Stefan, thanks a lot of taking care of my fallout!

• Revert 97ec6eba653a07. There shouldn't be a dependency of 'tmp' on
• Add 'tmp' to the list of FILESYSTEMS dependencies. Some scripts that

This may have a slight improvement when compiling from a non-branch point - a detached state. Now

Some extra cruft slept in previous update.

Separate decompression loops into their own subroutines. Reword
logical statement.

In D27245#608500, @trasz wrote:

Shouldn't this also include a change to default savecore_flags in /etc/defaults/rc.conf?

Add ZSTD support and address comments.

Definitely my miss. Thanks for fixing!

This looks good to me. Thanks a lot! Please commit if it passes tests.

spl calls is not something that gives any maintenance burden. They have been defined to nop for years. Of course it makes sense to cleanup those that aren't mentioned anywhere just like rS364918 did. However, I don't think just removing these nop defines is a holy grail we should aim for.

In D26027#577448, @mjg wrote:

I remember UMA warning about relevant allocations without even attempting to sleep, but maybe my memory is off here.

Add same check to UMA.

In D26027#577390, @kib wrote:

Why it is not WITNESS_WARN() ? Perhaps witness_warn should take td_no_sleeping as arbitrary non-sleepable lock as well.

In D22691#569217, @kevans wrote:

In D22691#569148, @melifaro wrote:

The resulting if_busy() / if_unbusy() / if_busy_wait() semantics looks very similar to sx locking. I'm wondering why not simply use sx locks as @glebius suggested?

I think, through the course of this, I became convinced that sx usage complicates some possible future busy'ers that would no longer be able to hold a mutex across calls here.