Hmm...
Two different opinions, next to each other (ignoring the page break - which users may have changed in their CP).
Readers can read and consider both.
As the link I included is quite clear about - not even a lawyer can say 'yes' or 'no' to the question about 'fair use'.
You can only know when you have been sued and won in court.
I.E: The court told you what you did was alright.
So doing a 2.5.8 -> 3.4.2 set of update patches and distributing them would be at the distributor's risk.
If Amazon sues you and you win, all is good.
Otherwise, all is expensive.
= = = = =
In a similar type of situation -
For a Linux kernel that uses an initial (ram resident) file system, the build can be done in two different ways:
bind the kernel and the initramfs builds into a single binary
or
keep them as two, individual, builds so they have only run-time linkage.
The single binary approach makes the initramfs binary (and its sources) subject to the kernel's GPLv2 license.
The two binary approach does not.
(Which is the reason why the method exists.)
Amazon chooses to use the single binary approach, and fails to release the initramfs source (a clear violation of the Kernel's license).
They have just decided that the risk of being sued (and losing) is small enough to get away with it.