Re: [PATCH 1/1] mm/vmalloc.c: Add an error message if two areas overlap

From: Uladzislau Rezki <urezki@gmail.com>
To: Matthew Wilcox <willy@infradead.org>
Cc: "Uladzislau Rezki (Sony)" <urezki@gmail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	linux-mm@kvack.org, LKML <linux-kernel@vger.kernel.org>,
	Hillf Danton <hdanton@sina.com>, Michal Hocko <mhocko@suse.com>,
	Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>,
	Steven Rostedt <rostedt@goodmis.org>
Subject: Re: [PATCH 1/1] mm/vmalloc.c: Add an error message if two areas overlap
Date: Fri, 10 Jul 2020 22:00:30 +0200	[thread overview]
Message-ID: <20200710200030.GB19487@pc636> (raw)
In-Reply-To: <20200710194406.GP12769@casper.infradead.org>

On Fri, Jul 10, 2020 at 08:44:06PM +0100, Matthew Wilcox wrote:
> On Fri, Jul 10, 2020 at 09:40:42PM +0200, Uladzislau Rezki (Sony) wrote:
> > Before triggering a BUG() it would be useful to understand
> > how two areas overlap between each other. Print information
> > about start/end addresses of both VAs and their addresses.
> > 
> > For example if both are identical it could mean double free.
> > 
> > Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
> > ---
> >  mm/vmalloc.c | 10 +++++++++-
> >  1 file changed, 9 insertions(+), 1 deletion(-)
> > 
> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index 5a2b55c8dd9a..1679b01febcd 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -549,8 +549,13 @@ find_va_links(struct vmap_area *va,
> >  		else if (va->va_end > tmp_va->va_start &&
> >  				va->va_start >= tmp_va->va_end)
> >  			link = &(*link)->rb_right;
> > -		else
> > +		else {
> > +			pr_err("Overlaps: 0x%px(0x%lx-0x%lx), 0x%px(0x%lx-0x%lx)\n",
> > +				va, va->va_start, va->va_end, tmp_va,
> > +				tmp_va->va_start, tmp_va->va_end);
> 
> It might be helpful to have a "vmalloc:" prefix to that string to indicate
> where to start searching.  And I don't think we're supposed to use %px
> without a really good justification these days.
>
That makes sense, i will add such prefix for sure. As for %px i can just
use a casting to (unsigned long) and 0x%lx prefix.

> 
> Can we do without the BUG()?  Maybe just break out and decline to insert
> the conflicting range?
> 
When i developed that logic, i was thinking about that but for some
reason i went with BUG_ON(), do not remember why. But i agree we shold
avoid of it. Let me think about it one more time.

> >  			BUG();
> > +		}
> >  	} while (*link);
> >  
> >  	*parent = &tmp_va->rb_node;
> > @@ -1993,6 +1998,9 @@ static void vmap_init_free_space(void)
> >  			insert_vmap_area_augment(free, NULL,
> >  				&free_vmap_area_root,
> >  					&free_vmap_area_list);
> > +			insert_vmap_area_augment(free, NULL,
> > +				&free_vmap_area_root,
> > +					&free_vmap_area_list);
> 
> This is surely testing code that you forgot to delete?
>
Yes, i forgot to remove it.

Thanks for your comments!

--
Vlad Rezki