[PATCH] Perverting cciss

[PATCH] Perverting cciss

[Hannes Reinecke]

[-- Attachment #1: Type: text/plain, Size: 568 bytes --]

Hi All,

this patch adds the SG_IO ioctl to the cciss driver.
As the driver is capable of sending SCSI CDBs to the controller there is
no reason why we shouldn't exploit it.
This way we get to use all the nice sg_utils for the cciss driver.
And a persistent device name for free.

Makes one wonder why this one is not implemented as a SCSI driver in the
first place.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		      zSeries & Storage
hare@suse.de			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)

[-- Attachment #2: cciss-add-SG_IO --]
[-- Type: text/plain, Size: 5048 bytes --]

Add SG_IO ioctl to cciss

The cciss driver is actually capable of sending SCSI CDB to the
controller. So it would only be logical to have it accepting 
SG_IO ioctl, too.
Now we can use all the nice sg_utils and get a persistent device name
for free, too.

Signed-off-by: Hannes Reinecke <hare@suse.de>

diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
index 0c716ee..33e7b83 100644
--- a/drivers/block/cciss.c
+++ b/drivers/block/cciss.c
@@ -45,6 +45,9 @@ #include <linux/dma-mapping.h>
 #include <linux/blkdev.h>
 #include <linux/genhd.h>
 #include <linux/completion.h>
+#include <scsi/scsi.h>
+#include <scsi/scsi_ioctl.h>
+#include <scsi/sg.h>
 
 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
 #define DRIVER_NAME "HP CISS Driver (v 3.6.14)"
@@ -1152,6 +1155,163 @@ #endif
 			kfree(ioc);
 			return status;
 		}
+	case SG_IO: {
+		struct sg_io_hdr hdr;
+		CommandList_struct *c;
+		char *buff = NULL;
+		u64bit temp64;
+		unsigned long flags;
+		DECLARE_COMPLETION_ONSTACK(wait);
+
+		if (!capable(CAP_SYS_RAWIO))
+			return -EPERM;
+
+		if (copy_from_user(&hdr, argp, sizeof(hdr)))
+			return -EFAULT;
+
+		if (hdr.interface_id != 'S')
+			return -EINVAL;
+
+		/* cciss only supports 16-byte commands */
+		if (hdr.cmd_len > 16)
+			return -EINVAL;
+
+		/* We don't support proper scatter-gather (yet) */
+		if (hdr.iovec_count)
+			return -EINVAL;
+
+		if ((hdr.dxfer_len < 1) &&
+		    (hdr.dxfer_direction != SG_DXFER_NONE))
+			return -EINVAL;
+
+		if (hdr.dxfer_len > 0) {
+			buff = kmalloc(hdr.dxfer_len, GFP_KERNEL);
+			if (buff == NULL)
+				return -EFAULT;
+		}
+		if ((hdr.dxfer_direction == SG_DXFER_TO_DEV) ||
+		    (hdr.dxfer_direction == SG_DXFER_TO_FROM_DEV)) {
+			/* Copy the data into the buffer we created */
+			if (copy_from_user (buff, hdr.dxferp,
+					    hdr.dxfer_len)) {
+				kfree(buff);
+				return -EFAULT;
+			}
+		} else
+			memset(buff, 0, hdr.dxfer_len);
+
+		if ((c = cmd_alloc(host, 0)) == NULL) {
+			kfree(buff);
+			return -ENOMEM;
+		}
+
+		/* Copy CDB */
+		if (copy_from_user(c->Request.CDB, hdr.cmdp, hdr.cmd_len))
+			return -EFAULT;
+
+		/* Fill in the command type */
+		c->cmd_type = CMD_IOCTL_PEND;
+		/* Fill in Command Header */
+		c->Header.ReplyQueue = 0;
+		if (hdr.dxfer_len > 0) {
+			c->Header.SGList = 1;
+			c->Header.SGTotal = 1;
+		} else {
+			c->Header.SGList = 0;
+			c->Header.SGTotal = 0;
+		}
+		/* Default to LUN the ioctl was directed to */
+		c->Header.LUN.LogDev.VolId = drv->LunID & 0x3FFFFFFF;
+		c->Header.LUN.LogDev.Mode = 0x01; /* Logical volume */
+		c->Header.Tag.lower = c->busaddr;
+
+		/* Fill in Request block */
+		c->Request.CDBLen = hdr.cmd_len;
+		c->Request.Type.Type = TYPE_CMD;
+		c->Request.Type.Attribute = ATTR_SIMPLE;
+		switch(hdr.dxfer_direction) {
+		    case SG_DXFER_NONE:
+			c->Request.Type.Direction = XFER_NONE;
+			break;
+		    case SG_DXFER_TO_DEV:
+			c->Request.Type.Direction = XFER_WRITE;
+			break;
+		    case SG_DXFER_FROM_DEV:
+			c->Request.Type.Direction = XFER_READ;
+			break;
+		    case SG_DXFER_TO_FROM_DEV:
+			c->Request.Type.Direction = XFER_RSVD;
+			break;
+		}
+		c->Request.Timeout = hdr.timeout;
+
+		/* Fill in the scatter gather information */
+		if (hdr.dxfer_len > 0) {
+			temp64.val = pci_map_single(host->pdev, buff,
+						    hdr.dxfer_len,
+						    PCI_DMA_BIDIRECTIONAL);
+			c->SG[0].Addr.lower = temp64.val32.lower;
+			c->SG[0].Addr.upper = temp64.val32.upper;
+			c->SG[0].Len = hdr.dxfer_len;
+			c->SG[0].Ext = 0;
+		}
+		c->waiting = &wait;
+
+		/* Put the request on the tail of the request queue */
+		spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
+		addQ(&host->reqQ, c);
+		host->Qdepth++;
+		start_io(host);
+		spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
+
+		wait_for_completion(&wait);
+
+		/* unlock the buffers from DMA */
+		temp64.val32.lower = c->SG[0].Addr.lower;
+		temp64.val32.upper = c->SG[0].Addr.upper;
+		pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
+				 hdr.dxfer_len,
+				 PCI_DMA_BIDIRECTIONAL);
+
+		/* Copy the error information out */
+		hdr.status = c->err_info->ScsiStatus;
+		if (c->err_info->SenseLen && hdr.mx_sb_len > 0) {
+			int sense_len = c->err_info->SenseLen;
+
+			if (sense_len > hdr.mx_sb_len)
+				sense_len = hdr.mx_sb_len;
+
+			if (copy_to_user(hdr.sbp, c->err_info->SenseInfo,
+					 sense_len)) {
+				kfree(buff);
+				cmd_free(host, c, 0);
+				return -EFAULT;
+			}
+			hdr.sb_len_wr = sense_len;
+		}
+		hdr.resid = c->err_info->ResidualCnt;
+		/* Copy out the header */
+		if (copy_to_user(argp, &hdr, sizeof(hdr))) {
+			kfree(buff);
+			cmd_free(host, c, 0);
+			return -EFAULT;
+		}
+
+		if ((hdr.dxfer_direction == SG_DXFER_FROM_DEV) ||
+		    (hdr.dxfer_direction == SG_DXFER_TO_FROM_DEV)) {
+			/* Copy the data out to the buffer we created */
+			if (copy_to_user
+			    (hdr.dxferp, buff, hdr.dxfer_len)) {
+				kfree(buff);
+				cmd_free(host, c, 0);
+				return -EFAULT;
+			}
+		}
+
+		kfree(buff);
+		cmd_free(host, c, 0);
+		return 0;
+		}
 	default:
 		return -ENOTTY;
 	}

Re: [PATCH] Perverting cciss

[Christoph Hellwig]

On Thu, Apr 05, 2007 at 11:58:06AM +0200, Hannes Reinecke wrote:
> Hi All,
> 
> this patch adds the SG_IO ioctl to the cciss driver.
> As the driver is capable of sending SCSI CDBs to the controller there is
> no reason why we shouldn't exploit it.
> This way we get to use all the nice sg_utils for the cciss driver.
> And a persistent device name for free.

Instead of adding yet another implementation of SG_IO please implement
support for REQ_TYPE_BLOCK_PC requests and add all the nice block layer
passthrough ioctls to it.

> Makes one wonder why this one is not implemented as a SCSI driver in the
> first place.

It probably should have been a scsi driver, but it's far too late now.

Re: [PATCH] Perverting cciss

[James Bottomley]

On Thu, 2007-04-05 at 11:03 +0100, Christoph Hellwig wrote:
> On Thu, Apr 05, 2007 at 11:58:06AM +0200, Hannes Reinecke wrote:
> > Hi All,
> > 
> > this patch adds the SG_IO ioctl to the cciss driver.
> > As the driver is capable of sending SCSI CDBs to the controller there is
> > no reason why we shouldn't exploit it.
> > This way we get to use all the nice sg_utils for the cciss driver.
> > And a persistent device name for free.
> 
> Instead of adding yet another implementation of SG_IO please implement
> support for REQ_TYPE_BLOCK_PC requests and add all the nice block layer
> passthrough ioctls to it.

Actually, I happen to know that HP is in the process of implementing
this correctly (via REQ_TYPE_BLOCK_PC).  I can't reveal the details but
it has something to do with a well known Linux High Availability company
needing SG_IO for sg_persist to work ...

James

RE: [PATCH] Perverting cciss

[Miller, Mike (OS Dev)]

James wrote: 
> 
> On Thu, 2007-04-05 at 11:03 +0100, Christoph Hellwig wrote:
> > On Thu, Apr 05, 2007 at 11:58:06AM +0200, Hannes Reinecke wrote:
> > > Hi All,
> > > 
> > > this patch adds the SG_IO ioctl to the cciss driver.
> > > As the driver is capable of sending SCSI CDBs to the controller 
> > > there is no reason why we shouldn't exploit it.
> > > This way we get to use all the nice sg_utils for the cciss driver.
> > > And a persistent device name for free.
> > 
> > Instead of adding yet another implementation of SG_IO 
> please implement 
> > support for REQ_TYPE_BLOCK_PC requests and add all the nice block 
> > layer passthrough ioctls to it.
> 
> Actually, I happen to know that HP is in the process of 
> implementing this correctly (via REQ_TYPE_BLOCK_PC).  I can't 
> reveal the details but it has something to do with a well 
> known Linux High Availability company needing SG_IO for 
> sg_persist to work ...
> 
> James

NAK. Please do not add this patch to cciss. We are working SG_IO
internally and will post the patch soon.

mikem


> 
> 
>