Re: You can already use ZFS as a bolt on
Yes, lots of people use ZOL (ZFS on Linux), me included. Works well, but building of ZFS modules is not something that everyone will want to do. And it has to be done after each kernel upgrade.
I understand what Canonical want to do it for their users. I do not know for sure whether or not they integrate ZOL into Linux sources, but I assume they do not. ZFS modules can be perfectly well built from the separate source tree - the resulting binaries won't allow boot from ZFS, but will allow ZFS to be used as a root filesystem (or any additional) which IMO is just right. Majority of users will normally want less esoteric filesystem for /boot anyway (and I do not see Canonical even suggesting use of ZFS for /boot , for support reasons). On the other hand, integrating ZOL into Linux kernel and then distributing the resulting binaries would IMO cross the legality barrier and there is no reason to do it, as explained above. Perhaps RMS assume they do it.
Also, I have trouble with RMS assertion that releasing combined work of Linux kernel with non-GPL module violates GPL, for the very simple reason that GPLv2 (under which Linux kernel is explicitly released) never mentions "combined work", hence this term cannot be applied to it. It does apply to derived work, which arguably is a different matter. The relevant section of GPLv2 license is cited below:
If identifiable sections of that work are not derived from the Program, and can be reasonably considered independent and separate works in themselves, then this License, and its terms, do not apply to those sections when you distribute them as separate works
The above section has been removed in GPLv3 and replaced with provisions for combined work. I understand RMS may wish Linux kernel was released under GPLv3, but it is not, and its license has different provisions from GPLv3. Also, since ZFS is maintained for many different operating systems (upstream is Illumos, which is descendant of Solaris - open source under CDDL, which incidentally is also ZFS license), it cannot be considered derived work of Linux kernel. The disagreement seems to be whether ZOL (i.e. port of ZFS to Linux) can be considered derived work of Linux kernel, or alternatively whether the act of releasing its modules as a part of a distribution makes it a derived work.
While I am very grateful to RMS and FSF for GPL which made Linux and GCC and many other great technologies available to me, for free, I do think they are going too far here. ZFS is open source even though the license is different from their preferred one. There are closed source blobs in Linux kernel I respectfully suggest they should focus on instead, e.g. nvidia module, firmware blobs and similar.
Oh, and Oracle has absolutely nothing to do with the current ZFS as maintained by Illumos team and others. Oracle have their own closed source version which is not compatible with ZFS as used by Illumos, FreeBSD or ZOL - they all form OpenZFS, which is open source as opposed to variant of ZFS used by Oracle and which no-one cares about (except for certain Oracle customers) and which is way behind OpenZFS. There is absolutely no reason to invite Oracle into this discussion.