linux-kernel - Re: [PATCH 2/2] vfree: Update documentation

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200922152240.GI32101@casper.infradead.org>
Date:   Tue, 22 Sep 2020 16:22:40 +0100
From:   Matthew Wilcox <willy@...radead.org>
To:     Christoph Hellwig <hch@....de>
Cc:     Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, Uladzislau Rezki <urezki@...il.com>
Subject: Re: [PATCH 2/2] vfree: Update documentation

On Tue, Sep 22, 2020 at 05:09:10PM +0200, Christoph Hellwig wrote:
> On Tue, Sep 22, 2020 at 04:06:03PM +0100, Matthew Wilcox wrote:
> > I don't think it makes sense to list all vmalloc-style allocators here.
> > It won't be updated by people who add new variations.  How about this?
> > 
> >  * Free the virtually continuous memory area starting at @addr, as
> >  * obtained from one of the vmalloc() family of APIs.  This will
> >  * usually also free the physical memory underlying the virtual
> >  * allocation, but that memory is reference counted, so it will not
> >  * be freed until the last user goes away.
> >  *
> >  * If @addr is NULL, no operation is performed.
> > 
> > I'm trying to strike a balance between being accurate and not requiring
> > device driver authors to learn all about struct page.  I may be too
> > close to the implementation to write good documentation for it.
> 
> I think the above is sensible, but not enough.  vmap really needs to
> be treated special, as by default area->pages for vmap is NULL.  So
> for vfree to be useful on a vmap mapping, the callers needs to
> manually set it up by poking into the internals.  Actually, I think
> we really want another API rather than vmap for that.  Let me respin
> my series to include that.

I've been thinking about somethng like:

void *vmap_mapping(struct address_space *mapping, pgoff_t start,
		unsigned long len);

but it doesn't quite work for the shmem cases because they need to use
shmem_read_mapping_page_gfp() instead of ->readpage.  I'd also want it
to work for DAX, but I don't have a user for that yet so it's hard to
justify adding it.

I have ideas for making shmem/tmpfs work more like other filesystems,
so we don't need shmem_read_mapping_page_gfp() but there are only so
many hours in the day.