From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mike Christie Date: Mon, 21 May 2018 19:45:31 +0000 Subject: Re: tcmu data consistency Message-Id: <5B0321DB.9090906@redhat.com> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit To: target-devel@vger.kernel.org On 05/20/2018 04:10 AM, 亓武强 wrote: > hi： > i have some questions about the tcmu, consider following scenario： > > 1，Iscsi initiator establishes two session connection with gw1,gw2 at > the same time. gw1 is the active path. > > 2, The active path is switched to gw 2 due to the interruption of the > connection from the initiator to gw1. > > 3, gw2 acquire exclusive lock and write some data to lun0. then gw1 > update the exclusive lock to UNLOCKED. > > 4, after step 3, some old iscsi session inflight write ios > arrival gw1 , gw1 acquire the exclusive lock again, the old data will > overwrite new data. With iscsi it should not happen when using explicit failover mode. We will process all outstanding IO before the STPG, so inflight writes would be failed then we would grab the lock. In implicit mode it could happen.