Allow recovery from guest failure.

[libguestfs.git] / src / guestfs.c

diff --git a/src/guestfs.c b/src/guestfs.c
index c2af08c..be7b0c7 100644 (file)
--- a/src/guestfs.c
+++ b/src/guestfs.c
@@ -650,7 +650,6 @@ guestfs_launch (guestfs_h *g)
   char *kernel = NULL, *initrd = NULL;
   char unixsock[256];
   struct sockaddr_un addr;
-  struct stat statbuf;
 
   /* Configured? */
   if (!g->cmdline) {
@@ -710,39 +709,14 @@ guestfs_launch (guestfs_h *g)
     goto cleanup0;
   }
 
-  /* Choose a suitable memory size (in MB).  This is more art
-   * than science, but you can help by doing
-   *   ./configure --enable-debug-command
-   * and then running:
-   *   debug sh free
-   *   debug mem ''
-   * and seeing how much free memory is left for particular
-   * configurations.
-   *
-   * It's also helpful to report both the compressed and uncompressed
-   * size of the initramfs (ls -lh initramfs*.img; du -sh initramfs).
-   *
-   * XXX KVM virtio balloon driver?
+  /* Choose a suitable memory size.  Previously we tried to choose
+   * a minimal memory size, but this isn't really necessary since
+   * recent QEMU and KVM don't do anything nasty like locking
+   * memory into core any more.  This we can safely choose a
+   * large, generous amount of memory, and it'll just get swapped
+   * on smaller systems.
    */
-  if (stat (initrd, &statbuf) != -1) {
-    /* Approximate size of the initramfs after it is decompressed
-     * in kernel memory.  The compression factor is ~2.5-3.
-     */
-    memsize = 3 * statbuf.st_size / 1024 / 1024;
-
-    /* Approximate size used by the kernel. */
-    memsize += 10;
-
-    /* Want to give userspace some room, so: */
-    memsize += 128;
-
-#if SIZEOF_LONG == 8
-    /* On 64 bit, assume some overhead. */
-    memsize += 64;
-#endif
-  } else
-    memsize = 512;
-  
+  memsize = 384;
 
   /* Make the temporary directory containing the socket. */
   if (!g->tmpdir) {
@@ -788,7 +762,7 @@ guestfs_launch (guestfs_h *g)
 
     /* Linux kernel command line. */
     snprintf (append, sizeof append,
-             "console=ttyS0 guestfs=%s:%d%s",
+             "panic=1 console=ttyS0 guestfs=%s:%d%s",
              VMCHANNEL_ADDR, VMCHANNEL_PORT,
              g->verbose ? " guestfs_verbose=1" : "");
 
@@ -799,6 +773,7 @@ guestfs_launch (guestfs_h *g)
 #if 0
     add_cmdline (g, "-no-kqemu"); /* Avoids a warning. */
 #endif
+    add_cmdline (g, "-no-reboot"); /* Force exit instead of reboot on panic */
     add_cmdline (g, "-kernel");
     add_cmdline (g, (char *) kernel);
     add_cmdline (g, "-initrd");
@@ -1120,6 +1095,25 @@ guestfs_set_busy (guestfs_h *g)
   return 0;
 }
 
+int
+guestfs_end_busy (guestfs_h *g)
+{
+  switch (g->state)
+    {
+    case BUSY:
+      g->state = READY;
+      break;
+    case CONFIG:
+    case READY:
+      break;
+    case LAUNCHING:
+    case NO_HANDLE:
+      error (g, "guestfs_end_busy: called when in state %d", g->state);
+      return -1;
+    }
+  return 0;
+}
+
 /* Structure-freeing functions.  These rely on the fact that the
  * structure format is identical to the XDR format.  See note in
  * generator.ml.
@@ -1151,6 +1145,39 @@ guestfs_free_lvm_lv_list (struct guestfs_lvm_lv_list *x)
   free (x);
 }
 
+/* We don't know if stdout_event or sock_read_event will be the
+ * first to receive EOF if the qemu process dies.  This function
+ * has the common cleanup code for both.
+ */
+static void
+child_cleanup (guestfs_h *g)
+{
+  if (g->verbose)
+    fprintf (stderr, "stdout_event: %p: child process died\n", g);
+  /*kill (g->pid, SIGTERM);*/
+  if (g->recoverypid > 0) kill (g->recoverypid, 9);
+  waitpid (g->pid, NULL, 0);
+  if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
+  if (g->stdout_watch >= 0)
+    g->main_loop->remove_handle (g->main_loop, g, g->stdout_watch);
+  if (g->sock_watch >= 0)
+    g->main_loop->remove_handle (g->main_loop, g, g->sock_watch);
+  close (g->fd[0]);
+  close (g->fd[1]);
+  close (g->sock);
+  g->fd[0] = -1;
+  g->fd[1] = -1;
+  g->sock = -1;
+  g->pid = 0;
+  g->recoverypid = 0;
+  g->start_t = 0;
+  g->stdout_watch = -1;
+  g->sock_watch = -1;
+  g->state = CONFIG;
+  if (g->subprocess_quit_cb)
+    g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
+}
+
 /* This function is called whenever qemu prints something on stdout.
  * Qemu's stdout is also connected to the guest's serial console, so
  * we see kernel messages here too.
@@ -1177,30 +1204,7 @@ stdout_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data,
   n = read (fd, buf, sizeof buf);
   if (n == 0) {
     /* Hopefully this indicates the qemu child process has died. */
-    if (g->verbose)
-      fprintf (stderr, "stdout_event: %p: child process died\n", g);
-    /*kill (g->pid, SIGTERM);*/
-    if (g->recoverypid > 0) kill (g->recoverypid, 9);
-    waitpid (g->pid, NULL, 0);
-    if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
-    if (g->stdout_watch >= 0)
-      g->main_loop->remove_handle (g->main_loop, g, g->stdout_watch);
-    if (g->sock_watch >= 0)
-      g->main_loop->remove_handle (g->main_loop, g, g->sock_watch);
-    close (g->fd[0]);
-    close (g->fd[1]);
-    close (g->sock);
-    g->fd[0] = -1;
-    g->fd[1] = -1;
-    g->sock = -1;
-    g->pid = 0;
-    g->recoverypid = 0;
-    g->start_t = 0;
-    g->stdout_watch = -1;
-    g->sock_watch = -1;
-    g->state = CONFIG;
-    if (g->subprocess_quit_cb)
-      g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
+    child_cleanup (g);
     return;
   }
 
@@ -1246,11 +1250,11 @@ sock_read_event (struct guestfs_main_loop *ml, guestfs_h *g, void *data,
   }
   n = read (g->sock, g->msg_in + g->msg_in_size,
            g->msg_in_allocated - g->msg_in_size);
-  if (n == 0)
-    /* Disconnected?  Ignore it because stdout_watch will get called
-     * and will do the cleanup.
-     */
+  if (n == 0) {
+    /* Disconnected. */
+    child_cleanup (g);
     return;
+  }
 
   if (n == -1) {
     if (errno != EINTR && errno != EAGAIN)