asusd: open scsi_generic /dev/sgN for SCSI Aura devices

[NB-Group]

SCSI Aura devices (e.g. ROG STRIX Arion, 0b05:1932) showed up on D-Bus and asusctl scsi reported success, but no traffic ever reached the enclosure. The LEDs never changed.

write_effect() opened the block node (/dev/sdX) and issued SG_IO with the vendor CDB. SG_IO with a vendor command on a block node requires CAP_SYS_RAWIO, but the hardened asusd unit strips every capability (empty CapabilityBoundingSet, NoNewPrivileges=true), so every ioctl returned EPERM. write_effect() then did perform(task).ok() and threw the error away, so asusd reported success while zero commands hit the wire. strace showed ioctl(N, SG_IO, ...) = -1 EPERM sixteen times per write, and usbmon showed only baseline usb-storage traffic.

Fix: resolve the scsi_generic node (/dev/sgN) backing the block device and open that instead. The sg driver gates access at open() through the file permissions, so SG_IO works with no capabilities, the same path sg3_utils and OpenRGB use. I also stopped swallowing the ioctl error in write_effect so this can't go silent again.

Verified on an Arion: 16/16 SG_IO return 0, the ENE vendor CDBs (mode 0x8021, colour registers 0x8160+, apply 0x80a0) show up in usbmon, and the LEDs change.

Two files, no kernel changes. This is the working userspace reference for the SCSI device-handler driver we discussed.

On hotplug the sg node can lag the block node by a moment, so init retries briefly before falling back to the block device; at startup the node already exists, so the first attempt hits with no delay.

[coderabbitai]

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Improved SCSI device detection for more reliable device initialization, including hotplug scenarios.
  • Added a fallback when a generic SCSI device node is unavailable.
  • SCSI effect execution failures are now logged while remaining effects continue processing.

Walkthrough

The SCSI path discovers and prefers existing /dev/sgN nodes. It retries discovery during hotplug and falls back to the block node when needed. SCSI task execution failures are logged while processing continues.

Changes

SCSI device handling

Layer / File(s)	Summary
Generic node discovery and initialization asusd/src/aura_manager.rs	The manager walks ancestor devices to find an existing /dev/sgN node. Initialization retries discovery up to eight times, then uses the generic node or logs a warning before falling back to the block node.
SCSI execution diagnostics asusd/src/aura_scsi/mod.rs	Failed SCSI tasks now log their errors. Remaining tasks continue, and write_effect returns Ok(()).

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• OpenGamingCollective/asusctl#211: Refactors the same SCSI device-handling areas.

Suggested labels: asusd, rog-scsi, fix

Suggested reviewers: ghoul4500, scardracs

Sequence Diagram(s)

sequenceDiagram
  participant AuraManager
  participant Sysfs
  participant DevSg
  participant ScsiAura

  AuraManager->>Sysfs: Walk ancestor devices
  Sysfs-->>AuraManager: Return scsi_generic entries
  AuraManager->>DevSg: Check /dev/sgN
  DevSg-->>AuraManager: Return generic node or unavailable
  AuraManager->>ScsiAura: Initialize with /dev/sgN or block node
  ScsiAura->>ScsiAura: Execute SCSI tasks
  ScsiAura-->>ScsiAura: Log task errors and continue

Loading

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the main change: opening the SCSI generic device for SCSI Aura devices.
Description check	✅ Passed	The description clearly explains the problem, motivation, implementation, testing, and hotplug behavior, but omits issue and environment details.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@asusd/src/aura_manager.rs`:
- Around line 317-335: Update the SCSI device-event handling around the
block-node initialization and monitor matching so a later scsi_generic
(/dev/sgN) add event triggers initialization retry for the same device. Replace
the retained /dev/sdX handle or re-run the relevant DeviceHandle initialization
once sg_node_for_block resolves, while preserving existing behavior when the
block event arrives after the generic node.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: bfefcb49-ea96-4492-aa70-c133dd53aeb2

📥 Commits

Reviewing files that changed from the base of the PR and between 3d3c048 and 538ed80.

📒 Files selected for processing (2)

• asusd/src/aura_manager.rs
• asusd/src/aura_scsi/mod.rs

📜 Review details

🔇 Additional comments (1)

asusd/src/aura_scsi/mod.rs (1)

30-35: LGTM!

[scardracs]

BTW, we own sg.rs too if you want to take a look into it too. The one on crates.io is not supported anymore, like the one made by luke back in 2025, so I took the liberty to pick his version up and import it directly into code.

[NB-Group]

Nice, took a look. sg.rs sidesteps the cmd_len trap I hit in the kernel driver: it sets cmd_len from the slice length, so the 16-byte vendor CDB works as-is. The kernel sizes the CDB by opcode, 0xec maps to 10, cdb[13] (the ENE data-length byte) drops silently, and I had to hand-build the request there to force 16.