Re: [PATCH v11 1/4] kasan: support backing vmalloc space with real shadow memory

[Daniel Axtens <dja@axtens.net>]

Qian Cai <cai@lca.pw> writes:

> On Thu, 2019-10-31 at 20:39 +1100, Daniel Axtens wrote:
>>  	/*
>>  	 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
>>  	 * flag. It means that vm_struct is not fully initialized.
>> @@ -3377,6 +3411,9 @@ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
>>  
>>  		setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC,
>>  				 pcpu_get_vm_areas);
>> +
>> +		/* assume success here */
>> +		kasan_populate_vmalloc(sizes[area], vms[area]);
>>  	}
>>  	spin_unlock(&vmap_area_lock);
>
> Here it is all wrong. GFP_KERNEL with in_atomic().

I think this fix will work, I will do a v12 with it included.

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index a4b950a02d0b..bf030516258c 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -3417,11 +3417,14 @@ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
 
                setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC,
                                 pcpu_get_vm_areas);
+       }
+       spin_unlock(&vmap_area_lock);
 
+       /* populate the shadow space outside of the lock */
+       for (area = 0; area < nr_vms; area++) {
                /* assume success here */
                kasan_populate_vmalloc(sizes[area], vms[area]);
        }
-       spin_unlock(&vmap_area_lock);
 
        kfree(vas);
        return vms;