Mobile hard disk drive: unexpected USB disconnect, and transient data errors, on Linux

[grahamperrin]

Kubuntu 25.04.

On a mobile drive that was previously used with FreeBSD-CURRENT, in an OpenZFS-encrypted dataset, I have data for various VirtualBox guests.

If I'm not mistaken:

• the USB disconnect below coincided with me using VirtualBox to save the state of a running guest

– a relatively intensive operation.

Tuesday 2025-07-08:

Jul 08 07:25:24 mowa219-gjp4 kernel: usb 4-5: USB disconnect, device number 3

I'm certain that neither the drive nor its cable was touched at the time. The drive was laid across (not pressing on) the keyboard of an HP ZBook 17 G2, docked to my right, whilst I used an external keyboard, mouse, and display to the left of the dock.

Question

Is anything tunable in Debian/Ubuntu, or OpenZFS, to reduce the likelihood of the kernel detecting/reporting a USB disconnect in a situation such as this?

Partial notes from the time of the incident

2025-07-08 07-25-24 pool Transcend device disconnected.txt

Observations

From accepted https://superuser.com/a/969710/84988:

Assuming you have a quality USB port, cable, and external drive, it should should be as safe and reliable as using an internal drive, especially if you have USB 3.0, …

The hardware here is fairly good quality.

With FreeBSD: I did occasionally (rarely) have difficulty with some VirtualBox guests that stored data in this pool. To the best of my recollection, none of the incidents involved data errors.

In this Linux case: the data errors were surprising, but not permanent. Easily cleared.

[AttilaFueloep]

Certain USB to SATA bridges show flaky behavior when used with UAS on Linux. This is a known issue for years. Connection drops and write errors are the usual indicators. The workaround is to use usb-storage mode for such a controller. I've added usb_storage.quirks=152d:0583:u,0bc2:61b7:u to my kernel command line years ago. Last time a tried without it was still broken. In usb-storage mode they are rock solid. This isn't a ZFS issue.

[grahamperrin]

… The workaround is to use usb-storage mode for such a controller. …

I was confused in an earlier comment about UAS, sorry.

usb-storage (not uas) is used, at the time of writing:

usbview

StoreJet Transcend
Manufacturer: StoreJet Transcend
Serial Number: S2S6J9FD203745      
Speed: 5Gb/s (super)
Bus:   4
Address:   4
USB Version:  3.00
Device Class: 00
Device Subclass: 00
Device Protocol: 00
Maximum Default Endpoint Size: 9
Number of Configurations: 1
Vendor Id: 174c
Product Id: 5106
Revision Number: 80.00

Config Number: 1
	Number of Interfaces: 1
	Attributes: c0
	MaxPower Needed: 144mA

	Interface Number: 0
		Name: usb-storage
		Alternate Number: 0
		Class: 08
		Sub Class: 06
		Protocol: 50
		Number of Endpoints: 2

			Endpoint Address: 02
			Direction: out
			Attribute: 2
			Type: Bulk
			Max Packet Size: 1024
			Interval: 0ms

			Endpoint Address: 81
			Direction: in
			Attribute: 2
			Type: Bulk
			Max Packet Size: 1024
			Interval: 0ms

[AttilaFueloep]

It also notes

With Linux, UAS is used

whatever, just my 2 ¢.

[grahamperrin]

Sorry, I misread https://superuser.com/a/930884/84988, back-to-front. Deleted my earlier comment.

[AttilaFueloep]

No worries. Well, if you're already using usb-storage then I've no idea how to improve stability, never encountered such a case myself.

[grahamperrin]

With FreeBSD: I did occasionally (rarely) have difficulty with some VirtualBox guests that stored data in this pool. To the best of my recollection, none of the incidents involved data errors.

In addition (if I recall correctly)

In FreeBSD host cases where no error was detectable by ZFS:

• the difficulties were most often with FreeBSD guests that used UFS, but not reproducible enough to draw a conclusion
• the guest became responsive as if an I/O error had occurred
• I might have looked at VirtualBox logs, I don't know whether any were saved for future diagnosis
• I probably never noted whether the host port used at the time was version 2.10 or 3.0 of USB.

In FreeBSD host cases where an error was detectable by ZFS:

• I don't recall any case where the actions suggested in Message ID: ZFS-8000-HC were possible

– I would habitually shut down FreeBSD (or force a stop of the computer, if a graceful shut down was impossible).

[grahamperrin]

Partial notes from the time of the incident

Additional notes from the time, clearance of the twenty-seven data errors:

2025-07-08 07-32 zpool clear Transcend, then I saved the VirtualBox guest.txt

In particular:

  scan: resilvered 336K in 00:00:01 with 0 errors on Tue Jul  8 07:32:16 2025

That, with Kubuntu, was extraordinary.

I feel certain that I never saw re-silvering, like that, with any FreeBSD use of ZFS on a mobile hard disk drive on USB.