launch: Close file descriptors after fork (RHBZ#1123007).

This refactors existing code to close file descriptors in the recovery
process, and also adds code to close file descriptors between the
fork() and exec() of QEMU or User-Mode Linux.

The reason is to avoid leaking main process file descriptors where the
main process (or other libraries in the main process) are not setting
O_CLOEXEC at all or not setting it atomically.  Python is a particular
culprit.

See also this OpenStack Nova bug report:
https://bugs.launchpad.net/nova/+bug/1313477

Thanks: Qin Zhao for identifying and characterizing the problem in Nova.

This is version 2 of this commit.  This commit is identical to the
reverted commit 115fcc3432 except that
we don't close stderr.

src/guestfs-internal-frontend.h

@@ -1,5 +1,5 @@
 /* libguestfs
- * Copyright (C) 2013 Red Hat Inc.
+ * Copyright (C) 2013-2014 Red Hat Inc.
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
@@ -164,4 +164,18 @@ extern GUESTFS_DLL_PUBLIC int guestfs___add_libvirt_dom (guestfs_h *g, virDomain
 #  define program_name "libguestfs"
 #endif
 
+/* Close all file descriptors matching the condition. */
+#define close_file_descriptors(cond) do {                               \
+    int max_fd = sysconf (_SC_OPEN_MAX);                                \
+    int fd;                                                             \
+    if (max_fd == -1)                                                   \
+      max_fd = 1024;                                                    \
+    if (max_fd > 65536)                                                 \
+      max_fd = 65536;          /* bound the amount of work we do here */ \
+    for (fd = 0; fd < max_fd; ++fd) {                                   \
+      if (cond)                                                         \
+        close (fd);                                                     \
+    }                                                                   \
+  } while (0)
+
 #endif /* GUESTFS_INTERNAL_FRONTEND_H_ */

src/launch-direct.c

@@ -717,6 +717,13 @@ launch_direct (guestfs_h *g, void *datav, const char *arg)
         goto dup_failed;
 
       close (sv[1]);
+
+      /* Close any other file descriptors that we don't want to pass
+       * to qemu.  This prevents file descriptors which didn't have
+       * O_CLOEXEC set properly from leaking into the subprocess.  See
+       * RHBZ#1123007.
+       */
+      close_file_descriptors (fd > 2);
     }
 
     /* Dump the command line (after setting up stderr above). */
@@ -747,7 +754,7 @@ launch_direct (guestfs_h *g, void *datav, const char *arg)
   if (g->recovery_proc) {
     r = fork ();
     if (r == 0) {
-      int i, fd, max_fd;
+      int i;
       struct sigaction sa;
       pid_t qemu_pid = data->pid;
       pid_t parent_pid = getppid ();
@@ -767,13 +774,7 @@ launch_direct (guestfs_h *g, void *datav, const char *arg)
       /* Close all other file descriptors.  This ensures that we don't
        * hold open (eg) pipes from the parent process.
        */
-      max_fd = sysconf (_SC_OPEN_MAX);
-      if (max_fd == -1)
-        max_fd = 1024;
-      if (max_fd > 65536)
-        max_fd = 65536; /* bound the amount of work we do here */
-      for (fd = 0; fd < max_fd; ++fd)
-        close (fd);
+      close_file_descriptors (1);
 
       /* It would be nice to be able to put this in the same process
        * group as qemu (ie. setpgid (0, qemu_pid)).  However this is

src/launch-uml.c

@@ -333,6 +333,9 @@ launch_uml (guestfs_h *g, void *datav, const char *arg)
         goto dup_failed;
 
       close (csv[1]);
+
+      /* RHBZ#1123007 */
+      close_file_descriptors (fd > 2 && fd != dsv[1]);
     }
 
     /* Dump the command line (after setting up stderr above). */
@@ -360,7 +363,7 @@ launch_uml (guestfs_h *g, void *datav, const char *arg)
   if (g->recovery_proc) {
     r = fork ();
     if (r == 0) {
-      int i, fd, max_fd;
+      int i;
       struct sigaction sa;
       pid_t vmlinux_pid = data->pid;
       pid_t parent_pid = getppid ();
@@ -380,13 +383,7 @@ launch_uml (guestfs_h *g, void *datav, const char *arg)
       /* Close all other file descriptors.  This ensures that we don't
        * hold open (eg) pipes from the parent process.
        */
-      max_fd = sysconf (_SC_OPEN_MAX);
-      if (max_fd == -1)
-        max_fd = 1024;
-      if (max_fd > 65536)
-        max_fd = 65536; /* bound the amount of work we do here */
-      for (fd = 0; fd < max_fd; ++fd)
-        close (fd);
+      close_file_descriptors (1);
 
       /* It would be nice to be able to put this in the same process
        * group as vmlinux (ie. setpgid (0, vmlinux_pid)).  However