vm_object_coalesce(): check that coalescing does not revive stale pages

[kib]

Summary

Also:

swap_pager_seek_data(): improve assertions

swap_pager_seek_data(): move the clipping at the object size to consumers

Check that the passed object is indeed anonymous.
Accept any kind of the object lock ownership, not only for read.

[markj]

sys/vm/swap_pager.c

2464

This function is called from tmpfs though, so won't this assertion fail? Should we check for OBJ_SWAP instead?

[kib]

swap_pager_seek_data(): check for the swap object, this is what the function assumes

[markj]

sys/vm/vm_object.c

2251 ↗

(On Diff #167999)

Should we also verify that there are no valid, resident pages in this range? That might catch cases where we have resident pages at indices above prev_object->size.

2256 ↗

(On Diff #167999)

prev_size is a page count. Note that we do prev_size >>= PAGE_SHIFT; near the beginning of the function.

And isn't pstart just the same as next_pindex?

[kib]

sys/vm/vm_object.c

2251 ↗

(On Diff #167999)

Indeed. I removed the clipping of the result from swap_pager_seek_data().

[kib]

Remove pstart.
Remove clipping from swap_pager_seek_data(), so that PG_ZERO check can see all pages in the object.

[alc]

sys/vm/vm_object.c

2256 ↗

(On Diff #168168)

[alc]

sys/vm/vm_object.c

2258 ↗

(On Diff #168168)

pend could also be eliminated without any loss of clarity.

[kib]

Fix pend (and eliminate).

[alc]

sys/vm/vm_object.c

2251 ↗

(On Diff #168174)

[kib]

Correct the article in comment.