Re: [PATCH] scsi: scsi_dh_alua: set handler_data before alua_initialize()
From: Laurence Oberman
Date: Sun Nov 15 2026 - 14:38:08 EST
On Sun, 2026-11-15 at 18:04 +0000, John Garry wrote:
> On 11/9/26 20:11, Laurence Oberman wrote:
> > Since commit 50759b881e1d ("scsi: device_handler: alua: Call
> > scsi_device_put() from non-atomic context"), alua_rtpg_queue() only
> > queues an RTPG when sdev->handler_data points at an alua_dh_data
> > whose
> > port group matches:
> >
> > struct alua_dh_data *h = sdev->handler_data;
> > ...
> > if (h && rcu_dereference(h->pg) == pg) {
>
> Do we still require this check (for h being non-NULL) after the
> change
> in this patch? I assume that we flush any calls to this function
> prior
> to unsetting sdev->handler_data.
>
> >
> > However alua_bus_attach() only assigns sdev->handler_data after
> > alua_initialize() returns, and alua_initialize() ->
> > alua_check_vpd()
> > is where the initial RTPG is queued. At that point handler_data is
> > still NULL, so the initial RTPG is silently dropped: no RTPG is
> > sent,
> > nothing is logged, and sdev->access_state keeps its zeroed default
> > of
> > active/optimized for every path.
> >
> > With dm-multipath this goes unnoticed, as pg_init calls
> > scsi_dh_activate() once the map is loaded, which queues an RTPG
> > after
> > handler_data is set, and multipathd's alua prioritizer issues its
> > own
> > RTPG from userspace. Anything relying on sdev->access_state without
> > a
> > subsequent activate (sysfs access_state/preferred_path, native SCSI
> > multipath path selection) sees every path as active/optimized until
> > a
> > Unit Attention or rescan triggers an RTPG.
> >
> > Seen on an HP P2000 G3 FC array (implicit ALUA, two port groups,
> > four
> > paths per LUN) with dm-multipath not in use: sg_rtpg reports port
> > group
> > 0 active/non-optimized and port group 1 active/optimized, but
> > /sys/class/scsi_device/H:C:T:L/device/access_state reads
> > active/optimized on all four paths after boot, and no
> > "alua: port group XX state ..." message is logged. Writing
> > "activate"
> > to dh_state sends the RTPG and corrects access_state.
> >
> > Kernels without commit 50759b881e1d queue the attach-time RTPG
> > unconditionally and are not affected.
> >
> > Fix by assigning sdev->handler_data before calling
> > alua_initialize(),
> > and clearing it again on failure. sdev->handler is only set once
> > scsi_dh_handler_attach() sees ->attach() succeed, so no handler
> > callback can observe the partially initialised alua_dh_data. On the
> > failure paths of alua_initialize(), h has not been added to a port
> > group and no RTPG has been queued, so freeing it is unchanged.
> >
> > Fixes: 50759b881e1d ("scsi: device_handler: alua: Call
> > scsi_device_put() from non-atomic context")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Laurence Oberman <loberman@xxxxxxxxxx>
> > ---
> >
> > Notes:
> > Tested on an HP P2000 G3 FC array (30 LUNs, 4 paths each):
> > with this
> > patch the "alua: port group 00 state N ..." messages are
> > logged during
> > boot and access_state shows two active/optimized and two
> > active/non-optimized paths per LUN without any manual
> > activate.
> > Without it, all paths read active/optimized until an RTPG is
> > triggered.
> >
> > I found this while testing John Garry's native SCSI multipath
> > series
> > (v9) on an HP P2000 G3 FC array (30 LUNs, 4 paths each, dm-
> > multipath
> > not in use). Native multipath uses sdev->access_state for path
> > selection, and every path showed active/optimized, although
> > sg_rtpg
> > and dm-multipath (alua prio 50/10) both report port group 0 as
> > active/non-optimized. No "alua: port group XX state ..."
> > message was
> > logged at boot. Writing "activate" to dh_state corrected
> > access_state,
> > which pointed at the attach-time RTPG never being queued.
> >
> > With this patch, the port group state messages are logged
> > during boot
> > and access_state shows two active/optimized and two
> > active/non-optimized paths per LUN without any manual
> > activate.
> >
> > This fix is independent of the native multipath series. It
> > affects any
> > user of access_state/preferred_path, and ALUA prep_fn fast-
> > fail, on
> > kernels with 50759b881e1d where nothing calls
> > scsi_dh_activate()
> > after attach.
> >
> > v9 series:
> > https://lore.kernel.org/linux-scsi/20261108204733.557995-1-
> > john.garry@xxxxxxxxx/
> >
> > drivers/scsi/device_handler/scsi_dh_alua.c | 9 ++++++++-
> > 1 file changed, 8 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/scsi/device_handler/scsi_dh_alua.c
> > b/drivers/scsi/device_handler/scsi_dh_alua.c
> > index 87ac0a6d63e5..84ff3aed2c46 100644
> > --- a/drivers/scsi/device_handler/scsi_dh_alua.c
> > +++ b/drivers/scsi/device_handler/scsi_dh_alua.c
> > @@ -1233,13 +1233,20 @@ static int alua_bus_attach(struct
> > scsi_device *sdev)
> > INIT_LIST_HEAD(&h->node);
> >
> > mutex_init(&h->init_mutex);
> > + /*
> > + * alua_rtpg_queue() only queues an RTPG when sdev-
> > >handler_data
> > + * points at this alua_dh_data, so set it before
> > alua_initialize()
> > + * otherwise the initial RTPG issued from alua_check_vpd()
> > is
> > + * silently dropped and sdev->access_state is never
> > populated.
> > + */
> > + sdev->handler_data = h;
>
> Note: I was checking which members of alua_dh_data are not
> initialized
> prior to publishing sdev->handler_data, and alua_dh_data.group_id is
> never accessed at all AFAICS.
>
> Thanks,
> John
On 11/15/26, John Garry wrote:
> Do we still require this check (for h being non-NULL) after the
change
> in this patch? I assume that we flush any calls to this function
prior
> to unsetting sdev->handler_data.
Hi John,
Thanks for looking at this.
You're right, with this patch the NULL check is redundant. All three
callers already have a valid h:
- alua_check_vpd() is called from attach/rescan with h passed in, and
sdev->handler_data now points at it before the initial RTPG is
queued
- alua_activate() and alua_check() both dereference
sdev->handler_data (h->init_mutex, h->pg) before calling
alua_rtpg_queue()
There is no explicit flush, but there doesn't need to be:
handler_data is only cleared in alua_bus_detach(), which runs from
scsi_dh_release_device() in scsi_device_dev_release(), i.e. after the
last sdev reference is dropped. Every caller holds a reference,
alua_rtpg_queue() takes its own scsi_device_get() before reading
handler_data, and queued RTPG work holds one through pg->rtpg_sdev.
The rcu_dereference(h->pg) == pg part is still needed, as a rescan can
move h to a different port group.
> Note: I was checking which members of alua_dh_data are not
initialized
> prior to publishing sdev->handler_data, and alua_dh_data.group_id is
> never accessed at all AFAICS.
Agreed, group_id is never read or written. All other members are
initialized before handler_data is set (h is kzalloc'd, and disabled
is set under init_mutex in alua_initialize()).
I would prefer to keep this patch minimal since it is tagged for
stable, and send a follow-up cleanup that drops the now-redundant
h check in alua_rtpg_queue() and removes the unused group_id member.
Does that work for you, or would you rather I fold both into a v2?
Thanks,
Laurence