[REVIEW] 16704 protolist corrupts inode numbers larger than 2^31

[REVIEW] 16704 protolist corrupts inode numbers larger than 2^31

[Bill Sommerfeld]

Issue: https://www.illumos.org/issues/16704
CR: https://code.illumos.org/c/illumos-gate/+/3634
Diff: 
https://code.illumos.org/~diff/b9ac120bc9e759ad1b83db83de3f6814432532cd

I've been experimenting with doing illumos-gate builds in /tmp, which
exposed a bug in the protolist tool used during nightly.

The output of protolist includes a column with the inode number of
each file in the proto area.  The ino_t type is an unsigned long, but
but protolist passes the value through a variable of type "int" and
then formats it for output as a signed integer.

The corresponding code in protocmp which reads protolist output then
refuses to accept a negative number in the inode column, and spits out
a "bad i-node" error for each one.

This has apparently gone unnoticed because most filesystems use the
inode number to index a compact table of some sort.  But tmpfs does
something different - its inode numbers are computed as a function of
the kernel virtual address of a per-file datastructure and on
large-memory machines tend to be scattered broadly between 0 and 2^32-1, 
so an appreciable fraction of them turn into negative values when cast 
to int.

Thanks in advance for your review.

					- Bill