Re: [PATCH] biosdisk / open_device() messing up offsets

[Pavel Roskin]

Quoting Bean <address@hidden>:

> On Fri, Jun 13, 2008 at 12:31 PM, Pavel Roskin <address@hidden> wrote:
> > Quoting Bean <address@hidden>:
> >
> > > After more thoughts, I think the current method still have problem.
> > > grub_disk_read reads up to 8192 bytes at a time, which can be larger
> > > than the block size.
> >
> > But we never ask grub_disk_read() to read across the block boundary in
> > grub_fshelp_read().  The code takes care of it.  I assume grub_disk_read()
> > would not write more than requested to the buffer.
>
> grub_disk_read would attempt to read up to GRUB_DISK_CACHE_SIZE and
> put the result in a cache, then copy the necessary part to output.

Then I don't see any problem with it.  Maybe you could give an example?

> > > Unless all or none of the sectors in the same
> > > block is mapped, we end up reading the wrong sector. So we need to do
> > > the mapping inside grub_disk_read.
> >
> > But mapping is done for blocks, not for sectors.  grub_fshelp_map_block()
> > clearly assumes that for every filesystem.
>
> block size is fs related, but most are less than 8192 bytes.

I don't understand why caching would have any effect.  It should be  
transparent for higher layers, like caching inside the hard drive.  If  
caching is not transparent, then it's broken.

--
Regards,
Pavel Roskin