usb_disk fails to handle USB stick, panic KDL when trying to boot from it

[jua]

The changes to usb_disk driver in hrev46099 broke booting an anyboot-image from one of my USB sticks for me. When the disk icon lights up in the boot splash, it shows a KDL "PANIC: could not mount boot device". (With a different USB stick it stills works fine...)

Here's what I found out:

The updated usb_disk driver wrongly detects the USB stick as read-only device (the KPartition::Dump printed to the serial console while booting shows the boot partition flags as 0xa... while with the usb_disk driver from hrev46098 it is the correct 0x2). This later leads to failed mounting.

The problem has its roots in usb_disk_device_added(), in the part starting at line 795. For some reason, the problematic USB stick responds to the 'unit ready'-inquiry with ASC 0x2800: "Not ready to ready change, medium may have changed" (maybe a device quirk?). In scsi_sense.h this is associated with err_act_fail, so it will drop out of the loop without retrying... lun->write_protected doesn't get set to false and it will then wrongly assume a read-only medium.

As a test-hack, I overrode it by just setting lun->write_protected always to true. It then works, boots fine, without any further changes.

A better fix would be to set the entry for 0x2800 in sSCSISenseAscTable to err_act_retry. This fixes the problem, it will retry the command on the stick and the second time it succeeds and can mount the boot volume. Attaching a patch which does this. I'm not sure whether this is the correct solution though.

[jua]

Proposed patch

[jua]

As a test-hack, I overrode it by just setting lun->write_protected always to true

Correction: I meant write false there.

[kallisti5]

Looks ok to me. I'm seeing this issue as well. Going to test the patch and commit it.

[kallisti5]

applied in hrev46154. Works for me. Thanks!

[korli]

This would need to be tested with a USB CD-ROM device and checked for no regression. Thanks!

[jua]

Just tested it with a USB CD/DVD drive: booting from it still works fine with the patch.

[korli]

Just tested here with a USB CD/DVD device, the media change isn't detected anymore for me. I suppose the fix isn't that good. Could you try with the following patch? Thanks!

[korli]

accept B_DEV_MEDIA_CHANGED

[korli]

Reverted hrev46154 for now in hrev46261, reopened the bug.

[jua]

Sorry, I had only tested it with booting from CD, didn't test media changing...

I will try with the new patch as soon as I find time.

[korli]

ok jua. kallisti5, feel free to try the second patch if you can reproduce.

[bonefish]

(In #10155) So, to sum it up, the Corsair stick works and always did, and booting off the Transcend stick broke between hrev46255 and hrev46263. This supports jua's theory. Closing as duplicate of #10043.

[jua]

(In #10174) Thanks for reporting. Most probably duplicate of #10043.

[x-ist]

The second patch seems to be working for me! Tested successfully with an SD card and three different USB sticks. Patch was applied to hrev46378. Sorry for the noise on the ML!

[korli]

Applied in hrev46380. Thanks for the feedback! Please reopen if needed.