Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 29 Sep 2021 12:54:48 +0100
From:   Will Deacon <will@kernel.org>
To:     yee.lee@mediatek.com
Cc:     linux-kernel@vger.kernel.org, nicholas.Tang@mediatek.com,
        Kuan-Ying.lee@mediatek.com, chinwen.chang@mediatek.com,
        Matthias Brugger <matthias.bgg@gmail.com>,
        Sami Tolvanen <samitolvanen@google.com>,
        "moderated list:ARM/Mediatek SoC support" 
        <linux-arm-kernel@lists.infradead.org>,
        "moderated list:ARM/Mediatek SoC support" 
        <linux-mediatek@lists.infradead.org>
Subject: Re: [PATCH] scs: Release kasan vmalloc poison in scs_free process
Message-ID: <20210929115447.GA21631@willie-the-truck>
References: <20210923095316.13867-1-yee.lee@mediatek.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210923095316.13867-1-yee.lee@mediatek.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Thu, Sep 23, 2021 at 05:53:12PM +0800, yee.lee@mediatek.com wrote:
> From: Yee Lee <yee.lee@mediatek.com>
> 
> Since scs allocation has been moved to vmalloc region, the
> shadow stack is protected by kasan_posion_vmalloc.
> However, the vfree_atomic operation needs to access
> its context for scs_free process and causes kasan error
> as the dump info below.
> 
> This patch Adds kasan_unpoison_vmalloc() before vfree_atomic,
> which aligns to the prior flow as using kmem_cache.
> The vmalloc region will go back posioned in the following
> vumap() operations.
> 
>  ==================================================================
>  BUG: KASAN: vmalloc-out-of-bounds in llist_add_batch+0x60/0xd4
>  Write of size 8 at addr ffff8000100b9000 by task kthreadd/2
> 
>  CPU: 0 PID: 2 Comm: kthreadd Not tainted 5.15.0-rc2-11681-(skip)
>  Hardware name: linux,dummy-virt (DT)
>  Call trace:
>   dump_backtrace+0x0/0x43c
>   show_stack+0x1c/0x2c
>   dump_stack_lvl+0x68/0x84
>   print_address_description+0x80/0x394
>   kasan_report+0x180/0x1dc
>   __asan_report_store8_noabort+0x48/0x58
>   llist_add_batch+0x60/0xd4
>   vfree_atomic+0x60/0xe0
>   scs_free+0x1dc/0x1fc
>   scs_release+0xa4/0xd4
>   free_task+0x30/0xe4

Thanks, I can reproduce this easily with mainline. We only recently gained
vmalloc support for kasan on arm64, so that's probably why we didn't spot
this earlier.

> diff --git a/kernel/scs.c b/kernel/scs.c
> index e2a71fc82fa0..25c0d8e416e6 100644
> --- a/kernel/scs.c
> +++ b/kernel/scs.c
> @@ -68,6 +68,7 @@ void scs_free(void *s)
>  
>  	__scs_account(s, -1);
>  
> +	kasan_unpoison_vmalloc(s, SCS_SIZE);
>  	/*
>  	 * We cannot sleep as this can be called in interrupt context,
>  	 * so use this_cpu_cmpxchg to update the cache, and vfree_atomic

I don't think we should be unpoisoning the stack pages if we're putting
them back on to the local cache -- unpoisoning happens on the alloc path
in that case anyway so that we can zero them.

So how about sending a v2 with this moved a bit later (see below)? With
that change:

Acked-by: Will Deacon <will@kernel.org>
Tested-by: Will Deacon <will@kernel.org>

Thanks,

Will

--->8

diff --git a/kernel/scs.c b/kernel/scs.c
index e2a71fc82fa0..579841be8864 100644
--- a/kernel/scs.c
+++ b/kernel/scs.c
@@ -78,6 +78,7 @@ void scs_free(void *s)
                if (this_cpu_cmpxchg(scs_cache[i], 0, s) == NULL)
                        return;
 
+       kasan_unpoison_vmalloc(s, SCS_SIZE);
        vfree_atomic(s);
 }