Re: NULL pointer dereference in swsusp_free with 3.17-rc5

["Bjørn Mork" <bjorn@mork.no>]

"Rafael J. Wysocki" <rjw@rjwysocki.net> writes:

> I've decided to go with a revert for 3.17, as we don't seem to have an immediate
> fix and the final 3.17 may be as close as this Sunday.  So I'm going to send my
> final pull request for 3.17 to Linus tomorrow or early on Friday.

Sounds safest to me, FWIW.

For the next round of this, I think the only missing part was some test
like

        if (!forbidden_pages_map || !free_pages_map)
           goto return_without_freeing_anything;

at the beginning of swsusp_free().  I think we can agree that it isn't
necessary to repeat that test for every page like the old code did :-)

But I I believe it would be useful to analyze exactly why this is
necessary, and possibly add a comment with that explanation.  I don't
have any other arguments for the test than the old code and the oops.
If this is how it is designed then that would be nice to spell out.  If
not, then maybe there is a flaw somewhere else that should be fixed
instead.

And BTW, I believe it would be useful if at least one more person in the
world tested hibernation between each release ;-)


Bjørn