Detection: HTTP Base64 Activity With JVM Errors

Description

The analytic associates each Zeek request containing a malformed Base64 value or a scripted decoder trigger with a nearby Base64-related HotSpot argument exception on the same server and Java PID. Sysmon child process, file creation, or process termination events from that PID add context and can raise confidence.

  1(
  2    `zeek_http_request_headers`
  3    header_name IN (
  4        "AUTHORIZATION",
  5        "HTTP2-SETTINGS",
  6        "CONTENT-TYPE",
  7        "CONTENT-DISPOSITION",
  8        "PRIORITY",
  9        "X-CONSOLEANNOTATOR",
 10        "X-CONNECT-AUTHORIZATION",
 11        "X-CSRF-TOKEN"
 12    )
 13    header_value="*=*"
 14    truncated=false
 15)
 16OR
 17(
 18    `zeek_http_post_body`
 19    body="*=*"
 20    truncated=false
 21)
 22OR
 23(
 24    `zeek_http_uri`
 25    uri IN ("*=*","*%3D*","*%3d*")
 26)
 27OR
 28(
 29    `java_hotspot_exceptions`
 30    (
 31        "*Input byte array has wrong*"
 32        OR
 33        "*IllegalArgumentException*"
 34        OR
 35        "*Base64$Decoder*"
 36        OR
 37        "*java/util/Base64*"
 38    )
 39)
 40OR
 41(
 42    `sysmon_linux`
 43    (
 44        (EventID=1 ParentImage="*/java")
 45        OR
 46        (EventID=5 Image="*/java")
 47        OR
 48        (
 49            EventID=11
 50            Image="*/java"
 51            NOT TargetFilename IN ("*/_state/*","*/indices/*")
 52        )
 53    )
 54)
 55
 56
 57| rex field=_raw "^\[(?<jvm_event_time_raw>[^\]]+)\]\[(?<jvm_pid_raw>\d+)\]\[(?<jvm_level_raw>[^\]]+)\]\[(?<jvm_tags_raw>[^\]]+)\]\s*(?<jvm_message_raw>.*)"
 58
 59
 60| eval jvm_pid=coalesce(jvm_pid_raw,jvm_pid),
 61       jvm_tags=coalesce(jvm_tags_raw,jvm_tags),
 62       jvm_message=coalesce(jvm_message_raw,jvm_message)
 63
 64
 65
 66| eval event_time=_time,
 67       is_network=if(isnotnull('id.resp_h'),1,0)
 68
 69
 70| eval observed_target=if(is_network=1,'id.resp_h',coalesce(dest,host,dest_ip))
 71
 72
 73| eval request_value=case(
 74      isnotnull(header_name),header_value,
 75      isnotnull(body),body,
 76      is_network=1,urldecode(uri)
 77  ),
 78  carrier=case(
 79      isnotnull(header_name),header_name,
 80      isnotnull(body),"POST body",
 81      is_network=1,"URI"
 82  )
 83
 84
 85| where is_network=0 OR len(request_value)>=65
 86
 87
 88| eval malformed_b64=if(
 89      is_network=1
 90      AND match(request_value,"(?<![A-Za-z0-9+/_-])(?:(?:[A-Za-z0-9+/]{64})+(?:[A-Za-z0-9+/]{0,2}=
 91|[A-Za-z0-9+/]?==)
 92|(?:[A-Za-z0-9_-]{64})+(?:[A-Za-z0-9_-]{0,2}=
 93|[A-Za-z0-9_-]?==))(?![A-Za-z0-9+/_=-])"),
 94      1,0
 95  ),
 96  scripted_b64=if(
 97      is_network=1 AND isnotnull(body)
 98      AND like(body,"%Base64.getDecoder()%")
 99      AND like(body,"%new byte[65]%")
100      AND like(body,"%[64] = (byte)61%"),
101      1,0
102  )
103
104
105| eval event_kind=case(
106      is_network=1 AND (malformed_b64=1 OR scripted_b64=1),"request",
107      isnotnull(jvm_pid) AND like(jvm_tags,"%exceptions%"),"jvm",
108      EventID=1 OR EventID=5 OR EventID=11,"sysmon"
109  )
110
111
112| eval java_pid=case(
113      event_kind="request","network",
114      event_kind="jvm",tostring(jvm_pid),
115      EventID=1,tostring(ParentProcessId),
116      EventID=5 OR EventID=11,tostring(ProcessId)
117  ),
118  request_uid=if(event_kind="request",uid,"-")
119
120
121| where isnotnull(event_kind)
122        AND isnotnull(observed_target)
123        AND isnotnull(java_pid)
124        AND isnotnull(request_uid)
125
126
127| eval argument_error=if(event_kind="jvm"
128      AND match(jvm_message,"(?i)IllegalArgumentException
129|Input byte array has wrong"),1,0),
130  decoder_evidence=if(event_kind="jvm"
131      AND match(jvm_message,"Base64[$]Decoder
132|java/util/Base64
133|Input byte array has wrong"),1,0),
134  shell_child=if(EventID=1
135      AND match(Image,"/(ash
136|bash
137|csh
138|dash
139|eshell
140|fish
141|ion
142|ksh
143|mksh
144|rbash
145|sh
146|tcsh
147|zsh)$"),1,0),
148  child_process=if(EventID=1,1,0),
149  file_created=if(EventID=11,1,0),
150  java_terminated=if(EventID=5,1,0),
151  suspicious_file=if(EventID=11
152      AND match(TargetFilename,"(?i)(?:[.](?:jsp
153|sh
154|so)$
155|/(?:tmp
156|var/tmp
157|dev/shm
158|webapps)/)"),1,0)
159
160
161| eval request_marker=case(
162      event_kind="request" AND scripted_b64=1,"scripted_decoder_trigger",
163      event_kind="request" AND malformed_b64=1,"malformed_base64_request"
164  ),
165  src_ip=if(event_kind="request",'id.orig_h',null()),
166  request_uri=if(event_kind="request",uri,null()),
167  http_method=if(event_kind="request",method,null()),
168  child_image=if(EventID=1,Image,null()),
169  created_file=if(EventID=11,TargetFilename,null())
170
171
172| bin _time span=1s
173
174
175| stats min(event_time) AS first_seen
176        max(event_time) AS last_seen
177        max(argument_error) AS argument_error
178        max(decoder_evidence) AS decoder_evidence
179        max(shell_child) AS shell_child
180        max(child_process) AS child_process
181        max(file_created) AS file_created
182        max(java_terminated) AS java_terminated
183        max(suspicious_file) AS suspicious_file
184        values(src_ip) AS src_ips
185        values(request_uri) AS uris
186        values(http_method) AS methods
187        values(carrier) AS carriers
188        values(request_marker) AS request_markers
189        values(child_image) AS child_images
190        values(created_file) AS created_files
191        BY _time observed_target java_pid request_uid event_kind
192
193
194| rename observed_target AS dest
195
196
197| `get_asset(dest)`
198
199
200| eval target=coalesce(dest_asset_id,dest),
201       bucket=floor(first_seen/15)
202
203
204| eval request_json=if(
205      event_kind="request",
206      json_object(
207          "uid",request_uid,
208          "time",first_seen,
209          "src",mvindex(src_ips,0),
210          "dest",dest,
211          "uri",mvindex(uris,0),
212          "method",mvindex(methods,0),
213          "carrier",mvjoin(carriers,","),
214          "marker",mvjoin(request_markers,",")
215      ),
216      null()
217  ),
218  bucket=if(
219      event_kind="request",
220      mvappend(tostring(bucket-1),tostring(bucket),tostring(bucket+1),tostring(bucket+2)),
221      bucket
222  )
223
224
225| mvexpand bucket
226
227
228| eventstats values(request_json) AS candidate_requests BY target bucket
229
230
231| where event_kind!="request" AND isnotnull(candidate_requests)
232
233
234| mvexpand candidate_requests
235
236
237| eval request_uid=spath(candidate_requests,"uid"),
238       request_time=tonumber(spath(candidate_requests,"time")),
239       src=spath(candidate_requests,"src"),
240       request_dest=spath(candidate_requests,"dest"),
241       uri=spath(candidate_requests,"uri"),
242       method=spath(candidate_requests,"method"),
243       carrier=spath(candidate_requests,"carrier"),
244       request_marker=spath(candidate_requests,"marker")
245
246
247| where (event_kind="jvm"
248         AND last_seen>=request_time-3
249         AND first_seen<=request_time+15)
250        OR
251        (event_kind="sysmon"
252         AND last_seen>=request_time-3
253         AND first_seen<=request_time+30)
254
255
256| stats min(request_time) AS request_time
257        min(eval(if(argument_error=1,first_seen,null()))) AS exception_time
258        max(last_seen) AS last_seen
259        max(argument_error) AS argument_error
260        max(decoder_evidence) AS decoder_evidence
261        max(shell_child) AS shell_child
262        max(child_process) AS child_process
263        max(file_created) AS file_created
264        max(java_terminated) AS java_terminated
265        max(suspicious_file) AS suspicious_file
266        values(src) AS src
267        values(request_dest) AS dest_ips
268        values(uri) AS uri
269        values(method) AS method
270        values(carrier) AS carrier
271        values(request_marker) AS request_markers
272        values(child_images) AS child_images
273        values(created_files) AS created_files
274        BY target request_uid java_pid
275
276
277| where argument_error=1 AND decoder_evidence=1
278
279
280| eval dest=mvindex(dest_ips,0),
281       exception_delay=round(exception_time-request_time,3),
282       confidence=case(
283           shell_child>0 OR suspicious_file>0,"high",
284           child_process>0 OR file_created>0 OR java_terminated>0,"medium",
285           true(),"low"
286       ),
287       asset_id=target,
288       firstTime=request_time,
289       lastTime=last_seen
290
291
292| fields - argument_error decoder_evidence shell_child child_process file_created java_terminated suspicious_file dest_ips
293
294
295| sort 0 - firstTime
296
297
298| `security_content_ctime(firstTime)`
299
300| `security_content_ctime(lastTime)`
301
302| `http_base64_activity_with_jvm_errors_filter`

Data Source

Name Platform Sourcetype Source
Sysmon for Linux EventID 11 Linux icon Linux 'sysmon:linux' 'Syslog:Linux-Sysmon/Operational'
Zeek HTTP Request Headers Web 'zeek:http_request_header:json' 'http_request_headers.log'
Sysmon for Linux EventID 5 Linux icon Linux 'sysmon:linux' 'Syslog:Linux-Sysmon/Operational'
Zeek HTTP POST Body Web 'zeek:http_post_body:json' 'http_post_body.log'
Sysmon for Linux EventID 1 Linux icon Linux 'sysmon:linux' 'Syslog:Linux-Sysmon/Operational'
Java HotSpot Exceptions Other 'java:hotspot:exceptions' 'exceptions-*.log'
Zeek HTTP Web 'zeek:http' 'http.log'

Macros Used

Name Value
java_hotspot_exceptions sourcetype="java:hotspot:exceptions"
http_base64_activity_with_jvm_errors_filter search *
http_base64_activity_with_jvm_errors_filter is an empty macro by default. It allows the user to filter out any results (false positives) without editing the SPL.

Annotations

CVE

Default Configuration

This detection is configured by default in Splunk Enterprise Security to run with the following settings:

Setting Value
Disabled true
Cron Schedule 0 * * * *
Earliest Time -70m@m
Latest Time -10m@m
Schedule Window auto
Creates Finding (Notable) No
Creates Intermediate Finding (Risk Event) Yes
Anomaly detections generate Intermediate Findings (Risk Events). They do not generate a Finding (Notable) directly.

Implementation

Configure zeek_http_uri, zeek_http_request_headers, zeek_http_post_body, and java_hotspot_exceptions for the indexes and sourcetypes in your environment. Configure sysmon_linux if you collect that optional source. Install the TA for Zeek for standard HTTP logs. To collect the optional header and POST-body streams, install the custom scripts from https://gist.github.com/nasbench/8581572459af92b7c30599f1068c791f (request headers) and https://gist.github.com/nasbench/4cd18a938056cbfbc8a9832edb26f4c4 (POST bodies), following the corresponding data source instructions. These scripts are not included with ESCU or the TA for Zeek. Ensure uid, uri, id.orig_h, id.resp_h, header_name, header_value, body, and truncated are extracted as applicable. The sensor must see request plaintext.

HotSpot exception diagnostics are not written to a standard file by default. Tomcat normally writes JULI logs under CATALINA_BASE/logs (and, on Unix with its startup scripts, console output to catalina.out); Spring Boot writes application logs to the console by default; and Elasticsearch application logs use /var/log/elasticsearch for Debian/RPM packages, ES_HOME/logs for tar/zip installs, or container stdout. Elasticsearch enables GC logs by default, but these are separate from the exception stream required here. Add this JVM startup option with a writable path chosen for the deployment: -Xlog:exceptions=debug:file=/var/log/myapp/exceptions-%p.log:utctime,pid,level,tags:filecount=8,filesize=16m For Tomcat scripts, place it in CATALINA_OPTS via CATALINA_BASE/bin/setenv.sh or setenv.bat; for a Windows Tomcat service, configure its Java options. For Spring Boot, add it to the Java service or container launch command. For Elasticsearch, add a .options file in config/jvm.options.d or /etc/elasticsearch/jvm.options.d, according to the installation method, and preserve existing JVM logging settings. Spring Boot logging.file.name and Tomcat logging.properties control application or container logs, not HotSpot unified logging. Forward the configured file and rotations as java:hotspot:exceptions. On the Splunk parsing tier, configure single-line event breaking and timestamps for utctime. Search-time extraction in props.conf is optional: use EXTRACT-jvm_unified (or a shared REPORT transform) for reusable jvm_pid, jvm_tags, and jvm_message fields, or leave those fields unconfigured and let this search parse the decorated _raw line with rex. The search coalesces raw and pre-extracted values, including jvm_tags, before correlating the Java PID. The Java HotSpot Exceptions data source gives the input and optional extraction pattern. Verify the raw format, host identity, file permissions, and retention before enabling the search. Exception logging can be high volume, and an exception event is not guaranteed during exploitation.

Optionally ingest Sysmon for Linux EventID 1, 5, and 11 with ParentImage, ParentProcessId, Image, ProcessId, and TargetFilename. Process creation joins on ParentProcessId; file creation and process termination join on ProcessId.

Configure Enterprise Security Assets and Identities so the Zeek destination IP and the JVM or Sysmon hostname resolve to the same dest_asset_id through get_asset(dest). Synchronize the sensor and server clocks. The search permits JVM events from 3 seconds before to 15 seconds after each request marker, and optional Sysmon events up to 30 seconds after it. Tune these windows to measured collection delays.

Known False Positives

One JVM can handle concurrent requests, so an exception may fall within 15 seconds of more than one malformed request. HotSpot logs do not carry the Zeek uid, and this search establishes time proximity rather than causation. Ordinary Java child processes and file writes can coincide with the request. Review the URI, Zeek uid, exception delay, Java PID, child image, and created path. Clock skew, PID reuse, and incorrect host/IP mapping can misassociate events. Sysmon evidence is supporting context and is not required for a result.

Associated Analytic Story

Intermediate Findings

Message Entity Field Entity Type Risk Score
HTTP Base64 marker in request [$request_uid$] from [$src$] to [$dest$] aligned with a Base64-related argument exception in Java PID [$java_pid$] at a [$exception_delay$]-second offset. Confidence: [$confidence$] dest system 20

References

Detection Testing

Test Type Status Dataset Source Sourcetype
Validation ✅ Passing N/A N/A N/A
Unit ✅ Passing Dataset http.log bro:http:json
Integration ✅ Passing Dataset http.log bro:http:json

Replay any dataset to Splunk Enterprise by using our replay.py tool or the UI. Alternatively you can replay a dataset into a Splunk Attack Range


Source: GitHub |

Version: 1