RDMA/odp: Ensure the mm is still alive before creating an implicit child

Registration of a mmu_notifier requires the caller to hold a mmget() on
the mm as registration is not permitted to race with exit_mmap(). There is
a BUG_ON inside the mmu_notifier to guard against this.

Normally creating a umem is done against current which implicitly holds
the mmget(), however an implicit ODP child is created from a pagefault
work queue and is not guaranteed to have a mmget().

Call mmget() around this registration and abort faulting if the MM has
gone to exit_mmap().

Before the patch below the notifier was registered when the implicit ODP
parent was created, so there was no chance to register a notifier outside
of current.

Fixes: c571fec ("RDMA/odp: use mmu_notifier_get/put for 'struct ib_ucontext_per_mm'")
Link: https://lore.kernel.org/r/20200227114118.94736-1-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

drivers/infiniband/core/umem_odp.c

@@ -181,14 +181,28 @@ ib_umem_odp_alloc_child(struct ib_umem_odp *root, unsigned long addr,
 	odp_data->page_shift = PAGE_SHIFT;
 	odp_data->notifier.ops = ops;
 
+	/*
+	 * A mmget must be held when registering a notifier, the owming_mm only
+	 * has a mm_grab at this point.
+	 */
+	if (!mmget_not_zero(umem->owning_mm)) {
+		ret = -EFAULT;
+		goto out_free;
+	}
+
 	odp_data->tgid = get_pid(root->tgid);
 	ret = ib_init_umem_odp(odp_data, ops);
-	if (ret) {
-		put_pid(odp_data->tgid);
-		kfree(odp_data);
-		return ERR_PTR(ret);
-	}
+	if (ret)
+		goto out_tgid;
+	mmput(umem->owning_mm);
 	return odp_data;
+
+out_tgid:
+	put_pid(odp_data->tgid);
+	mmput(umem->owning_mm);
+out_free:
+	kfree(odp_data);
+	return ERR_PTR(ret);
 }
 EXPORT_SYMBOL(ib_umem_odp_alloc_child);