proc: try to remove use of FOLL_FORCE entirely

We fixed the bugs in it, but it's still an ugly interface, so let's see
if anybody actually depends on it.  It's entirely possible that nothing
actually requires the whole "punch through read-only mappings"
semantics.

For example, gdb definitely uses the /proc/<pid>/mem interface, but it
looks like it mainly does it for regular reads of the target (that don't
need FOLL_FORCE), and looking at the gdb source code seems to fall back
on the traditional ptrace(PTRACE_POKEDATA) interface if it needs to.

If this breaks something, I do have a (more complex) version that only
enables FOLL_FORCE when somebody has PTRACE_ATTACH'ed to the target,
like the comment here used to say ("Maybe we should limit FOLL_FORCE to
actual ptrace users?").

Cc: Kees Cook <keescook@chromium.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Eric Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 9e3ac5c..45f6bf6 100644 (file)
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -821,10 +821,7 @@ static ssize_t mem_rw(struct file *file, char __user *buf,
        if (!mmget_not_zero(mm))
                goto free;
 
-       /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
-       flags = FOLL_FORCE;
-       if (write)
-               flags |= FOLL_WRITE;
+       flags = write ? FOLL_WRITE : 0;
 
        while (count > 0) {
                int this_len = min_t(int, count, PAGE_SIZE);