Permalink
Fetching contributors…
Cannot retrieve contributors at this time
325 lines (267 sloc) 14.9 KB
LTO4 and LTO5 drives and other modern tape drives
support hardware encryption.
There are several ways of using encryption with these drives
The following three types of key management are available for
doing encryption. The transmission of the keys to the volumes
is accomplished by:
- A backup application that supports Application Managed Encryption (AME)
- A tape library that supports Library Managed Encryption (LME)
- Using a Key Management Appliance (KMA).
We added support for Application Managed Encryption (AME) scheme where
on labeling a crypto key is generated for a volume and when the volume
is mounted the crypto key is loaded and when unloaded the key is cleared
from the memory of the Tape Drive using the SCSI SPOUT command set.
If you have implemented Library Managed Encryption (LME) or
a Key Management Appliance (KMA) there is no need to have support
from Bareos on loading and clearing the encryption keys as either
the Library knows the per volume encryption keys itself or it
will ask the KMA for the encryption key when it needs it. For
big installations you might consider using a KMA but the Application
Managed Encryption implemented in Bareos should also scale rather
well and has low overhead as the keys are only loaded and cleared
when needed.
How does it all work:
- the libbareos library has some new features:
- crypto_wrap.c - Implements a RFC3394 based wrapping of crypto keys
- crypto_cache.c - Implements a cache of wrapped crypto keys used when
we cannot ask the director for the key e.g. on
startup of the storage daemon.
- passphrase.c - Implements generation of semi-random passphrases
- scsi_lli.c - Implements a lowlevel interface to the tape drive for
several ioctl interfaces available on some modern UNIX
platforms. Current supported platforms are:
- Linux (SG_IO ioctl interface) (tested)
- Solaris (USCSI ioctl interface) (tested)
- FreeBSD (libcam interface)
- NetBSD (SCIOCCOMMAND ioctl interface)
- OpenBSD (SCIOCCOMMAND ioctl interface)
- scsi_crypto.c - Implements sending of SCSI Security Protocol IN (SPIN)
and SCSI Security Protocol OUT (SPOUT) pages using
the scsi_lli interface.
- A new tool named bscrypto allows you to manipulate the tape drive.
It is mostly used for Disaster Recovery (DR) purposes. The storage
daemon and the btools (bls, bextract, bscan, btape, bextract) will
use a so called storage daemon plugin to perform the setting and
clearing of the encryption keys. To bootstrap the encryption support
and for populating things like the crypto cache with encryption keys
of volumes that you want to scan you need to use the bscrypto tool.
The bscrypto tools has the following capabilities:
- Generate a new passphrase
- to be used as a so called Key Encryption Key (KEK) for
wrapping a passphrase using RFC3394 key wrapping with aes-wrap
- or -
- for usage as a clear text encryption key loaded into the tape drive.
- Base64 encode a key if requested
- Generate a wrapped passphrase which performs the following steps:
- generate a semi random clear text passphrase
- wrap the passphrase using the Key Encryption Key using RFC3394
- base64 encode the wrapped key (as the wrapped key is binary, we
always need to base64-encode it in order to be able to pass the
data as part of the director to storage daemon protocol
- show the content of a wrapped or unwrapped keyfile
This can be used to reveal the content of the passphrase when
a passphrase is stored in the database and you have the urge to
change the Key Encryption Key. Normally I would urge people to not
change their Key Encryption Key as this means that you have to redo
all your stored encryption keys as they are stored in the database
wrapped using the Key Encryption Key available in the config during
the label phase of the volume
- Clear the crypto cache on the machine running the bareos-sd which keeps
a cache of used encryption keys which can be used when the bareos-sd is
restarted without the need to connect to the bareos-dir to retrieve the
encryption keys.
- Set the encryption key of the drive
- Clear the encryption key of the drive
- Show the encryption status of the drive
- Show the encryption status of the next block (e.g. volume)
- Populate the crypto cache with data
- A new storage daemon plugin is added named scsicrypto-sd which
hooks into the "unload", "label read", "label write" and "label verified"
events for loading and clearing the key. It checks the drive if it
needs to clear it by either using a internal state if it loaded
a key before or when enabled via a special option which first issues
an encrytion status query. When there is a connection to the director
and the volume information is not available it will ask the director
for the data on the currently loaded volume. When no connection is
available a cache is used which should contain the most recently mounted
volumes. When an encryption key is available it is loaded into the
drives memory.
- The director is extended with additional code for handling
hardware data encryption. On a label of a volume the extra
keyword "encrypt" will force the director to generate a new
semi random passphrase for the volume and this passphrase
is stored in the database as part of the media information.
A passphrase is always stored in the database base64 encoded
and when a so called Key Encryption Key is set in the config
of the director the passphrase is first wrapped using RFC3394
key wrapping and then base64 encoded. By using key wrapping
the keys in the database are save against people sniffing
the info as the data is still encrypted using the Key
Encryption Key (which in essense is just an extra passphrase
of the same length as the volume passphrases used)
When the storage daemon needs to mount the volume it
will ask the director for the volume information and
that protocol is extended with the exchange of the
base64 wrapped encryption key (passphrase). The storage
daemon has an extra config option in which it records
the Key Encryption Key of the particular director and
as such can unwrap the key sended into the original
passphrase.
As can be seen from the above info we don't allow the
user to enter a passphrase but generate a semi random
passphrase using the openssl random functions (if available)
and convert that into a readable ASCII stream of letters,
numbers and most other characters other than the quotes
and space etc. This will give much stronger passphrase than
when requesting the info from a user, as we store things in
the database the user never has to enter these passphrases.
The volume label is written unencrypted to the volume so
we can always recognize a Bareos volume. When the key is
loaded onto the drive we set the decryption mode to mixed
so we can read both unencrypted and encrypted data from the
volume. When there is no key loaded or the wrong key is
loaded the drive will give an IO error when trying to
read the volume.
For disaster recovery you can store the Key Encryption Key
and the content of the wrapped encryption keys somewhere save
and the bscrypto tool together with the scsicrypto-sd plugin
can be used to get access to your volumes when you ever lose
your complete environment.
When you don't want to use the scsicrypto-sd plugin when
doing DR and you are only reading one volume you can also
set the crypto key using the bscrypto tool because we use
the mixed decryption mode you can set the right encryption
key before reading the volume label as in mixed mode you can
read both encrypted and unencrypted data from a volume.
When you need to read more then one volume you better use
the scsicrypto-sd plugin with things like bscan/bextract
as the plugin will then auto load the correct encryption
key when it loads the volume just as what the storage
daemon does when performing backups and restores.
The volume label is unencrypted so a volume can also be
recognized by a non encrypted installation but it won't be
able to read the actual data from it. Using an encrypted
volume label doesn't add much security (there is no security
related info in the volume label anyhow) and it makes it
harder to recognize either a labeled volume with encrypted
data v.s. a unlabeled new volume (both would return an
IO-error on read of the label.)
The initial setup of SCSI crypto looks something like this:
- Run configure with your normal configure options and
add the --enable-scsi-crypto option.
- Build the Bareos programs and package/install them
in the normal way.
- Generate a Key Encryption Key e.g.
bscrypto -g -
=== Security Setup ===
Some security levels need to be increased for the storage
daemon to be able to use the low level SCSI interface for
setting and getting encryption status on a tape device.
The following additional security is needed for the following
operating systems:
- Linux (SG_IO ioctl interface)
The user running the storage daemon needs the following
additional capabilities:
CAP_SYS_RAWIO (See capabilities(7))
If bareos-sd does not have the appropriate capabilities, all
other tape operations may still work correctly, but you will
get "Unable to perform SG_IO ioctl" errors.
On older kernels it can be you need CAP_SYS_ADMIN try
CAP_SYS_RAWIO first and if that doesn't work try CAP_SYS_ADMIN
When you are running the storage daemon as another user
then root (which has the CAP_SYS_RAWIO capability) you
need to add it to the current set of capabilities.
When you are using systemd you could add this additional
capability to the CapabilityBoundingSet parameter.
For systemd add to the bareos-sd.service the following:
Capabilities=cap_sys_rawio+ep
You can also setup the extra capability on bscrypto and
bareos-sd by running the following cmd:
# setcap cap_sys_rawio=ep bscrypto
# setcap cap_sys_rawio=ep bareos-sd
Check the setting with
# getcap -v bscrypto
# getcap -v bareos-sd
getcap and setcap are part of libcap-progs which may not be
installed on your system.
- Solaris (USCSI ioctl interface)
The user running the storage daemon needs the following
additional privileges:
PRIV_SYS_DEVICES (See privileges(5))
When you are running the storage daemon as another user
then root (which has the PRIV_SYS_DEVICES privilege) you
need to add it to the current set of privileges.
This can be setup by setting this either as a project for
the user or as a extra privileges in the SMF definition
starting the storage daemon. The SMF setup is the cleanest.
For SMF make sure you have something like this in the instance
block.
<method_context working_directory=":default">
<method_credential user="bareos" group="bareos" privileges="basic,sys_devices"/>
</method_context>
=== Changes in bareos-sd.conf ===
- Put the Key Encryption Key into the bareos-sd.conf
under the director entry in that config for the specific
director you are creating the config for. e.g.
Key Encryption Key = "<passphrase>"
- Enable the loading of storage daemon plugins by
setting the plugin dir in the bareos-sd.conf e.g.
Plugin Directory = <path_to_sd_plugins>
- Enable the SCSI encryption option in the device configuration
section of the drive in the bareos-sd.conf. e.g.
Drive Crypto Enabled = Yes
- When you want the plugin to probe the drive for its encryption
status if it needs to clear a pending key enable the Query
Crypto Status option in the device configuration section of the
drive in the bareos-sd.conf e.g.
Query Crypto Status = Yes
=== Changes in bareos-dir.conf ===
- Put the Key Encryption Key into the bareos-dir.conf under
the director config item named Key Encryption Key e.g.
Key Encryption Key = "<passphrase>"
- restart sd and dir
- Label a volume with the encrypt option e.g.
label slots=1-5 barcodes encrypt
For Disaster Recovery (DR) you need the following information:
- Actual bareos-sd.conf with config options enabled as described
above, including things like a definition of a director with
the Key Encryption Key used for creating the encryption keys
of the volumes.
- The actual keys used for encryption of the volumes.
This data needs to be availabe as a so called crypto cache
file which is used by the plugin when no connection to the
director can be made to do a lookup (most likely on DR).
Most of the times the needed information e.g. bootstrap info
is available on recently written volumes and most of the
time the encryption cache will contain the most recent data
so a recent copy of the bareos-sd.<portnr>.cryptoc file in
the workingdir is most of the time enough. You can also save
the info from database in a save place and use bscrypto to
populate this info (VolumeName<tab>EncryptKey) into the crypto
cache file used by bextract and bscan. You can use bscrypto
with the following flags to create a new or update an existing
crypto cache file e.g. :
# bscrypto -p /var/lib/bareos/bareos-sd.<portnr>.cryptoc
- A valid BSR file with the location of the last save of the
database makes recovery much easier. Adding a post script
to the database save job could collect the needed info and
make sure its stored somewhere safe.
- Recover the database in the normal way e.g. for postgresql:
# bextract -D <director_name> -c bareos-sd.conf -V <volname> \
/dev/nst0 /tmp -b bootstrap.bsr
# /usr/lib64/bareos/create_bareos_database
# /usr/lib64/bareos/grant_bareos_privileges
# psql bareos < /tmp/var/lib/bareos/bareos.sql
Or something similar (change paths to follow where you
installed the software or where the package put it.)
NOTE: As described at the beginning of this README there are different
types of key management:
- A backup application that supports Application Managed Encryption (AME)
- A tape library that supports Library Managed Encryption (LME)
- Using a Key Management Appliance (KMA).
If the Library is setup for LME or KMA it probably won't allow our AME setup
and the scsi-crypto plugin will fail to set/clear the encryption key. To be
able to use AME you need to "Modify Encryption Method" and set it to something
like "Application Managed". If you decide to use LME or KMA you don't have to
bother with the whole setup of AME which may for big libraries be easier, although
the overhead of using AME even for very big libraries should be minimal.