Physical hard disk drive for win32

Qemu-devel In review Kazu 8 participants 16 messages Sep 24, 2006 – Oct 25, 2007

Share & tools

Kazu

author

2006-09-24 04:01 UTC #0 lore raw

Hi,

An attached patch fixes to use raw access to physical hard disk drive for win32.
It can be used as -hdb \\.\PhysicalDrive0 or -hdb //./PhysicalDrive0.

For example, if the guest OS is Linux, the second partition can be accessed by /dev/hdb2.

Regards,
Kazu

Index: block-raw.c
===================================================================
RCS file: /sources/qemu/qemu/block-raw.c,v
retrieving revision 1.10

block-raw.c +20 -0

diff -u -r1.10 block-raw.c
@@ -833,6 +833,7 @@
 
 #define FTYPE_FILE 0
 #define FTYPE_CD     1
+#define FTYPE_HARDDISK 2
 
 typedef struct BDRVRawState {
     HANDLE hfile;

@@ -1090,6 +1091,9 @@
     BDRVRawState *s = bs->opaque;
     LARGE_INTEGER l;
     ULARGE_INTEGER available, total, total_free; 
+    DISK_GEOMETRY dg;
+    DWORD count;
+    BOOL status;
 
     switch(s->type) {
     case FTYPE_FILE:

@@ -1102,6 +1106,14 @@
             return -EIO;
         l.QuadPart = total.QuadPart;
         break;
+    case FTYPE_HARDDISK:
+        status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY,
+                                 NULL, 0, &dg, sizeof(dg), &count, NULL);
+        if (status != FALSE) {
+            l.QuadPart = dg.Cylinders.QuadPart * dg.TracksPerCylinder
+                * dg.SectorsPerTrack * dg.BytesPerSector;
+        }
+        break;
     default:
         return -EIO;
     }

@@ -1201,9 +1213,17 @@
     BDRVRawState *s = bs->opaque;
     UINT type;
     const char *p;
+    char *q;
+    int i;
+    char physical_drive[14];
 
     if (strstart(filename, "\\\\.\\", &p) ||
         strstart(filename, "//./", &p)) {
+        for (i = 0, q = p; i < 13; i++, q++) {
+            physical_drive[i] = tolower(*q);
+        }
+        if (strstart(physical_drive, "physicaldrive", NULL))
+            return FTYPE_HARDDISK;
         snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", p[0]);
         type = GetDriveType(s->drive_path);
         if (type == DRIVE_CDROM)

Alexey Eremenko

2006-09-24 09:14 UTC #1 lore raw

Is it hard to access "raw" hardware under Windows ? like Hard Disk...?

Under Linux it is easy.

Kazu

author

2006-09-24 12:22 UTC #2 lore raw

Quoted 3 lines · Alexey Eremenko

Is it hard to access "raw" hardware under Windows ? like Hard Disk...?
Under Linux it is easy.

CVS build cannot access host's physical hard disk drive by -hdb //./PhysicalDrive0 after AIO was introduced.
So I made a patch to fix it.

Regards,
Kazu

Stefan Weil

2007-10-19 20:12 UTC #3 lore raw

PATCH 1/1 Physical hard disk drive for win32

Hi,

raw harddisk access for Windows (//./PhysicalDrive0) gives wrong disk sizes. You can check this by booting a harddisk with the GRUB bootloader installed and using GRUB's geometry command.

The current QEMU code uses IOCTL_DISK_GET_DRIVE_GEOMETRY which is marked as obsolete by MS.

The appended patch gives correct results. Please apply it to CVS.

Stefan

Description of block-raw.patch:

* replace IOCTL_DISK_GET_DRIVE_GEOMETRY by IOCTL_DISK_GET_DRIVE_GEOMETRY_EX

Index: block-raw.c
===================================================================
RCS file: /sources/qemu/qemu/block-raw.c,v
retrieving revision 1.24

block-raw.c +4 -5

diff -u -b -B -r1.24 block-raw.c
@@ -1174,7 +1174,7 @@
     BDRVRawState *s = bs->opaque;
     LARGE_INTEGER l;
     ULARGE_INTEGER available, total, total_free;
-    DISK_GEOMETRY dg;
+    DISK_GEOMETRY_EX dg;
     DWORD count;
     BOOL status;

@@ -1190,11 +1190,10 @@
         l.QuadPart = total.QuadPart;
         break;
     case FTYPE_HARDDISK:
-        status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY,
+        status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY_EX,
                                  NULL, 0, &dg, sizeof(dg), &count, NULL);
-        if (status != FALSE) {
-            l.QuadPart = dg.Cylinders.QuadPart * dg.TracksPerCylinder
-                * dg.SectorsPerTrack * dg.BytesPerSector;
+        if (status != 0) {
+            l = dg.DiskSize;
         }
         break;
     default:

Ivan Kalvachev

2007-10-22 11:33 UTC #4 lore raw

2007/10/19, Stefan Weil <[email protected]>:

Quoted 9 lines

Hi,
raw harddisk access for Windows (//./PhysicalDrive0) gives wrong disk sizes. You can check this by booting a harddisk with the GRUB bootloader installed and using GRUB's geometry command.
The current QEMU code uses IOCTL_DISK_GET_DRIVE_GEOMETRY which is marked as obsolete by MS.

Quoted 7 lines

The appended patch gives correct results. Please apply it to CVS.
Stefan
Description of block-raw.patch:
* replace IOCTL_DISK_GET_DRIVE_GEOMETRY by IOCTL_DISK_GET_DRIVE_GEOMETRY_EX

According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?

Stefan Weil

2007-10-22 18:25 UTC #5 lore raw

Quoted 4 lines · Ivan Kalvachev

According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?

Good question. The old IOCTL_DISK_GET_DRIVE_GEOMETRY was also supported for W2K. Will QEMU support W2K as a host longer than MS does? It would make things easier if we dropped support for W2K hosts.

If we need host support for W2K, I'll check another method to get the disk size (reading size of partition 0) and send a new patch. Please send your feedback in this case.

I hope we don't need W95 or W98 host support!?

Stefan

Johannes Schindelin

2007-10-22 19:08 UTC #6 lore raw

Hi,

Quoted 9 lines · Stefan Weil

Quoted 4 lines · Ivan Kalvachev
According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?
Good question. The old IOCTL_DISK_GET_DRIVE_GEOMETRY was also supported for W2K. Will QEMU support W2K as a host longer than MS does? It would make things easier if we dropped support for W2K hosts.

Do we really want to be as evil as Microsoft and force all users to upgrade?

It should not be _that_ hard to stay at least a _little_ backwards compatible.

Ciao,
Dscho

Shaddy Baddah

2007-10-23 00:31 UTC #7 lore raw

In reply to Stefan Weil

Hi,

Quoted 3 lines · Stefan Weil

longer than MS does? It would make things easier if we dropped support for W2K hosts.

Surely not?!? I've heard of people being off hand about supporting Win95/98 (and I would agree with that) et al..., but surely W2K support should be sacrosanct ?!?

Regards,
Shaddy

Jamie Lokier

2007-10-23 03:34 UTC #8 lore raw

Quoted 6 lines · Shaddy Baddah

Quoted 3 lines
longer than MS does? It would make things easier if we dropped support for W2K hosts.
Surely not?!? I've heard of people being off hand about supporting Win95/98 (and I would agree with that) et al..., but surely W2K support should be sacrosanct ?!?

I would hope so, as I get the impression W2K is the last Windows version which can be run reliably in a VM, due to the "legitimacy" checking done by XP and later. :-)

-- Jamie

Ivan Kalvachev

2007-10-24 09:46 UTC #9 lore raw

In reply to Stefan Weil

2007/10/22, Stefan Weil <[email protected]>:

Quoted 10 lines

Quoted 4 lines · Ivan Kalvachev
According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?
Good question. The old IOCTL_DISK_GET_DRIVE_GEOMETRY was also supported for W2K. Will QEMU support W2K as a host longer than MS does? It would make things easier if we dropped support for W2K hosts.

Quoted 5 lines

If we need host support for W2K, I'll check another method to get the disk size (reading size of partition 0) and send a new patch. Please send your feedback in this case.
I hope we don't need W95 or W98 host support!?

Why not. :P

Anyway, calling the old ioctl code if the new one fails, should be compatible enough. (and the new structure contains the old one, so no need of having 2 variables).

It would be even better if somebody finds out what the error for not supported function is (I don't feel like checking 16000 error codes), but it is not critical.

Johannes Schindelin

2007-10-24 10:04 UTC #10 lore raw

Hi,

Quoted 15 lines · Ivan Kalvachev

2007/10/22, Stefan Weil <[email protected]>:
Quoted 14 lines
Quoted 4 lines · Ivan Kalvachev
According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?
Good question. The old IOCTL_DISK_GET_DRIVE_GEOMETRY was also supported for W2K. Will QEMU support W2K as a host longer than MS does? It would make things easier if we dropped support for W2K hosts.
If we need host support for W2K, I'll check another method to get the disk size (reading size of partition 0) and send a new patch. Please send your feedback in this case.

Quoted 7 lines

Quoted 1 line
I hope we don't need W95 or W98 host support!?
Why not. :P
Anyway, calling the old ioctl code if the new one fails, should be compatible enough. (and the new structure contains the old one, so no need of having 2 variables).

It is not just that. You'll have to guard the new ioctl within #ifdef's so that it still compiles on MinGW, even if that version does not define the constant.

Staying backwards compatible in this case is so easy, it is not even funny. (And certainly no reason to groan over having to support it.)

Ciao,
Dscho

Thiemo Seufer

2007-10-24 12:34 UTC #11 lore raw

Quoted 27 lines · Johannes Schindelin

Hi,
Quoted 15 lines · Ivan Kalvachev
2007/10/22, Stefan Weil <[email protected]>:
Quoted 14 lines
Quoted 4 lines · Ivan Kalvachev
According to http://msdn2.microsoft.com/en-us/library/aa365171.aspx
this function requires Win XP or Vista.
Is qemu supported only on these?
Good question. The old IOCTL_DISK_GET_DRIVE_GEOMETRY was also supported for W2K. Will QEMU support W2K as a host longer than MS does? It would make things easier if we dropped support for W2K hosts.
If we need host support for W2K, I'll check another method to get the disk size (reading size of partition 0) and send a new patch. Please send your feedback in this case.
Quoted 7 lines
Quoted 1 line
I hope we don't need W95 or W98 host support!?
Why not. :P
Anyway, calling the old ioctl code if the new one fails, should be compatible enough. (and the new structure contains the old one, so no need of having 2 variables).

Quoted 6 lines

It is not just that. You'll have to guard the new ioctl within #ifdef's so that it still compiles on MinGW, even if that version does not define the constant.
Staying backwards compatible in this case is so easy, it is not even funny. (And certainly no reason to groan over having to support it.)

A undirected remark to the general public: If somebody comes up with a patch for better backward compatibility I'll certainly consider it.

Thiemo

Quoted 1 line

Johannes Schindelin

2007-10-24 17:32 UTC #12 lore raw

Hi,

Quoted 7 lines · Thiemo Seufer

Quoted 2 lines · Johannes Schindelin
Staying backwards compatible in this case is so easy, it is not even funny. (And certainly no reason to groan over having to support it.)
A undirected remark to the general public: If somebody comes up with a patch for better backward compatibility I'll certainly consider it.

I think I fixed it. Since I am not your regular Windows user (I only use Windows when I have to), I cannot tell if it works, though.

It is here: http://repo.or.cz/w/qemu/dscho.git?a=shortlog;h=diskgeometry

Any testing (especially the physical drive on Windows < XP) would be very much appreciated.

There is also a compilation bugfix with the current CVS on Windows: the tap down script patch left a for() loop without a body.

A question: on some other lists, the preferred way to submit patches is to inline them into the mail, but here I saw many attachments and/or weblinks to patches. The easiest way for me would be like I just did: provide a link to repo.or.cz.

What is the preferred way on this list?

Ciao,
Dscho

Stefan Weil

2007-10-24 18:41 UTC #13 lore raw

Quoted 11 lines · Johannes Schindelin

I think I fixed it. Since I am not your regular Windows user (I only use
Windows when I have to), I cannot tell if it works, though.
It is here: http://repo.or.cz/w/qemu/dscho.git?a=shortlog;h=diskgeometry
Any testing (especially the physical drive on Windows < XP) would be very much appreciated.
...
Ciao,
Dscho

With your patch, W2K will still see a wrong disk size. This is caused by the fact that the size of a modern harddisk has nothing to do with tracks per cylinder or sectors per track
- it is not simply the product of four values.

You can use a disk to some degree even when some part at the end is inaccessible, but it is also possible to destroy it (for example if a partition tool "corrects" the disk size in the boot sector) or get other kinds of errors.

I prefer a patch which gives correct values for all Windows versions and will send one as soon as possible.

Even without a patch, the current code in CVS HEAD still allows compilation using MinGW on any Windows. QEMU will also run on the same Windows variants as before. The only thing no longer working on W2K and older variants is physical hard disk access. I think this is better than working with a wrong disk size!

Regards,
Stefan

Johannes Schindelin

2007-10-25 00:27 UTC #14 lore raw

Hi,

Quoted 11 lines · Stefan Weil

Quoted 8 lines · Johannes Schindelin
I think I fixed it. Since I am not your regular Windows user (I only use Windows when I have to), I cannot tell if it works, though.
It is here:
http://repo.or.cz/w/qemu/dscho.git?a=shortlog;h=diskgeometry
Any testing (especially the physical drive on Windows < XP) would be very much appreciated. ... Ciao, Dscho
With your patch, W2K will still see a wrong disk size.

Funny. I thought that since kazu did the original patch, and kazu is a Windows user, that it worked at _some_ stage.

Also I have to note that IOCTL_DISK_GET_DRIVE_GEOMETRY is a Windows 2000 feature, so unfortunately there is really no backwards compatibility to Windows 95/98/ME.

Quoted 3 lines

This is caused by the fact that the size of a modern harddisk has nothing to do with tracks per cylinder or sectors per track - it is not simply the product of four values.

Since it obviously worked in one setting, there might be a chance that it is fixable. Could you be a little bit more specific in what is going wrong? I.e. what is the real size, and what are the four values?

Quoted 2 lines

Even without a patch, the current code in CVS HEAD still allows compilation using MinGW on any Windows.

Yes, but a setup that was evidently working before was broken.

Ciao,
Dscho

Thiemo Seufer

2007-10-25 10:05 UTC #15 lore raw

In reply to Johannes Schindelin

Quoted 23 lines · Johannes Schindelin

Hi,
Quoted 7 lines · Thiemo Seufer
Quoted 2 lines · Johannes Schindelin
Staying backwards compatible in this case is so easy, it is not even funny. (And certainly no reason to groan over having to support it.)
A undirected remark to the general public: If somebody comes up with a patch for better backward compatibility I'll certainly consider it.
I think I fixed it. Since I am not your regular Windows user (I only use Windows when I have to), I cannot tell if it works, though.
It is here: http://repo.or.cz/w/qemu/dscho.git?a=shortlog;h=diskgeometry
Any testing (especially the physical drive on Windows < XP) would be very much appreciated.
There is also a compilation bugfix with the current CVS on Windows: the tap down script patch left a for() loop without a body.

Quoted 6 lines

A question: on some other lists, the preferred way to submit patches is to inline them into the mail, but here I saw many attachments and/or weblinks to patches. The easiest way for me would be like I just did: provide a link to repo.or.cz.
What is the preferred way on this list?

I prefer inlined or attached patches, so they get archived and don't develop a 404 problem over time.

Thiemo