linux-kernel - Re: [PATCH] drm/nouveau: Use kmemdup

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <a61d9781-7c6e-46b8-ab1b-cf4fc1c76ba3@redhat.com>
Date: Mon, 17 Jun 2024 17:55:33 +0200
From: Danilo Krummrich <dakr@...hat.com>
To: Dan Carpenter <dan.carpenter@...aro.org>
Cc: Lyude Paul <lyude@...hat.com>, David Airlie <airlied@...il.com>,
 Daniel Vetter <daniel@...ll.ch>, dri-devel@...ts.freedesktop.org,
 nouveau@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
 kernel-janitors@...r.kernel.org, Karol Herbst <kherbst@...hat.com>
Subject: Re: [PATCH] drm/nouveau: Use kmemdup_array() instead of kmemdup()

On 6/17/24 11:33, Dan Carpenter wrote:
> Use kmemdup_array() because we're allocating an array.
> 
> The main difference between kmemdup() and kmemdup_array() is that the
> kmemdup_array() function has integer overflow checking built it.  The
> "args->in_sync.count" variable is a u32 so integer overflows would only
> be a concern on 32bit systems.  Fortunately, however, the u_memcpya()
> function has integer overflow checking which means that it is not an
> issue.
> 
> Still using kmemdup_array() is more appropriate and makes auditing the
> code easier.

Indeed, we shouldn't rely on the previous check here, good catch.

> 
> Signed-off-by: Dan Carpenter <dan.carpenter@...aro.org>
> ---
>   drivers/gpu/drm/nouveau/nouveau_sched.c | 16 ++++++++--------
>   1 file changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/nouveau_sched.c b/drivers/gpu/drm/nouveau/nouveau_sched.c
> index 32fa2e273965..53d8b0584a56 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_sched.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_sched.c
> @@ -45,10 +45,10 @@ nouveau_job_init(struct nouveau_job *job,
>   		if (job->sync)
>   			return -EINVAL;
>   
> -		job->in_sync.data = kmemdup(args->in_sync.s,
> -					 sizeof(*args->in_sync.s) *
> -					 args->in_sync.count,
> -					 GFP_KERNEL);
> +		job->in_sync.data = kmemdup_array(args->in_sync.s,
> +					args->in_sync.count,
> +					sizeof(*args->in_sync.s),
> +					GFP_KERNEL);
>   		if (!job->in_sync.data)
>   			return -ENOMEM;

Not sure if this is what we want for kmemdup_array(). It just saturates the
size. This doesn't prevent accessing the array out of bounds later on. I mean,
it's rather unlikely to get such a huge amount of physically contiguous memory
anyways, but wouldn't it be cleaner to let kmemdup_array() return
ERR_PTR(-EOVERFLOW) on overflow, just like memdup_array_user()[1]?

AFAICS, there's just two users of kmemdup_array(), hence it should be an easy
change. :-)

[1] https://elixir.bootlin.com/linux/latest/source/include/linux/string.h#L30

>   	}
> @@ -60,10 +60,10 @@ nouveau_job_init(struct nouveau_job *job,
>   			goto err_free_in_sync;
>   		}
>   
> -		job->out_sync.data = kmemdup(args->out_sync.s,
> -					  sizeof(*args->out_sync.s) *
> -					  args->out_sync.count,
> -					  GFP_KERNEL);
> +		job->out_sync.data = kmemdup_array(args->out_sync.s,
> +					args->out_sync.count,
> +					sizeof(*args->out_sync.s),
> +					GFP_KERNEL);
>   		if (!job->out_sync.data) {
>   			ret = -ENOMEM;
>   			goto err_free_in_sync;