svn commit: r195817 - head/usr.sbin/sysinstall

Alexander Motin mav at FreeBSD.org
Wed Jul 22 11:34:06 UTC 2009


Erik Trulsson wrote:
> On Wed, Jul 22, 2009 at 01:28:14PM +0300, Alexander Motin wrote:
>> Juli Mallett wrote:
>>> On Wed, Jul 22, 2009 at 01:42, Alexander Motin<mav at freebsd.org> wrote:
>>>> Colin Percival wrote:
>>>>>   Remove the "dedicated disk mode" partitioning option from sysinstall, in
>>>>>   both the disk partitioning screen (the 'F' key) and via install.cfg (the
>>>>>   VAR_DEDICATED_DISK option).  This functionality is currently broken in 8.x
>>>>>   due to libdisk and geom generating different partition names; this commit
>>>>>   merely acts to help steer users away from the breakage.
>>>> Is there any other way to not align FS block to the ugly legacy 63
>>>> sectors per track boundary with sysinstall now? I think RAIDs won't be
>>>> happy. May be it would be better to fix it?
>>> If you're interested in fixing this issue, you might want to look at
>>> the need for compatibility names so that existing DD installs aren't
>>> broken, and so DD installs work as-is without correcting libdisk's
>>> expectations about slice/partition names for DD disks, which is pretty
>>> invasive, too.  Not breaking new installs by not letting users install
>>> broken systems is the absolute bare minimum approach, and given the
>>> late date and the lack of movement on the kernel side, I've been
>>> advocating for it for a while.
>>>
>>> See this message and others in the thread for some background:
>>>
>>> http://lists.freebsd.org/pipermail/freebsd-geom/2009-June/003567.html
>> Sorry, ENOTIME. I am not advocating DD mode, it is really a hack. Offset
>> 0 is just an easiest choice to align FS. Instead, I would really like
>> sysinstall to honor real disk geometry instead of fake one.
> 
> "real disk geometry" ?  How would sysinstall find that?

As I have said, GEOM is able to provide such info to user-level when
provider reports it.

> Keep in mind that any disk geometry reported by disks is completely fake
> these days and is just an attempt to fit the total number of blocks into
> the limitations of the PC BIOS.
> 
> In short the whole notion of 'disk geometry' is mostly obsolete these days
> and ought to be avoided as far as possible.

You are right about regular HDDs, but I am speaking mostly about RAID
geometry: stripe (native block) size and it's offset. geom_stripe
provides that kind of information, and I believe most of hardware RAIDs
could also do it.

Also it could be used for different SSD/FLASH storages, which have media
erase block of more then one sector. Proper FS alignment could reduce
number of media erases. mmcsd(4) driver reports SD/MMC card's erase
block size there.

-- 
Alexander Motin


More information about the svn-src-head mailing list