vm_map: Record the maximum size of the map over time

[markj]

Summary

This might be useful for some emulation layers, see bugzilla report

PR: 259940

[kib]

In kern/kern_exec.c:exec_new_vmspace(), in case the existing vmspace is reused, I believe the maxspace needs to be reset to zero,

sys/vm/vm_map.c

1761–1764

This place should be also handled.

4926–4928

And this.

[kib]

sys/sys/user.h

216

Why the new field is put there and not in kinfo_vm_layout?

[markj]

sys/sys/user.h

216

Mostly because we already export the map size there. IMO it should be added to kinfo_vm_layout as well (the map size is missing there), but we might as well have it here as well so that existing users can use it conveniently.

[markj]

Address kib's comments, export map size and max size in the vm_layout sysctl as well.

[kib]

sys/sys/user.h

216

kinfo_vm_layout is much easier to expand, even if spaces are fully utilized. We only return kinfo_vm_layout for single process, so we can detect old programs requests by comparing the size.

While for kinfo_proc, the structure is often is returned in the array, so exhausting the spares means that we need to add yet another oid with all backward-compat machinery.

This is why I asked to move the new field to vm_layout.

sys/vm/vm_map.c

4338

I remember the previous version simply copied the vm_map.size on fork. Why change?

[markj]

Avoid modifying kinfo_proc.

linprocfs will require more changes to provide VmPeak.
Drop that part of the diff for now, it is logically separate anyway.