Check for device type (block/character) in grub-setup?

Check for device type (block/character) in grub-setup?

[Grégoire Sutre]

Hi,

I'm wondering why there is no check for device type (block/character) in 
grub-setup.c whereas the function probe() in grub-probe.c exits with 
error if the device is not of the expected type. Shouldn't there be a 
similar check in grub-setup?

Currently, on NetBSD (with in-progress patches), grub-setup fails with 
an `out of disk error' on a block device as the detected disk size is 0, 
whereas grub-probe exits gracefully with an error message telling that 
the input device is not a character device.  I guess that a similar 
behavior could be observed on FreeBSD, but I can't actually test this.

Thanks,

Grégoire

Re: Check for device type (block/character) in grub-setup?

[Robert Millan]

On Sun, Jan 03, 2010 at 03:54:49AM +0100, Grégoire Sutre wrote:
> Hi,
>
> I'm wondering why there is no check for device type (block/character) in  
> grub-setup.c whereas the function probe() in grub-probe.c exits with  
> error if the device is not of the expected type. Shouldn't there be a  
> similar check in grub-setup?

Yes.

> I guess that a similar  
> behavior could be observed on FreeBSD, but I can't actually test this.

Please make the code generic if possible (i.e. accept both character and
block devices).

-- 
Robert Millan

  "Be the change you want to see in the world" -- Gandhi

Re: Check for device type (block/character) in grub-setup?

[Grégoire Sutre]

Robert Millan wrote:
> On Sun, Jan 03, 2010 at 03:54:49AM +0100, Grégoire Sutre wrote:
>> Hi,
>>
>> I'm wondering why there is no check for device type (block/character) in  
>> grub-setup.c whereas the function probe() in grub-probe.c exits with  
>> error if the device is not of the expected type. Shouldn't there be a  
>> similar check in grub-setup?
> 
> Yes.
> 
>> I guess that a similar  
>> behavior could be observed on FreeBSD, but I can't actually test this.
> 
> Please make the code generic if possible (i.e. accept both character and
> block devices).

I don't understand what you mean here.  I assumed that it's better to 
use character devices when accessing disks in GRUB utils, isn't it so? 
Moreover, at least on NetBSD, when a block device is mounted, it cannot 
be opened (device busy), and this would be a problem with grub-probe.

However, from a user view-point, it could be nice on NetBSD to accept 
stripped device names (e.g. only `wd0d' or even `wd0') and automatically 
get the associated device file (with opendisk(3)). This is what system 
tools do, e.g. `disklabel wd0' or `fdisk wd0' actually opens 
`/dev/rwd0d' (on i386).

Grégoire

Re: Check for device type (block/character) in grub-setup?

[Robert Millan]

On Thu, Jan 07, 2010 at 06:31:02PM +0100, Grégoire Sutre wrote:
> Robert Millan wrote:
>> On Sun, Jan 03, 2010 at 03:54:49AM +0100, Grégoire Sutre wrote:
>>> Hi,
>>>
>>> I'm wondering why there is no check for device type (block/character) 
>>> in  grub-setup.c whereas the function probe() in grub-probe.c exits 
>>> with  error if the device is not of the expected type. Shouldn't 
>>> there be a  similar check in grub-setup?
>>
>> Yes.
>>
>>> I guess that a similar  behavior could be observed on FreeBSD, but I 
>>> can't actually test this.
>>
>> Please make the code generic if possible (i.e. accept both character and
>> block devices).
>
> I don't understand what you mean here.  I assumed that it's better to  
> use character devices when accessing disks in GRUB utils, isn't it so?  
> Moreover, at least on NetBSD, when a block device is mounted, it cannot  
> be opened (device busy), and this would be a problem with grub-probe.

I don't think it matters to us whether a device is character or block
based, other than to perform sanity checks in it.  Tools like grub-setup
are supposed to be happy with either.

(I find it a bit odd that disks are represented by NetBSD as character
devices, but, whatever...)

> However, from a user view-point, it could be nice on NetBSD to accept  
> stripped device names (e.g. only `wd0d' or even `wd0') and automatically  
> get the associated device file (with opendisk(3)). This is what system  
> tools do, e.g. `disklabel wd0' or `fdisk wd0' actually opens  
> `/dev/rwd0d' (on i386).

I find this a bit confusing TBH.  It seems unpredictable and would make
GRUB inconsistent across platforms.

BUT

I think it's fine if you patch this up in the NetBSD version of GRUB,
if you want it to be consistent with the other utilities.

-- 
Robert Millan

  "Be the change you want to see in the world" -- Gandhi