[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <874cce7f-4545-450f-9d6b-4f3e3357281b@linux.alibaba.com>
Date: Mon, 18 Dec 2023 11:34:59 +0800
From: Baolin Wang <baolin.wang@...ux.alibaba.com>
To: Matthew Wilcox <willy@...radead.org>
Cc: akpm@...ux-foundation.org, david@...hat.com, ying.huang@...el.com,
ziy@...dia.com, xuyu@...ux.alibaba.com, linux-mm@...ck.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm: migrate: fix getting incorrect page mapping during
page migration
On 12/15/2023 11:01 PM, Matthew Wilcox wrote:
> On Fri, Dec 15, 2023 at 02:51:43PM +0000, Matthew Wilcox wrote:
>> I'm not saying no to this fix, but dump_mapping() is supposed to be
>> resilient against this. Is the issue that 'dentry' is NULL, or is it
>> some field within dentry that is NULL? eg, would this fix your
>> case?
>
> Uh, dentry is an on-stack dentry. So obviously it's a pointer within it
> that's NULL. Maybe this, having stared at the implementation of %pd?
Yes, I think below check can avoid crash of dentry_name() used by '%pd'.
Maybe we should also initialize on-stack dentry by
struct dentry dentry = { };
I can send a patch with your suggestion. Thanks.
> +++ b/fs/inode.c
> @@ -588,7 +588,8 @@ void dump_mapping(const struct address_space *mapping)
> }
>
> dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
> - if (get_kernel_nofault(dentry, dentry_ptr)) {
> + if (get_kernel_nofault(dentry, dentry_ptr) ||
> + !dentry->d_parent || !dentry->d_name) {
> pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
> a_ops, ino, dentry_ptr);
> return;
Powered by blists - more mailing lists