[PATCH 1/6] proc: Don't allow empty /proc/PID/cmdline for user tasks

[Tejun Heo]

Kernel threads have empty /proc/PID/cmdline and some userland tools
including ps(1) and older versions of systemd use this to detect
kernel threads.  However, any userland program can emulate the
behavior by making its argvs unavailable and trick the affected tools
into thinking that the task is a kernel thread.  Linus's reproducer
follows.

  #include <sys/prctl.h>
  #include <sys/mman.h>

  int main(void)
  {
	  char empty[16384];
	  unsigned long ptr;

	  asm volatile("" :"=r" (ptr) : "0" (empty):"memory");
	  ptr = (ptr+4095) & ~4095;
	  munmap((void *)ptr, 32768);

	  sleep(1000);
	  return 0;
  }

Compiling the above program into nullcmdline and running it on an
unpatche kernel shows the following behavior.

  $ ./nullcmdline &
  [1] 2382031
  [devbig577 ~/tmp]$ hexdump -C /proc/2382031/comm
  00000000  6e 75 6c 6c 63 6d 64 6c  69 6e 65 0a              |nullcmdline.|
  0000000c
  $ hexdump -C /proc/2382031/cmdline
  $ ps 2382031
      PID TTY      STAT   TIME COMMAND
  2382031 pts/2    S      0:00 [nullcmdline]

The empty cmdline makes ps(1) think that nullcmdline is a kernel
thread and put brackets around its name (comm), which is mostly a
nuisance but it's possible that this confusion can lead to more
harmful confusions.

This patch fixes the issue by making proc_pid_cmdline_read() never
return empty string for user tasks.  If the result is empty for
whatever reason, comm string is returned.  Even when the comm string
is empty, it still returns the null termnation character.  On a
patched kernel, running the same command as above gives us.

  $ ./nullcmdline &
  [1] 2317
  [test ~]# hexdump -C /proc/2317/comm
  00000000  6e 75 6c 6c 63 6d 64 6c  69 6e 65 0a              |nullcmdline.|
  0000000c
  $ hexdump -C /proc/2317/cmdline
  00000000  6e 75 6c 6c 63 6d 64 6c  69 6e 65 00              |nullcmdline.|
  0000000c
  $ ps 2317
    PID TTY      STAT   TIME COMMAND
   2317 pts/0    S      0:00 nullcmdline

Note that cmdline is a dup of comm and ps(1) is no longer confused.

Signed-off-by: Tejun Heo <tj@xxxxxxxxxx>
Suggested-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
---
 fs/proc/base.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 1b2ede6..2eee4d7 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -224,9 +224,10 @@ static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 	if (!tsk)
 		return -ESRCH;
 	mm = get_task_mm(tsk);
-	put_task_struct(tsk);
-	if (!mm)
-		return 0;
+	if (!mm) {
+		rv = 0;
+		goto out_put_task;
+	}
 	/* Check if process spawned far enough to have cmdline. */
 	if (!mm->env_end) {
 		rv = 0;
@@ -367,8 +368,23 @@ static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 	free_page((unsigned long)page);
 out_mmput:
 	mmput(mm);
+out_put_task:
+	/*
+	 * Some userland tools use empty cmdline to distinguish kthreads.
+	 * Avoid empty cmdline for user tasks by returning tsk->comm with
+	 * \0 termination when empty.
+	 */
+	if (*pos == 0 && rv == 0 && !(tsk->flags & PF_KTHREAD)) {
+		char tcomm[TASK_COMM_LEN];
+
+		get_task_comm(tcomm, tsk);
+		rv = min(strlen(tcomm) + 1, count);
+		if (copy_to_user(buf, tsk->comm, rv))
+			rv = -EFAULT;
+	}
 	if (rv > 0)
 		*pos += rv;
+	put_task_struct(tsk);
 	return rv;
 }
 
-- 
2.9.5