of: overlay: set 'overlay_tree' and 'fdt' fields only on success

From: Slawomir Stepien <slawomir.stepien@nokia.com>

Before this change, the memory pointed by fields 'overlay_tree' and
'fdt' will be double freed by a call to free_overlay_changeset() from
of_overlay_apply(), when the init_overlay_changeset() fails.

The first free will happen under 'err_free_tree' label and for the
second time under 'err_free_overlay_changeset' label, where we call
free_overlay_changeset().

This could happen for example, when you are applying an overlay to a
target path that does not exists.

By setting the pointers only when we are sure that
init_overlay_changeset() will not fail, will prevent this double free.

Signed-off-by: Slawomir Stepien <slawomir.stepien@nokia.com>
---
 drivers/of/overlay.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Hi Slawomir,

Thank you for the patch!

On 31/03/2022 11:56, Slawomir Stepien wrote:
> From: Slawomir Stepien <slawomir.stepien@nokia.com>
> 
> Before this change, the memory pointed by fields 'overlay_tree' and
> 'fdt' will be double freed by a call to free_overlay_changeset() from
> of_overlay_apply(), when the init_overlay_changeset() fails.
> 
> The first free will happen under 'err_free_tree' label and for the
> second time under 'err_free_overlay_changeset' label, where we call
> free_overlay_changeset().
> 
> This could happen for example, when you are applying an overlay to a
> target path that does not exists.
> 
> By setting the pointers only when we are sure that
> init_overlay_changeset() will not fail, will prevent this double free.
> 
> Signed-off-by: Slawomir Stepien <slawomir.stepien@nokia.com>

Reviewed-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>

> ---
>  drivers/of/overlay.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/of/overlay.c b/drivers/of/overlay.c
> index d80160cf34bb..a72a9a415f8f 100644
> --- a/drivers/of/overlay.c
> +++ b/drivers/of/overlay.c
> @@ -750,9 +750,6 @@ static int init_overlay_changeset(struct overlay_changeset *ovcs,
>  	if (!of_node_is_root(tree))
>  		pr_debug("%s() tree is not root\n", __func__);
>  
> -	ovcs->overlay_tree = tree;
> -	ovcs->fdt = fdt;
> -
>  	INIT_LIST_HEAD(&ovcs->ovcs_list);
>  
>  	of_changeset_init(&ovcs->cset);
> @@ -829,6 +826,8 @@ static int init_overlay_changeset(struct overlay_changeset *ovcs,
>  		goto err_free_fragments;
>  	}
>  
> +	ovcs->overlay_tree = tree;
> +	ovcs->fdt = fdt;
>  	ovcs->id = id;
>  	ovcs->count = cnt;
>  	ovcs->fragments = fragments;

On Thu, Mar 31, 2022 at 11:56:33AM +0200, Slawomir Stepien wrote:
> From: Slawomir Stepien <slawomir.stepien@nokia.com>
> 
> Before this change, the memory pointed by fields 'overlay_tree' and
> 'fdt' will be double freed by a call to free_overlay_changeset() from
> of_overlay_apply(), when the init_overlay_changeset() fails.
> 
> The first free will happen under 'err_free_tree' label and for the
> second time under 'err_free_overlay_changeset' label, where we call
> free_overlay_changeset().
> 
> This could happen for example, when you are applying an overlay to a
> target path that does not exists.
> 
> By setting the pointers only when we are sure that
> init_overlay_changeset() will not fail, will prevent this double free.

It looks to me like the free should just be moved from 
init_overlay_changeset() to of_overlay_fdt_apply() where the allocation 
is done. Frank?

Also, I believe there's a bug that of_overlay_apply() should be passed 
new_fdt_align rather than new_fdt. It's only a bug if we expect 
overlay_changeset.fdt to point to a valid fdt rather than the memory 
allocation.

Rob

On 4/4/22 15:38, Rob Herring wrote:
> On Thu, Mar 31, 2022 at 11:56:33AM +0200, Slawomir Stepien wrote:
>> From: Slawomir Stepien <slawomir.stepien@nokia.com>
>>
>> Before this change, the memory pointed by fields 'overlay_tree' and
>> 'fdt' will be double freed by a call to free_overlay_changeset() from
>> of_overlay_apply(), when the init_overlay_changeset() fails.
>>
>> The first free will happen under 'err_free_tree' label and for the
>> second time under 'err_free_overlay_changeset' label, where we call
>> free_overlay_changeset().
>>
>> This could happen for example, when you are applying an overlay to a
>> target path that does not exists.
>>
>> By setting the pointers only when we are sure that
>> init_overlay_changeset() will not fail, will prevent this double free.
> 
> It looks to me like the free should just be moved from 
> init_overlay_changeset() to of_overlay_fdt_apply() where the allocation 
> is done. Frank?

This patch is next on my list to look over.

-Frank

> 
> Also, I believe there's a bug that of_overlay_apply() should be passed 
> new_fdt_align rather than new_fdt. It's only a bug if we expect 
> overlay_changeset.fdt to point to a valid fdt rather than the memory 
> allocation.
> 
> Rob

Hi Slawomir,

 On 4/4/22 16:02, Frank Rowand wrote:
> On 4/4/22 15:38, Rob Herring wrote:
>> On Thu, Mar 31, 2022 at 11:56:33AM +0200, Slawomir Stepien wrote:
>>> From: Slawomir Stepien <slawomir.stepien@nokia.com>
>>>
>>> Before this change, the memory pointed by fields 'overlay_tree' and
>>> 'fdt' will be double freed by a call to free_overlay_changeset() from
>>> of_overlay_apply(), when the init_overlay_changeset() fails.
>>>
>>> The first free will happen under 'err_free_tree' label and for the
>>> second time under 'err_free_overlay_changeset' label, where we call
>>> free_overlay_changeset().
>>>
>>> This could happen for example, when you are applying an overlay to a
>>> target path that does not exists.
>>>
>>> By setting the pointers only when we are sure that
>>> init_overlay_changeset() will not fail, will prevent this double free.
>>
>> It looks to me like the free should just be moved from 
>> init_overlay_changeset() to of_overlay_fdt_apply() where the allocation 
>> is done. Frank?
> 
> This patch is next on my list to look over.

Thanks for finding this problem.  While investigating what you reported
I found that there are additional related issues.  I am in the process
of testing a patch to fix all of the issues.

-Frank

> 
> -Frank
> 
>>
>> Also, I believe there's a bug that of_overlay_apply() should be passed 
>> new_fdt_align rather than new_fdt. It's only a bug if we expect 
>> overlay_changeset.fdt to point to a valid fdt rather than the memory 
>> allocation.
>>
>> Rob
>

On kwi 07, 2022 14:53, Frank Rowand wrote:
> Hi Slawomir,

Hello

>  On 4/4/22 16:02, Frank Rowand wrote:
> > On 4/4/22 15:38, Rob Herring wrote:
> >> On Thu, Mar 31, 2022 at 11:56:33AM +0200, Slawomir Stepien wrote:
> >>> From: Slawomir Stepien <slawomir.stepien@nokia.com>
> >>>
> >>> Before this change, the memory pointed by fields 'overlay_tree' and
> >>> 'fdt' will be double freed by a call to free_overlay_changeset() from
> >>> of_overlay_apply(), when the init_overlay_changeset() fails.
> >>>
> >>> The first free will happen under 'err_free_tree' label and for the
> >>> second time under 'err_free_overlay_changeset' label, where we call
> >>> free_overlay_changeset().
> >>>
> >>> This could happen for example, when you are applying an overlay to a
> >>> target path that does not exists.
> >>>
> >>> By setting the pointers only when we are sure that
> >>> init_overlay_changeset() will not fail, will prevent this double free.
> >>
> >> It looks to me like the free should just be moved from 
> >> init_overlay_changeset() to of_overlay_fdt_apply() where the allocation 
> >> is done. Frank?
> > 
> > This patch is next on my list to look over.
> 
> Thanks for finding this problem.  While investigating what you reported
> I found that there are additional related issues.  I am in the process
> of testing a patch to fix all of the issues.

That sounds great! Thank you!

> -Frank
> 
> > 
> > -Frank
> > 
> >>
> >> Also, I believe there's a bug that of_overlay_apply() should be passed 
> >> new_fdt_align rather than new_fdt. It's only a bug if we expect 
> >> overlay_changeset.fdt to point to a valid fdt rather than the memory 
> >> allocation.
> >>
> >> Rob
> > 
>

On 4/4/22 15:38, Rob Herring wrote:
> On Thu, Mar 31, 2022 at 11:56:33AM +0200, Slawomir Stepien wrote:
>> From: Slawomir Stepien <slawomir.stepien@nokia.com>
>>
>> Before this change, the memory pointed by fields 'overlay_tree' and
>> 'fdt' will be double freed by a call to free_overlay_changeset() from
>> of_overlay_apply(), when the init_overlay_changeset() fails.
>>
>> The first free will happen under 'err_free_tree' label and for the
>> second time under 'err_free_overlay_changeset' label, where we call
>> free_overlay_changeset().
>>
>> This could happen for example, when you are applying an overlay to a
>> target path that does not exists.
>>
>> By setting the pointers only when we are sure that
>> init_overlay_changeset() will not fail, will prevent this double free.
> 
> It looks to me like the free should just be moved from 
> init_overlay_changeset() to of_overlay_fdt_apply() where the allocation 
> is done. Frank?

Yes, that would be much cleaner and proper.  v2 of my follow on patch set
will do so.

> 
> Also, I believe there's a bug that of_overlay_apply() should be passed 
> new_fdt_align rather than new_fdt. It's only a bug if we expect 
> overlay_changeset.fdt to point to a valid fdt rather than the memory 
> allocation.

new_fdt is correct instead of new_fdt_align.  It is only used by
of_overlay_apply() and children for the purpose of kfree().  The
only place that new_fdt_align is derefenced to access the data
in the FDT is in of_fdt_unflatten_tree() and children, which is
called by of_overlay_fdt_apply.
> 
> Rob