[PATCH] x86/ioremap: tighten integer overflow checking

[PATCH] x86/ioremap: tighten integer overflow checking

[Dan Carpenter]

The current check is a bit off in the case where "phys_addr + size"
wraps to zero because then "last_addr" is set to ULONG_MAX which is >=
phys_addr.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
---
 arch/x86/mm/ioremap.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
index 5378d10f1d31..ee43df3ebe66 100644
--- a/arch/x86/mm/ioremap.c
+++ b/arch/x86/mm/ioremap.c
@@ -146,9 +146,9 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr,
 	void __iomem *ret_addr;
 
 	/* Don't allow wraparound or zero size */
-	last_addr = phys_addr + size - 1;
-	if (!size || last_addr < phys_addr)
+	if (!size || phys_addr + size < phys_addr)
 		return NULL;
+	last_addr = phys_addr + size - 1;
 
 	if (!phys_addr_valid(phys_addr)) {
 		printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
-- 
2.11.0

Re: [PATCH] x86/ioremap: tighten integer overflow checking

[Juergen Gross]

On 25/10/2018 08:16, Dan Carpenter wrote:
> The current check is a bit off in the case where "phys_addr + size"
> wraps to zero because then "last_addr" is set to ULONG_MAX which is >=
> phys_addr.

And -2 would be okay?

For 32-bit systems I believe ULONG_MAX is a perfectly valid physical
address.

> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> ---
>  arch/x86/mm/ioremap.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
> index 5378d10f1d31..ee43df3ebe66 100644
> --- a/arch/x86/mm/ioremap.c
> +++ b/arch/x86/mm/ioremap.c
> @@ -146,9 +146,9 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr,
>  	void __iomem *ret_addr;
>  
>  	/* Don't allow wraparound or zero size */
> -	last_addr = phys_addr + size - 1;
> -	if (!size || last_addr < phys_addr)
> +	if (!size || phys_addr + size < phys_addr)
>  		return NULL;
> +	last_addr = phys_addr + size - 1;
>  
>  	if (!phys_addr_valid(phys_addr)) {

Wouldn't it make more sense to test last_addr for being a valid physical
address here?

>  		printk(KERN_WARNING "ioremap: invalid physical address %llx\n",
> 


Juergen

Re: [PATCH] x86/ioremap: tighten integer overflow checking

[Dan Carpenter]

On Thu, Oct 25, 2018 at 08:28:44AM +0200, Juergen Gross wrote:
> On 25/10/2018 08:16, Dan Carpenter wrote:
> > The current check is a bit off in the case where "phys_addr + size"
> > wraps to zero because then "last_addr" is set to ULONG_MAX which is >=
> > phys_addr.
> 
> And -2 would be okay?
> 

If you see any patches from me it's static checker stuff...

I guess the original is actually fine because the -1 re-wraps back to
what we wanted it to be.  We also do verify that size is correct in
__ioremap_check_mem() so I can't see a problem here after all.

regards,
dan carpenter

Re: [PATCH] x86/ioremap: tighten integer overflow checking

[Thomas Gleixner]

Dan,

On Thu, 25 Oct 2018, Dan Carpenter wrote:

> The current check is a bit off in the case where "phys_addr + size"
> wraps to zero because then "last_addr" is set to ULONG_MAX which is >=
> phys_addr.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> ---
>  arch/x86/mm/ioremap.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
> index 5378d10f1d31..ee43df3ebe66 100644
> --- a/arch/x86/mm/ioremap.c
> +++ b/arch/x86/mm/ioremap.c
> @@ -146,9 +146,9 @@ static void __iomem *__ioremap_caller(resource_size_t phys_addr,
>  	void __iomem *ret_addr;
>  
>  	/* Don't allow wraparound or zero size */
> -	last_addr = phys_addr + size - 1;
> -	if (!size || last_addr < phys_addr)
> +	if (!size || phys_addr + size < phys_addr)

Assume the following (resource_size_t == u32, which is the case when
CONFIG_PHYS_ADDR_T_64BIT=n):

       phys_addr = 0xFFFF0000
       size	 = 0x00010000

       sum	 = 0x00000000  which is < phys_addr

But the existing code does:

       last_addr = phys_addr + size - 1

       		 = 0xFFFFFFFF

which is correct. last_addr is the last valid address in the to be remapped
range starting @phys_addr.

Thanks,

	tglx