Re: [PATCH 1/1] scsi: target: core: Add 8Fh VPD page

[David Disseldorp <ddiss@suse.de>]

On Mon, 16 Aug 2021 21:13:59 +0300, Roman Bolshakov wrote:

> On Fri, Aug 13, 2021 at 04:52:55PM +0200, David Disseldorp wrote:
> > Hi Sergey,
> > 
> > On Thu, 29 Jul 2021 23:19:43 +0300, Sergey Samoylenko wrote:
> >   
> > > The 8Fh VPD page announces the capabilities supported by
> > > the TCM XCOPY manager. It helps to expand the coverage of
> > > the third-party copy manager with SCSI testing utilities.  
> > 
> > Please list which initiators use this VPD page, if you know of any.
> > Also, is there any test coverage for this? I don't see anything in
> > libiscsi...
> >   
> 
> Hi David,
> 
> ESXi is one of the hosts that inspects Third Party Copy VPD Page.
> Windows detects ODX support using the page [1][2].

Thanks for the links. I haven't seen ESXi attempt to use it, but also
haven't checked for some time. It'd be good to get some of this
information in the commit message.

> The page is also used by libiscsi to detect presence and features of
> copy manager as was agreed with Bart in the PR [3]:

I'm probably missing something, but why wasn't the 3PC flag in the
standard inquiry page an option for this check?

> "Implementing REPORT SUPPORTED OPERATION CODES in LIO would require more
> work than implementing the third-party copy VPD page.
> 
> I'm fine with relying on the third-party copy VPD page, or in other
> words, to skip the copy offloading tests if that page is not supported.
> 
> There are plans to implement XCOPY support in the Linux kernel sd
> driver. If nobody else volunteers I plan to work on this myself. I'm
> considering to only support SCSI targets that support the third-party
> copy VPD page. Or in other words, we will need support for that VPD page
> anyway."

Okay, fair enough.

> 
> 1. https://www.slideshare.net/CalvinChen5/a-joint-effort-of-the-storage-industry
> 2. http://sg.danny.cz/sg/ddpt_xcopy_odx.html
> 3. https://github.com/sahlberg/libiscsi/pull/353
> 
> > > Reviewed-by: Konstantin Shelekhin <k.shelekhin@yadro.com>
> > > Reviewed-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
> > > Reviewed-by: Anastasia Kovaleva <a.kovaleva@yadro.com>
> > > Signed-off-by: Sergey Samoylenko <s.samoylenko@yadro.com>
> > > ---
> > >  drivers/target/target_core_spc.c | 230 ++++++++++++++++++++++++++++++-
> > >  1 file changed, 226 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/drivers/target/target_core_spc.c b/drivers/target/target_core_spc.c
> > > index 22703a0dbd07..169341712b10 100644
> > > --- a/drivers/target/target_core_spc.c
> > > +++ b/drivers/target/target_core_spc.c  
> > ...  
> > > +/* Third-party Copy VPD page */
> > > +static sense_reason_t
> > > +spc_emulate_evpd_8f(struct se_cmd *cmd, unsigned char *buf)
> > > +{
> > > +	struct se_device *dev = cmd->se_dev;
> > > +	int off;
> > > +	u16 page_len;
> > > +
> > > +	if (!dev->dev_attrib.emulate_3pc)
> > > +		return TCM_INVALID_CDB_FIELD;
> > > +
> > > +	/*
> > > +	 * Since the Third-party copy manager in TCM is quite simple
> > > +	 * and supports only two commands, the function sets
> > > +	 * many descriptor parameters as constants.
> > > +	 *
> > > +	 * As the Copy manager supports the EXTENDED COPY(LID1) command,
> > > +	 * the Third-party Copy VPD page should include five mandatory
> > > +	 * Third-party copy descriptors. Its are:
> > > +	 *   0001h - Supported Commands
> > > +	 *   0004h - Parameter Data
> > > +	 *   0008h - Supported Descriptors
> > > +	 *   000Ch - Supported CSCD Descriptor IDs
> > > +	 *   8001h - General Copy Operations
> > > +	 *
> > > +	 * See spc4 section 7.8.17
> > > +	 */
> > > +
> > > +	off = 4;
> > > +
> > > +	/* fill descriptors */
> > > +	off += spc_evpd_8f_encode_supp_cmds(&buf[off]);
> > > +	off += spc_evpd_8f_encode_param_data(&buf[off]);
> > > +	off += spc_evpd_8f_encode_supp_descrs(&buf[off]);
> > > +	off += spc_evpd_8f_encode_supp_cscd_descr_id(&buf[off]);
> > > +	off += spc_evpd_8f_encode_general_copy_ops(&buf[off]);  
> > 
> > This looks risky in terms of buf overrun. I think it'd be good to pass
> > a @remaining or @buf_end param to these helper functions.
> >   
> 
> It's doable but would require to change the signature of all existing
> VPD handlers. SE_INQUIRY_BUF is hardcoded to 1kb but it's also capped by
> EDTL to avoid buffer overruns:
> 
>   memcpy(rbuf, buf, min_t(u32, SE_INQUIRY_BUF, cmd->data_length));

That's checking the amount copied into the response buffer. My concern
is the prior writes to the staging buf.

Cheers, David