SZTPD is an implementation of a “bootstrap server”, as defined in the Terminology section of RFC 8572, also known as an “SZTP server”, as defined in the Terminology section of draft-kwatsen-netconf-sztp- csr".
SZTPD is provided as software, an asynchronous event-driven Python-based executable. SZTPD is an application (not a library) that includes a northbound API for configuration, a southbound API for device bootstrapping requests, and eastbound hooks for integration with other business systems.
This documentation is in preparation for a “1.0.0” release, using the common “major.minor.patch” semantic versioning convention.
SZTPD is installed using the command:
$ pip install sztpd
or, if Python 3 is installed separately:
$ pip3 install sztpd
On some systems it may be necessary to install the sqlite package first (e.g., ‘yum install sqlite’).
SZTPD has resource requirements as described in the following sections.
SZTPD runs as a single asynchronous I/O event driven process. It is ideal to dedicate a whole core to the ‘sztpd’ process, thus, in addition to the operating system, the machine should have at least two processors.
SZTPD uses significant memory resources1. The following sections consider SZTPD’s constant and transitive memory demands.
SZTPD has a constant memory demand that grows in proportion to its configuration (minus all large base64 blobs stored in the configuration). Sizing numbers for this haven’t been measured yet.
SZTPD has transitive memory demands when processing requests from clients. The highest memory-impacting request is a “PUT” request used to replace the entire SZTPD configuration. In this case, the transitive memory demand is proportional to the sum of the old and new configuration sizes. Sizing numbers for this haven’t been measured yet.
Unless SZTPD is configured to use a file-based database, SZTPD has effectively no interaction with the filesystem, all configuration and logs are stored in the database.
That said, a filesystem must exist if the connection to the database is to be encrypted, as the requisit certificates are passed into the SZTPD executable as command line parameters as file paths. Additionally, it is expected that SZTPD’s stdout/stderr will be piped to a file (e.g., /var/log/sztpd.log).
When using a file-based database, the database file grows proportionally to the size of the configuration plus operational state (e.g., audit logs, bootstrapping logs, etc.). Sizing numbers for this haven’t been measured yet.
SZTPD’s networking demands vary by interface, as different APIs are presented for northbound administrators and southbound devices.
While the Northbound API is exclusively a programmatic interface, it is expected to be used to drive human-facing interfaces and thus the speed of the network interface may matter.
The Southbound API is used exclusively for machine-to-machine interactions (i.e., SZTPD and the bootstrapping devices), and thus the speed of the network interface need only be fast enough to not cause timeouts. That said, the devices access the southbound interface when they are bootstrapping, which typically occurs immediately when being powered-up for the first time, which typically entails a human (the installer) manually plugging in a power cable. In some use cases, the installer will be instructed to wait for the device to provide some indication that the bootstrapping process has completed (e.g., an LED pattern, audible sound, etc.). From this perspective only, the speed of the network interface may matter2.
SZTPD is offered as software that depends on Python 3.7 or above. Any operating system that supports Python 3.7 or above should be able to run SZTPD.
Regardless the operating system selected3, standard hardening guidelines should be observed (e.g., disable unnecessary processes, especially those that open listening ports of their own.
SZTPD’s networking demands are described in the following two sections. The first section regards the networking needs for inbound connections. The second section regards the networking needs for outbound connections.
SZTPD can listen to a multiplicity of ports and, for each such interface, can present a multiplicity of its APIs. Each port may be bound to a source address and/or source port, enabling integration into virtual network interfaces presented by the host operating system.
The protocol running on each port can be configured to be “HTTP” or “HTTPS”.
If a port is configured to use “HTTP”, it is only intended so that the TLS termination can be offloaded to a device fronting SZTPD (e.g., a firewall). That said, SZTPD cannot know or enforce that TLS was performed, so care must be taken to not accidentally expose an unprotected port to a untrusted network.
When “HTTP” is used, SZTPD may be configured with information about the external endpoint so that, e.g., when SZTPD sends callback information to remote peers (e.g., links in account activation emails), it can address the callback to the external endpoint instead of its local endpoint.
The APIs that may be selected for each listening port are:
The native interface presents an [Northbound Interface] API that can configure everything and anything that can be configured.
Based on the product mode, the API this interface presents varies slightly, as described in Prompting for Desired Mode Selection.
There must always be at least one “native-interface” configured.
For deployments using the mode ‘x’ product mode, the tenant interface presents a isolated view of each tenant’s data. No tenant is aware of the existance of any other tenant, nor has access to any other tenant’s data.
For deployments using the mode ‘x’ product mode, SZTPD offers an additional API for subtenant access.
The tenant interface presents an API that is almost exactly like mode-1’s “native” view.
The RFC 8752 selection presents the bootstrap server API defined in Section 7 of RFC 8572.
This interface does not provide any access other than as described in RFC 8572.
While the above suggests numerous possible combinations, some common options include:
While SZTPD having a single listening port is theoretically useful in production environments, it is unlikely for the reasons mentioned in the next section4.
That said, SZTPD’s default configuration opens a single listening port through which all configuration begins.
SZTPD having two listening ports is suitable when it is desired to isolate the network traffic for administrators and bootstrapping devices. Each port may5 have a unique TLS-level end-entity certificate. Notably, when the SZTPD instance is intended to be a “trusted” (as defined in RFC 8572) bootstrap server, this enables the southbound interface to present a device-friendly6 certificate while the northbound interface presents an administrator-friendly7 certificate.
SZTPD having a third listening port may also be useful, for deployments using the mode ‘x’ product mode, to present a distinct “tenant-interface” API, thus isolating networking traffic for the host-level and tenant-level administrators.
Administrators may configure any number of any of these interfaces. That said, the only meaningful reason to do so is in order to present distinct TLS end-entity certificates. This point has no bearing when external TLS termination is deployed.
Two cases for wanting to do this:
Additional listening ports may used to present a distinct TLS end-entity certificate to different groups of devices, such as may be necessary in order to accommodate devices manufactured using different trust anchors.
Additional listening ports may used to present a distinct TLS end-entity certificate to different groups of tenants. In particular, it is possible to use a certificate signed by a tenant-specific issuer.
SZTPD initiates outbound connections for the reasons discussed in this section.
SZTPD sends email for the following purposes:
In order to deliver the email, SZTPD, or the host operating system on its behalf, will initiate an outbound SMTP/S connection to a configured SMTP server. Traditionally this is a connection to port 25, 587, or 2525.
SZTPD sends notifications for the following reasons:
In order to deliver notification, SZTPD uses mutually authenticated HTTPS connection to configured URIs. The default destination port for HTTPS is port 443, but the URIs may encode alternate port number values as needed.
For deployments using the mode ‘x’ product mode, SZTPD enables the host system to ensure that the serial-numbers configured by tenants are rightfully owned by the tenant.
In order to verify device ownership, SZTPD initiates a mutually authenticated HTTPS connection to the URI configured when enabling this feature.
The default destination port for HTTPS is port 443, but the URIs may encode alternate port number values as needed.
SZTPD may need to verify certificate paths for the following reasons:
In order to properly verify certificate paths, it is sometimes necessary for SZTPD to check the current revocation status of certificate paths. Each certificate may specify URIs for where the CRL or OCSP can be obtained and, when needed, SZTPD will initiate a connection to the URI specified in the certificate.
When configured to use a remote database (see Persistence Selection), SZTPD will initiate connections to the remote database, specified by the URI given on the command line.
While the database is “remote” to the ‘sztpd’ process it could be running on the same machine.
The remote port accessed varies by database and TLS configuration,
When initiating any of the previously discussed outbound connections, the configurations for the remote host may be given as either an IP address or a domain name. When a domain name is provided, SZTPD resolves the domain name to IP addresses using the system configured DNS resolver. Depending on how the DNS resolver is configured, the host system may initiate a connection to a network-based resolver. Traditionally this is a TCP-based connection destined to port 53.
SZTPD includes a host of security features for both data in motion as well as data at rest.
All of SZTPD’s APIs are presented as mutually-authenticated HTTPS connections 8.
Two factor authentication (client certificate + password) may be configured.
Passwords, when used, may be required to have a minimal length9.
Each request sent from clients is tested against access control and a corresponding entry is recorded in the audit log.
SZTPD hashes passwords used to authenticate clients, but otherwise leaves it to the administrator to implement database-level encryption.
Passwords are hashed when configured for both administrators and devices. For instance, a clear-text input password value, such as ‘0
However, passwords used to authenticate to a remote system (e.g., an SMTP server) are not hashed when, as the cleartext password is needed in such cases. These password may be encrypted in a future release (e.g., as a symmetric-key in the [Keystore]).
For these reasons, database-level encryption is recommended.
When the installation completes, the executable “sztpd” is installed in your shell’s path.
To test running SZTPD and see its “help” page:
$ sztpd --help
Which produces:
usage: sztpd [-h] [-v] [-d] [-c CACERT] [-k KEY] database-url
The SZTP Server (sztpd) implements the "bootstrap server" defined in RFC 8572.
positional arguments:
database-url see below for details.
optional arguments:
-h, --help show this help message and exit
-v, --version show version number and exit.
-d, --debug increase debug output level up to 3x (e.g., -ddd)
-c CACERT, --cacert CACERT
path to trust anchor certificates used to authenticate
the database (see below for details).
-k KEY, --key KEY path to pkcs#12 used to authenticate into the database
(see below for details).
Exit status code: 0 on success, non-0 on error. Error output goes to stderr.
The "cacert" argument is a filepath to a PEM file that contains one or more X.509
certificates. The list of certificates must be ordered from the certificate of
the issuer to the database's certificate to the self-signed root certificate.
The "key" argument is a filepath to a PEM file that contains a PKCS#12, encoding
both its private key and end-entity certificate.
The "database-url" argument has the form "<dialect>:<dialect-specific-path>".
Three dialects are supported: "sqlite", "postgresql", and "mysql+pymysql".
The <dialect-specific-path> for each of these is described below.
For the "sqlite" dialect, <dialect-specific-path> follows the format
"///<sqlite-path>", where <sqlite-path> can be one of:
:memory: - an in-memory database (only useful for testing)
<filepath> - an OS-specific filepath to a persisted database file
Examples:
$ sztpd sqlite:///:memory: (memory)
$ sztpd sqlite:///relative/path/to/sztpd.db (unix)
$ sztpd sqlite:////absolute/path/to/sztpd.db (unix)
$ sztpd sqlite:///C:\path\to\sztpd.db (windows)
For both the "postgresql" and "mysql+pymysql" dialects, the dialect-specific-path
follows the format "//<user>:<passwd>@<host>:<port>/<database-name>".
Examples:
The following two examples assume the database is called "sztpd" and
that the database server listens on the loopback address with no TLS.
$ sztpd postgresql://user:pass@localhost:5432/sztpd
$ sztpd mysql+pymysql://user:pass@localhost:3306/sztpd
Please see the documentation for more information.
SZTPD persists all data10 into the database provided on the command line (i.e., the database URL passed into the ‘sztpd’ command). Varying the database URL enables use of different databases. Currently supported databases include Postgres, MySQL, MariaDB, and SQLite.
The in-memory database type “persists” all data in memory, which is automatically lost when the ‘sztpd’ process ends. In-memory databases are fast, but demand additional process memory usage proportional to the size of the data stored.
An in-memory database is an excellent choice for development and test efforts, as it automatically disappears when the ‘sztpd’ process ends. The simulator (see the Administrator’s Guide) uses the in-memory database for this reason.
An in-memory database may also have use in a production environment (e.g., in an SDN context) whereby the intent is for the SZTPD instance to be ephemeral, lasting only as long as needed to bootstrap a specific set of devices.
Use of the in-memory database type is specified by the database URL provided on the command line having the form:
sqlite:///:memory:
The file-based database storage persists all the data into a single file located by the path-component in the database URL specified on the command line. File-based databases may be desirable in cases where persistence across power cycles is important and yet dependency on a remote RDBMS is not desirable.
The specified file may reside on a filesystem mounted from a RAID system providing resiliency against disk failures. Care should be to ensure fast and secure access to a remote system (e.g., a NAS or SAN).
The file itself may be backed-up and restored as needed to ensure disaster recovery. If the filesystem does not support taking snapshots while a system is running, it is recommended that the ‘sztpd’ process is either stopped or suspended during the backup operation.
Use of the file-based database type is specified by the database URL provided on the command line having one of the following form:
sqlite:///relative/path/to/sztpd.db (unix)
sqlite:////absolute/path/to/sztpd.db (unix)
sqlite:///C:\path\to\sztpd.db (windows)
The RDBMS database engine persists all the data into one of several well-known database systems (e.g., MariaDB, Postgres, Oracle, etc.). In theory, SZTPD can use any RDBMS supported by SQLAlchemy, though it is recommended to only use those that SZTPD has been tested against.
Use of an RDBMS database enables large data sets with fast-access to keyed data. Scaling of the database tier supports a variety of performance and availability targets. It is recommended to use database-level encryption to ensure to protection of the SZTPD data at rest.
Whilst the database may run as another process on the same machine, and hence the two processes could communicated with each other in a protected manner over the ‘loopback’ address, it is more likely that the RDBMS is running remotely. It is recommended that the connection to a remote RDBMS is protected by transport level security (TLS), which entails both configuring the server to listen for TLS connections and configuring SZTPD to connect to the the RDBMS using TLS. Details for how to do this are provided in the RDBMS Configuration section.
Use of the RDBMS database type is specified by the database URL provided on the command line having the form:
<engine>://<user>:<passwd>@<host>:<port>/<database-name>
For instance:
postgresql://sztpd-admin:secret@db.example.com:5432/sztpd-db
mysql+pymysql://sztpd-admin:secret@db.example.com:3306/sztpd-db
The first time SZTPD runs11 the server will detect that the database is uninitialized and hence prompt for:
Example output:
<Non-Production Use Contract>
First time initialization. Please accept the license terms.
By entering "Yes" below, you agree to be bound to the terms and conditions contained on this screen with Watsen Networks.
Please enter "Yes" or "No": <SZTPD waits for input here>
Entering “Yes” allows SZTPD to proceed to prompting for the mode to run. SZTPD will exit for any other entered response, not just “No”.
SZTPD can run in one of two modes12:
Mode | Use Level | Purpose |
---|---|---|
1 | Single Tenant | Ideal for single-domain deployments (e.g., enterprises). |
x | Multi Tenant | Ideal to provide a self-service API for downstream customers. |
The mode character represents, roughly, the number or tenants. For instance, ‘1’ is for “one tenant” (i.e., you), while ‘x’ is for “many tenants” (i.e., you plus your customers).
Mode:
1 - single-tenant
x - multi-tenant
Mode: <SZTPD waits for input here>
Each of these modes are described in the following sections.
Mode ‘1’ enables a multiplicity of devices to be configured, but without an additional “tenants” layer. From an API perspective, the difference, in comparison to the mode ‘x’ tree diagram, is illustrated below.
Mode ‘1’
+--rw devices
+--rw device* [serial-number]
+--rw serial-number string
+-- ... // additional device parameters here.
Mode ‘x’ enables multi-tenancy, whereby each tenant effectively has their own Mode ‘1’ instance, with the only aspect missing being the ability to configure system-level information (e.g., networking parameters).
The host view is not only a superset of all tenant views, but it also comes with the unique ability to configure plugins, define device-types, and define an “device ownership verification” callback function13.
The difference in the APIs is illustrated below using tree diagrams (RFC 8340).
Mode ‘x’:
+--rw preferences
+--rw device-ownership-verification!
| +-- ... // ownership verification parameters here
+--rw plugins
+--rw plugin* [name]
+-- // plugin parameters here
+--rw device-types
+--rw device-type* [name]
+-- // device-type parameters here
+--rw tenants
+--rw tenant* [name]
+--rw name string
+--rw device
+--rw device* [serial-number]
+--rw serial-number string
+-- ... // additional device parameters here.
The follow sections describe defaults the freshly installed system uses.
A freshly installed version will open a single port by default.
This port uses the following parameters:
These values may be changed using environment variables:
For instance:
[Note: '\' line endings per RFC 8792]
$ export SZTPD_DEFAULT_ADDR="A.B.C.D"; \
export SZTPD_DEFAULT_PORT="9090"; \
sztpd --key my-pkcs12.pem --cacert rdbms-cacert.pem postgresql://user:pass@localhost:5432/sztpd
This default port presents an HTTP (without TLS) protocol over which the RESTCONF API is presented. The SZTPD server must use this port to configure the system to use TLS.
The default server (i.e., one that has never been configured) does not have any administrator accounts defined.
Prior to configuring an admin, no client authentication needs to be supplied.
The very first configuration write request must, at least or in addition to other changes, configure at least one administrator account and, further, this account must have the “unrestricted” access level.
The ‘sztpd’ executable takes an optional parameter ‘–cacert’ to specify trust anchor certificates that SZTPD must use to authenticate the RDBMS with using TLS. SZTPD will not fallback to an unencrypted TCP connection to the RDBMS when the ‘–cacert’ command line parameter is set.
$ sztpd --cacert db-cacert.pem postgresql://user:pass@localhost:5432/sztpd
The ‘sztpd’ executable takes an optional parameter ‘–key’ that encodes a PKCS#12, containing both a private key as well as accompanying public key certificates. The ‘–key’ parameter is dependent on presence of the ‘–cacert’ parameter.
$ sztpd --key my-pkcs12.pem --cacert db-cacert.pem postgresql://user:pass@localhost:5432/sztpd
Passing the key as a parameter releases SZTPD from having to determine where to store it15, while providing the installer the ability to extract the key from a secure location (e.g., TPM-protected storage).
The ‘sztpd’ executable should be executed as a daemon, gracefully starting and stopping with the host system. Many systems use “rc” scripts located in ‘/etc/rc.d’ or the like for this purpose.
The ‘sztpd’ executable responds the SIGNALS as follows:
Notably, SZTPD automatically sends itself a ‘SIGHUP’ signal whenever the system-level “transport” configuration is updated (see the Administrator’s Guide for information about “transport” configuration).
Unless SZTPD is asked to open a listening port below port number 1024, the ‘sztpd’ does not require any special user privileges (other than be able to read any input file and write any output files), and hence it is recommended to run the ‘sztpd’ executable using an unprivileged user account. Note that the ‘sztpd’ executable does not itself drop user privileges.
SZTPD is a single-threaded asynchronous event-driven executable. While the CPU demand has yet to be measured, the demand is not expected to be excessive.
It is ideal if the ‘sztpd’ process is locked down to a core, thus avoiding delays introduced from process swapping. This means that the machine should have at least two core, at least on other being for the underlying operating system.
The ‘sztpd’ process does not produce any log files, all audit logs and bootstrapping logs are held within the database layer (see Persistence Selection).
The ‘sztpd’ process itself may emit output to STDOUT and or STDERR. For instance, if an unexpected condition is encountered, a Python stack trace is generated.
It is recommended to pipe the ‘sztpd’ process’s output to a file. For instance:
sztpd sqlite:///:memory: >> /var/logs/sztpd.log 2>&1
It is recommended to have a cold standby instance on the ready in another geographic location, with active database replication configured between the two sites.
By “cold” standby, what is meant is that the remote system can be booted (the operating system is running), but the ‘sztpd’ process is not running. The reason why this form of “cold” standby is recommended is primarily due to the software not yet having the notion of being in a “standby” mode, but also because the ‘sztpd’ process takes very little time to startup. The SZTPD process takes a couple seconds to startup on a fast CPU with a small database.].
Database-level replication varies by database. The File-based Database type entails copying a file to the remote system. Replication using an RDBMS entails using the replication mechanism provided by that RDBMS.
Another form of high-availability can be had through RDBMS-based database clustering, which must be configured using the clustering mechanism provided by the RDBMS.
Scale testing numbers have not been measured yet.
Prior deployment experience suggests low interaction rates on both the southbound and northbound interfaces, scaling proportionally to the number of bootstrapping devices. Prior deployment experience suggests that a high number of sustained interactions can be supported with SZTPD’s current implementation strategy. Adjustments to the implementation will be made as needed to meet operator requirements.
This section will be updated when scale testing numbers become available.
Upgrades will be distributed using ‘pip’ and hence can be applied using the command:
pip install --upgrade sztpd
Before running this command, the ‘sztpd’ process should be shutdown16 and a database backed up is taken.
When an upgrade is installed, there may be a need to migrate the database17, which may take some time depending on the nature of the upgrade and how much data there is in the database. If not database migration is required (typical case) then the upgrade will take only seconds to complete.
This section provides details to set up a persistent relational data store for the SZTPD server. This document does not cover database configuration options in full detail and is not meant to replace a database server guide. This section uses the term “schema” to denote a set of database objects owned by a user account. This means a database may have more than one schemas similarly to Oracle RDBMS, where as in MySQL, a schema is referenced synonymously as a database.
Generally you will want to configure the database server to have as much RAM as possible and to fit as much data and indexes as possible in memory. We give some considerations towards an initial database set up but afterwards some database tuning, optimization, and parameter rezing would be necessary as a database grows.
SZTPD has been tested against MySQL 8.0.19 .
This document section uses “MySQL” interchangeably for MySQL on AWS EC2, AWS Aurora, AWS RDS MySQL, RDS MariaDB, MariaDB, Percona XtraDB and various MySQL forks and flavors.
Assuming a freshly installed MySQL instance running on the same machine as SZTPD, with an admin account called “root” with no password set, the following commands might be used to initialize a user account for SZTPD:
$ mysql -u root -e "CREATE USER 'testuser'@'localhost' IDENTIFIED BY 'secret';"
$ mysql -u root -e "GRANT ALL ON sztpd.* TO 'testuser'@'localhost';"
$ mysql -u root -e "FLUSH PRIVILEGES;"
Note: the grant above is to the yet-to-be-created “sztpd” database.
At which point SZTPD can be run as follows:
$ sztpd mysql+pymysql://testuser:secret@localhost:3306/sztpd"
Note that SZTPD will itself create a database called “sztpd”18 and initialize all the tables in it19.
The following command returns the database to the state before SZTPD ran:
$ mysql -u root -e "DROP DATABASE sztpd;"
And this command returns the database to the state before creating the user:
mysql -u root -e "DROP USER 'testuser'@'localhost';"
Note that, in all of the commands above, the string “localhost” could have been replaced with the string “127.0.0.1”. However, doing so switches from using a UNIX socket file to the TCP/IP stack. This may require setting the “bind-address” variable (e.g., in the “my.cnf” file) to “127.0.0.1” and restarting the MySQL server (i.e., ‘mysql.server restart’).
Assuming the MySQL instance is installed on another machine, the commands replace “localhost” with the appropriate IP addresses for the two machines.
For instance, assuming SZTPD is running on “11.11.11.11” and the MySQL instance is running on “22.22.22.22”, the MySQL user-creation commands would be:
$ mysql -u root -e "CREATE USER 'testuser'@'11.11.11.11' IDENTIFIED BY 'secret';"
$ mysql -u root -e "GRANT ALL ON sztpd.* TO 'testuser'@'11.11.11.11';"
$ mysql -u root -e "FLUSH PRIVILEGES;"
And for SZTPD to connect to it, the command would be:
$ sztpd mysql+pymysql://testuser:secret@22.22.22.22:3306/sztpd
Note that, in order for the MySQL server to support accepting connections from remote systems, its “bind-address” variable (e.g., in the “my.cnf” file) must either be unset, or set to either a specific IP address (i.e., “22.22.22.22”) or the wildcard address appropriate for the IP stack (e.g., “0.0.0.0” for IPv4). Be sure to restart the MySQL server afterwards (i.e., ‘mysql.server restart’).
Assume the following certificate chain with associated keys and certificates for the MySQL server’s certificate:
Root CA -> Intermediate 1 -> Intermediate 2 -> Intermediate 3 -> End Entity
(mysql-root.pem) (mysql-int1.pem) (mysql-int2.pem) (mysql-int3.pem) (mysql-ee.pem)
(mysql-root.key) (mysql-int1.key) (mysql-int2.key) (mysql-int3.key) (mysql-ee.key)
The chain is split as follows:
The following commands setup some additionally needed files:
# cat mysql-root.pem mysql-int1.pem > mysqld-trust-anchor-certs.pem
# cat mysql-int2.pem mysql-int3.pem > mysqld-intermediate-cacerts.pem
While it is possible to configure MySQL dyanmically, editing its configuration file (e.g., in the “my.cnf” file) is more understandable. Building on top of the previous example to bind the local address as well, the following is set:
[mysqld]
bind-address = 22.22.22.22
ssl_ca=mysqld-intermediate-cacerts.pem
ssl_cert=mysql-ee.pem
ssl_key=mysql-ee.key
require_secure_transport=ON
Be sure to restart the MySQL server afterwards (i.e., ‘mysql.server restart’).
AWS Aurora configures TLS by default, but it does NOT set the “require_secure_transport” variable. An equivalent setting is to use the “REQUIRE SSL” parameter on a per-user basis via in the “CREATE USER” clause 5.7 or “GRANT USAGE” clause in 5.6.
SZTPD is “configured” to authenticate the server’s TLS certificate solely on the command line20. Specifically:
$ sztpd --cacert mysqld-trust-anchor-certs.pem mysql+pymysql://testuser:secret@22.22.22.22:3306/sztpd
AWS Aurora instances use AWS-created certificates. It is thus necessary to pass the AWS-specific CA certificates file (e.g., rds-ca-2019-root.pem) via the “cacert” parameters. Please refer to this document to obtain the PEM file appropriate for your region.
Assume the following certificate chain with associated keys and certificates for the SZTPD client’s identity certificate:
Root CA -> Intermediate 1 -> End Entity
(sztpd-root.pem) (sztpd-int1.pem) (sztpd-ee.pem)
(sztpd-root.key) (sztpd-int1.key) (sztpd-ee.key)
The chain is split as follows:
Note that MySQL client does not support sending a partial chain for the client certificate.
The following commands setup some additionally needed files:
# cat sztpd-root.pem sztpd-int1.pem > sztpd-trust-anchor-cacerts.pem
# cat sztpd-trust-anchor-cacerts.pem mysqld-intermediate-cacerts.pem > mysqld-all-cacerts.pem
Note that MySQL does not maintain separate files for its intermediate CA certificates and the CA certificates used to authenticate clients. It is for this reason that the two sets of certificates are merged above into the one file “mysqld-all-cacerts.pem”.
Building on top of the previous example, the “my.cnf” file is updated to:
[mysqld]
bind-address = 22.22.22.22
ssl_ca=mysqld-all-cacerts.pem
ssl_cert=mysql-ee.pem
ssl_key=mysql-ee.key
require_secure_transport=ON
Be sure to restart the MySQL server afterwards (i.e., ‘mysql.server restart’).
It is additionally necessary to change the “user” definition from before.
$ mysql -u root -e "CREATE USER 'testuser'@'11.11.11.11' REQUIRE X509;"
$ mysql -u root -e "GRANT ALL ON sztpd.* TO 'testuser'@'11.11.11.11';"
$ mysql -u root -e "FLUSH PRIVILEGES;"
Note that additional “REQUIRE” options exist to ensure that the client certificate contains a specific ‘SUBJECT’ and/or is signed by a specific ‘ISSUER’. These tests are in addition to the MySQL server testing that the client certificate has a chain of trust to any of the CA certificates contained in the file specified by the “ssl_ca” variable.
It is also possible to create a user that requires both a client certificate and a password:
$ mysql -u root -e "CREATE USER 'testuser'@'11.11.11.11' IDENTIFIED BY 'secret' REQUIRE X509;"
$ mysql -u root -e "GRANT ALL ON sztpd.* TO 'testuser'@'11.11.11.11';"
$ mysql -u root -e "FLUSH PRIVILEGES;"
Client-certificate based authentication to an Aurora MySQL instance is not well supported.
AWS Aurora MySQL databases do not enable filesystem access to update the CA file pointed at by the “ssl_ca” variable set in the “my.cnf” file.
The command “SHOW VARIABLES LIKE ‘%SSL%’;” suggests the that files might be found in “/rdsdbdata/rds-metadata/ca-cert.pem”, but this folder doesn’t appear to be accessble.
This AWS page states that the CA file may be updatable via an API call, but it is only available for MySQL 5.6.
This MySQL page states that may be dynamically (i.e., via API) updated stating in release 8.0.16 that, at this time, is not yet supported by AWS Aurora.
SZTPD is “configured” to send a client certificate solely on the command line21. Specifically:
# NOTE: '\' line wrapping per RFC 8792
$ sztpd --cacert mysqld-trust-anchor-certs.pem --cert=sztpd-ee.pem \
--key=sztpd-ee.key mysql+pymysql://testuser@22.22.22.22:3306/sztpd
Or, if also needing to pass a password:
# NOTE: '\' line wrapping per RFC 8792
$ sztpd --cacert mysqld-trust-anchor-certs.pem --cert=sztpd-ee.pem \
--key=sztpd-ee.key mysql+pymysql://testuser:secret@22.22.22.22:3306/sztpd
[Disclaimer: the section needs to be reviewed]
We installed and tested PostgresSQL release 11.4 or later. Once the installation is completed, start up a PostgresSQL instance, hosted locally on port 5432. The server will be run out of the directory /usr/local/var/postgres.
By default, a Postgres installation has three databases defined template0, template1 and postgres. template0 and template1 are skeleton databases that are or can be used by the CREATE DATABASE command. postgres is the default database you will connect to before you have created any other databases. Once you have created another database you will want to switch to it in order to create tables and insert data. Often, when working with servers that manage multiple databases, you’ll find the need to jump between databases frequently. This can be done with the connect meta-command or its shortcut c.
In the event that the postgres command is not found, you can locate it by issuing the ‘locate bin/postgres’ command:
$ locate bin/postgres
/usr/lib/postgresql/11.4/bin/postgres
Now with the direct path to the postgres utility, you can call it with the -V flag:
$ /usr/lib/postgresql/11.4/bin/postgres -V
To view the client version, again simply pass the -V flag to the psql client utility command:
psql -V
After verifying the PostgreSQL software, you can configure a different database server/data directory with a name of your choice - this example uses [Data Dirctory] as follow:
$ initdb [Data Directory] -E utf8
Start manually with the database directory in /usr/local/var/postgres:
$ pg_ctl -D /usr/local/var/postgres start
Stop manually:
$ pg_ctl -D /usr/local/var/postgres stop
Stop manually with launchd, which starts postgresql now and restarts at login:
$ brew services restart postgresql
Check for error messages in the server.log and verify that postgres is in /var/lib/pgsql and a running process with “ps -ef | grep postgres” (its on port 5432).
To enable TLS 1.2 with server certificate validation, edit Postgres server configuration parameters in the postgresql.conf file, which specifies server behavior with regards to auditing, authentication, encryption, and other behaviors. The postgresql.conf file usually resides in the data directory under your installation:
password_encryption = scram-sha-256 # md5 prior to 10 or scram-sha-256 post version 10
# - SSL -
ssl = on
ssl_ca_file = ''
ssl_cert_file = 'server.crt'
ssl_crl_file = ''
ssl_key_file = 'server.key'
ssl_ciphers = 'TLSv1.2:!aNULL' #'HIGH:MEDIUM:+3DES' or TLSv1.3
or a list of ciphers but TLSv1.2 is a safe bet
ssl_prefer_server_ciphers = on
pg_ctl reload
If you have a Postgres release 11.4, set ssl_ciphers to TLSv1.2.
PostgreSQL release 12 contains two new server settings (ssl_min_protocol_version and ssl_max_protocol_version) that are are used to control the oldest (minimum) and newest (maximum) version of the SSL and TLS protocol family that the server will accept. Set these to TLSv1.2 and TLSv1.3 respectively.
Client authentication is controlled by a configuration file, which is named pg_hba.conf and is stored in the database data directory. Make sure tcp localhost connections are enabled in pg_hba.conf and modify your pg_hba.conf file to use scram-sha-256 algorithm. For more information, see
- https://www.postgresql.org/docs/12/auth-pg-hba-conf.html
TYPE DATABASE USER ADDRESS METHOD
local all all scram-sha-256
Procure the Certificate Authority (CA) signed certificate for the PostgreSQL database from the system administrator of your organization. Ensure that the certificate is in x509 format. For example, postgres.crt. Save the procured certificate file in the following locations:
SZTPD Server:
To start a PostgresSQL shell client connecting to the default Postgres database, type:
$ psql postgres
psql (12)
Type "help" for help.
You can start the server from a specific directory. To do this use the command and substitute in for the specified values:
pg_ctl -D [Data Directory] -l [Log file] start
The “Data Directory” refers to the directory that was just initialized. The “Log file” is a file that will record server events for later analysis. Generally log files are formatted to contain the date in the file name (e.g. “2018-05-27.log” or “myData-logfile-2018-05-27.log”) and should be stored outside of the database that they are logging so as to avoid unnecessary risks.
The server will only start if the port is free. If the default server is running it must first be stopped using command:
pg_ctl -D /usr/local/var/postgres stop
Once started, the database instance can be connected using an open source admin and development tool such as pgAdmin or simply a PostgreSQL shell:
$ psql —help
Connection options:
-h, --host=HOSTNAME database server host or socket directory (default: "local socket")
-p, --port=PORT database server port (default: "5432")
-U, --username=USERNAME database user name (default: "kristen")
-w, --no-password never prompt for password
-W, --password force password prompt (should happen automatically)
$ psql postgres
Once inside PostgresSQL shell, create a user account with a password and permission to create a database, as follow:
postgres=> CREATE USER my_user WITH LOGIN ENCRYPTED PASSWORD 'my_pass’ CREATEDB;
For more details, see https://www.postgresql.org/docs/9.1/sql-createrole.html
Exit out and login back in to verify that the user was created successfully
postgres=>exit
$ psql sztpd -U my_user
postgres=conninfo
You are connected to database "sztpd" as user "my_user" via socket in "/tmp" at port "5432".
Then you can start the SZTPD server, which creates a schema and populate seed data for the tables.
While the SZTPD server is running and waiting for client connections, you will want to verify that the schema and tables have been created successfully.
sztpd=> dn
List of schemas
Name | Owner
--------+----------
public | kristen
sztpd | my_user
And doubly so when using an in-memory database, though doing so in a production environment doesn’t make sense.↩︎
A profile of each kind of bootstrapping device is needed but, in general, device boot times may be an order of magnitude more time than any networking activity, thus the speed of the networking interface may be unimportant in the end.↩︎
Some more so than others. For instance, OpenBSD is pretty secure out of the box.↩︎
Even when external TLS termination is deployed, it still seems better to dedicate a port for each interface from both performance and monitoring perspectives.↩︎
When “HTTPS” is configured, SZTPD asserts that each port must have a unique private key and hence certificate. When “HTTP” is configured how the TLS certificates are managed is outside of SZTPD’s scope.↩︎
For instance, using a private PKI.↩︎
For instance, using a common public PKI.↩︎
There is an ability to offload the TLS termination to a device fronting it (e.g., a firewall). In either case, the traffic to the client endpoints is always protected by TLS.↩︎
Note that long passwords, such as might better known as passphrases, are recommended. Setting the minimum length to, e.g., 25 characters encourages that best practice.↩︎
SZTPD does not use any configuration files of any sort.↩︎
Really each time it runs against an uninitialized database, in case the database is ever reinitialized.↩︎
Warning: As per the Release Notes, the ‘product mode’ concept may disappear in the future.↩︎
The “device ownership verification” callback function is used to ensure that the devices configured by tenants are rightfully owned by the tenants.↩︎
Not yet implemented.↩︎
Note that the key cannot be stored in the database since it is needed in order to establish a connection to the database.↩︎
For instance, by sending the ‘SIGTERM’ signal to the ‘sztpd’ process. Alternatively, if using a process manager, via a command such as ‘stop sztpd’.↩︎
Only occurring only on major SZTPD product version boundaries↩︎
The name of the database SZTPD creates/uses is specified at the end of the database URL string passed into the ‘sztpd’ command.↩︎
SZTPD initializing the database only happens the firsttime it is run (i.e., when the database doesn’t exist). Upon subsequent invocations, SZTPD will simply use the already created database, which is then presumably populated with data.↩︎
These command line parameters cannot be stored in the database as they are needed in order to establish a connection to the database.↩︎
These command line parameters cannot be stored in the database as they are needed in order to establish a connection to the database.↩︎