Re: [2.6.15-rc1+ regression] do_file_page bug introduced in recentrework

[Hugh Dickins]

On Fri, 2 Dec 2005, Blaisorblade wrote:
> I recently found a bug introduced in your commit 
> 65500d234e74fc4e8f18e1a429bc24e51e75de4a, i.e. between 2.6.14 and 2.6.15-rc1, 
> about do_file_page changes wrt remap_file_pages and MAP_POPULATE.
> 
> Quoting from the changelog (which is wrong):
> 
>     do_file_page's fallback to do_no_page dates from a time when we were 
> testing
>     pte_file by using it wherever possible: currently it's peculiar to 
> nonlinear
>     vmas, so just check that.  BUG_ON if not?  Better not, it's probably page
>     table corruption, so just show the pte: hmm, there's a pte_ERROR macro, 
> let's
>     use that for do_wp_page's invalid pfn too.
> 
> This is false:
> 
> do_mmap_pgoff:
>         if (flags & MAP_POPULATE) {
>                 up_write(&mm->mmap_sem);
>                 sys_remap_file_pages(addr, len, 0,
>                                         pgoff, flags & MAP_NONBLOCK);
>                 down_write(&mm->mmap_sem);
>         }
> 
> So, with MAP_POPULATE|MAP_NONBLOCK passed, you can get a linear PAGE_FILE pte 
> in a !VM_NONLINEAR vma.
> 
> That PTE is very useless since it doesn't add any information, I know that, so 
> avoiding that possible installation is a possible fix, but for now it's 
> simpler to change the test in do_file_page(). Btw, in fact I discovered this 
> bug while I was implementing this optimization (working again on 
> remap_file_pages() patches of this summer).
> 
> Indeed, the condition to test (and to possibly BUG_ON/pte_ERROR) is that 
> ->populate must exist for the sys_remap_file_pages call to work.

I'm puzzled.  Both filemap_populate and shmem_populate
now test VM_NONLINEAR before calling install_file_pte.

Hugh
-
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/