Gleb Popov
e158e655cf
PR: 293585 Tested by: alven, kenrap Pull Request: https://github.com/freebsd/freebsd-ports/pull/512
38 lines
1.2 KiB
C
38 lines
1.2 KiB
C
When a process without full /dev/input access enumerates devices via libudev-devd,
|
|
the udev_device structs do not get udev properties that mark them as inputs, keyboards, etc,
|
|
and get rejected as not being input devices.
|
|
|
|
libinput reopens devices just to check path equality.
|
|
The udev_devices from reopening do have the right properties,
|
|
so we just use them instead of the original (enumerated) ones.
|
|
|
|
--- src/evdev.c.orig 2026-04-02 01:04:12 UTC
|
|
+++ src/evdev.c
|
|
@@ -1007,7 +1007,7 @@ evdev_sync_device(struct libinput *libinput, struct ev
|
|
|
|
evdev_device_dispatch_frame(libinput, device, frame);
|
|
|
|
- return rc == -EAGAIN ? 0 : rc;
|
|
+ return (rc == -EAGAIN || rc == -EINVAL)? 0 : rc;
|
|
}
|
|
|
|
static inline void
|
|
@@ -1105,6 +1105,17 @@ evdev_device_dispatch(void *data)
|
|
|
|
if (rc != -EAGAIN && rc != -EINTR) {
|
|
libinput_remove_source(libinput, device->source);
|
|
+ /*
|
|
+ * Dirty hack to allow cuse-based evdev backends to release
|
|
+ * character device file when device has been detached
|
|
+ * but still have it descriptor opened.
|
|
+ * Issuing evdev_device_suspend() here leads to SIGSEGV
|
|
+ */
|
|
+ int dummy_fd = open("/dev/null", O_RDONLY | O_CLOEXEC);
|
|
+ if (dummy_fd >= 0) {
|
|
+ dup2(dummy_fd, device->fd);
|
|
+ close(dummy_fd);
|
|
+ }
|
|
device->source = NULL;
|
|
}
|
|
}
|