vfs_domount(): ensure that v_mountedhere and VIRF_MOUNTPOINT are set under the vnode lock

[kib]

Summary

Reported by: pho
Fixes: f7833196bd6ba9bfc060a41b353422b15d6aa95b

[jah]

Oops, I didn't realize that the mount path doesn't hold the covered vnode lock while updating the relevant fields.
It looks as though the unmount path does hold the lock across the relevant operations, so we should be safe from these fields being cleared out from under us, correct?
Would it make sense to add a comment here to that effect?

sys/kern/vfs_lookup.c

1263 ↗

(On Diff #112340)

v_mounthere -> v_mountedhere

[jah]

sys/kern/vfs_lookup.c

1267 ↗

(On Diff #112340)

Hmm, thinking about it a little more:
Given the potential cost of a load-load barrier on this path, would it make more sense to instead just replace the KASSERT below with a check that exits the loop if mp is NULL?

[kib]

sys/kern/vfs_lookup.c

1267 ↗

(On Diff #112340)

The cost of acquire fence is zero on amd64, and somewhat more on arm64.

But if this causes questions, I think we can just extend the lock region in vfs_domount().

[kib]

cache_purge() should not be called with the vnode locked, it does vdrop() for unrelated vnodes