Driver for Open Profile for DICE

Open Profile for DICE is a protocol for deriving unique secrets at boot used by some Android devices. The firmware/bootloader hands over secrets in a reserved memory region, which this driver takes ownership of and exposes it to userspace via a misc device.

Userspace obtains the region's size using an ioctl and mmaps the memory to its address space. This mapping cannot be write+shared, giving userspace a guarantee that the secrets have not been overwritten by another process.

Userspace can also issue an ioctl requesting that the memory be wiped by the driver. Because both the kernel and userspace mappings use write-combine semantics, all clients will observe the memory as zeroed after the ioctl has returned.

Cc Andrew Scull <[email protected]>

Cc Will Deacon <[email protected]>

Signed-off-by David Brazdil <[email protected]>

---
 .../userspace-api/ioctl/ioctl-number.rst      |   1 +
 drivers/misc/Kconfig                          |   8 +
 drivers/misc/Makefile                         |   1 +
 drivers/misc/dice.c                           | 161 ++++++++++++++++++

 include/uapi/linux/dice.h                     |  14 ++
 5 files changed, 185 insertions(+)
 create mode 100644 drivers/misc/dice.c
 create mode 100644 include/uapi/linux/dice.h

Documentation/userspace-api/ioctl/ioctl-number.rst +0 -0

index cfe6cccf0f44..4b8bee2ffd1e 100644

Documentation/userspace-api/ioctl/ioctl-number.rst +0 -0

 0xAF  00-1F  linux/fsl_hypervisor.h                                  Freescale hypervisor
 0xB0  all                                                            RATIO devices in development:
                                                                      <mailto:[email protected]>

drivers/misc/Kconfig +0 -0

index 0f5a49fc7c9e..7165f4b6c41b 100644

drivers/misc/Kconfig +1 -0

+
 source "drivers/misc/c2port/Kconfig"
 source "drivers/misc/eeprom/Kconfig"
 source "drivers/misc/cb710/Kconfig"

drivers/misc/Makefile +1 -0

index a086197af544..f73c6bb23ccd 100644
@@ -59,3 +59,4 @@ obj-$(CONFIG_UACCE)		+= uacce/
 obj-$(CONFIG_XILINX_SDFEC)	+= xilinx_sdfec.o
 obj-$(CONFIG_HISI_HIKEY_USB)	+= hisi_hikey_usb.o
 obj-$(CONFIG_HI6421V600_IRQ)	+= hi6421v600-irq.o
+obj-$(CONFIG_DICE)		+= dice.o

drivers/misc/dice.c new file +37 -0

new file mode 100644
index 000000000000..06f3754feb71
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 - Google LLC
+ * Author: David Brazdil <[email protected]>
+ *
+ * Driver for Open Profile for DICE.
+ *
+ * This driver takes ownership of a reserved memory region containing secrets
+ * derived following the Open Profile for DICE. The contents of the memory
+ * region are not interpreted by the kernel but can be mapped into a userspace
+ * process via a misc device. The memory region can also be wiped, removing
+ * the secrets from memory.
+ *
+ * Userspace can access the data by (w/o error handling):
+ *
+ *     int fd = open("/dev/dice", O_RDONLY | O_CLOEXEC);
+ *     size_t size = ioctl(fd, DICE_GET_SIZE);
+ *     void *data = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+ *     ioctl(fd, DICE_WIPE);
+ *     close(fd);
+ */
+
+#include <linux/dice.h>
+#include <linux/io.h>
+#include <linux/miscdevice.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/of_reserved_mem.h>
+#include <linux/platform_device.h>
+
+static int dice_mmap(struct file *filp, struct vm_area_struct *vma);
+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg);
+
+static const struct file_operations dice_fops = {
+	.mmap = dice_mmap,
+	.unlocked_ioctl = dice_ioctl,
+};

drivers/misc/dice.c new file +17 -0

+
+static int dice_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+	/* Do not allow userspace to modify the underlying data. */
+	if ((vma->vm_flags & VM_WRITE) && (vma->vm_flags & VM_SHARED))
+		return -EPERM;
+
+	/* Create write-combine mapping so all clients observe a wipe. */
+	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTDUMP;
+	return vm_iomap_memory(vma, dice_rmem->base, dice_rmem->size);
+}
+
+static int dice_wipe(void)
+{
+	void *kaddr;
+

drivers/misc/dice.c new file +5 -0

+	spin_unlock(&dice_lock);
+	return 0;
+}
+
+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)

drivers/misc/dice.c line 74

Quoted 2 lines

drivers/misc/dice.c new file +2 -0

+	memzero_explicit(kaddr, dice_rmem->size);
+	devm_memunmap(dice_misc.this_device, kaddr);

Do you really need to call memzero_explicit()? This isn't "local" memory, if the compiler "optimizes away" a normal call, it would be _VERY_ broken.

Hi!

Quoted 6 lines

drivers/misc/dice.c line 74

Quoted 2 lines

drivers/misc/dice.c new file +2 -0

+	memzero_explicit(kaddr, dice_rmem->size);
+	devm_memunmap(dice_misc.this_device, kaddr);

Do you really need to call memzero_explicit()? This isn't "local" memory, if the compiler "optimizes away" a normal call, it would be _VERY_ broken.

For clearing secrets, I believe memzero_explicit is nice documentation.

Quoted 15 lines · David Brazdil

Open Profile for DICE is a protocol for deriving unique secrets at boot used by some Android devices. The firmware/bootloader hands over secrets in a reserved memory region, which this driver takes ownership of and exposes it to userspace via a misc device.

Userspace obtains the region's size using an ioctl and mmaps the memory to its address space. This mapping cannot be write+shared, giving userspace a guarantee that the secrets have not been overwritten by another process.

Userspace can also issue an ioctl requesting that the memory be wiped by the driver. Because both the kernel and userspace mappings use write-combine semantics, all clients will observe the memory as zeroed after the ioctl has returned.

Quoted 8 lines

Cc Andrew Scull <[email protected]>

Cc Will Deacon <[email protected]>

Signed-off-by David Brazdil <[email protected]>

---
 .../userspace-api/ioctl/ioctl-number.rst      |   1 +
 drivers/misc/Kconfig                          |   8 +
 drivers/misc/Makefile                         |   1 +
 drivers/misc/dice.c                           | 161 ++++++++++++++++++

Nice, almost 100 lines shorter than before!

Much better, thanks for the changes, but it can be made simpler, see comments below:

Quoted 6 lines

 include/uapi/linux/dice.h                     |  14 ++
 5 files changed, 185 insertions(+)
 create mode 100644 drivers/misc/dice.c
 create mode 100644 include/uapi/linux/dice.h

Documentation/userspace-api/ioctl/ioctl-number.rst Code Seq# Include File Comments · line 344

Quoted 9 lines

Documentation/userspace-api/ioctl/ioctl-number.rst +1 -0

index cfe6cccf0f44..4b8bee2ffd1e 100644
@@ -341,6 +341,7 @@ Code  Seq#    Include File                                           Comments
 0xAE  40-FF  linux/kvm.h                                             Kernel-based Virtual Machine
                                                                      <mailto:[email protected]>
 0xAE  20-3F  linux/nitro_enclaves.h                                  Nitro Enclaves
+0xAE  40-5F  uapi/linux/dice.h                                       Open Profile for DICE driver

Why the huge range? You are only really using 40 and 41. Stick to that please.

Quoted 9 lines

Documentation/userspace-api/ioctl/ioctl-number.rst +0 -0

 0xAF  00-1F  linux/fsl_hypervisor.h                                  Freescale hypervisor
 0xB0  all                                                            RATIO devices in development:
                                                                      <mailto:[email protected]>

drivers/misc/Kconfig +0 -0

index 0f5a49fc7c9e..7165f4b6c41b 100644

drivers/misc/Kconfig config HISI_HIKEY_USB · line 479

Quoted 10 lines

drivers/misc/Kconfig +7 -0

@@ -470,6 +470,14 @@ config HISI_HIKEY_USB
 	  switching between the dual-role USB-C port and the USB-A host ports
 	  using only one USB controller.
 
+config DICE
+	tristate "Open Profile for DICE driver"
+	depends on OF_RESERVED_MEM
+	help
+	  This driver allows to ownership of a reserved memory region
+	  containing DICE secrets and expose them to userspace via
+	  a character device.

What is the module name, please add that here.

And "dice" is a very generic name. I don't mind, but if you want to name it a bit more specific, that might be better.

Quoted 56 lines

drivers/misc/Kconfig +1 -0

+
 source "drivers/misc/c2port/Kconfig"
 source "drivers/misc/eeprom/Kconfig"
 source "drivers/misc/cb710/Kconfig"

drivers/misc/Makefile +1 -0

index a086197af544..f73c6bb23ccd 100644
@@ -59,3 +59,4 @@ obj-$(CONFIG_UACCE)		+= uacce/
 obj-$(CONFIG_XILINX_SDFEC)	+= xilinx_sdfec.o
 obj-$(CONFIG_HISI_HIKEY_USB)	+= hisi_hikey_usb.o
 obj-$(CONFIG_HI6421V600_IRQ)	+= hi6421v600-irq.o
+obj-$(CONFIG_DICE)		+= dice.o

drivers/misc/dice.c new file +37 -0

new file mode 100644
index 000000000000..06f3754feb71
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 - Google LLC
+ * Author: David Brazdil <[email protected]>
+ *
+ * Driver for Open Profile for DICE.
+ *
+ * This driver takes ownership of a reserved memory region containing secrets
+ * derived following the Open Profile for DICE. The contents of the memory
+ * region are not interpreted by the kernel but can be mapped into a userspace
+ * process via a misc device. The memory region can also be wiped, removing
+ * the secrets from memory.
+ *
+ * Userspace can access the data by (w/o error handling):
+ *
+ *     int fd = open("/dev/dice", O_RDONLY | O_CLOEXEC);
+ *     size_t size = ioctl(fd, DICE_GET_SIZE);
+ *     void *data = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+ *     ioctl(fd, DICE_WIPE);
+ *     close(fd);
+ */
+
+#include <linux/dice.h>
+#include <linux/io.h>
+#include <linux/miscdevice.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/of_reserved_mem.h>
+#include <linux/platform_device.h>
+
+static int dice_mmap(struct file *filp, struct vm_area_struct *vma);
+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg);
+
+static const struct file_operations dice_fops = {
+	.mmap = dice_mmap,
+	.unlocked_ioctl = dice_ioctl,
+};

drivers/misc/dice.c line 47

Quoted 10 lines

drivers/misc/dice.c new file +10 -0

+
+static struct miscdevice dice_misc = {
+	.name	= "dice",
+	.minor	= MISC_DYNAMIC_MINOR,
+	.fops	= &dice_fops,
+	.mode	= 0400,
+};
+
+static struct reserved_mem *dice_rmem;
+static DEFINE_SPINLOCK(dice_lock);

These should be per-device, not global, right? Please put them in a local structure that lives off of the platform device. Then everything is dynamic and nothing is static making it all much simpler again.

Do that and you can get rid of the pre-definitions of dice_mmap() and dice_ioctl() above as well.0

Quoted 23 lines

drivers/misc/dice.c new file +23 -0

+
+static int dice_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+	/* Do not allow userspace to modify the underlying data. */
+	if ((vma->vm_flags & VM_WRITE) && (vma->vm_flags & VM_SHARED))
+		return -EPERM;
+
+	/* Create write-combine mapping so all clients observe a wipe. */
+	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTDUMP;
+	return vm_iomap_memory(vma, dice_rmem->base, dice_rmem->size);
+}
+
+static int dice_wipe(void)
+{
+	void *kaddr;
+
+	spin_lock(&dice_lock);
+	kaddr = devm_memremap(dice_misc.this_device, dice_rmem->base,
+			      dice_rmem->size, MEMREMAP_WC);
+	if (IS_ERR(kaddr)) {
+		spin_unlock(&dice_lock);
+		return PTR_ERR(kaddr);

drivers/misc/dice.c line 74

Quoted 4 lines

drivers/misc/dice.c new file +4 -0

+	}
+
+	memzero_explicit(kaddr, dice_rmem->size);
+	devm_memunmap(dice_misc.this_device, kaddr);

Do you really need to call memzero_explicit()? This isn't "local" memory, if the compiler "optimizes away" a normal call, it would be _VERY_ broken.

Quoted 9 lines

drivers/misc/dice.c new file +9 -0

+	spin_unlock(&dice_lock);
+	return 0;
+}
+
+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+	switch (cmd) {
+	case DICE_GET_SIZE:
+		/* Checked against INT_MAX in dice_probe(). */

drivers/misc/dice.c line 89

Quoted 6 lines

drivers/misc/dice.c new file +6 -0

+		return dice_rmem->size;
+	case DICE_WIPE:
+		return dice_wipe();
+	}
+
+	return -ENOIOCTLCMD;

-ENOTTY please.

As you only have 2 ioctls, why not just use read/write for this? Write would cause dice_wipe() to happen, and read would return the size in the buffer provided. Then no ioctl is needed at all.

drivers/misc/dice.c line 100

Quoted 12 lines

drivers/misc/dice.c new file +11 -0

+}
+
+static int __init dice_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct device_node *rmem_np;
+	struct reserved_mem *rmem;
+	int ret;
+
+	if (dice_rmem) {
+		dev_err(dev, "only one instance of device allowed\n");

Why? That's a device-specific thing, not a driver-specific thing. Make this all dynamic and you don't need to care about it at all.

Again, simplification :)

thanks,

greg k-h

Quoted 4 lines · Greg Kroah-Hartman

What is the module name, please add that here.

And "dice" is a very generic name. I don't mind, but if you want to name it a bit more specific, that might be better.

Does "open-dice" sound good? I think that's the shorthand used on the official website.

Quoted 5 lines

Quoted 5 lines

drivers/misc/dice.c new file +5 -0

+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+	switch (cmd) {
+	case DICE_GET_SIZE:
+		/* Checked against INT_MAX in dice_probe(). */

Quoted 8 lines

drivers/misc/dice.c line 89

Quoted 6 lines

drivers/misc/dice.c new file +6 -0

+		return dice_rmem->size;
+	case DICE_WIPE:
+		return dice_wipe();
+	}
+
+	return -ENOIOCTLCMD;

-ENOTTY please.

I have no personal attachment to ENOIOCTLCMD, but it is documented as "no ioctl command" and converted to ENOTTY before returning to userspace. That made me think this was the right thing to do.

Quoted 3 lines

As you only have 2 ioctls, why not just use read/write for this? Write would cause dice_wipe() to happen, and read would return the size in the buffer provided. Then no ioctl is needed at all.

Fine by me but does feel like a bit of a hack. Is that a common pattern?

Quoted 7 lines · David Brazdil

Quoted 4 lines · Greg Kroah-Hartman

What is the module name, please add that here.

And "dice" is a very generic name. I don't mind, but if you want to name it a bit more specific, that might be better.

Does "open-dice" sound good? I think that's the shorthand used on the official website.

That might be better.

Naming is hard.

Quoted 13 lines

Quoted 5 lines

Quoted 5 lines

drivers/misc/dice.c new file +5 -0

+static long dice_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+	switch (cmd) {
+	case DICE_GET_SIZE:
+		/* Checked against INT_MAX in dice_probe(). */

Quoted 8 lines

drivers/misc/dice.c line 89

Quoted 6 lines

drivers/misc/dice.c new file +6 -0

+		return dice_rmem->size;
+	case DICE_WIPE:
+		return dice_wipe();
+	}
+
+	return -ENOIOCTLCMD;

-ENOTTY please.

Quoted 3 lines

I have no personal attachment to ENOIOCTLCMD, but it is documented as "no ioctl command" and converted to ENOTTY before returning to userspace. That made me think this was the right thing to do.

ENOTTY is better please.

Quoted 3 lines

Quoted 3 lines

As you only have 2 ioctls, why not just use read/write for this? Write would cause dice_wipe() to happen, and read would return the size in the buffer provided. Then no ioctl is needed at all.

Quoted 1 line

Fine by me but does feel like a bit of a hack. Is that a common pattern?

ioctls are hacks too :)

read/write like this is fine to do, might make the code simpler, and allow the code to be used by scripts easier. At the very least, wipe can be done by any language instead of only those that allow ioctls.

In your first email you also mentioned removing the check in dice_probe() that only allows a single instance. On a second thought, I think it's simpler to keep it there for now, even if the memory is dynamically allocated, which I agree makes the code cleaner.

The reason being that if we allowed multiple instances, we'd also need some static unique identifier that ties the cdev filename to the DT entry, same as /dev/disk/by-uuid/. Just adding an index number to the misc device nodename based on DT probe order sounds very fragile, and anything more sophisticated sounds like too much trouble for something we don't have a clear use case for right now.

Hi!

drivers/misc/Kconfig config HISI_HIKEY_USB · line 480

Quoted 8 lines

drivers/misc/Kconfig +8 -0

+config DICE
+	tristate "Open Profile for DICE driver"
+	depends on OF_RESERVED_MEM
+	help
+	  This driver allows to ownership of a reserved memory region
+	  containing DICE secrets and expose them to userspace via
+	  a character device.
+

Explaining what DICE is (and what Open Profile is) would be useful.

I see it is for some kind of DRM? Why is in non-evil and why do we want it in Linux?