Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <1562410493-8661-1-git-send-email-s.mesoraca16@gmail.com>
 <1562410493-8661-12-git-send-email-s.mesoraca16@gmail.com> <CAG48ez0uFX4AniOk1W0Vs6j=7Q5QfSFQTrBBzC2qL2bpWn_yCg@mail.gmail.com>
In-Reply-To: <CAG48ez0uFX4AniOk1W0Vs6j=7Q5QfSFQTrBBzC2qL2bpWn_yCg@mail.gmail.com>
From:   Salvatore Mesoraca <s.mesoraca16@gmail.com>
Date:   Sun, 7 Jul 2019 18:15:09 +0200
Message-ID: <CAJHCu1K-x1tCehO1CxTf9ZzVKLh44dE9hwWWSCxnW1A4SHX=kQ@mail.gmail.com>
Subject: Re: [PATCH v5 11/12] S.A.R.A.: /proc/*/mem write limitation
To:     Jann Horn <jannh@google.com>
Cc:     kernel list <linux-kernel@vger.kernel.org>,
        Kernel Hardening <kernel-hardening@lists.openwall.com>,
        Linux-MM <linux-mm@kvack.org>,
        linux-security-module <linux-security-module@vger.kernel.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Brad Spengler <spender@grsecurity.net>,
        Casey Schaufler <casey@schaufler-ca.com>,
        Christoph Hellwig <hch@infradead.org>,
        Kees Cook <keescook@chromium.org>,
        PaX Team <pageexec@freemail.hu>,
        "Serge E. Hallyn" <serge@hallyn.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        James Morris <jmorris@namei.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Jann Horn <jannh@google.com> wrote:
>
> On Sat, Jul 6, 2019 at 12:55 PM Salvatore Mesoraca
> <s.mesoraca16@gmail.com> wrote:
> > Prevent a task from opening, in "write" mode, any /proc/*/mem
> > file that operates on the task's mm.
> > A process could use it to overwrite read-only memory, bypassing
> > S.A.R.A. restrictions.
> [...]
> > +static void sara_task_to_inode(struct task_struct *t, struct inode *i)
> > +{
> > +       get_sara_inode_task(i) = t;
>
> This looks bogus. Nothing is actually holding a reference to `t` here, right?

I think you are right, I should probably store the PID here.

> > +}
> > +
> >  static struct security_hook_list data_hooks[] __lsm_ro_after_init = {
> >         LSM_HOOK_INIT(cred_prepare, sara_cred_prepare),
> >         LSM_HOOK_INIT(cred_transfer, sara_cred_transfer),
> >         LSM_HOOK_INIT(shm_alloc_security, sara_shm_alloc_security),
> > +       LSM_HOOK_INIT(task_to_inode, sara_task_to_inode),
> >  };
> [...]
> > +static int sara_file_open(struct file *file)
> > +{
> > +       struct task_struct *t;
> > +       struct mm_struct *mm;
> > +       u16 sara_wxp_flags = get_current_sara_wxp_flags();
> > +
> > +       /*
> > +        * Prevent write access to /proc/.../mem
> > +        * if it operates on the mm_struct of the
> > +        * current process: it could be used to
> > +        * bypass W^X.
> > +        */
> > +
> > +       if (!sara_enabled ||
> > +           !wxprot_enabled ||
> > +           !(sara_wxp_flags & SARA_WXP_WXORX) ||
> > +           !(file->f_mode & FMODE_WRITE))
> > +               return 0;
> > +
> > +       t = get_sara_inode_task(file_inode(file));
> > +       if (unlikely(t != NULL &&
> > +                    strcmp(file->f_path.dentry->d_name.name,
> > +                           "mem") == 0)) {
>
> This should probably at least have a READ_ONCE() somewhere in case the
> file concurrently gets renamed?

My understanding here is that /proc/$pid/mem files cannot be renamed.
t != NULL implies we are in procfs.
Under these assumptions I think that that code is fine.
Am I wrong?

> > +               get_task_struct(t);
> > +               mm = get_task_mm(t);
> > +               put_task_struct(t);
>
> Getting and dropping a reference to the task_struct here is completely
> useless. Either you have a reference, in which case you don't need to
> take another one, or you don't have a reference, in which case you
> also can't take one.

Absolutely agree.

> > +               if (unlikely(mm == current->mm))
> > +                       sara_warn_or_goto(error,
> > +                                         "write access to /proc/*/mem");
>
> Why is the current process so special that it must be protected more
> than other processes? Is the idea here to rely on other protections to
> protect all other tasks? This should probably come with a comment that
> explains this choice.

Yes, I should have spent some more words here.
Access to /proc/$pid/mem from other processes is already regulated by
security_ptrace_access_check (i.e. Yama).
Unfortunately, that hook is ignored when "mm == current->mm".
It seems that there is some user-space software that relies on /proc/self/mem
being writable (cfr. commit f511c0b17b08).

Thank you for your suggestions.