Skip to contentSkip to conversation

HID: usbhid: sanitize hid->uniq against non-printable bytes

PATCH In review Taylor Hewetson 5 participants 8 messages Apr 18 – Apr 23

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 HID: usbhid: sanitize hid->uniq against non-printable bytes

Some USB HID devices (observed on ASUS ROG Azoth via its 2.4GHz dongle, USB ID 0b05:1a85) report an iSerialNumber string whose USB string descriptor declares a longer length than the actual serial, leaving uninitialized firmware memory - including control characters such as 0x18 - appended to the returned string.

These non-printable bytes propagate into hid->uniq, which in turn populates /sys/class/input/inputN/uniq. Downstream userspace components (systemd sd-device property_is_valid(), and by extension mutter input enumeration on GNOME Wayland sessions) reject devices with control characters in their uniq, rendering otherwise- functional input devices unusable in graphical sessions despite the kernel input layer correctly translating keypresses.

Truncate hid->uniq at the first byte outside the printable ASCII range (0x20..0x7e) after the serial is read.

---

drivers/hid/usbhid/hid-core.c +10 -1
@@ -1427,8 +1427,17 @@
 		snprintf(hid->phys + len, sizeof(hid->phys) - len,
 			 "%d", intf->altsetting[0].desc.bInterfaceNumber);
 
-	if (usb_string(dev, dev->descriptor.iSerialNumber, hid->uniq, 64) <= 0)
+	if (usb_string(dev, dev->descriptor.iSerialNumber, hid->uniq, 64) <= 0) {
 		hid->uniq[0] = 0;
+	} else {
+		size_t i;
+		for (i = 0; i < sizeof(hid->uniq) && hid->uniq[i]; i++) {
+			if (hid->uniq[i] < 0x20 || hid->uniq[i] > 0x7e) {
+				hid->uniq[i] = 0;
+				break;
+			}
+		}
+	}
 
 	usbhid = kzalloc(sizeof(*usbhid), GFP_KERNEL);
 	if (usbhid == NULL) {
#1 lore raw
Quoted 14 lines · Taylor Hewetson

Some USB HID devices (observed on ASUS ROG Azoth via its 2.4GHz dongle, USB ID 0b05:1a85) report an iSerialNumber string whose USB string descriptor declares a longer length than the actual serial, leaving uninitialized firmware memory - including control characters such as 0x18 - appended to the returned string.

These non-printable bytes propagate into hid->uniq, which in turn populates /sys/class/input/inputN/uniq. Downstream userspace components (systemd sd-device property_is_valid(), and by extension mutter input enumeration on GNOME Wayland sessions) reject devices with control characters in their uniq, rendering otherwise- functional input devices unusable in graphical sessions despite the kernel input layer correctly translating keypresses.

Quoted 2 lines

Truncate hid->uniq at the first byte outside the printable ASCII range (0x20..0x7e) after the serial is read.

Why aren't we doing this in the USB core instead of forcing all users of this to do it instead?

thanks,

greg k-h

#2 lore raw
Quoted 15 lines · Greg KH
Quoted 14 lines · Taylor Hewetson

Some USB HID devices (observed on ASUS ROG Azoth via its 2.4GHz dongle, USB ID 0b05:1a85) report an iSerialNumber string whose USB string descriptor declares a longer length than the actual serial, leaving uninitialized firmware memory - including control characters such as 0x18 - appended to the returned string.

These non-printable bytes propagate into hid->uniq, which in turn populates /sys/class/input/inputN/uniq. Downstream userspace components (systemd sd-device property_is_valid(), and by extension mutter input enumeration on GNOME Wayland sessions) reject devices with control characters in their uniq, rendering otherwise- functional input devices unusable in graphical sessions despite the kernel input layer correctly translating keypresses.

Quoted 5 lines
Quoted 2 lines

Truncate hid->uniq at the first byte outside the printable ASCII range (0x20..0x7e) after the serial is read.

Why aren't we doing this in the USB core instead of forcing all users of this to do it instead?

Should it be up to the kernel to do this as well? Currently this is only a problem with systemd because they added this check, and it looks like they have something in mind to fix it as well [1].

[1] https://github.com/systemd/systemd/issues/41339#issuecomment-4266429563

Quoted 4 lines

thanks,

greg k-h

Signature · 2 lines
Regards,
  Eric
#3 lore raw
Quoted 19 lines · Eric Naim
Quoted 18 lines · Greg KH
Quoted 16 lines · Taylor Hewetson

Some USB HID devices (observed on ASUS ROG Azoth via its 2.4GHz dongle, USB ID 0b05:1a85) report an iSerialNumber string whose USB string descriptor declares a longer length than the actual serial, leaving uninitialized firmware memory - including control characters such as 0x18 - appended to the returned string.

These non-printable bytes propagate into hid->uniq, which in turn populates /sys/class/input/inputN/uniq. Downstream userspace components (systemd sd-device property_is_valid(), and by extension mutter input enumeration on GNOME Wayland sessions) reject devices with control characters in their uniq, rendering otherwise- functional input devices unusable in graphical sessions despite the kernel input layer correctly translating keypresses.

Truncate hid->uniq at the first byte outside the printable ASCII range (0x20..0x7e) after the serial is read.

Quoted 8 lines
Quoted 2 lines

Why aren't we doing this in the USB core instead of forcing all users of this to do it instead?

Should it be up to the kernel to do this as well? Currently this is only a problem with systemd because they added this check, and it looks like they have something in mind to fix it as well [1].

[1] https://github.com/systemd/systemd/issues/41339#issuecomment-4266429563

It's either up to the kernel, or every single userspace program that reads the strings from a device. Might as well do it in one place, right?

thanks,

greg k-h

#4 lore raw
Quoted 27 lines · Greg KH
Quoted 26 lines · Eric Naim
Quoted 15 lines · Greg KH
Quoted 14 lines · Taylor Hewetson

Some USB HID devices (observed on ASUS ROG Azoth via its 2.4GHz dongle, USB ID 0b05:1a85) report an iSerialNumber string whose USB string descriptor declares a longer length than the actual serial, leaving uninitialized firmware memory - including control characters such as 0x18 - appended to the returned string.

These non-printable bytes propagate into hid->uniq, which in turn populates /sys/class/input/inputN/uniq. Downstream userspace components (systemd sd-device property_is_valid(), and by extension mutter input enumeration on GNOME Wayland sessions) reject devices with control characters in their uniq, rendering otherwise- functional input devices unusable in graphical sessions despite the kernel input layer correctly translating keypresses.

Quoted 5 lines
Quoted 2 lines

Truncate hid->uniq at the first byte outside the printable ASCII range (0x20..0x7e) after the serial is read.

Why aren't we doing this in the USB core instead of forcing all users of this to do it instead?

Should it be up to the kernel to do this as well? Currently this is only a problem with systemd because they added this check, and it looks like they have something in mind to fix it as well [1].

Quoted 5 lines
Quoted 1 line

[1] https://github.com/systemd/systemd/issues/41339#issuecomment-4266429563

It's either up to the kernel, or every single userspace program that reads the strings from a device. Might as well do it in one place, right?

That is a very good point. Thanks for the insight!

Quoted 4 lines

thanks,

greg k-h

Signature · 2 lines
Regards,
  Eric
#5 lore raw
In reply to Greg KH
Quoted 8 lines · Greg KH
Quoted 3 lines · Eric Naim

On 4/18/26 3:14 PM, Greg KH wrote:

[1] https://github.com/systemd/systemd/issues/41339#issuecomment-4266429563

It's either up to the kernel, or every single userspace program that reads the strings from a device. Might as well do it in one place, right?

No, because that puts the assumption that user space is not interested in what the device actually returns and uses these strings just for printing.
Eg. you can no longer use this in a udev rule.

Regards
Oliver

#6 lore raw
Quoted 8 lines · Oliver Neukum
Quoted 5 lines · Greg KH
Quoted 3 lines · Eric Naim

On 4/18/26 3:14 PM, Greg KH wrote:

[1] https://github.com/systemd/systemd/issues/41339#issuecomment-4266429563

Quoted 8 lines
Quoted 3 lines

It's either up to the kernel, or every single userspace program that reads the strings from a device. Might as well do it in one place, right?

No, because that puts the assumption that user space is not interested in what the device actually returns and uses these strings just for printing.
Eg. you can no longer use this in a udev rule.

Also, the USB spec doesn't say anything about what characters are or aren't allowed in a string descriptor. Anything, even a control character, is valid. The kernel should not rule them and certainly shouldn't truncate a string descriptor.

Alan Stern

#7 lore raw

USB: core: sanitize string descriptors against C0 control characters

In reply to Taylor Hewetson

Some USB devices report string descriptors with a declared length greater than the actual string, leaving uninitialized firmware memory
- often including C0 control characters such as 0x18 - appended to the returned string. This has been observed on the ASUS ROG Azoth 2.4GHz dongle (USB ID 0b05:1a85), where the trailing bytes make their way into hid->uniq and then /sys/class/input/inputN/uniq.

Downstream userspace components then reject the device. systemd's sd-device property_is_valid() treats any string property containing control characters as invalid and refuses to set ID_SERIAL_SHORT, which in turn prevents the device from being tagged with seat. On GNOME Wayland, mutter silently declines to open input devices that are missing this tagging, leaving the keyboard visible and producing keycodes at the kernel layer but dead to the user in a graphical session.

Truncate the returned UTF-8 string at the first C0 control character (0x00..0x1F) or DEL (0x7F). Printable Unicode beyond ASCII is left intact, so legitimate non-ASCII serials (e.g. European manufacturers) continue to work. Callers that previously received a string with trailing garbage now receive the clean leading portion, which is well-formed UTF-8 and safe for all downstream consumers.

---

Changes since v1:
- Move the sanitization from drivers/hid/usbhid/hid-core.c to drivers/usb/core/message.c so that all usb_string() callers benefit, not just usbhid. (Greg KH)
- Broaden the scope from "ASUS Azoth workaround" to "well-formed string guarantee for usb_string()"; update commit message accordingly.

v1: https://lore.kernel.org/all/[email protected]/

drivers/usb/core/message.c +19 -0
@@ -1052,6 +1052,25 @@
 			UTF16_LITTLE_ENDIAN, buf, size);
 	buf[err] = 0;
 
+	/*
+	 * Some devices report string descriptors with a declared length
+	 * greater than the actual serial, leaving uninitialized firmware
+	 * memory (often including C0 control characters) appended to the
+	 * returned string. Truncate at the first control character so
+	 * callers get a clean, well-formed string.
+	 */
+	{
+		int i;
+		for (i = 0; i < err; i++) {
+			unsigned char c = buf[i];
+			if (c < 0x20 || c == 0x7f) {
+				buf[i] = 0;
+				err = i;
+				break;
+			}
+		}
+	}
+
 	if (tbuf[1] != USB_DT_STRING)
 		dev_dbg(&dev->dev,
 			"wrong descriptor type %02x for string %d (\"%s\")\n",