On 29-Mar-99 Karel van Houten wrote:
> Harald Koerfgen wrote:
>> AFAIK ext2, unlike ufs and ffs, doesn't reserve space for a bootloader
>> except block 0 of a partition. I hope I am wrong here, but I haven't
>> anything in that direction.
>> That means that a two stage boot process, like Ultrix and NetBSD do it,
>> a no go if you want to have an ext2 filesystem as the first partition
>> a disk.
>> Maybe I am making the wrong assumptions here and I'd appreciate if
>> with more knowledge of ext2 can correct me.
> I don't know more about ext2fs, but I think an approach like silo for
> sparclinux should be possible.
> My ideas are as follows:
> - Use MSDOS partition table, that only uses 66 bytes at the end of the
> first sector.
> - Port the second stage loader of silo to DECStation linux.
> - Use (a modified version of) Harald's bootprep to install a map table
> to load the second stage loader.
> - I've already modified bootprep.c in such a way that it leaves my
> partition table in place, and can install 52 maps (instead of 61).
That's *exactly* what I had in mind, except using sun partition tables.
Well, a part of, just enough that the kernel will recognise them without
modification. fdisk has to be modified anyway and my proposal limits the
maps to 51 but allows eight partitions instead of four. We simply have to
decide what we want.
> As an alternative, I'm trying to compile the netbsd loaders under linux.
If I understand things correctly, I haven't looked into the source yet,
the second stage loader may fit our needs, but I don't know if it
understands ext2. silo's second stage *might* be less work.