perf: Introduce a flag to enable close-on-exec in perf_event_open()

Unlike recent modern userspace API such as:

  epoll_create1 (EPOLL_CLOEXEC), eventfd (EFD_CLOEXEC),
  fanotify_init (FAN_CLOEXEC), inotify_init1 (IN_CLOEXEC),
  signalfd (SFD_CLOEXEC), timerfd_create (TFD_CLOEXEC),
  or the venerable general purpose open (O_CLOEXEC),

perf_event_open() syscall lack a flag to atomically set FD_CLOEXEC
(eg. close-on-exec) flag on file descriptor it returns to userspace.

The present patch adds a PERF_FLAG_FD_CLOEXEC flag to allow
perf_event_open() syscall to atomically set close-on-exec.

Having this flag will enable userspace to remove the file descriptor
from the list of file descriptors being inherited across exec,
without the need to call fcntl(fd, F_SETFD, FD_CLOEXEC) and the
associated race condition between the current thread and another
thread calling fork(2) then execve(2).

Links:

 - Secure File Descriptor Handling (Ulrich Drepper, 2008)
   http://udrepper.livejournal.com/20407.html

 - Excuse me son, but your code is leaking !!! (Dan Walsh, March 2012)
   http://danwalsh.livejournal.com/53603.html

 - Notes in DMA buffer sharing: leak and security hole
   http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/Documentation/dma-buf-sharing.txt?id=v3.13-rc3#n428

Signed-off-by: Yann Droneaud <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Linus Torvalds <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/8c03f54e1598b1727c19706f3af03f98685d9fe6.1388952061.git.ydroneaud@opteya.com
Signed-off-by: Ingo Molnar <[email protected]>

include/uapi/linux/perf_event.h

@@ -724,6 +724,7 @@ enum perf_callchain_context {
 #define PERF_FLAG_FD_NO_GROUP		(1U << 0)
 #define PERF_FLAG_FD_OUTPUT		(1U << 1)
 #define PERF_FLAG_PID_CGROUP		(1U << 2) /* pid=cgroup id, per-cpu mode only */
+#define PERF_FLAG_FD_CLOEXEC		(1U << 3) /* O_CLOEXEC */
 
 union perf_mem_data_src {
 	__u64 val;

kernel/events/core.c

@@ -119,7 +119,8 @@ static int cpu_function_call(int cpu, int (*func) (void *info), void *info)
 
 #define PERF_FLAG_ALL (PERF_FLAG_FD_NO_GROUP |\
 		       PERF_FLAG_FD_OUTPUT  |\
-		       PERF_FLAG_PID_CGROUP)
+		       PERF_FLAG_PID_CGROUP |\
+		       PERF_FLAG_FD_CLOEXEC)
 
 /*
  * branch priv levels that need permission checks
@@ -6982,6 +6983,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	int event_fd;
 	int move_group = 0;
 	int err;
+	int f_flags = O_RDWR;
 
 	/* for future expandability... */
 	if (flags & ~PERF_FLAG_ALL)
@@ -7010,7 +7012,10 @@ SYSCALL_DEFINE5(perf_event_open,
 	if ((flags & PERF_FLAG_PID_CGROUP) && (pid == -1 || cpu == -1))
 		return -EINVAL;
 
-	event_fd = get_unused_fd();
+	if (flags & PERF_FLAG_FD_CLOEXEC)
+		f_flags |= O_CLOEXEC;
+
+	event_fd = get_unused_fd_flags(f_flags);
 	if (event_fd < 0)
 		return event_fd;
 
@@ -7132,7 +7137,8 @@ SYSCALL_DEFINE5(perf_event_open,
 			goto err_context;
 	}
 
-	event_file = anon_inode_getfile("[perf_event]", &perf_fops, event, O_RDWR);
+	event_file = anon_inode_getfile("[perf_event]", &perf_fops, event,
+					f_flags);
 	if (IS_ERR(event_file)) {
 		err = PTR_ERR(event_file);
 		goto err_context;