linux-kernel - Re: [PATCH v5 07/10] xfs: Commit CoW-based atomic writes atomically

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: <f65f317d-cf00-4a96-9d85-6aa27a95ce17@oracle.com>
Date: Wed, 12 Mar 2025 15:01:42 +0000
From: John Garry <john.g.garry@...cle.com>
To: Christoph Hellwig <hch@...radead.org>
Cc: brauner@...nel.org, djwong@...nel.org, cem@...nel.org,
        linux-xfs@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        linux-kernel@...r.kernel.org, ojaswin@...ux.ibm.com,
        ritesh.list@...il.com, martin.petersen@...cle.com,
        catherine.hoang@...cle.com
Subject: Re: [PATCH v5 07/10] xfs: Commit CoW-based atomic writes atomically

On 12/03/2025 13:54, Christoph Hellwig wrote:

+

> On Wed, Mar 12, 2025 at 09:04:07AM +0000, John Garry wrote:
>>> As already mentioned in a previous reply:  "all" might be to much.
>>> The code can only support a (relatively low) number of extents
>>> in a single transaction safely.
>>
>> Then we would need to limit the awu max to whatever can be guaranteed
>> (to fit).
> 
> Yes.  And please add a testcase that creates a badly fragmented file
> and verifies that we can handle the worst case for this limit.

ok, we can do that.

I have my own stress test for atomic writes on fragmented FSes, but I 
have not encountered a such a problem. But we need to formalize 
something though.


> 
> (although being able to reproduce the worst case btree splits might
> be hard, but at least the worst case fragmentation should be doable)
> 
>>> Assuming we could actually to the multi extent per transaction
>>> commit safely, what would be the reason to not always do it?
>>>
>>
>> Yes, I suppose that it could always be used. I would suggest that as a later
>> improvement, if you agree.
> 
> I remember running into some problems with my earlier version, but I'd
> have to dig into it.  Maybe it will resurface with the above testing,
> or it was due to my optimizations for the extent lookups.
> 

ok.

Thanks,
John