Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.34 as permitted sender) client-ip=23.128.96.34;
Date:   Fri, 15 Sep 2023 11:01:47 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Jinjie Ruan <ruanjinjie@huawei.com>
Cc:     tglx@linutronix.de, linux-kernel@vger.kernel.org,
        Christophe Leroy <christophe.leroy@csgroup.eu>,
        Josh Poimboeuf <jpoimboe@kernel.org>
Subject: Re: [PATCH] static_call: Fix a wild-memory-access bug in
 static_call_del_module()
Message-ID: <20230915090147.GB6721@noisy.programming.kicks-ass.net>
References: <20230915082126.4187913-1-ruanjinjie@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230915082126.4187913-1-ruanjinjie@huawei.com>
Precedence: bulk

On Fri, Sep 15, 2023 at 04:21:25PM +0800, Jinjie Ruan wrote:
> Inject fault while probing btrfs.ko, if the first kzalloc() fails
> in __static_call_init(), key->mods will no be initialized. And then
> in static_call_del_module() the site_mod->mod will cause
> wild-memory-access as below:
> 
> So assign key->mods to NULL in __static_call_init() if it fails
> to fix the issue. And if kzalloc fails, it will just return in init
> func, so it should break if it the key->mods is NULL in exit func.
> 

I don't think we need that full splat.

> 
> Fixes: 8fd4ddda2f49 ("static_call: Don't make __static_call_return0 static")

And that looks wrong, that just moved code around.

> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
> ---
>  kernel/static_call_inline.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/static_call_inline.c b/kernel/static_call_inline.c
> index 639397b5491c..e7aa70d33530 100644
> --- a/kernel/static_call_inline.c
> +++ b/kernel/static_call_inline.c
> @@ -256,8 +256,10 @@ static int __static_call_init(struct module *mod,
>  			}
>  
>  			site_mod = kzalloc(sizeof(*site_mod), GFP_KERNEL);
> -			if (!site_mod)
> +			if (!site_mod) {
> +				key->mods = NULL;
>  				return -ENOMEM;
> +			}
>  
>  			/*
>  			 * When the key has a direct sites pointer, extract
> @@ -422,7 +424,7 @@ static void static_call_del_module(struct module *mod)
>  			;
>  
>  		if (!site_mod)
> -			continue;
> +			break;
>  
>  		*prev = site_mod->next;
>  		kfree(site_mod);

The actual patch looks okay.