Drain a keg of a UMA_ZONE_NOFREE zone when it is destroyed.

[glebius]

Summary

Looks like my (Jonathan's) change to GELI created a precedent
when an UMA_ZONE_NOFREE is being destroyed. Without this change
we are leaking zero items, but leaking pages:

Freed UMA keg (GELI buffers) was not empty (0 items). Lost 528 pages of memory.

Fixes: 2dbc9a388ee
PR: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=258787

[markj]

From my reading, GELI sets _NOFREE to ensure that the reserve is not drained during a memory shortage. But UMA should not have done this in the first place, and it was fixed in commit f09cbea31a3ffc03361687cb39f264b6c8775212 . So does GELI still need to specify _NOFREE? If not, then my preferred solution is to drop the _NOFREE flag there. We should try hard to avoid adding new _NOFREE zones in general.

[glebius]

In D32242#727897, @markj wrote:

From my reading, GELI sets _NOFREE to ensure that the reserve is not drained during a memory shortage. But UMA should not have done this in the first place, and it was fixed in commit f09cbea31a3ffc03361687cb39f264b6c8775212 . So does GELI still need to specify _NOFREE? If not, then my preferred solution is to drop the _NOFREE flag there. We should try hard to avoid adding new _NOFREE zones in general.

I believe you are 100% right. This flag we were carrying in our tree from the times that predate your fix.

[glebius]

The problem will be addressed differently.