ipif: use tun, not slip; compiles, untested

1 files changed, 226 insertions, 219 deletions

diff --git a/ipif/service.c b/ipif/service.c
index 08b0ed9..0a29f06 100644
--- a/ipif/service.c
+++ b/ipif/service.c
@@ -121,11 +121,23 @@
 #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 50
 #define ATXTLEN 16
@@ -155,53 +167,11 @@ static struct pplace {
 } *cpplace;
 
 
-static int slpipe[2], ptmaster, undoslattach;
-static const char *ifname;
-static const char *ptyname;
-
-#define NPIDS 4
-
-static union {
-  struct { pid_t sl, cout, cin, task; } byname;
-  pid_t bynumber[NPIDS];
-} pids;
-sigset_t emptyset, fullset;
+static int tunfd;
+static char *ifname;
 
 
-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);
 }
 
@@ -656,195 +626,73 @@ 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,WNOHANG);
+    if (pidr!=(pid_t)-1) break;
+    if (errno==EINTR) continue;
+    sysfatal("waitpid for task");
   }
-}
+  assert(pidr==pid);
 
-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);
-      }
-    }
-  } else {
+  if (WIFEXITED(status)) {
+    fprintf(stderr,
+	    "userv-ipif service: %s unexpectedly exited with exit status %d\n",
+	    desc, WEXITSTATUS(status));
+  } else if (WIFSIGNALED(status)) {
     fprintf(stderr,
-	    "userv-ipif service: received signal %d, terminating\n",
-	    signum);
+	    "userv-ipif service: %s unexpectedly killed by signal %s%s\n",
+	    desc, strsignal(WTERMSIG(status)),
+	    WCOREDUMP(status) ? " (core dumped)" : "");
+  } else {
+    fprintf(stderr, "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,
@@ -854,7 +702,7 @@ static void netconfigure(void) {
   }
 
   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);
@@ -863,18 +711,175 @@ 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) {
+  for (;;) {
+    int r= write(tunfd, packet, len);
+    if (r<0) {
+      if (errno==EINTR) continue;
+      if (errno==EAGAIN) 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;
+  }
+
+  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;
   }
-  pids.byname.cout= makesubproc(cout_entry);
+  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;
 
-  for (;;) sigsuspend(&emptyset);
+  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.
+   */
+
+  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(r < 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) {
@@ -883,8 +888,10 @@ int main(int argc, const char *const *argv) {
   checkpermit();
   if (!proto) dumpdebug();
 
-  startup();
-  startslattach();
+  createif();
   netconfigure();
+  setnonblock(tunfd);
+  setnonblock(0);
+  setnonblock(1);
   copydata();
 }