Re: [PATCH 2/3] mm: Consolidate vma destruction into remove_vma.

[ebiederm@xmission.com (Eric W. Biederman)]

Pekka Enberg <penberg@kernel.org> writes:

> (Fixing Hugh's email address.)

Sorry about that somehow a typo crept it.

> On Mon, Sep 27, 2010 at 9:37 AM, Pekka Enberg <penberg@kernel.org> wrote:
>> Hi Eric,
>>
>> On Sun, Sep 26, 2010 at 2:34 AM, Eric W. Biederman
>> <ebiederm@xmission.com> wrote:
>>> Consolidate vma destruction in remove_vma.   This is slightly
>>> better for code size and for code maintenance.  Avoiding the pain
>>> of 3 copies of everything needed to tear down a vma.
>>>
>>> Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
>>> ---
>>>  mm/mmap.c |   21 +++++----------------
>>>  1 files changed, 5 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/mm/mmap.c b/mm/mmap.c
>>> index 6128dc8..17dd003 100644
>>> --- a/mm/mmap.c
>>> +++ b/mm/mmap.c
>>> @@ -643,16 +643,10 @@ again:                    remove_next = 1 + (end > next->vm_end);
>>>                spin_unlock(&mapping->i_mmap_lock);
>>>
>>>        if (remove_next) {
>>> -               if (file) {
>>> -                       fput(file);
>>> -                       if (next->vm_flags & VM_EXECUTABLE)
>>> -                               removed_exe_file_vma(mm);
>>> -               }
>>>                if (next->anon_vma)
>>>                        anon_vma_merge(vma, next);
>>> +               remove_vma(next);
>>
>> remove_vma() does vma->vm_ops->close() but we don't do that here. Are
>> you sure the conversion is safe?

Definitely.  It actually isn't possible to reach that point with a
vma that has a close method.

Until I had traced through all of the code paths I suspect calling
remove_vma there might have been a bug fix.

Eric