KDL when booting anyboot image hrev53173

[MelanieFox]

I've tried booting the x86-64 anyboot iso in VirtualBox as usual. Unfortunately i get an "did not find any boot partitions!" KDL.

Screenshot attached.

[waddlesplash]

Please use VirtualBox's serial port emulation to get a syslog.

[MelanieFox]

Replying to waddlesplash:

Please use VirtualBox's serial port emulation to get a syslog.

Done.

[waddlesplash]

There does not appear to be a CD drive in the PCI device list, only a SATA controller with an empty hard drive attached...

How precisely did you connect the anyboot?

[MelanieFox]

The VBox CD-Rom drive is attached to the sata controller.

Using an IDE Controller with a CDROM Drive gave the same KDL.

[MelanieFox]

Some more testing turned out that the last working hrev is hrev53167. After that every hrev fails to boot.

[waddlesplash]

That would mean korli's SCSI changes are to blame here.

[korli]

I can't reproduce with VirtualBox 5.2.18 (both IDE and SATA). Is there anything to check in the settings?

[MelanieFox]

I use 6.0.8 for the tests, worked since hrev53167. Oddly i can reproduce the problem on recently enough baremetal.

Currently don't have a clue where the problem could be.

Very strange thing.

[simonsouth]

I've added a proposed fix for review here: ​https://review.haiku-os.org/c/haiku/+/1582

The issue turned to be a pair of calls to ioctl() that passed in a buffer but omitted the optional parameter specifying its length, causing ​a garbage value from va_arg() to be passed to _kern_ioctl() in its place.

With korli's changes in hrev53168 scsi_periph's periph_ioctl() actually does check this length value, meaning these calls would normally fail whenever they were directed at a SCSI device---like a CD-ROM drive. As a result, when Haiku was booted from CD KDiskDeviceManager would find the CD-ROM drive but later drop it from its list, removing it from consideration as a boot device and eventually leading to the kernel panic.

Once booted, the same issue in partitioning_system's Disc.cpp caused the desktop to hang with no Tracker.

For completeness my commit updates all the relevant ioctl calls in the affected modules (file_systems/cdda, partitioning_systems and disk_device_manager).

[waddlesplash]

Hey simonsouth, good to see you around again! :)

Very nice find! IMO we should make the kernel ioctl() always take 4 arguments. That would be the systemic solution ... but of course out of scope of this ticket.

Likely that patch will also fix #15075 then, so I'm adding it as a blocking bug.

[waddlesplash]

Merged in hrev53243.