[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250907110506.czjh3td4qf5ieb4m@pali>
Date: Sun, 7 Sep 2025 13:05:06 +0200
From: Pali Rohár <pali@...nel.org>
To: Stefan Metzmacher <metze@...ba.org>
Cc: Steve French <sfrench@...ba.org>, Paulo Alcantara <pc@...guebit.com>,
ronnie sahlberg <ronniesahlberg@...il.com>,
Ralph Böhme <slow@...ba.org>,
linux-cifs@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 00/35] cifs: Fix SMB rmdir() and unlink() against Windows
SMB servers
On Tuesday 02 September 2025 18:30:53 Pali Rohár wrote:
> On Tuesday 02 September 2025 17:17:14 Stefan Metzmacher wrote:
> > Do you have network captures showing the old and new behavior
> > that's often easier to understand than looking at patches alone.
> >
> > metze
>
> I do not have them right now, but I can run test scenario and capture
> them, this is not problem. Test case is pretty straightforward.
In attachments are network traces of both the old and new behaviors.
In the old behavior is visible that after calling "rm object", the
object is still in the followup "ls" output and calling "mkdir object"
is failing, also "stat object" is failing.
In the new behavior is visible that "rm object" is using exclusive
removal, which fails, and then fallback to rename+deletepending which
success. After that in the followup "ls" output the object entry is not
there, there is only renamed ".smb<num>", and "mkdir object" pass and
creates new directory "object".
Download attachment "smb_old.pcap.gz" of type "application/gzip" (4848 bytes)
Download attachment "smb_new.pcap.gz" of type "application/gzip" (5488 bytes)
Powered by blists - more mailing lists