Re: [PATCH 4/9] mm: use mm_populate() for blocking remap_file_pages()

[Hillf Danton]

On Tue, Mar 12, 2013 at 8:24 AM, Michel Lespinasse <walken@xxxxxxxxxx> wrote:
> (Sorry for the late reply)
>
> On Mon, Mar 11, 2013 at 4:03 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>> On Sun, 10 Mar 2013 20:55:21 +0200 Tommi Rantala <tt.rantala@xxxxxxxxx> wrote:
>>
>>> 2012/12/21 Michel Lespinasse <walken@xxxxxxxxxx>:
>>> > Signed-off-by: Michel Lespinasse <walken@xxxxxxxxxx>
>>>
>>> Hello, this patch introduced the following bug, seen while fuzzing with trinity:
>>>
>>> [  396.825414] BUG: unable to handle kernel NULL pointer dereference
>>> at 0000000000000050
>
> Good catch...
>
>> From: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> Subject: mm/fremap.c: fix oops on error path
>>
>> If find_vma() fails, sys_remap_file_pages() will dereference `vma', which
>> contains NULL.  Fix it by checking the pointer.
>>
>> (We could alternatively check for err==0, but this seems more direct)
>>
>> (The vm_flags change is to squish a bogus used-uninitialised warning
>> without adding extra code).
>>
>> Reported-by: Tommi Rantala <tt.rantala@xxxxxxxxx>
>> Cc: Michel Lespinasse <walken@xxxxxxxxxx>
>> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> ---
>>
>>  mm/fremap.c |    6 ++++--
>>  1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff -puN mm/fremap.c~mm-fremapc-fix-oops-on-error-path mm/fremap.c
>> --- a/mm/fremap.c~mm-fremapc-fix-oops-on-error-path
>> +++ a/mm/fremap.c
>> @@ -163,7 +163,8 @@ SYSCALL_DEFINE5(remap_file_pages, unsign
>>          * and that the remapped range is valid and fully within
>>          * the single existing vma.
>>          */
>> -       if (!vma || !(vma->vm_flags & VM_SHARED))
>> +       vm_flags = vma->vm_flags;
>> +       if (!vma || !(vm_flags & VM_SHARED))
>>                 goto out;
>
> Your commit message indicates the vm_flags load here doesn't generate any code, but this seems very brittle and compiler dependent. If the compiler was to generate an actual load here, the issue with vma == NULL would reappear.
>
>>         if (!vma->vm_ops || !vma->vm_ops->remap_pages)
>> @@ -254,7 +255,8 @@ get_write_lock:
>>          */
>>
>>  out:
>> -       vm_flags = vma->vm_flags;
>> +       if (vma)
>> +               vm_flags = vma->vm_flags;
>>         if (likely(!has_write_lock))
>>                 up_read(&mm->mmap_sem);
>>         else
>
>
>
> Would the following work ? I think it's simpler, and with the compiler
> I'm using here it doesn't emit warnings:
>
> diff --git a/mm/fremap.c b/mm/fremap.c
> index 0cd4c11488ed..329507e832fb 100644
> --- a/mm/fremap.c
> +++ b/mm/fremap.c
> @@ -254,7 +254,8 @@ get_write_lock:
>          */
>
>  out:
> -       vm_flags = vma->vm_flags;
> +       if (!err)
> +               vm_flags = vma->vm_flags;
>         if (likely(!has_write_lock))
>                 up_read(&mm->mmap_sem);
>         else
>
Is it still necessary to populate mm if bail out due
to a linear mapping encountered?

Hillf
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/