Re: vm: weird behaviour when munmapping

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Fri, 2006-11-17 at 11:31 +0000, moreau francis wrote:
> Hmm, I'm probably missing something but I don't see what. Please be
> nice even if the question is really stupid ;)
> 
> I'm looking at mmap.c code and to understand it I decided to implement
> a dumb char device that implement its own foo_mmap() method. In this
> method it defined its own vma ops:
> 
>     static void foo_vma_open(struct vm_area_struct *vma)
>     static void foo_vma_close(struct vm_area_struct *vma)
> 
> A dumb application mmap the device in order to make foo_mmap() install
> the vma ops.
> 
>     mmap(NULL, 16384, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
> 
> mmap returned 0x2aaae000 for example. Until now, foo_vma_open() and
> foo_vma_close() are not called.
> 
> Now I want to unmap the first part of the previous mapping to see how
> vma ops are called. So I did:
> 
>     munmap(0x2aaae000, 1024);
> 
> and here's what happen:
> 
>     foo_vma_open(vma) is called with:
>         vma->vm_start = 0x2aaae000
>         vma->vm_end = 0x2aaaf000
> 
>     foo_vma_close(vma) is called with:
>         vma->vm_start = 0x2aaae000
>         vma->vm_end = 0x2aaaf000
> 
> However I would have expected:
> 
>     foo_vma_open(vma) is called with:
>         vma->vm_start = 0x2aaaf000
>         vma->vm_end = 0x2aaab2000
> 
>     foo_vma_close(vma) is called with:
>         vma->vm_start = 0x2aaae000
>         vma->vm_end = 0x2aaaf000
> 
> Can anybody tell me why I get this behaviour ?
> 

http://lwn.net/Kernel/LDD3/

Chapter 15. Section 'Virtual Memory Areas'.

Basically; vm_ops->open() is not called on the first vma. With this
munmap() you split the area in two, and it so happens the new vma is the
lower one.