Hi,

On Sat, 2003-05-10 at 17:33, Andrea Arcangeli wrote:
> On Fri, May 09, 2003 at 01:34:21PM +0100, Stephen C. Tweedie wrote:
> > When a new vma can be merged simultaneously with its two immediate
> > neighbours in both directions, vma_merge() extends the predecessor vma
> > and deletes the successor.  However, if the vma maps a file, it fails to
> > fput() when doing the delete, leaving the file's refcount inconsistent.

> great catch! nobody could notice it in practice

Yep --- I only noticed it because I was running a quick-and-dirty vma
merging test and wanted to test on a shmfs file, and noticed that the
temporary shmfs filesystem became unmountable afterwards.  Test
attached, in case anybody is interested (it's the third test, mapping a
file page by page in two interleaved passes, which triggers this case.)

> I'm attaching for review what I'm applying to my -aa tree, to fix the
> above and the other issue with the non-ram vma merging fixed in 2.5.

Looks OK.

Cheers,
 Stephen