@rew see D34952 for an alternate fix.

LGTM, but see my comment in the PR. I doubt unmapped I/O is actually being used.

OTOH, on my system the only collisions are caused by the ZFS sysctls marked "LEGACY", and vm.uma.tcp_log_bucket.size/vm.uma.tcp_log.bucket_size. So a less disruptive option would simply be to ignore the LEGACY sysctls, and rename vm.uma.tcp_log.bucket_size.

In D32886#791065, @rew wrote:

If the name of the metric being exported changes...how will that not break existing setups?

The code change looks sensible to me. But what happens if there are two sysctls named foo..bar and foo_bar? Won't this still result in a collision? Or are such cases considered too rare to worry about? Also, the man page ought to reflect this change.

• 32-bit compatible printf format specifiers

LGTM except for the commit message. The problems with ustosbt are in the range of 2s - 4295s. For nstosbt it's 2s - 4s. And for mstosbt it's 2s - 4294967s.

Release root inode even if the server is already detached

LGTM

@0mp beat me to it in https://reviews.freebsd.org/D34439 .

Hehe, yeah. C++ that's not actually C is surprisingly uncommon. BTW, did you see the related PR at https://github.com/pjd/pjdfstest/pull/58 ?

Yep, that's exactly the idea.

@rmacklem will you be able to review this PR? I'd like to get it into FreeBSD 13.1.

Thanks @lwhsu ! I was actually working on this myself at this very moment. But I see that you're ahead of me.

• Use ses_status_common_get_element_status_code

• nvd and nvme devices are not part of CAM

In D31809#767688, @ken wrote:

I'm assuming this will need to run on a system with a SES device.

That's fine. Someone might write one for CTL someday, but that would be too much work for just a test rig.

• Properly handle IO_SYNC during VOP_DEALLOCATE for fuse

• Prohibit exporting file systems during VOP_VPTOFH, not VOP_MOUNT

What's the difference between CAM_CMD_TIMEOUT and CAM_SEL_TIMEOUT ?

LGTM

LGTM, except for the style nit on line 256 of ggatec.c .