Re: [PATCH 4.4 34/47] userns: move user access out of the mutex

[Rafael David Tinoco]

Greg,

On Fri, Sep 7, 2018 at 6:41 PM Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> 4.4-stable review patch.  If anyone has any objections, please let me know.
>
> ------------------
>
> From: Jann Horn <jannh@xxxxxxxxxx>
>
> commit 5820f140edef111a9ea2ef414ab2428b8cb805b1 upstream.
>
> The old code would hold the userns_state_mutex indefinitely if
> memdup_user_nul stalled due to e.g. a userfault region. Prevent that by
> moving the memdup_user_nul in front of the mutex_lock().
>
> Note: This changes the error precedence of invalid buf/count/*ppos vs
> map already written / capabilities missing.
>
> Fixes: 22d917d80e84 ("userns: Rework the user_namespace adding uid/gid...")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
> Acked-by: Christian Brauner <christian@xxxxxxxxxx>
> Acked-by: Serge Hallyn <serge@xxxxxxxxxx>
> Signed-off-by: Eric W. Biederman <ebiederm@xxxxxxxxxxxx>
> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>
> ---
>  kernel/user_namespace.c |   22 ++++++++++------------
>  1 file changed, 10 insertions(+), 12 deletions(-)
>
> --- a/kernel/user_namespace.c
> +++ b/kernel/user_namespace.c
> @@ -604,7 +604,16 @@ static ssize_t map_write(struct file *fi
>         struct uid_gid_extent *extent = NULL;
>         unsigned long page = 0;
>         char *kbuf, *pos, *next_line;
> -       ssize_t ret = -EINVAL;
> +       ssize_t ret;
> +
> +       /* Only allow < page size writes at the beginning of the file */
> +       if ((*ppos != 0) || (count >= PAGE_SIZE))
> +               return -EINVAL;
> +
> +       /* Slurp in the user data */
> +       if (copy_from_user(kbuf, buf, count))
> +               return -EFAULT;
> +       kbuf[count] = '\0';

Naresh will soon report issues found by LKFT on user_ns for 4.4 kernel
for this review round.

selftests: mount_run_tests.sh [FAIL]
write to /proc/self/uid_map failed: Bad address

LTP: user_namespace2 1 TBROK : safe_macros.c:452: userns02.c:95:
write(6,0x7ffc133113d0,18446744073709551615) failed: errno=EFAULT(14):
Bad address

I believe the EFAULT was caused because when changing code from
"memdup_user_nul" to "copy_from_user", for the older kernels, you
missed allocating the slab object for "kbuf", like memdup_user_nul()
does.

Note: This likely applies to 3.18 as well.

We are finishing functional tests without this patch, but we wanted to
make you aware right away.

Best Regards,
Rafael