Messages in this thread | | | Date | Tue, 17 Nov 2015 17:33:46 +0100 | From | Oleg Nesterov <> | Subject | Re: [PATCH] mm: fix incorrect behavior when process virtual address space limit is exceeded |
| |
On 11/17, Oleg Nesterov wrote: > > On 11/16, Piotr Kwapulinski wrote: > > > > @@ -1551,7 +1552,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr, > > * MAP_FIXED may remove pages of mappings that intersects with > > * requested mapping. Account for the pages it would unmap. > > */ > > - if (!(vm_flags & MAP_FIXED)) > > + if (!(flags & MAP_FIXED)) > > return -ENOMEM; > > Agree, "vm_flags & MAP_FIXED" makes no sense and just wrong... > > Can't we simply remove this check? Afaics it only helps to avoid > count_vma_pages_range() in the unlikely case when may_expand_vm() fails. > And without MAP_FIXED count_vma_pages_range() should be cheap, > find_vma_intersection() should fail.
Or we can simply move this may_expand_vm() block to the caller, do_mmap().
> And afaics arch/tile/mm/elf.c can use do_mmap(MAP_FIXED ...) rather than > mmap_region(), it can be changed by a separate patch. In this case we can > unexport mmap_region(). > > OTOH, I won't insist, this patch looks fine to me.
Yes, but what I actually tried to say is that it would be nice to unexport mmap_region(), arch/tile is the only caller outside of mmap.c.
Oleg.
| |