Can't upgrade jails to 8.0 using freebsd-update

[Eirik Øverby]

On 15. okt. 2009, at 20.56, Martin Turgeon wrote:

> Eirik Øverby a écrit :
>>
>> On 15. okt. 2009, at 19.44, Martin Turgeon wrote:
>>
>>> Eirik Øverby a écrit :
>>>>
>>>> On 8. okt. 2009, at 21.04, Martin Turgeon wrote:
>>>>
>>>>> Eirik Øverby a écrit :
>>>>>> On 8. okt. 2009, at 19.58, Martin Turgeon wrote:
>>>>>>
>>>>>>> Hi everyone!
>>>>>>>
>>>>>>> I just upgraded a 7.2-REL to 8.0RC1 using freebsd-update. The  
>>>>>>> upgrade
>>>>>>> went fine on the base system following the procedure written  
>>>>>>> in the
>>>>>>> announcement email by Ken Smith. My problem is when I try to  
>>>>>>> upgrade my
>>>>>>> jails, I get this message:
>>>>>>>
>>>>>>> # freebsd-update -b /usr/jail/mysql/ fetch install
>>>>>>> Looking up update.FreeBSD.org mirrors... 3 mirrors found.
>>>>>>> Fetching metadata signature for 8.0-RC1 from  
>>>>>>> update5.FreeBSD.org... done.
>>>>>>> Fetching metadata index... done.
>>>>>>> Inspecting system... done.
>>>>>>> Preparing to download files... done.
>>>>>>>
>>>>>>> No updates needed to update system to 8.0-RC1-p0.
>>>>>>> No updates are available to install.
>>>>>>> Run '/usr/sbin/freebsd-update fetch' first.
>>>>>>>
>>>>>>> But, if I compare the dates of the files in the base system to  
>>>>>>> the files
>>>>>>> in the jails, it's obvious that the jails are not up to date.
>>>>>>>
>>>>>>> It seems like freebsd-update doesn't care about the basedir I  
>>>>>>> specified.
>>>>>>
>>>>>> It does, but if you do a 'uname -a' - inside or outside the  
>>>>>> jail - you'll see that it reports the OS revision of the host.  
>>>>>> So you should have updated your jails first, then the host ...
>>>>>>
>>>>> Ok but if I update in the process of upgrading the first jail,  
>>>>> the new kernel will be installed and asked to reboot. After  
>>>>> that, I will have the same problem when upgrading the other  
>>>>> jails and the base system, right? There must be something I  
>>>>> don't  understand well. Thanks a lot for your answer.
>>>>
>>>> The kernel will be installed inside the jail, and the message  
>>>> about rebooting can be safely ignored. Just run the install  
>>>> command once more, and you're done and can move on to the next  
>>>> jail. :)
>>>>
>>>> /Eirik
>>>>
>>>>
>>>>> Martin
>>>>>> One way to get around it is to replace /usr/bin/uname with a  
>>>>>> shell script, which calls the original uname (which you have  
>>>>>> renamed) and pipes through something like sed to replace the  
>>>>>> revision with what you used to have:
>>>>>>
>>>>>> #!/bin/sh
>>>>>> /usr/bin/uname.org $* | sed s/"8.0-RC1-p0"/"7.2-RELEASE_p3"/g
>>>>>>
>>>>>> And this is a seriously butt ugly hack.
>>>>>>
>>>>>> /Eirik
>>>>>>
>>>>>>> Thanks a lot for your help,
>>>>>>>
>>>>>>> Martin
>>>>>>>
>>>>>>>
>>> Thanks a lot! It worked great, but I'm still concerned by the fact  
>>> that the world in the jails are from 8.0 while the kernel is still  
>>> at 7.2 during the updates of the jails. In the normal update  
>>> procedure, the kernel is upgraded first, rebooted and then the  
>>> world is updated. It must have a good reason for this. Why can I  
>>> jail be an exception to this rule?
>>
>> Because when you upgrade the host, the very binaries you are  
>> installing are being installed by ... the binaries you are  
>> installing. Which is why you'll want them to be reasonably in-sync  
>> with the kernel.
>>
>> When upgrading jails using freebsd-update, my understanding is that  
>> it uses the host binaries to  push files to the target directory  
>> (basedir). The risk of trouble is therefore very low (I've upgraded  
>> dozens of jails from 7.x to 8.0-RC1 without any issues), though I  
>> can probably imagine situations where this is not the case - i.e.  
>> if the upgrade depends on functionality in an upgraded binary in  
>> order to be able to complete the upgrade, however that sounds like  
>> a very unlikely scenario to me. And *any* use of the basedir option  
>> to freebsd-update would break in such a case.
>>
>> /Eirik
>>
> Sounds reasonable to me. Thanks for your quick reply.

Reasonable != true. And for that matter, reasonable != reasonable.  
YMMV etc. etc. I haven't peeked (much) into the code, so I'm just  
guessing from my experiences. ;)

People in the know should correct me if I'm wrong..

/Eirik