mem: fix the alloc size roundup overflow

  The size checking is done in the caller. The size parameter is an
unsigned (64b wide) right now, so the comparison with zero should be
enough in most cases. But it won't help in the following case.
If the allocating request input a huge number by mistake, e.g., some
overflow after the calculation (especially subtraction), the checking
in the caller will succeed since it is not zero. Indeed, there is not
enough space in the system to support such huge memory allocation.
Usually it will return failure in the following code. But if the
input size is just a little smaller than the UINT64_MAX, like -2 in
signed type.
The roundup will cause an overflow and then "reset" the size to 0,
and then only a header (128B now) with zero length will be returned.
The following will be the previous allocation header.
It should be OK in most cases if the application won't access the
memory body. Or else, some critical issue will be caused and not easy
to debug. So this issue should be prevented at the beginning, like
other big size failure, NULL pointer should be returned also.

Fixes: fdf20fa7bee9 ("add prefix to cache line macros")
Cc: sergio.gonzalez.monroy@intel.com
Cc: stable@dpdk.org

Signed-off-by: Bing Zhao <bingz@mellanox.com>
---
 lib/librte_eal/common/malloc_heap.c | 3 +++
 1 file changed, 3 insertions(+)
  

On 07-Apr-20 11:46 AM, Bing Zhao wrote:
> The size checking is done in the caller. The size parameter is an
> unsigned (64b wide) right now, so the comparison with zero should be
> enough in most cases. But it won't help in the following case.
> If the allocating request input a huge number by mistake, e.g., some
> overflow after the calculation (especially subtraction), the checking
> in the caller will succeed since it is not zero. Indeed, there is not
> enough space in the system to support such huge memory allocation.
> Usually it will return failure in the following code. But if the
> input size is just a little smaller than the UINT64_MAX, like -2 in
> signed type.
> The roundup will cause an overflow and then "reset" the size to 0,
> and then only a header (128B now) with zero length will be returned.
> The following will be the previous allocation header.
> It should be OK in most cases if the application won't access the
> memory body. Or else, some critical issue will be caused and not easy
> to debug. So this issue should be prevented at the beginning, like
> other big size failure, NULL pointer should be returned also.
> 
> Fixes: fdf20fa7bee9 ("add prefix to cache line macros")
> Cc: sergio.gonzalez.monroy@intel.com
> Cc: stable@dpdk.org
> 
> Signed-off-by: Bing Zhao <bingz@mellanox.com>
> ---
>   lib/librte_eal/common/malloc_heap.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/lib/librte_eal/common/malloc_heap.c b/lib/librte_eal/common/malloc_heap.c
> index 842eb9d..bd50656 100644
> --- a/lib/librte_eal/common/malloc_heap.c
> +++ b/lib/librte_eal/common/malloc_heap.c
> @@ -241,6 +241,9 @@
>   	size = RTE_CACHE_LINE_ROUNDUP(size);
>   	align = RTE_CACHE_LINE_ROUNDUP(align);
>   
> +	/* roundup might cause an overflow */
> +	if (size == 0)
> +		return NULL;
>   	elem = find_suitable_element(heap, size, flags, align, bound, contig);
>   	if (elem != NULL) {
>   		elem = malloc_elem_alloc(elem, size, align, bound, contig);
> 

Can we add a unit test for this in malloc_autotest?

Otherwise,

Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
  

> -----Original Message-----
> From: Burakov, Anatoly <anatoly.burakov@intel.com>
> Sent: Monday, April 20, 2020 7:40 PM
> To: Bing Zhao <bingz@mellanox.com>
> Cc: dev@dpdk.org; sergio.gonzalez.monroy@intel.com;
> stable@dpdk.org
> Subject: Re: [PATCH] mem: fix the alloc size roundup overflow
> 
> On 07-Apr-20 11:46 AM, Bing Zhao wrote:
> > The size checking is done in the caller. The size parameter is an
> > unsigned (64b wide) right now, so the comparison with zero should
> be
> > enough in most cases. But it won't help in the following case.
> > If the allocating request input a huge number by mistake, e.g., some
> > overflow after the calculation (especially subtraction), the checking
> > in the caller will succeed since it is not zero. Indeed, there is not
> > enough space in the system to support such huge memory allocation.
> > Usually it will return failure in the following code. But if the input
> > size is just a little smaller than the UINT64_MAX, like -2 in signed
> > type.
> > The roundup will cause an overflow and then "reset" the size to 0,
> and
> > then only a header (128B now) with zero length will be returned.
> > The following will be the previous allocation header.
> > It should be OK in most cases if the application won't access the
> > memory body. Or else, some critical issue will be caused and not easy
> > to debug. So this issue should be prevented at the beginning, like
> > other big size failure, NULL pointer should be returned also.
> >
> > Fixes: fdf20fa7bee9 ("add prefix to cache line macros")
> > Cc: sergio.gonzalez.monroy@intel.com
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Bing Zhao <bingz@mellanox.com>
> > ---
> >   lib/librte_eal/common/malloc_heap.c | 3 +++
> >   1 file changed, 3 insertions(+)
> >
> > diff --git a/lib/librte_eal/common/malloc_heap.c
> > b/lib/librte_eal/common/malloc_heap.c
> > index 842eb9d..bd50656 100644
> > --- a/lib/librte_eal/common/malloc_heap.c
> > +++ b/lib/librte_eal/common/malloc_heap.c
> > @@ -241,6 +241,9 @@
> >   	size = RTE_CACHE_LINE_ROUNDUP(size);
> >   	align = RTE_CACHE_LINE_ROUNDUP(align);
> >
> > +	/* roundup might cause an overflow */
> > +	if (size == 0)
> > +		return NULL;
> >   	elem = find_suitable_element(heap, size, flags, align, bound,
> contig);
> >   	if (elem != NULL) {
> >   		elem = malloc_elem_alloc(elem, size, align, bound,
> contig);
> >
> 
> Can we add a unit test for this in malloc_autotest?
> 
> Otherwise,
> 
> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
> 

Thanks, Burakov. I can try to add one case in the unit for this.

> --
> Thanks,
> Anatoly
  

21/04/2020 05:18, Bing Zhao:
> From: Burakov, Anatoly <anatoly.burakov@intel.com>
> > Can we add a unit test for this in malloc_autotest?
> > 
> > Otherwise,
> > 
> > Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
> 
> Thanks, Burakov. I can try to add one case in the unit for this.

Any update about writing an unit test?