The code works but has a nasty corner case -- what if the process has only 2 fds opened, say 0 and 8364742 (or some other high number). Then you copy out out rather huge bitmap and force it to scan it.

I think this is a pretty weird choice.

this description is rather misleading, suggesting states get fetched one at a time. instead they are already all fetched, except there is a per-state callback

sorry for the delay

by "here" I meant the vfs layer ;)

sorry mate, I'm no longer involved here, you have to prod someone else

for example there was panic under load when running poudriere, i don't know if that is fixed

For this to be an option this review has to provide a list of bugs reported against block cloning and commits which fix them, as is I'm not even sure it was all sorted out.

Have you seen debug.vfscache (hidden behind DEBUG_CACHE) and the vfs.cache_fast_lookup sysctls?