> On Nov 25, 2015, at 10:54 AM, Dan McDonald <[email protected]> wrote:
>
> I'm going to take a crack at an accompanying kernel fix this afternoon, and
> report back on this (new) thread when I'm done.
Done!!!
I've updated this webrev in place:
http://kebe.com/~danmcd/webrevs/4986/
The design idea is that some properties mustn't be set until AFTER the dataset
is received. I call these "delayable" properties in the kernel code I've
changed. And it MUST accompany the libzfs changes from earlier for maximum
safety and fewer errors. The kernel changes alone will yield correct results,
but at the cost of error output from zfs(1M) even when the right things happen
eventually. The libzfs changes are not sufficient, but they are necessary for
correct operation.
I've tested it successfully with the reproduction in bug 4986. As I pointed
out earlier, a distinct commit is the zfs test for it:
http://kebe.com/~danmcd/webrevs/test-4986/
I will eventually be running the full zfs test suite on things, including the
new bits above in test-4986.
My timing is interesting, given it's a 4-day holiday weekend in the US, so I
won't expect much feedback until next week. Nonetheless, I think I fixed this,
and I look forward to your comments. I suspect some of the libzfs changes will
need to be cleaned up. I tried reducing parameters, and may have caused more
clutter than I removed, for example.
Thanks!!!
Dan
p.s. Happy Thanksgiving to the US readers in the audience.
_______________________________________________
developer mailing list
[email protected]
http://lists.open-zfs.org/mailman/listinfo/developer
