Audit Logging

The Audit Logging entries are delivered over syslog using UDP transport. This transport must be configured before Audit Logging is enabled in order to collect the initial messages.

For the overall procedure, see Creating a Security World using new-world.

Key considerations:

When you configure an Audit Logging Security World:

When you indoctrinate a new HSM into an Audit Logging Security World:

Check for AuditLogging on the state line in the output of nfkminfo.

Check AuditLogging on the active modes line in the output of enquiry.

Audit Logging is set for the lifetime of the Security World.

To disable Audit Logging on an HSM:

The Audit logging functionality is based on that described in RFC-5848 ( https://tools.ietf.org/html/rfc5848 ). This describes a mechanism also known as syslog-sign that adds the following capabilities to syslog:

It is implemented on top of a standard syslog data stream and does not use any additional protocol. The syslog-sign logging scheme adds a number of control messages to the log entries that are to be audited. These are also implemented as syslog messages. The following sections outline the audit log entries that are present in the syslog data stream for nShield Audit Logging. The signing mechanism used is DSA with a 3072 bit key and SHA-256 as the hashing mechanism.

Audit log entry

This is the log message from the entity being audited. It is in a standard format and includes operation specific data required to provide an auditing capability. As each log message is generated on the HSM, a digest operation is performed on it and the digest buffered in the HSM.

Signature Block

When sufficient digests have been accumulated (N), a Signature Block is generated as a standard log entry containing the following:

The number of messages is dictated by the transport medium, the size of the digests, the size of the signature and the size of other data contained in the message. There is a limit to the size of messages that can be transported over syslog. The signature is performed using a log signing key. This key is generated and the private half is held securely in the HSM.

Certifier Block

Verification of the Signature Blocks requires that you, or the application performing the verification, has access to the public half of the log signing key. The Certifier Block provides a mechanism for the log verifier to get access to this key. The key is packaged in a form allowing the source of the audit logs to be verified. As the size of this information may be too large for the syslog transport medium it can be broken down into Certifier Block Fragments which are compatible with the syslog transport mechanism. When all of these fragments are received by the log verifier, it can reconstruct the public half of the log signing key and perform any consistency checks and origin verification that is needed.

Given the public half of the log signing key, a Signature block and its corresponding log entries, the verifier can check the signature on the Signature Block. When this is verified, the log entry digests in the Signature Block are implicitly verified. The integrity of the corresponding log entries can be verified by performing a digest on received log entries and comparing them to the corresponding verified digests in the Signature block. The image below shows the basics of this approach. For more information, see Audit Log Verification.

The nShield Audit Logging capability uses the RFC-3164 (https://tools.ietf.org/html/rfc3164) standard for distributing audit log messages. All audit log messages will have the following header prepended. This header is applied by the hardserver before sending the message and does not form part of the signed audit log messages. The signed portion of the audit log message starts at the CEF:0 CEF identifier and continues to the end of the message.

The PRI value of this header <134> indicates the facility local0 and a severity of info.

The syslog infrastructure used for Log distribution is out of the scope of this guide and your responsibility to implement. Log distribution for Audit Logging uses syslog as the transport medium which allows a standard protocol and message format to be used for the Audit Logging messages. This transport is compatible with SIEM systems and the wider syslog infrastructure in an organization can be used to further distribute or process the log stream. There are many possible configurations of syslog deployment, as shown below.

rsyslog can be configured to not reformat messages using the following approach:

syslog-ng does not appear to rewrite messages in the same way as rsyslog. Refer to the syslog-ng documentation for information on formatting.

The audit log entries are emitted from the HSM in CEF format. This provides both human readable log messages and compatibility with SIEM applications. As indicated in the previous section the audit log entries are distributed using the syslog transport mechanism.

A CEF format log message is shown below:

The rest of the log message is made up of CEF extensions. These are name/value pairs that are used to convey specific information for the log message. The name-value pairs can be processed by SIEM applications such as Arcsight and can be displayed in tabular reports of the messages received. They can be used for filtering and further processing within the SIEM application. The following table specifies the meaning and format of the extensions used by the Audit Logging facility.

The Audit Logging Implementation requires a number of infrastructure CEF extensions to provide data necessary for the RFC-5848 based signed syslog approach used. Please refer to RFC-5848 for further details on these infrastructure extensions. These CEF extensions replace the RFC-5424 Structured Data used in the original scheme but have the same meaning.

There are two counters that are sent with all Audit Logging command log messages. The Reboot Session ID is also sent with Certifier Block and Signature Block messages.

The following extensions are used in the Certifier Block where the CEF header name element is ssign-cert. The Certifier Block is used to distribute the log signing public key. It is sent by the HSM when logging is enabled and every time the HSM is restarted. This provides for redundancy as any Certifier Block from an HSM that has been configured for Audit Logging will contain the same log signing public key. The Certifier Block can extend over multiple syslog messages. The extensions identified here allow the data of a Certifier Block to be rebuilt from multiple fragments. Sufficient fragments are sent in separate ssign-cert messages to rebuild the payload block. See Certifier Block example for the details of the data included in the certifier block.

The following extensions are used in the Signature Block where the CEF header name element is ssign. The Signature Block supports the verification of Audit Logging messages. The Signature Block is sized to fit within a syslog message which dictates the number of audit log messages it covers. This release supports a maximum of 10 audit log messages per Signature Block. The main data for the Signature Block is the SHA-256 hashes of the log messages covered by the block.

This section shows example Certifier Block, Signature Block and Audit Logging messages and shows how the CEF extensions are used together.

By default the nShield Audit Logging Facility does not log usage of keys for cryptographic operations such as sign, verify, encrypt and decrypt or their usage in channels for these purposed. Creation, Deletion and a number of other key operations are unconditionally logged by default. The Audit Logging feature provides the capability to optionally log these operations. This is determined on a per-key basis by the LogKeyUsage permission group flag on the ACL group authorizing the operation for which logging is desired. See the nCore Developer Tutorial for further information on ACLs.

The generatekey utility (see Key generation options and parameters) provides the ability to set this permission group flag when a key is generated by either:

When generatekey is used this flag is applied to all permission groups but is only checked by the HSM on the group authorizing the desired action.

The following example shows this set on permission group 0 of a key’s ACL.

In the following sections, the tables will indicate if this mechanism is required to generate a log message for a specific command or key.

The following tables list the nCore commands that generate Audit Logging messages. For each command they identify command specific data that is contained in the log message and the CEF extension used to identify it.

nCore Key and Logical Token hashes are the standard nCore identifying hashes. They are used to identify a key or logical token as it is an invariant for the key or logical token. These hashes are logged as lower-case hex encoding. In some cases a short hash may be presented. This is the first 10 bytes of the hash in a lower-case hex encoding.

For each command logged the command is specified by the name element of the CEF header. The other elements of the CEF header are filled as detailed in the previous section. All commands being logged will also include the following CEF extensions:

Commands with Yes in the Requires logkeyusage ACL column will only be logged if the Key’s ACL contains the LogKeyUsage Flag in the permission group authorizing the operation.

These commands do not use the logkeyusage ACL mechanism and log unconditionally.

These commands are logged unconditionally.

These commands are logged unconditionally.

The heartbeat is a periodic audit log message sent every 15 minutes. This audit log message indicates that the HSM is still active. After a heartbeat event is logged a Signature Block is generated including the heartbeat log message and any outstanding audit log messages. Waiting until the heartbeat is logged before restarting the HSM will ensure outstanding log messages can be verified.

The nShield HSM has a number of commands and errors that cannot be logged directly when they occur. This applies primarily to errors detected during processing or self test and the reboot command Cmd_ClearUnit. The strategy adopted for these is to persist sufficient information and replay them as log entries after a successful reboot of the HSM. These reboot event messages occur after the Certifier Block has been emitted.

Each of these messages are emitted with rsid and seqNo relating to the current session and will have a prevrtc CEF element recording the RTC at the time of the event. The name element will identify the event. If the event is associated with a nCore SOS code this will be indicated by a sos CEF extension and an appropriate code. The Device Event Class Id is set to 5 and Severity will be set to 10 for errors or 6 for shutdown events. The source CEF extension will be internal. The following table lists the events replayed in a post reboot log. The available events depend on the type of HSM.

As an example, the following shows a post reboot log of Cmd_ClearUnit. In this excerpt, it can be seen after the last fragment of the Certifier Block. A Signature Block is generated after the reboot log entries.

With the information logged as detailed in the preceding sections it is possible to trace back from a Key Command to the loading of the Key, then to loading the Logical Token and reading the Shares that constitute the Logical Token.

The following example shows the notional traceback from a Cmd_Encrypt operation. This command logs the nCore Key Hash KKKKKKK. Prior to this the Key was loaded onto the HSM using Cmd_LoadBlob which correlates the nCore Key Hash with the ncore Hash of the Logical Token that authorized loading the Key. Tracing further back we can identify the shares used to reconstruct that Logical Token. In this example two shares are required identified by share indices S1 and S2. The share index identifies a specific card in an OCS cardset.

The example verification program can be found in the following location:

This program requires the use of the nShield Python interpreter. This is necessary to provide support for the nShield specific marshalling functions used to export the log signing public key. The example verification program also requires the presence of an nShield HSM accessible to the machine on which the verification is to be performed. This is required to perform the cryptographic operations necessary to verify the log signing public key and the Signature Blocks. This HSM does not need to be the same HSM on which the logs were generated, nor does it need to be in a Security World.

The Audit Log verifier program is run with a command of the form:

Where:

The program takes and reads the input syslog file containing the log messages. It optionally sets the ESN of module for which log events are to be validated, if this was passed in. If an ESN is not provided as input then the first ESN found in the syslog will be processed.

The verifier calls its parse function which segregates the messages based on ESN, and creates lists of Certifier fragments, Signature Blocks and Log events, based on matching with regular expressions.

Syslog may have gathered logs from multiple sources. As such, the verifier has a concept of a logging world, which represents a set of logs, sigblocks and certblocks that belong together, from a Security World. Based on Reboot Sequence ID, Sequence Number of the Log event, Global block counter of the Signature block and Fragment index of the Certifier block, a logging world is identified and a logging instance is created.

All records are thus given a log-instance number, such that records with the same instance number belong together.

Each event can thus be uniquely identified via a tuple. For the log messages, signature blocks and certifier blocks these are respectively (rsid and sequence number), (rsid and gbc) and (rsid and findex).

The reconstruct_CBs function is then called to validate the certifier fragments (using calls to an nShield HSM for crypto functionality). It then reconstructs the certifier blocks from the certifier fragments.

A list of valid and verified Certifier Blocks is created.

For any log instance one valid Certifier Block is enough to validate the events, so further certifier blocks are ignored after the first.

Next the process_sbs function is called. Signature Blocks for a supplied ESN are validated per log instance (once again via calls to the module for crypto functionality), using the KAL value taken from the Certifier block previously.

The validated Signature block hashes are maintained as a dictionary of hashes with keys as unique ids. These unique ids per instance are generated based on rsid and sequence numbers.

The process_logs function is finally called. This generates the hash of each of the log events and matches against hashes from corresponding signature blocks. Verified and Tampered log events are then written to different files in the LogResult folder.

While the example verifier uses an HSM for cryptographic operation, it would be possible to use 3rd party cryptographic libraries to provide this functionality. This is outside the scope of this document.

Currently the log messages are verified against the hash in the signature blocks, and the signature of the signature blocks is verified against the key extracted from the certifier block. The certifier block its self is not verified. A potential extension to the verifier tool would be to verify the certifier block. The certifier block is signed by KLF2. This can be checked against the KLF2 value found within the module’s warrant. This would complete the chain of trust.

Additionally, the example verifier does not cope with fields that rotate back around to zero when their max size is exceeded. (for example, gbk, rsid or seqno). Currently logs, SBs and CBs are uniquely identified by (rsid and sequence number), (rsid and gbc) and (rsid and findex). This means that, if any of those values rotate back around to zero, we are no longer able to uniquely identify them. As a potential extension, RTC or line number values could be used to solve this.

The example verifier does not detect missing/deleted log messages in the case where a complete group of log messages are deleted, along with their corresponding SignatureBlock. Given that the SeqNo field increases for each log message, spotting missing SeqNos would reveal missing or deleted log messages. This is a potential extension.

The example verifier expects a static, unchanging log file to be supplied to it. This would be compatible with verifying a batch of log files at the end of each day, for example. A possible extension would be to extend the verifier to cope with a live stream of logs, continuously verifying them as they are generated.