Re: [dm-devel] LVM snapshot broke between 4.14 and 4.16

[Theodore Y. Ts'o]

On Fri, Aug 03, 2018 at 09:31:03AM -0400, Mike Snitzer wrote:
> 
> Debian is notorious for having a stale and/or custom lvm2.
> Generally speaking, it is recommended that lvm2 not be older than the
> kernel (but the opposite is fine).

On Fri, Aug 03, 2018 at 03:31:18PM +0200, Zdenek Kabelac wrote:
> IMHO (as the author of fixing lvm2 patch) user should not be upgrading
> kernels and keep running older lvm2 user-land tool (and there are very good
> reasons for this).

I'm going to have to strenuously disagree.

In *general* it's quite common for users to update their kernel
without updating their userspace.  For example, I as a *developer*, I
am often running bleeding kernels (e.g., at the moment I am running
something based on 4.18-rc6 on a Debian testing system; and it's not
at all uncommon for users to run a newer kernel on older
distribution).

This is the *first* I've heard that I should be continuously updating
lvm because I'm running bleeding edge kernels --- and I would claim
that this is entirely unreasonable.

I'll also note that very often users will update kernels while running
distribution userspace.  And if you are using Linode, very often
*Linode* will offer a newer kernel to better take advantage of the
Linode VM, and this is done without needing to install the Linode
kernel into the userspace.

It *used* to be the case that users running RHEL 2 or RHEL 3 could try
updating to the latest upstream kernel, and everything would break and
fall apart.  This was universally considered to be a failure, and a
Bad Thing.  So if LVM2 is not backwards compatible, and breaks in the
face of newer kernels running older distributions, that is a bug.

If there is a fundamental bug in the userspace API, and it can't be
fixed without a serious security bug, sometimes we need to have an
exception to the "you can't mandate newer userspace" rule.  But I
don't think this falls into this category; how would a user "exploit"
what people are calling a "security bug" to break root?

    	      	      	      	    		   	- Ted