nvme: nvd/nda call disk_resize() when namespace changed async event happened

[wanpengqian_gmail.com]

Summary

Take D32963 and make nvd/nda resize after namespace changed async event trigger.

create a new thread and this thread will update namespace data when needed.

nvd can now call resize_disk() and update the disk size properly.

for nda, after NVME_PROBE_IDENTIFY_NS is finish, fire a AC_GETDEV_CHANGED event,
and it will call resize_disk() to adjust disk size.

Test Plan

When bhyve vm resize disk, verify diskinfo output inside guest, both nvd and nda.

[wanpengqian_gmail.com]

remove nvd modification.

[imp]

Like I said before, we shouldn't be basing this off the async callback, but instead the namespace change stuff.
Namespaces can arrive and depart, and this doesn't cope with that aspect of things, just the size change.

sys/cam/nvme/nvme_da.c

717

You can't do this in the kernel. the stack is too small.

939

I think this should be after we set the other parameters.

sys/dev/nvme/nvme.c

281 ↗

(On Diff #98635)

No. Like I said before, we do not want to parse the NVME log pages in the clients. This change is wrong.

sys/dev/nvme/nvme_ctrlr.c

760

I think this is wrong on many levels.

sys/dev/nvme/nvme_sim.c

382

this is also wrong. It doesn't handle the multiple name spaces arriving or departing. It only handles the size change.

[wanpengqian_gmail.com]

rebase as reviewer suggested.

[wanpengqian_gmail.com]

fix unrelated change.

[wanpengqian_gmail.com]

sys/cam/nvme/nvme_da.c

717

OK, remove this output.

939

Yes, I did not match the original sequence.

sys/dev/nvme/nvme_ctrlr.c

760

since notify_ns will query the namespace, that will cause kernel panic. here we post a query request to queue. other thread will query and construct_ns and call notify_ns after that.

sys/dev/nvme/nvme_sim.c

382

I will not use async callback this time.

[wanpengqian_gmail.com]

after NVME_PROBE_IDENTIFY_NS is done, fire a AC_GETDEV_CHANGED async event.

[wanpengqian_gmail.com]

I tested with 14.0-CURRENT Oct 10 2022, and it works. Is it the right direction for dynamically changing the disk size after hosts resize the disk?

[imp]

Ah, I missed the update on this review. It's largely the right direction and I'm sorry I missed it...
I've since updated different parts of the code, so this isn't quite right... But the glue to the bulk of this code is change, the bulk of this code hasn't changed, so I'll commit this to the tree shortly.

[wanpengqian_gmail.com]

@imp I'm not very familiar with the CAM subsystem, so it would be great if you could make the necessary adjustments and commit this to current. Thank you very much.