Re: Race between vmtruncate and mapped areas?

Rik van Riel (riel@redhat.com)
Wed, 14 May 2003 22:36:23 -0400 (EDT)


On Thu, 15 May 2003, Andrea Arcangeli wrote:

> --- x/include/linux/fs.h.~1~ 2003-05-14 23:26:19.000000000 +0200
> +++ x/include/linux/fs.h 2003-05-15 02:35:57.000000000 +0200
> @@ -421,6 +421,8 @@ struct address_space {
> struct vm_area_struct *i_mmap; /* list of private mappings */
> struct vm_area_struct *i_mmap_shared; /* list of shared mappings */
> spinlock_t i_shared_lock; /* and spinlock protecting it */
> + int truncate_sequence1; /* serialize ->nopage against truncate */
> + int truncate_sequence2; /* serialize ->nopage against truncate */

How about calling them truncate_start and truncate_end ?

> --- x/mm/vmscan.c.~1~ 2003-05-14 23:26:12.000000000 +0200
> +++ x/mm/vmscan.c 2003-05-15 00:22:57.000000000 +0200
> @@ -165,11 +165,10 @@ drop_pte:
> goto drop_pte;
>
> /*
> - * Anonymous buffercache pages can be left behind by
> + * Anonymous buffercache pages can't be left behind by
> * concurrent truncate and pagefault.
> */
> - if (page->buffers)
> - goto preserve;
> + BUG_ON(page->buffers);

I wonder if there is nothing else that can leave behind
buffers in this way.

> + mb(); /* spin_lock has inclusive semantics */
> + if (unlikely(truncate_sequence != mapping->truncate_sequence1)) {
> + struct inode *inode;

This code looks like it should work, but IMHO it is very subtle
so it should really get some documentation.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/