On Tue, Oct 01, 2013 at 06:39:00PM -0700, Andy Lutomirski wrote:
> On 10/01/2013 01:26 PM, Djalal Harouni wrote:
> > Some fields of the /proc/*/stat are sensitive fields that need
> > appropriate protection.
> >
> > However, /proc file descriptors can be passed to a more privileged
> > process (e.g. a suid-exec) which will pass the classic
> > ptrace_may_access() permission check during read().
> >
> > To prevent it, use proc_same_open_cred() to detect if current's cred
> > have changed between ->open() and ->read(), if so, call
> > proc_allow_access() to check if the original file's opener had enough
> > permissions to read these sensitive fields. This will prevent passing
> > file descriptors to a more privileged process to leak data.
> >
> > The patch also adds a previously missing signal->cred_guard_mutex lock.
> >
> > This patch does not break userspace since it only hides the fields that
> > were supposed to be protected.
> >
> > Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> > Cc: Eric W. Biederman <ebiederm@xxxxxxxxxxxx>
> > Signed-off-by: Djalal Harouni <tixxdz@xxxxxxxxxx>
> > ---
> > fs/proc/array.c | 16 ++++++++++++++--
> > 1 file changed, 14 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/proc/array.c b/fs/proc/array.c
> > index cbd0f1b..f034e05 100644
> > --- a/fs/proc/array.c
> > +++ b/fs/proc/array.c
> > @@ -394,7 +394,7 @@ static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
> > char state;
> > pid_t ppid = 0, pgid = -1, sid = -1;
> > int num_threads = 0;
> > - int permitted;
> > + int permitted = 0;
> > struct mm_struct *mm;
> > unsigned long long start_time;
> > unsigned long cmin_flt = 0, cmaj_flt = 0;
> > @@ -404,10 +404,22 @@ static int do_task_stat(struct seq_file *m, struct pid_namespace *ns,
> > unsigned long rsslim = 0;
> > char tcomm[sizeof(task->comm)];
> > unsigned long flags;
> > + struct file *file = m->private;
> > + int same_cred = proc_same_open_cred(file->f_cred);
> > + unsigned int ptrace_mode = PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT;
> >
> > state = *get_task_state(task);
> > vsize = eip = esp = 0;
> > - permitted = ptrace_may_access(task, PTRACE_MODE_READ | PTRACE_MODE_NOAUDIT);
> > +
> > + if (!mutex_lock_killable(&task->signal->cred_guard_mutex)) {
> > + permitted = ptrace_may_access(task, ptrace_mode);
> > + if (permitted && !same_cred)
> > + permitted = proc_allow_access(file->f_cred,
> > + task, ptrace_mode);
> > +
> > + mutex_unlock(&task->signal->cred_guard_mutex);
> > + }
> > +
>
> else permitted = false?
permitted is initialized to 0
First the original ptrace_may_access() check did not hold
cred_guard_mutex, so add it. If we can't grab mutex then let permitted
to be zero. Yes this a change in behaviour and I think it's correct, IOW
we were not able to perform the ptrace_may_access() check, otherwise
permitted will depend on checks result.
However, there is still a race here since we set the permitted value
before gathering the appropriate info about task. At the read() data moment
this target task may have been gone privileged... , acquiring an X lock
on target task, will just break/slow things, as it has been shown before...
Not to mention that the race window is small...
> But surely this would be *much* more comprehensible if you had
> proc_allow_access do the entire check.
I don't understand what you mean by "do the entire check" ?
the /proc/pid/stat file is used by "ps", "top" ... it's a vital file
We don't want to break it, otherwise "ps" will hide processes. So just
do the ptrace_may_access() check correctly with the help of
proc_allow_access() and set the permitted variable only. The patch did
not touch /proc/pid/stat fields.
I've done tests on ps and top, and they work fine.
--
Djalal Harouni
http://opendz.org
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
