Index | Thread | Search

From:
"James J. Lippard" <lippard-openbsd@discord.org>
Subject:
uhidev: fix USB_GET_REPORT returning EIO on report-ID devices
To:
tech@openbsd.org
Date:
Thu, 5 Nov 2026 10:27:13 -0700

Download raw body.

Thread
  • James J. Lippard:

    uhidev: fix USB_GET_REPORT returning EIO on report-ID devices

On any device with report IDs, USB_GET_REPORT on a uhid(4) child
always fails with EIO; USB_SET_REPORT works. I hit this with an APC
Back-UPS NS 1500M2, where all 57 uhid children fail while upd(4)
works.

uhidev_get_report() adds a byte for the report ID and strips it from
the data, but returns the transfer length including that byte, so the
!= size check in uhidev_ioctl() never
passes. uhidev_get_report_async_cb() already subtracts the ID byte,
which is why upd(4) is unaffected. This diff does the same in the sync
path, and treats a zero-length reply as an error, as 1.114 does in the
async path.

--- uhidev.c.orig	Thu Nov  5 10:22:00 2026
+++ uhidev.c	Thu Nov  5 10:24:23 2026
@@ -874,6 +874,10 @@
 
 	/* Skip the reportID. */
 	if (id > 0) {
+	  	if (actlen == 0)
+		  	actlen = -1;
+		else if (actlen > 0)
+			actlen--;
 		memcpy(data, buf + 1, len - 1);
 		free(buf, M_TEMP, len);
 	}

Unrelated, while looking at this file: commit 1.115 drops any
interrupt report longer than its declared size. Some HID devices pad
interrupt reports out to the full USB packet size; would it be better
to truncate those to the declared size rather than drop them? I
haven’t confirmed a device that does this.

-- 
Jim Lippard        lippard-openbsd@discord.org       https://discord.org/
GPG Key ID: 0x99FD5CD6