crash on device unplug on macOS Catalina (10.15)

[Youw]

Typical usage scenario (simplified):

hid_init();
dev = hid_open(...);
// use the device
// ...

// device physically disconnected somewhere here

// ...
// all further hid_* funcitons are gracefully failing

hid_close(dev); // ok here

hid_exit(); // <-- macOS Catalina - crash here

The cause (in hid_close):

hidapi/mac/hid.c

Lines 1122 to 1124 in 65d22a9

	if (!dev->disconnected) {
	IOHIDDeviceClose(dev->device_handle, kIOHIDOptionsTypeSeizeDevice);
	}

+

hidapi/mac/hid.c

Lines 1097 to 1104 in 65d22a9

	if (!dev->disconnected) {
	IOHIDDeviceRegisterInputReportCallback(
	dev->device_handle, dev->input_report_buf, dev->max_input_report_len,
	NULL, dev);
	IOHIDDeviceRegisterRemovalCallback(dev->device_handle, NULL, dev);
	IOHIDDeviceUnscheduleFromRunLoop(dev->device_handle, dev->run_loop, dev->run_loop_mode);
	IOHIDDeviceScheduleWithRunLoop(dev->device_handle, CFRunLoopGetMain(), kCFRunLoopDefaultMode);
	}

The original comment from Alan:

hidapi/mac/hid.c

Lines 1119 to 1121 in 65d22a9

	/* Close the OS handle to the device, but only if it's not
	been unplugged. If it's been unplugged, then calling
	IOHIDDeviceClose() will crash. */

it's been unplugged, then calling
IOHIDDeviceClose() will crash

Apparently that was the case for the version of macOS, where this was originally developed.
In Catalina (10.15) update - it is the opposite case: if we skip closing the device, the app crashes on hid_exit() (IOHIDManagerClose).

---

Ideally, of course, we should always close all handles, and avoid any resource leaks. Unfortunately, cannot do it unconditionally as some users on older systems might experience crashes as the original author.

---

Currently I have access only to macOS Catalina and cannot check the behaviour on other systems.
Would be great to check how it behaves on older systems, at least up(down) to 10.10.

[vitalys]

@Youw I'd like to help you with this issue, but I can't reproduce it.

What I tried:

• Native Instruments Kontrol F1 device
• Test app that writes to device (to highlight buttons/segment indicators)
• My test app doesn't read for the device (is this an issue?)
• I used hidapi from master branch

What my app does:

1. hid_init
2. hid_open
3. hid_write, until Ctrl-C is pressed (during this step I log if device is active or unplugged)
4. hid_close
5. hid_exit
6. Additional printf before exiting

As result I can see final printf output, so my app doesn't crash
I tested using:

• macOS 10.15.6 (Macbook Pro)
• macOS 10.15.1 (VirtualBox)
• macOS 11.0 Beta 5 (VirtualBox)

Later today I'm going to try with macOS 11.0 Beta 6 using Macbook Pro

Do you have any specific scenario to reproduce this crash?

[Youw]

Try next:

1. hid_init
2. hid_open
3. Physically unplug the device
4. hid_close
5. hid_exit

[vitalys]

I've unplugged my device during write cycle, but it doesn't crash at the end
Let me simplify my test app and share it with you

[vitalys]

There is simplified test code:

    hid_device *handle;
    hid_init();
    handle = hid_open(0x17cc, 0x1120, NULL);

    printf("\nUnplug your device and press any key\n");
    getchar();

    hid_close(handle);
    hid_exit();

    printf("\nBye!\n");

It doesn't crash, I can see Bye! in my console

$ sw_vers 
ProductName:	Mac OS X
ProductVersion:	10.15.6
BuildVersion:	19G2021

[vitalys]

@Youw Is my test correct? Should I try other macOS versions?

[Youw]

It looks correct.
Its been a while since I tested this last time. I may have a different environment, when I did the test, like run all this in a separate thread, and had a UI windows in a main thread.

[vitalys]

I've just tried test UI app and made all calls to hidapi from separate queue (and separate thread)
It doesn't crash.

I tried to force call IOHIDDeviceClose in hid_close:

// if (!dev->disconnected) { 
 	IOHIDDeviceClose(dev->device_handle, kIOHIDOptionsTypeSeizeDevice); 
// } 

In this case app crashed in hid_exit

[Youw]

This is interesting, let me check on my side.
Would you be able to check on older version of macOS? (10.10-10-14) ?

[vitalys]

Would you be able to check on older version of macOS? (10.10-10-14) ?

Yes. That's why I proposed my help

[Youw]

I tried to force call IOHIDDeviceClose in hid_close

NOTE: if you do so, you also need to force call IOHIDDeviceRegisterInputReportCallback

	/*if (!dev->disconnected)*/ {
		IOHIDDeviceRegisterInputReportCallback( /* ... */ );
        }

a few lines before that.

---

Just checked on my machine (10.15.6):
hid_init/hid_open then a few hid_get/set_intput/output_report, (all in non-main thread) -> physically remove the device -> hid_close -> hid_exit <- crash here if I don't do IOHIDDeviceRegisterInputReportCallback+IOHIDDeviceClose

[Youw]

Apparently, I cannot reproduce the original crash in simple environments (e.g. a console application, or an app with just a single working thread.
The project, where I can reproduce it, uses a lot of stuff: it opens hid device (hidapi) + raw usb device (libusb) + UVC device (IOKit Framework) - those are parts of the single composite device. And the application communicates with all the devices.
Maybe a combination of all of them causes the issue I see.