-scanbus broken with hardware raid

[Joe Feise]

I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel card.
Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
cdrecord -scanbus shows this output:
Linux sg driver version: 3.5.34
cdrecord: Warning Linux Bus mapping botch.
cdrecord: Warning Linux Bus mapping botch.
cdrecord: Warning Linux Bus mapping botch.
cdrecord: Warning Linux Bus mapping botch.
Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
1995-2009 Jörg Schilling
Using libscg version 'schily-0.9'.
scsibus0:
        0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
        0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
        0,2,0     2) *
        0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
        0,4,0     4) *
        0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable CD-ROM
        0,6,0     6) *
        0,7,0     7) *

However, sgdiag (from http://scsirastools.sourceforge.net/) shows the correct
output:
-- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
 0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
 1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
 2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
 3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
 4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
 5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]

In other words, the RAID disks are listed at channel 0, the physical disks on
channel 1 are listed at channel 1, and the cdrom on channel 2 is listed on
channel 2, which is the correct layout.

-Joe

_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Joe Feise]

Joe Feise wrote on 11/06/09 08:44:

> I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel card.
> Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
> cdrecord -scanbus shows this output:
> Linux sg driver version: 3.5.34
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
> 1995-2009 Jörg Schilling
> Using libscg version 'schily-0.9'.
> scsibus0:
>         0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
>         0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
>         0,2,0     2) *
>         0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
>         0,4,0     4) *
>         0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable CD-ROM
>         0,6,0     6) *
>         0,7,0     7) *
>
> However, sgdiag (from http://scsirastools.sourceforge.net/) shows the correct
> output:
> -- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
>  0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
>  1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
>  2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
>  3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
>  4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
>  5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]
>
> In other words, the RAID disks are listed at channel 0, the physical disks on
> channel 1 are listed at channel 1, and the cdrom on channel 2 is listed on
> channel 2, which is the correct layout.

I forgot to mention, the system is a Linux box.
$: uname -a
Linux joe 2.6.29.6 #7 SMP PREEMPT Tue Oct 27 01:39:17 PDT 2009 i686 Intel(R)
Pentium(R) 4 CPU 3.40GHz GenuineIntel GNU/Linux

-Joe

_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Antonio Olivares-2]

Joe,

The kernel was not known from your output, but the following also
could show that it was a linux box:

......
 cdrecord -scanbus shows this output:
 Linux sg driver version: 3.5.34

So the above line would imply that it is no a linux box.  No offense
intended just an observation.

Also,
If you boot to an older kernel if possible, does the same thing happen?

Regards,

Antonio

On 11/6/09, Joe Feise <jfeise@...> wrote:

> Joe Feise wrote on 11/06/09 08:44:
>
>> I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel
>> card.
>> Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
>> cdrecord -scanbus shows this output:
>> Linux sg driver version: 3.5.34
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
>> 1995-2009 Jörg Schilling
>> Using libscg version 'schily-0.9'.
>> scsibus0:
>>         0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
>>         0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
>>         0,2,0     2) *
>>         0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
>>         0,4,0     4) *
>>         0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable CD-ROM
>>         0,6,0     6) *
>>         0,7,0     7) *
>>
>> However, sgdiag (from http://scsirastools.sourceforge.net/) shows the
>> correct
>> output:
>> -- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
>>  0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
>>  1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
>>  2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
>>  3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
>>  4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
>>  5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]
>>
>> In other words, the RAID disks are listed at channel 0, the physical disks
>> on
>> channel 1 are listed at channel 1, and the cdrom on channel 2 is listed on
>> channel 2, which is the correct layout.
>
> I forgot to mention, the system is a Linux box.
> $: uname -a
> Linux joe 2.6.29.6 #7 SMP PREEMPT Tue Oct 27 01:39:17 PDT 2009 i686 Intel(R)
> Pentium(R) 4 CPU 3.40GHz GenuineIntel GNU/Linux
>
> -Joe
>
> _______________________________________________
> Cdrecord-support mailing list
> Cdrecord-support@...
> https://lists.berlios.de/mailman/listinfo/cdrecord-support
>

_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Joerg Schilling-3]

Joe Feise <jfeise@...> wrote:

> I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel card.
> Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
> cdrecord -scanbus shows this output:
> Linux sg driver version: 3.5.34
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> cdrecord: Warning Linux Bus mapping botch.
> Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
> 1995-2009 Jörg Schilling
> Using libscg version 'schily-0.9'.
> scsibus0:
>         0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
>         0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
>         0,2,0     2) *
>         0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
>         0,4,0     4) *
>         0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable CD-ROM
>         0,6,0     6) *
>         0,7,0     7) *
>
> However, sgdiag (from http://scsirastools.sourceforge.net/) shows the correct
> output:
> -- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
>  0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
>  1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
>  2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
>  3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
>  4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
>  5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]
>
> In other words, the RAID disks are listed at channel 0, the physical disks on
> channel 1 are listed at channel 1, and the cdrom on channel 2 is listed on
> channel 2, which is the correct layout.

Linux drivers make a strange assumptions by calling the different SCSI busses
on a single card "channels" but without implementing a kernel interface that
allows to ask about the number of busses on that card. This makes it impossible
to implement a mapping from the Linux Scheme to what the SCSI standard defines
for the CAM interface.

Jörg

--
 EMail:joerg@... (home) Jörg Schilling D-13353 Berlin
       js@...                (uni)  
       joerg.schilling@... (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/private/ ftp://ftp.berlios.de/pub/schily
_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Michael Shell]

On Tue, 10 Nov 2009 16:40:05 +0100
Joerg.Schilling@... (Joerg Schilling) wrote:

> Linux drivers make a strange assumptions by calling the different SCSI
> busses on a single card "channels" but without implementing a kernel
> interface that allows to ask about the number of busses on that card.
> This makes it impossible to implement a mapping from the Linux Scheme
> to what the SCSI standard defines for the CAM interface.


I assume this is not an issue on Solaris. Just for comparison, how does
say, FreeBSD, rate in this regard?


  Cheers,

  Mike Shell

_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Joe Feise]

On Tue, November 10, 2009 07:40, Joerg Schilling wrote:

> Joe Feise <jfeise@...> wrote:
>
>> I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel
>> card.
>> Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
>> cdrecord -scanbus shows this output:
>> Linux sg driver version: 3.5.34
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
>> 1995-2009 Jörg Schilling
>> Using libscg version 'schily-0.9'.
>> scsibus0:
>>         0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
>>         0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
>>         0,2,0     2) *
>>         0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
>>         0,4,0     4) *
>>         0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable
>> CD-ROM
>>         0,6,0     6) *
>>         0,7,0     7) *
>>
>> However, sgdiag (from http://scsirastools.sourceforge.net/) shows the
>> correct
>> output:
>> -- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
>>  0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
>>  1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
>>  2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
>>  3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
>>  4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
>>  5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]
>>
>> In other words, the RAID disks are listed at channel 0, the physical
>> disks on
>> channel 1 are listed at channel 1, and the cdrom on channel 2 is listed
>> on
>> channel 2, which is the correct layout.
>
> Linux drivers make a strange assumptions by calling the different SCSI
> busses
> on a single card "channels" but without implementing a kernel interface
> that
> allows to ask about the number of busses on that card. This makes it
> impossible
> to implement a mapping from the Linux Scheme to what the SCSI standard
> defines
> for the CAM interface.

Well, it may be that the assumptions are strange, but as the sgdiag output
shows, it is possible to get this information.
Further, /proc/scsi/scsi also shows the correct channel, id, and lun.

-Joe


_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Joerg Schilling-3]

Michael Shell <list1@...> wrote:

> On Tue, 10 Nov 2009 16:40:05 +0100
> Joerg.Schilling@... (Joerg Schilling) wrote:
>
> > Linux drivers make a strange assumptions by calling the different SCSI
> > busses on a single card "channels" but without implementing a kernel
> > interface that allows to ask about the number of busses on that card.
> > This makes it impossible to implement a mapping from the Linux Scheme
> > to what the SCSI standard defines for the CAM interface.
>
>
> I assume this is not an issue on Solaris. Just for comparison, how does
> say, FreeBSD, rate in this regard?

FreeBSD implements a CAM standard interface and thus only offers the standard
addressing scheme bus,target,lun.

Jörg

--
 EMail:joerg@... (home) Jörg Schilling D-13353 Berlin
       js@...                (uni)  
       joerg.schilling@... (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/private/ ftp://ftp.berlios.de/pub/schily
_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support

[Matthias Andree]

Am 10.11.2009, 16:40 Uhr, schrieb Joerg Schilling  
<Joerg.Schilling@...>:

> Joe Feise <jfeise@...> wrote:
>
>> I have a SCSI hardware RAID system, using an Adaptec 2200S dual channel  
>> card.
>> Channel 1 has a RAID 1 and a RAID 0 disk, channel 2 has a CD-ROM drive.
>> cdrecord -scanbus shows this output:
>> Linux sg driver version: 3.5.34
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.
>> cdrecord: Warning Linux Bus mapping botch.

Rather missing feature or support for Linux interfaces.

Spare the world such useless warnings.
If that were a technical complaint, it might make sense.

>> Cdrecord-ProDVD-ProBD-Clone 2.01.01a67 (i686-pc-linux-gnu) Copyright (C)
>> 1995-2009 Jörg Schilling
>> Using libscg version 'schily-0.9'.
>> scsibus0:
>>         0,0,0     0) 'Adaptec ' '2200S Legacy    ' 'V1.0' Disk
>>         0,1,0     1) 'Adaptec ' 'Media           ' 'V1.0' Disk
>>         0,2,0     2) *
>>         0,3,0     3) 'MAXTOR  ' 'ATLAS10K5_147WLS' 'GVV0' Disk
>>         0,4,0     4) *
>>         0,5,0     5) 'YAMAHA  ' 'CRW-F1S         ' '1.0g' Removable  
>> CD-ROM
>>         0,6,0     6) *
>>         0,7,0     7) *
>>
>> However, sgdiag (from http://scsirastools.sourceforge.net/) shows the  
>> correct
>> output:
>> -- sgdiag v1.60 log started at Fri Nov  6 08:32:13 2009
>>  0 /dev/sg0 [0:0:0:0] Disk Adaptec  2200S Legacy     V1.0  [em]
>>  1 /dev/sg1 [0:0:1:0] Disk Adaptec  Media            V1.0  [em]
>>  2 /dev/sg2 [0:1:0:0] Disk ModusLnk                                 [em]
>>  3 /dev/sg3 [0:1:1:0] Disk ModusLnk MXJ3300SC800600W M108   ML0001  [em]
>>  4 /dev/sg4 [0:1:3:0] Disk MAXTOR   ATLAS10K5_147WLS GVV0 D4Y00QYK  [em]
>>  5 /dev/sg5 [0:2:5:0] CDRM YAMAHA   CRW-F1S          1.0g 16/04/03  [em]

                 | | | \___ LUN
                 | | \___ device ID
                 | \___ channel
                 \___ bus/host adaptor

...

> Linux drivers make a strange assumptions by calling the different SCSI  
> busses
> on a single card "channels" but without implementing a kernel interface  
> that
> allows to ask about the number of busses on that card. This makes it  
> impossible
> to implement a mapping from the Linux Scheme to what the SCSI standard  
> defines
> for the CAM interface.

This is false. How could sgdiag then have obtained the bus:channel:id:lun  
information that Joe showed? The truth is that sgdiag uses ioctl()s to  
obtain this information. Look at lines 695 - 706 of  
<http://scsirastools.svn.sourceforge.net/viewvc/scsirastools/trunk/src/sgcommon.c?revision=22&view=markup>,  
695-698, 703-706, 713f in particular.

And IIRC your code iterates over all /dev/sg* nodes anyways, so you get  
information about all relevant buses (those with devices attached).

Please don't ask me where the official documentation for these interfaces  
resides, I'm too unmotivated to dig that up for you, and it's not my  
software that invents (rather than queries the kernel) such IDs...

Even after all those years I fail to see why we conceptually need to map  
devices to the bus/ch/id/lun thing only to have cdrecord maps it back  
later. That's two needless indirections and *heaps* of code that needs to  
be cared for all the time. We should just be using a device node and be  
done.

--
Matthias Andree
_______________________________________________
Cdrecord-support mailing list
Cdrecord-support@...
https://lists.berlios.de/mailman/listinfo/cdrecord-support