How to do IO across hardsector boundries

How to do IO across hardsector boundries

[Thomas S. Iversen]

Hi There

As part of an assignment I am trying to port a piece of software from 
FreeBSD to linux. Essentially this software (crypto) makes a virtual 
blockdevice with "virtual" sectors on top. Under FreeBSD these virtual 
sectors are just read/written using a simple command:

buf=g_read(dev, offset, len)
error=g_write(dev,offset,buf,len)

In linux however I have only seen the BIO layer which operates on IO on 
hardsector boundaries.

So my question really is, how do I go about updating for instance the 
512 bytes located for at byte 64 to 64+511 on the actual media without 
getting in trouble regarding the data from offset 0-63 and 64+512->1023?

Regards Thomas

Re: How to do IO across hardsector boundries

[Alan Cox]

On Sul, 2004-08-01 at 01:22, Thomas S. Iversen wrote:
> So my question really is, how do I go about updating for instance the 
> 512 bytes located for at byte 64 to 64+511 on the actual media without 
> getting in trouble regarding the data from offset 0-63 and 64+512->1023?

If you want to use the BIO layer then you need to read the two blocks,
write to the bits you care about and then write them back. That is what
will always have to happen at a lower level for this anyway. In essence
you would be implementing the small g_read/g_write equivalent routines.

Re: How to do IO across hardsector boundries

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 534 bytes --]

On Sun, 2004-08-01 at 02:22, Thomas S. Iversen wrote:
> Hi There
> 
> As part of an assignment I am trying to port a piece of software from 
> FreeBSD to linux. Essentially this software (crypto) makes a virtual 
> blockdevice with "virtual" sectors on top. Under FreeBSD these virtual 
> sectors are just read/written using a simple command:

I can recommend that you look at the Device Mapper layer and consider
porting your code to that, it's designed to take care of some of the
hard work for you for usages like this.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: How to do IO across hardsector boundries

[Richard B. Johnson]

On Sun, 1 Aug 2004, Thomas S. Iversen wrote:

> Hi There
>
> As part of an assignment I am trying to port a piece of software from
> FreeBSD to linux. Essentially this software (crypto) makes a virtual
> blockdevice with "virtual" sectors on top. Under FreeBSD these virtual
> sectors are just read/written using a simple command:
>
> buf=g_read(dev, offset, len)
> error=g_write(dev,offset,buf,len)
>
> In linux however I have only seen the BIO layer which operates on IO on
> hardsector boundaries.
>
> So my question really is, how do I go about updating for instance the
> 512 bytes located for at byte 64 to 64+511 on the actual media without
> getting in trouble regarding the data from offset 0-63 and 64+512->1023?
>
> Regards Thomas

There are no hard-sector boundaries in Linux. The geometry of a drive
is really ignored. It's just used to determine the size of a disk
partition, etc. The stated Heads/cylinders/sectors is a throw-back
to the old int 0x13, DOS days where stuff had to fit into registers
for access.

If your driver expects to write physical sectors, it's broken.
Even physical sectors (on the media) don't really exist anymore.
They are now "logical" sectors and they are never really accessed
as a single unit (except in floppies). They are "extracted" from a
sector-buffer that could contain nearly a whole track.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.26 on an i686 machine (5570.56 BogoMips).
            Note 96.31% of all statistics are fiction.