This has us hold the tv_tpg_mutex when decrementing the
tv_tpg_vhost_count in vhost_scsi_set_endpoint's failure path. We
currently don't need to because we are holding the vhost_scsi_mutex and
the dev.mutex when incrementing/decrementing in the normal paths, and we
can't hit the tv_tpg_port_count check in vhost_scsi_drop_nexus during
this time because there is a nexus and we have a refcount to the tpg.
In the next patch we will change when we hold the vhost_scsi_mutex, so it's
not always held in set/clear endpoint and in the future we might change
the dev.mutex use, so this future proofs us.
Signed-off-by: Mike Christie <michael.christie@xxxxxxxxxx>
---
drivers/vhost/scsi.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c
index d5ecb8876fc9..c31659aa5466 100644
--- a/drivers/vhost/scsi.c
+++ b/drivers/vhost/scsi.c
@@ -1658,7 +1658,9 @@ vhost_scsi_set_endpoint(struct vhost_scsi *vs,
for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
tpg = vs_tpg[i];
if (tpg) {
+ mutex_lock(&tpg->tv_tpg_mutex);
tpg->tv_tpg_vhost_count--;
+ mutex_unlock(&tpg->tv_tpg_mutex);
target_undepend_item(&tpg->se_tpg.tpg_group.cg_item);
}
}
--
2.25.1
