[6/9] vhost-scsi: new device supporting the tcm_vhost Linux kernel module

On Wed, 2013-05-29 at 21:29 -0700, Nicholas A. Bellinger wrote:
> On Thu, 2013-05-30 at 06:17 +0800, Asias He wrote:
> > On Wed, May 29, 2013 at 08:10:44AM -0700, Badari Pulavarty wrote:
> > > On 05/29/2013 02:05 AM, Wenchao Xia wrote:
> > > >于 2013-5-28 17:00, Wenchao Xia 写道:
> 
> <SNIP>
> 
> > > >  I have done a basic test of vhost-scsi, following is the result I'd
> > > >like to post, generally it seems fine:
> > > >
> > > >Result:
> > > >  fdisk/mkfs: fdisk can find it, mke2fs works fine.
> > > >  mount: can mount it.
> > > >  file I/O: dd 90M zero to a file in that disk succeed.
> > > 
> > > 
> > > 
> > > I tried without nested kvm.
> > > 
> > > >
> > > >Issues:
> > > >  1) in fdisk -l, sometime timeout with dmesg "end_request: I/O error,
> > > >dev  fd0, sector 0", I guess it is caused by nested KVM that failed
> > > >to kick host kernel?
> > > 
> > > 
> > > I don't see this issue. Are you sure "fd0" is actually the scsi device ?
> > > what is "fd0" ?
> > > 
> > > >  2) in fdisk -l, it shows 512 bytes larger than the parameter I
> > > >specified in fd_dev_size parameter in configfs on host.(shows
> > > >104858112 bytes, see the invocation script below)
> > > >
> > > 
> > > I see the same. For some reason "fdisk -l" in the VM shows
> > > 512-bytes more than the actual size for the file (on the host).
> > 
> > Hmm, interesting. Will look into it.
> > 
> > Nick, Any ideas here?
> > 
> 
> Mmm, fd_get_blocks() is not returning the expected minus one logical
> blocks with a !S_ISBLK() setup.
> 
> This is happening for every other backend ->get_blocks() call already,
> and should be happening for the fd_dev_size case as well.
> 
> Applying the following to target-pending.git now.
> 

Actually sorry, that last patch is not correct..

Here's a better one to properly set fd_dev->fd_block_size at configure
time, and use dev_attrib.block_size in fd_get_blocks() to allow for user
defined block_sizes.

Thanks,

--nab

commit 9e309f9307fe644dee8718980bfcb77de91ce38e
Author: Nicholas Bellinger <nab@linux-iscsi.org>
Date:   Wed May 29 21:35:23 2013 -0700

    target/file: Fix off-by-one READ_CAPACITY bug for !S_ISBLK export
    
    This patch fixes a bug where FILEIO was incorrectly reporting the number
    of logical blocks (+ 1) when using non struct block_device export mode.
    
    It changes fd_get_blocks() to follow all other backend ->get_blocks() cases,
    and reduces the calculated dev_size by one dev->dev_attrib.block_size
    number of bytes, and fixes the initial block_size assignment within
    fd_configure_device()
    
    Reported-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
    Reported-by: Badari Pulavarty <pbadari@us.ibm.com>
    Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>

On 05/29/2013 10:36 PM, Nicholas A. Bellinger wrote:
> On Wed, 2013-05-29 at 21:29 -0700, Nicholas A. Bellinger wrote:
>> On Thu, 2013-05-30 at 06:17 +0800, Asias He wrote:
>>> On Wed, May 29, 2013 at 08:10:44AM -0700, Badari Pulavarty wrote:
>>>> On 05/29/2013 02:05 AM, Wenchao Xia wrote:
>>>>> 于 2013-5-28 17:00, Wenchao Xia 写道:
>> <SNIP>
>>
>>>>>   I have done a basic test of vhost-scsi, following is the result I'd
>>>>> like to post, generally it seems fine:
>>>>>
>>>>> Result:
>>>>>   fdisk/mkfs: fdisk can find it, mke2fs works fine.
>>>>>   mount: can mount it.
>>>>>   file I/O: dd 90M zero to a file in that disk succeed.
>>>>
>>>>
>>>> I tried without nested kvm.
>>>>
>>>>> Issues:
>>>>>   1) in fdisk -l, sometime timeout with dmesg "end_request: I/O error,
>>>>> dev  fd0, sector 0", I guess it is caused by nested KVM that failed
>>>>> to kick host kernel?
>>>>
>>>> I don't see this issue. Are you sure "fd0" is actually the scsi device ?
>>>> what is "fd0" ?
>>>>
>>>>>   2) in fdisk -l, it shows 512 bytes larger than the parameter I
>>>>> specified in fd_dev_size parameter in configfs on host.(shows
>>>>> 104858112 bytes, see the invocation script below)
>>>>>
>>>> I see the same. For some reason "fdisk -l" in the VM shows
>>>> 512-bytes more than the actual size for the file (on the host).
>>> Hmm, interesting. Will look into it.
>>>
>>> Nick, Any ideas here?
>>>
>> Mmm, fd_get_blocks() is not returning the expected minus one logical
>> blocks with a !S_ISBLK() setup.
>>
>> This is happening for every other backend ->get_blocks() call already,
>> and should be happening for the fd_dev_size case as well.
>>
>> Applying the following to target-pending.git now.
>>
> Actually sorry, that last patch is not correct..
>
> Here's a better one to properly set fd_dev->fd_block_size at configure
> time, and use dev_attrib.block_size in fd_get_blocks() to allow for user
> defined block_sizes.
>
> Thanks,
>
> --nab
>
> commit 9e309f9307fe644dee8718980bfcb77de91ce38e
> Author: Nicholas Bellinger <nab@linux-iscsi.org>
> Date:   Wed May 29 21:35:23 2013 -0700
>
>      target/file: Fix off-by-one READ_CAPACITY bug for !S_ISBLK export
>      
>      This patch fixes a bug where FILEIO was incorrectly reporting the number
>      of logical blocks (+ 1) when using non struct block_device export mode.
>      
>      It changes fd_get_blocks() to follow all other backend ->get_blocks() cases,
>      and reduces the calculated dev_size by one dev->dev_attrib.block_size
>      number of bytes, and fixes the initial block_size assignment within
>      fd_configure_device()
>      
>      Reported-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
>      Reported-by: Badari Pulavarty <pbadari@us.ibm.com>
>      Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
>
> diff --git a/drivers/target/target_core_file.c b/drivers/target/target_core_file.c
> index 1b1d544..b11890d 100644
> --- a/drivers/target/target_core_file.c
> +++ b/drivers/target/target_core_file.c
> @@ -153,6 +153,7 @@ static int fd_configure_device(struct se_device *dev)
>                  struct request_queue *q = bdev_get_queue(inode->i_bdev);
>                  unsigned long long dev_size;
>
> +               fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
>                  /*
>                   * Determine the number of bytes from i_size_read() minus
>                   * one (1) logical sector from underlying struct block_device
> @@ -199,6 +200,7 @@ static int fd_configure_device(struct se_device *dev)
>                          goto fail;
>                  }
>
> +               fd_dev->fd_block_size = FD_BLOCKSIZE;
>                  /*
>                   * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
>                   */
> @@ -217,9 +219,7 @@ static int fd_configure_device(struct se_device *dev)
>                  dev->dev_attrib.max_write_same_len = 0x1000;
>          }
>
> -       fd_dev->fd_block_size = dev->dev_attrib.hw_block_size;
> -
> -       dev->dev_attrib.hw_block_size = FD_BLOCKSIZE;
> +       dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
>          dev->dev_attrib.hw_max_sectors = FD_MAX_SECTORS;
>          dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
>
> @@ -694,11 +694,12 @@ static sector_t fd_get_blocks(struct se_device *dev)
>           * to handle underlying block_device resize operations.
>           */
>          if (S_ISBLK(i->i_mode))
> -               dev_size = (i_size_read(i) - fd_dev->fd_block_size);
> +               dev_size = i_size_read(i);
>          else
>                  dev_size = fd_dev->fd_dev_size;
>
> -       return div_u64(dev_size, dev->dev_attrib.block_size);
> +       return div_u64(dev_size - dev->dev_attrib.block_size,
> +                      dev->dev_attrib.block_size);
>   }
>
>   static struct sbc_ops fd_sbc_ops = {
>
>
Verified and it shows the correct value now.

Thanks,
Badari

Hi Nicholas,

I am trying to test vhost-scsi with pscsi lio. I am able to create a 
target and
pass it to QEMU. I don't see any error from tcm_vhost or qemu. But the guest
VM doesn't see any SCSI device.

What am I missing ? Here is how I am using pscsi target.


wwpn="naa.6001405bd4e8476e"
tpgt=/sys/kernel/config/target/vhost/$wwpn/tpgt_0
nexus=$tpgt/nexus
mkdir -p $tpgt
echo -n $wwpn > $nexus

n=0
lun=$tpgt/lun/lun_${n}
data=/sys/kernel/config/target/core/pscsi_0/data_${n}
dsk=/dev/sdf
mkdir -p $lun
mkdir -p $data
echo -n ${dsk} > $data/udev_path
echo "scsi_host_id=1,scsi_channel_id=0,scsi_target_id=2,scsi_lun_id=0" > 
$data/control
echo -n 1 > $data/enable
ln -s $data $lun

Thanks,
Badari