nanobsd fdisk layout broken

Hi,

For some Soekris boxes I will use soon, I thought I'd give nanobsd a
try and encounterd the following problem.

My Kingston compact flash has 695 cyls, 15 heads and 48 secs/trk
according to dmesg when I insert it into my notebook.

When I want a 10MB data slice (DATASIZE=10240) the i386.diskimage
script is unable to populate the flash image, from _.md.tmp:

    + cat /tmp/nanobsd.WhzaHkUW
    g c695 h15 s48
    p 1 165 48 245056
    p 2 165 245104 245056
    p 3 165 490160 10240
    + fdisk -i -f /tmp/nanobsd.WhzaHkUW md0
    fdisk: invalid fdisk partition table found
    fdisk: WARNING: adjusting size of partition 1 from 245056 to 244752
	to end on a cylinder boundary
    fdisk: WARNING: adjusting start offset of partition 2
	from 245104 to 245136, to fall on a head boundary
    fdisk: WARNING: adjusting size of partition 2 from 245056 to 244464
	to end on a cylinder boundary
    fdisk: WARNING: adjusting start offset of partition 3
	from 490160 to 490176, to fall on a head boundary
    fdisk: WARNING: adjusting size of partition 3 from 10240 to 10224
	to end on a cylinder boundary
    ******* Working on device /dev/md0 *******
    + fdisk md0
    ******* Working on device /dev/md0 *******
    parameters extracted from in-core disklabel are:
    cylinders=695 heads=15 sectors/track=48 (720 blks/cyl)

and a bit later:

    + dd if=/dev/md0s1 of=/dev/md0s2 bs=64k
    dd: /dev/md0s2: short write on character device
    dd: /dev/md0s2: end of device
    1910+0 records in
    1909+1 records out
    125165568 bytes transferred in 32.334543 secs (3870955 bytes/sec)

Note how fdisk decides to make md0s2 slightly smaller than md0s1.

Fdisk wants things to be aligned at cylinder or track boundaries
(it's not clear to me when it wants one and when the other).

The i386 architecture uses the first sector of any disk to store
the MBR and reserves a whole track for this, so the first slice
normally starts at the second track of the first cylinder.

So with the formula used by i386.diskimage to calculate the size
of a boot slice (total_sectors - sectors_per_track - datasize) / 2
the rounding formulas in fdisk may cause the second slice to be
shorter than the first which breaks the nanobsd scripts.

Therefore I propose to alter the calculations done in i386.diskimage
such that:

  1. The data slice (/etc) physically comes first, starts at the
     second track of the first cylinder and ends at a cylinder
     boundary.
  2. The boot slices are physically located after the data slice
     and their size is an integral multiple of the cylinder size.

The attached patch for i386.diskimage implements this resulting in
two equally sized boot partitions starting and ending at cylinder
boundaries and a data partition starting at the second track and
ending at a cylinder boundary.  Because of all the roundings, the
actual size of the data slice will normally be larger that the
value of DATASIZE in make.conf.

Note that there is one more exercise left: change this calculation
or the newfs parameters such that newfs will use all slices entirely
instead of throwing away some sectors at the end of each slice.

Regards,

Paul Schenkeveld

------------------------------------------------------------------------

--- src/tools/tools/nanobsd/i386.diskimage.orig	Mon Sep  6 22:57:28 2004
+++ src/tools/tools/nanobsd/i386.diskimage	Sun Oct 10 18:43:46 2004
_at__at_ -56,12 +56,14 _at__at_
 MD=`mdconfig -a -t vnode -f ${TMPFILE0} -x ${SC} -y ${HD}`
 rm -f ${TMPFILE0}
 (
-sl=`expr "(" ${SECTS} - ${SC} - ${DATASIZE} ")" / 2`
 cyl=`expr ${SECTS} / ${SC} / ${HD}`
+sl_cyl=`expr "(" ${SECTS} - ${SC} - ${DATASIZE} ")" / 2 / ${HD} / ${SC}`
+sl=`expr $sl_cyl "*" ${HD} "*" ${SC}`
+ds=`expr ${SECTS} - ${SC} - ${sl} - ${sl}`
 echo g c${cyl} h${HD} s${SC}
-echo p 1 165 ${SC} $sl
-echo p 2 165 `expr ${SC} + $sl` $sl
-echo p 3 165 `expr ${SC} + $sl + $sl` ${DATASIZE}
+echo p 1 165 `expr ${SC} + $ds` $sl
+echo p 2 165 `expr ${SC} + $ds + $sl` $sl
+echo p 3 165 ${SC} $ds
 ) > ${TMPFILE1}
 cat ${TMPFILE1}
 fdisk -i -f ${TMPFILE1} ${MD}