Skip to contentSkip to conversation

iio: buffer: Ensure bounce buffer used for unaligned case is zeroed.

PATCH Superseded Jonathan Cameron 3 participants 6 messages Jun 4 – Jun 14 IIO

lore
Share & tools
lore

Jonathan Cameron

author maintainer
#0 lore raw

PATCH v2 1/1 iio: buffer: Ensure bounce buffer used for unaligned case is zeroed.

iio_push_to_buffers_with_ts_unaligned() leaks uninitialized heap memory to userspace if the data passed in is not a multiple of 8 bytes and the timestamp is enabled. Use memset() to zero it after resizing.

Fixes 95ec3fdf2b79 ("iio: core: Introduce iio_push_to_buffers_with_ts_unaligned()")

---
v2: Zero the new buffer not the old one (which might not exist)
Drop bonus Reported-by:

---
 drivers/iio/industrialio-buffer.c | 1 +
 1 file changed, 1 insertion(+)
drivers/iio/industrialio-buffer.c +1 -0
index 5c3df993bea2..157b0415091a 100644
@@ -2445,6 +2445,7 @@ int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev,
 				   indio_dev->scan_bytes, GFP_KERNEL);
 		if (!bb)
 			return -ENOMEM;
+		memset(bb, 0, indio_dev->scan_bytes);
 		iio_dev_opaque->bounce_buffer = bb;
 		iio_dev_opaque->bounce_buffer_size = indio_dev->scan_bytes;
 	}
Signature · 1 line
2.54.0
#1 lore raw
Quoted 3 lines · Jonathan Cameron

iio_push_to_buffers_with_ts_unaligned() leaks uninitialized heap memory to userspace if the data passed in is not a multiple of 8 bytes and the timestamp is enabled. Use memset() to zero it after resizing.

...

drivers/iio/industrialio-buffer.c int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev, · line 2448

Quoted 4 lines
drivers/iio/industrialio-buffer.c +1 -0
 				   indio_dev->scan_bytes, GFP_KERNEL);
 		if (!bb)
 			return -ENOMEM;
+		memset(bb, 0, indio_dev->scan_bytes);

May I suggest different approach, id est use __GFP_ZERO instead of hunting correct pointers?

With Best Regards,
Andy Shevchenko

author maintainer
#2 lore raw
Quoted 7 lines · On Thu, 4 Jun 2026 12:10:42 +0300 Andy Shevchenko…
Quoted 3 lines · Jonathan Cameron

iio_push_to_buffers_with_ts_unaligned() leaks uninitialized heap memory to userspace if the data passed in is not a multiple of 8 bytes and the timestamp is enabled. Use memset() to zero it after resizing.

...

Quoted 7 lines

drivers/iio/industrialio-buffer.c int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev, · line 2448

Quoted 4 lines
drivers/iio/industrialio-buffer.c +1 -0
 				   indio_dev->scan_bytes, GFP_KERNEL);
 		if (!bb)
 			return -ENOMEM;
+		memset(bb, 0, indio_dev->scan_bytes);  

May I suggest different approach, id est use __GFP_ZERO instead of hunting correct pointers?

That's a weird beast when combined with a krealloc so I was a bit nervous about readability (and less so whether it was correct). It should be fine in that we will either get stale data or zeros because we always use this path to allocate the buffer so if you think it is obviously fine then I don't mind.

The oddities are that if an object grows within a slab but doesn't need a new one we are relying on those bits happening to be zero based on the original allocation doing __GFP_ZERO as well (as it's the same call)

I'm nervous though as that region off the end is sometimes used for debug objects and I really don't understand that bit of slab well enough. If it actually does this, then seems like we'd end up with a lot of nasty corner cases so I assume it doesn't. However, I couldn't convince myself enough not to just force a memset of the whole thing.

Quoted 6 lines

With Best Regards,
Andy Shevchenko

#3 lore raw
Quoted 3 lines · Jonathan Cameron
Quoted 1 line · On Thu, 4 Jun 2026 12:10:42 +0300 Andy Shevchenko…

On Thu, Jun 04, 2026 at 09:43:07AM +0100, Jonathan Cameron wrote:

...

Quoted 14 lines
Quoted 7 lines

drivers/iio/industrialio-buffer.c int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev, · line 2448

Quoted 4 lines
drivers/iio/industrialio-buffer.c +1 -0
 				   indio_dev->scan_bytes, GFP_KERNEL);
 		if (!bb)
 			return -ENOMEM;
+		memset(bb, 0, indio_dev->scan_bytes);  

May I suggest different approach, id est use __GFP_ZERO instead of hunting correct pointers?

That's a weird beast when combined with a krealloc so I was a bit nervous about readability (and less so whether it was correct). It should be fine in that we will either get stale data or zeros because we always use this path to allocate the buffer so if you think it is obviously fine then I don't mind.

Quoted 8 lines

The oddities are that if an object grows within a slab but doesn't need a new one we are relying on those bits happening to be zero based on the original allocation doing __GFP_ZERO as well (as it's the same call)

I'm nervous though as that region off the end is sometimes used for debug objects and I really don't understand that bit of slab well enough. If it actually does this, then seems like we'd end up with a lot of nasty corner cases so I assume it doesn't.

Such a bug will be a serious issue in mm. I don't think it exists. But, of course, chances are not completely 0.

Current users (not a comprehensive list)

arch/arm64/kvm/nested.c:92 drivers/firmware/efi/capsule-loader.c:61 drivers/gpio/gpiolib.c:5198 drivers/media/platform/amphion/vdec.c:329 drivers/net/ethernet/intel/ice/ice_lib.c:3044 drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c:785 drivers/net/wireless/ti/wl18xx/main.c:1582 drivers/nvme/target/pci-epf.c:708 drivers/spi/spi.c:1411 drivers/usb/gadget/function/uvc_configfs.c:880 fs/ceph/mdsmap.c:322
fs/smb/server/ksmbd_work.c:123 lib/tests/slub_kunit.c:273 sound/soc/meson/meson-card-utils.c:49

Also note the kernel-doc for krealloc_array(). It has some note WRT __GFP_ZERO and it means the flag must be supported. Also there is a note in mm/slub.c.

Quoted 2 lines

However, I couldn't convince myself enough not to just force a memset of the whole thing.

I personally would go with __GFP_ZERO.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#4 lore raw
Quoted 50 lines · Andy Shevchenko
Quoted 3 lines · Jonathan Cameron
Quoted 1 line · On Thu, 4 Jun 2026 12:10:42 +0300 Andy Shevchenko…

On Thu, Jun 04, 2026 at 09:43:07AM +0100, Jonathan Cameron wrote:

...

Quoted 14 lines
Quoted 7 lines

drivers/iio/industrialio-buffer.c int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev, · line 2448

Quoted 4 lines

     indio_dev->scan_bytes, GFP_KERNEL); if (!bb)
  return -ENOMEM;

drivers/iio/industrialio-buffer.c +1 -0
+		memset(bb, 0, indio_dev->scan_bytes);  

May I suggest different approach, id est use __GFP_ZERO instead of hunting correct pointers?

That's a weird beast when combined with a krealloc so I was a bit nervous about readability (and less so whether it was correct). It should be fine in that we will either get stale data or zeros because we always use this path to allocate the buffer so if you think it is obviously fine then I don't mind.

Quoted 8 lines

The oddities are that if an object grows within a slab but doesn't need a new one we are relying on those bits happening to be zero based on the original allocation doing __GFP_ZERO as well (as it's the same call)

I'm nervous though as that region off the end is sometimes used for debug objects and I really don't understand that bit of slab well enough. If it actually does this, then seems like we'd end up with a lot of nasty corner cases so I assume it doesn't.

Such a bug will be a serious issue in mm. I don't think it exists. But, of course, chances are not completely 0.

Current users (not a comprehensive list)

arch/arm64/kvm/nested.c:92 drivers/firmware/efi/capsule-loader.c:61 drivers/gpio/gpiolib.c:5198 drivers/media/platform/amphion/vdec.c:329 drivers/net/ethernet/intel/ice/ice_lib.c:3044 drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c:785 drivers/net/wireless/ti/wl18xx/main.c:1582 drivers/nvme/target/pci-epf.c:708 drivers/spi/spi.c:1411 drivers/usb/gadget/function/uvc_configfs.c:880 fs/ceph/mdsmap.c:322
fs/smb/server/ksmbd_work.c:123 lib/tests/slub_kunit.c:273 sound/soc/meson/meson-card-utils.c:49

Quoted 7 lines

Also note the kernel-doc for krealloc_array(). It has some note WRT __GFP_ZERO and it means the flag must be supported. Also there is a note in mm/slub.c.

Quoted 2 lines

However, I couldn't convince myself enough not to just force a memset of the whole thing.

I personally would go with __GFP_ZERO.

If I'm not missing anything, this should also be proof that the flag should be fine:

https://elixir.bootlin.com/linux/v7.1-rc7/source/lib/tests/slub_kunit.c#L275

- Nuno Sá

author maintainer
#5 lore raw
Quoted 54 lines · On Mon, 08 Jun 2026 21:00:30 +0100 Nuno Sá <[email protected]> wrote:
Quoted 53 lines · Andy Shevchenko
Quoted 3 lines · Jonathan Cameron
Quoted 1 line · On Thu, 4 Jun 2026 12:10:42 +0300 Andy Shevchenko…

On Thu, Jun 04, 2026 at 09:43:07AM +0100, Jonathan Cameron wrote:

...

Quoted 14 lines
Quoted 7 lines

drivers/iio/industrialio-buffer.c int iio_push_to_buffers_with_ts_unaligned(struct iio_dev *indio_dev, · line 2448

Quoted 4 lines

     indio_dev->scan_bytes, GFP_KERNEL); if (!bb)
  return -ENOMEM;

drivers/iio/industrialio-buffer.c +1 -0
+		memset(bb, 0, indio_dev->scan_bytes);    

May I suggest different approach, id est use __GFP_ZERO instead of hunting correct pointers?

That's a weird beast when combined with a krealloc so I was a bit nervous about readability (and less so whether it was correct). It should be fine in that we will either get stale data or zeros because we always use this path to allocate the buffer so if you think it is obviously fine then I don't mind.

Quoted 8 lines

The oddities are that if an object grows within a slab but doesn't need a new one we are relying on those bits happening to be zero based on the original allocation doing __GFP_ZERO as well (as it's the same call)

I'm nervous though as that region off the end is sometimes used for debug objects and I really don't understand that bit of slab well enough. If it actually does this, then seems like we'd end up with a lot of nasty corner cases so I assume it doesn't.

Such a bug will be a serious issue in mm. I don't think it exists. But, of course, chances are not completely 0.

Current users (not a comprehensive list)

arch/arm64/kvm/nested.c:92 drivers/firmware/efi/capsule-loader.c:61 drivers/gpio/gpiolib.c:5198 drivers/media/platform/amphion/vdec.c:329 drivers/net/ethernet/intel/ice/ice_lib.c:3044 drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c:785 drivers/net/wireless/ti/wl18xx/main.c:1582 drivers/nvme/target/pci-epf.c:708 drivers/spi/spi.c:1411 drivers/usb/gadget/function/uvc_configfs.c:880 fs/ceph/mdsmap.c:322
fs/smb/server/ksmbd_work.c:123 lib/tests/slub_kunit.c:273 sound/soc/meson/meson-card-utils.c:49

Also note the kernel-doc for krealloc_array(). It has some note WRT __GFP_ZERO and it means the flag must be supported. Also there is a note in mm/slub.c.

Quoted 9 lines
Quoted 4 lines
Quoted 2 lines

However, I couldn't convince myself enough not to just force a memset of the whole thing.

I personally would go with __GFP_ZERO.

If I'm not missing anything, this should also be proof that the flag should be fine:

https://elixir.bootlin.com/linux/v7.1-rc7/source/lib/tests/slub_kunit.c#L275

Perfect! Thanks. I'll hopefully spin an updated version in next few days

Jonathan

Quoted 1 line

- Nuno Sá