[UTMapp-bot]

Disconnecting a USB drive from a running VM froze UTM when the guest was using the drive through USB Attached SCSI, which most USB 3 drives and SSD enclosures do. The USB menu and the rest of the app stopped answering. After UTM was ended, the drive did not come back to macOS until it was unplugged and plugged in again.

When a device is released, usbredirhost_clear_device() cancels its transfers and waits for them to complete. libusb on macOS refused to cancel bulk stream transfers (darwin_cancel_transfer() returned LIBUSB_ERROR_INVALID_PARAM for LIBUSB_TRANSFER_TYPE_BULK_STREAM), and USB Attached SCSI always has one pending on its status pipe. The wait never ended, and it holds the lock that UTM's main thread takes in spice_usb_device_manager_is_device_connected().

This picks up c1c01e6593 from the UTM libusb fork (utmapp/libusb#1), which sends bulk stream transfers to darwin_abort_transfers(); that function already aborts a stream with AbortStreamsPipe(). The change is by @giovariot, who included it in #7928; this is that change on its own, with them as the patch's author. Upstream libusb's master has the same gap.

Resolves #5851

Related: #7530

Testing

Tested by an AI agent on a MacBook Pro M2 Max, macOS 27.2 (26B123), with a Seagate Backup Plus 500 GB drive (0bc2:ab24) on a USB 3 port (5 Gb/s link). Guest: Ubuntu 26.04 arm64 (Linux 7.0), USB 3.0 controller, the drive attached with the scripting connect command. In the guest the drive is Driver=uas, 5000M. Only reads were done on the drive.

• Before (the released UTM 5.0.8 (129)): the drive attaches and the guest sees the 500 GB disk. disconnect never returns ("AppleEvent timed out"), and every later scripting command times out too; a click on the USB menu in the VM window freezes the window. A sample of UTM shows a worker thread in spice_usb_backend_channel_detach → usbredirhost_set_device → usbredirhost_clear_device, and the main thread waiting in spice_usb_device_manager_is_device_connected → g_mutex_lock. After ending UTM, diskutil list did not show the drive again until it was replugged.
• After (a Debug build of main with libusb rebuilt from the patched source in its sysroot):
  • disconnect returns within a second, twice in a row, and the disk is gone from the guest each time. connect brings it back, and 200 MiB reads run at 102 MB/s.
  • disconnect while dd was reading from the drive returns within a second and UTM keeps answering. The drive connects again afterwards.
  • Powering the guest off with the drive attached ends the VM normally, and the drive is back in diskutil list without a replug.
  • The same sequence was repeated on two more builds that also carry a QEMU change for usb-redir (not part of this PR, and not active for a drive attached at SuperSpeed), with a 1000 MiB read at 105 MB/s before the disconnects: no freeze in any of them.

The patch file applies to a fresh extract of libusb-1.0.25.tar.bz2 with no fuzz or rejects. The tests above ran before the change was merged in the fork; the code line is the same.

Not tested: iOS, other drives, a drive using the older bulk-only protocol (its transfers were already cancelled), and a full dependency build (left to CI).