13
Sep
17

Kubuntu 17.04 on Dell e5470 for extreme battery life

So apparently it’s been 5 years since I last updated this series: https://www.totalnetsolutions.net/2012/12/09/lenovo-t430-running-kubuntu-12-10-for-extreme-battery-life/

i’ve restarted most of these configurations over the past 5 years, especially as I’ve switched away from WWAN to tethering, and from spinning rust to SSD, but a lot of the core concepts remain: about 6 years ago my battery died after 3.5 hours of VM troubleshooting, while on a flight, and I lost some data in the emergency “go to sleep, not hibernate”, which cost me 2 hours of rework in the hotel at midnight. My goal, now, is “be able to work multiple simultaneous tasks, have a VM running, and still get super-long battery life when I need it, but not impact performance noticeably.”

With powertop reporting <5W (14h remaining) power consumption while idle, and <6W (11 hours) with firefox open while I start to write this post in August, 2017, I think I’ve hit the mark reasonably well.

As with the Lenovo T430 in the previous post, everything I care about works right out of the box, but when I first started my custom kernels, I missed a few things that I had to add back in before writing this up.

Hardware


CPU: Intel(R) Core(TM) i5-6300U CPU @ 2.40GHz
Memory: 16GB RAM
VGA compatible controller: Intel Corporation HD Graphics 520 (rev 07)
Ethernet controller: Intel Corporation Ethernet Connection I219-LM (rev 21)
Network controller: Intel Corporation Wireless 8260 (rev 3a)

Audio device: Intel Corporation Sunrise Point-LP HD Audio (rev 21)
Bluetooth device: Intel (integrated on the USB)
Synaptics Touchpad and Twiddler Mouse

Jump to main sections with these links:
CPU Configuration
Network Configuration
Video Configuration
Encryption / Security configuration
Battery saving configuration
Custom kernel .config

CPU and Battery


I have a series of posts on my love of getting the most performance and battery life I can from my systems, see the last and first of the series for a bit more, or dig through my twitter on the subject. What’s new this year is the latest i5 Core CPU with Linux 4.4 has a new “pstate” performance governor that’s not actually buggy anymore, if you configure it right. I used to use the acpi_cpufreq governors “ondemand” on AC and “conservative” on battery. But the new pstate drivers apparently perform better (thanks Phoronix) AND scale down for battery savings better, so I needed to switch that. Since I was switching governors, I figured it was time to re-check my 2007 finding that moving from “GENERIC_CPU” to “MCORE2” saved me 30+ minutes alone.

Well, it does. But I didn’t keep the data, sorry. What this means is I once again needed to custom compile a kernel to get the right CPU options, and to get the new pstate driver. Since I was in doing that, and since the 2012 post, I’ve moved away from the default kernel Scheduler to Colin Kalvis’ BFS scheduler, so we get to patch THAT in as well. More on those options down at the custom kernel config section, but the point here is that BFS added some stability to heavy “running multiple VMs, and processing 4GB of raw data in Perl” swapping problems I was having, even with 16GB RAM, as well as not hurting my battery life, with a high possibility of 10-20 minutes extra life on normal operations.

The battery in the system is designed at 8157000 mWh, and after 6 months is down to 6621000 mWh. The “amount of time running” is based on the past 2 months, not day 1 of receiving the laptop.

Lastly, I’m still using cpufreqd, but the configuration is vastly simplified – pstate “powersave” when my AC is not plugged in, or the battery’s below 70%, and pstate “performance” otherwise. Ubuntu fixed the broken cpufreqd daemon sometime in 2014, so I’m back to the distribution default version of that, yay!

My custom cpufreqd.conf.

Network


I stopped using my jumbo frames script from here in Kubuntu 16.10, because apparently NetworkManager can figure that out on its own, and it’s been relatively successful. My wireless adapter connects to the new Netgear T6400 at near gigabit speed, but the R6400 doesn’t support jumbo frames itself, so I’m segmenting off some new VLANs to break the Jumbo Frames hosts from the wireless / nonintelligent hosts. That’ll mean resurrecting my jumbo frames script to instead set the VLAN Tag when I’m home.h

Sound


Sound has always been a joke for Linux users, but the Intel HD-Audio has been really solid for me for several years, especially with pulseaudio actually being relatively stable for me. When I recieved the laptop, I was having a problem where full-duplex audio was causing what appeared to be a storm of interrupts that hung the entire laptop. But about 2 months of debugging resulted in “I built a new kernel, and now it works fine.” I don’t know if it was a bug in the codec in the kernel, or something that silently patched. How I have Bluetooth audio headset, bluetooth headset for online conference calls, and appropriate switching for apps and reminders (reminders / alerts go to speakers and bluetooth in my config, in case I take the headphones off), with the options int he PulseAudio configuration in KDE. I have no “.asoundrc” or /etc/asound or /etc/pulse or ~/.pulse/client.conf anymore either, which is great!

Video


For the first time in years, I do not have a multi-graphics card system to deal with. The i915 driver works out of the box and is unremarkable, but functional. And great for battery life. But boring to discuss.

Encryption and Security

Encryption

During installation, I chose the option to use an encrypted LVM volume. This uses DM-Crypt to encrypt the full HDD, so that it has to be unlocked at boot time. The Kubuntu installer seems to forget this fact, so it also asks you to set up ecryptfs private home directories, which is NOT neccessary for a single-user laptop, since the whole OS is already encrypted. The only oddity with dm-crypt is that sometimes the splash screen prompt to unlock the computer doesn’t show. Originally, if I just wait for disk activity to disappear, and have a blank screen, I can just type the passphrase, and it’ll still unlock successfully. But I instead made a change to /etc/default/grub:

GRUB_CMDLINE_LINUX_DEFAULT="noquiet nosplash"

And now I don’t get the splash screen, and the prompt comes up properly right away. And I get all the hacker-y looking boot errors from systemd.

Security

Because this laptop has some sensitive work information on it, I wanted to get a bit more paranoid with the “unattended on a conference table” and “connected to a public wifi network” situations, especially since I actually have OpenSSH listening on all interfaces (yes, I ssh into my laptop from my phone more often than you do). I purchased a multi-protocol Yubikey and downloaded and installed the Yubikey PAM module for Challenge-Response, with the instructions on their website, here. Combined with Active Directory authentication, my cached user can only log in if the Yubikey is inserted into the laptop. So when I step away in meetings, the laptop locks, and my password can’t be cracked.

For additional security, I enrolled the root account on my laptop into my BeyondTrust Password Safe (since I’m working for them right now). This product rotates the root password every 14 days with a 50+-character random passcode, so even an attacker getting physical access once it’s booted (decrypted) will have little chance of breaking into the box even when I have the yubikey in place.

Additional Power Savings


I still run laptop-mode tools to cut down on power utilization from non-CPU peripherals. I could get more by having the ethernet port actually turn off when on battery, but I actually use it on battery quite a lot, so I’m not sure the hassle of re-enabling it is worth the battery savings. Here are the configurations I use:
intel-hda-powersave
intel-sata-powermgmt
intel_pstate
laptop-mode
runtime-pm
wireless-iwl-power
cpuhotplug
bluetooth
battery-level-polling
ethernet

What I’m now getting is 5-7W of power utilization while online with firefox and chrome both running, bluetooth running, and no VMs. Booting my Windows VM in VMware Workstation bumps me up to 15-20W, but I’m still getting 5 hours of battery life with no features disabled AND running a full Windows VM (the Windows VM has battery detection disabled, too). My non-VM battery life is reporting in the 9-11 hour range, but I’ve never had to use it that long to worry.

Kernel Config


I use the Ubuntu Kernel sources, mostly because the laptop tells me when there’s new Kernel sources with security fixes. I’m using BFS as my scheduler, which is fantastic when I get into “3 VMs using 12GB RAM and a reporting job wanting another 6GB” swap death. I have enough keyboard control to kill the reporting job, then shut down the VMs, and try the reporting job again. Before BFS, I either waited 6 hours, or rebooted the whole damn laptop.
BFS patches are here. If that doesn’t make sense, don’t use them. Please.
My custom kernel .config is here.

Building

The build system I use is the same as in 2012:

sudo apt-get install fakeroot build-dep linux-image-`uname -r`
sudo apt-get install linux-source
sudo usermod -a -G src YOUR_USERNAME

Now log out and back in, so that you’re a member of the “src” group.

cd /usr/src
sudo chown -R $USER:src .
tar -jxf ./linux-source-4.4.0/linux-source-4.4.0.tar.bz2
ln -s linux-source-4.4.0 linux
cd linux
wget http://www.totalnetsolutions.net/wp-content/uploads/2012/12/rob-config-20121204c.txt
mv rob-config-20121204c.txt .config
make oldconfig
make menuconfig

Make any changes you want in here, then exit and save

make-kpkg --initrd --rootcmd fakeroot --append-to-version=.20170912a kernel_image kernel_headers

You’ll get 2 DEB files in /usr/src that you can then install and boot to. the “append-to-version” I use as a dating system for my kernels. “20170912a” means the 2nd kernel attempt on September 12, 2017, the day I’m writing this post (first attempts get no letter).

13
Sep
17

Getting Group Policy information via LDAP and SMB only

Sometimes it’s nice to know what’s happening under the hood, so let’s talk about how Group Policy is built, by tearing down how to access a particular policy. First, Group Policy is implemented in 2 parts, an LDAP part and a file part, delivered via SMB (CIFS if you’re oldschool) via DFS (Distributed FIle SYstem). Because the DFS part is replicated completely differently than the AD part, there’s a version number for each Group Policy object that’s stored in both places to keep them in sync. Most GPO engines remember the last version they applied by remembering the lowest of the 2 numbers (the LDAP version and the file version in the GPT.INI), if they don’t match.

Let’s talk about the “Default Domain Policy” which everyone will have one of. To find where that policy lives, you have to ask AD. The policy doesn’t actually live in the OU or Domain where it’s linked, so we have to back out the link:
ldap_search_s(ld, "dc=company,dc=com", 2, "(objectClass=organizationalUnit)", gpLink, base, &msg)
We’ll get back something like:

gPLink: [LDAP://CN={6AC1786C-016F-11D2-945F-00C04fB984F9},CN=Policies,CN=System,DC=company,Dc=com;0]

Now, this is a multi-valued array, because multiple GPOs can be linked, in order, to a single OU or Domain or Site. But we only care about this one, so let’s see what’s in it:

ldap_search_s(ld, "CN={6AC1786C-016F-11D2-945F-00C04fB984F9},CN=Policies,CN=System,DC=company,DC=com", 2, "(objectClass=*)", gPCMachineExtensionNames;gPCFileSysPath;displayName;versionNumber, 0, &msg)

That’ll get us the Client Side Extensions (where the work actually happens), and what the file path to the files in the estension are stored, as well as the pretty name of the Group Policy Object:

displayName: Default Domain Policy;
gPCFileSysPath: \\child1.lwtest.corp\sysvol\child1.lwtest.corp\Policies\{31B2F340-016D-11D2-945F-00C04FB984F9};
gPCMachineExtensionNames: [{35378EAC-683F-11D2-A89A-00C04FBBCFA2}{53D6AB1B-2488-11D1-A28C-00C04FB94F17}{53D6AB1D-2488-11D1-A28C-00C04FB94F17}{D02B1F72-3407-48AE-BA88-E8213C6761F1}][{827D319E-6EAC-11D2-A4EA-00C04F79F83A}{803E14A0-B4FB-11D0-A0D0-00A0C90F574B}][{B1BE8D72-6EAC-11D2-A4EA-00C04F79F83A}{53D6AB1B-2488-11D1-A28C-00C04FB94F17}];
versionNumber: 15;

So we have the Default Domain Policy, as desired, but there are a bunch of client side extensions here. It’d be nice to know what they all do generically, without having to inspect each one.
And TechNet delivers on that desire: a list of all Client Side Extensions (in 2010) by GUID for easy reference. Now, I’m writing this, because someone asked where the Password Policy for the domain was stored. Well, that appears to be in: {827D319E-6EAC-11D2-A4EA-00C04F79F83A} Security, which our Default Domain policy applies. So, let’s go find the data!

One of the attributes in the list we last requested was gPCFileSysPath which returned a normal SMB share. If you browse to that share, you’ll see 3 objects:

  • A folder named “MACHINE”
  • A folder named “USER”
  • a file named “GPT.INI”

The GPT.INI will only have 2 lines:

[General]
Version=15

That’s the version number, that you can compare to the “versionNumber” property from the object. If they’re the same, you’re good. If not, your AD isn’t in sync.

In the “MACHINE” Folder are all the Computer Policy settings, and in the “User” folder are all the User Policy settings. Since we were talking about the Password Policy, which is affected on the SAM on the server, it’s a MACHINE setting. If you were to poke through, eventually you’d find this file:

\\domain\sysvol\domain\\MACHINE\Microsoft\Windows NT\SecEdit\GptTmpl.inf

with this data:

[Registry Settings]
MACHINE\System\CurrentControlSet\Services\Netlogon\Parameters\MaximumPasswordAge=4,15

And there’s your password policy, via LDAP and SMB only.

For a bit of additional background, when a computer processes this data, in this order, it will actually only apply CSEs from the gPCMachineExtensionNames that the computer recognizes and has DLLs (or whatever code, if it’s non-Microsoft vendor CSE) that can apply the CSE. This makes it technically safe to put multiple GPOs for multiple Operating sytsems on the same OU structure, knowing that the client computer won’t even bother downloading the files for un-recognized CSEs.

Now, that’s a lot of stuff to type into ldp.exe, how can we make a report on this a bit easier? Well, PowerShell could do it, but one of the products I work on is PowerBroker Open https://github.com/BeyondTrust/pbis-open and https://www.beyondtrust.com/products/powerbroker-identity-services-open/ which includes a CLI for browing ldap called “adtool”. With a bit of bash, we can list out all the group policy objects by name attached to a single OU:

$ cat report-gpos-by-ou.sh
GP=`adtool -a lookup-object --dn "$@" --attr gPLink`;
GPO=`echo $GP | sed -e 's/\[LDAP:\/\///g' -e 's/;[[:digit:]]\]/ /g'`
if [ -n "$GPO" ]; then
echo "";
echo "$OU";
for P in $GPO; do
G=`adtool -a lookup-object --dn "$P" --attr displayName`;
grep -q "$G" /tmp/gpos.txt;
if [ $? -ne 0 ]; then
echo $G >> /tmp/gpos.txt;
fi;
echo "$G";
done;
fi;
$ ./report-gpos-by-ou.sh "OU=Company,DC=domain,DC=com"
OU=Company,DC=domain,DC=com
PBUL Basics
GP-Preferences
$

24
Aug
17

Invalid ISTG in RODC-only Site

Our primary lab here at totalnetsolutions.net is a bit space-constrained (RAM constantly 50% overcommited in ESX), so we have it limited to only 2 domains, a parent, and a child. There’s 2 sites, one of them is a “BranchOffice” with only an RODC. This serves us well for the majority of customer cases, especially since the DHCP pool serves the BranchOffice site, which is routed via DSL to the CorpOffice site.

Recently we needed a 3rd domain (2nd child) for testing a particular user/group creation scheme for a customer. Upon completing testing in August, to reduce RAM utilization on the ESX host (and because it makes the AD Schema even more “real-world”, we dcpromo’ed the DC for the 3rd domain, deleting the entire domain. We deleted the objects, and ran the ntdsutil metadata cleanup, like good AD admins. Then we watched the logs fill up with Information events only for a few weeks, and forgot about the episode.

The thing was: the DC in the 3rd domain, because it was short-lived, was left with its DHCP address (it was never a DNS server), so it got added to the BranchOffice site. According to this 2009 Technet AD Troubleshooting Blog entry from Ingolfur Strangeland, RODCs won’t register as the Intersite Topology Generator (ISTG) for the site, but they will perform that role for themselves. (Background on the ISTG.) Because of this, the *new* DC, which was writeable, took over the role of the ISTG for the site, and made sure that all the replication connections went through it, as the only writeable DC in the site.

The problem is, we removed this server. When we removed the writeable DC in the BranchOffice site, there was no other server available to write “I’m now the ISTG” and replicate that setting outbound to the other site(s)… because the only remaining server was an RODC.

We discovered the problem 2 months later (Rob had a baby and wasn’t paying any attention to the lab for a while) when, after creating 1,000,000 new users in the lab, replication was surprisingly slow, but only into the Branch office. So slow that when we joined new computers, they’d boot up with “Service Principal Unknown” errors, and AD users couldn’t log in. In Active Directory Sites and Services we saw that the ISTG Server and Site were both “Invalid”.  This post from 2011 discusses how to move this, but not if it works for RODCs… the good news is that it does:

  1. open the Configuration container, like with adsiedit.msc
  2. Expand CN=Sites, and then the site with the broken ISTG (likely the site with the RODC)
  3. Double-click to open the properties of CN=NTDS Settings
  4. Find the value: “InterSiteTopologyGenerator” and paste in the full DN (from the Configuration Container, not the RootDN) of the RODC
    1. This is the “distinguishedName” value of the CN=<servername>,CN=<sitename>,CN=Sites,CN=Configuration object of the server in the site in question that *should* be the ISTG.
  5. Click “OK” to Save, use ‘repadmin’ or dssite.msc (AD Sites and Services) to force replication and wait 15 minutes (or your own inter-site replication time)

 

24
Aug
17

Signing RPM packages for Multiple RHEL versions

Signing RPMs is supposed to be easy:

gpg --list-keys
cat > ~/.rpmmacros < <EOL
%_topdir /home/rob/rpmbuild
%debug_package %{nil}
%_signature gpg
%_gpg_name Rob A <me@totalnetsolutions.net>
%{__gpg} \
gpg --digest-algo=sha1 --batch --no-verbose --no-armor \
--force-v3-sigs --passphrase-fd 3 --no-secmem-warning -u "%{_gpg_name}" \
-sbo %{__signature_filename} %{__plaintext_filename}
EOL
rpm --resign

But there are a lot of caveats:

  1. If you run this on RHEL or CentOS or Scientific Linux 6.x and have an RSA, rather than a DSA, GPG key, any older systems (5.x or 4.x) won’t be able to properly decode the signatures. If you have a DSA key (and only v6+), the --force-v3-sigs is enough.
  2. If you run this on RHEL or CentOS 5.x, you need to remove the “–force-v3-sigs” option from the .rpmmacros line
  3. http://technosorcery.net/blog/2010/10/10/pitfalls-with-rpm-and-gpg/ is a bit wrong – RSA keys don’t always work, DSA do (there’s apparently a RHN KB article about this, if you have support and licenses to read it. I don’t right now).
  4. RHEL / CentOS / Scientific Linux 7 won’t accept (without warning) RPMs signed with weak keys (or weak digests, in some cases, like sha1)

So, based on the systems you’re trying to build your package for, ensure the signing key you’re using and the digest algorithm you’re using, are supported across all the versions you expect to support… or build multiple RPMs, and put them in separate repos for each OS version.

04
Oct
13

Updating CentOS 4.9

I recently booted up a long-powered-off test system for a customer, and realized it was still CentOS 4.7. up2date gives one of these error messages:


Error: Bad/Outdated Mirrorlist and Baseurl

[Errno -1] Header is not complete.
Trying other mirror.

To fix, first run this to update to a newer version of yum:

for i in \
libxml2-2.6.16-12.6.i386.rpm \
libxml2-python-2.6.16-12.6.i386.rpm \
readline-4.3-13.i386.rpm \
python-2.3.4-14.7.el4.i386.rpm \
python-elementtree-1.2.6-5.el4.centos.i386.rpm \
sqlite-3.3.6-2.i386.rpm \
python-sqlite-1.1.7-1.2.1.i386.rpm \
elfutils-0.97.1-5.i386.rpm \
popt-1.9.1-32_nonptl.i386.rpm \
rpm-libs-4.3.3-32_nonptl.i386.rpm \
rpm-4.3.3-32_nonptl.i386.rpm \
rpm-python-4.3.3-32_nonptl.i386.rpm \
python-urlgrabber-2.9.8-2.noarch.rpm \
yum-metadata-parser-1.0-8.el4.centos.i386.rpm \
yum-2.4.3-4.el4.centos.noarch.rpm
do
rpm -Uvh http://archive.kernel.org/centos/4.9/os/i386/CentOS/RPMS/${i};
done

(note: the order is important for the dependencies, and the trailing “\” prevents bash from spitting out interpretation errors, but makes it readable.)

Then edit /etc/yum.repos.d/CentOS-Base.repo: comment out mirrorlist directives, and in each enabled section add baseurl=http://archive.kernel.org/centos/4.9/os/$basearch, baseurl=http://archive.kernel.org/centos/4.9/updates/$basearch, etc.

Note the change from vault.centos.org to archive.kernel.org/centos – the CentOS Vault is handling RPM headers differently than yum in CentOS 4.9 expects. The Kernel.org archive does not.




About Us

Complete networking solutions for business.
September 2017
M T W T F S S
« Aug    
 123
45678910
11121314151617
18192021222324
252627282930