Contents
This document summarizes the most important changes in the current Bro
release. For an exhaustive list of changes, see the CHANGES
file
(note that submodules, such as BroControl and Broccoli, come with
their own CHANGES
.)
Bro 2.5.5 primarily addresses security issues.
The following changes address potential vectors for Denial of Service reported by Christian Titze & Jan Grashöfer of Karlsruhe Institute of Technology:
to some tunable parameters:
Those options can be changed if one needs the previous behavior of a “net_weird”, “flow_weird”, or “conn_weird” event being raised for every single event. Otherwise, there is a new weird_stats.log which contains concise summaries of weird counts per type per time period and the original weird.log may not differ much either, except in the cases where a particular weird type exceeds the sampling threshold. These changes help improve performance issues resulting from excessive numbers of weird events.
Improved handling of empty lines in several text protocol analyzers that can cause performance issues when seen in long sequences.
Add ‘smtp_excessive_pending_cmds’ weird which serves as a notification for when the “pending command” queue has reached an upper limit and been cleared to prevent one from attempting to slowly exhaust memory.
Bro 2.5.4 primarily fixes security issues:
There’s also the following bug fixes:
Bro 2.5.3 fixes a security issue in Binpac generated code. In some cases the code generated by binpac could lead to an integer overflow which can lead to out of bound reads and allow a remote attacker to crash Bro; there is also a possibility that this can be exploited in other ways.
Bro 2.5.2 fixes a security issue in the ContentLine analyzer. In rare cases a bug in the ContentLine analyzer can lead to an out of bound write of a single byte. This allows a remote attacker to crash Bro; there also is a possibility this can be exploited in other ways. CVE-2017-1000458 has been assigned to this issue.
SMB analyzer. This is the rewrite that has been in development for several years. The scripts are currently not loaded by default and must be loaded manually by loading policy/protocols/smb. The next release will load the smb scripts by default.
- Implements SMB1+2.
- Fully integrated with the file analysis framework so that files transferred over SMB can be analyzed.
- Includes GSSAPI and NTLM analyzer and reimplements the DCE-RPC analyzer.
- New logs: smb_cmd.log, smb_files.log, smb_mapping.log, ntlm.log, and dce_rpc.log
- Not every possible SMB command or functionality is implemented, but generally, file handling should work whenever files are transferred. Please speak up on the mailing list if there is an obvious oversight.
Bro now includes the NetControl framework. The framework allows for easy interaction of Bro with hard- and software switches, firewalls, etc. New log files: netcontrol.log, netcontrol_catch_release.log, netcontrol_drop.log, and netcontrol_shunt.log.
Bro now includes the OpenFlow framework which exposes the data structures necessary to interface to OpenFlow capable hardware.
Bro’s Intelligence Framework was refactored and new functionality has been added:
There is a new file entropy analyzer for files.
Bro now supports the remote framebuffer protocol (RFB) that is used by VNC servers for remote graphical displays. New log file: rfb.log.
Bro now supports the Radiotap header for 802.11 frames.
Bro now has rudimentary IMAP and XMPP analyzers examining the initial phases of the protocol. Right now these analyzers only identify STARTTLS sessions, handing them over to TLS analysis. These analyzers do not yet analyze any further IMAP/XMPP content.
New funtionality has been added to the SSL/TLS analyzer:
Bro now tracks VLAN IDs. To record them inside the connection log, load protocols/conn/vlan-logging.bro.
A new dns_CAA_reply() event gives access to DNS Certification Authority Authorization replies.
A new per-packet event raw_packet() provides access to layer 2 information. Use with care, generating events per packet is expensive.
A new built-in function, decode_base64_conn() for Base64 decoding. It works like decode_base64() but receives an additional connection argument that will be used for decoding errors into weird.log (instead of reporter.log).
A new get_current_packet_header() bif returns the headers of the current packet.
Three new built-in functions for handling set[subnet] and table[subnet]:
Several built-in functions for handling IP addresses and subnets were added:
The IRC analyzer now recognizes StartTLS sessions and enables the SSL analyzer for them.
The misc/stats.bro script is now loaded by default and logs more Bro execution statistics to the stats.log file than it did previously. It now also uses the standard Bro log format.
A set of new built-in functions for gathering execution statistics:
get_net_stats(), get_conn_stats(), get_proc_stats(), get_event_stats(), get_reassembler_stats(), get_dns_stats(), get_timer_stats(), get_file_analysis_stats(), get_thread_stats(), get_gap_stats(), get_matcher_stats()
Two new functions haversine_distance() and haversine_distance_ip() for calculating geographic distances. The latter function requires that Bro be built with libgeoip.
Table expiration timeout expressions are evaluated dynamically as timestamps are updated.
The pcap buffer size can be set through the new option Pcap::bufsize.
Input framework readers stream types Table and Event can now define a custom event (specified by the new “error_ev” field) to receive error messages emitted by the input stream. This can, e.g., be used to raise notices in case errors occur when reading an important input source.
The logging framework now supports user-defined record separators, renaming of column names, as well as extension data columns that can be added to specific or all logfiles (e.g., to add new names).
The new “bro-config” script can be used to determine the Bro installation paths.
New BroControl functionality in aux/broctl:
New Bro plugins in aux/plugins:
- af_packet: Native AF_PACKET support.
- kafka : Log writer interfacing to Kafka.
- myricom: Native Myricom SNF v3 support.
- pf_ring: Native PF_RING support.
- postgresql: A PostgreSQL reader/writer.
- redis: An experimental log writer for Redis.
- tcprs: A TCP-level analyzer detecting retransmissions, reordering, and more.
Log changes:
Connections
The ‘history’ field gains two new flags: ‘^’ indicates that Bro heuristically flipped the direction of the connection. ‘t/T’ indicates the first TCP payload retransmission from originator or responder, respectively.
Intelligence
New field ‘matched’ to indicate which indicator type(s) caused the hit.
DNS
New ‘rtt’ field to indicate the round trip time between when a request was sent and when a reply started.
SMTP
New ‘cc’ field which includes the ‘Cc’ header from MIME messages sent over SMTP.
Changes in ‘mailfrom’ and ‘rcptto’ fields to remove some non-address cruft that will tend to be found. The main example is the change from
"<user@domain>"
to"user@domain.com"
.HTTP
Removed ‘filename’ field (which was seldomly used).
New ‘orig_filenames’ and ‘resp_filenames’ fields which each contain a vector of filenames seen in entities transferred.
stats.log
The following fields have been added: active_tcp_conns, active_udp_conns, active_icmp_conns, tcp_conns, udp_conns, icmp_conns, timers, active_timers, files, active_files, dns_requests, active_dns_requests, reassem_tcp_size, reassem_file_size, reassem_frag_size, reassem_unknown_size.
The following fields have been renamed: lag -> pkt_lag.
The following fields have been removed: pkts_recv.
The BrokerComm and BrokerStore namespaces were renamed to Broker. The Broker “print()” function was renamed to Broker::send_print(), and the “event()” function was renamed to Broker::send_event().
The constant SSH::skip_processing_after_detection
was removed. The
functionality was replaced by the new constant
SSH::disable_analyzer_after_detection
.
The net_stats()
and resource_usage()
functions have been
removed, and their functionality is now provided by the new execution
statistics functions (see above).
Some script-level identifiers have changed their names:
- snaplen -> Pcap::snaplen
- precompile_pcap_filter() -> Pcap::precompile_pcap_filter()
- install_pcap_filter() -> Pcap::install_pcap_filter()
- pcap_error() -> Pcap::error()
TCP analysis was changed to process connections without the initial SYN packet. In the past, connections without a full handshake were treated as partial, meaning that most application-layer analyzers would refuse to inspect the payload. Now, Bro will consider these connections as complete and all analyzers will process them normally.
The policy/misc/capture-loss.bro
script is now loaded by default.
The traceroute detection script package policy/misc/detect-traceroute
is no longer loaded by default.
Changed BroControl functionality in aux/broctl:
--analyze
, --set-seed
, and
--md5-hashkey
have been removed.Bro now has support for external plugins that can extend its core functionality, like protocol/file analysis, via shared libraries. Plugins can be developed and distributed externally, and will be pulled in dynamically at startup (the environment variables BRO_PLUGIN_PATH and BRO_PLUGIN_ACTIVATE can be used to specify the locations and names of plugins to activate). Currently, a plugin can provide custom protocol analyzers, file analyzers, log writers, input readers, packet sources and dumpers, and new built-in functions. A plugin can furthermore hook into Bro’s processing at a number of places to add custom logic.
See https://www.bro.org/sphinx-git/devel/plugins.html for more information on writing plugins.
Bro now has support for the MySQL wire protocol. Activity gets logged into mysql.log.
Bro now parses DTLS traffic. Activity gets logged into ssl.log.
Bro now has support for the Kerberos KRB5 protocol over TCP and UDP. Activity gets logged into kerberos.log.
Bro now has an RDP analyzer. Activity gets logged into rdp.log.
Bro now has a file analyzer for Portable Executables. Activity gets logged into pe.log.
Bro now has support for the SIP protocol over UDP. Activity gets logged into sip.log.
Bro now features a completely rewritten, enhanced SSH analyzer. The
new analyzer is able to determine if logins failed or succeeded in
most circumstances, logs a lot more more information about SSH
sessions, supports v1, and introduces the intelligence type
Intel::PUBKEY_HASH
and location SSH::IN_SERVER_HOST_KEY
. The
analayzer also generates a set of additional events
(ssh_auth_successful
, ssh_auth_failed
, ssh_auth_attempted
,
ssh_auth_result
, ssh_capabilities
, ssh2_server_host_key
,
ssh1_server_host_key
, ssh_encrypted_packet
,
ssh2_dh_server_params
, ssh2_gss_error
, ssh2_ecc_key
). See
next section for incompatible SSH changes.
Bro’s file analysis now supports reassembly of files that are not
transferred/seen sequentially. The default file reassembly buffer
size is set with the Files::reassembly_buffer_size
variable.
Bro’s file type identification has been greatly improved (new file types, bug fixes, and performance improvements).
Bro’s scripting language now has a while
statement:
while ( i < 5 )
print ++i;
next
and break
can be used inside the loop’s body just like
with for
loops.
Bro now integrates Broker, a new communication library. See aux/broker/README for more information on Broker, and doc/frameworks/broker.rst for the corresponding Bro script API.
With Broker, Bro has the similar capabilities of exchanging events and logs with remote peers (either another Bro process or some other application that uses Broker). It also includes a key-value store API that can be used to share state between peers and optionally allow data to persist on disk for longer-term storage.
Broker support is by default off for now; it can be enabled at configure time with –enable-broker. It requires CAF version 0.13+ (https://github.com/actor-framework/actor-framework) as well as a C++11 compiler (e.g. GCC 4.8+ or Clang 3.3+).
Broker will become a mandatory dependency in future Bro versions and replace the current communication and serialization system.
Add –enable-c++11 configure flag to compile Bro’s source code in C++11 mode with a corresponding compiler. Note that 2.4 will be the last version of Bro that compiles without C++11 support.
The SSL analysis now alerts when encountering SSL connections with old protocol versions or unsafe cipher suites. It also gained extended reporting of weak keys, caching of already validated certificates, and full support for TLS record defragmentation. SSL generally became much more robust and added several fields to ssl.log (while removing some others).
A new icmp_sent_payload event provides access to ICMP payload.
The input framework’s raw reader now supports seeking by adding an option “offset” to the config map. Positive offsets are interpreted to be from the beginning of the file, negative from the end of the file (-1 is end of file).
One can now raise events when a connection crosses a given size threshold in terms of packets or bytes. The primary API for that functionality is in base/protocols/conn/thresholds.bro.
There is a new command-line option -Q/–time that prints Bro’s execution time and memory usage to stderr.
BroControl now has a new command “deploy” which is equivalent to running the “check”, “install”, “stop”, and “start” commands (in that order).
BroControl now has a new option “StatusCmdShowAll” that controls whether or not the broctl “status” command gathers all of the status information. This option can be used to make the “status” command run significantly faster (in this case, the “Peers” column will not be shown in the output).
BroControl now has a new option “StatsLogEnable” that controls whether or not broctl will record information to the “stats.log” file. This option can be used to make the “broctl cron” command run slightly faster (in this case, “broctl cron” will also no longer send email about not seeing any packets on the monitoring interfaces).
BroControl now has a new option “MailHostUpDown” which controls whether or not the “broctl cron” command will send email when it notices that a host in the cluster is up or down.
BroControl now has a new option “CommandTimeout” which specifies the number of seconds to wait for a command that broctl ran to return results.
bro-cut has been rewritten in C, and is hence much faster.
File analysis
- Removed
fa_file
record’smime_type
andmime_types
fields. The eventfile_sniff
has been added which provides the same information. Themime_type
field ofFiles::Info
also still has this info.- The earliest point that new mime type information is available is in the
file_sniff
event which comes after thefile_new
andfile_over_new_connection
events. Scripts which inspected mime type info within those events will need to be adapted. (Note: for users that worked w/ versions of Bro from git, for a while there was also an event calledfile_mime_type
which is now replaced with thefile_sniff
event).- Removed
Files::add_analyzers_for_mime_type
function.- Removed
offset
parameter of thefile_extraction_limit
event. Since file extraction now internally depends on file reassembly for non-sequential files, “offset” can be obtained with other information already available – adding togetherseen_bytes
andmissed_bytes
fields of thefa_file
record gives how many bytes have been written so far (i.e. the “offset”).
The SSH changes come with a few incompatibilities. The following events have been renamed:
SSH::heuristic_failed_login
tossh_auth_failed
SSH::heuristic_successful_login
tossh_auth_successful
The SSH::Info
status field has been removed and replaced with
the auth_success
field. This field has been changed from a
string that was previously success
, failure
or
undetermined
to a boolean. a boolean that is T
, F
, or
unset.
The has_valid_octets function now uses a string_vec parameter instead of string_array.
conn.log gained a new field local_resp that works like local_orig, just for the responder address of the connection.
GRE tunnels are now identified as Tunnel::GRE
instead of
Tunnel::IP
.
The default name for extracted files changed from extract-protocol-id to extract-timestamp-protocol-id.
The weird named “unmatched_HTTP_reply” has been removed since it can be detected at the script-layer and is handled correctly by the default HTTP scripts.
When adding a logging filter to a stream, the filter can now inherit
a default path
field from the associated Log::Stream
record.
When adding a logging filter to a stream, the
Log::default_path_func
is now only automatically added to the
filter if it has neither a path
nor a path_func
already
explicitly set. Before, the default path function would always be set
for all filters which didn’t specify their own path_func
.
BroControl now establishes only one ssh connection from the manager to each remote host in a cluster configuration (previously, there would be one ssh connection per remote Bro process).
BroControl now uses SQLite to record state information instead of a plain text file (the file “spool/broctl.dat” is no longer used). On FreeBSD, this means that there is a new dependency on the package “py27-sqlite3”.
BroControl now records the expected running state of each Bro node right before each start or stop. The “broctl cron” command uses this info to either start or stop Bro nodes as needed so that the actual state matches the expected state (previously, “broctl cron” could only start nodes in the “crashed” state, and could never stop a node).
BroControl now sends all normal command output (i.e., not error messages) to stdout. Error messages are still sent to stderr, however.
The capability of processing NetFlow input has been removed for the time being. Therefore, the -y/–flowfile and -Y/–netflow command-line options have been removed, and the netflow_v5_header and netflow_v5_record events have been removed.
The -D/–dfa-size command-line option has been removed.
The -L/–rule-benchmark command-line option has been removed.
The -O/–optimize command-line option has been removed.
The deprecated fields “hot” and “addl” have been removed from the connection record. Likewise, the functions append_addl() and append_addl_marker() have been removed.
Log files now escape non-printable characters consistently as “xXX’. Furthermore, backslashes are escaped as “", making the representation fully reversible.
The split* family of functions are to be replaced with alternate versions that return a vector of strings rather than a table of strings. This also allows deprecation for some related string concatenation/extraction functions. Note that the new functions use 0-based indexing, rather than 1-based.
The full list of now deprecated functions is:
- split: use split_string instead.
- split1: use split_string1 instead.
- split_all: use split_string_all instead.
- split_n: use split_string_n instead.
- cat_string_array: see join_string_vec instead.
- cat_string_array_n: see join_string_vec instead.
- join_string_array: see join_string_vec instead.
- sort_string_array: use sort instead.
- find_ip_addresses: use extract_ip_addresses instead.
string slices now exclude the end index (e.g., “123”[1:2] returns “2”). Generally, Bro’s string slices now behave similar to Python.
ssl_client_hello() now receives a vector of ciphers, instead of a set, to preserve their order.
Notice::end_suppression() has been removed.
Bro now parses X.509 extensions headers and, as a result, the corresponding event got a new signature:
event x509_extension(c: connection, is_orig: bool, cert: X509, ext: X509_extension_info);
In addition, there are several new, more specialized events for a number of x509 extensions.
Generally, all x509 events and handling functions have changed their signatures.
X509 certificate verification now returns the complete certificate chain that was used for verification.
Bro no longer special-cases SYN/FIN/RST-filtered traces by not reporting missing data. Instead, if Bro never sees any data segments for analyzed TCP connections, the new base/misc/find-filtered-trace.bro script will log a warning in reporter.log and to stderr. The old behavior can be reverted by redef’ing “detect_filtered_trace”.
We have removed the packet sorter component.
Bro no longer uses libmagic to identify file types but instead now comes with its own signature library (which initially is still derived from libmagic’s database). This leads to a number of further changes with regards to MIME types:
- The second parameter of the “identify_data” built-in function can no longer be used to get verbose file type descriptions, though it can still be used to get the strongest matching file magic signature.
- The “file_transferred” event’s “descr” parameter no longer contains verbose file type descriptions.
- The BROMAGIC environment variable no longer changes any behavior in Bro as magic databases are no longer used/installed.
- Removed “binary” and “octet-stream” mime type detections. They don’t provide any more information than an uninitialized mime_type field.
- The “fa_file” record now contains a “mime_types” field that contains all magic signatures that matched the file content (where the “mime_type” field is just a shortcut for the strongest match).
dns_TXT_reply() now supports more than one string entry by receiving a vector of strings.
BroControl now runs the “exec” and “df” broctl commands only once per host, instead of once per Bro node. The output of these commands has been changed slightly to include both the host and node names.
Several performance improvements were made. Particular emphasis was put on the File Analysis system, which generally will now emit far fewer file handle request events due to protocol analyzers now caching that information internally.
A completely overhauled intelligence framework for consuming
external intelligence data. It provides an abstracted mechanism
for feeding data into the framework to be matched against the
data available. It also provides a function named Intel::match
which makes any hits on intelligence data available to the
scripting language.
Using input framework, the intel framework can load data from
text files. It can also update and add data if changes are
made to the file being monitored. Files to monitor for
intelligence can be provided by redef-ing the
Intel::read_files
variable.
The intel framework is cluster-ready. On a cluster, the manager is the only node that needs to load in data from disk, the cluster support will distribute the data across a cluster automatically.
Scripts are provided at policy/frameworks/intel/seen
that
provide a broad set of sources of data to feed into the intel
framwork to be matched.
A new file analysis framework moves most of the processing of file
content from script-land into the core, where it belongs. See
doc/file-analysis.rst
, or the online documentation, for more
information.
Much of this is an internal change, but the framework also comes with the following user-visible functionality (some of that was already available before but is done differently, and more efficiently, now):
- HTTP:
- Identify MIME type of messages.
- Extract messages to disk.
- Compute MD5 for messages.
- SMTP:
- Identify MIME type of messages.
- Extract messages to disk.
- Compute MD5 for messages.
- Provide access to start of entity data.
- FTP data transfers:
- Identify MIME types of data.
- Record to disk.
- IRC DCC transfers: Record to disk.
- Support for analyzing data transferred via HTTP range requests.
- A binary input reader interfaces the input framework with the file analysis, allowing to inject files on disk into Bro’s content processing.
A new framework for computing a wide array of summary statistics,
such as counters and thresholds checks, standard deviation and mean,
set cardinality, top K, and more. The framework operates in
real-time, independent of the underlying data, and can aggregate
information from many independent monitoring points (including
clusters). It provides a transparent, easy-to-use user interface,
and can optionally deploy a set of probabilistic data structures for
memory-efficient operation. The framework is located in
scripts/base/frameworks/sumstats
.
A number of new applications now ship with Bro that are built on top of the summary statistics framework:
- Scan detection: Detectors for port and address scans. See
policy/misc/scan.bro
(these scan detectors used to exist in Bro versions <2.0; it’s now back, but quite different).- Tracerouter detector:
policy/misc/detect-traceroute.bro
- Web application detection/measurement:
policy/misc/app-stats/*
- FTP and SSH brute-forcing detector:
policy/protocols/ftp/detect-bruteforcing.bro
,policy/protocols/ssh/detect-bruteforcing.bro
- HTTP-based SQL injection detector:
policy/protocols/http/detect-sqli.bro
(existed before, but now ported to the new framework)
GridFTP support. This is an extension to the standard FTP analyzer and includes:
- An analyzer for the GSI mechanism of GSSAPI FTP AUTH method. GSI authentication involves an encoded TLS/SSL handshake over the FTP control session. For FTP sessions that attempt GSI authentication, the
service
field of the connection log will includegridftp
(as well as alsoftp
andssl
).- An example of a GridFTP data channel detection script. It relies on the heuristics of GridFTP data channels commonly default to SSL mutual authentication with a NULL bulk cipher and that they usually transfer large datasets (default threshold of script is 1 GB). For identified GridFTP data channels, the
services
fields of the connection log will includegridftp-data
.
Modbus and DNP3 support. Script-level support is only basic at this
point but see src/analyzer/protocol/{modbus,dnp3}/events.bif
, or
the online documentation, for the events Bro generates. For Modbus,
there are also some example policies in
policy/protocols/modbus/*
.
The documentation now includes a new introduction to writing Bro
scripts. See doc/scripting/index.rst
or, much better, the online
version. There’s also the beginning of a chapter on “Using Bro” in
doc/using/index.rst
.
GPRS Tunnelling Protocol (GTPv1) decapsulation.
The scripting language now provide “hooks”, a new flavor of
functions that share characteristics of both standard functions and
events. They are like events in that multiple bodies can be defined
for the same hook identifier. They are more like functions in the
way they are invoked/called, because, unlike events, their execution
is immediate and they do not get scheduled through an event queue.
Also, a unique feature of a hook is that a given hook handler body
can short-circuit the execution of remaining hook handlers simply by
exiting from the body as a result of a break
statement (as
opposed to a return
or just reaching the end of the body). See
doc/scripts/builtins.rst
, or the online documentation, for more
informatin.
Bro’s language now has a working switch
statement that generally
behaves like C-style switches (except that case labels can be
comprised of multiple literal constants delimited by commas). Only
atomic types are allowed for now. Case label bodies that don’t
execute a return
or break
statement will fall through to
subsequent cases. A default
case label is supported.
Bro’s language now has a new set of types opaque of X
. Opaque
values can be passed around like other values but they can only be
manipulated with BiF functions, not with other operators. Currently,
the following opaque types are supported:
opaque of md5
opaque of sha1
opaque of sha256
opaque of cardinality
opaque of topk
opaque of bloomfilter
These go along with the corrsponding BiF functions md5_*
,
sha1_*
, sha256_*
, entropy_*
, etc. . Note that where
these functions existed before, they have changed their signatures
to work with opaques types rather than global state.
The scripting language now supports constructing sets, tables, vectors, and records by name:
type MyRecordType: record {
c: count;
s: string &optional;
};
global r: MyRecordType = record($c = 7);
type MySet: set[MyRec];
global s = MySet([$c=1], [$c=2]);
Strings now support the subscript operator to extract individual
characters and substrings (e.g., s[4]
, s[1:5]
). The index
expression can take up to two indices for the start and end index of
the substring to return (e.g. mystring[1:3]
).
Functions now support default parameters, e.g.:
global foo: function(s: string, t: string &default="abc", u: count &default=0);
Scripts can now use two new “magic constants” @DIR
and
@FILENAME
that expand to the directory path of the current
script and just the script file name without path, respectively.
ssl.log
now also records the subject client and issuer
certificates.
The ASCII writer can now output CSV files on a per filter basis.
New SQLite reader and writer plugins for the logging framework allow to read/write persistent data from on disk SQLite databases.
A new packet filter framework supports BPF-based load-balancing, shunting, and sampling; plus plugin support to customize filters dynamically.
Bro now provides Bloom filters of two kinds: basic Bloom filters supporting membership tests, and counting Bloom filters that track the frequency of elements. The corresponding functions are:
bloomfilter_basic_init(fp: double, capacity: count, name: string &default=""): opaque of bloomfilter
bloomfilter_basic_init2(k: count, cells: count, name: string &default=""): opaque of bloomfilter
bloomfilter_counting_init(k: count, cells: count, max: count, name: string &default=""): opaque of bloomfilter
bloomfilter_add(bf: opaque of bloomfilter, x: any)
bloomfilter_lookup(bf: opaque of bloomfilter, x: any): count
bloomfilter_merge(bf1: opaque of bloomfilter, bf2: opaque of bloomfilter): opaque of bloomfilter
bloomfilter_clear(bf: opaque of bloomfilter)
See src/probabilistic/bloom-filter.bif
, or the online
documentation, for full documentation.
Bro now provides a probabilistic data structure for computing “top k” elements. The corresponding functions are:
topk_init(size: count): opaque of topk
topk_add(handle: opaque of topk, value: any)
topk_get_top(handle: opaque of topk, k: count)
topk_count(handle: opaque of topk, value: any): count
topk_epsilon(handle: opaque of topk, value: any): count
topk_size(handle: opaque of topk): count
topk_sum(handle: opaque of topk): count
topk_merge(handle1: opaque of topk, handle2: opaque of topk)
topk_merge_prune(handle1: opaque of topk, handle2: opaque of topk)
See src/probabilistic/top-k.bif
, or the online documentation,
for full documentation.
Bro now provides a probabilistic data structure for computing set cardinality, using the HyperLogLog algorithm. The corresponding functions are:
hll_cardinality_init(err: double, confidence: double): opaque of cardinality
hll_cardinality_add(handle: opaque of cardinality, elem: any): bool
hll_cardinality_merge_into(handle1: opaque of cardinality, handle2: opaque of cardinality): bool
hll_cardinality_estimate(handle: opaque of cardinality): double
hll_cardinality_copy(handle: opaque of cardinality): opaque of cardinality
See src/probabilistic/cardinality-counter.bif
, or the online
documentation, for full documentation.
base/utils/exec.bro
provides a module to start external
processes asynchronously and retrieve their output on termination.
base/utils/dir.bro
uses it to monitor a directory for changes,
and base/utils/active-http.bro
for providing an interface for
querying remote web servers.
BroControl can now pin Bro processes to CPUs on supported platforms:
To use CPU pinning, a new per-node option pin_cpus
can be
specified in node.cfg if the OS is either Linux or FreeBSD.
BroControl now returns useful exit codes. Most BroControl commands return 0 if everything was OK, and 1 otherwise. However, there are a few exceptions. The “status” and “top” commands return 0 if all Bro nodes are running, and 1 if not all nodes are running. The “cron” command always returns 0 (but it still sends email if there were any problems). Any command provided by a plugin always returns 0.
BroControl now has an option “env_vars” to set Bro environment variables. The value of this option is a comma-separated list of environment variable assignments (e.g., “VAR1=value, VAR2=another”). The “env_vars” option can apply to all Bro nodes (by setting it in broctl.cfg), or can be node-specific (by setting it in node.cfg). Environment variables in node.cfg have priority over any specified in broctl.cfg.
BroControl now supports load balancing with PF_RING while sniffing multiple interfaces. Rather than assigning the same PF_RING cluster ID to all workers on a host, cluster ID assignment is now based on which interface a worker is sniffing (i.e., all workers on a host that sniff the same interface will share a cluster ID). This is handled by BroControl automatically.
BroControl has several new options: MailConnectionSummary (for disabling the sending of connection summary report emails), MailAlarmsInterval (for specifying a different interval to send alarm summary emails), CompressCmd (if archived log files will be compressed, this specifies the command that will be used to compress them), CompressExtension (if archived log files will be compressed, this specifies the file extension to use).
BroControl comes with its own test-suite now. make test
in
aux/broctl
will run it.
In addition to these, Bro 2.2 comes with a large set of smaller extensions, tweaks, and fixes across the whole code base, including most submodules.
Previous versions of $prefix/share/bro/site/local.bro
(where
“$prefix” indicates the installation prefix of Bro), aren’t compatible
with Bro 2.2. This file won’t be overwritten when installing over a
previous Bro installation to prevent clobbering users’ modifications,
but an example of the new version is located in
$prefix/share/bro/site/local.bro.example
. So if no modification
has been done to the previous local.bro, just copy the new example
version over it, else merge in the differences. For reference,
a common error message when attempting to use an outdated local.bro
looks like:
fatal error in /usr/local/bro/share/bro/policy/frameworks/software/vulnerable.bro, line 41: BroType::AsRecordType (table/record) (set[record { min:record { major:count; minor:count; minor2:count; minor3:count; addl:string; }; max:record { major:count; minor:count; minor2:count; minor3:count; addl:string; }; }])
The type of Software::vulnerable_versions
changed to allow
more flexibility and range specifications. An example usage:
const java_1_6_vuln = Software::VulnerableVersionRange( $max = Software::Version($major = 1, $minor = 6, $minor2 = 0, $minor3 = 44) ); const java_1_7_vuln = Software::VulnerableVersionRange( $min = Software::Version($major = 1, $minor = 7), $max = Software::Version($major = 1, $minor = 7, $minor2 = 0, $minor3 = 20) ); redef Software::vulnerable_versions += { ["Java"] = set(java_1_6_vuln, java_1_7_vuln) };
The interface to extracting content from application-layer protocols (including HTTP, SMTP, FTP) has changed significantly due to the introduction of the new file analysis framework (see above).
Removed the following, already deprecated, functionality:
- Scripting language:
&disable_print_hook attribute
.
- BiF functions:
parse_dotted_addr()
,dump_config()
,make_connection_persistent()
,generate_idmef()
,split_complete()
md5_*
,sha1_*
,sha256_*
, andentropy_*
have all changed their signatures to work with opaque types (see above).
Removed a now unused argument from do_split
helper function.
this
is no longer a reserved keyword.
The Input Framework’s update_finished
event has been renamed to
end_of_data
. It will now not only fire after table-reads have
been completed, but also after the last event of a whole-file-read
(or whole-db-read, etc.).
Renamed the option defining the frequency of alarm summary mails to
Logging::default_alarm_mail_interval
. When using BroControl, the
value can now be set with the new broctl.cfg option
MailAlarmsInterval
.
We have completely rewritten the notice_policy
mechanism. It now
no longer uses a record of policy items but a hook
, a new
language element that’s roughly equivalent to a function with
multiple bodies (see above). For existing code, the two main changes
are:
What used to be a
redef
ofNotice::policy
now becomes a hook implementation. Example:Old:
redef Notice::policy += { [$pred(n: Notice::Info) = { return n$note == SSH::Login && n$id$resp_h == 10.0.0.1; }, $action = Notice::ACTION_EMAIL] };New:
hook Notice::policy(n: Notice::Info) { if ( n$note == SSH::Login && n$id$resp_h == 10.0.0.1 ) add n$actions[Notice::ACTION_EMAIL]; }notice() is now likewise a hook, no longer an event. If you have handlers for that event, you’ll likely just need to change the type accordingly. Example:
Old:
event notice(n: Notice::Info) { ... }New:
hook notice(n: Notice::Info) { ... }
The notice_policy.log
is gone. That’s a result of the new notice
policy setup.
Removed the byte_len()
and length()
bif functions. Use the
|...|
operator instead.
The SSH::Login
notice has been superseded by an corresponding
intelligence framework observation (SSH::SUCCESSFUL_LOGIN
).
PacketFilter::all_packets
has been replaced with
PacketFilter::enable_auto_protocol_capture_filters
.
We removed the BitTorrent DPD signatures pending further updates to that analyzer.
In previous versions of BroControl, running “broctl cron” would create
a file $prefix/logs/stats/www
(where “$prefix” indicates the
installation prefix of Bro). Now, it is created as a directory.
Therefore, if you perform an upgrade install and you’re using BroControl,
then you may see an email (generated by “broctl cron”) containing an
error message: “error running update-stats”. To fix this problem,
either remove that file (it is not needed) or rename it.
Due to lack of maintenance the Ruby bindings for Broccoli are now
deprecated, and the build process no longer includes them by
default. For the time being, they can still be enabled by
configuring with --enable-ruby
, however we plan to remove
Broccoli’s Ruby support with the next Bro release.
Bro now comes with extensive IPv6 support. Past versions offered only basic IPv6 functionality that was rarely used in practice as it had to be enabled explicitly. IPv6 support is now fully integrated into all parts of Bro including protocol analysis and the scripting language. It’s on by default and no longer requires any special configuration.
Some of the most significant enhancements include support for IPv6 fragment reassembly, support for following IPv6 extension header chains, and support for tunnel decapsulation (6to4 and Teredo). The DNS analyzer now handles AAAA records properly, and DNS lookups that Bro itself performs now include AAAA queries, so that, for example, the result returned by script-level lookups is a set that can contain both IPv4 and IPv6 addresses. Support for the most common ICMPv6 message types has been added. Also, the FTP EPSV and EPRT commands are now handled properly. Internally, the way IP addresses are stored has been improved, so Bro can handle both IPv4 and IPv6 by default without any special configuration.
In addition to Bro itself, the other Bro components have also been made IPv6-aware by default. In particular, significant changes were made to trace-summary, PySubnetTree, and Broccoli to support IPv6.
Bro now decapsulates tunnels via its new tunnel framework located in scripts/base/frameworks/tunnels. It currently supports Teredo, AYIYA, IP-in-IP (both IPv4 and IPv6), and SOCKS. For all these, it logs the outer tunnel connections in both conn.log and tunnel.log, and then proceeds to analyze the inner payload as if it were not tunneled, including also logging that session in conn.log. For SOCKS, it generates a new socks.log in addition with more information.
Bro now features a flexible input framework that allows users to integrate external information in real-time into Bro while it’s processing network traffic. The most direct use-case at the moment is reading data from ASCII files into Bro tables, with updates picked up automatically when the file changes during runtime. See doc/input.rst for more information.
Internally, the input framework is structured around the notion of “reader plugins” that make it easy to interface to different data sources. We will add more in the future.
BroControl now has built-in support for host-based load-balancing when using either PF_RING, Myricom cards, or individual interfaces. Instead of adding a separate worker entry in node.cfg for each Bro worker process on each worker host, it is now possible to just specify the number of worker processes on each host and BroControl configures everything correctly (including any neccessary enviroment variables for the balancers).
This change adds three new keywords to the node.cfg file (to be used with worker entries): lb_procs (specifies number of workers on a host), lb_method (specifies what type of load balancing to use: pf_ring, myricom, or interfaces), and lb_interfaces (used only with “lb_method=interfaces” to specify which interfaces to load-balance on).
Bro’s default ASCII log format is not exactly the most efficient way for storing and searching large volumes of data. An alternatives, Bro now comes with experimental support for two alternative output formats:
- DataSeries: an efficient binary format for recording structured bulk data. DataSeries is developed and maintained at HP Labs. See doc/logging-dataseries for more information.
- ElasticSearch: a distributed RESTful, storage engine and search engine built on top of Apache Lucene. It scales very well, both for distributed indexing and distributed searching. See doc/logging-elasticsearch.rst for more information.
Note that at this point, we consider Bro’s support for these two formats as prototypes for collecting experience with alternative outputs. We do not yet recommend them for production (but welcome feedback!)
The following summarizes the most important differences in existing functionality. Note that this list is not complete, see CHANGES for the full set.
Changes in dependencies:
Bro now requires CMake >= 2.6.3.
On Linux, Bro now links in tcmalloc (part of Google perftools) if found at configure time. Doing so can significantly improve memory and CPU use.
On the other platforms, the new configure option –enable-perftools can be used to enable linking to tcmalloc. (Note that perftools’s support for non-Linux platforms may be less reliable).
The configure switch –enable-brov6 is gone.
DNS name lookups performed by Bro now also query AAAA records. The results of the A and AAAA queries for a given hostname are combined such that at the scripting layer, the name resolution can yield a set with both IPv4 and IPv6 addresses.
The connection compressor was already deprecated in 2.0 and has now been removed from the code base.
We removed the “match” statement, which was no longer used by any of the default scripts, nor was it likely to be used by anybody anytime soon. With that, “match” and “using” are no longer reserved keywords.
The syntax for IPv6 literals changed from “2607:f8b0:4009:802::1012”
to “[2607:f8b0:4009:802::1012]”. When an IP address variable or IP
address literal is enclosed in pipes (for example,
|[fe80::db15]|
) the result is now the size of the address in
bits (32 for IPv4 and 128 for IPv6).
Bro now spawns threads for doing its logging. From a user’s perspective not much should change, except that the OS may now show a bunch of Bro threads.
We renamed the configure option –enable-perftools to –enable-perftools-debug to indicate that the switch is only relevant for debugging the heap.
Bro’s ICMP analyzer now handles both IPv4 and IPv6 messages with a joint set of events. The icmp_conn record got a new boolean field ‘v6’ that indicates whether the ICMP message is v4 or v6.
Log postprocessor scripts get an additional argument indicating the type of the log writer in use (e.g., “ascii”).
BroControl’s make-archive-name script also receives the writer type, but as its 2nd(!) argument. If you’re using a custom version of that script, you need to adapt it. See the shipped version for details.
Signature files can now be loaded via the new “@load-sigs” directive. In contrast to the existing (and still supported) signature_files constant, this can be used to load signatures relative to the current script (e.g., “@load-sigs ./foo.sig”).
The options “tunnel_port” and “parse_udp_tunnels” have been removed. Bro now supports decapsulating tunnels directly for protocols it understands.
ASCII logs now record the time when they were opened/closed at the beginning and end of the file, respectively (wall clock). The options LogAscii::header_prefix and LogAscii::include_header have been renamed to LogAscii::meta_prefix and LogAscii::include_meta, respectively.
The ASCII writers “header_*” options have been renamed to “meta_*” (because there’s now also a footer).
Some built-in functions have been removed: “addr_to_count” (use “addr_to_counts” instead), “bro_has_ipv6” (this is no longer relevant because Bro now always supports IPv6), “active_connection” (use “connection_exists” instead), and “connection_record” (use “lookup_connection” instead).
The “NFS3::mode2string” built-in function has been renamed to “file_mode”.
Some built-in functions have been changed: “exit” (now takes the exit code as a parameter), “to_port” (now takes a string as parameter instead of a count and transport protocol, but “count_to_port” is still available), “connect” (now takes an additional string parameter specifying the zone of a non-global IPv6 address), and “listen” (now takes three additional parameters to enable listening on IPv6 addresses).
Some Bro script variables have been renamed: “LogAscii::header_prefix” has been renamed to “LogAscii::meta_prefix”, “LogAscii::include_header” has been renamed to “LogAscii::include_meta”.
Some Bro script variables have been removed: “tunnel_port”, “parse_udp_tunnels”, “use_connection_compressor”, “cc_handle_resets”, “cc_handle_only_syns”, and “cc_instantiate_on_data”.
A couple events have changed: the “icmp_redirect” event now includes the target and destination addresses and any Neighbor Discovery options in the message, and the last parameter of the “dns_AAAA_reply” event has been removed because it was unused.
The format of the ASCII log files has changed very slightly. Two new lines are automatically added, one to record the time when the log was opened, and the other to record the time when the log was closed.
In BroControl, the option (in broctl.cfg) “CFlowAddr” was renamed to “CFlowAddress”.
As the version number jump from 1.5 suggests, Bro 2.0 is a major upgrade and lots of things have changed. Most importantly, we have rewritten almost all of Bro’s default scripts from scratch, using quite different structure now and focusing more on operational deployment. The result is a system that works much better “out of the box”, even without much initial site-specific configuration. The down-side is that 1.x configurations will need to be adapted to work with the new version. The two rules of thumb are:
- If you have written your own Bro scripts that do not depend on any of the standard scripts formerly found in
policy/
, they will most likely just keep working (although you might want to adapt them to use some of the new features, like the new logging framework; see below).- If you have custom code that depends on specifics of 1.x default scripts (including most configuration tuning), that is unlikely to work with 2.x. We recommend to start by using just the new scripts first, and then port over any customizations incrementally as necessary (they may be much easier to do now, or even unnecessary). Send mail to the Bro user mailing list if you need help.
Below we summarize changes from 1.x to 2.x in more detail. This list
isn’t complete, see the CHANGES
file in the distribution.
for the full story.
In versions before 2.0, Bro scripts were all maintained in a flat
directory called policy/
in the source tree. This directory is now
renamed to scripts/
and contains major subdirectories base/
,
policy/
, and site/
, each of which may also be subdivided
further.
The contents of the new scripts/
directory, like the old/flat
policy/
still gets installed under the share/bro
subdirectory of the installation prefix path just like previous
versions. For example, if Bro was compiled like ./configure
--prefix=/usr/local/bro && make && make install
, then the script
hierarchy can be found in /usr/local/bro/share/bro
.
The main subdirectories of that hierarchy are as follows:
base/
contains all scripts that are loaded by Bro by default
(unless the -b
command line option is used to run Bro in a
minimal configuration). Note that is a major conceptual change:
rather than not loading anything by default, Bro now uses an
extensive set of default scripts out of the box.
The scripts under this directory generally either accumulate/log useful state/protocol information for monitored traffic, configure a default/recommended mode of operation, or provide extra Bro scripting-layer functionality that has no significant performance cost.
policy/
contains all scripts that a user will need to explicitly
tell Bro to load. These are scripts that implement
functionality/analysis that not all users may want to use and may have
more significant performance costs. For a new installation, you
should go through these and see what appears useful to load.
site/
remains a directory that can be used to store locally
developed scripts. It now comes with some preinstalled example
scripts that contain recommended default configurations going beyond
the base/
setup. E.g. local.bro
loads extra scripts from
policy/
and does extra tuning. These files can be customized in
place without being overwritten by upgrades/reinstalls, unlike
scripts in other directories.
With version 2.0, the default BROPATH
is set to automatically
search for scripts in policy/
, site/
and their parent
directory, but not base/
. Generally, everything under
base/
is loaded automatically, but for users of the -b
option,
it’s important to know that loading a script in that directory
requires the extra base/
path qualification. For example, the
following two scripts:
$PREFIX/share/bro/base/protocols/ssl/main.bro
$PREFIX/share/bro/policy/protocols/ssl/validate-certs.bro
are referenced from another Bro script like:
@load base/protocols/ssl/main @load protocols/ssl/validate-certs
Notice how policy/
can be omitted as a convenience in the second
case. @load
can now also use relative path, e.g., @load
../main
.
ssl.log
.conn.log
service
field is derived from DPD instead of
well-known ports (while that was already possible in 1.5, it was
not the default).conn.log
now reports raw number of packets/bytes per
endpoint.connection
records so that
state can be collected until enough is seen to log a coherent unit
of information regarding the activity of that connection. This
state is now frequently seen/accessible in event handlers, for
example, like c$<protocol>
where <protocol>
is replaced by
the name of the protocol. This field is added to the connection
record by redef
’ing it in a
base/protocols/<protocol>/main.bro
script.The way users interact with “notices” has changed significantly in order to make it easier to define a site policy and more extensible for adding customized actions.
PacketFilter::all_packets
for how to revert to old
behavior.@prefixes
directive works differently now.
Any added prefixes are now searched for and loaded after all input
files have been parsed. After all input files are parsed, Bro
searches BROPATH
for prefixed, flattened versions of all of the
parsed input files. For example, if lcl
is in @prefixes
, and
site.bro
is loaded, then a file named lcl.site.bro
that’s in
BROPATH
would end up being automatically loaded as well. Packages
work similarly, e.g. loading protocols/http
means a file named
lcl.protocols.http.bro
in BROPATH
gets loaded automatically.make_addr
BIF now returns a subnet
versus an addr
Module
is more widely used for namespacing. E.g. the new
site.bro
exports the local_nets
identifier (among other
things) into the Site
module.We have remove a bunch of functionality that was rarely used and/or had not been maintained for a while already:
- The
net
script data type.- The
alarm
statement; use the notice framework instead.- Trace rewriting.
- DFA state expiration in regexp engine.
- Active mapping.
- Native DAG support (may come back eventually)
- ClamAV support.
- The connection compressor is now disabled by default, and will be removed in the future.
BroControl looks pretty much similar to the version coming with Bro 1.x, but has been cleaned up and streamlined significantly internally.
BroControl has a new process
command to process a trace on disk
offline using a similar configuration to what BroControl installs for
live analysis.
BroControl now has an extensive plugin interface for adding new commands and options. Note that this is still considered experimental.
We have removed the analysis
command, and BroControl currently
does not send daily alarm summaries anymore (this may be restored
later).
Bro development has moved from using SVN to Git for revision control. Users that want to use the latest Bro development snapshot by checking it out from the source repositories should see the development process. Note that all the various sub-components now reside in their own repositories. However, the top-level Bro repository includes them as git submodules so it’s easy to check them all out simultaneously.
Bro now uses CMake for its build system so that is a new required dependency when building from source.
Bro now comes with a growing suite of regression tests in
testing/
.