Re: [PATCH] Fix OOPS in mmap_region() when merging adjacent VM_LOCKED file segments

[Hugh Dickins <hugh@veritas.com>]

On Fri, 30 Jan 2009, Linus Torvalds wrote:
> On Fri, 30 Jan 2009, Hugh Dickins wrote:
> > 
> > By the same (used to merge, now won't) argument, one could say
> > VM_INSERTPAGE should be there too; but whereas VM_MIXEDMAP is used
> > in one place only, quite a lot of drivers use vm_insert_page(), so
> > I feel more comfortable with the idea that it's stopping merges -
> > though in that case, shouldn't we add it to VM_SPECIAL?
> 
> But VM_INSERTPAGE - no.
> 
> Why? Because VM_INSERTPAGE implies that mmap() itself quite possibly 
> actually _does_ something (it may have inserted _all_ the pages, and may 
> not even have any ->fault handler at all).
> 
> So we can't just expand a current mapping and forget about it, we need to 
> do the mmap(). So the "only do merges early" fundamentally cannot merge 
> such a vma, even if the old code did.

Good point indeed.  Same as why we test for it in copy_page_range().
Should it be added to VM_SPECIAL?  Not for this particular (need to
call ->mmap) reason, but we probably ought to add it anyway.

Hugh

> 
> Of course, we could look at whether it has a ->fault handler as an 
> indication of whether it's possible to merge or not. We already do that 
> for ->close, and in many ways ->fault would likely be a better indicator 
> of whether something is mergeable or not.
> 
> But there's really no point. VM_INSERTPAGE implies a very special mapping: 
> it's used by things like the SCSI target user space ring map, or the magic 
> network packet mmap thing. If you use those things, you really don't care 
> about merging adjacent VM's anyway, and at least the two I looked at 
> really do the whole "pre-populate at mmap time" thing.
> 
> And at least the packet one really does have a ->close function, and lacks 
> a ->fault function, so it wouldn't have merged before either (and looking 
> at ->fault again seems to be as valid as looking at ->close).
> 
> 			Linus