+ /* Get qemu help text and version. */
+ if (test_qemu (g) == -1)
+ goto cleanup0;
+
+ /* Make the vmchannel socket. */
+ snprintf (unixsock, sizeof unixsock, "%s/sock", g->tmpdir);
+ unlink (unixsock);
+
+ if (pipe (wfd) == -1 || pipe (rfd) == -1) {
+ perrorf (g, "pipe");
+ goto cleanup0;
+ }
+
+ r = fork ();
+ if (r == -1) {
+ perrorf (g, "fork");
+ close (wfd[0]);
+ close (wfd[1]);
+ close (rfd[0]);
+ close (rfd[1]);
+ goto cleanup0;
+ }
+
+ if (r == 0) { /* Child (qemu). */
+ char buf[256];
+
+ /* 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_supports (g, "-chardev") && qemu_supports (g, "guestfwd")) {
+ /* 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.
+ */
+ 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:%s:%d-chardev:guestfsvmc",
+ GUESTFWD_ADDR, GUESTFWD_PORT);
+
+ add_cmdline (g, "-net");
+ add_cmdline (g, buf);
+ } else {
+ /* Not guestfwd. HOPEFULLY this qemu uses the older -net channel
+ * syntax, or if not then we'll get a quick failure.
+ */
+ snprintf (buf, sizeof buf,
+ "channel,%d:unix:%s,server,nowait", GUESTFWD_PORT, 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");
+ }
+ 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 */ \
+ "cgroup_disable=memory " /* saves us about 5 MB of RAM */
+
+ /* Linux kernel command line. */
+ snprintf (buf, sizeof buf,
+ LINUX_CMDLINE
+ "%s" /* (selinux) */
+ "%s" /* (verbose) */
+ "%s", /* (append) */
+ g->selinux ? "selinux=1 enforcing=0 " : "selinux=0 ",
+ 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);
+
+ /* 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
+
+ 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).
+ */
+ 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;
+
+ /* Start the clock ... */
+ time (&g->start_t);
+
+ /* 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 */
+
+ /* Open the Unix socket. The vmchannel implementation that got
+ * merged with qemu sucks in a number of ways. Both ends do
+ * connect(2), which means that no one knows what, if anything, is
+ * connected to the other end, or if it becomes disconnected. Even
+ * worse, we have to wait some indeterminate time for qemu to create
+ * the socket and connect to it (which happens very early in qemu's
+ * start-up), so any code that uses vmchannel is inherently racy.
+ * Hence this silly loop.
+ */
+ g->sock = socket (AF_UNIX, SOCK_STREAM, 0);
+ if (g->sock == -1) {
+ perrorf (g, "socket");
+ goto cleanup1;
+ }
+
+ if (fcntl (g->sock, F_SETFL, O_NONBLOCK) == -1) {
+ perrorf (g, "fcntl");
+ goto cleanup2;
+ }
+
+ addr.sun_family = AF_UNIX;
+ strncpy (addr.sun_path, unixsock, UNIX_PATH_MAX);
+ addr.sun_path[UNIX_PATH_MAX-1] = '\0';
+
+ tries = 100;
+ /* Always sleep at least once to give qemu a small chance to start up. */
+ usleep (10000);
+ while (tries > 0) {
+ r = connect (g->sock, (struct sockaddr *) &addr, sizeof addr);
+ if ((r == -1 && errno == EINPROGRESS) || r == 0)
+ goto connected;
+
+ if (errno != ENOENT)
+ perrorf (g, "connect");
+ tries--;
+ usleep (100000);
+ }
+
+ error (g, _("failed to connect to vmchannel socket"));
+ goto cleanup2;
+
+ connected:
+ g->state = LAUNCHING;
+
+ /* Wait for qemu to start and to connect back to us via vmchannel and
+ * send the GUESTFS_LAUNCH_FLAG message.
+ */
+ uint32_t size;
+ void *buf = NULL;
+ r = recv_from_daemon (g, &size, &buf);
+ free (buf);
+
+ if (r == -1) return -1;
+
+ if (size != GUESTFS_LAUNCH_FLAG) {
+ error (g, _("guestfs_launch failed, see earlier error messages"));
+ goto cleanup2;
+ }
+
+ /* This is possible in some really strange situations, such as
+ * guestfsd starts up OK but then qemu immediately exits. Check for
+ * it because the caller is probably expecting to be able to send
+ * commands after this function returns.
+ */
+ if (g->state != READY) {
+ error (g, _("qemu launched and contacted daemon, but state != READY"));
+ goto cleanup2;
+ }
+
+ return 0;
+
+ cleanup2:
+ close (g->sock);
+
+ cleanup1:
+ close (wfd[1]);
+ close (rfd[0]);
+ kill (g->pid, 9);
+ if (g->recoverypid > 0) kill (g->recoverypid, 9);
+ waitpid (g->pid, NULL, 0);
+ if (g->recoverypid > 0) waitpid (g->recoverypid, NULL, 0);
+ g->fd[0] = -1;
+ g->fd[1] = -1;
+ g->sock = -1;
+ g->pid = 0;
+ g->recoverypid = 0;
+ g->start_t = 0;
+
+ cleanup0:
+ free (kernel);
+ free (initrd);
+ return -1;
+}
+
+/* This function is used to print the qemu command line before it gets
+ * executed, when in verbose mode.
+ */
+static void
+print_cmdline (guestfs_h *g)
+{
+ int i = 0;
+ int needs_quote;
+
+ while (g->cmdline[i]) {
+ if (g->cmdline[i][0] == '-') /* -option starts a new line */
+ fprintf (stderr, " \\\n ");
+
+ if (i > 0) fputc (' ', stderr);
+
+ /* Does it need shell quoting? This only deals with simple cases. */
+ needs_quote = strcspn (g->cmdline[i], " ") != strlen (g->cmdline[i]);
+
+ if (needs_quote) fputc ('\'', stderr);
+ fprintf (stderr, "%s", g->cmdline[i]);
+ if (needs_quote) fputc ('\'', stderr);
+ i++;
+ }
+
+ fputc ('\n', stderr);
+}
+
+/* This function does the hard work of building the supermin appliance
+ * on the fly. 'path' is the directory containing the control files.
+ * 'kernel' and 'initrd' are where we will return the names of the
+ * kernel and initrd (only initrd is built). The work is done by
+ * an external script. We just tell it where to put the result.
+ */
+static int
+build_supermin_appliance (guestfs_h *g, const char *path,
+ char **kernel, char **initrd)
+{
+ char cmd[4096];
+ int r, len;
+
+ len = strlen (g->tmpdir);
+ *kernel = safe_malloc (g, len + 8);
+ snprintf (*kernel, len+8, "%s/kernel", g->tmpdir);
+ *initrd = safe_malloc (g, len + 8);
+ snprintf (*initrd, len+8, "%s/initrd", g->tmpdir);
+
+ snprintf (cmd, sizeof cmd,
+ "PATH='%s':$PATH "
+ "libguestfs-supermin-helper '%s' %s %s",
+ path,
+ path, *kernel, *initrd);
+
+ r = system (cmd);
+ if (r == -1 || WEXITSTATUS(r) != 0) {
+ error (g, _("external command failed: %s"), cmd);
+ free (*kernel);
+ free (*initrd);
+ *kernel = *initrd = NULL;
+ return -1;
+ }
+
+ return 0;
+}
+
+static int read_all (guestfs_h *g, FILE *fp, char **ret);
+
+/* Test qemu binary (or wrapper) runs, and do 'qemu -help' and
+ * 'qemu -version' so we know what options this qemu supports and
+ * the version.
+ */
+static int
+test_qemu (guestfs_h *g)
+{
+ char cmd[1024];
+ FILE *fp;
+
+ free (g->qemu_help);
+ free (g->qemu_version);
+ g->qemu_help = NULL;
+ g->qemu_version = NULL;
+
+ snprintf (cmd, sizeof cmd, "'%s' -help", g->qemu);
+
+ fp = popen (cmd, "r");
+ /* qemu -help should always work (qemu -version OTOH wasn't
+ * supported by qemu 0.9). If this command doesn't work then it
+ * probably indicates that the qemu binary is missing.
+ */
+ if (!fp) {
+ /* XXX This error is never printed, even if the qemu binary
+ * doesn't exist. Why?
+ */
+ error:
+ perrorf (g, _("%s: command failed: If qemu is located on a non-standard path, try setting the LIBGUESTFS_QEMU environment variable."), cmd);
+ return -1;
+ }
+
+ if (read_all (g, fp, &g->qemu_help) == -1)
+ goto error;
+
+ if (pclose (fp) == -1)
+ goto error;
+
+ snprintf (cmd, sizeof cmd, "'%s' -version 2>/dev/null", g->qemu);
+
+ fp = popen (cmd, "r");
+ if (fp) {
+ /* Intentionally ignore errors. */
+ read_all (g, fp, &g->qemu_version);
+ pclose (fp);
+ }
+
+ return 0;
+}
+
+static int
+read_all (guestfs_h *g, FILE *fp, char **ret)
+{
+ int r, n = 0;
+ char *p;
+
+ again:
+ if (feof (fp)) {
+ *ret = safe_realloc (g, *ret, n + 1);
+ (*ret)[n] = '\0';
+ return n;
+ }
+
+ *ret = safe_realloc (g, *ret, n + BUFSIZ);
+ p = &(*ret)[n];
+ r = fread (p, 1, BUFSIZ, fp);
+ if (ferror (fp)) {
+ perrorf (g, "read");
+ return -1;
+ }
+ n += r;
+ goto again;
+}
+
+/* Test if option is supported by qemu command line (just by grepping
+ * the help text).
+ */
+static int
+qemu_supports (guestfs_h *g, const char *option)
+{
+ return g->qemu_help && strstr (g->qemu_help, option) != NULL;
+}
+
+/* You had to call this function after launch in versions <= 1.0.70,
+ * but it is now a no-op.
+ */
+int
+guestfs__wait_ready (guestfs_h *g)
+{
+ if (g->state != READY) {
+ error (g, _("qemu has not been launched yet"));
+ return -1;
+ }
+
+ return 0;
+}
+
+int
+guestfs__kill_subprocess (guestfs_h *g)
+{
+ if (g->state == CONFIG) {
+ error (g, _("no subprocess to kill"));
+ return -1;
+ }
+
+ if (g->verbose)
+ fprintf (stderr, "sending SIGTERM to process %d\n", g->pid);
+
+ kill (g->pid, SIGTERM);
+ if (g->recoverypid > 0) kill (g->recoverypid, 9);
+
+ return 0;
+}
+
+/* Access current state. */
+int
+guestfs__is_config (guestfs_h *g)
+{
+ return g->state == CONFIG;
+}
+
+int
+guestfs__is_launching (guestfs_h *g)
+{
+ return g->state == LAUNCHING;
+}
+
+int
+guestfs__is_ready (guestfs_h *g)
+{
+ return g->state == READY;
+}
+
+int
+guestfs__is_busy (guestfs_h *g)
+{
+ return g->state == BUSY;
+}
+
+int
+guestfs__get_state (guestfs_h *g)
+{
+ return g->state;
+}
+
+void
+guestfs_set_log_message_callback (guestfs_h *g,
+ guestfs_log_message_cb cb, void *opaque)
+{
+ g->log_message_cb = cb;
+ g->log_message_cb_data = opaque;
+}
+
+void
+guestfs_set_subprocess_quit_callback (guestfs_h *g,
+ guestfs_subprocess_quit_cb cb, void *opaque)
+{
+ g->subprocess_quit_cb = cb;
+ g->subprocess_quit_cb_data = opaque;
+}
+
+void
+guestfs_set_launch_done_callback (guestfs_h *g,
+ guestfs_launch_done_cb cb, void *opaque)
+{
+ g->launch_done_cb = cb;
+ g->launch_done_cb_data = opaque;
+}
+
+/*----------------------------------------------------------------------*/
+
+/* This is the code used to send and receive RPC messages and (for
+ * certain types of message) to perform file transfers. This code is
+ * driven from the generated actions (src/guestfs-actions.c). There
+ * are five different cases to consider:
+ *
+ * (1) A non-daemon function. There is no RPC involved at all, it's
+ * all handled inside the library.
+ *
+ * (2) A simple RPC (eg. "mount"). We write the request, then read
+ * the reply. The sequence of calls is:
+ *
+ * guestfs___set_busy
+ * guestfs___send
+ * guestfs___recv
+ * guestfs___end_busy
+ *
+ * (3) An RPC with FileOut parameters (eg. "upload"). We write the
+ * request, then write the file(s), then read the reply. The sequence
+ * of calls is:
+ *
+ * guestfs___set_busy
+ * guestfs___send
+ * guestfs___send_file (possibly multiple times)
+ * guestfs___recv
+ * guestfs___end_busy
+ *
+ * (4) An RPC with FileIn parameters (eg. "download"). We write the
+ * request, then read the reply, then read the file(s). The sequence
+ * of calls is:
+ *
+ * guestfs___set_busy
+ * guestfs___send
+ * guestfs___recv
+ * guestfs___recv_file (possibly multiple times)
+ * guestfs___end_busy
+ *
+ * (5) Both FileOut and FileIn parameters. There are no calls like
+ * this in the current API, but they would be implemented as a
+ * combination of cases (3) and (4).
+ *
+ * During all writes and reads, we also select(2) on qemu stdout
+ * looking for messages (guestfsd stderr and guest kernel dmesg), and
+ * anything received is passed up through the log_message_cb. This is
+ * also the reason why all the sockets are non-blocking. We also have
+ * to check for EOF (qemu died). All of this is handled by the
+ * functions send_to_daemon and recv_from_daemon.
+ */
+
+int
+guestfs___set_busy (guestfs_h *g)
+{
+ if (g->state != READY) {
+ error (g, _("guestfs_set_busy: called when in state %d != READY"),
+ g->state);
+ return -1;
+ }
+ g->state = BUSY;
+ 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:
+ default:
+ error (g, _("guestfs_end_busy: called when in state %d"), g->state);
+ return -1;
+ }
+ return 0;
+}
+
+/* This is called if we detect EOF, ie. qemu died. */
+static void
+child_cleanup (guestfs_h *g)
+{
+ if (g->verbose)
+ fprintf (stderr, "child_cleanup: %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);
+ 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->state = CONFIG;
+ if (g->subprocess_quit_cb)
+ g->subprocess_quit_cb (g, g->subprocess_quit_cb_data);
+}
+
+static int
+read_log_message_or_eof (guestfs_h *g, int fd)
+{
+ char buf[BUFSIZ];
+ int n;
+
+#if 0
+ if (g->verbose)
+ fprintf (stderr,
+ "read_log_message_or_eof: %p g->state = %d, fd = %d\n",
+ g, g->state, fd);
+#endif
+
+ /* QEMU's console emulates a 16550A serial port. The real 16550A
+ * device has a small FIFO buffer (16 bytes) which means here we see
+ * lots of small reads of 1-16 bytes in length, usually single
+ * bytes.
+ */
+ n = read (fd, buf, sizeof buf);
+ if (n == 0) {
+ /* Hopefully this indicates the qemu child process has died. */
+ child_cleanup (g);
+ return -1;
+ }
+
+ if (n == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ return 0;
+
+ perrorf (g, "read");
+ return -1;
+ }
+
+ /* In verbose mode, copy all log messages to stderr. */
+ if (g->verbose)
+ ignore_value (write (STDERR_FILENO, buf, n));
+
+ /* It's an actual log message, send it upwards if anyone is listening. */
+ if (g->log_message_cb)
+ g->log_message_cb (g, g->log_message_cb_data, buf, n);
+
+ return 0;
+}
+
+static int
+check_for_daemon_cancellation_or_eof (guestfs_h *g, int fd)
+{
+ char buf[4];
+ int n;
+ uint32_t flag;
+ XDR xdr;
+
+ if (g->verbose)
+ fprintf (stderr,
+ "check_for_daemon_cancellation_or_eof: %p g->state = %d, fd = %d\n",
+ g, g->state, fd);
+
+ n = read (fd, buf, 4);
+ if (n == 0) {
+ /* Hopefully this indicates the qemu child process has died. */
+ child_cleanup (g);
+ return -1;
+ }
+
+ if (n == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ return 0;
+
+ perrorf (g, "read");
+ return -1;
+ }
+
+ xdrmem_create (&xdr, buf, 4, XDR_DECODE);
+ xdr_uint32_t (&xdr, &flag);
+ xdr_destroy (&xdr);
+
+ if (flag != GUESTFS_CANCEL_FLAG) {
+ error (g, _("check_for_daemon_cancellation_or_eof: read 0x%x from daemon, expected 0x%x\n"),
+ flag, GUESTFS_CANCEL_FLAG);
+ return -1;
+ }
+
+ return -2;
+}
+
+/* This writes the whole N bytes of BUF to the daemon socket.
+ *
+ * If the whole write is successful, it returns 0.
+ * If there was an error, it returns -1.
+ * If the daemon sent a cancellation message, it returns -2.
+ *
+ * It also checks qemu stdout for log messages and passes those up
+ * through log_message_cb.
+ *
+ * It also checks for EOF (qemu died) and passes that up through the
+ * child_cleanup function above.
+ */
+static int
+send_to_daemon (guestfs_h *g, const void *v_buf, size_t n)
+{
+ const char *buf = v_buf;
+ fd_set rset, rset2;
+ fd_set wset, wset2;
+
+ if (g->verbose)
+ fprintf (stderr,
+ "send_to_daemon: %p g->state = %d, n = %zu\n", g, g->state, n);
+
+ FD_ZERO (&rset);
+ FD_ZERO (&wset);
+
+ FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
+ FD_SET (g->sock, &rset); /* Read socket for cancellation & EOF. */
+ FD_SET (g->sock, &wset); /* Write to socket to send the data. */
+
+ int max_fd = g->sock > g->fd[1] ? g->sock : g->fd[1];
+
+ while (n > 0) {
+ rset2 = rset;
+ wset2 = wset;
+ int r = select (max_fd+1, &rset2, &wset2, NULL, NULL);
+ if (r == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ continue;
+ perrorf (g, "select");
+ return -1;
+ }
+
+ if (FD_ISSET (g->fd[1], &rset2)) {
+ if (read_log_message_or_eof (g, g->fd[1]) == -1)
+ return -1;
+ }
+ if (FD_ISSET (g->sock, &rset2)) {
+ r = check_for_daemon_cancellation_or_eof (g, g->sock);
+ if (r < 0)
+ return r;
+ }
+ if (FD_ISSET (g->sock, &wset2)) {
+ r = write (g->sock, buf, n);
+ if (r == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ continue;
+ perrorf (g, "write");
+ if (errno == EPIPE) /* Disconnected from guest (RHBZ#508713). */
+ child_cleanup (g);
+ return -1;
+ }
+ buf += r;
+ n -= r;
+ }
+ }
+
+ return 0;
+}
+
+/* This reads a single message, file chunk, launch flag or
+ * cancellation flag from the daemon. If something was read, it
+ * returns 0, otherwise -1.
+ *
+ * Both size_rtn and buf_rtn must be passed by the caller as non-NULL.
+ *
+ * *size_rtn returns the size of the returned message or it may be
+ * GUESTFS_LAUNCH_FLAG or GUESTFS_CANCEL_FLAG.
+ *
+ * *buf_rtn is returned containing the message (if any) or will be set
+ * to NULL. *buf_rtn must be freed by the caller.
+ *
+ * It also checks qemu stdout for log messages and passes those up
+ * through log_message_cb.
+ *
+ * It also checks for EOF (qemu died) and passes that up through the
+ * child_cleanup function above.
+ */
+static int
+recv_from_daemon (guestfs_h *g, uint32_t *size_rtn, void **buf_rtn)
+{
+ fd_set rset, rset2;
+
+ if (g->verbose)
+ fprintf (stderr,
+ "recv_from_daemon: %p g->state = %d, size_rtn = %p, buf_rtn = %p\n",
+ g, g->state, size_rtn, buf_rtn);
+
+ FD_ZERO (&rset);
+
+ FD_SET (g->fd[1], &rset); /* Read qemu stdout for log messages & EOF. */
+ FD_SET (g->sock, &rset); /* Read socket for data & EOF. */
+
+ int max_fd = g->sock > g->fd[1] ? g->sock : g->fd[1];
+
+ *size_rtn = 0;
+ *buf_rtn = NULL;
+
+ char lenbuf[4];
+ /* nr is the size of the message, but we prime it as -4 because we
+ * have to read the message length word first.
+ */
+ ssize_t nr = -4;
+
+ while (nr < (ssize_t) *size_rtn) {
+ rset2 = rset;
+ int r = select (max_fd+1, &rset2, NULL, NULL, NULL);
+ if (r == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ continue;
+ perrorf (g, "select");
+ free (*buf_rtn);
+ *buf_rtn = NULL;
+ return -1;
+ }
+
+ if (FD_ISSET (g->fd[1], &rset2)) {
+ if (read_log_message_or_eof (g, g->fd[1]) == -1) {
+ free (*buf_rtn);
+ *buf_rtn = NULL;
+ return -1;
+ }
+ }
+ if (FD_ISSET (g->sock, &rset2)) {
+ if (nr < 0) { /* Have we read the message length word yet? */
+ r = read (g->sock, lenbuf+nr+4, -nr);
+ if (r == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ continue;
+ int err = errno;
+ perrorf (g, "read");
+ /* Under some circumstances we see "Connection reset by peer"
+ * here when the child dies suddenly. Catch this and call
+ * the cleanup function, same as for EOF.
+ */
+ if (err == ECONNRESET)
+ child_cleanup (g);
+ return -1;
+ }
+ if (r == 0) {
+ error (g, _("unexpected end of file when reading from daemon"));
+ child_cleanup (g);
+ return -1;
+ }
+ nr += r;
+
+ if (nr < 0) /* Still not got the whole length word. */
+ continue;
+
+ XDR xdr;
+ xdrmem_create (&xdr, lenbuf, 4, XDR_DECODE);
+ xdr_uint32_t (&xdr, size_rtn);
+ xdr_destroy (&xdr);
+
+ if (*size_rtn == GUESTFS_LAUNCH_FLAG) {
+ if (g->state != LAUNCHING)
+ error (g, _("received magic signature from guestfsd, but in state %d"),
+ g->state);
+ else {
+ g->state = READY;
+ if (g->launch_done_cb)
+ g->launch_done_cb (g, g->launch_done_cb_data);
+ }
+ return 0;
+ }
+ else if (*size_rtn == GUESTFS_CANCEL_FLAG)
+ return 0;
+ /* If this happens, it's pretty bad and we've probably lost
+ * synchronization.
+ */
+ else if (*size_rtn > GUESTFS_MESSAGE_MAX) {
+ error (g, _("message length (%u) > maximum possible size (%d)"),
+ (unsigned) *size_rtn, GUESTFS_MESSAGE_MAX);
+ return -1;
+ }
+
+ /* Allocate the complete buffer, size now known. */
+ *buf_rtn = safe_malloc (g, *size_rtn);
+ /*FALLTHROUGH*/
+ }
+
+ size_t sizetoread = *size_rtn - nr;
+ if (sizetoread > BUFSIZ) sizetoread = BUFSIZ;
+
+ r = read (g->sock, (char *) (*buf_rtn) + nr, sizetoread);
+ if (r == -1) {
+ if (errno == EINTR || errno == EAGAIN)
+ continue;
+ perrorf (g, "read");
+ free (*buf_rtn);
+ *buf_rtn = NULL;
+ return -1;
+ }
+ if (r == 0) {
+ error (g, _("unexpected end of file when reading from daemon"));
+ child_cleanup (g);
+ free (*buf_rtn);
+ *buf_rtn = NULL;
+ return -1;
+ }
+ nr += r;
+ }
+ }
+
+ /* Got the full message, caller can start processing it. */
+#ifdef ENABLE_PACKET_DUMP
+ if (g->verbose) {
+ ssize_t i, j;
+
+ for (i = 0; i < nr; i += 16) {
+ printf ("%04zx: ", i);
+ for (j = i; j < MIN (i+16, nr); ++j)
+ printf ("%02x ", (*(unsigned char **)buf_rtn)[j]);
+ for (; j < i+16; ++j)
+ printf (" ");
+ printf ("|");
+ for (j = i; j < MIN (i+16, nr); ++j)
+ if (isprint ((*(char **)buf_rtn)[j]))
+ printf ("%c", (*(char **)buf_rtn)[j]);
+ else
+ printf (".");
+ for (; j < i+16; ++j)
+ printf (" ");
+ printf ("|\n");
+ }
+ }
+#endif
+
+ return 0;