The server for the BSim Database is distinct from the traditional Ghidra server, although for many use cases it is convenient to have both running and view the BSim server as a loosely coupled extension to the base Ghidra Server. In terms of start-up, shutdown, and configuration however, the two servers are completely separate.
There are two choices for deploying a shared server for the BSim Database: PostgreSQL or Elasticsearch. Both options support multiple users and allow multiple simultaneous connections to the remote server. A single database can ingest large datasets, while still maintaining short query times.
Alternately, a single user can create a BSim database on their local file-system, without a server, by utilizing the H2 database engine integrated into Ghidra. This option is intended for querying against small datasets and does not require installation of additional server software.
A PostgreSQL server, which must be built with a BSim specific extension, runs on a single host and makes efficient use of whatever CPU, memory, and disk resources are made available to it. PostgreSQL is a robust and capable server that should perform well on minimally configured workstations up to high-end production hardware. Source for the BSim extension to PostgreSQL is included as part of the Ghidra installation, but the PostgreSQL source may need to be obtained separately by the database administrator. See “Building the Server”
An Elasticsearch server, which must have a BSim specific plug-in installed, runs as a scalable database that automatically distributes itself across machines in a cluster, allowing individual database queries and requests to be serviced in parallel. The Elasticsearch BSim plug-in is included with the Ghidra installation, but the core server software must be obtained separately by the database administrator. See “Installing the Plug-in”
BSim clients included in the base Ghidra distribution can interface to any of these databases. Users that just want to connect to an existing shared server via a BSim client do not need to install any server software themselves.
In order to use PostgreSQL as a BSim server, it must be built with a BSim specific
extension, provided as part of the Ghidra installation. Prebuilt servers, like those
provided as OS distribution packages, will not work as is with BSim. For users on Linux
and macOS, the Ghidra installation provides a script, make-postgres.sh
,
in the module directory Ghidra/Features/BSim
that builds both the PostgreSQL
server and the BSim extension from source and prepares the installation for use with
Ghidra. If not already included in the Ghidra installation, the source distribution
file, currently postgresql-15.3.tar.gz
, can be obtained from the PostgreSQL
website at
https://www.postgresql.org/ftp/source/v15.3
|
The steps to build the PostgreSQL server with the BSim extension then are:
1) If not already present, place the PostgreSQL source distribution file
postgresql-15.3.tar.gz
in the Ghidra installation at
$(ROOT)/Ghidra/Features/BSim/postgresql-15.3.tar.gz
|
2) From the command-line, within the same directory, run the script make-postgres.sh
cd $(ROOT)/Ghidra/Features/BSim
|
./make-postgres.sh
|
Additional packages or software may need to be installed on the host OS in order for the build to complete successfully, OpenSSL in particular is required for BSim. For the full list of PostgreSQL software dependencies, refer to:
https://www.postgresql.org/docs/current/install-requirements.html
|
Once the build has completed successfully, the bsim_ctl command-line script is ready to use for starting a server (see “Starting and Stopping the Server”). The PostgreSQL server software will run out of the Ghidra installation at
$(ROOT)/Ghidra/Features/BSim/build/os/linux64/postgresql
OR |
$(ROOT)/Ghidra/Features/BSim/build/os/osx64/postgresql |
Other than having the extension itself, a BSim enabled PostgreSQL server is completely standard, and can be configured like any other stand-alone PostgreSQL server. There are no direct restrictions on modifying the configuration values. A default configuration is provided with this installation that has been tuned specifically for the BSim Database application, so in practice there may be little reason to modify it. But there are a few standard configuration values for the server that might need adjusting. See “Additional Configuration”.
The basic start-up and shut-down is accomplished with the same command-line script, which takes either the keyword start or stop as the first parameter. The second parameter must be an absolute path to the chosen data directory.
$(ROOT)/support/bsim_ctl start
/path/to/datadir |
$(ROOT)/support/bsim_ctl start /path/to/datadir
--port 8000 |
$(ROOT)/support/bsim_ctl stop
/path/to/datadir |
$(ROOT)/support/bsim_ctl stop /path/to/datadir
force |
The data directory should already exist and should initially not contain any files. The first time a server is started for a particular data directory, a large number of configuration files and other sub-directories associated with the PostgreSQL server will automatically be created. Upon subsequent restarts the existing configuration will be reused.
The start command can take an optional --port parameter. This can be used to specify a non-standard port for the PostgreSQL server to listen on. In this case, any subsequent reference to the BSim server, in the Ghidra client, or with the bsim command described below, must specify the port. When using the bsim command, a non-default port must be explicitly specified with the BSim postgresql:// URL (see “Ghidra and BSim URLs” for more details).
The stop command can take the keyword force as an optional parameter. Without this, the shutdown of the server will wait until all currently connected clients finish their sessions. Adding this parameter will cause all clients to be disconnected immediately, rolling back any transactions, and the server will shutdown immediately.
BSim makes use of PostgreSQL security mechanisms to enforce privileges and authenticate users. The bsim_ctl command wraps the subset of functionality described here, but other adjustments are possible by connecting directly to the server and issuing SQL commands.
The PostgreSQL server, as configured for BSim, only accepts connections via SSL, so communications in transit are always encrypted regardless of the authentication settings.
PostgreSQL uses the concept of roles to grant access privileges based on particular users. Generally, a user's role is determined by the username used to establish the connection. For BSim, each user role is granted one of two privilege levels: user, which allows read-only access to the server for normal queries, and admin, which additionally allows database creation, ingest, update, and deletion.
BSim supports three different authentication methods, when connecting as a client or during database ingest and maintenance. This method is established for a server by the initial start command.
bsim_ctl start /path/to/datadir
--auth trust
This is currently the default. No authentication is performed and privilege is granted based on the user name presented. Masquerading is possible.
bsim_ctl start /path/to/datadir
--auth password
Users are authenticated via password. A default password 'changeme' is established when the new user is created. Passwords can be changed by the user from the BSim client or can be reset by an administrator via the resetpassword command.
bsim_ctl start /path/to/datadir --auth pki
--cafile "/path/to/rootcert"
Users are authenticated by PKI certificates. Upon initialization, the BSim server must be provided (via the --cafile option) a file containing the public keys for the certificate authorities used to issue user's certificates. The file consists of the authoritative certificates in PEM format concatenated together.
BSim users must register their certificate with the Ghidra client using the Edit->Set PKI Certificate... menu option from the Project dialog. The BSim client will automatically submit the certificate to a server that requests it, and the password to unlock it will be requested as needed. This is the same mechanism used to a access a PKI protected Ghidra server, and if a user needs access to both a BSim server and Ghidra server that are PKI protected, the servers should probably be configured with the same certificate authorities so that they will accept the same certificate from the user.
With PKI authentication enabled, at the time a new user role is established with the server, the X.509 Distinguished Name, as bound to the user's certificate, must be associated with the user name via the --dn option. See “Adding Users to the Database”.
The authentication method should be established once, the first time the start command is issued for the server on an empty data directory. Subsequent restarts of the server will not change these settings. If the settings really need to be changed, the changeauth command can be issued. It takes the same options as the start command and can only be run if the server is shutdown first.
$(ROOT)/support/bsim_ctl changeauth
/datadir/path --auth password |
Using the changeauth command on a server with an established set of users will likely require other disruptive changes to create passwords or associate Distinguished Names with users, if they didn't exist before.
If it is determined that only the database administrators have OS level, local, access to the server's host machine, they can choose to use the noLocalAuth option as part of the start or changeauth commands. This disables authentication for users connecting to the server by the 'localhost' interface. This may facilitate the use of scripts for ingest etc., where working with passwords is cumbersome. Authentication is still enforced for any remote connection.
The username used to start the server for the first time, causing the initialization of the data directory, becomes the administrator for that server. No other username/role is initially known to the server. New usernames/roles can be added to the server using the following command:
$(ROOT)/support/bsim_ctl adduser username |
$(ROOT)/support/bsim_ctl adduser username --dn "C=US,ST=MD,CN=Firstname User" |
If password authentication has been set for the server, the new user's password will initially be set to 'changeme'. If PKI authentication has been set for the server, The Distinguished Name, as bound to the new user's certificated must be provided when issuing the adduser command, via the --dn option. The Distinguished Name must be presented as a string containing a comma separated sequence of attribute/value pairs that uniquely identifies a certificate. Currently, the Common Name (CN=) is the only attribute inspected by the PostgreSQL server, so other attributes can be omitted.
New users are by default only given user permissions, meaning that they can only place queries to the database and cannot ingest, update, or delete data. The new user can be given admin privileges (by an existing administrator) by issuing the command:
$(ROOT)/support/bsim_ctl changeprivilege username admin |
The relevant configuration files are at the top level of the data directory:
postgresql.conf |
pg_hba.conf |
The most important configuration parameters in postgresql.conf
are:
This controls the amount of RAM available for caching database pages across all connections to the server. The default should be reasonable in most situations, but for large databases or many simultaneous connections it might make sense to increase this.
These control how often the server forces database pages to be written back out to the file-system. The defaults are set to minimize disk writes when ingesting large numbers of records in one session. There should be little reason to change these values.
This controls the minimum SSL/TLS protocol version used when the server negotiates a connection. The current default is 'TLSv1.2'
The pg_hba.conf
file is used to configure which
connections the server accepts for a particular outward facing IP address and what
security mechanisms are enforced for those connections. Currently all addresses are
configured to accept SSL connections only, except possibly for 'localhost'.
Administrators can currently filter connections
based on usernames and the particular database (which corresponds to Ghidra's concept
of repository).
By default, the server accepts all connections from all users.
There is a serverconfig.xml
which contains a few of
the default configuration values that are most crucial for the BSim Database. Beware: This file is currently parsed only once
for the entire lifetime of a particular data
directory: it is read only when the data directory is first initialized, i.e. the first
time the bsim_ctl start command is
invoked on the empty directory. This file is intended to provide reasonable defaults
that are different from the standard PostgreSQL defaults. To provide site specific
configuration, changes should be made to the normal PostgreSQL configuration files.
A full description of how to configure an Elasticsearch cluster is beyond the scope of this document. In particular, the bsim_ctl command-line, as described in “PostgreSQL Configuration”, does not apply to Elasticsearch. Complete documentation for administering a database is available on-line from the Elasticsearch website.
The following discussion describes how to set up a toy, or single node, server, using the
free and open Elasticsearch distribution. This distribution includes a REST API for administering
a database, which can be accessed using curl
commands or some other method
to send HTTP requests directly to the node.
In order to make use of Elasticsearch with BSim, the database administrator must install the lsh.zip plug-in as part of the Elasticsearch deployment. The plug-in is available in the Ghidra add-on named BSimElasticPlugin, which unpacks into a standard Ghidra installation. The file lsh.zip is a standard Elasticsearch plug-in that must be installed on every node of the cluster before a BSim repository can be created. The description below shows how to enable the BSim plug-in for a single node, but this will need to be repeated for any additional nodes.
Assuming the add-on has been unpacked, the plug-in can be installed to a single node using the elasticsearch-plugin command in the bin directory of the node's Elasticsearch installation.
bin/elasticsearch-plugin install
file:///path/to/ghidra/Ghidra/contrib/BSimElasticPlugin/data/lsh.zip |
Replace the initial portion of the absolute path in the URL to point to the Ghidra installation. Once the plug-in is installed, the toy deployment can be (re)started from the command-line by running
bin/elasticsearch |
This will dump logging messages to the console, and you should see [lsh]
listed among the loaded plug-ins as the node starts
up.
The open Elasticsearch distribution starts with up with password authentication
enabled by default. When a node is started up for the first time, as described above, an
elastic user is created with a randomly
generated password that is reported, once, to the console. For a toy deployment, it may
be convenient to add additional users via curl
commands. The
following example creates a user named ghidrauser
with a default password "changeme", using the elastic users credentials.
The generated password for the elastic user must be substituted for the XXXXXX
at the beginning of the command.
curl -k -u elastic:XXXXXX -X POST "https://localhost:9200/_security/user/ghidrauser?pretty" -H 'Content-Type: application/json' -d' { "password" : "changeme", "roles" : [ "viewer" ], "full_name" : "Ghidra User", "email" : "ghidrauser@example.com" } '
Elasticsearch uses the concept of roles to grant access privileges to particular users. The built-in role viewer, as in the example above, can be used to grant users read-only access to a database. The built-in superuser role grants administrator privileges.
Assuming an Elasticsearch cluster is running and the plug-in has been properly installed, all other parts of BSim interact transparently with the cluster. The bsim command, described in Ingesting Executables, and the Ghidra/BSim client, described in Querying a BSim Database, require no additional configuration to work with Elasticsearch, except users must provide the correct URL to establish a connection. Elasticsearch communicates over https, and BSim clients automatically assume they are communicating with Elasticsearch when they see this protocol. Alternatively, the protocol may be specified as elastic when using the bsim command. Elasticsearch use by BSim assumes a default port of 9200 unless otherwise specified when specifying the server host. See “Ghidra and BSim URLs” for additional information about URLs.
If using either PostgreSQL or Elasticsearch the server must be properly configured and running before a database can be created. In the case of an H2 file-based database there is no server requirement. Only after a database has been created can data be ingested or queries performed. In this context, a database is a single container of reverse engineered functions. Metadata pertaining to executables and call-graph relationships is also stored, but the principle database record describes a function. A single PostgreSQL or Elasticsearch server can hold multiple independent databases.
A database is created using the bsim command script. The basic command looks like
$(ROOT)/support/bsim createdatabase bsimURL config_template |
A BSim database is completely distinct from the Ghidra Server or Ghidra project, so the executables and functions contained within do not need to coincide at all.
The Ghidra GUI client specifies a BSim database with its explicit characteristics (i.e., DB type, name, host/port if applicable, etc.), while the bsim command accepts a bsimURL which includes similar details (see “Ghidra and BSim URLs” for more details).
The config_template parameter passed to bsim createdatabase names a collection of specific configuration values for the newly created database. A standard Ghidra distribution provides a number of predefined templates (See below) designed for specific database use cases. It is simplest to use a predefined template when creating a database, but it is possible to edit an existing template or create a new template (See “Creating Database Templates”).
There are two critical database properties being determined by the template that need to be kept in mind: the index tuning and the weighting scheme relative to the size of the database. The two pieces of the template name, separated by the '_' character, refer to these concerns.
The index tuning affects the use of the database by trading off between, the time required to perform individual queries, the amount of variation between matching functions a query can tolerate, and the amount of storage required per database record. Ideally, the database is tuned, before the initial ingest occurs, to the anticipated size of the database. The database can trade off storage size (per record) and latency for overall query response time, but the decision needs to be made before the database is populated. Currently there is a medium tuning that is ideal for repositories that will store on the order of 10 million functions. There is also a large tuning, which uses more storage but can maintain fast query times for databases with 100 million functions or more. There is a large overlap for these tunings, so if its unclear how large a database might grow, go ahead and use the medium tuning.
The weighting scheme affects how BSim views the relative importance of individual code constructs within a function. Code constructions are extracted as features, and each feature is assigned a weight. The basic schemes are: 32 for 32-bit compiled code, 64 for 64-bit code. The scheme that matches the predominant form of code in the repository being ingested should be used. Mixed schemes are possible, but a corpus which is predominantly 32-bit, even with a small number of 64-bit executables mixed in, should still use the 32-bit weights.
There are some weighting schemes designed for more specialized code. The 64_32 scheme is for 64-bit code using 32-bit pointers. The nosize scheme allows better matching of 32-bit functions to 64-bit functions, when they are compiled from the same source. The cpool scheme is designed for Java byte-code or Dalvik executables. For more discussion of weighting, see “Weighting Software Features”.
The full template name incorporates both an index tuning and a weight scheme. Some common examples of template names:
A medium index tuning with a weighting scheme designed for 32-bit executables.
A medium index tuning with a weighting scheme designed for 64-bit executables.
A 32-bit weighting scheme with tuning for a large database size.
A medium index tuning with a weighting scheme for Java executables.
A medium index tuning with a weighting scheme allowing matches between 32-bit and 64-bit code.
There is some facility to tailor a specific BSim database instance so that it can ingest and/or report information about executables or functions to make results more useful for a specific project or user. Capabilities can be added after a database has been created and is running by issuing specific bsim commands, but they can also be added to a configuration template prior to creating the database, which provides a record of the specific additions should the database instance need to be recreated or multiple tailored instances be deployed. For additions that allow the ingest of more metadata about executables or functions, users must provide additional scripts to Ghidra during the ingest process in order to read in or discover the new metadata.
The Name, Owner, and Description associated with a BSim instance can be trivially tailored with the bsim setmetadata command.
$(ROOT)/support/bsim setmetadata bsimURL --name "BSim Database" |
$(ROOT)/support/bsim setmetadata bsimURL --owner "Administrators" |
$(ROOT)/support/bsim setmetadata bsimURL --description "Files of interest" |
This information is displayed in various windows by the BSim client. The values can be changed at any time and do not otherwise affect the records contained in the database. Multiple command-line parameters can be fed to bsim setmetadata so long as each one starts with --name, --owner, or --description respectively. Quoting of values may be necessary to get some strings to be interpreted as a single command-line parameter.
BSim provides the powerful ability to associate new types of metadata with each executable that the database ingests. Any method of categorizing executables that describes an executable with a simple string value, referred to here as an executable category, can be added as a field to the database. With only minor adjustments to the ingest process, new category values can be automatically attached to incoming executables and are treated like any other executable field that BSim understands. Category values are retrieved with queries, can be used for filtering, and show up as sortable columns in result tables.
All categories have a formal name (or type), which is used both in the ingest process (See below) and as the label for table columns. The name can contain alphanumeric characters or punctuation from the limited set, ' ._:/()'. For each executable there can be zero, one, or more string values associated with the category. No value is required for the executable, and any single value can be used for filtering (either the executable is labeled with the value or it is not) even if there are multiple values for that category. If there are multiple values, a query that matches the executable will return all the values as a single sorted column entry.
It is also possible to create a special time-based category. This category can have any name as above, but instead of associating string values with the executable, it associates a single time-stamp. The time-stamp has precision down to the millisecond and provides filtering and sorting based on time. Internally, this new category repurposes the column storage originally providing an executable's Ingest Date field. As a result, any BSim instance can have only one time category and only one time-stamp within it. The ingest scripting must provide any actual time-stamp value for the executable, or the database will fill in the "epoch", 12:00 am, Jan 1, 1970.
A new category can be added to the database at any time using the bsim addexecategory command.
$(ROOT)/support/bsim addexecategory bsimURL MyCategoryName |
$(ROOT)/support/bsim addexecategory bsimURL MyTimeField date |
The single time-stamp field can be renamed by appending the keyword "date" to the command after the name of the category. Once a category, the corresponding program options set for any new executables will automatically read into the database as part of the ingest process. Previously ingested executables, assuming they have the new program options set, can be updated within the BSim database using one of the bsim updaterepo command variants. In either case, the relevant program options typically need to be filled by running a Ghidra script (See “Ingesting Executable Categories”). There is currently no method for deleting a category once it has been created.
BSim can be configured to recognize specific Function Tags, which are named Boolean properties on individual functions within an executable. Within a Ghidra program, any number of different function tags can be established by the user and are used to label individual functions or specific subsets of functions that share a particular property. This would typically be used to label classes of functions that are important to analysts, unpacked functions could be labeled with the tag UNPACKED for instance.
In order for BSim to recognize specific function tags, they must be individually registered with the BSim database. These tags are then automatically ingested into the database, along with the other standard metadata describing functions, and can be used to filter match results when querying the database. A function tag has a formal name, which can be displayed as part of the function header within the main code browser and is used for BSim columns and filter labels. Once the tag is created for a program, functions universally have the tag as a Boolean property, either the name applies to a function or it doesn't, and arbitrary subsets can be tagged with that name.
A tag must be registered with a BSim database before it can be used as a filter or seen in results. A tag can be registered at any time with the bsim addfunctiontag command.
$(ROOT)/support/bsim addfunctiontag bsimURL MyTagName |
The new tag will automatically be read in when any new executables are ingested. If previously ingested executables already had the new tags before they were registered, their metadata within BSim database can be updated using the bsim updaterepo command variants. BSim is limited to 29 registered tag names, and there is currently no way to remove a tag once it has been registered.
It is possible to create tailored database configuration templates so that implementers have a permanent and accessible record of a particular set-up and don't need to repeatedly issue bsim setmetadata and bsim addexecategory when creating a database. Other aspects of a database can also be manipulated, like weighting schemes and index tuning, but doing this properly is beyond the scope of this document. A database template is the basic set of configuration parameters used to set up BSim database instance. The configuration parameters are established for a particular database when the bsim createdatabase command is run (See “Creating a Database”). The template name passed on the command-line actually identifies an XML file-name, appended with the '.xml' suffix, in the directory:
$(ROOT)/Ghidra/Features/BSim/data |
The file has a root tag of <dbconfig>, and the first child tag of this root is the <info> tag. This tag contains all the metadata tags that can be easily changed or added to the database. A list of the metadata tags follows. The metadata is provided as formal text content within the tag, and none of the tags currently take attributes.
XML Tag | Description |
<name> |
Name of the database |
<owner> |
Owner of the database |
<description> |
An overarching description of the database |
<major> |
Major decompiler version used for ingest (Should be set to zero) |
<minor> |
Minor decompiler version used for ingest (Should be set to zero) |
<settings> |
Specific settings for the signature strategy (Should be set to zero) |
<execategory> |
The name of an executable category (to be) defined for this database instance |
<datename> |
The name of the timestamp column |
<functiontag> |
The name of a function tag (to be) registered with this database instance |
There can be multiple <execategory> tags if more than one category is desired and both <execategory> and <datename> are optional tags. The date column name defaults to 'Ingest Date' and is drawn from the corresponding Ghidra program option. The tag order needs to be preserved. There can be multiple <functiontag> tags, one for each function tag to be registered with the database.
It is easiest to copy an existing template and just edit the tags described above. The remaining tags in the file are more dangerous to manipulate. The <k> and <L> tags pertain to the index tuning. The <weightsfile> tag gives the name of the weights file, within the same directory, which is also another XML file. It is simplest to choose from the existing weight files provided with the distribution. See “Weighting Software Features”.