
 
 static void netconfigure(void) {
   char mtutxt[100];
   int i;
 
-  if (task()) {
+  if (task("ifconfig")) {
     sprintf(mtutxt,"%lu",mtu);
   
     execlp("ifconfig", "ifconfig", ifname, localtxt,
-	   "netmask","255.255.255.255", "-broadcast", "pointopoint",peertxt,
+	   "netmask","255.255.255.255", "pointopoint",peertxt, "-broadcast",
 	   "mtu",mtutxt, "up", (char*)0);
     sysfatal("cannot exec ifconfig");
   }
 
   for (i=0; i<nexroutes; i++) {
-    if (task()) {
+    if (task("route")) {
       execlp("route","route", "add", "-net",exroutes[i].prefixtxt,
 	     "netmask",exroutes[i].masktxt,
 	     "gw",peertxt, "dev",ifname, (char*)0);
@@ -708,18 +710,184 @@ static void netconfigure(void) {
   }
 }
 
-static void copydata(void) __attribute__((noreturn));
-static void copydata(void) {
+static void setnonblock(int fd) {
   int r;
+  r= fcntl(fd,F_GETFL); 
+  if (r==-1) sysfatal("fcntl F_GETFL");
+  r= fcntl(fd,F_SETFL, r|O_NONBLOCK);
+  if (r==-1) sysfatal("fcntl F_SETFL O_NONBLOCK");
+}
+
+static void rx_packet(const uint8_t *packet, int len) {
+  if (!len)
+    return;
+  for (;;) {
+    int r= write(tunfd, packet, len);
+    if (r<0) {
+      if (errno==EINTR) continue;
+      if (errno==EAGAIN || errno==ENOMEM) return; /* oh well */
+      sysfatal("error writing packet to tun (transmitting)");
+    }
+    assert(r==len);
+    return;
+  }
+}
+
+static int output_waiting, input_waiting;
+
+#define SLIP_END     0300
+#define SLIP_ESC     0333
+#define SLIP_ESC_END 0334
+#define SLIP_ESC_ESC 0335
+
+static void more_rx_data(uint8_t *input_buf, uint8_t *output_buf) {
+  /* we make slip_data never contain continuation of a packet */
+  /* input_buf is passed as a parameter since it's in copydata's stack frame */
+  static int scanned;
+  static int output_len;
+
+  uint8_t *op= output_buf + output_len;
+  const uint8_t *ip= input_buf + scanned;
+  const uint8_t *ip_end= input_buf + input_waiting;
+  int eaten= 0;
   
-  pids.byname.cin= makesubproc(cin_entry);
   for (;;) {
-    r= write(1, "\300", 1); if (r==1) break;
-    assert(r==-1);  if (errno != EINTR) sysfatal("send initial delim to confirm");
+    if (ip>=ip_end) break;
+    uint8_t c= *ip++;
+    if (c==SLIP_END) {
+      rx_packet(output_buf, op-output_buf);
+      op= output_buf;
+      eaten= ip - input_buf;
+      continue;
+    }
+    if (c==SLIP_ESC) {
+      if (ip>=ip_end) { /* rescan this when there's more */ ip--; break; }
+      c= *ip++;
+      if (c==SLIP_ESC_END) c=SLIP_END;
+      else if (c==SLIP_ESC_ESC) c=SLIP_ESC;
+      else fatal("unexpected byte 0%o after SLIP_ESC",c);
+    }
+    if (op == output_buf+mtu)
+      fatal("SLIP packet exceeds mtu");
+    *op++= c;
   }
-  pids.byname.cout= makesubproc(cout_entry);
 
-  for (;;) sigsuspend(&emptyset);
+  output_len= op - output_buf;
+  scanned= ip - input_buf;
+
+  input_waiting -= eaten;
+  memmove(input_buf, input_buf+eaten, input_waiting);
+  scanned -= eaten;
+}
+
+static void tx_packet(uint8_t *output_buf, const uint8_t *ip, int inlen) {
+  /* output_buf is passed as a parameter since it's in copydata's stack frame */
+  assert(!output_waiting);
+  uint8_t *op= output_buf;
+
+  *op++= SLIP_END;
+  while (inlen-- >0) {
+    uint8_t c= *ip++;
+    if (c==SLIP_END) { *op++= SLIP_ESC; *op++= SLIP_ESC_END; }
+    else if (c==SLIP_ESC) { *op++= SLIP_ESC; *op++= SLIP_ESC_ESC; }
+    else *op++= c;
+  }
+  *op++= SLIP_END;
+  assert(op <= output_buf + mtu*2+2);
+
+  output_waiting= op - output_buf;
+}
+
+static void copydata(void) __attribute__((noreturn));
+static void copydata(void) {
+  uint8_t output_buf[mtu*2+2];
+  uint8_t input_buf[mtu*2+2];
+  uint8_t rx_packet_buf[mtu];
+
+  int r, i;
+
+  struct pollfd polls[3];
+  memset(polls, 0, sizeof(polls));
+  
+  polls[0].fd= 0;      polls[0].events= POLLIN;
+  polls[1].fd= 1;
+  polls[2].fd= tunfd;
+
+  /* We don't do flow control on input packets; instead, we just throw
+   * away ones which the kernel doesn't accept.  So we always poll for
+   * those.
+   *
+   * Output packets we buffer, so we poll only as appropriate for those.
+   */
+
+  /* Start by transmitting one END byte to say we're ready. */
+  output_buf[0]= SLIP_END;
+  output_waiting= 1;
+
+  for (;;) {
+    if (output_waiting) {
+      r= write(1, output_buf, output_waiting);
+      if (r<0) {
+	if (errno==EINTR) continue;
+	if (errno!=EAGAIN)
+	  sysfatal("error writing SLIP output (packets being received)");
+      } else {
+	assert(r>0);
+	output_waiting -= r;
+	memmove(output_buf, output_buf+r, output_waiting);
+      }
+    }
+    if (output_waiting) {
+      polls[1].events |= POLLOUT;
+      polls[2].events &= ~POLLIN;
+    } else {
+      polls[1].events &= ~POLLOUT;
+      polls[2].events |= POLLIN;
+    }
+    r= poll(polls,3,-1);
+
+    if (r<0) {
+      if (errno==EINTR) continue;
+      sysfatal("poll() failed");
+    }
+    assert(r>0); /* we used an infinite timeout */
+
+    for (i=0; i<sizeof(polls)/sizeof(polls[0]); i++)
+      if (polls[i].revents & ~polls[i].events)
+	fatal("unexpected revents 0x%x for fd=%d",
+	      polls[i].revents, polls[i].fd);
+
+    if (polls[0].events & POLLIN) {
+      int want= sizeof(input_buf) - input_waiting;
+      if (want<0) fatal("incoming packet necessarily exceeds MTU");
+      r= read(0, input_buf + input_waiting, want);
+      if (r>0) {
+	input_waiting += r;
+	assert(input_waiting <= sizeof(input_buf));
+	more_rx_data(input_buf, rx_packet_buf);
+      } else if (r==0) {
+	terminate(0);
+      } else {
+	if (!(errno==EINTR || errno==EAGAIN))
+	  sysfatal("error reading input SLIP data (packets to transmit)");
+      }
+    }
+
+    /* We handle what would be (polls[1].events & POLLOUT) above,
+     * unconditionally.  That eliminates the need to poll in the usual case */
+      
+    if (polls[2].events & POLLIN) {
+      uint8_t packet_buf[mtu];
+      r= read(tunfd, packet_buf, mtu);
+      if (r>0) {
+	tx_packet(output_buf, packet_buf, r);
+      } else {
+	assert(r<0);
+	if (!(errno==EAGAIN || errno==EWOULDBLOCK))
+	  sysfatal("error reading packet (being transmitted) from tun");
+      }
+    }
+  } 
 }
 
 int main(int argc, const char *const *argv) {
@@ -728,8 +896,10 @@ int main(int argc, const char *const *argv) {
   checkpermit();
   if (!proto) dumpdebug();
 
-  startup();
-  startslattach();
+  createif();
   netconfigure();
+  setnonblock(tunfd);
+  setnonblock(0);
+  setnonblock(1);
   copydata();
 }