Received-SPF: pass (google.com: domain of linux-kernel+bounces-5021-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: Yuntao Wang <ytcoode@gmail.com>
To: fuqiang.wang@easystack.cn
Cc: bhe@redhat.com,
	dyoung@redhat.com,
	kexec@lists.infradead.org,
	linux-kernel@vger.kernel.org,
	vgoyal@redhat.com,
	ytcoode@gmail.com
Subject: Re: [PATCH] kexec: avoid out of bounds in crash_exclude_mem_range()
Date: Tue, 19 Dec 2023 18:39:28 +0800
Message-ID: <20231219103928.98465-1-ytcoode@gmail.com>
In-Reply-To: <3765549d-892e-4102-9b56-9add1d0a8089@easystack.cn>
References: <3765549d-892e-4102-9b56-9add1d0a8089@easystack.cn>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

On Tue, 19 Dec 2023 16:55:16 +0800, fuqiang wang <fuqiang.wang@easystack.cn> wrote:

> Thank you very much for your patient comment. This change does indeed improve
> readability. But as a combination of these two, how do you feel about moving
> crash_setup_memmap_entries() behind vzalloc().

I don't quite understand what you're trying to express.

> The image->elf_load_addr is determined by arch_kexec_locate_mem_hole(), this
> function can ensure that the value is within the range of [crashk_res.start,
> crashk_res.end), but it seems that it cannot guarantee that its value will
> always be equal to crashk_res.start. Perhaps I have some omissions, please
> point them out.

Because elfcorehdr is the first one and only one that allocates memory from the
starting address of crashk_res, and the starting address of crashk_res meets
the alignment requirement of elfcorehdr.

elfcorehdr requires 4k alignment, and the starting address of crashk_res is
16M aligned.

Therefore, image->elf_load_addr should be equal to crashk_res.start.