Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for May 23, 2013
An "enum" for Python 3
An unexpected perf feature
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
To me this sounds more like "weird hack" than "promising technology".
I mean, is it really worth that much to save the reboot compared to
having a strangely binary patched kernel ?
Ksplice: kernel patches without reboots
Posted Apr 29, 2008 21:11 UTC (Tue) by proski (subscriber, #104)
Posted Apr 30, 2008 2:44 UTC (Wed) by clugstj (subscriber, #4020)
But many more weird hacks started as weird hacks.
Posted Apr 30, 2008 12:47 UTC (Wed) by hmh (subscriber, #3838)
And remain as weird hacks, causing trouble and maintenance nightmares.
Which in fact IS a big issue.
Posted Apr 30, 2008 4:32 UTC (Wed) by jamesh (guest, #1159)
If there is a critical security vulnerability that must be fixed as soon as possible, but you
have jobs running that you'd prefer not to kill, then it might be worth using such a tool
(assuming it doesn't risk causing corruption).
In the best case, things run smoothly and the vulnerability is patched. You can then perform
the reboot at your leisure. If things don't work out, you can reboot into the new kernel as
you would have done previously.
Posted Apr 30, 2008 8:40 UTC (Wed) by tzafrir (subscriber, #11501)
Such hack seems to be applicable to distro kernels: exactly the same kernel on many
machines. Hence this work only needs to be done once.
Posted Apr 30, 2008 12:58 UTC (Wed) by miahfost (subscriber, #51602)
Think of a small switch somewhere running linux. An embedded version of linux perhaps. Now
imagine it routes telephone calls, and is placed in a desert, 100 kilometers from anywhere,
10,000 kilometers from where it was developed.
Now imagine it needs a new patch _without_ taking down 10,000 subscribers' service. Would be
nice to have a thing like Ksplice, no?
Posted May 1, 2008 4:09 UTC (Thu) by brouhaha (guest, #1698)
Which is why phone switches like the AT&T 5ESS were designed such that the control processor
(running Unix) could be rebooted without disrupting phone calls. At worst it introduced a
slight delay in a subscriber getting a dial tone if they picked up just as the system was
No clever patch mechanism necessary.
For something more complex than a phone switch, in-service-upgrade is more difficult, so this
patch mechanism may be very useful.
Posted Jun 11, 2008 23:03 UTC (Wed) by telcoman (guest, #52500)
HW failure will tear down the calls anyway, if you rely purely on Ksplice. For telco systems
you need HW redundancy, too. Especially if you are running a commercial network, you had
better not loosing charging data due to HW issues.
Nevertheless, Ksplice alone could help stateless services, like Google's search, where you can
allow data loss due to HW issues. And it can help small systems, too, where HW failure is rare
and less disturbing.
Using Ksplice along with clustering solutions could improve the situation. Keep in mind,
though, that you are just patching the kernel, but the applications remain unfixed. So you did
not end up with a real solution.
Posted May 9, 2008 6:35 UTC (Fri) by xamal (guest, #51976)
I need to take the time to respond to this. This sort of thing is NOT new and has been done for a long long long time.
NASA does this as part of normal practice. Basically diff(objectold, objectnew) ... then carefully smush the diff in place. In an old Dr. Dobbs (of which I have a hard copy) the following was said:
DDJ: And when you had to hot-patch in flight?
GR: That's standard procedure. You always build in the ability to change it. ( Read ~ http://www.ddj.com/184411097 )
As further evidence that this is done on Spacecraft please read: http://www.flightsoftware.org/files/FSW07_Calder.ppt
If you really want to get into this read http://en.wikipedia.org/wiki/Binary_File_Descriptor_library
As I said before this stuff has been done since time immemorial (or at least when the first computers showed up) ... also read up self-modifying-code.
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds