[PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()

[PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()

[Miquel van Smoorenburg]

This should go into 2.6.28 and -stable, I think.

dpt_i2o-fix-xferlen-for-scsi_set_resid.patch

[PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()

dpt_i2o.c::adpt_i2o_to_scsi() reads the value at (reply+5) which
should contain the length in bytes of the transferred data. This
would be correct if reply was a u32 *. However it is a void * here,
so we need to read the value at (reply+20) instead.

The value at (reply+5) is usually 0xff0000, which is apparently
'large enough' and didn't cause any trouble until 2.6.27 where
427e59f09fdba387547106de7bab980b7fff77be caused this to become
visible through e.g. iostat -x .

Signed-off-by: Miquel van Smoorenburg <mikevs@xs4all.net>

--- linux-2.6.27.4/drivers/scsi/dpt_i2o.c.ORIG	2008-10-26 00:05:07.000000000 +0200
+++ linux-2.6.27.4/drivers/scsi/dpt_i2o.c	2008-11-04 23:43:13.000000000 +0100
@@ -2445,7 +2445,7 @@
 	hba_status = detailed_status >> 8;
 
 	// calculate resid for sg 
-	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(reply+5));
+	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(reply+20));
 
 	pHba = (adpt_hba*) cmd->device->host->hostdata[0];
 
@@ -2456,7 +2456,7 @@
 		case I2O_SCSI_DSC_SUCCESS:
 			cmd->result = (DID_OK << 16);
 			// handle underflow
-			if(readl(reply+5) < cmd->underflow ) {
+			if (readl(reply+20) < cmd->underflow) {
 				cmd->result = (DID_ERROR <<16);
 				printk(KERN_WARNING"%s: SCSI CMD underflow\n",pHba->name);
 			}

Re: [PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()

[Boaz Harrosh]

Miquel van Smoorenburg wrote:
> This should go into 2.6.28 and -stable, I think.
> 
> dpt_i2o-fix-xferlen-for-scsi_set_resid.patch
> 
> [PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()
> 
> dpt_i2o.c::adpt_i2o_to_scsi() reads the value at (reply+5) which
> should contain the length in bytes of the transferred data. This
> would be correct if reply was a u32 *. However it is a void * here,
> so we need to read the value at (reply+20) instead.
> 
> The value at (reply+5) is usually 0xff0000, which is apparently
> 'large enough' and didn't cause any trouble until 2.6.27 where
> 427e59f09fdba387547106de7bab980b7fff77be caused this to become
> visible through e.g. iostat -x .
> 
> Signed-off-by: Miquel van Smoorenburg <mikevs@xs4all.net>
> 
> --- linux-2.6.27.4/drivers/scsi/dpt_i2o.c.ORIG	2008-10-26 00:05:07.000000000 +0200
> +++ linux-2.6.27.4/drivers/scsi/dpt_i2o.c	2008-11-04 23:43:13.000000000 +0100
> @@ -2445,7 +2445,7 @@
>  	hba_status = detailed_status >> 8;
>  
>  	// calculate resid for sg 
> -	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(reply+5));
> +	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(reply+20));
>  

I do wish that someone could do a:
+	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(REG_BYTE_COUNT));

or what ever. To prevent just these errors from happening.

>  	pHba = (adpt_hba*) cmd->device->host->hostdata[0];
>  
> @@ -2456,7 +2456,7 @@
>  		case I2O_SCSI_DSC_SUCCESS:
>  			cmd->result = (DID_OK << 16);
>  			// handle underflow
> -			if(readl(reply+5) < cmd->underflow ) {
> +			if (readl(reply+20) < cmd->underflow) {

Look here it is used a gain. I would say it calls for a constant

>  				cmd->result = (DID_ERROR <<16);
>  				printk(KERN_WARNING"%s: SCSI CMD underflow\n",pHba->name);
>  			}

I understand that this is not your fault, but Just for the record, as a
"TODO:" comment for us all.

Thanks
Boaz

Re: [PATCH] dpt_i2o.c: fix transferred data length for scsi_set_resid()

[Miquel van Smoorenburg]

On Wed, 2008-11-05 at 16:05 +0200, Boaz Harrosh wrote:
> Miquel van Smoorenburg wrote:
> > This should go into 2.6.28 and -stable, I think.
> > 
> > dpt_i2o-fix-xferlen-for-scsi_set_resid.patch
>
> 
> I do wish that someone could do a:
> +	scsi_set_resid(cmd, scsi_bufflen(cmd) - readl(REG_BYTE_COUNT));
> 
> or what ever. To prevent just these errors from happening.
>
> I understand that this is not your fault, but Just for the record, as a
> "TODO:" comment for us all.

Yes, dpt_i2o could use some more cleaning up. A while ago I said I'd
look at that (after I did the 64-bit support patches), but due to
circumstances I have even less time available than usual.

Anyway, this is the minimal patch for 2.6.28 and 2.6.27-stable, which is
correct for -rc and -stable patches, right ?

Mike.