Don't use sleeping allocations for ufs dirhash blocks when holding directory vnode

[dgmorris_earthlink.net]

Summary

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=253992

System with a based-on-FreeBSD-12 kernel but with a timeout panic for the UFS root vnode lock paniced due to a ufsdirhash_build() call with M_WAITOK, UMA_ZONE_FIRSTTOUCH (the default) and a domain which was OOM. The ufsdirhash_build() code is invoked with the directory vnode exclusively locked (per ufs_lookup.c / the caller) and in all other places takes care to allocate M_NOWAIT to avoid such issues.

So it seems reasonable to make this one allocation also M_NOWAIT instead of M_WAITOK, especially given the FIRSTTOUCH default behavior where only the calling cpu's domain can satisfy the request.

Test Plan

Organizational stress testing, no unit test planned -- change is pretty straightforward.

[markj]

sys/ufs/ufs/ufs_dirhash.c

111

I would just drop the NOWAIT suffix, or get rid of the UMA wrappers entirely.

[dgmorris_earthlink.net]

Took Mark's suggestion on dropping _NOWAIT suffix on allocation macro.
Left the macro for symmetry with the FREE macros in the file.

[mckusick]

This seems like a reasonable solution to the problem. The allocation will fail in a few cases where it previously would have succeeded, but hopefully those will be rare. The effect of failing will simply be slower lookups rather than unexpected errors to applications.