[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [RFC PATCH] blkif.h: document scsi/0x12/0x83 node

Bob Liu writes ("[RFC PATCH] blkif.h: document scsi/0x12/0x83 node"):
> Sometimes, we need to query VPD page=0x83 data from underlying
> storage so that vendor supplied software can run inside the VM and
> believe it's talking to the vendor's own storage.  But different
> vendors may have different special features, so it's not suitable to
> export through "feature-xxxx".
> One solution is query the whole VPD page through Xenstore node, which has
> already been used by windows pv driver.
> http://xenbits.xen.org/gitweb/?p=pvdrivers/win/xenvbd.git;a=blob;f=src/xenvbd/pdoinquiry.c

Thanks for your contribution.

Thanks also to Konrad for decoding the numbers, which really helps me
understand what is going on here and helped me find the relevant

(For background: I have just double-checked the SCSI spec and: INQUIRY
lets you query either the standard page, or one of a number of `vital
product data' pages, each identified by an 8-bit page number.  The VPD
pages are mostly full of vendor-specific data in vendor-specific

I have some qualms about the approach you have adopted.  It is
difficult to see how this feature could be used safely without
knowledge specific to the storage vendor.

But I think it is probably OK to define a specification along these
lines provided that it is very clear that if you aren't the storage
vendor and you use this and something breaks, you get to keep all the

> + * scsi/0x12/0x83
> + *   Values: string
> + *   A base64 formatted string providing VPD pages read out from backend
> + *   device.

I think this probably isn't the prettiest name for this node or
necessarily the best format but given that this protocol is already
deployed, and this syntax will do, I don't want to quibble.

I would like the base64 encoding to specified much more explicitly.
Just `base64 formatted' is too vague.

> + *   The backend driver or the toolstack should write this node with VPD
> + *   informations when attaching devices.

I think this is the wrong semantics.  I certainly don't want to
encourage backends to use this feature.

Rather, I would prefer something like this:

 * scsi/0x12/0x<NN>

   This optional node contains SCSI INQUIRY VPD information.
   <NN> is the hexadecimal representation of the VPD page code.

   A frontend which represents a Xen VBD to its containing operating
   system as a (virtual) SCSI target may return the specified data in
   response to INQUIRY commands from its containing OS.

   A frontend which supports this feature must return the backend-
   specified data for every INQUIRY command with the EVPD bit set.
   For EVPD=1 INQUIRY commands where the corresponding xenstore node
   does not exist, the frontend must report (to its containing OS) an
   appropriate failure condition.

   A frontend which does not support this feature (ie, which does not
   use these xenstore nodes), and which presents as a SCSI target to
   its containing OS, should support and provide whatever VPD
   information it considers appropriate, and should disregard these
   xenstore nodes.

   A frontend need not - and often will not - present to its
   containing OS as a device addressable with SCSI CDBs.  Such a
   frontend has no use for SCSI INQUIRY VPD information.

   A backend should set this information with caution.  Pages
   containing device-vendor-specific information should not be
   specified without the appropriate device-vendor-specific knowledge.

Also I have two other observations:

Firstly, AFAICT you have not provided any way to set the standard
INQUIRY response.  Is it not necessary in your application to provide
synthetic vendorid and productid, at the very least ?

Secondly, I think your hope that

> blkfront in Linux ... can use the same mechanism.

is I think misguided.  blkfront does not present the disk (to the rest
of the Linux storage system) as a SCSI device.  Rather, Linux allows
blkfront to present as a block device, directly, and this is what
blkfront does.

So blkfront would not consume this data.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.