linux-kernel - Re: [GIT PULL] non-MM updates for 6.19-rc1

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <nrgonfemkkz2bnwfkpzycfbz3luspjm65dakyjt4e4nuawklbw@cw3jqnedbyjf>
Date: Mon, 8 Dec 2025 15:58:44 +0800
From: Heming Zhao <heming.zhao@...e.com>
To: Joseph Qi <joseph.qi@...ux.alibaba.com>
Cc: Linus Torvalds <torvalds@...uxfoundation.org>, 
	Andrew Morton <akpm@...ux-foundation.org>, Dmitry Antipov <dmantipov@...dex.ru>, 
	Mark Fasheh <mark@...heh.com>, linux-kernel@...r.kernel.org, linux-mm@...ck.org, 
	mm-commits@...r.kernel.org, Pasha Tatashin <pasha.tatashin@...een.com>
Subject: Re: [GIT PULL] non-MM updates for 6.19-rc1

On Mon, Dec 08, 2025 at 12:01:09PM +0800, Joseph Qi wrote:
> 
> 
> On 2025/12/8 10:21, Linus Torvalds wrote:
> > On Mon, 8 Dec 2025 at 10:11, Joseph Qi <joseph.qi@...ux.alibaba.com> wrote:
> >>
> >> On 2025/12/7 07:40, Linus Torvalds wrote:
> >>>
> >>> See commit c06c303832ec ("ocfs2: fix xattr array entry __counted_by
> >>> error") which simply makes the 'last' pointer point to the actual last
> >>> entry, rather than point to one _past_ the last entry.
> >>
> >> It looks fine to me.
> >> Sorry for the noise and thank you for pointing out my mistake.
> > 
> > Note that if my fix is correct, the mistake is most certainly not
> > yours and goes all the way back to 2009, and commit 0fe9b66c65f3
> > ("ocfs2: Add preserve to reflink").
> > 
> > But I would really love to have somebody who knows that code go
> > through it, and see if my fix actually is the right thing to do.
> > 
> 
> Go through the code logic again, maybe I'm wrong before.
> 
> Take the following flow for example:
> 
> ocfs2_reflink_xattr_in_block
>   ocfs2_create_empty_xattr_block
>     ocfs2_create_xattr_block
>       memset(xblk, 0, inode->i_sb->s_blocksize);
>   ocfs2_reflink_xattr_block
>     ocfs2_reflink_xattr_header

In ocfs2_reflink_xattr_block(), the call to memcpy(new_blk_bh->.., blk_bh->.., ...)
occurs before ocfs2_reflink_xattr_header(). Therefore, the new_xh->xh_count is
very likely not ZERO.

> 
> It seems now new_xh->xh_count is 0. That's means, it is trying to add a
> new entry and here 'last' is pointing the next free entry.
> If so, the original code seems right.
> 
> Heming, could you please verify this? I'm not familiar with this logic...
> 
> Thanks,
> Joseph

In ocfs2_reflink_xattrs(), when preserve_security is false, the function sets
'args.xattr_reflinked'. IMUC, it means that if users want to preserve security
xattr items, the args.xattr_reflinked is NULL, this then results in
ocfs2_reflink_xattr_header() not executing the cleanup jobs, (in the code area
defined by "if (args->xattr_reflinked && !args->xattr_reflinked(xe))").

Regarding the clean jobs:
- by default, it only handles three types:
  OCFS2_XATTR_INDEX_[SECURITY|POSIX_ACL_ACCESS|POSIX_ACL_DEFAULT]
- it cleans the first item and shifts forward the remaining items.

It seems Linus' patch does not handle all cases. If new_xh->xh_count is 1,
le16_add_cpu(&new_xh->xh_count, -1) makes it to 0, and it will miss the cleanup
job.

Thanks,
Heming