Patchwork Use SCSI command to get size of SG device

login
register
mail settings
Submitter Chen HanXiao
Date Oct. 9, 2012, 9:42 a.m.
Message ID <00c501cda602$583de850$08b9b8f0$@cn.fujitsu.com>
Download mbox | patch
Permalink /patch/190307/
State New
Headers show

Comments

Chen HanXiao - Oct. 9, 2012, 9:42 a.m.
When we use SCSI generic device as disk image, function lseek
could not get the size of this kind of device.
So try to use SCSI command Read Capacity(10) when lseek failed to get
the size of SCSI generic device.

Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
---
 block/raw-posix.c |   46 ++++++++++++++++++++++++++++++++++++++++++++--
 1 files changed, 44 insertions(+), 2 deletions(-)
Christoph Hellwig - Oct. 9, 2012, 5:21 p.m.
On Tue, Oct 09, 2012 at 05:42:01PM +0800, Chen HanXiao wrote:
> When we use SCSI generic device as disk image, function lseek
> could not get the size of this kind of device.
> So try to use SCSI command Read Capacity(10) when lseek failed to get
> the size of SCSI generic device.

Eww, this is ugly as hell.  Why would you even need the size for a raw
passthrough device?
Chen HanXiao - Oct. 10, 2012, 2:11 a.m.
Hi
> -----Original Message-----
> From: Christoph Hellwig [mailto:hch@lst.de]
> Sent: Wednesday, October 10, 2012 1:21 AM
> To: Chen HanXiao
> Cc: qemu-devel@nongnu.org
> Subject: Re: [Qemu-devel] [PATCH] Use SCSI command to get size of SG
device
> 
> On Tue, Oct 09, 2012 at 05:42:01PM +0800, Chen HanXiao wrote:
> > When we use SCSI generic device as disk image, function lseek could
> > not get the size of this kind of device.
> > So try to use SCSI command Read Capacity(10) when lseek failed to get
> > the size of SCSI generic device.
> 
> Eww, this is ugly as hell.  Why would you even need the size for a raw
> passthrough device?

If we want to enable snapshot for SCSI generic device as disk image, the
size of 
SCSI generic device is needed. Function lseek could not get this, SCSI
command 
can finish the job.
Only when lseek failed would Read Capacity command be sent. 

Regards
Paolo Bonzini - Oct. 10, 2012, 6:33 a.m.
Il 10/10/2012 04:11, Chen HanXiao ha scritto:
>> > 
>> > On Tue, Oct 09, 2012 at 05:42:01PM +0800, Chen HanXiao wrote:
>>> > > When we use SCSI generic device as disk image, function lseek could
>>> > > not get the size of this kind of device.
>>> > > So try to use SCSI command Read Capacity(10) when lseek failed to get
>>> > > the size of SCSI generic device.
>> > 
>> > Eww, this is ugly as hell.  Why would you even need the size for a raw
>> > passthrough device?
>
> If we want to enable snapshot for SCSI generic device as disk image, the size
> of SCSI generic device is needed. Function lseek could not get this, SCSI
> command can finish the job.
> Only when lseek failed would Read Capacity command be sent. 

You need to use scsi-block instead of scsi-generic.  However, I don't
see how this can work.  After the snapshot, the image will be qcow2, not
raw, and thus it will not support bdrv_aio_ioctl.  Hence any SCSI
command (for scsi-generic) or any non-data SCSI command (for scsi-block)
will fail.

Can you give an example of what exactly you are trying to do?

Paolo
Chen HanXiao - Oct. 10, 2012, 8:33 a.m.
> -----Original Message-----
> From: Paolo Bonzini [mailto:paolo.bonzini@gmail.com] On Behalf Of Paolo
> Bonzini
> Sent: Wednesday, October 10, 2012 2:34 PM
> To: Chen HanXiao
> Cc: 'Christoph Hellwig'; qemu-devel@nongnu.org
> Subject: Re: [PATCH] Use SCSI command to get size of SG device
> 
> Il 10/10/2012 04:11, Chen HanXiao ha scritto:
> >> >
> >> > On Tue, Oct 09, 2012 at 05:42:01PM +0800, Chen HanXiao wrote:
> >>> > > When we use SCSI generic device as disk image, function lseek
> >>> > > could not get the size of this kind of device.
> >>> > > So try to use SCSI command Read Capacity(10) when lseek failed
> >>> > > to get the size of SCSI generic device.
> >> >
> >> > Eww, this is ugly as hell.  Why would you even need the size for a
> >> > raw passthrough device?
> >
> > If we want to enable snapshot for SCSI generic device as disk image,
> > the size of SCSI generic device is needed. Function lseek could not
> > get this, SCSI command can finish the job.
> > Only when lseek failed would Read Capacity command be sent.
> 
> You need to use scsi-block instead of scsi-generic.  However, I don't see
how
> this can work.  After the snapshot, the image will be qcow2, not raw, and
thus
> it will not support bdrv_aio_ioctl.  Hence any SCSI command (for
scsi-generic)
> or any non-data SCSI command (for scsi-block) will fail.
> 
That's the issue what I also encountered.
Do you mean that it is impossible for us to enable snapshot for
scsi-generic?
Or patched for qcow2 would solve this?
> Can you give an example of what exactly you are trying to do?
> 
I could enable snapshot for scsi-block
device, but failed when using scsi-generic with parameter 'snapshot =on'. 
The first issue is failing to get the size of SG device. So I tried to fix
it.

Command line:
-drive if=none,id=hd,file=/dev/sg2,snapshot=on \
-device virtio-scsi-pci,id=scsi --enable-kvm \
-device scsi-generic,drive=hd,id=vd1
> Paolo

Regards
Paolo Bonzini - Oct. 10, 2012, 9:01 a.m.
Il 10/10/2012 10:33, Chen HanXiao ha scritto:
>> You need to use scsi-block instead of scsi-generic.  However, I don't see
>> how this can work.  After the snapshot, the image will be qcow2, not raw, and
>> thus it will not support bdrv_aio_ioctl.  Hence any SCSI command (for
>> scsi-generic) or any non-data SCSI command (for scsi-block) will fail.
>
> That's the issue what I also encountered.
> Do you mean that it is impossible for us to enable snapshot for
> scsi-generic?

It's impossible, and rightly so: the user could anyway bypass the
snapshotting and write to the disk (WRITE SAME for example).

Not just snapshot=on, which fails because it is early enough that the
disk fails to initialize.  Live snapshots also ought to be blocked, it
is a bug that they are not.

Paolo

Patch

diff --git a/block/raw-posix.c b/block/raw-posix.c
index 28d439f..0be7db8 100644
--- a/block/raw-posix.c
+++ b/block/raw-posix.c
@@ -53,6 +53,7 @@ 
 #include <linux/cdrom.h>
 #include <linux/fd.h>
 #include <linux/fs.h>
+#include <scsi/sg.h>
 #endif
 #ifdef CONFIG_FIEMAP
 #include <linux/fiemap.h>
@@ -147,6 +148,7 @@  typedef struct BDRVRawReopenState {
 } BDRVRawReopenState;
 
 static int fd_open(BlockDriverState *bs);
+static int64_t raw_getlength_ioctl(int fd);
 static int64_t raw_getlength(BlockDriverState *bs);
 
 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
@@ -653,13 +655,53 @@  static int64_t raw_getlength(BlockDriverState *bs)
 {
     BDRVRawState *s = bs->opaque;
     int ret;
-
+    int64_t len;
+    
     ret = fd_open(bs);
     if (ret < 0) {
         return ret;
     }
 
-    return lseek(s->fd, 0, SEEK_END);
+    len = lseek(s->fd, 0, SEEK_END);
+    if ( len >= 0) {
+        return len;
+    } else {
+        len = raw_getlength_ioctl(s->fd);
+        return len;
+    }
+}
+
+/* Use SCSI Read Capacity(10) Command to get length */
+static int64_t raw_getlength_ioctl(int fd)
+{
+    unsigned char CDB[10] =
+        {0x25, 0, 0, 0, 0, 0, 0, 0, 0, 0};
+    unsigned char sense_buffer[32];
+    unsigned char resp_buffer[32];
+    uint32_t block_size;
+    uint64_t last_blk_addr;
+    struct sg_io_hdr io_hdr;
+    int64_t ret;
+ 
+    memset(&io_hdr, 0, sizeof(struct sg_io_hdr));
+    memset(sense_buffer, 0, sizeof(sense_buffer));
+    memset(sense_buffer, 0, sizeof(resp_buffer));
+    io_hdr.interface_id = 'S';
+    io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
+    io_hdr.cmd_len = sizeof(CDB);
+    io_hdr.cmdp = CDB;
+    io_hdr.sbp = sense_buffer;
+    io_hdr.dxferp = resp_buffer;
+    io_hdr.dxfer_len = sizeof(resp_buffer);
+    if((ret = ioctl(fd, SG_IO, &io_hdr)) < 0)
+        return ret;
+
+    last_blk_addr = ((resp_buffer[0] << 24) | (resp_buffer[1] << 16) |
+        (resp_buffer[2] << 8) | resp_buffer[3]);
+    block_size = ((resp_buffer[4] << 24) | (resp_buffer[5] << 16) |
+        (resp_buffer[6] << 8) | resp_buffer[7]);
+    ret = (int64_t)((last_blk_addr + 1) * block_size);
+    return ret;
 }
 #endif