Currently if ipcomp_alloc_scratches() fails to allocate memory
ipcomp_scratches holds obsolete address. So when we try to free the
percpu scratches using ipcomp_free_scratches() it tries to vfree non
existent vm area. Described below:

static void * __percpu *ipcomp_alloc_scratches(void)
{
        ...
        scratches = alloc_percpu(void *);
        if (!scratches)
                return NULL;
ipcomp_scratches does not know about this allocation failure.
Therefore holding the old obsolete address.
        ...
}

So when we free,

static void ipcomp_free_scratches(void)
{
        ...
        scratches = ipcomp_scratches;
Assigning obsolete addresses from ipcomp_scratches

        if (!scratches)
                return;

        for_each_possible_cpu(i)
               vfree(*per_cpu_ptr(scratches, i));
Trying to free non existent page, causing warning: trying to vfree
existent vm area.
        ...
}


Fix this breakage by: 
(1) Update ipcomp_scratches with NULL if the above mentioned 
allocation fails.
(2) Update ipcomp_scrtches with NULL when scratches is freed

Reported-by: syzbot+5ec9bb042ddfe9644773@syzkaller.appspotmail.com
Tested-by: syzbot+5ec9bb042ddfe9644773@syzkaller.appspotmail.com
Signed-off-by: Khalid Masum <khalid.masum.92@gmail.com>
---
Changes since v2:
- Set ipcomp_scratches to NULL when scratches is freed.
- Update commit message.
- v2 Link: https://lore.kernel.org/lkml/20220831142938.5882-1-khalid.masum.92@gmail.com/

Changes since v1:
- Instead of altering usercount, update ipcomp_scratches to NULL
- Update commit message.
- v1 Link: https://lore.kernel.org/lkml/20220831014126.6708-1-khalid.masum.92@gmail.com/

diff --git a/net/xfrm/xfrm_ipcomp.c b/net/xfrm/xfrm_ipcomp.c
index cb40ff0ff28d..3774d07c5819 100644
--- a/net/xfrm/xfrm_ipcomp.c
+++ b/net/xfrm/xfrm_ipcomp.c
@@ -203,6 +203,7 @@ static void ipcomp_free_scratches(void)
 		vfree(*per_cpu_ptr(scratches, i));
 
 	free_percpu(scratches);
+	ipcomp_scratches = NULL;
 }
 
 static void * __percpu *ipcomp_alloc_scratches(void)
@@ -215,7 +216,7 @@ static void * __percpu *ipcomp_alloc_scratches(void)
 
 	scratches = alloc_percpu(void *);
 	if (!scratches)
-		return NULL;
+		return ipcomp_scratches = NULL;
 
 	ipcomp_scratches = scratches;

On Thu, Sep 01, 2022 at 10:03:07AM +0600, Khalid Masum wrote:
> 
> diff --git a/net/xfrm/xfrm_ipcomp.c b/net/xfrm/xfrm_ipcomp.c
> index cb40ff0ff28d..3774d07c5819 100644
> --- a/net/xfrm/xfrm_ipcomp.c
> +++ b/net/xfrm/xfrm_ipcomp.c
> @@ -203,6 +203,7 @@ static void ipcomp_free_scratches(void)
>  		vfree(*per_cpu_ptr(scratches, i));
>  
>  	free_percpu(scratches);
> +	ipcomp_scratches = NULL;
>  }

Good catch! This is probably the root cause of all the crashes.

>  static void * __percpu *ipcomp_alloc_scratches(void)
> @@ -215,7 +216,7 @@ static void * __percpu *ipcomp_alloc_scratches(void)
>  
>  	scratches = alloc_percpu(void *);
>  	if (!scratches)
> -		return NULL;
> +		return ipcomp_scratches = NULL;

This is unnecessary as with your first hunk, ipcomp_scratches
is guaranteed to be NULL.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

On Thu, Sep 1, 2022 at 10:18 AM Herbert Xu <herbert@gondor.apana.org.au> wrote:
>
> On Thu, Sep 01, 2022 at 10:03:07AM +0600, Khalid Masum wrote:
> >
> > diff --git a/net/xfrm/xfrm_ipcomp.c b/net/xfrm/xfrm_ipcomp.c
> > index cb40ff0ff28d..3774d07c5819 100644
> > --- a/net/xfrm/xfrm_ipcomp.c
> > +++ b/net/xfrm/xfrm_ipcomp.c
> > @@ -203,6 +203,7 @@ static void ipcomp_free_scratches(void)
> >               vfree(*per_cpu_ptr(scratches, i));
> >
> >       free_percpu(scratches);
> > +     ipcomp_scratches = NULL;
> >  }
>
> Good catch! This is probably the root cause of all the crashes.
>
> >  static void * __percpu *ipcomp_alloc_scratches(void)
> > @@ -215,7 +216,7 @@ static void * __percpu *ipcomp_alloc_scratches(void)
> >
> >       scratches = alloc_percpu(void *);
> >       if (!scratches)
> > -             return NULL;
> > +             return ipcomp_scratches = NULL;
>
> This is unnecessary as with your first hunk, ipcomp_scratches
> is guaranteed to be NULL.
>
> Thanks,
> --

You are right. Instead of setting it to NULL at both places, it makes
more sense to
do it when memory is freed.

I shall send a v4 with the suggested change.

thanks,
 -- Khalid Masum