[PATCH] bus: fsl-mc: fix improper free of mc_dev

[PATCH] bus: fsl-mc: fix improper free of mc_dev

[trix]

From: Tom Rix <trix@redhat.com>

Clang static analysis reports this error

fsl-mc-bus.c:891:2: warning: Attempt to free released memory
        kfree(mc_dev);
        ^~~~~~~~~~~~~

In this block of code

if (strcmp(obj_desc->type, "dprc") == 0) {
..
  mc_bus = kzalloc(..)
  mc_dev = &mc_bus->mc_dev;

mc_dev is not alloc-ed, so it should not be freed.
Old handler triggers a false positive from checkpatch, so add a
comment and change logic a bit.

Fixes: a042fbed0290 ("staging: fsl-mc: simplify couple of deallocations")
Signed-off-by: Tom Rix <trix@redhat.com>
---
 drivers/bus/fsl-mc/fsl-mc-bus.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
index 380ad1fdb745..fb3e1d8a7f63 100644
--- a/drivers/bus/fsl-mc/fsl-mc-bus.c
+++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
@@ -887,8 +887,10 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
 
 error_cleanup_dev:
 	kfree(mc_dev->regions);
+	/* mc_dev is only allocated when it is not part of mc_bus */
+	if (!mc_bus)
+		kfree(mc_dev);
 	kfree(mc_bus);
-	kfree(mc_dev);
 
 	return error;
 }
-- 
2.26.3

Re: [PATCH] bus: fsl-mc: fix improper free of mc_dev

[Nick Desaulniers]

On Tue, Apr 27, 2021 at 11:36 AM <trix@redhat.com> wrote:
>
> From: Tom Rix <trix@redhat.com>
>
> Clang static analysis reports this error
>
> fsl-mc-bus.c:891:2: warning: Attempt to free released memory
>         kfree(mc_dev);
>         ^~~~~~~~~~~~~
>
> In this block of code
>
> if (strcmp(obj_desc->type, "dprc") == 0) {
> ..
>   mc_bus = kzalloc(..)
>   mc_dev = &mc_bus->mc_dev;

Thanks for the patch.

Aren't the allocations for mc_bus and mc_dev mutually exclusive based
on that conditional? If so...

>
> mc_dev is not alloc-ed, so it should not be freed.
> Old handler triggers a false positive from checkpatch, so add a
> comment and change logic a bit.
>
> Fixes: a042fbed0290 ("staging: fsl-mc: simplify couple of deallocations")
> Signed-off-by: Tom Rix <trix@redhat.com>
> ---
>  drivers/bus/fsl-mc/fsl-mc-bus.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
> index 380ad1fdb745..fb3e1d8a7f63 100644
> --- a/drivers/bus/fsl-mc/fsl-mc-bus.c
> +++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
> @@ -887,8 +887,10 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
>
>  error_cleanup_dev:
>         kfree(mc_dev->regions);
> +       /* mc_dev is only allocated when it is not part of mc_bus */
> +       if (!mc_bus)
> +               kfree(mc_dev);
>         kfree(mc_bus);
> -       kfree(mc_dev);

The error handling here seems quite wrong (regardless of your patch).
mc_dev->regions is allocated by fsl_mc_device_get_mmio_regions() IIUC.
Wouldn't the first `goto error_cleanup_dev;` taken end up passing an
uninitialized pointer to kfree()?

what if `strcmp(obj_desc->type, "dprc") == 0` is false? We allocate
`mc_dev`, but then call kfree on `mc_bus`?

I think it would be safer to locally save the result of
`strcmp(obj_desc->type, "dprc") == 0`, then check that throughout this
function, including the error handling at the end, or use multiple
labels to unwind the allocations correctly.

>
>         return error;
>  }

> --


--
Thanks,
~Nick Desaulniers

Re: [PATCH] bus: fsl-mc: fix improper free of mc_dev

[Tom Rix]

On 4/27/21 12:19 PM, Nick Desaulniers wrote:
> On Tue, Apr 27, 2021 at 11:36 AM <trix@redhat.com> wrote:
>> From: Tom Rix <trix@redhat.com>
>>
>> Clang static analysis reports this error
>>
>> fsl-mc-bus.c:891:2: warning: Attempt to free released memory
>>          kfree(mc_dev);
>>          ^~~~~~~~~~~~~
>>
>> In this block of code
>>
>> if (strcmp(obj_desc->type, "dprc") == 0) {
>> ..
>>    mc_bus = kzalloc(..)
>>    mc_dev = &mc_bus->mc_dev;
> Thanks for the patch.
>
> Aren't the allocations for mc_bus and mc_dev mutually exclusive based
> on that conditional? If so...
>
>> mc_dev is not alloc-ed, so it should not be freed.
>> Old handler triggers a false positive from checkpatch, so add a
>> comment and change logic a bit.
>>
>> Fixes: a042fbed0290 ("staging: fsl-mc: simplify couple of deallocations")
>> Signed-off-by: Tom Rix <trix@redhat.com>
>> ---
>>   drivers/bus/fsl-mc/fsl-mc-bus.c | 4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/bus/fsl-mc/fsl-mc-bus.c b/drivers/bus/fsl-mc/fsl-mc-bus.c
>> index 380ad1fdb745..fb3e1d8a7f63 100644
>> --- a/drivers/bus/fsl-mc/fsl-mc-bus.c
>> +++ b/drivers/bus/fsl-mc/fsl-mc-bus.c
>> @@ -887,8 +887,10 @@ int fsl_mc_device_add(struct fsl_mc_obj_desc *obj_desc,
>>
>>   error_cleanup_dev:
>>          kfree(mc_dev->regions);
>> +       /* mc_dev is only allocated when it is not part of mc_bus */
>> +       if (!mc_bus)
>> +               kfree(mc_dev);
>>          kfree(mc_bus);
>> -       kfree(mc_dev);
> The error handling here seems quite wrong (regardless of your patch).
> mc_dev->regions is allocated by fsl_mc_device_get_mmio_regions() IIUC.
> Wouldn't the first `goto error_cleanup_dev;` taken end up passing an
> uninitialized pointer to kfree()?

On the first goto, mc_dev->regions, because of the kzalloc, the value 
would be

mc_bus->mc_dev.regions , should be 0 or

mc_dev->regions, which also should be 0

and kfree handles 0.

>
> what if `strcmp(obj_desc->type, "dprc") == 0` is false? We allocate
> `mc_dev`, but then call kfree on `mc_bus`?

mc_bus is initialized to NULL, which makes the call to kfree safe.

The original handler was

if (mc_bus)

   kfree(mc_bus)

else

   kfree(mc_dev)

I tried this first, which works, but checkpatch throw a warning for 
kfree(mc_bus).

This change makes the 'else' with the !mc_bus

> I think it would be safer to locally save the result of
> `strcmp(obj_desc->type, "dprc") == 0`, then check that throughout this
the local mc_bus is only set in this block, so I don't think another 
local is needed.
> function, including the error handling at the end, or use multiple
> labels to unwind the allocations correctly.

The goto's could be finer grained because some of the mc_dev->regions 
are known to be unallocated.

Changing these would not be a fix and it could be argued the simpler, 
less efficent error handling works as designed.

Tom

>
>>          return error;
>>   }
>> --
>
> --
> Thanks,
> ~Nick Desaulniers
>