+ if (access (filename, F_OK) == -1) {
+ perrorf (g, "%s", filename);
+ return -1;
+ }
+
+ return guestfs__config (g, "-cdrom", filename);
+}
+
+/* Returns true iff file is contained in dir. */
+static int
+dir_contains_file (const char *dir, const char *file)
+{
+ int dirlen = strlen (dir);
+ int filelen = strlen (file);
+ int len = dirlen+filelen+2;
+ char path[len];
+
+ snprintf (path, len, "%s/%s", dir, file);
+ return access (path, F_OK) == 0;
+}
+
+/* Returns true iff every listed file is contained in 'dir'. */
+static int
+dir_contains_files (const char *dir, ...)
+{
+ va_list args;
+ const char *file;
+
+ va_start (args, dir);
+ while ((file = va_arg (args, const char *)) != NULL) {
+ if (!dir_contains_file (dir, file)) {
+ va_end (args);
+ return 0;
+ }
+ }
+ va_end (args);
+ return 1;
+}
+
+static void print_timestamped_message (guestfs_h *g, const char *fs, ...);
+static int build_supermin_appliance (guestfs_h *g, const char *path, char **kernel, char **initrd);
+static int test_qemu (guestfs_h *g);
+static int qemu_supports (guestfs_h *g, const char *option);
+static void print_cmdline (guestfs_h *g);
+
+static const char *kernel_name = "vmlinuz." REPO "." host_cpu;
+static const char *initrd_name = "initramfs." REPO "." host_cpu ".img";
+static const char *supermin_name =
+ "initramfs." REPO "." host_cpu ".supermin.img";
+static const char *supermin_hostfiles_name =
+ "initramfs." REPO "." host_cpu ".supermin.hostfiles";
+
+int
+guestfs__launch (guestfs_h *g)
+{
+ const char *tmpdir;
+ char dir_template[PATH_MAX];
+ int r, pmore;
+ size_t len;
+ int wfd[2], rfd[2];
+ int tries;
+ char *path, *pelem, *pend;
+ char *kernel = NULL, *initrd = NULL;
+ int null_vmchannel_sock;
+ char unixsock[256];
+ struct sockaddr_un addr;
+
+ /* Start the clock ... */
+ gettimeofday (&g->launch_t, NULL);
+
+#ifdef P_tmpdir
+ tmpdir = P_tmpdir;
+#else
+ tmpdir = "/tmp";
+#endif
+
+ tmpdir = getenv ("TMPDIR") ? : tmpdir;
+ snprintf (dir_template, sizeof dir_template, "%s/libguestfsXXXXXX", tmpdir);
+
+ /* Configured? */
+ if (!g->cmdline) {
+ error (g, _("you must call guestfs_add_drive before guestfs_launch"));
+ return -1;
+ }
+
+ if (g->state != CONFIG) {
+ error (g, _("qemu has already been launched"));
+ return -1;
+ }
+
+ /* Make the temporary directory. */
+ if (!g->tmpdir) {
+ g->tmpdir = safe_strdup (g, dir_template);
+ if (mkdtemp (g->tmpdir) == NULL) {
+ perrorf (g, _("%s: cannot create temporary directory"), dir_template);
+ goto cleanup0;
+ }
+ }
+
+ /* First search g->path for the supermin appliance, and try to
+ * synthesize a kernel and initrd from that. If it fails, we
+ * try the path search again looking for a backup ordinary
+ * appliance.
+ */
+ pelem = path = safe_strdup (g, g->path);
+ do {
+ pend = strchrnul (pelem, ':');
+ pmore = *pend == ':';
+ *pend = '\0';
+ len = pend - pelem;
+
+ /* Empty element of "." means cwd. */
+ if (len == 0 || (len == 1 && *pelem == '.')) {
+ if (g->verbose)
+ fprintf (stderr,
+ "looking for supermin appliance in current directory\n");
+ if (dir_contains_files (".",
+ supermin_name, supermin_hostfiles_name,
+ "kmod.whitelist", NULL)) {
+ if (build_supermin_appliance (g, ".", &kernel, &initrd) == -1)
+ return -1;
+ break;
+ }
+ }
+ /* Look at <path>/supermin* etc. */
+ else {
+ if (g->verbose)
+ fprintf (stderr, "looking for supermin appliance in %s\n", pelem);
+
+ if (dir_contains_files (pelem,
+ supermin_name, supermin_hostfiles_name,
+ "kmod.whitelist", NULL)) {
+ if (build_supermin_appliance (g, pelem, &kernel, &initrd) == -1)
+ return -1;
+ break;
+ }
+ }
+
+ pelem = pend + 1;
+ } while (pmore);
+
+ free (path);
+
+ if (kernel == NULL || initrd == NULL) {
+ /* Search g->path for the kernel and initrd. */
+ pelem = path = safe_strdup (g, g->path);
+ do {
+ pend = strchrnul (pelem, ':');
+ pmore = *pend == ':';
+ *pend = '\0';
+ len = pend - pelem;
+
+ /* Empty element or "." means cwd. */
+ if (len == 0 || (len == 1 && *pelem == '.')) {
+ if (g->verbose)
+ fprintf (stderr,
+ "looking for appliance in current directory\n");
+ if (dir_contains_files (".", kernel_name, initrd_name, NULL)) {
+ kernel = safe_strdup (g, kernel_name);
+ initrd = safe_strdup (g, initrd_name);
+ break;
+ }
+ }
+ /* Look at <path>/kernel etc. */
+ else {
+ if (g->verbose)
+ fprintf (stderr, "looking for appliance in %s\n", pelem);
+
+ if (dir_contains_files (pelem, kernel_name, initrd_name, NULL)) {
+ kernel = safe_malloc (g, len + strlen (kernel_name) + 2);
+ initrd = safe_malloc (g, len + strlen (initrd_name) + 2);
+ sprintf (kernel, "%s/%s", pelem, kernel_name);
+ sprintf (initrd, "%s/%s", pelem, initrd_name);
+ break;
+ }
+ }
+
+ pelem = pend + 1;
+ } while (pmore);
+
+ free (path);
+ }
+
+ if (kernel == NULL || initrd == NULL) {
+ error (g, _("cannot find %s or %s on LIBGUESTFS_PATH (current path = %s)"),
+ kernel_name, initrd_name, g->path);
+ goto cleanup0;
+ }
+
+ if (g->verbose)
+ print_timestamped_message (g, "begin testing qemu features");
+
+ /* Get qemu help text and version. */
+ if (test_qemu (g) == -1)
+ goto cleanup0;
+
+ /* Choose which vmchannel implementation to use. */
+ if (qemu_supports (g, "-net user")) {
+ /* The "null vmchannel" implementation. Requires SLIRP (user mode
+ * networking in qemu) but no other vmchannel support. The daemon
+ * will connect back to a random port number on localhost.
+ */
+ struct sockaddr_in addr;
+ socklen_t addrlen = sizeof addr;
+
+ g->sock = socket (AF_INET, SOCK_STREAM, IPPROTO_TCP);
+ if (g->sock == -1) {
+ perrorf (g, "socket");
+ goto cleanup0;
+ }
+ addr.sin_family = AF_INET;
+ addr.sin_port = htons (0);
+ addr.sin_addr.s_addr = htonl (INADDR_LOOPBACK);
+ if (bind (g->sock, (struct sockaddr *) &addr, addrlen) == -1) {
+ perrorf (g, "bind");
+ goto cleanup0;
+ }
+
+ if (listen (g->sock, 256) == -1) {
+ perrorf (g, "listen");
+ goto cleanup0;
+ }
+
+ if (getsockname (g->sock, (struct sockaddr *) &addr, &addrlen) == -1) {
+ perrorf (g, "getsockname");
+ goto cleanup0;
+ }
+
+ if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
+ perrorf (g, "fcntl");
+ goto cleanup0;
+ }
+
+ null_vmchannel_sock = ntohs (addr.sin_port);
+ if (g->verbose)
+ fprintf (stderr, "null_vmchannel_sock = %d\n", null_vmchannel_sock);
+ } else {
+ /* Using some vmchannel impl. We need to create a local Unix
+ * domain socket for qemu to use.
+ */
+ snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
+ unlink (unixsock);
+ null_vmchannel_sock = 0;
+ }
+
+ if (!g->direct) {
+ if (pipe (wfd) == -1 || pipe (rfd) == -1) {
+ perrorf (g, "pipe");
+ goto cleanup0;
+ }
+ }
+
+ if (g->verbose)
+ print_timestamped_message (g, "finished testing qemu features");
+
+ r = fork ();
+ if (r == -1) {
+ perrorf (g, "fork");
+ if (!g->direct) {
+ close (wfd[0]);
+ close (wfd[1]);
+ close (rfd[0]);
+ close (rfd[1]);
+ }
+ goto cleanup0;
+ }
+
+ if (r == 0) { /* Child (qemu). */
+ char buf[256];
+ const char *vmchannel = NULL;
+
+ /* Set up the full command line. Do this in the subprocess so we
+ * don't need to worry about cleaning up.
+ */
+ g->cmdline[0] = g->qemu;
+
+ snprintf (buf, sizeof buf, "%d", g->memsize);
+ add_cmdline (g, "-m");
+ add_cmdline (g, buf);
+
+ add_cmdline (g, "-no-reboot"); /* Force exit instead of reboot on panic */
+ add_cmdline (g, "-nographic");
+ add_cmdline (g, "-serial");
+ add_cmdline (g, "stdio");
+
+ /* These options recommended by KVM developers to improve reliability. */
+ if (qemu_supports (g, "-no-hpet"))
+ add_cmdline (g, "-no-hpet");
+
+ if (qemu_supports (g, "-rtc-td-hack"))
+ add_cmdline (g, "-rtc-td-hack");
+
+ /* If qemu has SLIRP (user mode network) enabled then we can get
+ * away with "no vmchannel", where we just connect back to a random
+ * host port.
+ */
+ if (null_vmchannel_sock) {
+ add_cmdline (g, "-net");
+ add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
+
+ snprintf (buf, sizeof buf,
+ "guestfs_vmchannel=tcp:10.0.2.2:%d", null_vmchannel_sock);
+ vmchannel = strdup (buf);
+ }
+
+ /* New-style -net user,guestfwd=... syntax for guestfwd. See:
+ *
+ * http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=c92ef6a22d3c71538fcc48fb61ad353f7ba03b62
+ *
+ * The original suggested format doesn't work, see:
+ *
+ * http://lists.gnu.org/archive/html/qemu-devel/2009-07/msg01654.html
+ *
+ * However Gerd Hoffman privately suggested to me using -chardev
+ * instead, which does work.
+ */
+ else if (qemu_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
+ snprintf (buf, sizeof buf,
+ "socket,id=guestfsvmc,path=%s,server,nowait", unixsock);
+
+ add_cmdline (g, "-chardev");
+ add_cmdline (g, buf);
+
+ snprintf (buf, sizeof buf,
+ "user,vlan=0,net=10.0.2.0/8,"
+ "guestfwd=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT
+ "-chardev:guestfsvmc");
+
+ add_cmdline (g, "-net");
+ add_cmdline (g, buf);
+
+ vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
+ }
+
+ /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
+ * syntax, or if not then we'll get a quick failure.
+ */
+ else {
+ snprintf (buf, sizeof buf,
+ "channel," GUESTFWD_PORT ":unix:%s,server,nowait", unixsock);
+
+ add_cmdline (g, "-net");
+ add_cmdline (g, buf);
+ add_cmdline (g, "-net");
+ add_cmdline (g, "user,vlan=0,net=10.0.2.0/8");
+
+ vmchannel = "guestfs_vmchannel=tcp:" GUESTFWD_ADDR ":" GUESTFWD_PORT;
+ }
+ add_cmdline (g, "-net");
+ add_cmdline (g, "nic,model=" NET_IF ",vlan=0");
+
+#define LINUX_CMDLINE \
+ "panic=1 " /* force kernel to panic if daemon exits */ \
+ "console=ttyS0 " /* serial console */ \
+ "udevtimeout=300 " /* good for very slow systems (RHBZ#480319) */ \
+ "noapic " /* workaround for RHBZ#502058 - ok if not SMP */ \
+ "acpi=off " /* we don't need ACPI, turn it off */ \
+ "printk.time=1 " /* display timestamp before kernel messages */ \
+ "cgroup_disable=memory " /* saves us about 5 MB of RAM */
+
+ /* Linux kernel command line. */
+ snprintf (buf, sizeof buf,
+ LINUX_CMDLINE
+ "%s " /* (selinux) */
+ "%s " /* (vmchannel) */
+ "%s " /* (verbose) */
+ "%s", /* (append) */
+ g->selinux ? "selinux=1 enforcing=0" : "selinux=0",
+ vmchannel ? vmchannel : "",
+ g->verbose ? "guestfs_verbose=1" : "",
+ g->append ? g->append : "");
+
+ add_cmdline (g, "-kernel");
+ add_cmdline (g, (char *) kernel);
+ add_cmdline (g, "-initrd");
+ add_cmdline (g, (char *) initrd);
+ add_cmdline (g, "-append");
+ add_cmdline (g, buf);
+
+ /* Finish off the command line. */
+ incr_cmdline_size (g);
+ g->cmdline[g->cmdline_size-1] = NULL;
+
+ if (g->verbose)
+ print_cmdline (g);
+
+ if (!g->direct) {
+ /* Set up stdin, stdout. */
+ close (0);
+ close (1);
+ close (wfd[1]);
+ close (rfd[0]);
+
+ if (dup (wfd[0]) == -1) {
+ dup_failed:
+ perror ("dup failed");
+ _exit (1);
+ }
+ if (dup (rfd[1]) == -1)
+ goto dup_failed;
+
+ close (wfd[0]);
+ close (rfd[1]);
+ }
+
+#if 0
+ /* Set up a new process group, so we can signal this process
+ * and all subprocesses (eg. if qemu is really a shell script).
+ */
+ setpgid (0, 0);
+#endif
+
+ setenv ("LC_ALL", "C", 1);
+
+ execv (g->qemu, g->cmdline); /* Run qemu. */
+ perror (g->qemu);
+ _exit (1);
+ }
+
+ /* Parent (library). */
+ g->pid = r;
+
+ free (kernel);
+ kernel = NULL;
+ free (initrd);
+ initrd = NULL;
+
+ /* Fork the recovery process off which will kill qemu if the parent
+ * process fails to do so (eg. if the parent segfaults).
+ */
+ g->recoverypid = -1;
+ if (g->recovery_proc) {
+ r = fork ();
+ if (r == 0) {
+ pid_t qemu_pid = g->pid;
+ pid_t parent_pid = getppid ();
+
+ /* Writing to argv is hideously complicated and error prone. See:
+ * http://anoncvs.postgresql.org/cvsweb.cgi/pgsql/src/backend/utils/misc/ps_status.c?rev=1.33.2.1;content-type=text%2Fplain
+ */
+
+ /* Loop around waiting for one or both of the other processes to
+ * disappear. It's fair to say this is very hairy. The PIDs that
+ * we are looking at might be reused by another process. We are
+ * effectively polling. Is the cure worse than the disease?
+ */
+ for (;;) {
+ if (kill (qemu_pid, 0) == -1) /* qemu's gone away, we aren't needed */
+ _exit (0);
+ if (kill (parent_pid, 0) == -1) {
+ /* Parent's gone away, qemu still around, so kill qemu. */
+ kill (qemu_pid, 9);
+ _exit (0);
+ }
+ sleep (2);
+ }
+ }
+
+ /* Don't worry, if the fork failed, this will be -1. The recovery
+ * process isn't essential.
+ */
+ g->recoverypid = r;
+ }
+
+ if (!g->direct) {
+ /* Close the other ends of the pipe. */
+ close (wfd[0]);
+ close (rfd[1]);
+
+ if (fcntl (wfd[1], F_SETFL, O_NONBLOCK) == -1 ||
+ fcntl (rfd[0], F_SETFL, O_NONBLOCK) == -1) {
+ perrorf (g, "fcntl");
+ goto cleanup1;
+ }
+
+ g->fd[0] = wfd[1]; /* stdin of child */
+ g->fd[1] = rfd[0]; /* stdout of child */
+ } else {
+ g->fd[0] = open ("/dev/null", O_RDWR);
+ if (g->fd[0] == -1) {
+ perrorf (g, "open /dev/null");
+ goto cleanup1;
+ }
+ g->fd[1] = dup (g->fd[0]);
+ if (g->fd[1] == -1) {
+ perrorf (g, "dup");
+ close (g->fd[0]);
+ goto cleanup1;
+ }
+ }
+
+ if (null_vmchannel_sock) {
+ int sock = -1;
+ uid_t uid;
+
+ /* Null vmchannel implementation: We listen on g->sock for a
+ * connection. The connection could come from any local process
+ * so we must check it comes from the appliance (or at least
+ * from our UID) for security reasons.
+ */
+ while (sock == -1) {
+ sock = accept_from_daemon (g);
+ if (sock == -1)
+ goto cleanup1;
+
+ if (check_peer_euid (g, sock, &uid) == -1)
+ goto cleanup1;
+ if (uid != geteuid ()) {
+ fprintf (stderr,
+ "libguestfs: warning: unexpected connection from UID %d to port %d\n",
+ uid, null_vmchannel_sock);
+ close (sock);
+ continue;
+ }
+ }