fix(linux): report input access as denied when the event node is unreadable

[AalmanSadath]

Summary

Settings → Permissions checked /dev/uinput and a Logitech /dev/hidraw*, and
called both being accessible Granted. The hook also needs to read the mouse's
/dev/input/event* node, and that is not implied by the other two: logind
does not grant it on its own for a Bluetooth-direct mouse, whose node hangs off
/devices/virtual/misc/uhid and has no seat. docs/INSTALL-linux.md already
lists it as a third requirement.

So a machine carrying udev rules left by another Logitech manager reported
Granted while button remapping did nothing. HID++ rides hidraw, so the device
appeared and DPI and SmartShift worked; evdev::enumerate silently skips a node
it cannot open, so the mouse never reached the hook and nothing was logged.
Reporting that state as Granted is the one thing that page exists to prevent.

Changes

• openlogi-permissions: new probe_logitech_event(), identifying Logitech
  by sysfs device/id/vendor the way probe_logitech_hidraw already does, and
  returning the same tri-state (Some(true) readable / Some(false) present
  but denied / None none found). classify takes it as a third input.

  A None event probe deliberately does not block Granted. Only a
  definite denial downgrades, so a machine where sysfs cannot be read is never
  reported as broken.

  It counts any Logitech event node, not only the mouse's. Telling them apart
  means decoding sysfs capability bitmasks, and uaccess is granted per node by
  the same rule, so they stand or fall together.

• openlogi-desktop: the Denied hint named only the first two
  requirements, which would have sent users to check what was already working.
  It now names the event node and says another manager's rules often cover the
  first two but not the third.

PermissionStatus carries no serde derives and is not referenced from
openlogi-ipc, so this is not a wire change and PROTOCOL_VERSION is
untouched.

Testing

cargo fmt --all -- --check
cargo clippy --workspace --all-targets -- -D warnings
cargo test --workspace
RUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-deps --document-private-items \
  --exclude openlogi-ui --exclude openlogi-desktop --exclude openlogi-overlay --exclude openlogi-agent
cargo check -p openlogi-permissions --target x86_64-pc-windows-gnu
cargo xtask ci   # 8 passed, 1 skipped

All on Linux with RUSTFLAGS="-D warnings". The skipped xtask ci job is
tests (macos), not reproducible on this host, so not claimed green. Both
changed files are cfg-gated to Linux; the Windows cross-check covers the gating.

Verified on hardware, Fedora 44, MX Master 3S connected Bluetooth-direct,
the case this fix is about. Its event node sits at
/sys/devices/virtual/misc/uhid/0005:046D:B034.0024/input/input98, so logind
grants no ACL on its own and the bundled rule is what makes it readable.
Revoking that ACL with setfacl leaves /dev/uinput and /dev/hidraw*
accessible, which is the exact state this fixes:

Build	uinput	hidraw	event node	Reported
before	OK	OK	denied	Granted
after	OK	OK	denied	Not granted

Both return to Granted once the ACL is restored. Four new classify cases
cover the regression and the guard against over-eager denial.

Before fix:

After fix:

Follow-on from #529 / #530, which added the event node to the udev rules and
listed it in docs/INSTALL-linux.md as a third requirement. The permissions
probe was not updated at the time and still checked only uinput and hidraw,
so it reports Granted for a machine missing exactly the access #530 was about.

[greptile-apps]

Greptile Summary

The PR strengthens Linux permission reporting by checking every hook-relevant Logitech event node and ensuring a denied node cannot be masked by a readable sibling.

• Adds Logitech event-node discovery and capability-bitmap decoding.
• Gives permission denial precedence across all matching event nodes, independent of enumeration order.
• Extends classification and regression tests for mixed-access nodes.
• Updates the Settings guidance to describe all three required Linux device interfaces.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains, and the previously reported readable-node masking issue is fixed by scanning all matching nodes and giving permission denial precedence.

Important Files Changed

Filename	Overview
crates/openlogi-permissions/src/linux.rs	The event-node probe now scans all hook-relevant Logitech mouse nodes and deterministically reports denial when any matching node is inaccessible.
crates/openlogi-permissions/src/tests.rs	Tests cover event-node classification, denial precedence in both enumeration orders, transient errors, and capability-bitmap decoding.
crates/openlogi-desktop/src/windows/settings/permissions.rs	Linux permission guidance now explains the event-node access required for button hooks.

_{Reviews (2): Last reviewed commit: "fix(linux): count only hookable mouse no..." | Re-trigger Greptile}