Upcoming ABI Breakage in RELENG_7

[Kostik Belousov]

On Wed, Jul 30, 2008 at 01:26:33PM +0200, Heiko Wundram wrote:
> Am Mittwoch, 30. Juli 2008 13:03:34 schrieb Ronald Klop:
> > I think in this case (the ABI breakage) it is more nice to say: "If you
> > don't know what to do, you wil probably not see any problem because of
> > this."
> > The edge case of what can go wrong is so small, that you must be doing
> > quite specialized stuff to see breakage. In that case you would understand
> > what is going on. (IMHO)
> 
> Err, no.
> 
> As someone else has already noted, a prominent KLM that's distributed 
> separately from the kernel which uses the vnode structure extensively (which 
> I didn't think of at all until I read the respective mail) is fuse. A 
> pre-upgrade compiled fuse is most certainly going to break because of this 
> change. Some people have dual installations of Windows/FreeBSD (at least I'd 
> presume that's the case with the fiddlers like me that track -STABLE as a 
> hobby, not as a developer, or those developers who program for Windows as a 
> day-job, also like me) who use ntfs-3g to mount their NTFS-data; 
> additionally, I also extensively use ssh-fs, which is also fuse-based, and I 
> guess there are also others who use it, and so the reach of this ABI-change, 
> at least IMHO, is much larger than the original message makes you believe.
> 
> Now, after the update, a lot can go wrong, because the fuse KLM is loaded by 
> an init-script, and your system is most probably going to Oops while booting 
> if you didn't think of disabling the fuse init-script before you update your 
> kernel, and will NOT fail "gracefully". If the respective person doesn't know 
> how he/she should boot to single-user-mode, update rc.conf to disable this, 
> reboot, rebuild the module to get the system back up, the only thing I can 
> possibly say is: "don't track stable."
> 
> It might've sounded a bit harsh what I wrote, but tracking -STABLE means 
> knowing your system enough so that you know how to fix things if they come 
> back to bite you (especially after getting a HEADS UP). And that doesn't seem 
> to be the case here if the respective person asks for SPECIFIC instructions 
> what to do.
> 
> So, again: DON'T track -STABLE if you can't fix the system if it breaks, and 
> AFAICT this change is most certainly going to break quite a few systems.

I do not use the fuse fs myself. But the issue was raised during the
internal discussion of the MFC, and it seems that fuse could be not
affected.

The change only adds the field at the end of the vnode structure, so all
existing fields offsets are intact.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20080730/8e03b930/attachment.pgp