Re: [PATCH 01/10] hbitmap: introduce HBITMAP_MAX_ORIG_SIZE

[Eric Blake <eblake@redhat.com>]

On 9/30/19 10:14 AM, Vladimir Sementsov-Ogievskiy wrote:
> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>

A bit light on the commit message for explaining why.

> ---
>   include/qemu/hbitmap.h | 7 +++++++
>   util/hbitmap.c         | 2 ++
>   2 files changed, 9 insertions(+)
> 
> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h
> index 1bf944ca3d..82317c5364 100644
> --- a/include/qemu/hbitmap.h
> +++ b/include/qemu/hbitmap.h
> @@ -33,6 +33,13 @@ typedef struct HBitmapIter HBitmapIter;
>    */
>   #define HBITMAP_LEVELS         ((HBITMAP_LOG_MAX_SIZE / BITS_PER_LEVEL) + 1)
>   
> +/*
> + * We have APIs which returns signed int64_t, to be able to return error.
> + * Therefore we can't handle bitmaps with absolute size larger than
> + * (INT64_MAX+1). Still, keep it INT64_MAX to be a bit safer.
> + */
> +#define HBITMAP_MAX_ORIG_SIZE INT64_MAX

That, and bitmaps represent disk images, but disk images can't exceed 
INT64_MAX bytes (thanks to off_t being signed).  But does introducing a 
new constant really help?

> +
>   struct HBitmapIter {
>       const HBitmap *hb;
>   
> diff --git a/util/hbitmap.c b/util/hbitmap.c
> index 757d39e360..df192234e3 100644
> --- a/util/hbitmap.c
> +++ b/util/hbitmap.c
> @@ -708,6 +708,7 @@ HBitmap *hbitmap_alloc(uint64_t size, int granularity)
>       HBitmap *hb = g_new0(struct HBitmap, 1);
>       unsigned i;
>   
> +    assert(size <= HBITMAP_MAX_ORIG_SIZE);

or can we just inline INT64_MAX here?

>       hb->orig_size = size;
>   
>       assert(granularity >= 0 && granularity < 64);
> @@ -738,6 +739,7 @@ void hbitmap_truncate(HBitmap *hb, uint64_t size)
>       uint64_t num_elements = size;
>       uint64_t old;
>   
> +    assert(size <= HBITMAP_MAX_ORIG_SIZE);
>       hb->orig_size = size;
>   
>       /* Size comes in as logical elements, adjust for granularity. */
> 

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org