Ignore object->handle for OBJ_ANON objects.

[kib]

Summary

Note that the change in vm_object_collapse() is arguably a correctness fix. We must not collapse into content-identity carrying objects.

[This is a preparational commit, the plan is to use handle for OBJ_ANON to point to the end of the shadow chain].

[jeff]

sys/vm/vm_object.c

569–570

This could be an assert that it is an ANON object. Non-anon should never have a shadow_count now.

603

I'm a little confused at some of the dead checks in the shadow chains. I don't know where they are possible and where they are not. I suspect some code may be unnecessarily checking and handling them. a vnode may become dead but another anon with a ref should not right?

1752

Here again, can this actually become dead now that we limit to anon objects?

[kib]

sys/vm/vm_object.c

603

Object collapse creates dead anon objects with refcount > 0.

[jeff]

sys/vm/vm_object.c

603

But it should only do so when ref_count == 1. Which means if we hold the object with the final ref it can't disappear out from under us?

[kib]

shadow_count > 0 implies OBJ_ANON.

[kib]

sys/vm/vm_object.c

603

Definitely it cannot disappear, but I am not sure that collapse cannot run in parallel while we temporary unlock the object, the case when this code is run is not when the last ref was deallocated, it is pre-last.

[jeff]

sys/vm/vm_object.c

603

I see now that we drop locks in collapse scan and that could possibly expose OBJ_DEAD in some cases.

[pho]

I ran the full stress2 test set without finding any problems.