MySQL Java JDBC connector TLSv1 deprecation in CentOS 8
Friday, November 12. 2021
Yeah, a mouthful. Running CentOS 8 Linux, in Java (JRE) a connection to MySQL / MariaDB there seems to be trouble. I think this is a transient issue and eventually it will resolve itself. Right now the issue is real.
Here is the long story.
I was tinkering with Databricks. The nodes for my bricks were on CentOS 8 and I was going to a MariaDB in AWS RDS. with MySQL Connector/J. As you've figured out, it didn't work! Following errors were in exception backtrace:
com.mysql.cj.jdbc.exceptions.CommunicationsException: Communications link failure
The last packet sent successfully to the server was 0 milliseconds ago. The driver has not received any packets from the server.
com.mysql.cj.jdbc.exceptions.CommunicationsException: Communications link failure
Caused by: com.mysql.cj.exceptions.CJCommunicationsException: Communications link failure
javax.net.ssl.SSLHandshakeException: No appropriate protocol (protocol is disabled or cipher suites are inappropriate)
Weird.
Going to the database with a simple CLI-command of (test run on OpenSUSE):
$ mysql -h db-instance-here.rds.amazonaws.com -P 3306 \
-u USER-HERE -p \
--ssl-ca=/var/lib/ca-certificates/ca-bundle.pem \
--ssl-verify-server-cert
... works ok.
Note: This RDS-instance enforces encrypted connection (see AWS docs for details).
Note 2: Term used by AWS is SSL. However, SSL was deprecated decades ago and the protocol used is TLS.
Taking a look on the wire:
Two details popped out instantly: TLSv1 and TLS_ECDHE_RSA_WITH_AES_256_CBC_SHA cipher. Both deprecated. Both deemed highly insecure and potentially leaking your private information.
Why would anybody using those? Don't MySQL/MariaDB/AWS -people remove insecure stuff from their software? What! Why!
Troubleshooting started. First I found SSLHandShakeException No Appropriate Protocol on Stackoverflow. It contains a hint about JVM security settings. Then MySQL documentation 6.3.2 Encrypted Connection TLS Protocols and Ciphers, where they explicitly state "As of MySQL 5.7.35, the TLSv1 and TLSv1.1 connection protocols are deprecated and support for them is subject to removal in a future MySQL version." Well, fair enough, but the bad stuff was still there in AWS RDS. I even found Changes in MySQL 5.7.35 (2021-07-20, General Availability) which clearly states TLSv1 and TLSv1.1 removal to be quite soon.
No amount of tinkering with jdk.tls.disabledAlgorithms in file /etc/java/*/security/java.security helped. I even created a simple Java-tester to make my debugging easier:
import java.sql.*;
// Code from: https://www.javatpoint.com/example-to-connect-to-the-mysql-database
// 1) Compile: javac mysql-connect-test.java
// 2) Run: CLASSPATH=.:./mysql-connector-java-8.0.27.jar java MysqlCon
class MysqlCon {
public static void main(String args[]) {
try {
Class.forName("com.mysql.cj.jdbc.Driver");
Connection con = DriverManager.getConnection("jdbc:mysql://db.amazonaws.com:3306/db", "user", "password");
Statement stmt = con.createStatement();
ResultSet rs = stmt.executeQuery("select * from emp");
while (rs.next())
System.out.println(rs.getInt(1) + " " + rs.getString(2) + " " + rs.getString(3));
con.close();
} catch (Exception e) {
System.out.println(e);
e.printStackTrace(System.out);
}
}
}
Hours passed by, but no avail. Then I found command update-crypto-policies. RedHat documentation Chapter 8. Security, 8.1. Changes in core cryptographic components, 8.1.5. TLS 1.0 and TLS 1.1 are deprecated contains mention of command:
update-crypto-policies --set LEGACY
As it does the trick, I followed up on it. In CentOS / RedHat / Fedora there is /etc/crypto-policies/back-ends/java.config. A symlink pointing to file containing:
jdk.tls.ephemeralDHKeySize=2048
jdk.certpath.disabledAlgorithms=MD2, MD5, DSA, RSA keySize < 2048
jdk.tls.disabledAlgorithms=DH keySize < 2048, TLSv1.1, TLSv1, SSLv3, SSLv2, DHE_DSS, RSA_EXPORT, DHE_DSS_EXPORT, DHE_RSA_EXPORT, DH_DSS_EXPORT, DH_RSA_EXPORT, DH_anon, ECDH_anon, DH_RSA, DH_DSS, ECDH, 3DES_EDE_CBC, DES_CBC, RC4_40, RC4_128, DES40_CBC, RC2, HmacMD5
jdk.tls.legacyAlgorithms=
That's the culprit! It turns out any changes in java.security -file won't have any effect as the policy is loaded later. Running the policy change and set it into legacy-mode has the desired effect. However, running ENTIRE system with such a bad security policy is bad. I only want to connect to RDS, why cannot I lower the security on that only? Well, that's not how Java works.
Entire troubleshooting session was way too much work. People! Get the hint already, no insecure protocols!
Wi-Fi 6 - Part 2 of 2: Practical wireless LAN with Linksys E8450
Sunday, August 15. 2021
There is a previous post in this series about wireless technology.
Wi-Fi 6 hardware is available, but uncommon. Since its introduction three years ago, finally it is gaining popularity. A practial example of sometimes-difficult-to-obtain part is an USB-dongle. Those have existed at least 15 years now. There simply is none with Wi-Fi 6 capability.
Additional twist is thrown at me, a person living in EU-reagion. For some weird (to me) reason, manufacturers aren't getting their radio transmitters licensed in EU. Only in US/UK. This makes Wi-Fi 6 appliance even less common here.
When I throw in my absolute non-negotiable requirement of running a reasonable firmware in my access point, I'll limit my options to almost nil. Almost! I found this in OpenWRT Table-of-Hardware: Linksys E8450 (aka. Belkin RT3200) It is an early build considered as beta, but hey! All of my requirements align there, so I went for it in Amazon UK:
Wi-Fi 6 Access Point: Belkin RT3200
Couple of days waiting for UPS delivery, and here goes:
This is exactly what I wanted and needed! A four-port gigabit switch for wired LAN, incoming Internet gigabit connector. 12 VDC / 2 A barrel connector for transformer. Given UK power plugs are from 1870s they're widely incompatible with EU-ones. Luckily manufacturers are aware of this and this box contains both UK and EU plugs in an easily interchangeable form. Thanks for that!
Notice how this is a Belkin "manufactured" unit. In reality it is a relabled Linksys RT3200. Even the OpenWRT-firmware is exactly same. Me personally, I don't care what the cardobard box says as long as my Wi-Fi is 6, is fast and is secure.
Illustrated OpenWRT Installation Guide
The thing with moving away from vendor firmware to OpenWRT is that it can be tricky. It's almost never easy, so this procedure is not for everyone.
To achieve this, there are a few steps needed. Actual documentation is at https://openwrt.org/toh/linksys/e8450, but be warned: amount of handholding there is low, for newbie there is not much details. To elaborate the process of installation, I'm walking trough what I did to get me OpenWRT running in the box.
Step 0: Preparation
You will need:
- Linksys/Belkin RT3200 access point
- Wallsocket to power the thing
- A computer with Ethernet port
- Any Windows / Mac / Linux will do, no software needs to be installed, all that is required is a working web browser
- Ethernet cable with RJ-45 connectors to access the access point's admin panel via LAN
- OpenWRT firmware from https://github.com/dangowrt/linksys-e8450-openwrt-installer
- Download files into a laptop you'll be doing your setup from
- Linksys-compatible firmware is at at:https://github.com/dangowrt/linksys-e8450-openwrt-installer/releases, get
openwrt-mediatek-mt7622-linksys_e8450-ubi-initramfs-recovery-installer.itb - Also download optimized firmware
openwrt-mediatek-mt7622-linksys_e8450-ubi-squashfs-sysupgrade.itb
- Skills and rights to administer your workstation to have its Ethernet port a fixed IPv4-address from net 192.168.1.1/24
- Any other IPv4 address on that net will do, I used 192.168.1.10
- No DNS nor gateway will be needed for this temporary setup
Make sure not to connect the WAN / Internet into anything. The Big Net is scary and don't rush into that yet. You can do that later when all installing and setupping is done.
Mandatory caution:
If you just want to try OpenWrt and still plan to go back to the vendor firmware, use the non-UBI version of the firmware which can be flashed using the vendor's web interface.
Process described here is the UBI-version which does not allow falling back to vendor firmware.
Step 1: Un-box and replace Belkin firmware
After plugging the Access Point to a wall socket, flicking the I/O-switch on, attaching an Ethernet cable to one of the LAN-switch ports and other end directly to a laptop, going to http://192.168.1.1 with your browser will display you something like this:
What you need to do is try to exit the out-of-box-experience setup wizard:
For the "Ethernet cable is not connected" you need to click Exit. When you think of the error message bit harder, if you get the message, your Ethernet IS connected. Ok, ok. It is for the WAN Ethernet, not LAN.
Notice how setup "did not complete succesfully". That is fully intentional. Click "Do not set up". Doing that will land you on a login:
This is your unconfigured admin / admin -scenario. Log into your Linksys ... erhm. Belkin.
Select Configuration / Administration / Firmware Upgrade. Choose File. Out of the two binaries you downloaded while preparing, go for the ubi-initramfs-recovery-installer.itb. That OpenWRT firmware file isn't from manufacturer, but the file is packaged in a way which makes it compatible to allow easy installation:
On "Start Upgrade" there will be a warning. Click "Ok" and wait patiently for couple minutes.
Step 2: Upgrade your OpenWRT recovery into a real OpenWRT
When all the firmware flashing is done, your factory firmware is gone:
There is no password. Just "Login". An OpenWRT welcome screen will be shown:
Now that you're running OpenWRT, your next task is to go from recovery to real thing. I'm not sure if I'll ever want to go back, but as recommended by OpenWRT instructions, I did take backups of all four mtdblocks: bl2, fip, factory and ubi. This step is optinal:
When you're ready, go for the firmware upgrade. This time select openwrt-mediatek-mt7622-linksys_e8450-ubi-squashfs-sysupgrade.itb:
To repeat the UBI / non-UBI firmware: This is the UBI-version. It is recommended as it has better optimization for layout and management of SPI flash, but it does not allow fallbacking to vendor firmware.
I unchecked the "Keep settings and retain the current configuration" to make sure I got a fresh start with OpenWRT. On "Continue", yet another round of waiting will occur:
Step 3: Setup your wireless AP
Déjà vu!
You have seen this exact screen before. Login (there is no password yet):
Second time, same screen but with this time there is a proper firmware in the AP. Go set the admin account properly to get rid of the "There is no password set on this router" -nag. Among all settings, go to wireless configuration to verify both 2.4 and 5 GHz radios are off:
Go fix that. Select "Edit" for the 5 GHz radio and you'll be greeted by a regular wireless access point configuration dialog. It will include section about wireless security:
As I wanted to improve my WLAN security, I steer away from WPA2 and went for a WPA3-SAE security. Supporting both at the same time is possible, but securitywise it isn't wise. If your system allows wireless clients to associate with a weaker solution, they will.
Also for security, check KRACK attack countermeasures. For more details on KRACK, see: https://www.krackattacks.com/
When you've done, you should see radio enabled on a dialog like this:
Step 4: Done! Test.
That's it! Now you're running a proper firmware on our precious Wi-Fi 6 AP. But how fast it is?
As I said, I don't have many Wi-Fi 6 clients to test with. On my 1 gig fiber, iPad seems to be pretty fast. Also my Android phone speed is ... well ... acceptable. 
For that speed test I didn't even go for the "one foot distance" which manufacturers love to do. As nobody uses their mobile devices right next to their AP, I tested this on a real life -scenario where both AP and I were located the way I would use Internet in my living room.
Final words
After three year wait Wi-Fi 6 is here! Improved security, improved speed, improved everything!
DynDNS updates to your Cloud DNS - Updated
Monday, July 12. 2021
Python 3.9 in RedHat Enterprise Linux 8.4
Tuesday, May 25. 2021
Behind the scenes: Reality of running a blog - Story of a failure
Monday, March 22. 2021
... or any (un)social media activity.
IMHO the mentioned "social" media isn't. There are statistics and research to establish the un-social aspect of it. Dopamin-loop in your brain keeps feeding regular doses to make person's behaviour addicted to an activity and keep the person leeching for more material. This very effectively disconnects people from the real world and makes the dive deeper into the rabbit hole of (un)social media.
What most of the dopamin-dosed viewer of any published material keep ignoring is the peak-of-an-iceberg -phenomenon. What I mean is a random visitor gets to see something amazingly cool. A video or picture depicting something that's very impressive and assume that person's life consists of a series of such events. Also humans tend to compare. What that random visitor does next is compares the amazing thing to his/hers own "dull" personal life, which does not consist of a such imaginary sequence of wonderful events. Imaginary, because reality is always harsh. As most of the time we don't know the real story, it is possible for 15 seconds of video footage to take months or preparation, numerous failures, reasonable amounts of money and a lot of effort to happen.
An example of harsh reality, the story of me trying to get a wonderful piece of tech-blogging published.
I started tinkering with a Raspberry Pi 4B. That's something I've planned for a while, ordered some parts and most probably will publish the actual story of the success later. Current status of the project is, well planned, underway, but nowhere near finished.
What happened was for the console output of the Linux to look like this:
That's "interesting" at best. Broken to say the least.
For debugging of this, I rebooted the Raspi into previous Linux kernel of 5.8 and ta-daa! Everything was working again. Most of you are running Raspian, which has Linux 5.4. As I have the energy to burn into hating all of those crappy debians and ubuntus, my obvious choice is a Fedora Linux Workstation AArch64-build.
To clarify the naming: ARM build of Fedora Linux is a community driven effort, it is not run by Red Hat, Inc. nor The Fedora Project.
Ok, enough name/org -talk, back to Raspi.
When in a Linux graphics go that wrong, I always disable the graphical boot in Plymouth splash-screen. Running plymouth-set-default-theme details --rebuild-initrd will do the trick of displaying all-text at the boot. However, it did not fix the problem on my display. Next I had a string of attempts doing all kinds of Kernel parameter tinkering, especially with deactivating Frame Buffer, learning all I could from KMS or Kernel Mode Setting, attempting to build Raspberry Pi's userland utilities to gain insight of EDID-information just to realize they'll never build on a 64-bit Linux, failing with nomodeset and vga=0 as Kernel Parameters to solve the problem. No matter what I told the kernel, display would fail. Every. Single. Time.
It hit me quite late in troubleshooting. While observing the sequence of boot-process, during early stages of boot everything worked and display was un-garbled. Then later when Feodra was starting system services everything fell. Obviously something funny happened with GPU-driver of Broadcom BCM2711 -chip of VideoCore 4, aka. vc4 in that particular Linux-build when the driver was loaded. Creating file /etc/modprobe.d/vc4-blacklist.conf with contents of blacklist vc4 to prevent VideoCore4 driver from ever loading did solve the issue! Yay! Finally found the problem.
All of this took several hours, I'd say 4-5 hours straight work. What happened next was surprising. Now that I had the problem isolated into GPU-driver, on IRC's #fedora-arm -channel, people said vc4 HDMI-output was a known problem and was already fixed in Linux 5.11. Dumbfounded by this answer, I insisted version 5.10 of being the latest and 5.11 lacking availability. They insisted back. Couple hours before me asking, 5.11 was deployed into mirrors sites for everybody to receive. This happened while I was investigating failing and investigating more.
dnf update, reboot and pooof. Problem was gone!
There is no real story here. In pursuit of getting the thing fixed, it fixed itself by time. All I had to do is wait (which obviously I did not do). Failure after failure, but no juicy story on how to fix the HDMI-output. On a typical scenario, this type of story would not get published. No sane person would shine any light on a failure and time wasted.
However, this is what most of us do with computers. Fail, retry and attempt to get results. No glory, just hard work.
podman - Running containers in Fedora 31+
Tuesday, November 10. 2020
openSUSE Leap 15.2 in-place upgrade
Sunday, July 12. 2020
Blog server upgrade to CentOS 8
Saturday, December 7. 2019
Since the inception of this blog back in January 2013, my weapon-of-choice has been CentOS Linux. When looking at the release chart @ https://en.wikipedia.org/wiki/CentOS#Latest_version_information it becomes obvious this is the 3rd major version of CentOS I'm running my blog on. In 2013 only version 6 was available, I must have upgraded into version 7 during 2014, and now 2019 I'm running on version 8. Given how RedHat and their organization(s) operate, the base Linux for my system is Fedora 28. See Fedora Project releases from https://fedoraproject.org/wiki/Releases.
The only motivation for me to upgrade is technology. RHEL/CentOS almost never upgrade their component versions. They do back-port any security patches even if authors of the original ones give up on their obsoleted stuff. RedHat does not. For people loving things how they are, that's a good thing. For people like me, its not that good.
Absolutely necessary things I had earlier, but lost and again have:
- HTTP/2
- For how and why this differs from HTTP/1.1 everybody else is still using, dive into Mr. Curl's book http2 explained. Its freely available @ https://http2-explained.haxx.se/content/en/
- TLS 1.3
- TLS versions 1 and 1.1 have been obsoleted. That leaves TLS 1.2 as the almost-only viable secure protocol.
- Obvious disclaimer for TLS 1.3: As of writing, it is still experimental. In reality not so much. Chrome and Firefox (among other platforms) support TLS 1.3 fully.
- Cloudflare's Head of Research Nick Sullivan is a known 1.3 enthusiast. Read his thoughts @ https://blog.cloudflare.com/rfc-8446-aka-tls-1-3/.
Other highlights:
- PHP 7.3
- My blog software runs on PHP. I upgraded 7.2, but am too scared to go for 7.4 yet.
- Native dual-stack IPv6/IPv4 networking. This is courtesy of my service provider.
- TLS 1.2 configured to not support any CBC-ciphers, for details see Why did TLS 1.3 drop AES-CBC? as an example
- Inspiration for this taken from Cipherli.st and Security/Server Side TLS on Mozilla wiki.
- Apologies for anybody using IE 11 on Windows Phone 8.1, or Safari versions 6-8 on iOS 6-9/OS X 10.9 or 10.10. You won't see this text as your devices/operating systems won't support my reasonably secure settings.
- For everybody else: Congratulations on having a decently secure device to do your Internet browsing with.
- tmux
- Terminal multiplexer, https://github.com/tmux/tmux/wiki
- Most of you just SSH into a server and be happy with it. I almost always run my sessions trough something that will keep my work safe if a disconnection occurs. To my surprise I keep bumping into sysadmins who don't either know about this or don't see this as a necessary approach.
- I've ran GNU Screen for over 25 years now. Not anymore. Uff!
- nftables (https://wiki.nftables.org/), courtesy of RHEL 8 / CentOS 8
- the new packet classification framework that replaces the existing {ip,ip6,arp,eb}_tables infrastructure
- I've ran IPchains / IPtables for 21 years now. Not anymore.
Arf!
Qualsys report on my blog now:
Nice!
Next up: CentOS Stream.
A new attempt to allow change of software versions. This will effectively detach CentOS from RHEL and gear it towards Fedora. This enables CentOS to get newer software as a rolling release Linux-distro, but keep the changes not-so-aggressive.
I won't run this yet on my blog server. This is so new at this point, but I'll have it running on a devel-box.
New Weather Station - Davis Vantage Vue - Part 2 of 2: Linux installation
Tuesday, November 5. 2019
This is a part two out of two on my Davis Vantage Vue weather station installation story. Previous part was about hardware installation.
Datalogger expansion
The indoors console has a slot for expansion devices. I went for the RS-232 WeatherLink datalogger expansion:
RS232 Cabling
The datalogger has a very short cable and a RJ-11 connector on the other end. The obvious good thing is the common availability of telephone extension cords to get past the 5 cm cable lenght of the datalogger. A regular landline telephone typically has such RJ-11 connectors in it and what I did was to get an inexpensive extension cord with suitable lenght.
For computer connectivity, with datalogger box has a blue RJ-11 to RS232 converter. The four connected pins of the converter are as follows:
RS232 and Linux
Most computers today don't have a RS232-port in them. To overcome this, years ago I bought a quad-RS232-port USB-thingie:
If you look closely, Port 1 of the unit has a DIY RS232-connector attached into it. That's my Vantage Vue cable connected to the indoors console. Also, note the lack of the blue RJ-11 to RS232 converter unit. I reverse engineered the pins and soldered my cable directly to a D-9 connector to get the same result.
Now the hardware parts is done. All the connectors are connected and attached to a PC.
Software
Half of the work is done. Next some software is needed to access the data in the data logger.
For Windows and macOS
Those not running Linux, there is WeatherLink software freely available at https://www.davisinstruments.com/product/weatherlink-computer-software/. It goes without saying, the software is useless without appropriate hardware it gets the inputs from. I never even installed the software, as using a non-Linux was never an option for me. So, I have no idea if the software is good or not.
For Linux
As you might expect, when going to Linux, there are no commercial software options available. However, number of open-source ones are. My personal choice is WeeWX, its available at http://www.weewx.com/ and source code at https://github.com/weewx/weewx.
Install WeeWX:
- Download software:
git clone https://github.com/weewx/weewx.git - In git-directory, create RPM-package:
make -f makefile rpm-package SIGN=0 - As root, install the newly created RPM-package:
rpm --install -h dist/weewx-3.9.2-1.rhel.noarch.rpm - That's it!
Configure WeeWX:
- (dependency) Python 2 is a requirement. Given Python version 2 deprecation (see https://pythonclock.org/ for details), at the time of writing, there is less than two months left of Python 2 lifetime, this project really should get an upgrade to 3. No such joy yet.
- (dependency) pyserial-package:
pip install pyserial - Run autoconfig:
wee_config --install --dist-config /etc/weewx/weewx.conf.dist --output /etc/weewx/weewx.conf- In the array of questions, when being asked to choose a driver, go for
Vantage (weewx.drivers.vantage)
- In the array of questions, when being asked to choose a driver, go for
- Inspect the resulting
/etc/weewx/weewx.confand edit if necessary:- Section
[Station]week_start = 0
- Section [StdReport]
HTML_ROOT = /var/www/html/weewx
- (optional for Weather Underground users), Section
[StdRESTful]- Subsection
[[Wunderground]] - Enter station name and password
enable = true
- Subsection
- Section
- Configuring done!
Now you're ready (as root) to start the thing with a systemctl start weewx.
On a working system, you should get LOOP-records out of weewxd instantly. After a little while of gathering data, you should start having set of HTML/CSS/PNG-files in /var/www/html/weewx. It's a good idea to set up a web server to publish those files for your own enjoyment. This is something you don't have to do, but I strongly advice to enable HTTP-endpoint to your results. I will immensly help determining if your system works or not.
One reason I love having a weather station around is to publish my data to The Net. I've been tagging along with Weather Underground for years. There has been good years and really bad years, when Wunderground's servers have been misbehaving a lot. Now that IBM owns the thing, there has been some improvements. The most important thing is there is somebody actually maintaining the system and making it run. Obviously, there has been some improvements in the service too.
Aggregation problem
When I got my system stable and running, I realized my wind data is flatline. In nature it is almost impossible for not to be windy for multiple days. I visually inspected the wind speed and direction gauges, they were working unobstructed. However, my console did not indicate any wind at all.
After multiple days of running, I was about to give up and RMA the thing back to Davis for replacement, the console started working! That was totally unexpected. Why: 1) the system did not work, 2) the system started working without any actions from me. That was the case for me. As the problem has not occurred since, it must have been some kind of newness.
What next
Now that everything is up, running and stays that way. There's always something to improve. What I'm planning is to do is pump the LOOP-records to Google BigQuery.
As a side note: when new information is available, it will be emitted by weewxd as LOOP: <the data here> into your system logs. However, that data isn't used. After a period of time, your Vantage Vue will aggregate all those records into a time-slot record. That will be used as your measurement. Since those LOOP-records are simply discarded I thought it might be a good idea to base some analytics on those. I happen to know Google BigQuery well from past projects, all that I need is to write a suitable subsystem into weewx to pump the data into a correct place. Then it would be possible to do some analytics of my own on those records.
Python script for creating Linux iptables port forwarding rules based on an YAML-file
Saturday, September 14. 2019
Notes on running openSUSE on a MacBook Pro
Sunday, August 18. 2019
This one is related to a previous post of mine Installing openSUSE Leap 15.1 into a MacBook Pro with encrypted root drive.
kworker CPU-hog
Symptoms:
CPU-load is high. Top consumer is kworker:
top - 11:16:47 up 6 min, 4 users, load average: 0.93, 0.70, 0.36
Tasks: 248 total, 2 running, 246 sleeping, 0 stopped, 0 zombie
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
58 root 20 0 0 0 0 R 73.33 0.000 4:29.74 kworker/0:
What a kernel worker (or kworker) is, a sensible explanation can be found from Kworker, what is it and why is it hogging so much CPU?. In this case, high CPU on a kworker is a symptom, not the fault.
This same issue is faced by multiple users on different hardware, for example Kworker is at 100% - I think I've tried everything!. Also Fedora Linux users on MBP are suffering from the same Bug 1192856 - ACPI Interrupt storm causes high kworker CPU usage.
Checking some kernel performance counters (more about those, see perf: Linux profiling with performance counters), perf-report:
Children Self Command Shared Object Symbol
+ 50.51% 0.01% swapper [kernel.kallsyms] [k] cpu_startup_entry
+ 41.67% 0.31% swapper [kernel.kallsyms] [k] acpi_hw_read_port
+ 39.99% 39.99% swapper [kernel.kallsyms] [k] acpi_os_read_port
+ 37.62% 0.00% kworker/0:2 [kernel.kallsyms] [k] ret_from_fork
Something really fishy is going on with acpi_os_read_port. To get the amount of interrupts generated, most people run a simple grep . /sys/firmware/acpi/interrupts/*, but not me. My solution is to do a simple(?) Perl one-liner wrapped here on multiple lines for readability:
perl -ne 'next if (!/^\s*(\d+)\s+/);
next if (!$1);
$intrs{$ARGV}=$1;
END {foreach (sort {$intrs{$b} <=> $intrs{$a}} keys(%intrs)) {
printf("%s: %d\n", $_, $intrs{$_});
};
}' /sys/firmware/acpi/interrupts/*
It will output interrupt counts. On my system, the counters are something like this:
/sys/firmware/acpi/interrupts/gpe_all: 4695534
/sys/firmware/acpi/interrupts/sci: 4694806
/sys/firmware/acpi/interrupts/gpe06: 4694582
/sys/firmware/acpi/interrupts/gpe17: 940
If you're really interested in what's happening inside ACPI and how General Purpose Events (GPE) work, read the document ACPI in LinuxArchitecture, Advances, and Challenges.
Anyway, it looks like ACPI event 06 is firing a lot. Way too lot keeping kworker busy handling the interrupts. This interrupt-handling on the other hand, takes a lot of CPU-power making your system run hotter than expected.
Fix:
As root, a simple echo disable > /sys/firmware/acpi/interrupts/gpe06 will do the trick. Resulting in CPU cooling down:
The effect is instantaneous and easy to notice. To persist this setting over reboots, things get bit more trickier. A lot of people suggest putting the above disable into crontab to be run on a @reboot. My opinion is clear: this is a system problem, system fix and needs to be addressed by system, not user. The @reboot-magic doesn't work in /etc/crontab, only on user's crontab-file.
A very good alternative is to go with acpi_mask_gpe-kernel parameter. It is documented in The kernel’s command-line parameters. And docs state "This facility can be used to prevent such uncontrolled GPE floodings". Nice! Exactly what I need. The tricky part is to edit /etc/default/grub and add acpi_mask_gpe=0x06 into GRUB_CMDLINE_LINUX-setting.
That's not all. Simply having the setting in a file won't change a anything yet. To make the new setting stick, on most Linuxes you would run update-grub. Some smart brain chose not to implement that helper into openSUSE, so that's not an option. To achieve the same, go for a: grub2-mkconfig -o /boot/grub2/grub.cfg
Now you're all set. Reboot and confirm. Of course you didn't fix the actual problem with ACPI, you simply made kernel ignore the interrupt-flood. You may want to settle for that at this point. I did file a bug report about this (Bug 1146023), but it seems the problem is limited to a specific set of MBPs and Fedora kernel guys are having hard time reproducing the issue. So, I'm not expecting SuSE guys to fix this anytime soon either.
Touchpad scroll direction
In 2011 when OS X 10.7 Lion was released, Apple made a drastic change on how touchpad (or how Apple calls it: Trackpad) scrolling works. Personally I totally understand this new logic, they wanted the scrolling user experience to be exactly the same and you would do on your phone's touch screen. You place your finger on the screen and pull the finger to the direction you want the screen to scroll. Before macOS 10.7 (note: the name change from OS X to macOS happened on 10.12 sierra) touchpad scrolling was targeted for the scrollbar, not to the actual content like on your touchscreen effectively reversing the direction.
So, as a macOS / iPad user, I want my scrolling to happen correctly, not the Windows way. To change, navigate to Touchpad settings and reverse the vertical scrolling:
Reversing the reverse makes your head spin, but there is a logic there.
Touchpad gestures
On any typical use case, I would use a mouse. As in real external device, not the skin-on-my-fingertips damaging trackpad. For reasons unknown to me, it literally hurts to use those trackpads for too many hours. So, I normally steer away from them. It looks like I'm pretty alone with this one, but all I can do is to complain and use a mouse.
Typical this you may want to do:
- Right-click: On macOS, hold Ctrl and tap is a right click. This won't work on a Linux. You need to do the two-finger-tap for right-clicking.
- Scrolling: You can go with keyboard or keep dragging the scrollbars, that's ok. Doing a two-finger scroll with the touchpad, however, is much faster and less error prone.
- Zoom: Pinch won't work on a Linux, you need to use the actual zoom-control of whatever application you want to zoom
All the Apple's gestures are documented in HT204895 suppor article Use Multi-Touch gestures on your Mac, but as already mentioned, not all gestures are supported on a Linux.
Something else?
Nothing more pops into my mind. If you have anything to comment, please do so.
Installing openSUSE Leap 15.1 into a MacBook Pro with encrypted root drive
Thursday, August 15. 2019
I happen to have a spare 3rd generation 15" MBP. It is late 2013 model, used for years but recently just gathering dust. Back in the days, when the laptop was shiny and new, it was a serious piece of hardware and cost north of 2000 €. Even today, a fully working second hand 15" late 2013 would be priced around 25% of the original value. Couple years later Apple started manufacturing MBPs with faulty keyboards, faulty touch bar (the really weird top row) and faulty batteries. I'm in luck, as my unit pre-dates all of those. Yes, there are also some fatures missing which I'd love to have, fingerprint reader, USB-C -port and improved display to mention the most important ones.
Since the dust-gathering, I decided to see how a fresh Linux installation would go. Back in the days, I had a plastic MacBook running Linux. Given 32-bits and not-so-powerful CPU, I recycled it at the point I got tired of waiting for pages to load too long on Firefox. The amount of JavaScript modern web pages run is staggering. All that requires huge amounts of CPU to parse, process and run. Old low-end models simply won't cope making web browsing nearly impossible. So, I didn't have a Linux-Mac for couple years and now I wanted one.
Known facts
This I could figure out before doing any real installing:
- Installation can be done. openSUSE docs describe the process @ https://en.opensuse.org/SDB:Installation_on_a_Mac
- The built-in Broadcom WiFi isn't properly supported by most Linuxes, that includes openSUSE
- Debian Linux wiki @ https://wiki.debian.org/MacBook/Wireless states:
"MacBooks before the late 2008 models use Atheros chipsets, while late 2008 are now using Broadcom 4322 chipset." - StackExchange article @ https://unix.stackexchange.com/a/175813/241396:
"Broadcom 4360 actually comes with either of two distinct chips, 14E4:4360 and 14E4:43A0. There is no driver in Linux for the first one, while wl is an appropriate driver for the second one." - Arc Linux wiki @ https://wiki.archlinux.org/index.php/broadcom_wireless#broadcom-wl only states the fact wl-driver not being open-source.
- It is unclear which exact chipset is in the Mac, but it seems not to be properly supported in Linux.
- Debian Linux wiki @ https://wiki.debian.org/MacBook/Wireless states:
- My MBP doesn't have Ethernet RJ-45 port
- The existing USB-ports or Thunderbolt-ports could be used to host a network adapter, either wired or wireless, but I choose not to go down that path. Having an USB-dongle stick out of the Mac isn't something I'm willing to do.
- There is a good chance, that a Linux-driver for the built-in Broadcom WiFi exists. The existing driver can be downloaded from the net, but require some trickery to install during openSUSE installer.
- In macOS, drives are encrypted. In Windows using BitLocker is a good option to keep your data secure. Linux supports drive encryption as well.
Prerequisites
- A MacBook Pro with SSD full of data you don't care about. Also emptied SSD-drives will do.
- openSUSE Leap 15.1 ISO-image from https://www.opensuse.org/
- I personally will go for the full DVD, but it is not absolutely necessary. Small network installer will work as well.
- Skills to create a bootable USB-stick from the install image
- Your wireless network credentials and skills to connect a new device into it
- Two USB-sticks:
- One to boot openSUSE installer from. This stick will contain installation image and cannot be easily written into and must be attached to the Mac during installation.
- Second to transfer the WiFi-driver files
- Broadcom Linux wl-driver from http://download.opensuse.org/repositories/home:/Sauerland:/hardware/openSUSE_Leap_15.1/x86_64/
- Installer will need file
broadcom-wl-kmp-default-6.30.223.271_k4.12.14_lp151.27-lp151.121.1.x86_64.rpm. - Store the file into the second USB-stick.
- Installer will need file
- Since kernel-default -package has an update, the above RPMs needed for installer won't work in your installed openSUSE. To fix this, you will need also Broadcom Linux wl-driver from http://download.opensuse.org/repositories/home:/Sauerland:/hardware/openSUSE_Leap_15.1_Update/x86_64/
- Your installed openSUSE will need two files
broadcom-wl-6.30.223.271-lp151.121.4.x86_64.rpmandbroadcom-wl-kmp-default-6.30.223.271_k4.12.14_lp151.28.10-lp151.121.4.x86_64.rpm. - Store these files into the second USB-stick.
- Your installed openSUSE will need two files
- To make your life easier, having access to another computer while doing the install will help the process greatly. Especially, if you need to get a newer version of the Broadcom-drivers.
- That's it!
Disclaimer for RPM-files:
Those versions of files are valid at the time of writing. Both the install ISO and installed & updated openSUSE Linux are likely to eventually get updated and the versions of those files WILL change. Keep reading! There are instructions for you on how to compensate for updated versions. However, download locations will not change for openSUSE 15.1. The locations won't be valid for 15.2, but I'm sure you got that already.
Step 1: Prepare openSUSE 15.1 install
As instructed in https://en.opensuse.org/SDB:Installation_on_a_Mac, copy the installer ISO-file to the USB-stick. On macOS and Linux, figuring out the exact output drive can be bit tricky, so be careful with that. On Windows, the best utility for USB-drives is Rufus. You can get it from https://rufus.ie/.
While at it, download the required driver from above URLs pointing to Sauerland repository. Both versions are eventually needed, so just download them.
Note: for installer, only broadcom-wl-kmp-default -package is needed. We can safely ignore any RPM dependencies, as it is not possible to actually install the package. For installed Linux both files are needed, they have a dependency between packages. More about that when you're actually tinkering with the files.
Step 2: Boot the Mac into openSUSE installer
Plug in the installer USB (the other USB isn't needed yet). Power up the Mac. When you hear the boot-sound, press and hold the Alt/option-key. When you see the boot-menu, you can release the key. Displayed on your screen, there is a boot drive selection menu:
Go for the USB UEFI, that will start openSUSE installer.
Old geezers like me will remember the times when Macs required special bootloaders and tons of tinkering just to get the USB-boot to work. For past years, thanks to UEFI, USB-booting a Mac isn't any different than booting a PC.
Step 3: Establish facts
When the graphical installer launches, you'll see the license agreement screen:
On top-right corner of that screen is also keyboard map choice. This is your chance to make sure your keys will work correctly on the GUI. This won't affect the text-console.
Switch to text-mode console
There are two critical pieces of information needing to be confirmed to get the WiFi-driver working. Since the openSUSE installer is running in graphical mode, you need to switch to text-mode. In a Linux, the console can be chosen with Alt-<function key> -combo. When graphical interface is running, key combo is Ctrl-Alt-<function key>. To make things even trickier, in a Mac <function key>s require the actual Fn-key to be pressed, so ultimately you'll need to press and hold Fn-Ctrl-Alt and then tap F2-key to go for a text-console.
Confirm kernel version
On a bash-prompt, check the exact Linux kernel version with a:
cat /proc/version
As a result, my install image will display:
Linux version 4.12.14-lp151.27-default
In Linux, all device modules are stamped with the specific kernel version they'll work with, so it is imperative to get the Broadcom driver for that exact kernel version.
Confirm broadcom WiFi hardware
On a bash-prompt, get a listing for wireless network devices. List PCI-devices with class 280, that's for all network controllers (listing is at https://pci-ids.ucw.cz/read/PD/), as already established, there shouldn't be too many of those:
lspci -d '::280' -vv -n
On my Mac, the response is:
02:00.0 0280: 14e4:43a0 (rev 03)
Remember from above: "Broadcom 4360 actually comes with either of two distinct chips, 14E4:4360 and 14E4:43A0. There is no driver in Linux for the first one, while wl is an appropriate driver for the second one." So, this is the confirmation we needed. wl is the driver needed and we can move forward with installing the driver.
Step 4: Install the Broadcom driver
This screenshot depicts my process:
Now that you know the kernel version (4.12.14_lp151.27 in my installer) and copied the downloaded RPMs into the other USB-stick, plug it in. Remember: the installer USB is needed and cannot be removed.
Determine the correct device and mount it
In the above screenhot, the USB-stick having the drivers is /dev/sdb. First I checked out which storage devices Linux found. There are number of those for: Apple SSD, installer USB and driver USB. Since I knew, my driver USB is a Kingston Datatraveller, displaying the block-device partition table by running parted /dev/sdb print helped to identify the correct drive.
I mounted the USB-stick with a:
mkdir /tmp/usb
mount /dev/sdb1 /tmp/usb
Locate the files from the USB-stick.
Unpack the driver RPM
At this point, the Broadcom RPMs versioned for openSUSE installer kernel need to be popped open. The ones for your running Linux will install as-is, so no need to unpack them. It would be easy & smart to just install the RPM, but given Linux installer, it is not possible. Most of the mounted filesystems in an installer are SquashFS filesystems (more info from https://en.wikipedia.org/wiki/SquashFS) and are read-only. Well... you CAN alter a SquashFS, but it is more complicated than just running couple commands to extract the driver from RPM and injecting it into a running system.
To unpack an RPM, in a temporary directory run following (this is a single-liner, I just wrapped it into two lines for readability):
rpm2cpio broadcom-wl-kmp-default-6.30.223.271_k4.12.14_lp151.27-lp151.121.1.x86_64.rpm \
| cpio -ivd
This will create any subdirectories and files into the current directory where cpio-command is run. That's why a temporary location is suggested.
Install wl-driver
As already established, Broadcom on Linux is a messy business. And this Broadcom chip in your MBP is especially messed up. So beware! Trickery ahead.
The other file you won't be needing during installation, but you will for your running openSUSE would contain a file etc/modprobe.d/50-broadcom-wl-blacklist.conf. This blacklisting mechanism exists in Linux to prevent device-modules from being loaded when a particular device is being used. Now, at the point you're running the commands on your Linux-installer its too late. Your installer already loaded bunch of device drivers you won't be needing. So, they must go. Contents of the file suggests what to remove:
# modules blacklisted for broadcom-wl
blacklist bcm43xx
blacklist ssb
blacklist b43
blacklist ndiswrapper
blacklist brcm80211
blacklist bcma
blacklist brcmsmac
You can run command rmmod for each of those, but I choose to go for a for-loop:
for module in bcm43xx ssb b43 ndiswrapper brcm80211 bcma brcmsmac; do
rmmod $module
done
In my Mac, at least modules b43, ssb and bcma were loaded.
Warning: If you DON'T rmmod conflicting modules, your wl-module will NOT work.
Finally, it's time to go for the good driver. When you unpacked the driver RPM, it contains a subdirectory. My driver is at lib/modules/4.12.14-lp151.27-default/updates. Go there and run:
insmod wl.ko
There is zero feedback for your command. To get an indication of success, run iwconfig. It will display something like:
wlan0 IEEE 802.11 ESSID:off/any
Mode:Managed Access Point: Not-Associated Tx-Power=200 dBm
Retry short limit:7 RTS thr:off Fragment thr:off
Encryption key:off
Power Management:off
When you see an existing wlan0, you're golden! Now you're good to go forward with graphical installer.
Step 5: Install openSUSE Leap 15.1
First, go back to graphical installer. The key combination for that is:
Press and hold Fn-Alt and then tap F7-key
Add your Mac's Broadcom to Linux
In the graphical installer, go Next from license screen and witness a blank Network Settings screen. The installer wasn't able to determine any operational network interfaces in your system. Not to worry, now that your Broadcom is on-line, you can add it. Click Add, and add a Wireless device using Kernel module wl:
Moving forward with Wireless Network Card Configuration, a screen to connect to your WiFi will appear:
Scanning for any available networks will provide you a drop-list of found networks. Your own network needs to be in that list. Just add your authentication details to the dialog and you're good to go.
Instruct openSUSE installer to encrypt your root drive
(This is completely optional, you don't NEED to have an encrypted hard-drive. I just love doing stuff like this.) When your installer wizard will reach Suggested Partitioning, that's your cue. Going for encrypted root partition is actually surprisingly easy on openSUSE:
Go to Expert Partitioner using the suggested partition as a baseline for your custom partitioning. The only thing you need to change is to enable encryption on your root partition. Find mount point / and Edit it. Just click Encrypt Device and choose a wise password for your encryption. That's it. Then you can Accept the partition plan and move forward.
Add an user to new Linux
For security reasons, I'll always separate my administrators and regular users. What level of security you're targeting for is completely up to you. I, personally won't use Automatic Login and won't use same passwords for super-user and regular users. The Local Users dialog will look like this:
Hang tight, you're almost there.
Install!
When you reach the Installation Settings, you're about to start destroying any previous data on your Mac and start installing the Linux:
Enabling SSH and drilling a hole for it into firewall is up to you. Also, the choice of tooling for network management is up to you. I do dislike NetworkManager and obviously choose to drop it and go with wicked. If you want to learn more, go read wicked FAQ at https://github.com/openSUSE/wicked/wiki/FAQ.
When you click Install in the Confirm Installation -dialog, the ball starts rolling. If you missed the previous warning, before this point your data is still intact in the drive, after this point it won't be anymore. This is the part where you can go grab a cup of coffee:
At the successful end, your system will automatically reboot. When the installer reboots the computer, you can remove the installer USB. It won't be needed anymore.
Step 6: Reboot into your newly installed Linux
If you went with encrypted partition, GRUB will require the password to decrypt the drive to load Linux kernel:
Hint: If you already didn't, just yank out the installer USB.
The "funny" thing about encrypted root partition is: when your kernel is loaded by GRUB, kernel will need the same password again!
Yes, punch in the same password again to reach login-screen.
Add your Mac's Broadcom to Linux - Again!
Once logged in all the hard work you did with the Broadcom-driver are lost. Your newly installed openSUSE will NOT have the Broadcom-driver installed. What! What! What?
But wait! This time installing is much easier. You actually CAN simply install the RPMs from the other USB-stick. On an installed Linux, the system can be written into.
Confirm the kernel version (again) and notice how it is different. My system has:
Linux version 4.12.14-lp151.28.10-default
Just like before, figure out which one the device is (with a great likelihood, it is not the same device anymore) and mount it. Once mounted, as root you can do something like:
rpm --install -h *.rpm
Installing the RPMs won't make your network operational. In a Linux, there are always ways to avoid rebooting, but in this instance I found it simply easier to go the Windows-way and reboot. You can do that at this point. Alternatively if you don't just yet need a network, you can also remove the second query for encryption password and then reboot to fix two problems on a single reboot.
Remove the 2nd password prompt
This is something I wouldn't do on a shared machine. Since I'm almost certain, nobody else will use my laptop, in this instance I'm willing to add ease of use with cost of potentially losing my security. If a logged-in user can access the encryption key-file, my encryption security is gone! Before allowing a second encryption key to be used, the plain-text password is not stored anywhere. To keep my system still secure, I need to make sure the permissions for the file are correctly set and also keep any malware or unwanted software from stealing my precious key-file.
Process for setting up encrypted root in openSUSE is described well in https://en.opensuse.org/SDB:Encrypted_root_file_system. As instructed there, create a keyfile, add the keyfile as a valid key into LUKS and secure the setup:
For relaxing my system's security, the general idea is to load some random bytes from /dev/urandom and create a random-content key-file. The docs are using 1024 bytes, I went double that for a 2048 byte keyfile. In my sequence, I first create an empty file, then lock all other users out of it and then populate the file with random bytes:
touch /.partition.35d5293d-6fe0-4eaf-987f-4f2eca8ca6d7.key
chmod 600 /.partition.35d5293d-6fe0-4eaf-987f-4f2eca8ca6d7.key
dd if=/dev/urandom \
/.partition.35d5293d-6fe0-4eaf-987f-4f2eca8ca6d7.key \
bs=1024 count=2
Then this newly created filename needs to be added into two places. First /etc/crypttab. After setup, there are only two columns. As instructed in the documentation, a third column needs to be added. Note: This single line having three columns is split into three lines here for readability. If you fail to have this as a single line, your system will fail to reboot:
cr_ata-APPLE_SSD_SM0256F_S1K4NYBF401257-part3
UUID=7e49a147-b626-41cc-bc89-49e8a13b9cb0
/.partition.35d5293d-6fe0-4eaf-987f-4f2eca8ca6d7.key
Second into initRD by creating a new Dracut configuration segment /etc/dracut.conf.d/99-root-key.conf:
echo -e 'install_items+=" /.partition.35d5293d-6fe0-4eaf-987f-4f2eca8ca6d7.key "' \
| sudo tee --append /etc/dracut.conf.d/99-root-key.conf
Yet again, to keep your system secure, make sure nobody else can access your /boot-stuff:
chmod -R go= /boot/
To have this Dracut segment have any effect, you need to run mkinitrd as root. If you want more information about initRDs or Dracut, go to https://dracut.wiki.kernel.org/.
Now, you're ready to reboot the Linux. Both WiFi and password setups are done.
Step 7: Reboot and verify setup
You will need to enter the encryption password for GRUB, no change there. However, your Linux should go directly to login-screen. If it doesn't, you failed somewhere.
Also, after a login, your WiFi should work. To confirm, go to a prompt:
Since installer saved your wireless connection details, now you have network connectivity also in your installed Linux!
Step 8: Finalize
Remember the part where your installer used a different version of Linux kernel than your ready system? Kinda hard to not remember it, right. Now some additional work is needed to keep your system working:
As root, run this in a bash-prompt (again, a single-liner):
zypper addrepo \
https://download.opensuse.org/repositories/home:Sauerland/openSUSE_Leap_15.1_Update/home:Sauerland.repo
The setup information is available at https://software.opensuse.org/download.html?project=home%3ASauerland&package=broadcom-wl.
Finally: Go to YaST2 and use the GUI to inspect all the software repositories your system has. It must have Sauerland's repo enabled for the Broadcom driver to get updates as openSUSE's kernel-default -package does.
Step last: All done
That's it. Now your Mac is successfully running an openSUSE 15.1 Leap Linux.
Enjoy!
systemd setting overrides
Monday, June 17. 2019
In couple past posts of mine, I've instructed people to do alter systemd service descriptions. Examples from Randomness in computers and Handling /run with systemd. As we all know, systemd is quite a beast. It just keeps on giving complexity as we stumble along trying to grasp all of it.
For example:
To add more command arguments to rngd.service, you can do what I instructed earlier. You can copy /usr/lib/systemd/system/rngd.service to /etc/systemd/system/, edit the file and do a systemctl daemon-reload to update the changes.
If you want to choose more systemd'ic (Python developers always say "pythonic") way, you can do a:
systemctl edit rngd
This command will open a text-editor. The editor is your default one from EDITOR environment variable, or if you haven't picked one, a default will be chosen for you by your favorite Linux distro vendor.
What you should make the text file contain, is following three (3) lines:
[Service]
ExecStart=
ExecStart=/sbin/rngd -f --rng-device=/dev/TrueRNG --fill-watermark=4000
Those lines will re-define service description. First by clearing out the previous value(s) and then defining a new one. On save, this setting will take automatically effect and you don't have to do anything else. Also, this setting is persisted in a file called /etc/systemd/system/rngd.service.d/override.conf. Of course, you can create that particular file yourself, but then the familiar systemctl daemon-reload is needed to make your changes stick.
Nice and easy!
Adding a source of randomness to a Linux
Monday, June 3. 2019
Randomness in computers
You don't need to know much about computers to understand, that computers cannot do random things. Yes, all programming languages and libraries do offer you a rand()-function to emulate randomness. However, the resulting output will follow the carefully crafted programming implementing this "randomness". The most trivial pseudo-random functions will merely provide a sequence of numbers appearing random, but this sequence can be reset to start from beginning making the "randomness" predicatable. That's not really very random, huh!
Improved randomness in computers
To be fair, there does exist improved pseudo-random algorithms which take their initial seed-values from something volatile (time is one such volatile parameter) making the quality of randomness better. Still, even high-quality pseudo-random algorithm is just complex sequence of operations, which will produce duplicate results on same input values. Sometimes its just very tricky to craft a situation where all of the input values would match.
If somebody is capable of doing that, your randomness changes into predictability. Read the story of Dual_EC_DRBG on Wikipedia https://en.wikipedia.org/wiki/Dual_EC_DRBG. When you're generating your precious private keys, you don't want anybody (including NSA) to be able to guess what you have there.
Random source in Linux
Since a proper random source is something every single user, developer and sysadmin would love to have, the problem has been approached on your Linux by authors of the operating system. An excellent description can be found from Wikipedia article https://en.wikipedia.org/wiki//dev/random#Linux. Briefly put, your Linux will collect environmental entropy from number of sources (including human interaction with keyboard and mouse) to a pool, which can then be used to produce naturally random numbers. It actually works very well, the quality of randomness is top-notch.
Obvious problem with this approach is, that you cannot pull too many random numbers out of this source without exhausting it. The fix is to keep typing something while moving your mouse (not a joke!) to generate entropy for the random source. This will eventually help fill the entropy pool and /dev/random will spit couple bytes more.
Those users who have exhausted their /dev/random on an idling rack server without a console keyboard, mouse and video know that it takes painfully long for the entropy pool to fill. A busy server doing something will be able to fill the pool much faster.
A real random source
If you need a real proper random source, which works without human intervention and can provide really good randomness as a stream, there are possibilities on hardware. I know of two good ones, Simtec Electronics Entropy Key and ubld.it TrueRNG Hardware Random Number Generator.
Note: if you consider getting one, get the TrueRNG version 3 (http://ubld.it/truerng_v3). Its just that I have the 1st gen version at hand and haven't found the reason to upgrade.
My TrueRNG looks like this:
It is essentially an USB-stick.
Linux lsusb info essentially identifies it as a Microchip (vendor ID 0x04d8) manufactured USB-device (with ID 0xf5fe) providing RS-232 communications:
Bus 002 Device 009: ID 04d8:f5fe Microchip Technology, Inc.
Device Descriptor:
bLength 18
bDescriptorType 1
bcdUSB 2.00
bDeviceClass 2 Communications
bDeviceSubClass 0
bDeviceProtocol 0
bMaxPacketSize0 8
idVendor 0x04d8 Microchip Technology, Inc.
idProduct 0xf5fe
bcdDevice 1.00
iManufacturer 1 ubld.it
iProduct 2 TrueRNG
iSerial 0
bNumConfigurations 1
Configuration Descriptor:
bLength 9
bDescriptorType 2
wTotalLength 0x0043
bNumInterfaces 2
bConfigurationValue 1
iConfiguration 0
bmAttributes 0x80
(Bus Powered)
MaxPower 100mA
Interface Descriptor:
bLength 9
bDescriptorType 4
bInterfaceNumber 0
bAlternateSetting 0
bNumEndpoints 1
bInterfaceClass 2 Communications
bInterfaceSubClass 2 Abstract (modem)
bInterfaceProtocol 1 AT-commands (v.25ter)
iInterface 0
CDC Header:
bcdCDC 1.10
CDC ACM:
bmCapabilities 0x02
line coding and serial state
CDC Union:
bMasterInterface 0
bSlaveInterface 1
CDC Call Management:
bmCapabilities 0x00
bDataInterface 1
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x81 EP 1 IN
bmAttributes 3
Transfer Type Interrupt
Synch Type None
Usage Type Data
wMaxPacketSize 0x000a 1x 10 bytes
bInterval 1
Interface Descriptor:
bLength 9
bDescriptorType 4
bInterfaceNumber 1
bAlternateSetting 0
bNumEndpoints 2
bInterfaceClass 10 CDC Data
bInterfaceSubClass 0
bInterfaceProtocol 0
iInterface 0
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x02 EP 2 OUT
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
Endpoint Descriptor:
bLength 7
bDescriptorType 5
bEndpointAddress 0x82 EP 2 IN
bmAttributes 2
Transfer Type Bulk
Synch Type None
Usage Type Data
wMaxPacketSize 0x0040 1x 64 bytes
bInterval 0
Device Status: 0x0001
Self Powered
And by looking at /dev/, there is a /dev/ttyACM0. That's how udevd will populate a CDC-device when it sees one.
How is this a "true" random source?
Oh, that's easy. The device will produce a random 0 or 1 bit constantly when its on. Or to be precise, there is an internal algorithm producing those based on a constant flow of electrons on a transistor PN-surface. The exact phenomenon is called avalance effect or avalance breakdown. For those who can do electronics, there is a good explanation about this in Difference Between Avalanche Breakdown and Zener Breakdown (I borrowed the visualisation pic from above link).
To (over)simplify that, in a carefully constructed electronic circuit, inside a transistor an electron may or may not be emitted on the other side of a semiconducting surface. The occurrence is as random as it can be in nature. Other circuitry will detect this random flow of electrons (or lack of flow) to produce ones and zeros.
What makes this a really good for randomness, as it is well established that this avalance of electrons will happen. Also, it will happen often enough to produce a stream of events. It's just that we don't know exactly WHEN the avalance of electrons will happen. If you time-slice this to slots, a slot can be empty (no avalance) or full (electrons avalanching).
Linux tweaking:
udev
Anybody having multiple devices in their Linuxes knows, that you really cannot control which device name some specific device will get on reboot. To overcome that, udevd can be instructed to do things when it sees a device. My rules for TrueRNG include setting it to highest possible speed and creating a symlink-device so, that I can point to a known source of random. Also, I'm loosening access to that source of randomness to any users belonging to dialout-group. If I wouldn't do that, only root would have access to this fine random-source.
My /etc/udev/rules.d/99-TrueRNG.rules contains:
SUBSYSTEM=="tty", ATTRS{product}=="TrueRNG", SYMLINK+="TrueRNG", RUN+="/bin/stty raw -echo -ixoff -F /dev/%k speed 3000000"
ATTRS{idVendor}=="04d8", ATTRS{idProduct}=="f5fe", ENV{ID_MM_DEVICE_IGNORE}="1", GROUP="dialout", MODE="0664"
If you want to take your random-device for a spin, you can do something like:
dd if=/dev/TrueRNG of=random.bytes bs=64 count=1024
That would create a file of 64 KiB containing very very random bytes. In theory you can just cp data out of the character device, but since it has an infite flow, you'll need to cut it at one point.
rngd
Remember the part I said earlier about Linux using your keypresses and mouse movements as entropy source for randomness. Even with the USB-stick popped into a PC, that still remains the case. What needs to be done next is to offer a helping hand to the Linux kernel and make sure the entropy pool is always full.
My Fedora has package called rng-tools. It is packaged from Mr. Horman's https://github.com/nhorman/rng-tools. What's in there are the tools for pumping those precious truly random bits out of the USB-source to Linux kernel's entropy pool. As default, rngd will use /dev/hwrng as the source for randomness. Some Linuxes don't have that device at all, some Linuxes point that into CPU's random source. What's guaranteed, it will not point to your USB-stick! We need to change that.
Btw. you might be horrified by the fact, that something is fidding with your randomness. The exact bits transferred from USB to entropy pool won't be the actual bits getting out of /dev/random. Your keypresses and many other events are still a factor. Its still a good idea to not run randomness-monitoring malware or spyware in your Linux.
Systemd works so, that I did create a copy of /usr/lib/systemd/system/rngd.service into /etc/systemd/system/rngd.service. The contents of the copy in /etc/systemd/system/ can be freely modified and it has priority over the /usr/lib/systemd/system/ one. The only change I made was to have the ExecStart-line say as:
ExecStart=/sbin/rngd -f --rng-device=/dev/TrueRNG --fill-watermark=4000
When rngd-service would be started, it will use the USB-stick as source and make sure, there are at least 4000 bits of entropy in the pool.
Making sure rngd setup works
At any given point, you can query how many bits are available in the Linux entropy-pool:
cat /proc/sys/kernel/random/entropy_avail
Since my setup is working correctly, it will display a number greater than 4000 and smaller than 4096. The upper limit comes from /proc/sys/kernel/random/poolsize, which is a hard-coded number from Linux kernel source.
Hint: If you do the stupid thing like I did and set the /proc/sys/kernel/random/write_wakeup_threshold (using --fill-watermark) into 4096 (or above), your rngd will keep hogging CPU like there is no tomorrow. It is impossible for the pool to contain maximum number of bits at any given time. Give your system a break and set the threshold bit lower than max.
Finally
It's always nice to know for a fact, that random numbers are random. This fact can be verified and has been verified by number of other people.
Enjoy!
Bacula 9 vchanger: Tape count fix
Sunday, June 2. 2019
One of the first ever blog posts I've written here is about Bacula, the open-source backup software (more at https://www.bacula.org/). I published Fedora 17 binaries for the virtual tape changer running for Bacula 7. The post from year 2013 is here.
Running Bacula in Fedora Linux isn't much of a trick, ready-made binaries are available by the distro and configuring one is covered in Bacula's documentation. Then again, running Bacula with a NAS (see Wikipedia for Network-attached storage) as storage backend is where things get very very tricky. I've written about my Qnap NAS-device's support earlier, see the post about that.
Since its inception, Bacula is baked to require a tape drive (or drives) and a set of tapes (a single tape is supported also). Given modern day computing environment, actual physical tapes aren't used that much. Even I stopped using DLT (Wikipedia Digital Linear Tape) or LTO (Wikipedia Linear Tape-Open) tapes years ago and went for an easy, fast and inexpensive solution for storing my backups on a NAS. So, I really do need to have a concept of a "tape" somehow. That's where the virtual Bacula tape changer steps in. It is a piece of software attaching to Bacula autochanger API emulating a virtual "tape" drive and set of tapes with all the necessary operations, but doing all that on a filesystem. More details about autochangers can be found from Bacula Autochanger Resource page.
The obvious idea is to create a set of files to act as a set of "tapes". For system administration purposes, the tapes are just files in a subdirectory. Smart thing to do is to make that particular subdirectory located on a NAS to store the backups where there is plenty of external capacity outside your system. In my case, I'll access them over an iSCSI-mounted filesystem. More details about iSCSI on a Linux can be found from RedHat Enterprise Linux 7 manual pages at https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/storage_administration_guide/online-storage-management#osm-target-setup. For those planning something similar: I chose NOT to go with a NFS. A NFS-connection get stuck very easily and you will be rebooting your stuff a lot. See How-To: Release Stuck NFS Mounts without a Reboot @ Linux Journal for more about that.
When I went to Fedora 29, my Bacula setup got an automatic bump to version 9. My previous setup was for Bacula version 7 and quite soon I realized that I needed to alter my vchanger somehow to get it to support version 9. Bacula-guys did make changes to autochanger-API in their version-bump process. Luckily vchanger author was ahead of me and I got the code from http://sourceforge.net/projects/vchanger/. Soon realized that when I did a simple command of vchanger /etc/qnap.conf LIST, it displayed an extra tape which didn't exist in reality. I was puzzled. Old setup displayed the tape count correctly.
I did some C++ debugging and found out an obvious bug in the code. In src/diskchanger.cpp, DiskChanger-class InitializeVirtSlots()-method calculates the last changer slot numer incorrectly. It is guaranteed to be one-off. After fixing this, I contacted the vchanger author Mr. J. Fisher about my findings, and he agreed, there was a bug in his code.
Unfortunately, couple of months have passed and there is no 1.0.3 release yet, so the fix isn't in the SourceForge git-repo yet. For Fedora-users, my RPMs are available at http://opensource.hqcodeshop.com/Bacula/vchanger for B9/. Go get them there! I've been using those since last December, so I think my fix is correct and doesn't introduce any issues.
