Pages

Monday, 11 April 2016

Using TCP-LP with pycurl in Python

Intro


TCP-LP (low priority) is a TCP congestion control algorithm that is meant to be used by TCP connections that don't want to compete with other connections for bandwidth. Its goal is to use the idle bandwidth for file transfers. The details of TCP-LP are here.

With Linux' plugable congestion control algorithms, it is possible to change both the default algorithm for the whole system and the one used per connection. For the latter, one needs to be root.

Note: Changing the CC algorithm will only affect the transmissions. You cannot alter the remote end's behavior. This means that the below only make sense when you are going to upload data.

PyCurl


Changing the CC algorithm is a matter of using setsockopt on a socket. Doing this with pycurl can be a bit tricky. Even though pycurl supports the SOCKOPTFUNCTION, this is only for newer pycurl versions. For older, one can exploit pycurl's OPENSOCKETFUNCTION instead.

The trick is done with this piece of code:

[code language="python"]
import pycurl
import socket

def _getsock(family, socktype, protocol, addr):
    s=socket.socket(family, socktype, protocol)
    s.setsockopt(socket.IPPROTO_TCP, 13, 'lp' )
    return s

c = pycurl.Curl()
c.setopt(c.OPENSOCKETFUNCTION, _getsock)
c.setopt(c.URL, 'http://127.0.0.1/')
c.perform()
c.close()
[/code]

In the above, pycurl will call _getsock and expect it to return a socket. The function creates a new socket, then calls setsockopt with IPPPROTO_TCP and 13 (which is TCP_CONGESTION - see /usr/include/linux/tcp.h,  /usr/include/netinet/tcp.h). It then attempts to set the algorithm to "lp" which is the TCP-LP congestion control algorithm.

You most probably want to wrap the setsockopt around a try/except clause as it may fail if "lp" is not available (needs the module tcp_lp loaded) or if the program doesn't run as root.

The _getsock function also depends on the pycurl version, as its arguments have changed over time. Consult the docs for the fine details.

Results


Example of uploading two 500MB files in parallel on an already busy production network. One is with TCP-LP and the other with the default (TCP Cubic):

TCP-Cubic: 9.38 seconds
TCP-LP: 23.08 seconds

Same test, for 100MB files, again in parallel, on the same network:

TCP-Cubic: 3.14 seconds
TCP-LP: 5.38 seconds

Note: The above are random runs, presented to give an idea of the impact. For actual experimental results we would need to have  multiple runs and also monitor the background traffic.

Monday, 15 February 2016

Running an NTP server in a VM using KVM

The setup


Having physical server pA, running VMs using KVM. One of theVMs (vA) acts as an NTP server. pA gets the time from vA and vA gets it from the Internet.

It's not a great idea to run an NTP server in a VM, but in this case there was need for it.

The problem


NTP server gets frequently out of sync.

If you use nagios, you may get errors like this:
SERVICE ALERT: pA;ntpd;CRITICAL;SOFT;4;NTP CRITICAL: Offset unknown

Both for the physical server and other servers that fetch the time from vA.

The reason


There's some guessing involved here, but this should be pretty accurate:

VM vA needs to correct its clock every now and then by slowing down or speeding up things per ntpd/adjtimex. As expected, this creates a small discrepancy between vA and pA, as now the physical server gets out of sync and needs to correct its time using vA's reference time.

Once vA attempts to correct its time, again by slowing down or speeding up its clock, this has a direct effect on vA, as vA's clock is now affected by pA's ongoing adjustment. This happens because KVM by default uses kvmclock as its clock source (the source that ticks and not the source that returns the time of the day).

This action sometimes causes pA's ntpd to get even more out of sync and may even make it consider its peers inaccurate and become fully out of sync.

The problem gets even worse if you have two ntp servers (vA and vB) running on two different physical servers (pA and pB), because the amount of desync between the two is mostly random. Assuming that all your servers, including pA and pB, fetch the time from vA and vB, the discrepancy between them will make them mark at least one of them as wrong, as the stratum of vA and vB does not permit such difference between their clocks.

You can see the above by looking at the falsetick result in ntpq's associations:
ind assid status  conf reach auth condition  last_event cnt
===========================================================
  1 33082  961a   yes   yes  none  sys.peer    sys_peer  1
  2 33083  911a   yes   yes  none falsetick    sys_peer  1

Overall, the problem is that the physical servers will try to fix their clocks, thus affecting the clocks of the NTP servers running in VMs under them.

The solution


The problem is with the VMs using the kvmclock source. You can see that using dmesg:
$ dmesg | grep clocksource
Switching to clocksource kvm-clock

The way to disable this is to pass the "no-kvmclock" parameter to the kernel of your VMs. This will not always work though. The reason is that the kernel (at least the CentOS kernels) will panic very early in the boot process as it will still try to initialize the kvmclock even if it's not going to use it, and will fail.

The solution is to pass two parameters to your VM kernels: "no-kvmclock no-kvmclock-vsyscall". The second one is a bit undocumented, but will do the trick.

After that you can verify it through dmesg:
$ dmesg | grep Switching
Switching to clocksource refined-jiffies
Switching to clocksource acpi_pm
Switching to clocksource tsc

Example


Below is the output of a server running in such an environment. In this case the first ntp server (vA) runs with the extra kernel parameters and the other (vB) runs without them. The clock of the physical servers (pA and pB) was slowed down by hand using adjtimex in order to test the effect of the physical server's clock on the VM clocks. As you can see, this server is still in sync with vA and has a very large offset with vB. Note that this server is not a VM under pA or pB.
$ ntpq -nc peers
     remote           refid      st t when poll reach delay   offset  jitter
==============================================================================
*10.93.XXX.XXX   216.218.254.202  2 u   81  256  377 0.433  -87.076  20.341
 10.93.XXX.XXX   216.218.254.202  2 u  290  512  377 0.673  11487.6 9868.84

I.e., what happened is that the first one, using the extra parameters, kept its clock accurate while the second did not.

 

Saturday, 8 August 2015

Debian packaging for python2 and python3 at the same time

The problem


The scenario was like this:

  • Python code that provides a library and a binary

  • The code is compatible with both Python v2 and v3


The requirements were:

  • Generate a package with the library part for python v2

  • Generate a package with the library part for python v3

  • Generate a binary package with the executable for python v3


I.e, from one source package (vdns) I wanted to create python-vdns (python2 lib), python3-vdns (python3 lib) and vdns (executable).

The approach


After trying other methods, I ended up using debhelper 9 and pybuild. Before that I tried using CDBS but had no luck there.

With DH9 it's easy to package a python library for multiple python versions as it handles everything itself. The only catch was the package that contained the binaries and how to make it use the v3 version instead of the v2

The solution


The solution was this rules file:

[code lang="bash" gutter="0"]
#!/usr/bin/make -f

# see EXAMPLES in dpkg-buildflags(1) and read /usr/share/dpkg/*
DPKG_EXPORT_BUILDFLAGS = 1
include /usr/share/dpkg/default.mk

# Don't set this or else the .install files will fail
#export PYBUILD_NAME = vdns
#export PYBUILD_SYSTEM = custom

# Otherwise the usr/bin/vdns.py file is from python2
export PYBUILD_INSTALL_ARGS_python2 = --install-scripts=/dev/null

# main packaging script based on dh7 syntax
%:
dh $@ --with python3,python2 --buildsystem=pybuild
[/code]

The control file is like this:
Source: vdns
Section: unknown
Priority: optional
Maintainer: Stefanos Harhalakis <v13@v13.gr>;
Build-Depends: debhelper (>= 9), dh-python,
python-all (>=2.6.6-3~), python-setuptools,
python3-all (>=3.2), python3-setuptools
Standards-Version: 3.9.5
X-Python-Version: >= 2.7
X-Python3-Version: >= 3.2
XS-Python-Version: >= 2.7
XS-Python3-Version: >= 3.2

Package: python-vdns
Architecture: all
Depends: ${python:Depends}, ${misc:Depends}, python-psycopg2
Description: vdns python2 libraries
These libraries allow the reading and the creation of bind zone files

Package: python3-vdns
Architecture: all
Depends: ${python3:Depends}, ${misc:Depends}, python3-psycopg2
Description: vdns python3 libraries
These libraries allow the reading and the creation of bind zone files

Package: vdns
Architecture: all
Depends: ${python:Depends}, ${misc:Depends}, python3-vdns, python(>=3.2)
Description: Database-based DNS management
vdns is a database-based DNS management tool. It gets data from its database
and generates bind zone files. It supports A, AAAA, MX, NS, DS, PTR, CNAME,
TXT, DNSSEC, SSHFP, DKIM and SRV records
.
vdns uses a PostgreSQL database to store the data
.
vdns is not a 1-1 mapping of DB<->zone files. Instead the databsae
is meant to describe the data that are later generated. E.g:
* DKIM data are used to generate TXT records
* A, AAAA and PTR entries are generated from the same data
* NS glue records for sub-zones are auto-created

And the .install files are like this:
python-vdns.install:
usr/lib/python2*/dist-packages/vdns/*.py*
usr/lib/python2*/dist-packages/vdns/src

python3-vdns.install:
usr/lib/python3*/dist-packages/vdns/*.py*
usr/lib/python3*/dist-packages/vdns/src

Explanation


What the above does is to use DH9 with pybuild. Pybuild takes care of multiple versions by "compiling" the binaries  twice under build/scripts-2.x and build/scripts/3.x directories. After that it copies them  to debian/tmp and finally splits the contents of debian/tmp based on the .install files to debian/python-vdns, debian/python3-vdns and debian/vdns.

The biggest problem were the files that were meant for the vdns package as those were present in both build/scripts-2.x and build/scripts/3.x, each one prepared for the appropriate debian version:
-rwxr-xr-x 1 v13 v13 1197 Jul 19 21:12 build/scripts-2.7/vdns.py
-rwxr-xr-x 1 v13 v13 1198 Jul 19 21:12 build/scripts-3.4/vdns.py

The following line in rules takes care of the conflict by skipping a version:
export PYBUILD_INSTALL_ARGS_python2 = --install-scripts=/dev/null

This way, the python2 version never gets installed and thus only the python3 version is available to be copied to debian/tmp. Otherwise the behavior was random (it was picking the first one it was finding).

Other attempts


I also tried using autoconf with CDBS but that proved to be even more difficult.

Acknowledgements


The above was accomplished only because of the help of folks in #debian-python @ OFTC; namely: p1otr, mapreri and jcristau.

Sunday, 28 September 2014

Multiple relay configuration based on sender address with sendmail

One of the needs that came up was to be able to use separate relay configurations based on the sender email address, using sendmail. The problem is that sendmail is missing support for most parts of that sentence.

At the end the solution involved a combination of sendmail, smarttable, procmail and msmtp

The idea is the following:

  • Use smarttable to implement sender based rules

  • Use the procmail mailer support to use procmail to deliver the emails

  • Use procmailrc to pipe messages to msmtp

  • Use msmtp to relay via external hosts


Sender based rules


In order to be able to have sender-based rules I used smarttable.m4 from here.

Download the smarttable.m4 and (assuming sendmail config is under /etc/mail) place it under /etc/mail/m4/. Normally it should be placed along the rest of the sendmail features (/usr/share/sendmail/cf/features) but I don't like polluting system dirs. Then use the following config in sendmail.mc:
dnl Change the _CF_DIR for a bit to load the feature from /etc/mail/m4
dnl then change it back
define(`_CF_DIR_OLD', _CF_DIR_)dnl
define(`_CF_DIR_', `/etc/mail/m4/')dnl
dnl This has to be a hash. I.e. not text.
FEATURE(`smarttable',`hash -o /etc/mail/smarttable')dnl
define(`_CF_DIR_', _CF_DIR_OLD)dnl

Then configure smarttable (/etc/mail/smarttable) like this:
test@test.com    procmail:/etc/mail/persource/test.test.com.procmailrc

You can add as many lines as you like, one for each sender. See smarttable's web page for more information on the supported sender formats. Dont' forget to generate the hashed version (smarttable.db)

Procmail config


Configure sendmail for procmail mailer like this:
define(`PROCMAIL_MAILER_ARGS', `procmail -Y -t -m $h $f $u')dnl
MAILER(`procmail')dnl

You have to override the default procmail parameters in order to add the -t switch. This way delivery errors will be interpreted as softfails, otherwise mails will be rejected on the first failure.

Create /etc/mail/persource and put the procmail configs in there (nice and tidy). In this example create /etc/mail/persources/test.test.com.procmailrc as follows:
:0w
|/usr/bin/msmtp -C /etc/mail/persource/test.test.com.msmtprc -a test@test.com -t

The 'w' flag is essential in order to feed failures back to sendmail.

Msmtp config


Create the msmtp config file (/etc/mail/persource/test.test.com.msmtprc) as follows:
defaults
syslog on
# logfile /tmp/msmtp-test@test.com.log

account     test@test.com
host         smtp.gmail.com
from         test@test.com
user         test@test.com
password     xxx
auth         on
tls         on
tls_trust_file /etc/ssl/certs/ca-certificates.crt

Your mileage may vary. They above is good for gmail accounts on a debian system.

Done


And that's it. Sending an email as test@test.com will cause sendmail to use smarttable. This will match the sender and use procmail with our config to deliver the email. Procmail will pipe the email to msmtp which will send the email via google's email servers.

Thursday, 11 September 2014

OpenVPN and remote-cert-tls server

This required a bit of digging into OpenVPN's and OpenSSL's code to figure out.

The problem


This error:
Thu Sep 11 00:12:05 2014 Validating certificate key usage
Thu Sep 11 00:12:05 2014 ++ Certificate has key usage  00f8, expects 00a0
Thu Sep 11 00:12:05 2014 ++ Certificate has key usage  00f8, expects 0088

The condition


Using openvpn with the following option:
remote-cert-tls server

The solution


(for me) to add this to openvpn's config file:
remote-cert-ku f8

The explanation


Background


remote-cert-tls attempts to solve one problem: Lets say you run a CA and you distribute the certificates to 2 people including me and you. Then you setup a VPN server for us to use and you generate another certificate for the VPN server.

As always, the problem that certificates attempt to solve is "how do you know you're connecting to the remote end you assume you are. In normal SSL pages you trust a CA to verify that the CN of the certificate matches the owner of the domain. If you want to achieve the same thing with openvpn then you need to verify the CN of the remote end against either the hostname or a predefined string. If not then you need to use the "remote-cert-tls server" option.

If you don't use any of the above methods then I can fire up an openvpn server using the certificate you provided me with and since both my certificate and the actual VPN server's certificate are signed by the same CA you would be verifying both and be equally willing to connect to both, thus allowing me to spy on you.

To solve this kind of problems X509 has some properties for certificates that designate them for certain purposes. E.g. one of them is to run a VPN endpoint (TLS Web Client Authentication)

To be precise there are 2+1 such designations in X509:
X509v3 Key Usage: 
    Digital Signature, Non Repudiation, Key Encipherment, Data Encipherment, Key Agreement

X509v3 Extended Key Usage:
    TLS Web Server Authentication, TLS Web Client Authentication, IPSec End System, IPSec Tunnel, Time Stamping

Netscape Cert Type:
    SSL Client, SSL Server, S/MIME, Object Signing

"Netscape Cert Type" is kind of old. "Key Usage" is the main one and "Extended Key Usage" is the final addition. Ignoring NS Cert Type, "Key Usage" is a bitmap and thus has limited space for expansion. "Extended Key Usage" on the other hand is a list of object identifiers which allows for unlimited expansion.

The certificate


The certificate I was using for the server-side of the OpenVPN had the above attributes. Ignoring NS Cert Type once more, the other two correspond to the following data:
  494:d=5  hl=2 l=   3 prim: OBJECT            :X509v3 Key Usage
  499:d=5  hl=2 l=   4 prim: OCTET STRING      [HEX DUMP]:030203F8

  433:d=5  hl=2 l=   3 prim: OBJECT            :X509v3 Extended Key Usage
  438:d=5  hl=2 l=  52 prim: OCTET STRING      [HEX DUMP]:303206082B0601050507030106082B0601050507030206082B0601050507030506082B0601050507030606082B06010505070308

Starting with "Key Usage", the actual value is "F8". The meaning of each bit can be found in OpenSSL's code:
#define KU_DIGITAL_SIGNATURE    0x0080
#define KU_NON_REPUDIATION      0x0040
#define KU_KEY_ENCIPHERMENT     0x0020
#define KU_DATA_ENCIPHERMENT    0x0010
#define KU_KEY_AGREEMENT        0x0008
#define KU_KEY_CERT_SIGN        0x0004
#define KU_CRL_SIGN             0x0002
#define KU_ENCIPHER_ONLY        0x0001
#define KU_DECIPHER_ONLY        0x8000

On the other hand, the "Extended Key Usage" part contains the following Object IDs:
06082B 06010505070301 -> serverAuth (TLS Web Server Authentication)
06082B 06010505070302 -> clientAuth (TLS Web Client Authentication)
06082B 06010505070305 -> ipsecEndSystem (IPSec End System)
06082B 06010505070306 -> ipsecTunnel (IPSec Tunnel)
06082B 06010505070308 -> timeStamping (Time Stamping)

The "bug"


The first thing to notice is that the failure is for "Key Usage" and not for "Extended Key Usage" (took me some time to figure out).

After that, a bit of digging into the code confirms that OpenVPN attempts to verify a bitmap with equality. I.e. it gets the certificates' value and compares it against a predefined list of allowed values, which according to OpenVPN's documentation defaults to "a0 88" (which means one of them). However the actual certificates bitmap value is 0xf8 as mentioned above. And thus the comparison fails with the error:
Thu Sep 11 00:12:05 2014 Validating certificate key usage
Thu Sep 11 00:12:05 2014 ++ Certificate has key usage  00f8, expects 00a0
Thu Sep 11 00:12:05 2014 ++ Certificate has key usage  00f8, expects 0088

The reason I'm calling this a bug is because it's not sensible to use equality to compare against a bitmap. Instead one can use AND, in which case there would be:
( <certificate's value> & <desired value> ) == <desired value>
or:
( 0xf8 & 0xa0) == 0xa0 -> True

In order for the validation to succeed with the defaults the certificate should have one of the following designations:
0xa0: Digital Signature, Key Encipherment
0x88: Digital Signature, Key Agreement

The solution


So there, since the comparison is done with equality you can do one of the following:

  • Use the above Key Usage on the certificate (inconvenient)

  • Don't use "remote-cert-tls server" (bad)

  • Use "remote-cert-ku XX" where XX is the value of your certificate which can be seen in OpenVPN's messages (the last octet). In my case it's f8.


 

Thursday, 31 July 2014

Linux, multicast, bridging and IPv6 troubles (i.e. why my IPv6 connectivity goes missing)

For a long time now I had a very annoying problem with IPv6 under Linux.

My setup is as follows: Linux box <-> Switch <-> Router

The Linux box uses a bridge interface (br0) and usually only has one physical interface attached to it (eth0). That's a very convenient setup.

The problem is that after a couple of minutes the IPv6 connectivity of the host will go away. Now, the host has a static IPv6 assigned to it and it's not that it loses the address or any route. Instead it just stops communicating with everything.

Troubleshooting this showed that the box loses the MAC address of the router and the ND protocol does not work, so it never recovers.

When the problem occurs, the neighbor information becomes stale:
# ip neigh
2a01:XXX:YYY:1::1 dev br0 lladdr 00:11:12:13:14:c4 router STALE
fe80::20c:XXff:feXX:YYYY dev br0 lladdr 00:11:12:13:14:c4 router STALE

I.e the entry remains in a 'STALE' state and never recovers.

My workarounds so far have been:

  • Enable promiscuous mode on the interface (ifconfig br0 promisc)

  • Clear neighbors (ip neigh flush)


Everything pointed out to multicast issues (what IPv6 ND uses).

Long-story-short, this was an eye opener: http://troglobit.com/blog/2013/07/09/multicast-howto/

What needs to be done is to disable IGMP snooping on the bridge interface because it causes these issues. This is done with:
# echo 0 > /sys/devices/virtual/net/br0/bridge/multicast_snooping

So do yourself a favor and add this to /etc/network/interfaces, in the relevant interface:
    up    echo 0 > /sys/devices/virtual/net/$IFACE/bridge/multicast_snooping

 

Thursday, 8 August 2013

Installing package build dependencies from a .dsc file (Debian)

There are cases where one needs to install build-dependencies of a .dsc file in Debian.

Apparently this is not as trivial as:

[code gutter="false"]
# apt-get build-dep package
[/code]

The easiest way I've found so far is to use mk-build-deps (from the devscripts package):

[code gutter="false"]
# mk-build-deps -i vadm_1.0.4ci+r16.dsc -t apt-get --no-install-recommends -y
dh_testdir
dh_testroot
dh_prep
dh_testdir
dh_testroot
dh_install
dh_installdocs
dh_installchangelogs
dh_compress
dh_fixperms
dh_installdeb
dh_gencontrol
dh_md5sums
dh_builddeb
dpkg-deb: building package `vadm-build-deps' in `../vadm-build-deps_1.0.4ci+r16_all.deb'.

The package has been created.
Attention, the package has been created in the current directory,
not in ".." as indicated by the message above!
(Reading database ... 41052 files and directories currently installed.)
Preparing to replace vadm-build-deps 1.0.4ci+r16 (using vadm-build-deps_1.0.4ci+r16_all.deb) ...
Unpacking replacement vadm-build-deps ...

Reading package lists...
Building dependency tree...
Reading state information...

0 upgraded, 0 newly installed, 0 to remove and 2 not upgraded.
1 not fully installed or removed.
After this operation, 0 B of additional disk space will be used.
Setting up vadm-build-deps (1.0.4ci+r16) ...
[/code]

This godly script:

  • Creates a psudo package that depends on the build-depends of the .dsc file

  • Does a dpkg -i on the generated deb file (which may fail because of missing build depends)

  • Does apt-get -f install


Extra points for using -r which will remove the generated package after it's done.