linux-kernel - Re: [RFC PATCH] mm/vmalloc: make sure to traverse from the beginning when overflow occur

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <b295ec51-56fd-a377-95c9-a88cf5b151bc@intel.com>
Date:   Mon, 13 Apr 2020 08:02:49 -0700
From:   Dave Hansen <dave.hansen@...el.com>
To:     Liu Song <fishland@...yun.com>, gregkh@...uxfoundation.org,
        jroedel@...e.de, dave.hansen@...ux.intel.com, tglx@...utronix.de
Cc:     linux-mm@...ck.org, linux-kernel@...r.kernel.org,
        liu.song11@....com.cn
Subject: Re: [RFC PATCH] mm/vmalloc: make sure to traverse from the beginning
 when overflow occur

On 4/12/20 7:38 AM, Liu Song wrote:
> If overflow, should ensure that "free_vmap_cache" is set to NULL,
> so as to ensure that it can be traversed from the beginning.

This changelog is a bit sparse.

Does this fix a demonstrated problem?  Or was it just discovered via
code review and assumed to help a theoretical problem?

Which tree is it against?  'free_vmap_cache' doesn't show up in Linus's
tree.
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index d8e877365f9f..2638a20d36ce 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -441,7 +441,7 @@ static struct vmap_area *alloc_vmap_area(unsigned long size,
>  	if (!free_vmap_cache ||
>  			size < cached_hole_size ||
>  			vstart < cached_vstart ||
> -			align < cached_align) {
> +			align < cached_align || purged) {
>  nocache:
>  		cached_hole_size = 0;
>  		free_vmap_cache = NULL;

Is 'purged' the right way to do this?

There is a path through the overflow code that also sets purged=0.  Does
'free_vmap_cache' *not* need to get reset in that case?

It also seems more natural to me that the code mucking with the rbtree
would be the one to reset the cache.