Re: [PATCH] scsi: target: iscsi: set memalloc_noio with loopback network connections

From: Laurence Oberman
Date: Mon Feb 13 2023 - 11:27:30 EST


On Mon, 2023-02-13 at 10:22 -0600, Mike Christie wrote:
> On 2/13/23 5:59 AM, Maurizio Lombardi wrote:
> > st 8. 2. 2023 v 21:10 odesílatel David Jeffery <djeffery@xxxxxxxxxx
> > > napsal:
> > >
> > > + /*
> > > + * If the iscsi connection is over a loopback device from
> > > using
> > > + * iscsi and iscsit on the same system, we need to set
> > > memalloc_noio to
> > > + * prevent memory allocation deadlocks between target and
> > > initiator.
> > > + */
> > > + rcu_read_lock();
> > > + dst = rcu_dereference(conn->sock->sk->sk_dst_cache);
> > > + if (dst && dst->dev && dst->dev->flags & IFF_LOOPBACK)
> > > + loopback = true;
> > > + rcu_read_unlock();
> >
> > Hi Mike,
> > I tested it, it works. The customer also confirmed that it fixes
> > the
> > deadlock on his setup.
>
> You never responded about why/how it's used in production. Is it some
> sort
> of clustering or container or what?
>
> The login related code can still swing back on you if it's run for a
> relogin.
> It would happen if we overqueue and a nop timesout because the iscsi
> recv thread
> is waiting for backend resources like a request/queue slot, or if
> management tools
> disable/enable the tpgt for reconfigs, etc.
>

Hi Mike,
The use case described is as follows:


"This customer moved their on-premise system to the cloud.
Their on-premise system runs with two servers and one external storage
and uses data mirroring software to mirror data.

When moving to the cloud, customer wanted to implement a data mirror
using data mirror software with two instances to reduce the cost of
using the cloud infrastructure.
To build a system with two instances, we use iSCSI to mirror data
between a local disk on one instance and a local disk on the other
instance.
We coexist iSCSI initiator and target so that data mirroring software
can access each disk through a unified interface."

Thanks
Laurence