catatoinit: close fds >= 3

close any additional fd that was already leaked into the child
process.

Closes: #12

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

catatonit.c

@@ -31,6 +31,9 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <sys/syscall.h>
+#include <limits.h>
+#include <dirent.h>
 
 #include "config.h"
 
@@ -136,6 +139,82 @@ int kernel_signals[] = {SIGFPE, SIGILL, SIGSEGV, SIGBUS, SIGABRT, SIGTRAP, SIGSY
 
 #define ARRAY_LEN(arr) (sizeof(arr) / sizeof(*arr))
 
+static int syscall_close_range(unsigned int fd, unsigned int max_fd, unsigned int flags)
+{
+#ifdef __NR_close_range
+  return (int) syscall(__NR_close_range, fd, max_fd, flags);
+#else
+  errno = ENOSYS;
+  return -1;
+#endif
+}
+
+
+/*
+ * Close every fd >= n that is different from exclude_fd using close_range.
+ */
+static int close_range_fds_ge_than(int n, int exclude_fd)
+{
+	int r;
+
+        /* exclude_fd is not in the [n, UINT_MAX] range.  */
+	if (exclude_fd < n)
+		return syscall_close_range(n, UINT_MAX, 0);
+
+        /* exclude_fd is the first fd in the [n, UINT_MAX] range.  */
+	if (exclude_fd == n)
+		return syscall_close_range(n + 1, UINT_MAX, 0);
+
+        /* exclude_fd is between n and UINT_MAX.  */
+        r = syscall_close_range(n, exclude_fd - 1, 0);
+	if (r < 0)
+		return r;
+        if (exclude_fd < UINT_MAX)
+		return syscall_close_range(exclude_fd + 1, UINT_MAX, 0);
+
+          return 0;
+}
+
+/*
+ * Close every fd >= n that is different from exclude_fd.
+ */
+static void close_fds_ge_than(int n, int exclude_fd)
+{
+	struct dirent *next;
+	DIR *dir;
+	int fd;
+
+	if (close_range_fds_ge_than(n, exclude_fd) == 0)
+		return;
+
+	/* Fallback when close_range fails.  */
+
+	dir = opendir("/proc/self/fd");
+	if (dir == NULL) {
+		warn("cannot opendir /proc/self/fd: %m");
+		return;
+	}
+
+	fd = dirfd(dir);
+	for (next = readdir (dir); next; next = readdir (dir)) {
+		long long val;
+		const char *name = next->d_name;
+		int r;
+
+		if (name[0] == '.')
+			continue;
+
+		val = strtoll (name, NULL, 10);
+		if (val < n || val == fd || val == exclude_fd)
+			continue;
+
+		r = close(val);
+		if (r < 0)
+			warn("cannot close %d: %m", val);
+	}
+	closedir(dir);
+}
+
 /*
  * Makes the current process a "foreground" process, by making it the leader of
  * a process group and session leader. It also updates the sigmask to include
@@ -361,6 +440,8 @@ int main(int argc, char **argv)
 		bail("self-check that pid1 (%d) was spawned failed: %m", pid1);
 	debug("pid1 (%d) spawned: %s", pid1, argv[0]);
 
+	close_fds_ge_than(3, sfd);
+
 	/*
 	 * The "pid" we send signals to. With -g we send signals to the entire
 	 * process group which pid1 is in, which is represented by a -ve pid.