linux-ext4 - Re: [PATCH 1/3] xfs/018: remove inline xattr recovery tests

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: <aYWDTrxRhDAE2efB@infradead.org>
Date: Thu, 5 Feb 2026 21:59:42 -0800
From: Christoph Hellwig <hch@...radead.org>
To: "Darrick J. Wong" <djwong@...nel.org>
Cc: Christoph Hellwig <hch@...radead.org>, zlang@...hat.com,
	linux-ext4@...r.kernel.org, fstests@...r.kernel.org
Subject: Re: [PATCH 1/3] xfs/018: remove inline xattr recovery tests

On Thu, Feb 05, 2026 at 08:56:24AM -0800, Darrick J. Wong wrote:
> "In Linux 7.0 we've changed the extended attribute update code to try to
> take a shortcut for performance reasons.  Before walking through the
> attr intent state machine (slow), the update will check to see if the
> attr structure is in short format and will stay in that format after the
> change.  If so, then the incore inode can be updated and logged, and the
> update is complete (fast) in a single transaction.
> 
> "(Obviously, for complex attr structures or large changes we still walk
> through the intent machinery.)
> 
> "However, xfs/018 tests the behavior of the "larp" error injector, which
> only triggers from inside the attr intent state machine.  Therefore, the
> short format tests don't actually trip the injector.  It makes no sense
> to add a new larp injection callsite for the shortcut because either the
> single transaction gets written to disk or it doesn't."

Make sense, but from looking at the test I'm still a bit confused
why it fails (vs just not testing something too useful)

> 
> > Only vaguely related, but should we ensure to always clear error
> > tags after the test runs to ensure they don't leak into other tests?
> 
> They go away with _scratch_remount because error tags only live as long
> as the mount.

True.  While still leaves me puzzled why generic/753 now hits attr log
recovery for me sometimes.