• Buy the hardware.
• Plug in the hardware.
• Confirm it is detected.
• Install the appropriate sdr development-package(s).
• Install the magic software.
  • Written by @antirez, no less, you know it is gonna be good!

  $ lsusb
  ..
  Bus 003 Device 043: ID 0bda:2838 Realtek Semiconductor Corp. RTL2838 DVB-T
  ..

  idVendor           0x0bda Realtek Semiconductor Corp.
  idProduct          0x2838 RTL2838 DVB-T

  # apt-get install librtlsdr-dev libusb-1.0-0-dev

  $ git clone https://github.com/antirez/dump1090.git
  $ cd dump1090
  $ make

  $ sudo ./dump1090 --interactive --net

  $ sudo rmmod dvb_usb_rtl28xxu
  $ sudo ./dump1090 --interactive --net

  Hex    Flight   Altitude  Speed   Lat       Lon       Track  Messages Seen       .
  --------------------------------------------------------------------------------
  4601fc          14200     0       0.000     0.000     0     11        1 sec
  4601f2          9550      0       0.000     0.000     0     58        0 sec
  45ac52 SAS1716  2650      177     60.252    24.770    47    26        1 sec

• With the right udev-rules in place it is possible to run the tool as non-root, and blacklist the default kernel module.
• There are other forks of the dump1090 software that are more up-to-date to explore.
• SDR can do more than track planes.

Why store keys offline? Before we go into details about storing keys offline, it may be useful to do a small reminder of how the OpenPGP standard works. OpenPGP keys are made of a main public/private key pair, the certification key, used to sign user identifiers and subkeys. My public key, shown below, has the usual main certification/signature key (marked SC) but also an encryption subkey (marked E), a separate signature key (S), and two authentication keys (marked A) which I use as RSA keys to log into servers using SSH, thanks to the Monkeysphere project.

    pub   rsa4096/792152527B75921E 2009-05-29 [SC] [expires: 2018-04-19]
      8DC901CE64146C048AD50FBB792152527B75921E
    uid                 [ultimate] Antoine Beaupr  <anarcat@anarc.at>
    uid                 [ultimate] Antoine Beaupr  <anarcat@koumbit.org>
    uid                 [ultimate] Antoine Beaupr  <anarcat@orangeseeds.org>
    uid                 [ultimate] Antoine Beaupr  <anarcat@debian.org>
    sub   rsa2048/B7F648FED2DF2587 2012-07-18 [A]
    sub   rsa2048/604E4B3EEE02855A 2012-07-20 [A]
    sub   rsa4096/A51D5B109C5A5581 2009-05-29 [E]
    sub   rsa2048/3EA1DDDDB261D97B 2017-08-23 [S]

All the subkeys (sub) and identities (uid) are bound by the main certification key using cryptographic self-signatures. So while an attacker stealing a private subkey can spoof signatures in my name or authenticate to other servers, that key can always be revoked by the main certification key. But if the certification key gets stolen, all bets are off: the attacker can create or revoke identities or subkeys as they wish. In a catastrophic scenario, an attacker could even steal the key and remove your copies, taking complete control of the key, without any possibility of recovery. Incidentally, this is why it is so important to generate a revocation certificate and store it offline. So by moving the certification key offline, we reduce the attack surface on the OpenPGP trust chain: day-to-day keys (e.g. email encryption or signature) can stay online but if they get stolen, the certification key can revoke those keys without having to revoke the main certification key as well. Note that a stolen encryption key is a different problem: even if we revoke the encryption subkey, this will only affect future encrypted messages. Previous messages will be readable by the attacker with the stolen subkey even if that subkey gets revoked, so the benefits of revoking encryption certificates are more limited.

Common strategies for offline key storage Considering the security tradeoffs, some propose storing those critical keys offline to reduce those threats. But where exactly? In an attempt to answer that question, Jonathan McDowell, a member of the Debian keyring maintenance team, said that there are three options: use an external LUKS-encrypted volume, an air-gapped system, or a keycard. Full-disk encryption like LUKS adds an extra layer of security by hiding the content of the key from an attacker. Even though private keyrings are usually protected by a passphrase, they are easily identifiable as a keyring. But when a volume is fully encrypted, it's not immediately obvious to an attacker there is private key material on the device. According to Sean Whitton, another advantage of LUKS over plain GnuPG keyring encryption is that you can pass the --iter-time argument when creating a LUKS partition to increase key-derivation delay, which makes brute-forcing much harder. Indeed, GnuPG 2.x doesn't have a run-time option to configure the key-derivation algorithm, although a patch was introduced recently to make the delay configurable at compile time in gpg-agent, which is now responsible for all secret key operations. The downside of external volumes is complexity: GnuPG makes it difficult to extract secrets out of its keyring, which makes the first setup tricky and error-prone. This is easier in the 2.x series thanks to the new storage system and the associated keygrip files, but it still requires arcane knowledge of GPG internals. It is also inconvenient to use secret keys stored outside your main keyring when you actually do need to use them, as GPG doesn't know where to find those keys anymore. Another option is to set up a separate air-gapped system to perform certification operations. An example is the PGP clean room project, which is a live system based on Debian and designed by DD Daniel Pocock to operate an OpenPGP and X.509 certificate authority using commodity hardware. The basic principle is to store the secrets on a different machine that is never connected to the network and, therefore, not exposed to attacks, at least in theory. I have personally discarded that approach because I feel air-gapped systems provide a false sense of security: data eventually does need to come in and out of the system, somehow, even if only to propagate signatures out of the system, which exposes the system to attacks. System updates are similarly problematic: to keep the system secure, timely security updates need to be deployed to the air-gapped system. A common use pattern is to share data through USB keys, which introduce a vulnerability where attacks like BadUSB can infect the air-gapped system. From there, there is a multitude of exotic ways of exfiltrating the data using LEDs, infrared cameras, or the good old TEMPEST attack. I therefore concluded the complexity tradeoffs of an air-gapped system are not worth it. Furthermore, the workflow for air-gapped systems is complex: even though PGP clean room went a long way, it's still lacking even simple scripts that allow signing or transferring keys, which is a problem shared by the external LUKS storage approach.

Keycard advantages The approach I have chosen is to use a cryptographic keycard: an external device, usually connected through the USB port, that stores the private key material and performs critical cryptographic operations on the behalf of the host. For example, the FST-01 keycard can perform RSA and ECC public-key decryption without ever exposing the private key material to the host. In effect, a keycard is a miniature computer that performs restricted computations for another host. Keycards usually support multiple "slots" to store subkeys. The OpenPGP standard specifies there are three subkeys available by default: for signature, authentication, and encryption. Finally, keycards can have an actual physical keypad to enter passwords so a potential keylogger cannot capture them, although the keycards I have access to do not feature such a keypad. We could easily draw a parallel between keycards and an air-gapped system; in effect, a keycard is a miniaturized air-gapped computer and suffers from similar problems. An attacker can intercept data on the host system and attack the device in the same way, if not more easily, because a keycard is actually "online" (i.e. clearly not air-gapped) when connected. The advantage over a fully-fledged air-gapped computer, however, is that the keycard implements only a restricted set of operations. So it is easier to create an open hardware and software design that is audited and verified, which is much harder to accomplish for a general-purpose computer. Like air-gapped systems, keycards address the scenario where an attacker wants to get the private key material. While an attacker could fool the keycard into signing or decrypting some data, this is possible only while the key is physically connected, and the keycard software will prompt the user for a password before doing the operation, though the keycard can cache the password for some time. In effect, it thwarts offline attacks: to brute-force the key's password, the attacker needs to be on the target system and try to guess the keycard's password, which will lock itself after a limited number of tries. It also provides for a clean and standard interface to store keys offline: a single GnuPG command moves private key material to a keycard (the keytocard command in the --edit-key interface), whereas moving private key material to a LUKS-encrypted device or air-gapped computer is more complex. Keycards are also useful if you operate on multiple computers. A common problem when using GnuPG on multiple machines is how to safely copy and synchronize private key material among different devices, which introduces new security problems. Indeed, a "good rule of thumb in a forensics lab", according to Robert J. Hansen on the GnuPG mailing list, is to "store the minimum personal data possible on your systems". Keycards provide the best of both worlds here: you can use your private key on multiple computers without actually storing it in multiple places. In fact, Mike Gerwitz went as far as saying:

For users that need their GPG key on multiple boxes, I consider a smartcard to be essential. Otherwise, the user is just furthering her risk of compromise.

Keycard tradeoffs As Gerwitz hinted, there are multiple downsides to using a keycard, however. Another DD, Wouter Verhelst clearly expressed the tradeoffs:

Smartcards are useful. They ensure that the private half of your key is never on any hard disk or other general storage device, and therefore that it cannot possibly be stolen (because there's only one possible copy of it). Smartcards are a pain in the ass. They ensure that the private half of your key is never on any hard disk or other general storage device but instead sits in your wallet, so whenever you need to access it, you need to grab your wallet to be able to do so, which takes more effort than just firing up GnuPG. If your laptop doesn't have a builtin cardreader, you also need to fish the reader from your backpack or wherever, etc.

"Smartcards" here refer to older OpenPGP cards that relied on the IEC 7816 smartcard connectors and therefore needed a specially-built smartcard reader. Newer keycards simply use a standard USB connector. In any case, it's true that having an external device introduces new issues: attackers can steal your keycard, you can simply lose it, or wash it with your dirty laundry. A laptop or a computer can also be lost, of course, but it is much easier to lose a small USB keycard than a full laptop and I have yet to hear of someone shoving a full laptop into a washing machine. When you lose your keycard, unless a separate revocation certificate is available somewhere, you lose complete control of the key, which is catastrophic. But, even if you revoke the lost key, you need to create a new one, which involves rebuilding the web of trust for the key a rather expensive operation as it usually requires meeting other OpenPGP users in person to exchange fingerprints. You should therefore think about how to back up the certification key, which is a problem that already exists for online keys; of course, everyone has a revocation certificates and backups of their OpenPGP keys... right? In the keycard scenario, backups may be multiple keycards distributed geographically. Note that, contrary to an air-gapped system, a key generated on a keycard cannot be backed up, by design. For subkeys, this is not a problem as they do not need to be backed up (except encryption keys). But, for a certification key, this means users need to generate the key on the host and transfer it to the keycard, which means the host is expected to have enough entropy to generate cryptographic-strength random numbers, for example. Also consider the possibility of combining different approaches: you could, for example, use a keycard for day-to-day operation, but keep a backup of the certification key on a LUKS-encrypted offline volume. Keycards introduce a new element into the trust chain: you need to trust the keycard manufacturer to not have any hostile code in the key's firmware or hardware. In addition, you need to trust that the implementation is correct. Keycards are harder to update: the firmware may be deliberately inaccessible to the host for security reasons or may require special software to manipulate. Keycards may be slower than the CPU in performing certain operations because they are small embedded microcontrollers with limited computing power. Finally, keycards may encourage users to trust multiple machines with their secrets, which works against the "minimum personal data" principle. A completely different approach called the trusted physical console (TPC) does the opposite: instead of trying to get private key material onto all of those machines, just have them on a single machine that is used for everything. Unlike a keycard, the TPC is an actual computer, say a laptop, which has the advantage of needing no special procedure to manage keys. The downside is, of course, that you actually need to carry that laptop everywhere you go, which may be problematic, especially in some corporate environments that restrict bringing your own devices.

Quick keycard "howto" Getting keys onto a keycard is easy enough:

1. Start with a temporary key to test the procedure:

       export GNUPGHOME=$(mktemp -d)
       gpg --generate-key
   

2. Edit the key using its user ID (UID):

       gpg --edit-key UID
   

3. Use the key command to select the first subkey, then copy it to the keycard (you can also use the addcardkey command to just generate a new subkey directly on the keycard):

       gpg> key 1
       gpg> keytocard
   

4. If you want to move the subkey, use the save command, which will remove the local copy of the private key, so the keycard will be the only copy of the secret key. Otherwise use the quit command to save the key on the keycard, but keep the secret key in your normal keyring; answer "n" to "save changes?" and "y" to "quit without saving?" . This way the keycard is a backup of your secret key.
5. Once you are satisfied with the results, repeat steps 1 through 4 with your normal keyring (unset $GNUPGHOME)

When a key is moved to a keycard, --list-secret-keys will show it as sec> (or ssb> for subkeys) instead of the usual sec keyword. If the key is completely missing (for example, if you moved it to a LUKS container), the # sign is used instead. If you need to use a key from a keycard backup, you simply do gpg --card-edit with the key plugged in, then type the fetch command at the prompt to fetch the public key that corresponds to the private key on the keycard (which stays on the keycard). This is the same procedure as the one to use the secret key on another computer.

Conclusion There are already informal OpenPGP best-practices guides out there and some recommend storing keys offline, but they rarely explain what exactly that means. Storing your primary secret key offline is important in dealing with possible compromises and we examined the main ways of doing so: either with an air-gapped system, LUKS-encrypted keyring, or by using keycards. Each approach has its own tradeoffs, but I recommend getting familiar with keycards if you use multiple computers and want a standardized interface with minimal configuration trouble. And of course, those approaches can be combined. This tutorial, for example, uses a keycard on an air-gapped computer, which neatly resolves the question of how to transmit signatures between the air-gapped system and the world. It is definitely not for the faint of heart, however. Once one has decided to use a keycard, the next order of business is to choose a specific device. That choice will be addressed in a followup article, where I will look at performance, physical design, and other considerations.

This article first appeared in the Linux Weekly News.

Why store keys offline? Before we go into details about storing keys offline, it may be useful to do a small reminder of how the OpenPGP standard works. OpenPGP keys are made of a main public/private key pair, the certification key, used to sign user identifiers and subkeys. My public key, shown below, has the usual main certification/signature key (marked SC) but also an encryption subkey (marked E), a separate signature key (S), and two authentication keys (marked A) which I use as RSA keys to log into servers using SSH, thanks to the Monkeysphere project.

    pub   rsa4096/792152527B75921E 2009-05-29 [SC] [expires: 2018-04-19]
      8DC901CE64146C048AD50FBB792152527B75921E
    uid                 [ultimate] Antoine Beaupr  <anarcat@anarc.at>
    uid                 [ultimate] Antoine Beaupr  <anarcat@koumbit.org>
    uid                 [ultimate] Antoine Beaupr  <anarcat@orangeseeds.org>
    uid                 [ultimate] Antoine Beaupr  <anarcat@debian.org>
    sub   rsa2048/B7F648FED2DF2587 2012-07-18 [A]
    sub   rsa2048/604E4B3EEE02855A 2012-07-20 [A]
    sub   rsa4096/A51D5B109C5A5581 2009-05-29 [E]
    sub   rsa2048/3EA1DDDDB261D97B 2017-08-23 [S]

All the subkeys (sub) and identities (uid) are bound by the main certification key using cryptographic self-signatures. So while an attacker stealing a private subkey can spoof signatures in my name or authenticate to other servers, that key can always be revoked by the main certification key. But if the certification key gets stolen, all bets are off: the attacker can create or revoke identities or subkeys as they wish. In a catastrophic scenario, an attacker could even steal the key and remove your copies, taking complete control of the key, without any possibility of recovery. Incidentally, this is why it is so important to generate a revocation certificate and store it offline. So by moving the certification key offline, we reduce the attack surface on the OpenPGP trust chain: day-to-day keys (e.g. email encryption or signature) can stay online but if they get stolen, the certification key can revoke those keys without having to revoke the main certification key as well. Note that a stolen encryption key is a different problem: even if we revoke the encryption subkey, this will only affect future encrypted messages. Previous messages will be readable by the attacker with the stolen subkey even if that subkey gets revoked, so the benefits of revoking encryption certificates are more limited.

Common strategies for offline key storage Considering the security tradeoffs, some propose storing those critical keys offline to reduce those threats. But where exactly? In an attempt to answer that question, Jonathan McDowell, a member of the Debian keyring maintenance team, said that there are three options: use an external LUKS-encrypted volume, an air-gapped system, or a keycard. Full-disk encryption like LUKS adds an extra layer of security by hiding the content of the key from an attacker. Even though private keyrings are usually protected by a passphrase, they are easily identifiable as a keyring. But when a volume is fully encrypted, it's not immediately obvious to an attacker there is private key material on the device. According to Sean Whitton, another advantage of LUKS over plain GnuPG keyring encryption is that you can pass the --iter-time argument when creating a LUKS partition to increase key-derivation delay, which makes brute-forcing much harder. Indeed, GnuPG 2.x doesn't have a run-time option to configure the key-derivation algorithm, although a patch was introduced recently to make the delay configurable at compile time in gpg-agent, which is now responsible for all secret key operations. The downside of external volumes is complexity: GnuPG makes it difficult to extract secrets out of its keyring, which makes the first setup tricky and error-prone. This is easier in the 2.x series thanks to the new storage system and the associated keygrip files, but it still requires arcane knowledge of GPG internals. It is also inconvenient to use secret keys stored outside your main keyring when you actually do need to use them, as GPG doesn't know where to find those keys anymore. Another option is to set up a separate air-gapped system to perform certification operations. An example is the PGP clean room project, which is a live system based on Debian and designed by DD Daniel Pocock to operate an OpenPGP and X.509 certificate authority using commodity hardware. The basic principle is to store the secrets on a different machine that is never connected to the network and, therefore, not exposed to attacks, at least in theory. I have personally discarded that approach because I feel air-gapped systems provide a false sense of security: data eventually does need to come in and out of the system, somehow, even if only to propagate signatures out of the system, which exposes the system to attacks. System updates are similarly problematic: to keep the system secure, timely security updates need to be deployed to the air-gapped system. A common use pattern is to share data through USB keys, which introduce a vulnerability where attacks like BadUSB can infect the air-gapped system. From there, there is a multitude of exotic ways of exfiltrating the data using LEDs, infrared cameras, or the good old TEMPEST attack. I therefore concluded the complexity tradeoffs of an air-gapped system are not worth it. Furthermore, the workflow for air-gapped systems is complex: even though PGP clean room went a long way, it's still lacking even simple scripts that allow signing or transferring keys, which is a problem shared by the external LUKS storage approach.

Keycard advantages The approach I have chosen is to use a cryptographic keycard: an external device, usually connected through the USB port, that stores the private key material and performs critical cryptographic operations on the behalf of the host. For example, the FST-01 keycard can perform RSA and ECC public-key decryption without ever exposing the private key material to the host. In effect, a keycard is a miniature computer that performs restricted computations for another host. Keycards usually support multiple "slots" to store subkeys. The OpenPGP standard specifies there are three subkeys available by default: for signature, authentication, and encryption. Finally, keycards can have an actual physical keypad to enter passwords so a potential keylogger cannot capture them, although the keycards I have access to do not feature such a keypad. We could easily draw a parallel between keycards and an air-gapped system; in effect, a keycard is a miniaturized air-gapped computer and suffers from similar problems. An attacker can intercept data on the host system and attack the device in the same way, if not more easily, because a keycard is actually "online" (i.e. clearly not air-gapped) when connected. The advantage over a fully-fledged air-gapped computer, however, is that the keycard implements only a restricted set of operations. So it is easier to create an open hardware and software design that is audited and verified, which is much harder to accomplish for a general-purpose computer. Like air-gapped systems, keycards address the scenario where an attacker wants to get the private key material. While an attacker could fool the keycard into signing or decrypting some data, this is possible only while the key is physically connected, and the keycard software will prompt the user for a password before doing the operation, though the keycard can cache the password for some time. In effect, it thwarts offline attacks: to brute-force the key's password, the attacker needs to be on the target system and try to guess the keycard's password, which will lock itself after a limited number of tries. It also provides for a clean and standard interface to store keys offline: a single GnuPG command moves private key material to a keycard (the keytocard command in the --edit-key interface), whereas moving private key material to a LUKS-encrypted device or air-gapped computer is more complex. Keycards are also useful if you operate on multiple computers. A common problem when using GnuPG on multiple machines is how to safely copy and synchronize private key material among different devices, which introduces new security problems. Indeed, a "good rule of thumb in a forensics lab", according to Robert J. Hansen on the GnuPG mailing list, is to "store the minimum personal data possible on your systems". Keycards provide the best of both worlds here: you can use your private key on multiple computers without actually storing it in multiple places. In fact, Mike Gerwitz went as far as saying:

For users that need their GPG key on multiple boxes, I consider a smartcard to be essential. Otherwise, the user is just furthering her risk of compromise.

Keycard tradeoffs As Gerwitz hinted, there are multiple downsides to using a keycard, however. Another DD, Wouter Verhelst clearly expressed the tradeoffs:

Smartcards are useful. They ensure that the private half of your key is never on any hard disk or other general storage device, and therefore that it cannot possibly be stolen (because there's only one possible copy of it). Smartcards are a pain in the ass. They ensure that the private half of your key is never on any hard disk or other general storage device but instead sits in your wallet, so whenever you need to access it, you need to grab your wallet to be able to do so, which takes more effort than just firing up GnuPG. If your laptop doesn't have a builtin cardreader, you also need to fish the reader from your backpack or wherever, etc.

"Smartcards" here refer to older OpenPGP cards that relied on the IEC 7816 smartcard connectors and therefore needed a specially-built smartcard reader. Newer keycards simply use a standard USB connector. In any case, it's true that having an external device introduces new issues: attackers can steal your keycard, you can simply lose it, or wash it with your dirty laundry. A laptop or a computer can also be lost, of course, but it is much easier to lose a small USB keycard than a full laptop and I have yet to hear of someone shoving a full laptop into a washing machine. When you lose your keycard, unless a separate revocation certificate is available somewhere, you lose complete control of the key, which is catastrophic. But, even if you revoke the lost key, you need to create a new one, which involves rebuilding the web of trust for the key a rather expensive operation as it usually requires meeting other OpenPGP users in person to exchange fingerprints. You should therefore think about how to back up the certification key, which is a problem that already exists for online keys; of course, everyone has a revocation certificates and backups of their OpenPGP keys... right? In the keycard scenario, backups may be multiple keycards distributed geographically. Note that, contrary to an air-gapped system, a key generated on a keycard cannot be backed up, by design. For subkeys, this is not a problem as they do not need to be backed up (except encryption keys). But, for a certification key, this means users need to generate the key on the host and transfer it to the keycard, which means the host is expected to have enough entropy to generate cryptographic-strength random numbers, for example. Also consider the possibility of combining different approaches: you could, for example, use a keycard for day-to-day operation, but keep a backup of the certification key on a LUKS-encrypted offline volume. Keycards introduce a new element into the trust chain: you need to trust the keycard manufacturer to not have any hostile code in the key's firmware or hardware. In addition, you need to trust that the implementation is correct. Keycards are harder to update: the firmware may be deliberately inaccessible to the host for security reasons or may require special software to manipulate. Keycards may be slower than the CPU in performing certain operations because they are small embedded microcontrollers with limited computing power. Finally, keycards may encourage users to trust multiple machines with their secrets, which works against the "minimum personal data" principle. A completely different approach called the trusted physical console (TPC) does the opposite: instead of trying to get private key material onto all of those machines, just have them on a single machine that is used for everything. Unlike a keycard, the TPC is an actual computer, say a laptop, which has the advantage of needing no special procedure to manage keys. The downside is, of course, that you actually need to carry that laptop everywhere you go, which may be problematic, especially in some corporate environments that restrict bringing your own devices.

Quick keycard "howto" Getting keys onto a keycard is easy enough:

1. Start with a temporary key to test the procedure:

       export GNUPGHOME=$(mktemp -d)
       gpg --generate-key
   

2. Edit the key using its user ID (UID):

       gpg --edit-key UID
   

3. Use the key command to select the first subkey, then copy it to the keycard (you can also use the addcardkey command to just generate a new subkey directly on the keycard):

       gpg> key 1
       gpg> keytocard
   

4. If you want to move the subkey, use the save command, which will remove the local copy of the private key, so the keycard will be the only copy of the secret key. Otherwise use the quit command to save the key on the keycard, but keep the secret key in your normal keyring; answer "n" to "save changes?" and "y" to "quit without saving?" . This way the keycard is a backup of your secret key.
5. Once you are satisfied with the results, repeat steps 1 through 4 with your normal keyring (unset $GNUPGHOME)

When a key is moved to a keycard, --list-secret-keys will show it as sec> (or ssb> for subkeys) instead of the usual sec keyword. If the key is completely missing (for example, if you moved it to a LUKS container), the # sign is used instead. If you need to use a key from a keycard backup, you simply do gpg --card-edit with the key plugged in, then type the fetch command at the prompt to fetch the public key that corresponds to the private key on the keycard (which stays on the keycard). This is the same procedure as the one to use the secret key on another computer.

Conclusion There are already informal OpenPGP best-practices guides out there and some recommend storing keys offline, but they rarely explain what exactly that means. Storing your primary secret key offline is important in dealing with possible compromises and we examined the main ways of doing so: either with an air-gapped system, LUKS-encrypted keyring, or by using keycards. Each approach has its own tradeoffs, but I recommend getting familiar with keycards if you use multiple computers and want a standardized interface with minimal configuration trouble. And of course, those approaches can be combined. This tutorial, for example, uses a keycard on an air-gapped computer, which neatly resolves the question of how to transmit signatures between the air-gapped system and the world. It is definitely not for the faint of heart, however. Once one has decided to use a keycard, the next order of business is to choose a specific device. That choice will be addressed in a followup article, where I will look at performance, physical design, and other considerations.

This article first appeared in the Linux Weekly News.

• build-rdeps
  • Updated build-rdeps to work with compressed apt indices. (Debian bug #698240)
  • Added support for Build-Arch- Conflicts,Depends to build-rdeps. (adc87981)
  • Merged Andreas Henriksson's patch for setting remote.<name>.push-url when using debcheckout to clone a git repository. (Debian bug #753838)
• debsign
  • Updated bash completion for gpg keys to use gpg --with-colons, instead of manually parsing gpg -K output. Aside from being the Right Way to get machine parseable information out of gpg, it fixed completion when gpg is a 2.x version. (Debian bug #837380)

• Packaged the new upstream release (1.2.1)
• Basic package maintenance (-dbgsym package, policy update, enabled hardening flags).
• Uploaded 1.2.1-1

• Attempted to nudge lua-nvim's builds along on a couple architectures where they were waiting for neovim to be installable
  • x32: Temporarily removed lua-nvim Build-Depends to break the BD-Uninstallable cycle between lua-nvim and neovim.
  • powerpcspe: Temporarily removed luajit Build-Depends, reducing test scope, to fix the build.
    • If memory serves, the test failures are fixed upstream for the next release.
• Uploaded 0.2.0-4

• autobahn-cpp
• libdata-messagepack-perl
• ring

• Used the powerpc porterbox to debug and fix a 32-bit integer overflow that was causing test failures.
• Asked the vim-perl folks about getting updated runtime files to Bram, after Jakub Wilk filed Debian bug #873755. This had been fixed 4+ years earlier, but not yet merged back into Vim. Thanks to Rob Hoelz for pulling things together and sending the updates to Bram.
• I've continued to receive feedback from Debian users about their frustration with Vim's new "defaults.vim", both in regards to the actual default settings and its interaction with the system-wide vimrc file. While I still don't intend to deviate from upstream's behavior, I did push back some more on the existing behavior. I appreciate Christian Brabandt's effort, as always, to understand the issue at hand and have constructive discussions. His final suggestion seems like it will resolve the system vimrc interaction, so hopefully Bram is receptive to it.
• Uploaded 2:8.0.1144-1
• Thanks to a nudge from Salvatore Bonaccorso and Moritz M hlenhoff, I uploaded 2:8.0.0197-4+deb9u1 which fixes CVE-2017-11109. I had intended to do this much sooner, but it fell through the cracks. Due to Adam Barratt's quick responses, this should make it into the upcoming Stretch 9.2 release.

• Started work on updating the packaging
  • Converted to 3.0 (quilt) source format
  • Updated to debhelper 10 compat
  • Initial attempts at converting to a dh rules file
    • Running into various problems here and still trying to figure out whether they're in the upstream build system, Debian's patches, or both.

• Worked with Niko Dittmann to fix build failures Niko was experiencing on OpenBSD 6.1 #7298
• Merged upstream Vim patches into neovim from various contributors
  • Young Jean Han - #7233
  • KunMing Xie - #7258, #7311, #7309, #7310
  • James McCoy - #7325
• Discussed focus detection behavior after a recent change in the implementation (#7221)
  • While testing focus detection in various terminal emulators, I noticed pangoterm didn't support this. I submitted a merge request on libvterm to provide an API for reporting focus changes. If that's merged, it will be trivial for pangoterm to notify applications when the terminal has focus.
• Fixed a bug in our tooling around merging Vim patches, which was causing it to incorrectly drop certain files from the patches. #7328

• Submitted a pull request to Quadrapassel (the Gnome version of Tetris) to start a new game when the pause button is pressed outside of a game. This means you would no longer have to use the mouse to start a new game. [...]
• Made a large number of improvements to AptFS my FUSE-based filesystem that provides a view on unpacked Debian source packages as regular folders including moving away from manual parsing of package lists [...] and numerous code tidying/refactoring changes.
• Sent a small patch to django-sitetree, a Django library for menu and breadcrumb navigation elements to not mask test exit codes from the surrounding shell. [...]
• Updated travis.debian.net, my hosted service for projects that host their Debian packaging on GitHub to use the Travis CI continuous integration platform to test builds:
  • Add support for "sloppy" backports. Thanks to Bernd Zeimetz for the idea and ongoing testing. [...]
  • Merged a pull request from James McCoy to pass DEB_BUILD_PROFILES through to the build. [...]
  • Workaround Travis CI's HTTP proxy which does not appear to support SRV records. [...]
  • Run debc from devscripts if the build was successful [...] and output the .buildinfo file if it exists [...].
• Fixed a few issues in local-debian-mirror, my package to easily maintain and customise a local Debian mirror via the DebConf configuration tool:
  • Fix an issue where file permissions from the remote could result in a local archive that was impossible to access. [...]
  • Clear out empty directories on the local repository. [...]
• Updated django-staticfiles-dotd, my Django staticfiles adaptor to concatentate static media in .d-style directories to support Python 3.x by using bytes objects (commit) and move away from monkeypatch as it does not have a Python 3.x port yet (commit).
• I also posted a short essay to my blog entitled "Ask the Dumb Questions" as well as provided an update on the latest Lintian release.

[Unit]
Description=Beeps while a USB key is plugged
DefaultDependencies=false
StopWhenUnneeded=true
[Install]
WantedBy=dev-disk-by\x2dlabel-ERLUG.device
[Service]
Type=simple
ExecStart=/bin/sh -ec 'while true; do /usr/bin/aplay -q /tmp/beep.wav; sleep 2; done'

1. Run udevadm monitor --environment and plug the device
2. Look at the DEVNAME= values and pick one that addresses your device the way you prefer
3. udevadm info --attribute-walk --name=*the value of devname* will give you all you can use for matching in the udev rule.

• man systemd.device
• Using systemd and udev rules

$ sudo debootstrap stretch cross
$ echo "strech_cross"   sudo tee cross/etc/debian_chroot
$ sudo systemd-nspawn -D cross
# dpkg --add-architecture armhf
# echo "deb-src http://deb.debian.org/debian stretch main" >> /etc/apt/sources.list
# apt update
# apt install --no-install-recommends build-essential crossbuild-essential-armhf

# apt source bc
# cd bc-1.06.95/
# apt-get build-dep -a armhf .
# dpkg-buildpackage -aarmhf -j2 -b
 
dh_auto_configure -- --prefix=/usr --with-readline
        ./configure --build=x86_64-linux-gnu --prefix=/usr --includedir=\$ prefix /include --mandir=\$ prefix /share/man --infodir=\$ prefix /share/info --sysconfdir=/etc --localstatedir=/var --disable-silent-rules --libdir=\$ prefix /lib/arm-linux-gnueabihf --libexecdir=\$ prefix /lib/arm-linux-gnueabihf --disable-maintainer-mode --disable-dependency-tracking --host=arm-linux-gnueabihf --prefix=/usr --with-readline
 
dpkg-deb: building package 'dc-dbgsym' in '../dc-dbgsym_1.06.95-9_armhf.deb'.
dpkg-deb: building package 'bc-dbgsym' in '../bc-dbgsym_1.06.95-9_armhf.deb'.
dpkg-deb: building package 'dc' in '../dc_1.06.95-9_armhf.deb'.
dpkg-deb: building package 'bc' in '../bc_1.06.95-9_armhf.deb'.
 dpkg-genbuildinfo --build=binary
 dpkg-genchanges --build=binary >../bc_1.06.95-9_armhf.changes
dpkg-genchanges: info: binary-only upload (no source code included)
 dpkg-source --after-build bc-1.06.95
dpkg-buildpackage: info: binary-only upload (no source included)

# apt source pumpa
# cd pumpa-0.9.3/
# apt-get build-dep -a armhf .
# dpkg-buildpackage -aarmhf -j2 -b
 
        qmake -makefile -nocache "QMAKE_CFLAGS_RELEASE=-g -O2 -fdebug-prefix-map=/root/pumpa-0.9.3=.
          -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2"
          "QMAKE_CFLAGS_DEBUG=-g -O2 -fdebug-prefix-map=/root/pumpa-0.9.3=. -fstack-protector-strong
          -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2"
          "QMAKE_CXXFLAGS_RELEASE=-g -O2 -fdebug-prefix-map=/root/pumpa-0.9.3=. -fstack-protector-strong
          -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2"
          "QMAKE_CXXFLAGS_DEBUG=-g -O2 -fdebug-prefix-map=/root/pumpa-0.9.3=. -fstack-protector-strong
          -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2"
          "QMAKE_LFLAGS_RELEASE=-Wl,-z,relro -Wl,-z,now"
          "QMAKE_LFLAGS_DEBUG=-Wl,-z,relro -Wl,-z,now" QMAKE_STRIP=: PREFIX=/usr
qmake: could not exec '/usr/lib/x86_64-linux-gnu/qt5/bin/qmake': No such file or directory
 
debian/rules:19: recipe for target 'build' failed
make: *** [build] Error 2
dpkg-buildpackage: error: debian/rules build gave error exit status 2

# apt source caneda
# cd caneda-0.3.0/
# apt-get build-dep -a armhf .
# dpkg-buildpackage -aarmhf -j2 -b
 
        cmake .. -DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_VERBOSE_MAKEFILE=ON -DCMAKE_BUILD_TYPE=None
          -DCMAKE_INSTALL_SYSCONFDIR=/etc -DCMAKE_INSTALL_LOCALSTATEDIR=/var -DCMAKE_SYSTEM_NAME=Linux
          -DCMAKE_SYSTEM_PROCESSOR=arm -DCMAKE_C_COMPILER=arm-linux-gnueabihf-gcc
          -DCMAKE_CXX_COMPILER=arm-linux-gnueabihf-g\+\+
          -DPKG_CONFIG_EXECUTABLE=/usr/bin/arm-linux-gnueabihf-pkg-config
          -DCMAKE_INSTALL_LIBDIR=lib/arm-linux-gnueabihf
 
CMake Error at /usr/lib/arm-linux-gnueabihf/cmake/Qt5Core/Qt5CoreConfig.cmake:27 (message):
  The imported target "Qt5::Core" references the file
     "/usr/lib/arm-linux-gnueabihf/qt5/bin/moc"
  but this file does not exist.  Possible reasons include:
  * The file was deleted, renamed, or moved to another location.
  * An install or uninstall procedure did not complete successfully.
  * The installation package was faulty and contained
     "/usr/lib/arm-linux-gnueabihf/cmake/Qt5Core/Qt5CoreConfigExtras.cmake"
  but not all the files it references.

helmut> enrico: yes, cross building is painful. we have ~26000 source packages. of those, ~13000 build arch-dep packages. of those, ~6000 have cross-satisfiable build-depends. of those, I tried cross building ~2300. of those 1300 cross built. so we are at about 10% working. helmut> enrico: plus there are some 607 source packages affected by some 326 bugs with patches. helmut> enrico: gogo nmu them helmut> enrico: I've filed some 1000 bugs (most of them with patches) now. around 600 are fixed :)

mitya57> we have cross-building stuff on our TODO list, but it will likely require a lot of time and neither Lisandro nor I have it currently. mitya57> see https://gobby.debian.org/export/Teams/KDE/qt-cross for a summary of what needs to be done. mitya57> Any help or patches are always welcome :))

svuorela and others repeat that Qt upstream is not compatible with Debian's multiarch thinking, in that Qt upstream insists on having one toolchain for each pair of architectures, whereas the Debian way tends to be to make packages generic and split stuff such that it can be mixed and matched. An example being that you need to run qmake (thus you need qmake for the build architecture), but qmake also embeds the relevant paths and you need to query it for them (so you need qmake for the host architecture) Either you run it through qemu, or you have a particular cross qmake for your build/host pair, or you fix qt upstream to stop this madness

• can I have a qmake-$ARCH binary that can build a source using locally installed multiarch Qt libraries, do I need to recompile and ship the whole of Qt, or just qmake?
• is there a recipe for building a cross-building Qt environment that would be able use Debian development libraries installed the normal multiarch way?
• we can't do perfect yet, but can we do better than this?

• Christos Zoulas gave a talk entitled Reproducible builds on NetBSD at EuroBSDCon 2017

• Paul Eggert reported that TZ=UTC is not a portable setting for the TZ environment variable.

• Adrian Bunk:
  • #876615 filed against librsvg.
• Bernhard M. Wiedemann:
  • varnish (random IDs)
  • make (sort)
  • nautilus-dropbox (extended date)
  • fontforge (date)
  • votca-csg (merged, date)
  • freeipmi (merged, date)
  • libmypaint (merged, sort)
  • doomsday (merged, sort)
  • asciidoc (help make release with SOURCE_DATE_EPOCH patch)

• gtk_doc_api_index_full (fixed upstream: #779090)

• Adrian Bunk (56)
• Bas Couwenberg (1)
• Helmut Grohne (1)
• Nobuhiro Iwamatsu (2)

• Ximin Luo:
  • comparators/*:
    • Add a test for fallback_recognizes and improve the behaviour
    • Add a fallback_recognizes to work around file(1) #876316. (Closes: #875272)
    • If --force-details then don't skip files with identical md5sums either
    • Add a --force-details flag for debugging
  • presenters.html:
    • Restore the previous more-detailed comment
    • Don't show pointer-cursor when jQuery is disabled
    • Prune all descendants properly (Closes: #875281)
  • difference.py:
    • Also copy self._comment properly, compare self._visuals in equals()
    • In fmap/map_lines, don't forget about self._visuals
    • Use diff_split_lines everywhere
  • tests:
    • Add case for #875281
    • Make test_md5sums less brittle
    • Update test_deb for new string
  • readers: Convert bytes to str in the right place
  • config: Force-set a value if it must be < another and it was not set on purpose (Closes: #875451)
  • Bump minimum Python version to 3.5 as we use syntax introduced by PEP 448
• Chris Lamb:
  • Print a debugging message if we are reading diff from stdin.
  • Compare types with identity not equality.
  • diffoscope.presenters.html: Use logging.py's lazy argument interpolation.
  • debian/control: Bump Standards-Version to 4.1.0.
• Mattia Rizzolo:
  • debian/rules: Place "before" commands before "after" commands.

• Chris Lamb:
  • Log which handler procesed a file. (Closes: #876140)
  • Bump Standards-Version to 4.1.0.

• Ximin Luo:
  • Push use of UNIX return codes to the edges of the program
  • Add a --auto-build option to determine which variations cause unreproducibility
  • Allow umask and user_group to both vary at the same time
  • Generate build names in main instead of build, guard against dupes
  • Pull traceback-printing stuff out of the core code
  • More refactoring, make check() contain only logic that would be changed in an auto-detector
  • Split check() into a coroutine producer and consumer, prepares auto-detection

• Re-add and armhf build node that had been disabled due to performance issues, but works linux 4.14-rc1 now! #876212

• Use botch from stretch to fix the jenkins job which create the package sets. (botch is currently uninstallable in sid and from pre-stretch-release times we used a sid schroot to install and use botch.)

conn V2-1
  left        = 2001:db8:1::1
  leftsubnet  = 2001:db8:a1::/64
  right       = 2001:db8:2::1
  rightsubnet = 2001:db8:a2::/64
  authby      = psk
  auto        = route

$ ip xfrm policy
src 2001:db8:a1::/64 dst 2001:db8:a2::/64
        dir out priority 399999 ptype main
        tmpl src 2001:db8:1::1 dst 2001:db8:2::1
                proto esp reqid 4 mode tunnel
src 2001:db8:a2::/64 dst 2001:db8:a1::/64
        dir fwd priority 399999 ptype main
        tmpl src 2001:db8:2::1 dst 2001:db8:1::1
                proto esp reqid 4 mode tunnel
src 2001:db8:a2::/64 dst 2001:db8:a1::/64
        dir in priority 399999 ptype main
        tmpl src 2001:db8:2::1 dst 2001:db8:1::1
                proto esp reqid 4 mode tunnel
[ ]

• a direction (out, in or fwd²),
• a selector (source subnet, destination subnet, protocol, ports),
• a mode (transport or tunnel),
• an encapsulation protocol (esp or ah), and
• the endpoint source and destination addresses.

$ ip xfrm state
src 2001:db8:1::1 dst 2001:db8:2::1
        proto esp spi 0xc1890b6e reqid 4 mode tunnel
        replay-window 0 flag af-unspec
        auth-trunc hmac(sha256) 0x5b68[ ]8ba2904 128
        enc cbc(aes) 0x8e0e377ad8fd91e8553648340ff0fa06
        anti-replay context: seq 0x0, oseq 0x0, bitmap 0x00000000
[ ]

$ tcpdump -pni eth0 -c2 -s0 esp
13:07:30.871150 IP6 2001:db8:1::1 > 2001:db8:2::1: ESP(spi=0xc1890b6e,seq=0x222)
13:07:30.872297 IP6 2001:db8:2::1 > 2001:db8:1::1: ESP(spi=0xcf2426b6,seq=0x204)

conn V1-1-to-V2-1
  left        = 2001:db8:1::1
  leftsubnet  = 2001:db8:a1::/64,2001:db8:a6::cc:1/128,2001:db8:a6::cc:5/128
  right       = 2001:db8:2::1
  rightsubnet = 2001:db8:a2::/64,2001:db8:a6::/64,2001:db8:a8::/64
  authby      = psk
  keyexchange = ikev2
  auto        = route

1. Routing daemons can be used to distribute routes to be protected by the VPN. This decreases the administrative burden when many subnets are present on each side.
2. Encapsulation and decapsulation can be executed in a different routing instance or namespace. This enables a clean separation between a private routing instance (where VPN users are) and a public routing instance (where VPN endpoints are).

Route-based VPN on Juniper Before looking at how to achieve that on Linux, let s have a look at the way it works with a JunOS-based platform (like a Juniper vSRX). This platform as long-standing history of supporting route-based VPNs (a feature already present in the Netscreen ISG platform). Let s assume we want to configure the IPsec VPN from V3-2 to V1-1. First, we need to configure the tunnel interface and bind it to the private routing instance containing only internal routes (with IPv4, they would have been RFC 1918 routes):

interfaces  
    st0  
        unit 1  
            family inet6  
                address 2001:db8:ff::7/127;
             
         
     
 
routing-instances  
    private  
        instance-type virtual-router;
        interface st0.1;
     
 

The second step is to configure the VPN:

security  
    /* Phase 1 configuration */
    ike  
        proposal IKE-P1  
            authentication-method pre-shared-keys;
            dh-group group20;
            encryption-algorithm aes-256-gcm;
         
        policy IKE-V1-1  
            mode main;
            proposals IKE-P1;
            pre-shared-key ascii-text "d8bdRxaY22oH1j89Z2nATeYyrXfP9ga6xC5mi0RG1uc";
         
        gateway GW-V1-1  
            ike-policy IKE-V1-1;
            address 2001:db8:1::1;
            external-interface lo0.1;
            general-ikeid;
            version v2-only;
         
     
    /* Phase 2 configuration */
    ipsec  
        proposal ESP-P2  
            protocol esp;
            encryption-algorithm aes-256-gcm;
         
        policy IPSEC-V1-1  
            perfect-forward-secrecy keys group20;
            proposals ESP-P2;
         
        vpn VPN-V1-1  
            bind-interface st0.1;
            df-bit copy;
            ike  
                gateway GW-V1-1;
                ipsec-policy IPSEC-V1-1;
             
            establish-tunnels on-traffic;
         
     
 

We get a route-based VPN because we bind the st0.1 interface to the VPN-V1-1 VPN. Once the VPN is up, any packet entering st0.1 will be encapsulated and sent to the 2001:db8: 1::1 endpoint. The last step is to configure BGP in the private routing instance to exchange routes with the remote site:

routing-instances  
    private  
        routing-options  
            router-id 1.0.3.2;
            maximum-paths 16;
         
        protocols  
            bgp  
                preference 140;
                log-updown;
                group v4-VPN  
                    type external;
                    local-as 65003;
                    hold-time 6;
                    neighbor 2001:db8:ff::6 peer-as 65001;
                    multipath;
                    export [ NEXT-HOP-SELF OUR-ROUTES NOTHING ];
                 
             
         
     
 

The export filter OUR-ROUTES needs to select the routes to be advertised to the other peers. For example:

policy-options  
    policy-statement OUR-ROUTES  
        term 10  
            from  
                protocol ospf3;
                route-type internal;
             
            then  
                metric 0;
                accept;
             
         
     
 

The configuration needs to be repeated for the other peers. The complete version is available on GitHub. Once the BGP sessions are up, we start learning routes from the other sites. For example, here is the route for 2001:db8: a1::/64:

> show route 2001:db8:a1::/64 protocol bgp table private.inet6.0 best-path
private.inet6.0: 15 destinations, 19 routes (15 active, 0 holddown, 0 hidden)
+ = Active Route, - = Last Active, * = Both
2001:db8:a1::/64   *[BGP/140] 01:12:32, localpref 100, from 2001:db8:ff::6
                      AS path: 65001 I, validation-state: unverified
                      to 2001:db8:ff::6 via st0.1
                    > to 2001:db8:ff::14 via st0.2

It was learnt both from V1-1 (through st0.1) and V1-2 (through st0.2). The route is part of the private routing instance but encapsulated packets are sent/received in the public routing instance. No route-leaking is needed for this configuration. The VPN cannot be used as a gateway from internal hosts to external hosts (or vice-versa). This could also have been done with JunOS security policies (stateful firewall rules) but doing the separation with routing instances also ensure routes from different domains are not mixed and a simple policy misconfiguration won t lead to a disaster.

Route-based VPN on Linux Starting from Linux 3.15, a similar configuration is possible with the help of a virtual tunnel interface³. First, we create the private namespace:

# ip netns add private
# ip netns exec private sysctl -qw net.ipv6.conf.all.forwarding=1

Any private interface needs to be moved to this namespace (no IP is configured as we can use IPv6 link-local addresses):

# ip link set netns private dev eth1
# ip link set netns private dev eth2
# ip netns exec private ip link set up dev eth1
# ip netns exec private ip link set up dev eth2

Then, we create vti6, a tunnel interface (similar to st0.1 in the JunOS example):

# ip tunnel add vti6 \
   mode vti6 \
   local 2001:db8:1::1 \
   remote 2001:db8:3::2 \
   key 6
# ip link set netns private dev vti6
# ip netns exec private ip addr add 2001:db8:ff::6/127 dev vti6
# ip netns exec private sysctl -qw net.ipv4.conf.vti6.disable_policy=1
# ip netns exec private sysctl -qw net.ipv4.conf.vti6.disable_xfrm=1
# ip netns exec private ip link set vti6 mtu 1500
# ip netns exec private ip link set vti6 up

The tunnel interface is created in the initial namespace and moved to the private one. It will remember its original namespace where it will process encapsulated packets. Any packet entering the interface will temporarily get a firewall mark of 6 that will be used only to match the appropriate IPsec policy⁴ below. The kernel sets a low MTU on the interface to handle any possible combination of ciphers and protocols. We set it to 1500 and let PMTUD do its work. We can then configure strongSwan⁵:

conn V3-2
  left        = 2001:db8:1::1
  leftsubnet  = ::/0
  right       = 2001:db8:3::2
  rightsubnet = ::/0
  authby      = psk
  mark        = 6
  auto        = route
  keyexchange = ikev2
  keyingtries = %forever
  ike         = aes256gcm16-prfsha384-ecp384!
  esp         = aes256gcm16-prfsha384-ecp384!
  mobike      = no

The IKE daemon configures the following policies in the kernel:

$ ip xfrm policy
src ::/0 dst ::/0
        dir out priority 399999 ptype main
        mark 0x6/0xffffffff
        tmpl src 2001:db8:1::1 dst 2001:db8:3::2
                proto esp reqid 1 mode tunnel
src ::/0 dst ::/0
        dir fwd priority 399999 ptype main
        mark 0x6/0xffffffff
        tmpl src 2001:db8:3::2 dst 2001:db8:1::1
                proto esp reqid 1 mode tunnel
src ::/0 dst ::/0
        dir in priority 399999 ptype main
        mark 0x6/0xffffffff
        tmpl src 2001:db8:3::2 dst 2001:db8:1::1
                proto esp reqid 1 mode tunnel
[ ]

Those policies are used for any source or destination as long as the firewall mark is equal to 6, which matches the mark configured for the tunnel interface. The last step is to configure BGP to exchange routes. We can use BIRD for this:

router id 1.0.1.1;
protocol device  
   scan time 10;
 
protocol kernel  
   persist;
   learn;
   import all;
   export all;
   merge paths yes;
 
protocol bgp IBGP_V3_2  
   local 2001:db8:ff::6 as 65001;
   neighbor 2001:db8:ff::7 as 65003;
   import all;
   export where ifname ~ "eth*";
   preference 160;
   hold time 6;
 

Once BIRD is started in the private namespace, we can check routes are learned correctly:

$ ip netns exec private ip -6 route show 2001:db8:a3::/64
2001:db8:a3::/64 proto bird metric 1024
        nexthop via 2001:db8:ff::5  dev vti5 weight 1
        nexthop via 2001:db8:ff::7  dev vti6 weight 1

The above route was learnt from both V3-1 (through vti5) and V3-2 (through vti6). Like for the JunOS version, there is no route-leaking between the private namespace and the initial one. The VPN cannot be used as a gateway between the two namespaces, only for encapsulation. This also prevent a misconfiguration (for example, IKE daemon not running) from allowing packets to leave the private network. As a bonus, unencrypted traffic can be observed with tcpdump on the tunnel interface:

$ ip netns exec private tcpdump -pni vti6 icmp6
tcpdump: verbose output suppressed, use -v or -vv for full protocol decode
listening on vti6, link-type LINUX_SLL (Linux cooked), capture size 262144 bytes
20:51:15.258708 IP6 2001:db8:a1::1 > 2001:db8:a3::1: ICMP6, echo request, seq 69
20:51:15.260874 IP6 2001:db8:a3::1 > 2001:db8:a1::1: ICMP6, echo reply, seq 69

You can find all the configuration files for this example on GitHub. The documentation of strongSwan also features a page about route-based VPNs.

---

1. Everything in this post should work with Libreswan.
2. fwd is for incoming packets on non-local addresses. It only makes sense in transport mode and is a Linux-only particularity.
3. Virtual tunnel interfaces (VTI) were introduced in Linux 3.6 (for IPv4) and Linux 3.12 (for IPv6). Appropriate namespace support was added in 3.15. KLIPS, an alternative out-of-tree stack available since Linux 2.2, also features tunnel interfaces.
4. The mark is set right before doing a policy lookup and restored after that. Consequently, it doesn t affect other possible uses (filtering, routing). However, as Netfilter can also set a mark, one should be careful for conflicts.
5. The ciphers used here are the strongest ones currently possible while keeping compatibility with JunOS. The documentation for strongSwan contains a complete list of supported algorithms as well as security recommendations to choose them.

Context aka. Sections This is probably the most relevant of the three proposals. Consider a bunch of related functions, say analyseExpr and analyseAlt, like these:

analyseExpr :: Expr -> Expr
analyseExpr (Var v) = change v
analyseExpr (App e1 e2) =
  App (analyseExpr e1) (analyseExpr e2)
analyseExpr (Lam v e) = Lam v (analyseExpr flag e)
analyseExpr (Case scrut alts) =
  Case (analyseExpr scrut) (analyseAlt <$> alts)
analyseAlt :: Alt -> Alt
analyseAlt (dc, pats, e) = (dc, pats, analyseExpr e)

You have written them, but now you notice that you need to make them configurable, e.g. to do different things in the Var case. You thus add a parameter to all these functions, and hence an argument to every call:

type Flag = Bool
analyseExpr :: Flag -> Expr -> Expr
analyseExpr flag (Var v) = if flag then change1 v else change2 v
analyseExpr flag (App e1 e2) =
  App (analyseExpr flag e1) (analyseExpr flag e2)
analyseExpr flag (Lam v e) = Lam v (analyseExpr (not flag) e)
analyseExpr flag (Case scrut alts) =
  Case (analyseExpr flag scrut) (analyseAlt flag <$> alts)
analyseAlt :: Flag -> Alt -> Alt
analyseAlt flag (dc, pats, e) = (dc, pats, analyseExpr flag e)

I find this code problematic. The intention was: flag is a parameter that an external caller can use to change the behaviour of this code, but when reading and reasoning about this code, flag should be considered constant. But this intention is neither easily visible nor enforced. And in fact, in the above code, flag does change , as analyseExpr passes something else in the Lam case. The idiom is indistinguishable from the environment idiom, where a locally changing environment (such as variables in scope ) is passed around. So we are facing exactly the same problem as when reasoning about a loop in an imperative program with mutable variables. And we (pure functional programmers) should know better: We cherish immutability! We want to bind our variables once and have them scope over everything we need to scope over! The solution I d like to see in Haskell is common in other languages (Gallina, Idris, Agda, Isar), and this is what it would look like here:

type Flag = Bool
section (flag :: Flag) where
  analyseExpr :: Expr -> Expr
  analyseExpr (Var v) = if flag then change1 v else change2v 
  analyseExpr (App e1 e2) =
    App (analyseExpr e1) (analyseExpr e2)
  analyseExpr (Lam v e) = Lam v (analyseExpr e)
  analyseExpr (Case scrut alts) =
    Case (analyseExpr scrut) (analyseAlt <$> alts)
  analyseAlt :: Alt -> Alt
  analyseAlt (dc, pats, e) = (dc, pats, analyseExpr e)

Now the intention is clear: Within a clearly marked block, flag is fixed and when reasoning about this code I do not have to worry that it might change. Either all variables will be passed to change1, or all to change2. An important distinction! Therefore, inside the section, the type of analyseExpr does not mention Flag, whereas outside its type is Flag -> Expr -> Expr. This is a bit unusual, but not completely: You see precisely the same effect in a class declaration, where the type signature of the methods do not mention the class constraint, but outside the declaration they do. Note that idioms like implicit parameters or the Reader monad do not give the guarantee that the parameter is (locally) constant. More details can be found in the GHC proposal that I prepared, and I invite you to raise concern or voice support there. Curiously, this problem must have bothered me for longer than I remember: I discovered that seven years ago, I wrote a Template Haskell based implementation of this idea in the seal-module package!

Less parentheses 1: Bulleted argument lists The next two proposals are all about removing parentheses. I believe that Haskell s tendency to express complex code with no or few parentheses is one of its big strengths, as it makes it easier to visualy parse programs. A common idiom is to use the $ operator to separate a function from a complex argument without parentheses, but it does not help when there are multiple complex arguments. For that case I propose to steal an idea from the surprisingly successful markup language markdown, and use bulleted lists to indicate multiple arguments:

foo :: Baz
foo = bracket
          some complicated code
          that is evaluated first
          other complicated code for later
          even more complicated code

I find this very easy to visually parse and navigate. It is actually possible to do this now, if one defines ( ) = id with infixl 0 . A dedicated syntax extension (-XArgumentBullets) is preferable:

• It only really adds readability if the bullets are nicely vertically aligned, which the compiler should enforce.
• I would like to use $ inside these complex arguments, and multiple operators of precedence 0 do not mix. (infixl -1 would help).
• It should be possible to nest these, and distinguish different nesting levers based on their indentation.

Less parentheses 1: Whitespace precedence The final proposal is the most daring. I am convinced that it improves readability and should be considered when creating a new language. As for Haskell, I am at the moment not proposing this as a language extension (but could be convinced to do so if there is enough positive feedback). Consider this definition of append:

(++) :: [a] -> [a] -> [a]
[]     ++ ys = ys
(x:xs) ++ ys = x : (xs++ys)

Imagine you were explaining the last line to someone orally. How would you speak it? One common way to do so is to not read the parentheses out aloud, but rather speak parenthesised expression more quickly and add pauses otherwise. We can do the same in syntax!

(++) :: [a] -> [a] -> [a]
[]   ++ ys = ys
x:xs ++ ys = x : xs++ys

The rule is simple: A sequence of tokens without any space is implicitly parenthesised. The reaction I got in Oxford was horror and disgust. And that is understandable we are very used to ignore spacing when parsing expressions (unless it is indentation, of course. Then we are no longer horrified, as our non-Haskell colleagues are when they see our code). But I am convinced that once you let the rule sink in, you will have no problem parsing such code with ease, and soon even with greater ease than the parenthesised version. It is a very natural thing to look at the general structure, identify compact chunks of characters , mentally group them, and then go and separately parse the internals of the chunks and how the chunks relate to each other. More natural than first scanning everything for ( and ), matching them up, building a mental tree, and then digging deeper. Incidentally, there was a non-programmer present during my presentation, and while she did not openly contradict the dismissive groan of the audience, I later learned that she found this variant quite obvious to understand and more easily to read than the parenthesised code. Some FAQs about this:

• What about an operator with space on one side but not on the other? I d simply forbid that, and hence enforce readable code.
• Do operator sections still require parenthesis? Yes, I d say so.
• Does this overrule operator precedence? Yes! a * b+c == a * (b+c).
• What is a token? Good question, and I am not not decided. In particular: Is a parenthesised expression a single token? If so, then (Succ a)+b * c parses as ((Succ a)+b) * c, otherwise it should probably simply be illegal.
• Can we extend this so that one space binds tighter than two spaces, and so on? Yes we can, but really, we should not.
• This is incompatible with Agda s syntax! Indeed it is, and I really like Agda s mixfix syntax. Can t have everything.
• Has this been done before? I have not seen it in any language, but Lewis Wall has blogged this idea before.

Well, let me know what you think!

• Once again, we reviewed our current task state
• I submitted a series which fixed our test suite for Git 2.13 which was an FTBFS bug submitted against the Debian package for Gitano. Richard Maw reviewed and merged it.
• Richard Maw supplied a series to add testing for dangling HEAD syndrome. I reviewed and merged that.
• Richard Maw submitted a patch to improve the auditability of the 'as' command and I reviewed and merged that.
• Richard Ipsum submitted a patch to add reader/writer configs to ease simple project management in Gitano. I'm not proud to say that I was too busy to look at this and ended up saying it was unlikely it'd get in. Richard Maw, quite rightly, took umbrage at that and worked on the patch, eventually submitting a new series with tests which I then felt obliged to review and I merged the series eventually.

  This is an excellent example of where just because one person is too busy doesn't mean that a good idea should be dropped, and I am grateful to Richard Maw for getting this work mergeable and effectively guilt-tripping me into reviewing/merging. This is a learnable moment for me and I hope to do better into the future.

• During all that time, I was working on a plugin to support git-multimail.py in Gitano. This work ranged across hooks and caused me to spend a long time thinking about the semantics of configuration overriding etc. Fortunately I got there in the end, and with a massive review effort from Richard Maw, we got it merged into Gitano.
• Finally I submitted a patch which caused the tests we run in Gitano to run from an 'install' directory which ensures that we catch bugs such as those which happened in earlier series where we missed out rules files for installation etc. Richard Maw reviewed and merged that.
• And then we released the new version of Gitano and subsidiary libraries. There was Luxio version 13 which switched us to readdir() from readdir_r() thanks to Richard Ipsum; Gall 1.3 which contained a bunch of build cleanups, and also a revparse_single() implementation in the C code to speed things up thanks to Richard Maw; Supple 1.0.8 which improved wrapper environment cleanups thanks to Richard Ipsum, allowed baking of paths in which means Nix is easier to support (again thanks to Richard Ipsum), fixed setuid handling so that Nix is easier to support (guess what? Richard Ipsum again); Lace 1.4 which now verifies definition names in allow/deny/anyof/allof and also produces better error messages from nested includes. And, of course, Gitano 1.1 whose changes were somewhat numerous and so you are invited to read them in the Gitano NEWS file for the release.

Not Gitano Of course, not everything I did in August was Gitano related. In fact once I had completed the 1.1 release and uploaded everything to Debian I decided that I was going to take a break from Gitano until the next developer day. (In fact there's even some patch series still unread on the mailing list which I will get to when I start the developer day.) I have long been interested in STM32 microcontrollers, using them in a variety of projects including the Entropy Key which some of you may remember. Jorge Aparicio was working on Cortex-M3 support (among other microcontrollers) in Rust and he then extended that to include a realtime framework called RTFM and from there I got interested in what I might be able to do with Rust on STM32. I noticed that there weren't any pure Rust implementations of the USB device stack which would be necessary in order to make a device, programmed in Rust, appear on a USB port for a computer to control/use. This tweaked my interest. As many of my readers are aware, I am very bad at doing things without some external motivation. As such, I then immediately offered to give a talk at a conference which should be happening in November, just so that I'd be forced to get on with learning and implementing the stack. I have been chronicling my work in this blog, and you're encouraged to go back and read them if you have similar interests. I'm sure that as my work progresses, I'll be doing more and more of that and less of Gitano, for at least the next two months. To bring that into context as F/LOSS work, I did end up submitting some patches to Jorge's STM32F103xx repository to support a couple more clock configuration entries so that USB and ADCs can be set up cleanly. So at least there's that.

Remember, Mastodon is a new decentralized social network, based on a free software which is rapidly gaining users (already there is more than 1.5 million accounts). As I ve created my account in June, I was a fast addict and I ve already created several tools for this network, Feed2toot, Remindr and Boost (mostly written in Python).

Now, with all this experience I have to stress out the importance of choosing the correct Mastodon instance.

Some technical reminders on how Mastodon works

First, let s quickly clarify something about the decentralized part. In Mastodon, decentralization is made through a federation of dedicated servers, called instances , each one with a complete independent administration. Your user account is created on one specific instance. You have two choices:

• Create your own instance. Which requires advanced technical knowledge.
• Create your user account on a public instance. Which is the easiest and fastest way to start using Mastodon.

You can move your user account from one instance to another, but you have to follow a special procedure which can be quite long, considering your own interest for technical manipulation and the total amount of your followers you ll have to warn about your change. As such, you ll have to create another account on a new instance and import three lists: the one with your followers, the one with the accounts you have blocked, and the one with the account you have muted.

From this working process, several technical and human factors will interest us.

A good technical administration for instance

As a social network, Mastodon is truly decentralized, with more than 1.5 million users on more than 2350 existing instances. As such, the most common usage is to create an account on an open instance. To create its own instance is way too difficult for the average user. Yet, using an open instance creates a strong dependence on the technical administrator of the chosen instance.

The technical administrator will have to deal with several obligations to ensure its service continuity, with high-quality hardware and regular back-ups. All of these have a price, either in money and in time. Regarding the time factor, it would be better to choose an administration team over an individual, as life events can change quite fast everyone s interests. As such, Framasoft, a French association dedicated to promoting the Free software use, offers its own Mastodon instance named: Framapiaf. The creator of the mastodon project, also offers a quite solid instance, Mastodon.social (see below).

Regarding the money factor, many instance administrators with a large number of users are currently asking for donation via Patreon, as hosting an instance server or renting one cost money.

Mastodon.social, the first instance of the Mastodon network

The Ideological Trend Of Your Instance

If anybody could have guessed the previous technical points since the recent registration explosion on the Mastodon social network, the following point took almost everyone by surprise. Little by little, different instances show their culture , their protest action, and their propaganda on this social network.

As the instance administrator has all the powers over its instance, he or she can block the instance of interacting with some other instances, or ban its instance s users from any interaction with other instances users.

With everyone having in mind the main advantages to have federalized instance from, this partial independence of some instances from the federation was a huge surprise. One of the most recent example was when the Unixcorn.xyz instance administrator banned its users from reading Aeris account, which was on its own instance. It was a cataclysm with several consequences, which I ve named the #AerisGate as it shows the different views on moderation and on its reception by various Mastodon users.

If you don t manage your own instance, when you ll have to choose the one where to create your account, make sure that the content you plan to toot is within the rules and compatible with the ideology of said instance s administrator. Yes, I know, it may seem surprising but, as stated above, by entering a public instance you become dependent on someone else s infrastructure, who may have an ideological way to conceive its Mastodon hosting service. As such, if you re a nazi, for example, don t open your Mastodon account on a far-left LGBT instance. Your account wouldn t stay open for long.

The moderation rules are described in the about/more page of the instance, and may contain ideological elements.

To ease the process for newcomers, it is now possible to use a great tool to select what instance should be the best to host your account.

Remember that, as stated above, Mastodon is decentralized, and as such there is no central authority which can be reached in case you have a conflict with your instance administrator. And nobody can force said administrator to follow its own rules, or not to change them on the fly. Think Twice Before Creating Your Account

If you want to create an account on an instance you don t control, you need to check two elements: the availability of the instance hosting service in the long run, often linked to the administrator or the administration group of said instance, and the ideological orientation of your instance. With these two elements checked, you ll be able to let your Mastodon account growth peacefully, without fearing an outage of your instance, or simple your account blocked one morning because it doesn t align with your instance s ideological line.

in Conclusion

To help me get involved in free software and writing articles for this blog, please consider a donation through my Liberapay page, even if it s only a few cents per week. My contact Bitcoin and Monero are also available on this page. Follow me on Mastodon Translated from French to English by St phanie Chaptal.

Lookup trie implementation Looking up a prefix in a routing table comes down to find the most specific entry matching the requested destination. A common structure for this task is the trie, a tree structure where each node has its parent as prefix. With IPv4, Linux uses a level-compressed trie (or LPC-trie), providing good performances with low memory usage. For IPv6, Linux uses a more classic radix tree (or Patricia trie). There are three reasons for not sharing:

• The IPv6 implementation (introduced in Linux 2.1.8, 1996) predates the IPv4 implementation based on LPC-tries (in Linux 2.6.13, commit 19baf839ff4a).
• The feature set is different. Notably, IPv6 supports source-specific routing¹ (since Linux 2.1.120, 1998).
• The IPv4 address space is denser than the IPv6 address space. Level-compression is therefore quite efficient with IPv4. This may not be the case with IPv6.

The trie in the below illustration encodes 6 prefixes: For more in-depth explanation on the different ways to encode a routing table into a trie and a better understanding of radix trees, see the explanations for IPv4. The following figure shows the in-memory representation of the previous radix tree. Each node corresponds to a struct fib6_node. When a node has the RTN_RTINFO flag set, it embeds a pointer to a struct rt6_info containing information about the next-hop. The fib6_lookup_1() function walks the radix tree in two steps:

1. walking down the tree to locate the potential candidate, and
2. checking the candidate and, if needed, backtracking until a match.

Here is a slightly simplified version without source-specific routing:

static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
                                       struct in6_addr  *addr)
 
    struct fib6_node *fn;
    __be32 dir;
    /* Step 1: locate potential candidate */
    fn = root;
    for (;;)  
        struct fib6_node *next;
        dir = addr_bit_set(addr, fn->fn_bit);
        next = dir ? fn->right : fn->left;
        if (next)  
            fn = next;
            continue;
         
        break;
     
    /* Step 2: check prefix and backtrack if needed */
    while (fn)  
        if (fn->fn_flags & RTN_RTINFO)  
            struct rt6key *key;
            key = fn->leaf->rt6i_dst;
            if (ipv6_prefix_equal(&key->addr, addr, key->plen))  
                if (fn->fn_flags & RTN_RTINFO)
                    return fn;
             
         
        if (fn->fn_flags & RTN_ROOT)
            break;
        fn = fn->parent;
     
    return NULL;
 

Caching While IPv4 lost its route cache in Linux 3.6 (commit 5e9965c15ba8), IPv6 still has a caching mechanism. However cache entries are directly put in the radix tree instead of a distinct structure. Since Linux 2.1.30 (1997) and until Linux 4.2 (commit 45e4fd26683c), almost any successful route lookup inserts a cache entry in the radix tree. For example, a router forwarding a ping between 2001:db8:1::1 and 2001:db8:3::1 would get those two cache entries:

$ ip -6 route show cache
2001:db8:1::1 dev r2-r1  metric 0
    cache
2001:db8:3::1 via 2001:db8:2::2 dev r2-r3  metric 0
    cache

These entries are cleaned up by the ip6_dst_gc() function controlled by the following parameters:

$ sysctl -a   grep -F net.ipv6.route
net.ipv6.route.gc_elasticity = 9
net.ipv6.route.gc_interval = 30
net.ipv6.route.gc_min_interval = 0
net.ipv6.route.gc_min_interval_ms = 500
net.ipv6.route.gc_thresh = 1024
net.ipv6.route.gc_timeout = 60
net.ipv6.route.max_size = 4096
net.ipv6.route.mtu_expires = 600

The garbage collector is triggered at most every 500 ms when there are more than 1024 entries or at least every 30 seconds. The garbage collection won t run for more than 60 seconds, except if there are more than 4096 routes. When running, it will first delete entries older than 30 seconds. If the number of cache entries is still greater than 4096, it will continue to delete more recent entries (but no more recent than 512 jiffies, which is the value of gc_elasticity) after a 500 ms pause. Starting from Linux 4.2 (commit 45e4fd26683c), only a PMTU exception would create a cache entry. A router doesn t have to handle those exceptions, so only hosts would get cache entries. And they should be pretty rare. Martin KaFai Lau explains:

Out of all IPv6 RTF_CACHE routes that are created, the percentage that has a different MTU is very small. In one of our end-user facing proxy server, only 1k out of 80k RTF_CACHE routes have a smaller MTU. For our DC traffic, there is no MTU exception.

Here is how a cache entry with a PMTU exception looks like:

$ ip -6 route show cache
2001:db8:1::50 via 2001:db8:1::13 dev out6  metric 0
    cache  expires 573sec mtu 1400 pref medium

Performance We consider three distinct scenarios:

Excerpt of an Internet full view

In this scenario, Linux acts as an edge router attached to the default-free zone. Currently, the size of such a routing table is a little bit above 40,000 routes.

/48 prefixes spread linearly with different densities

Linux acts as a core router inside a datacenter. Each customer or rack gets one or several /48 networks, which need to be routed around. With a density of 1, /48 subnets are contiguous.

/128 prefixes spread randomly in a fixed /108 subnet

Linux acts as a leaf router for a /64 subnet with hosts getting their IP using autoconfiguration. It is assumed all hosts share the same OUI and therefore, the first 40 bits are fixed. In this scenario, neighbor reachability information for the /64 subnet are converted into routes by some external process and redistributed among other routers sharing the same subnet².

Route lookup performance With the help of a small kernel module, we can accurately benchmark³ the ip6_route_output_flags() function and correlate the results with the radix tree size: Getting meaningful results is challenging due to the size of the address space. None of the scenarios have a fallback route and we only measure time for successful hits⁴. For the full view scenario, only the range from 2400::/16 to 2a06::/16 is scanned (it contains more than half of the routes). For the /128 scenario, the whole /108 subnet is scanned. For the /48 scenario, the range from the first /48 to the last one is scanned. For each range, 5000 addresses are picked semi-randomly. This operation is repeated until we get 5000 hits or until 1 million tests have been executed. The relation between the maximum depth and the lookup time is incomplete and I can t explain the difference of performance between the different densities of the /48 scenario. We can extract two important performance points:

• With a full view, the lookup time is 450 ns. This is almost ten times the budget for forwarding at 10 Gbps which is about 50 ns.
• With an almost empty routing table, the lookup time is 150 ns. This is still over the time budget for forwarding at 10 Gbps.

With IPv4, the lookup time for an almost empty table was 20 ns while the lookup time for a full view (500,000 routes) was a bit above 50 ns. How to explain such a difference? First, the maximum depth of the IPv4 LPC-trie with 500,000 routes was 6, while the maximum depth of the IPv6 radix tree for 40,000 routes is 40. Second, while both IPv4 s fib_lookup() and IPv6 s ip6_route_output_flags() functions have a fixed cost implied by the evaluation of routing rules, IPv4 has several optimizations when the rules are left unmodified⁵. Those optimizations are removed on the first modification. If we cancel those optimizations, the lookup time for IPv4 is impacted by about 30 ns. This still leaves a 100 ns difference with IPv6 to be explained. Let s compare how time is spent in each lookup function. Here is a CPU flamegraph for IPv4 s fib_lookup(): Only 50% of the time is spent in the actual route lookup. The remaining time is spent evaluating the routing rules (about 30 ns). This ratio is dependent on the number of routes we inserted (only 1000 in this example). It should be noted the fib_table_lookup() function is executed twice: once with the local routing table and once with the main routing table. The equivalent flamegraph for IPv6 s ip6_route_output_flags() is depicted below: Here is an approximate breakdown on the time spent:

• 50% is spent in the route lookup in the main table,
• 15% is spent in handling locking (IPv4 is using the more efficient RCU mechanism),
• 5% is spent in the route lookup of the local table,
• most of the remaining is spent in routing rule evaluation (about 100 ns)⁶.

Why does the evaluation of routing rules is less efficient with IPv6? Again, I don t have a definitive answer.

History The following graph shows the performance progression of route lookups through Linux history: All kernels are compiled with GCC 4.9 (from Debian Jessie). This version is able to compile older kernels as well as current ones. The kernel configuration is the default one with CONFIG_SMP, CONFIG_IPV6, CONFIG_IPV6_MULTIPLE_TABLES and CONFIG_IPV6_SUBTREES options enabled. Some other unrelated options are enabled to be able to boot them in a virtual machine and run the benchmark. There are three notable performance changes:

• In Linux 3.1, Eric Dumazet delays a bit the copy of route metrics to fix the undesirable sharing of route-specific metrics by all cache entries (commit 21efcfa0ff27). Each cache entry now gets its own metrics, which explains the performance hit for the non-/128 scenarios.
• In Linux 3.9, Yoshifuji Hideaki removes the reference to the neighbor entry in struct rt6_info (commit 887c95cc1da5). This should have lead to a performance increase. The small regression may be due to cache-related issues.
• In Linux 4.2, Martin KaFai Lau prevents the creation of cache entries for most route lookups. The most sensible performance improvement comes with commit 4b32b5ad31a6. The second one is from commit 45e4fd26683c, which effectively removes creation of cache entries, except for PMTU exceptions.

Insertion performance Another interesting performance-related metric is the insertion time. Linux is able to insert a full view in less than two seconds. For some reason, the insertion time is not linear above 50,000 routes and climbs very fast to 60 seconds for 500,000 routes. Despite its more complex insertion logic, the IPv4 subsystem is able to insert 2 million routes in less than 10 seconds.

Memory usage Radix tree nodes (struct fib6_node) and routing information (struct rt6_info) are allocated with the slab allocator⁷. It is therefore possible to extract the information from /proc/slabinfo when the kernel is booted with the slab_nomerge flag:

# sed -ne 2p -e '/^ip6_dst/p' -e '/^fib6_nodes/p' /proc/slabinfo   cut -f1 -d:
   name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>
fib6_nodes         76101  76104     64   63    1
ip6_dst_cache      40090  40090    384   10    1

In the above example, the used memory is 76104 64+40090 384 bytes (about 20 MiB). The number of struct rt6_info matches the number of routes while the number of nodes is roughly twice the number of routes: The memory usage is therefore quite predictable and reasonable, as even a small single-board computer can support several full views (20 MiB for each): The LPC-trie used for IPv4 is more efficient: when 512 MiB of memory is needed for IPv6 to store 1 million routes, only 128 MiB are needed for IPv4. The difference is mainly due to the size of struct rt6_info (336 bytes) compared to the size of IPv4 s struct fib_alias (48 bytes): IPv4 puts most information about next-hops in struct fib_info structures that are shared with many entries.

Conclusion The takeaways from this article are:

• upgrade to Linux 4.2 or more recent to avoid excessive caching,
• route lookups are noticeably slower compared to IPv4 (by an order of magnitude),
• CONFIG_IPV6_MULTIPLE_TABLES option incurs a fixed penalty of 100 ns by lookup,
• memory usage is fair (20 MiB for 40,000 routes).

Compared to IPv4, IPv6 in Linux doesn t foster the same interest, notably in term of optimizations. Hopefully, things are changing as its adoption and use at scale are increasing.

---

1. For a given destination prefix, it s possible to attach source-specific prefixes:

   ip -6 route add 2001:db8:1::/64 \
     from 2001:db8:3::/64 \
     via fe80::1 \
     dev eth0
   

   Lookup is first done on the destination address, then on the source address.
2. This is quite different of the classic scenario where Linux acts as a gateway for a /64 subnet. In this case, the neighbor subsystem stores the reachability information for each host and the routing table only contains a single /64 prefix.
3. The measurements are done in a virtual machine with one vCPU and no neighbors. The host is an Intel Core i5-4670K running at 3.7 GHz during the experiment (CPU governor set to performance). The benchmark is single-threaded. Many lookups are performed and the result reported is the median value. Timings of individual runs are computed from the TSC.
4. Most of the packets in the network are expected to be routed to a destination. However, this also means the backtracking code path is not used in the /128 and /48 scenarios. Having a fallback route gives far different results and make it difficult to ensure we explore the address space correctly.
5. The exact same optimizations could be applied for IPv6. Nobody did it yet.
6. Compiling out table support effectively removes those last 100 ns.
7. There is also per-CPU pointers allocated directly (4 bytes per entry per CPU on a 64-bit architecture). We ignore this detail.