[PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Yang Yingliang]

When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
set to 'i - 1' to prevent double-free when cleanup resources.

BUG: KASAN: double-free or invalid-free in __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
Call Trace:
 kfree+0x117/0x4c0
 __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
 iio_buffers_alloc_sysfs_and_mask+0x60d/0x1570 [industrialio]
 __iio_device_register+0x483/0x1a30 [industrialio]
 ina2xx_probe+0x625/0x980 [ina2xx_adc]

Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: ee708e6baacd ("iio: buffer: introduce support for attaching more IIO buffers")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 drivers/iio/industrialio-buffer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iio/industrialio-buffer.c b/drivers/iio/industrialio-buffer.c
index a95cc2da56be..5f4bd0b73d03 100644
--- a/drivers/iio/industrialio-buffer.c
+++ b/drivers/iio/industrialio-buffer.c
@@ -1616,7 +1616,7 @@ int iio_buffers_alloc_sysfs_and_mask(struct iio_dev *indio_dev)
 		buffer = iio_dev_opaque->attached_buffers[i];
 		ret = __iio_buffer_alloc_sysfs_and_mask(buffer, indio_dev, i);
 		if (ret) {
-			unwind_idx = i;
+			unwind_idx = i - 1;
 			goto error_unwind_sysfs_and_mask;
 		}
 	}
-- 
2.25.1

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Alexandru Ardelean]

On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
<yangyingliang@huawei.com> wrote:
>
> When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
> set to 'i - 1' to prevent double-free when cleanup resources.
>
> BUG: KASAN: double-free or invalid-free in __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
> Call Trace:
>  kfree+0x117/0x4c0
>  __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
>  iio_buffers_alloc_sysfs_and_mask+0x60d/0x1570 [industrialio]
>  __iio_device_register+0x483/0x1a30 [industrialio]
>  ina2xx_probe+0x625/0x980 [ina2xx_adc]
>

Makes sense.
Thanks for the catch.

Reviewed-by: Alexandru Ardelean <ardeleanalex@gmail.com>

> Reported-by: Hulk Robot <hulkci@huawei.com>
> Fixes: ee708e6baacd ("iio: buffer: introduce support for attaching more IIO buffers")
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
> ---
>  drivers/iio/industrialio-buffer.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/iio/industrialio-buffer.c b/drivers/iio/industrialio-buffer.c
> index a95cc2da56be..5f4bd0b73d03 100644
> --- a/drivers/iio/industrialio-buffer.c
> +++ b/drivers/iio/industrialio-buffer.c
> @@ -1616,7 +1616,7 @@ int iio_buffers_alloc_sysfs_and_mask(struct iio_dev *indio_dev)
>                 buffer = iio_dev_opaque->attached_buffers[i];
>                 ret = __iio_buffer_alloc_sysfs_and_mask(buffer, indio_dev, i);
>                 if (ret) {
> -                       unwind_idx = i;
> +                       unwind_idx = i - 1;
>                         goto error_unwind_sysfs_and_mask;
>                 }
>         }
> --
> 2.25.1
>

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Andy Shevchenko]

On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
<ardeleanalex@gmail.com> wrote:
>
> On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> <yangyingliang@huawei.com> wrote:
> >
> > When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
> > set to 'i - 1' to prevent double-free when cleanup resources.
> >
> > BUG: KASAN: double-free or invalid-free in __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
> > Call Trace:
> >  kfree+0x117/0x4c0
> >  __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
> >  iio_buffers_alloc_sysfs_and_mask+0x60d/0x1570 [industrialio]
> >  __iio_device_register+0x483/0x1a30 [industrialio]
> >  ina2xx_probe+0x625/0x980 [ina2xx_adc]
> >
>
> Makes sense.
> Thanks for the catch.
>
> Reviewed-by: Alexandru Ardelean <ardeleanalex@gmail.com>

...

> >                 ret = __iio_buffer_alloc_sysfs_and_mask(buffer, indio_dev, i);
> >                 if (ret) {
> > -                       unwind_idx = i;
> > +                       unwind_idx = i - 1;
> >                         goto error_unwind_sysfs_and_mask;

I prefer to see

- for (; unwind_idx >= 0; unwind_idx--) {
+ while (unwind_idx--)

instead.

> >                 }
> >         }

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Joe Perches]

On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
> On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> <ardeleanalex@gmail.com> wrote:
> > 
> > On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> > <yangyingliang@huawei.com> wrote:
> > > 
> > > When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
> > > set to 'i - 1' to prevent double-free when cleanup resources.
[]
> 
> I prefer to see
> 
> - for (; unwind_idx >= 0; unwind_idx--) {
> + while (unwind_idx--)

Not the same code as unwind_idx would be decremented before entering
the code block.

You'd want

	do {
		...
	} while (unwind_idx--);

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Andy Shevchenko]

On Tue, Oct 12, 2021 at 8:43 PM Joe Perches <joe@perches.com> wrote:
>
> On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
> > On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> > <ardeleanalex@gmail.com> wrote:
> > >
> > > On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> > > <yangyingliang@huawei.com> wrote:
> > > >
> > > > When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
> > > > set to 'i - 1' to prevent double-free when cleanup resources.
> []
> >
> > I prefer to see
> >
> > - for (; unwind_idx >= 0; unwind_idx--) {
> > + while (unwind_idx--)
>
> Not the same code as unwind_idx would be decremented before entering
> the code block.

It's kinda cryptic what you are pointing out.

What's needed additionally is to change

- unwind_idx = iio_dev_opaque->attached_buffers_cnt - 1;
+ unwind_idx = i;

> You'd want
>
>         do {
>                 ...
>         } while (unwind_idx--);

Of course not. See above. The usual pattern is

  while (i--)
    do_clean_item(i);

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Joe Perches]

On Tue, 2021-10-12 at 23:48 +0300, Andy Shevchenko wrote:
> On Tue, Oct 12, 2021 at 8:43 PM Joe Perches <joe@perches.com> wrote:
> > 
> > On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
> > > On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> > > <ardeleanalex@gmail.com> wrote:
> > > > 
> > > > On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> > > > <yangyingliang@huawei.com> wrote:
> > > > > 
> > > > > When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
> > > > > set to 'i - 1' to prevent double-free when cleanup resources.
> > []
> > > 
> > > I prefer to see
> > > 
> > > - for (; unwind_idx >= 0; unwind_idx--) {
> > > + while (unwind_idx--)
> > 
> > Not the same code as unwind_idx would be decremented before entering
> > the code block.
> 
> It's kinda cryptic what you are pointing out.

Not really,

> What's needed additionally is to change
> 
> - unwind_idx = iio_dev_opaque->attached_buffers_cnt - 1;
> + unwind_idx = i;

You left out that 'additional change' above from your reply.

> Of course not. See above. The usual pattern is
> 
>   while (i--)
>     do_clean_item(i);

Of course, but that's not what you replied.
I was merely pointing out that your reply included a logic change
converting a loop from for to while.

cheers, Joe

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Andy Shevchenko]

On Tue, Oct 12, 2021 at 8:55 PM Joe Perches <joe@perches.com> wrote:
> On Tue, 2021-10-12 at 23:48 +0300, Andy Shevchenko wrote:
> > On Tue, Oct 12, 2021 at 8:43 PM Joe Perches <joe@perches.com> wrote:
> > > On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
> > > > On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> > > > <ardeleanalex@gmail.com> wrote:
> > > > > On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> > > > > <yangyingliang@huawei.com> wrote:

...

> > > > I prefer to see
> > > >
> > > > - for (; unwind_idx >= 0; unwind_idx--) {
> > > > + while (unwind_idx--)
> > >
> > > Not the same code as unwind_idx would be decremented before entering
> > > the code block.
> >
> > It's kinda cryptic what you are pointing out.
>
> Not really,

It's. It lacks the very same "additional" words to explain what you
meant and why.

> > What's needed additionally is to change
> >
> > - unwind_idx = iio_dev_opaque->attached_buffers_cnt - 1;
> > + unwind_idx = i;
>
> You left out that 'additional change' above from your reply.

Yes, that's true, but it took some time to decrypt your message.

> > Of course not. See above. The usual pattern is
> >
> >   while (i--)
> >     do_clean_item(i);
>
> Of course, but that's not what you replied.
> I was merely pointing out that your reply included a logic change
> converting a loop from for to while.

I expect that developers actually think about the changes they do and
double check what's proposed by reviewers. If they just copy'n'paste
whatever others propose, I wouldn't take any patch from such a
developer.

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Yang Yingliang]

Hi,

On 2021/10/13 4:58, Andy Shevchenko wrote:
> On Tue, Oct 12, 2021 at 8:55 PM Joe Perches <joe@perches.com> wrote:
>> On Tue, 2021-10-12 at 23:48 +0300, Andy Shevchenko wrote:
>>> On Tue, Oct 12, 2021 at 8:43 PM Joe Perches <joe@perches.com> wrote:
>>>> On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
>>>>> On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
>>>>> <ardeleanalex@gmail.com> wrote:
>>>>>> On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
>>>>>> <yangyingliang@huawei.com> wrote:
> ...
>
>>>>> I prefer to see
>>>>>
>>>>> - for (; unwind_idx >= 0; unwind_idx--) {
>>>>> + while (unwind_idx--)
>>>> Not the same code as unwind_idx would be decremented before entering
>>>> the code block.
>>> It's kinda cryptic what you are pointing out.
>> Not really,
> It's. It lacks the very same "additional" words to explain what you
> meant and why.
>
>>> What's needed additionally is to change
>>>
>>> - unwind_idx = iio_dev_opaque->attached_buffers_cnt - 1;
>>> + unwind_idx = i;
>> You left out that 'additional change' above from your reply.
> Yes, that's true, but it took some time to decrypt your message.
>
>>> Of course not. See above. The usual pattern is
>>>
>>>    while (i--)
>>>      do_clean_item(i);
>> Of course, but that's not what you replied.
>> I was merely pointing out that your reply included a logic change
>> converting a loop from for to while.
> I expect that developers actually think about the changes they do and
> double check what's proposed by reviewers. If they just copy'n'paste
> whatever others propose, I wouldn't take any patch from such a
> developer.
I think in alloc path is using for loop, and in error/free path also 
using for loop is better to read the code.
>

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Andy Shevchenko]

On Wed, Oct 13, 2021 at 12:17 PM Yang Yingliang
<yangyingliang@huawei.com> wrote:
> On 2021/10/13 4:58, Andy Shevchenko wrote:
> > On Tue, Oct 12, 2021 at 8:55 PM Joe Perches <joe@perches.com> wrote:
> >> On Tue, 2021-10-12 at 23:48 +0300, Andy Shevchenko wrote:
> >>> On Tue, Oct 12, 2021 at 8:43 PM Joe Perches <joe@perches.com> wrote:
> >>>> On Tue, 2021-10-12 at 23:30 +0300, Andy Shevchenko wrote:
> >>>>> On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> >>>>> <ardeleanalex@gmail.com> wrote:
> >>>>>> On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
> >>>>>> <yangyingliang@huawei.com> wrote:
> > ...
> >
> >>>>> I prefer to see
> >>>>>
> >>>>> - for (; unwind_idx >= 0; unwind_idx--) {
> >>>>> + while (unwind_idx--)
> >>>> Not the same code as unwind_idx would be decremented before entering
> >>>> the code block.
> >>> It's kinda cryptic what you are pointing out.
> >> Not really,
> > It's. It lacks the very same "additional" words to explain what you
> > meant and why.
> >
> >>> What's needed additionally is to change
> >>>
> >>> - unwind_idx = iio_dev_opaque->attached_buffers_cnt - 1;
> >>> + unwind_idx = i;
> >> You left out that 'additional change' above from your reply.
> > Yes, that's true, but it took some time to decrypt your message.
> >
> >>> Of course not. See above. The usual pattern is
> >>>
> >>>    while (i--)
> >>>      do_clean_item(i);
> >> Of course, but that's not what you replied.
> >> I was merely pointing out that your reply included a logic change
> >> converting a loop from for to while.
> > I expect that developers actually think about the changes they do and
> > double check what's proposed by reviewers. If they just copy'n'paste
> > whatever others propose, I wouldn't take any patch from such a
> > developer.
> I think in alloc path is using for loop, and in error/free path also
> using for loop is better to read the code.

I don't think so.

  while(idx--)

is kinda idiom which is really easy to read.

I could send a v2 on your behalf.

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Andy Shevchenko]

On Wed, Oct 13, 2021 at 12:28 PM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:
> On Wed, Oct 13, 2021 at 12:17 PM Yang Yingliang
> <yangyingliang@huawei.com> wrote:

> I could send a v2 on your behalf.

Looking into code further I think that your patch is good as is, see
v2 I have just sent.

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: buffer: Fix double-free in iio_buffers_alloc_sysfs_and_mask()

[Yang Yingliang]

Hi,

On 2021/10/13 4:30, Andy Shevchenko wrote:
> On Tue, Oct 12, 2021 at 2:37 PM Alexandru Ardelean
> <ardeleanalex@gmail.com> wrote:
>> On Tue, Oct 12, 2021 at 12:18 PM Yang Yingliang
>> <yangyingliang@huawei.com> wrote:
>>> When __iio_buffer_alloc_sysfs_and_mask() failed, 'unwind_idx' should be
>>> set to 'i - 1' to prevent double-free when cleanup resources.
>>>
>>> BUG: KASAN: double-free or invalid-free in __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
>>> Call Trace:
>>>   kfree+0x117/0x4c0
>>>   __iio_buffer_free_sysfs_and_mask+0x32/0xb0 [industrialio]
>>>   iio_buffers_alloc_sysfs_and_mask+0x60d/0x1570 [industrialio]
>>>   __iio_device_register+0x483/0x1a30 [industrialio]
>>>   ina2xx_probe+0x625/0x980 [ina2xx_adc]
>>>
>> Makes sense.
>> Thanks for the catch.
>>
>> Reviewed-by: Alexandru Ardelean <ardeleanalex@gmail.com>
> ...
>
>>>                  ret = __iio_buffer_alloc_sysfs_and_mask(buffer, indio_dev, i);
>>>                  if (ret) {
>>> -                       unwind_idx = i;
>>> +                       unwind_idx = i - 1;
>>>                          goto error_unwind_sysfs_and_mask;
> I prefer to see
>
> - for (; unwind_idx >= 0; unwind_idx--) {
> + while (unwind_idx--)
>
> instead.
With using while loop, 'unwind_idx = 
iio_dev_opaque->attached_buffers_cnt - 1;' need
be changed.
I think my change is clear and simple, do I need resend a new version 
with using while loop ?

Thanks,
Yang
>
>>>                  }
>>>          }