linux-kernel - Re: [PATCH 1/5] locking: Add rwsem_is_write

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: <ZPoift7B3UDQgmWB@casper.infradead.org>
Date:   Thu, 7 Sep 2023 20:20:30 +0100
From:   Matthew Wilcox <willy@...radead.org>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Ingo Molnar <mingo@...hat.com>, Will Deacon <will@...nel.org>,
        Waiman Long <longman@...hat.com>, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, Chandan Babu R <chandan.babu@...cle.com>,
        "Darrick J . Wong" <djwong@...nel.org>, linux-xfs@...r.kernel.org
Subject: Re: [PATCH 1/5] locking: Add rwsem_is_write_locked()

On Thu, Sep 07, 2023 at 09:08:10PM +0200, Peter Zijlstra wrote:
> On Thu, Sep 07, 2023 at 06:47:01PM +0100, Matthew Wilcox (Oracle) wrote:
> > Several places want to know whether the lock is held by a writer, instead
> > of just whether it's held.  We can implement this for both normal and
> > rt rwsems.  RWSEM_WRITER_LOCKED is declared in rwsem.c and exposing
> > it outside that file might tempt other people to use it, so just use
> > a comment to note that's what the 1 means, and help anybody find it if
> > they're looking to change the implementation.
> 
> I'm presuming this is deep in a callchain where they know they hold the
> lock, but they lost in what capacity?

No, it's just assertions.  You can see that in patch 3 where it's
used in functions called things like "xfs_islocked".

> In general I strongly dislike the whole _is_locked family, because it
> gives very poorly defined semantics if used by anybody but the owner.
> 
> If these new functions are indeed to be used only by lock holders to
> determine what kind of lock they hold, could we please put:
> 
> 	lockdep_assert_held()
> 
> in them?

Patch 2 shows it in use in the MM code.  We already have a
lockdep_assert_held_write(), but most people don't enable lockdep, so
we also have VM_BUG_ON_MM(!rwsem_is_write_locked(&mm->mmap_lock), mm)
to give us a good assertion when lockdep is disabled.

XFS has a problem with using lockdep in general, which is that a worker
thread can be spawned and use the fact that the spawner is holding the
lock.  There's no mechanism for the worker thread to ask "Does struct
task_struct *p hold the lock?".