Re: Review of Roland's >2GB ext2fs proposal

[Thomas Bushnell BSG]

FWIW, this "Roland's proposal" was also my original design.

So Neal adds wrong.  The indirect blocks are not always mapped in; you
might call them "metadata", but nothing here mandates they be handled
the same way.

And how could you, given the way they change?

The original idea was that there was a DINODE pager which was indeed
always mapped in.

But the SINDIR pager was a *per file* pager; so was the DINDIR pager.
These were mapped when the file was in use.

The TINDIR pager was one pager per filesystem, and always mapped in,
with one block per DINODE, but that could easily be changed to only
map when the file is in use.

Thomas