Re: [PATCH] xfs: mark sb_fname as nonstring

[Arnd Bergmann]

On Tue, Jun 5, 2018 at 8:44 PM, Eric Sandeen <sandeen@xxxxxxxxxxx> wrote:
> On 5/25/18 10:14 AM, Arnd Bergmann wrote:

>> diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
>> index 84fbf164cbc3..eb79f2bc4dcc 100644
>> --- a/fs/xfs/xfs_ioctl.c
>> +++ b/fs/xfs/xfs_ioctl.c
>> @@ -1819,12 +1819,12 @@ xfs_ioc_getlabel(
>>       BUILD_BUG_ON(sizeof(sbp->sb_fname) > FSLABEL_MAX);
>>
>>       spin_lock(&mp->m_sb_lock);
>> -     strncpy(label, sbp->sb_fname, sizeof(sbp->sb_fname));
>> +     strncpy(label, sbp->sb_fname, XFSLABEL_MAX);
>>       spin_unlock(&mp->m_sb_lock);
>>
>>       /* xfs on-disk label is 12 chars, be sure we send a null to user */
>>       label[XFSLABEL_MAX] = '\0';
>> -     if (copy_to_user(user_label, label, sizeof(sbp->sb_fname)))
>> +     if (copy_to_user(user_label, label, sizeof(label)))
>
> I /think/ this also runs the risk of copying out stack memory.
>
> I'll send another proposal based on this with slight modifications.

I assumed it's safe since the earlier strncpy() pads the local 'label'
with zero bytes up the XFSLABEL_MAX, and the last byte
is explicitly set to guarantee zero padding.

Using strlcpy() or strscpy() would guarantee a zero-terminated
string without the explicit ='\0' assignment but would risk
the data leak you were probably thinking of.

       Arnd