I would also move rest of generic core dump bits into e.g. kern_ucoredump.c (from kern_sig.c).

Tracing the lineage of everything here is challenging, but here are some notes from my basic scouting of some of the more fundamental changes

sys/ucoredump.h:
 - markj: core_write, core_output, coredump_params, compress_user*
   largely imgact_elf.c in aa14e9b7c9d90 / r279801
   moved to exec.h in 33621dfc196e3 by trasz
 - kevans: coredump_writer + friends, coredump_vnode + friends
 - cem: coredump_pack_* (2 lines)

kern_ucoredump.c:
 - kib / emaste did work under the FreeBSD Foundation
   2014 (539c9eef121048 / r272535)
   2017 (6edbe5616c761bf)
 - markj implemented compress_user_cores as it is today
   78f57a9cdef729 / r327707
 - ps implemented kern.coredump
   e5a28db9f5cf27 / r58416 (2000)
 - dg implemented kern.sugid_coredump
   c87e2930e6a89 / r37226 (1998)

coredump_vnode.c:
 - oshogbo rewrote a bunch of logic for %I, resulting in corefile_open_last
   0dea6e3c9834ba / r335844
 - sef implemented kern.corefile / expand_name(), basis for corefile_open
   c5edb423c6271 / r37496 (1998)
 - pjd closed races in vnode open
   c345faea5abdc6 / r244452 (WHEEL systems)
 - alfred implemented initial compressed coredumps
   e7228204343ac / r204552
 - rpaulo implemented the devd notification
   842ab62b051379 / r278479

I think the executive summary is:

• sys/ucoredump.h: BSD-2-Clause, @markj (2015) and kevans (2025)
• kern_ucoredump.c: BSD-3-Clause, Foundation, markj, ps, dg
• coredump_vnode.c: BSD-2-Clause, @rpaulo, @alfred, @pjd, @sef, @oshogbo

I'm quite certain I've missed something somewhere, but this seems like a reasonable start -- I've tagged all of the above in case they want to object or clarify... almsot all of this work was before my involvement in the project. The licenses used are taken from where the majority of the file came from in the most recent iteration, I'm not sure if we need to have a 3-clause vs. 2-clause fight since that's probably not right.

In D51349#1175669, @kevans wrote:

Tracing the lineage of everything here is challenging, but here are some notes from my basic scouting of some of the more fundamental changes

sys/ucoredump.h:
 - markj: core_write, core_output, coredump_params, compress_user*
   largely imgact_elf.c in aa14e9b7c9d90 / r279801
   moved to exec.h in 33621dfc196e3 by trasz
 - kevans: coredump_writer + friends, coredump_vnode + friends
 - cem: coredump_pack_* (2 lines)

kern_ucoredump.c:
 - kib / emaste did work under the FreeBSD Foundation
   2014 (539c9eef121048 / r272535)
   2017 (6edbe5616c761bf)
 - markj implemented compress_user_cores as it is today
   78f57a9cdef729 / r327707
 - ps implemented kern.coredump
   e5a28db9f5cf27 / r58416 (2000)
 - dg implemented kern.sugid_coredump
   c87e2930e6a89 / r37226 (1998)

coredump_vnode.c:
 - oshogbo rewrote a bunch of logic for %I, resulting in corefile_open_last
   0dea6e3c9834ba / r335844
 - sef implemented kern.corefile / expand_name(), basis for corefile_open
   c5edb423c6271 / r37496 (1998)
 - pjd closed races in vnode open
   c345faea5abdc6 / r244452 (WHEEL systems)
 - alfred implemented initial compressed coredumps
   e7228204343ac / r204552
 - rpaulo implemented the devd notification
   842ab62b051379 / r278479

I think the executive summary is:

• sys/ucoredump.h: BSD-2-Clause, @markj (2015) and kevans (2025)
• kern_ucoredump.c: BSD-3-Clause, Foundation, markj, ps, dg
• coredump_vnode.c: BSD-2-Clause, @rpaulo, @alfred, @pjd, @sef, @oshogbo

I'm quite certain I've missed something somewhere, but this seems like a reasonable start -- I've tagged all of the above in case they want to object or clarify... almsot all of this work was before my involvement in the project. The licenses used are taken from where the majority of the file came from in the most recent iteration, I'm not sure if we need to have a 3-clause vs. 2-clause fight since that's probably not right.

So I'm a little confused.... Did you move everything to kern_ucoredump.c. If. so: put all three copyrights there and get on with life. The code that I think is moved in this patch (I glanced only, didn't do per-line review).

If this is 'what to put on each file' then one per is fine. Generally copied code gets the copyright of where it was copied from... That's never wrong, but sometimes could be better...

The general rule isn't 'who changes it' so much as 'who contributed a significant amount of creative content' to the file. Those are the people that hold the copyright. If there's been such a significant rewrite of a file since the original author created it. If so, then that person no longer has a copyright to the file (but everybody else committed to it with the original license which implicitly accepts that license and is what their code should be considered to be).

Otherwise, I'm confused...

In any case BSD-3-clause vs BSD-2-clause isn't worth fighting before landing. ps and dg would sign-off on the change, but have high latencies when I've pinged them in the past. If the code they wrote is now de-minimus, then we could change w/o asking and changing the copyright owners. It's nicer to give an ack to the original author "This code was rewritten from code originally by Paul Saab" or similar if you want. I tend to do this when I copy the whole file and use it as a template (say to create a new driver, leaving only boilerplate elements from the original).

Clear as mud, eh?

tl;dr: copying the copyright/license is right, but might be suboptimal. It's necessary and sufficient to commit. However, one might optimize the attribution based on what actually remains, but we have a long history of not bothering except in extreme cases.