资源 · 29

准备好安全事件日志以供检测

选择事件、保护记录、关联系统并演练响应。

已更新 · 26 min

虚构的无人网络事件采集室,配有网络采集设备、线缆、抽象事件显示屏和酒红色文件夹 由人工智能生成的编辑插图

本指南有助于实现的目标

  • 定义事件
  • 保护记录
  • 关联信号
  • 测试检测

快速检查

  • 哪些系统支持关键路径?
  • 时钟和标识符能否重建时间线?
  • 谁可以读取或修改记录?
  • 什么场景会触发警报?
  • 是否有人知道下一步该做什么?

分步指南

  1. 1

    选择数据源

    从服务和威胁场景入手:登录、权限变更、数据访问、API、网络和基础设施。 定义所需事件及其负责人

    交付成果:源级日志记录策略

  2. 2

    规范化上下文

    指定时钟、时区、请求标识符、参与者、结果和源系统。 避免在消息中包含不必要的机密信息和个人数据。

    交付成果:事件模式和最小化规则。

  3. 3

    集中化和保护

    测试传输、可用性、权限、保留和篡改检测。 受损系统不应悄无声息地删除唯一有用的证据。

    交付成果:传输和完整性检查。

  4. 4

    关联场景

    模拟可疑的登录操作,随后进行 API 调用或敏感更改。 验证事件在时间和标识符上是否一致。

    交付成果:重建的时间线和差距列表。

  5. 5

    调整检测

    编写包含条件、阈值、例外情况、所有者和优先级的规则。 使用授权场景衡量误报和漏报事件。

    交付成果:已测试规则和验证用例。

  6. 6

    演练响应

    将警报发送给负责人,启动调查,保存证据并验证是否已结案。 演练后改进模式或规则。

    交付成果:带日期的演练和后续行动。

理解、计算与判断

日志:区分事件时间与接收时间

收到日志不等于已经检测或处理告警。这个虚构场景追踪同一账号的两次登录失败和一次成功。时间是从共同起点开始的秒数,并假定时钟已同步。

分别保留事件时间与接收时间。测试日志丢失、延迟、读取权限和响应责任人。不要记录密码、访问令牌或完整请求体。

观察到的行为 → 采集到的事件 → 生成的警报 → 验证过的响应
从日志到经过验证的检测 — 授权演练中,测试登录生成了日志,却没有警报。采集有效,但检测尚未得到证明。将ATT&CK技术标为已覆盖之前,应验证整个链路。 MITRE ATT&CK

指标与证据

指标含义所需证据
E1 0秒失败;2秒接收 延迟:2秒
E2 20秒失败;140秒接收 延迟:120秒
E3 35秒成功;36秒接收 延迟:1秒

查看每条日志的延迟

按事件时间,E1、E2和E3都在一分钟内。按接收时间,E2延迟到达,并晚于成功事件。应测试迟到事件策略和告警能否送达负责人。教学规则本身不能证明攻击。

查看每条日志的延迟

共同刻度为0至150秒。浅色圆点表示事件,金色方块表示接收,线段表示延迟。表格提供相同数值。

  • E10 s → 2 s (+2 s)
  • E220 s → 140 s (+120 s)
  • E335 s → 36 s (+1 s)

管理指标

指标衡量内容首要行动
覆盖面包含预期事件的关键源添加缺失的源
延迟从操作到可搜索记录的时间修复传输
关联端到端场景重建添加上下文或标识符
响应已测试的警报及其所有者和操作修复规则或值班路径

常见错误

  • 未对场景进行优先级排序而收集所有信息
  • 以明文形式记录密钥或敏感数据
  • 未进行完整性和权限检查而集中管理
  • 未添加处理程序而添加警报

常见问题解答

更多日志是否总是意味着更好的检测?

不是。质量、上下文和响应能力与日志量同样重要。

为什么要同步时钟?

不一致的时间线会降低关联和事件决策的可靠性。

规则何时准备就绪?

当授权场景产生预期信号且所有者可以进行分类时。

参考资料:数据、用途与局限

文档核查日期 .

此语言版本的详细参考说明以英语提供。

参考资料:数据、用途与局限
来源与目的已核实的数据或指标使用方法需要保留的局限
OASIS · STIX 2.1

Describe threat intelligence through structured objects and relationships.

STIX 2.1 is an OASIS Standard dated 10 June 2021. It defines objects, properties and JSON serialisation. Keep identifiers, dates, provenance and relationships. Check that the recipient interprets the same object type and validity period.

深入了解方法 →

A syntactically valid object can still be stale, misattributed or weakly supported.
MITRE ATT&CK

Connect observations with adversary tactics and techniques.

ATT&CK is a knowledge base grounded in observations. Its domains include Enterprise, Mobile and ICS. Connect a technique with a trace and detection hypothesis. Distinguish having a log source from actually testing detection.

深入了解方法 →

A shared technique cannot establish actor attribution. Technique coverage counts do not fully measure security.

STIX 指标与 TLP 共享

此语言版本的详细参考说明以英语提供。

Cyber threat intelligence: from an indicator to a defensible detection

A threat intelligence feed may deliver a domain, a detection pattern and a confidence assessment. A SOC still needs its provenance, useful time window, telemetry capable of recognising it and a retirement procedure. This reference helps prepare that decision and explain its limits to the recipient.

The properties and objects described here refer to the OASIS STIX 2.1 Standard dated 10 June 2021. Deployment checks, review deadlines and case decisions are local editorial proposals. This page is neither a STIX validator, a blocking rule nor an assessment of a real feed.

Version note: at review time the OASIS latest-stage URL points to a document dated 2 April 2025 titled STIX 2.1 Errata 01, incorporating Draft 01. Its draft errata status is retained in the citation; it is not presented as a newly approved final version. Agree marking encodings with the recipient and check the edition actually used.

Local indicator lifecycle and a separate sharing check
Local indicator lifecycle and a separate sharing check Four steps: preserve source and version; qualify validity and context; test detection against logs; review and propagate retirement. TLP scope is checked at every disclosure, independently of confidence. 1 · Preserve Source and version Receipt time 2 · Qualify Validity and context Explicit confidence 3 · Test Logs and coverage Exceptions and decision 4 · Review Review deadline Retirement propagated At every disclosure: check TLP, recipients and source restrictions.

Four steps: preserve source and version; qualify validity and context; test detection against logs; review and propagate retirement. TLP scope is checked at every disclosure, independently of confidence.

12 reading checks before importing an indicator

12 reading checks before importing an indicator
STIX 2.1 propertyWhat it tells youOperational check to record
id / modifiedA versioned object retains its id; modified identifies its version.Compare received versions before overwriting local state. A later delivery can contain an older version.
created_by_refOptional reference to the creator’s Identity; when absent, this property leaves the source undefined.Also preserve collection provenance. A TAXII relay is not necessarily the intelligence creator.
created / modifiedObject creation and version modification times; modified is at least equal to created.Do not treat them as first and last observed activity times.
pattern / pattern_typeBoth the pattern and its language are required; a STIX pattern describes a condition to recognise.Check language, version, parser and available telemetry fields before translating it into a SIEM rule.
valid_fromRequired start of the Indicator’s validity window.A future start calls for holding or review; it does not establish current activity.
valid_untilOptional end, strictly later than valid_from when present. At this time the indicator should no longer be considered valid.Without an end the standard sets no upper time bound. Still assign a local review date without inventing a producer expiry.
revokedPermanent withdrawal by the creator. No future version of the same id may be created after revocation.Propagate withdrawal to derived uses and preserve the record. Expiry or a local false positive does not automatically mean revoked=true.
confidenceOptional integer from 0 to 100 expressing the creator’s confidence in data correctness; absence means unspecified.Preserve the producer’s method. Do not turn 60 into a 60% probability of compromise, or absence into zero.
external_referencesLinks, descriptions or identifiers referring to external information.Check origin and access to cited evidence. Multiple copies of one report are not independent sources.
object_marking_refsReferences to marking-definition objects applying to the object.Resolve references and preserve restrictions in copies, tickets and summaries.
granular_markingsMarkings or languages applying to specific parts of an object.Check the relevant fields before quoting an excerpt; a title’s marking does not establish the pattern’s marking.
description / indicator_typesNarrative context and categories help interpretation; these properties are optional.Explain the behaviour sought, prerequisites, possible legitimate uses and person responsible for review.

Four objects answering four different questions

Four objects answering four different questions
ObjectQuestion it answersAnalytical limit
IndicatorWhat pattern could recognise suspicious activity?A usable pattern is not evidence that the activity occurred in your environment.
Observed DataWhat data was observed, and during which window?Observed facts alone do not establish attribution or malicious intent.
SightingWhich intelligence entity does the reporter believe they saw?The assertion may reference Observed Data. Sighting count and number_observed measure different things.
RelationshipWhat link is asserted between two objects?An indicates or based-on link has a specific meaning. A dense graph does not prove attribution.

Read a fictional object without turning it into an alert

A standalone training example written for this page; no real threat intelligence. RFC 2606 reserves .example for documentation. Dates and confidence 60 are fictional. Creator and marking are deliberately absent: this establishes neither a verified source nor permission to publish. This is neither a complete bundle nor an OASIS-tool compliance validation.

{
    "type": "indicator",
    "spec_version": "2.1",
    "id": "indicator--643d2844-11ca-4f63-8d5d-e5d938db1854",
    "created": "2026-10-22T09:00:00.000Z",
    "modified": "2026-10-22T09:00:00.000Z",
    "lang": "en",
    "name": "Fictional training indicator",
    "description": "Documentation only. Never deploy this example as a detection or blocking rule.",
    "pattern": "[domain-name:value = 'training-beacon.example']",
    "pattern_type": "stix",
    "pattern_version": "2.1",
    "valid_from": "2026-10-22T09:00:00Z",
    "valid_until": "2026-10-23T09:00:00Z",
    "confidence": 60
}

At 10:00 UTC on 22 October 2026, the time falls inside the stated window. This establishes neither a log match nor maliciousness. Deriving a detection would still require the producer, context, restrictions, telemetry, exceptions and retirement owner.

Eight fictional qualification and sharing cases

Each case isolates one difficulty. Suggested decisions belong to a local working method, rather than automatic verdicts from STIX syntax alone. The filter only hides examples already in the page; it sends no data.

Validity starts tomorrow

The pattern arrives today; valid_from is tomorrow.

Retain the object for review and schedule assessment at the appropriate time; receipt alone does not justify immediate blocking.

Keep receipt time, evaluation time and object version. Historical investigation has a different purpose.

Validity end reached

At 09:00 UTC, valid_until is reached; a derived rule remains active.

Remove the object from active candidates or trigger the planned retirement process, then check derived rules.

Retain history and the withdrawal rationale. The deadline does not prove past events were benign.

Revocation received after import

A new version of the same id contains revoked=true.

Propagate withdrawal to dependent uses and verify each consumer’s state.

Keep the received version and propagation record; do not reactivate the object from an older copy.

Confidence unspecified

The feed provides no confidence; an interface proposes displaying 0.

Display unspecified, request provenance and describe available corroboration.

An internal scoring convention may exist, but must be named and kept separate from the producer’s value.

Shared domain and DNS log

A cited domain appears in a DNS log; it also hosts legitimate uses.

Inspect the exact name, time, host and related events; prepare a focused search before broad action.

DNS resolution demonstrates neither download, execution nor compromise. Plan exceptions and rollback.

Sighting count=0

A producer publishes a Sighting with count=0.

Read this as a non-observation statement in its context, then ask about window and coverage.

Zero does not measure worldwide absence of the threat. It does not replace checks of your own sensors.

Client and AMBER+STRICT

A recipient wants to forward a TLP:AMBER+STRICT note to a client.

The restriction limits sharing to the organisation on a need-to-know basis; obtain explicit source permission to widen it.

Do not mistakenly apply the client-sharing scope associated with AMBER without STRICT.

GREEN message to a public channel

A TLP:GREEN note is proposed for a public article.

Keep the note inside the permitted community; ask the source for permission to share more widely.

A publicly accessible web page exceeds GREEN’s scope. Source attribution does not lift the restriction.

TLP 2.0: recipients, need to know and restrictions

FIRST defines four labels; AMBER+STRICT restricts AMBER rather than adding a fifth colour. Labels remain unchanged across languages. This matrix summarises TLP scope and does not replace additional source restrictions or applicable document rules.

TLP 2.0: recipients, need to know and restrictions
MarkingSharing scopeCheck
TLP:REDIndividual recipients only; no further disclosure.A colleague’s membership of the same organisation does not authorise forwarding.
TLP:AMBEROrganisation and clients on a need-to-know basis to protect them and prevent further harm.Check that the recipient needs the information and no additional restriction narrows this scope.
TLP:AMBER+STRICTOrganisation only, on a need-to-know basis.Clients are not included by default under this restriction.
TLP:GREENThe relevant community; no publicly accessible channels.When the community is undefined, FIRST assumes the cybersecurity/defence community.
TLP:CLEARNo TLP disclosure limit, subject in particular to copyright rules.Preserve origin, separate restrictions and applicable public-release procedures.

TLP describes a sharing scope; it certifies neither accuracy, official classification nor a usage licence. Encryption and retention rules need separate definition. Obtain explicit source permission before widening the received scope.

Interoperability check: predefined TLP markings in the 2021 STIX 2.1 edition include TLP:WHITE. FIRST’s TLP 2.0 vocabulary uses TLP:CLEAR and specifies AMBER+STRICT. Do not invent a tlp=clear encoding or silently rename historical markings: document the version and exchange profile actually supported by the tools.

Six elements to hand over with the decision

  • Object identifier and version, declared producer and collection provenance.
  • UTC timestamps: receipt, validity window and a clearly separate local review date.
  • Detection hypothesis, required logs, fields used and known sensor coverage.
  • Search results and limits: absence, match, ambiguity or a justified false positive.
  • Recipients, markings, additional restrictions and permission to widen scope where applicable.
  • Owner, derived rule or ticket, decision, planned retirement and propagation verification.

References and verification scope

Sources checked on

syslog代码与采集证据

此语言版本的详细参考说明以英语提供。

Syslog priority combines a message family (facility) and a declared severity. It helps interpret a log stream but measures neither attack likelihood nor business incident urgency. Check what the sender encodes and what the collector preserves.

32 official values: 24 facilities and 8 severities. IANA registry updated 26 April 2016, consulted 23 October 2026. Cases and acceptance questions below are original CIE editorial examples.

IANA · syslog Parameters · Registry updated 2016-04-26 · consulted 2026-10-23

How do you calculate and decode syslog priority?

PRI = facility × 8 + severity. Reverse it with facility = integer quotient of PRI / 8 and severity = remainder of PRI / 8. The 192 combinations span 0–191. The field uses angle brackets and no leading zero except <0>. Example: 20 × 8 + 5 = 165, written <165>.

PRI is not a global danger ranking: <184> combines local7 and Emergency while <7> combines kernel and Debug. Interpret severity in the source context. A declared value or hostname does not authenticate a message.

IETF · R. Gerhards · RFC 5424 · The Syslog Protocol · March 2009 · §6.2.1 PRI · §6 format · §8 security

计算PRI

Choose two codes or enter only a PRI integer, with or without angle brackets. Accepted examples: 0, 165, <165>, <191>. Spaces, signs, decimals, leading zeros and complete messages are rejected. Calculation stays on this page; it does not connect to your equipment.

Official facility and severity tables

The registry’s English labels are preserved. A local0–local7 code does not identify the product or service that emitted it: your organisation must document its operational meaning.

24 facilities · codes 0–23

24 facilities · codes 0–23
CodeOfficial IANA labelRegistry note
0kernel messages
1user-level messages
2mail system
3system daemons
4security/authorization messagesCodes 4, 10, 13 and 14 have similar security, authorization, audit and alert uses across operating systems.
5messages generated internally by syslogd
6line printer subsystem
7network news subsystem
8UUCP subsystem
9clock daemonBoth codes 9 and 15 are used for clock or scheduled-job messages across operating systems.
10security/authorization messagesCodes 4, 10, 13 and 14 have similar security, authorization, audit and alert uses across operating systems.
11FTP daemon
12NTP subsystem
13log auditCodes 4, 10, 13 and 14 have similar security, authorization, audit and alert uses across operating systems.
14log alertCodes 4, 10, 13 and 14 have similar security, authorization, audit and alert uses across operating systems.
15clock daemonBoth codes 9 and 15 are used for clock or scheduled-job messages across operating systems.
16local use 0 (local0)
17local use 1 (local1)
18local use 2 (local2)
19local use 3 (local3)
20local use 4 (local4)
21local use 5 (local5)
22local use 6 (local6)
23local use 7 (local7)

8 severities · codes 0–7

8 severities · codes 0–7
CodeOfficial IANA labelRegistry note
0Emergency: system is unusable
1Alert: action must be taken immediately
2Critical: critical conditions
3Error: error conditions
4Warning: warning conditions
5Notice: normal but significant condition
6Informational: informational messages
7Debug: debug-level messages

IANA · Licensing Terms · Protocol Registries CC0 1.0 · Revised 2021-11-10 · original CIE commentary identified separately

Read a message before building a rule

Fictional example, containing no production data:

<165>1 2026-10-23T09:14:00Z gateway.example.net app 742 EVT42 [meta sequenceId="12"] Cache refreshed
Read a message before building a rule
ElementReading this exampleAcceptance consideration
<165>facility 20, severity 5Document the sender’s convention.
1Format versionDistinguish the received format from collector-added labels.
TIMESTAMPSource timestamp in UTCCheck reception time and clock drift as well.
HOSTNAME / APP-NAMEgateway.example.net / appConnect the declaration to your equipment inventory.
PROCID / MSGID742 / EVT42Do not treat a process number as a universal identifier.
STRUCTURED-DATAmeta with sequenceId 12Preserve names and parameters without flattening distinct identifiers.
MSGCache refreshedDo not log secrets to enrich an event.

A NILVALUE hyphen marks unavailable information in fields that allow it; it does not mean zero. Decoding PRI alone validates neither the rest of a message, its timestamp nor its transport.

Eight explained calculations · fictional situations

Eight explained calculations · fictional situations
Facility / severityCalculation and PRIInterpretation and limit
0 / 00 × 8 + 0 = 0 · <0>Kernel / Emergency: encoding an outage does not establish its cause.
1 / 51 × 8 + 5 = 13 · <13>User-level / Notice: a notable condition may still be normal.
4 / 24 × 8 + 2 = 34 · <34>Security / Critical: qualify the rule and asset before escalating.
10 / 510 × 8 + 5 = 85 · <85>Security / Notice: one family can carry a different severity.
13 / 613 × 8 + 6 = 110 · <110>Audit / Informational: an informative event can be decisive in an investigation.
20 / 520 × 8 + 5 = 165 · <165>Local4 / Notice: document the local convention; no universal product meaning.
23 / 023 × 8 + 0 = 184 · <184>Local7 / Emergency: high PRI, most severe level on this severity scale.
23 / 723 × 8 + 7 = 191 · <191>Local7 / Debug: same facility, eighth level, not evidence of harmlessness.

Collection acceptance: what evidence should you request?

Correct decoding does not establish useful detection. The joint guidance published by ASD ACSC and partners in 2024 addresses policy, centralisation, protected storage and detection strategy. This CIE acceptance proposal should be adapted to the service.

Collection acceptance: what evidence should you request?
CIE checkPractical exerciseExpected evidence
Source coverageTrigger an identified test event on a critical source.Find source, collector, index and owner; record absent sources.
Severity conventionCompare a known event before and after normalisation.Preserve raw code, normalised value and versioned conversion rule.
Identity and correlationFollow a fictional sign-in and subsequent application action.Connect useful identifiers without logging tokens, passwords or sensitive content.
Time and delayCompare source time, reception and search availability.Separate clock drift, transport delay and indexing delay.
Loss and recoveryInterrupt an authorised test stream and restore it.Observe gaps, queues, possible duplicates and missing-stream alerts.
Truncation and parsingFeed a long message and malformed event into the test pipeline.Record received size, lost fields or rejection; do not equate rejection with inactivity.
Path protectionReview transfer, access and retention with the owners.Identify every hop, permissions and the protection mechanisms actually configured.
Detection and responseTrigger a rule in an authorised scenario and follow handling.Connect event, alert, qualification, decision and observed result.

ASD ACSC and partners · Best practices for event logging and threat detection · Published and reviewed 2024-08-22 · consulted 2026-10-23

Set retention, scope and thresholds according to investigation needs and the service’s applicable constraints. This table defines neither a universal retention period nor SIEM certification. A format check cannot replace an end-to-end exercise.

Value provenance and consulted edition

IANA · 32 codes · registry 2016-04-26 · reviewed 2026-10-23 · CC0 1.0

IANA XML SHA-256: 10eb91ad6e29dc6fdebe5320cc0b2711ce840c1ed7a6d2c33c4bc716423b4ec6

RFC 5424 TXT SHA-256: 24d91f90a05d2536730ada929c296f45ae9cd72226e22b6775ad8c04cae37a53

Commentary, cases and acceptance proposal: Cyber Intelligence Embassy. IANA labels are reproduced for reference.

官方参考文献

参考文献支持该方法。 请根据您的实际情况调整检查;它们并非认证。 原始参考文献标题和源文档可能使用其他语言。

参考资料查阅日期 .