sysinst segfaults on Alpha ("file system full") - install help

[commodorejohn]

Trying to get a NetBSD 9.0 install going on an Alpha 3000 300X workstation since Tru64 is being a pain in the butt. The install environment boots and runs, and I can do a bunch of stuff from the shell with no issues, but the installer itself craps out when partitioning the disk. I can partition the same disk with disklabel and create filesystems with newfs totally without issue, so it's not an issue with the drive - unfortunately, this doesn't give it whatever magic blessing sysinst needs to recognize the appropriate mount points automatically, and when I try to use the sysinst partition utility to do anything (even just set the mount points on already partitioned and formatted filesystems,) it kills with the following message:

/: write failed, file system is full
[1]   Segmentation fault

Someone else described the same issue here for the SGI port sixteen-plus years ago :/ So I'm not exactly thrilled about the chances of a fix coming along, but it's worth asking, I suppose. One thing I did notice was that /dev/mda0 (the in-memory filesystem for the installer) ends up very close to completely full, and part of that is a sysinst.core dump, but I'm not clear on whether sysinst is dumping core because the filesystem is full, or vice versa. It doesn't seem like memory constraints should be an issue, though - the box has 160 MB of RAM, way more than the listed minimum.

I think my better shot here is: can I just avoid the sysinst partitioning utility entirely? I can do everything needed with the disk from the shell, except apparently chant the proper blessing upon the resulting partitions. How does sysinst determine which partitions in the disklabel should map to which mount points in the target install? Where is that magic documented or stored?

[J65nko]

I cannot give you any assistance on your "sysinst" issue but .....

The mapping of BSD partitions/labels to file systems is done in the file "/etc/fstab".
The following is from my OpenBSD install:

hp_server$ cat /etc/fstab
e9298df9f130b613.b none swap sw
e9298df9f130b613.a / ffs rw 1 1
e9298df9f130b613.e /home ffs rw,nodev,nosuid 1 2
e9298df9f130b613.d /usr ffs rw,wxallowed,nodev 1 2

By convention partition/label "a" is used for the root file system "/", where the system boots from. Partition/label "b" is used by convention for swap. On NetBSD the "c" label covers the NetBSD slice (MBR partition), while on OpenBSD it covers the whole disk. For NetBSD label "d" covers the whole disk. See http://netbsd.org/docs/guide/en/chap...tall-partition

[Sehnsucht94]

Originally Posted by commodorejohn

when I try to use the sysinst partition utility to do anything (even just set the mount points on already partitioned and formatted filesystems,) it kills with the following message:

[code]/: write failed, file system is full

On i386 once I experienced the same problem when the MBR had been overwritten with a BSD label (resulting in a 'dangerously dedicated disk'), since...

unfortunately, this doesn't give it whatever magic blessing sysinst needs to recognize the appropriate mount points automatically

How does sysinst determine which partitions in the disklabel should map to which mount points in the target install? Where is that magic documented or stored?

sysinst relies on the MBR in order to get the correct BIOS geometry, or other partition table to get the disk layout. What about dropping to shell at boot and running disklabel -w /wd0 before launching sysinst?