Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 19 Jun 2019 14:32:10 +0200
From:   Pavel Machek <pavel@ucw.cz>
To:     pavel@ucw.cz
Cc:     linux-kernel@vger.kernel.org,
        Randall Huang <huangrandall@google.com>,
        Chao Yu <yuchao0@huawei.com>, Jaegeuk Kim <jaegeuk@kernel.org>,
        Sasha Levin <sashal@kernel.org>
Subject: Re: [PATCH 4.19 33/75] f2fs: fix to avoid accessing xattr across the
 boundary
Message-ID: <20190619123210.GA14477@xo-6d-61-c0.localdomain>
References: <20190617210752.799453599@linuxfoundation.org>
 <20190617210754.076823433@linuxfoundation.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190617210754.076823433@linuxfoundation.org>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi!

> When we traverse xattr entries via __find_xattr(),
> if the raw filesystem content is faked or any hardware failure occurs,
> out-of-bound error can be detected by KASAN.
> Fix the issue by introducing boundary check.

Ok, so this prevents fs corruption from causing problems,

> @@ -340,7 +347,11 @@ static int lookup_all_xattrs(struct inode *inode, struct page *ipage,
>  	else
>  		cur_addr = txattr_addr;
>  
> -	*xe = __find_xattr(cur_addr, index, len, name);
> +	*xe = __find_xattr(cur_addr, last_txattr_addr, index, len, name);
> +	if (!*xe) {
> +		err = -EFAULT;
> +		goto out;
> +	}

Is -EFAULT suitable here? We do not have userspace passing pointers to us, we
have fs corruption. -EUNCLEAN?

Should it do some kind of printk to let the user know fs is corrupted, and mark
it as needing fsck?

Thanks,
									Pavel