svn commit: r221972 - head/sys/geom/part

[Warner Losh]

On May 22, 2011, at 11:29 PM, Andrey V. Elsukov wrote:

> On 23.05.2011 6:29, Warner Losh wrote:
>> Looking at one of my flash drives that shows the problem:
>> 
>> da0: 15423MB (31588351 512 byte sectors: 255H 63S/T 1966C) GEOM_PART: partition 1 has end offset
>> beyond last LBA: 31588350 > 31588325
>> 
>> So why does gpart think the last LBA is 25 less than it really is.  Well, let's do some sanity
>> checks first.  fdisk -s da0 tells us:
> 
>> I'm pretty sure this problem is due to a bug in g_part_mbr.c:
>> 
>> basetable->gpt_last = msize - (msize % basetable->gpt_sectors) - 1;
>> 
>> This is wrong, or at least it is a widely disregarded part of what makes up an MBR.  When I
>> correct the size, the geom code is fine.  There's no requirement in MBR that a partition end of
>> any particular boundary, although sometimes you'll find mistaken documentation that suggests this
>> is the case.  Reading between the lines at http://www.boot-us.com/gloss03.htm suggests that this
>> restriction was only for disks < 8GB in size (from the fact it said that all the partitions can
>> be described with the CHS fields).  This is one of the things I learned when I tried to make
>> fdisk enforce that: this isn't a requirement of MBR as it is implemented in the wild today.
> 
> There is another opinion:
> http://www.boot-us.com/gloss02.htm
> 
> "There is the convention that partitions should always start and end on *cylinder boundaries*."

The current code doesn't do that.  The current code just ends on a track boundary...

> but
> 
> "Linux and Windows NT/2000/XP/2003/Vista/2008/7 seem to have no problems when partitions do not
> start or end on cylinder boundaries. However other operating systems (e.g. DOS) expect that this
> convention is obeyed."
> 
> What way will we choose?

We choose the way that works and not bother with alignment.  It is an artifact of the < 8GB disks and hasn't really been necessary for about 15 years now.

> I prefer to: leave the choice to the user how partition should be aligned, and remove automatic
> alignment to track boundaries from MBR (and probably from all other schemes).

I think we agree.  Allow, but do not force, alignment.

Warner

> -- 
> WBR, Andrey V. Elsukov
> 
> -- 
> WBR, Andrey V. Elsukov
>