Autodetection under OpenBSD assumes full SCSI for sd-Devices

[Martin Ziemer]

Under OpenBSD not only SCSI devices, but also sata devices have names like /dev/sd0.

When autodetectiong such a device, smartmontools uses the full scsi commandset instead of only sat commands.

One way to fix this would be to change line 592 in os_openbsd.cpp to something like this:
return new openbsd_scsi_device(this, name, "", true /*scanning*/);

[Christian Franke]

When autodetectiong such a device, smartmontools uses the full scsi commandset instead of only sat commands.

Which smartctl command is used?
Which SCSI commands are erroneously issued?
Is SAT detected later or not?

One way to fix this would be to change line 592 in os_openbsd.cpp to something like this:
return new openbsd_scsi_device(this, name, "", true /*scanning*/);

The scanning flag should only set during smartctl --scan or smartd.conf DEVICESCAN. The autodetect code in os_openbsd.cpp may need some rework.

[Martin Ziemer]

The command was just smartctl -a /dev/sd0c.

I dont know, how to see the scsi-command, which was issued.

I found the solution by reproducing the behaviour of smartmontools 7.1, where the sat-flag was given.

In the daemon-process it is detected correctly. Also there are no Problems, if i add "-d sat" to the commandline.

[Martin Ziemer]

To complete the Bug-Desription Command and output as copy/pasta:

(126) horrad@dozer:/usr/ports/sysutils/smartmontools $ doas /usr/obj/ports/smartmontools-7.2/build-amd64/smartctl -a /dev/sd0c        
smartctl 7.2 2020-12-30 r5155 [OpenBSD 6.9 amd64] (local build)
Copyright (C) 2002-20, Bruce Allen, Christian Franke, www.smartmontools.org

User Capacity:        2,000,398,934,016 bytes [2.00 TB]
Logical block size:   512 bytes
Physical block size:  4096 bytes
LB provisioning type: not reported [LBPME=1, LBPRZ=0]
Form Factor:          2.5 inches
>> Terminate command early due to bad response to IEC mode page
A mandatory SMART command failed: exiting. To continue, add one or more '-T permissive' options.

[Christian Franke]

Needs to be further investigated by an OpenBSD expert. Leaving open as undecided for now.

[Alex Samorukov]

I can try to take a look. Which OpenBSD version do you use?

[Alex Samorukov]

Ok, was able to reproduce it in vbox, using virtual ahci controller and OpenBSD 6.8.

Please try this patch:

os_openbsd.cpp

@@ -588 +588 @@
       // XXX get USB vendor ID, product ID and version from sd(4)/umass(4).
       // XXX check sat device via get_usb_dev_type_by_id().
 
-      // No USB bridge found, assume regular SCSI device
-      return get_scsi_device(name, "scsi");
+      // No USB bridge found, assume regular SCSI or SAT device
+      return get_scsi_device(name, "");
     }
     if (!strncmp(net_dev_scsi_tape, test_name, strlen(net_dev_scsi_tape)))
       return get_scsi_device(name, "scsi");

And let me know if it works for you.

TODO: check if same apply to netbsd which is based on the same code logic.

[Alex Samorukov]

P.S. had a quick look on the kernel to see if its easy/possible to get usb device detection attached to scsi and seems that this information is not exported to userland. I would recommend to request that from OpenBSD folks if you want to have this in the smartmontools. Specifically ioctl needs to return bus type or driver (e.g. usb or umass) and get easy way to get usb vid/pid using it.

[Martin Ziemer]

Replying to Alex Samorukov:

Ok, was able to reproduce it in vbox, using virtual ahci controller and OpenBSD 6.8.

Please try this patch:

os_openbsd.cpp

@@ -588 +588 @@
       // XXX get USB vendor ID, product ID and version from sd(4)/umass(4).
       // XXX check sat device via get_usb_dev_type_by_id().
 
-      // No USB bridge found, assume regular SCSI device
-      return get_scsi_device(name, "scsi");
+      // No USB bridge found, assume regular SCSI or SAT device
+      return get_scsi_device(name, "");
     }
     if (!strncmp(net_dev_scsi_tape, test_name, strlen(net_dev_scsi_tape)))
       return get_scsi_device(name, "scsi");

And let me know if it works for you.

TODO: check if same apply to netbsd which is based on the same code logic.

This patch works for me.
BUT i have only sata and USB disks to test at hand. -- So i dont know, of it breaks real scsi disks!

[Alex Samorukov]

Thank you for testing. It should not break real scsi. There is a way to be even more strict and do SAT check ONLY when sc_bus != SCSI, but i think it is overkill + device is not yet open when we doing return get_scsi_device. Also checked with netbsd - it should not be relevant as NetBSD always adds AHCI devices as wd* and not sd*. We also doing sat check by default on Linux/FreeBSD, so that should not harm.

[Alex Samorukov]

Should be fixed in r5215