On 2021/12/27 1:33, Christophe Leroy wrote:

Le 25/12/2021 à 13:06, Kefeng Wang a écrit :

quoted

virt_addr_valid() could be insufficient to validate the virt addr
on some architecture, which could lead to potential BUG which has
been found on arm64/powerpc64.

Let's add WARN_ON to check if the virt addr is passed virt_addr_valid()
but is a vmalloc/module address.

I think that's the responsibility of the architecture and doesn't
deserve a WARN_ON() in generic code.

The generic code cannot check all what architectures do wrong.

Eventually you can do some testing at startup, maybe with
CONFIG_DEBUG_VM_PGTABLE, but I don't think we should have such
verification in functions like check_heap_object()

Ok, we could find other better place to add a virt_to_valid() check

instead of this one.

quoted

Signed-off-by: Kefeng Wang <redacted>
---
   mm/usercopy.c | 2 ++
   1 file changed, 2 insertions(+)

diff --git a/mm/usercopy.c b/mm/usercopy.c
index b3de3c4eefba..ce83e0b137dd 100644
--- a/mm/usercopy.c
+++ b/mm/usercopy.c

@@ -228,6 +228,8 @@ static inline void check_heap_object(const void *ptr, unsigned long n,
   	if (!virt_addr_valid(ptr))
   		return;
   
+	WARN_ON_ONCE(is_vmalloc_or_module_addr(ptr));
+
   	/*
   	 * When CONFIG_HIGHMEM=y, kmap_to_page() will give either the
   	 * highmem page or fallback to virt_to_page(). The following