Jupyter Notebook 代码示例

本文提供了一些示例代码片段，这些代码片段演示如何使用 Jupyter 笔记本与 Microsoft Sentinel lake 数据（预览版）交互，以分析Microsoft Sentinel 数据湖中的安全数据。 这些示例演示如何访问和分析各种表中的数据，例如Microsoft Entra ID 登录日志、组信息和设备网络事件。 代码片段旨在使用 Microsoft Sentinel 扩展在 Visual Studio Code 中的 Jupyter 笔记本中运行。

若要运行这些示例，必须具有所需的权限，并使用 Microsoft Sentinel 扩展安装 Visual Studio Code。 有关详细信息，请参阅 Microsoft Sentinel 数据湖权限 ，并将 Jupyter 笔记本与 Microsoft Sentinel data lake 配合使用。

失败的登录尝试分析

此示例标识尝试登录失败的用户。 为此，此笔记本示例处理两个表中的登录数据：

• SigninLogs
• AAD非交互式用户登录日志 (AADNonInteractiveUserSignInLogs)

笔记本执行以下步骤：

1. 创建一个函数来处理来自指定表的数据，其中包括：
   1. 将数据从指定的表加载到 DataFrames 中。
   2. 分析“状态”JSON 字段以提取“errorCode”，并确定每次登录尝试是否成功还是失败。
   3. 聚合数据，以计算每个用户失败和成功登录尝试次数。
   4. 筛选数据以仅包含登录尝试失败超过 100 次的用户，以及至少一次成功的登录尝试。
   5. 按失败的登录尝试次数对结果进行排序。
2. 为 SigninLogs 表和 AADNonInteractiveUserSignInLogs 表调用函数。
3. 将这两个表的结果合并为单个数据帧。
4. 将 DataFrame 转换为 Pandas DataFrame。
5. 筛选 Pandas DataFrame 以显示失败尝试次数最多的前 20 个用户。
6. 创建条形图，以可视化尝试失败次数最多的用户。

# Import necessary libraries
import matplotlib.pyplot as plt
from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import col, when, count, from_json, desc
from pyspark.sql.types import StructType, StructField, StringType

data_provider = MicrosoftSentinelProvider(spark)

# Function to process data
def process_data(table_name,workspace_name):
    # Load data into DataFrame
    df = data_provider.read_table(table_name, workspace_name)
    
    # Define schema for parsing the 'Status' JSON field
    status_schema = StructType([StructField("errorCode", StringType(), True)])
    # Parse the 'Status' JSON field to extract 'errorCode'
    df = df.withColumn("Status_json", from_json(col("Status"), status_schema)) \
           .withColumn("ResultType", col("Status_json.errorCode"))
    # Define success codes
    success_codes = ["0", "50125", "50140", "70043", "70044"]
    
    # Determine FailureOrSuccess based on ResultType
    df = df.withColumn("FailureOrSuccess", when(col("ResultType").isin(success_codes), "Success").otherwise("Failure"))
    
    # Summarize FailureCount and SuccessCount
    df = df.groupBy("UserPrincipalName", "UserDisplayName", "IPAddress") \
           .agg(count(when(col("FailureOrSuccess") == "Failure", True)).alias("FailureCount"),
                count(when(col("FailureOrSuccess") == "Success", True)).alias("SuccessCount"))
    
    # Filter where FailureCount > 100 and SuccessCount > 0
    df = df.filter((col("FailureCount") > 100) & (col("SuccessCount") > 0))
    
    # Order by FailureCount descending
    df = df.orderBy(desc("FailureCount"))
         
    return df

# Process the tables to a common schema
workspace_name = "your-workspace-name"  # Replace with your actual workspace name
aad_signin = process_data("SigninLogs", workspace_name)
aad_non_int = process_data("AADNonInteractiveUserSignInLogs", workspace_name)

# Union the DataFrames
result_df = aad_signin.unionByName(aad_non_int)

# Show the result
result_df.show()

# Convert the Spark DataFrame to a Pandas DataFrame
result_pd_df = result_df.toPandas()

# Filter to show table with top 20 users with the highest failed sign-ins attempted
top_20_df = result_pd_df.nlargest(20, 'FailureCount')

# Create bar chart to show users by highest failed sign-ins attempted
plt.figure(figsize=(12, 6))
plt.bar(top_20_df['UserDisplayName'], top_20_df['FailureCount'], color='skyblue')
plt.xlabel('Users')
plt.ylabel('Number of Failed sign-ins')
plt.title('Top 20 Users with Failed sign-ins')
plt.xticks(rotation=45, ha='right')
plt.tight_layout()
plt.show()  

以下屏幕截图显示了上述代码的输出示例，其中显示了条形图格式中失败登录尝试次数最多的前 20 个用户。

访问湖层 Microsoft Entra ID 组表

下面的代码示例演示如何访问 EntraGroups Microsoft Sentinel data lake 中的表。 它显示各种字段，例如displayName、、groupTypesmail、mailNicknamedescription和tenantId。

from sentinel_lake.providers import MicrosoftSentinelProvider
data_provider = MicrosoftSentinelProvider(spark)
 
table_name = "EntraGroups"  
df = data_provider.read_table(table_name)  
df.select("displayName", "groupTypes", "mail", "mailNickname", "description", "tenantId").show(100, truncate=False)   

以下屏幕截图显示了上述代码的输出示例，以数据帧格式显示Microsoft Entra ID 组信息。

访问特定用户的 Microsoft Entra ID 登录日志

下面的代码示例演示如何访问Microsoft Entra ID SigninLogs 表并筛选特定用户的结果。 它检索各种字段，例如 UserDisplayName、UserPrincipalName、UserId 等。

from sentinel_lake.providers import MicrosoftSentinelProvider
data_provider = MicrosoftSentinelProvider(spark)
 
table_name = "SigninLogs"  
workspace_name = "your-workspace-name"  # Replace with your actual workspace name
df = data_provider.read_table(table_name, workspace_name)  
df.select("UserDisplayName", "UserPrincipalName", "UserId", "CorrelationId", "UserType", 
 "ResourceTenantId", "RiskLevelDuringSignIn", "ResourceProvider", "IPAddress", "AppId", "AADTenantId")\
    .filter(df.UserPrincipalName == "bploni5@contoso.com")\
    .show(100, truncate=False) 

检查登录位置

下面的代码示例演示如何从 Microsoft Entra ID SigninLogs 表中提取和显示登录位置。 它使用 from_json 函数分析字段的 LocationDetails JSON 结构，使你能够访问特定位置属性，例如城市、州和国家/地区或区域。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import from_json, col  
from pyspark.sql.types import StructType, StructField, StringType  
 
data_provider = MicrosoftSentinelProvider(spark)  
workspace_name = "your-workspace-name"  # Replace with your actual workspace name
table_name = "SigninLogs"  
df = data_provider.read_table(table_name, workspace_name)  
 
location_schema = StructType([  
  StructField("city", StringType(), True),  
  StructField("state", StringType(), True),  
  StructField("countryOrRegion", StringType(), True)  
])  
 
# Extract location details from JSON  
df = df.withColumn("LocationDetails", from_json(col("LocationDetails"), location_schema))  
df = df.select("UserPrincipalName", "CreatedDateTime", "IPAddress", 
 "LocationDetails.city", "LocationDetails.state", "LocationDetails.countryOrRegion")  
 
sign_in_locations_df = df.orderBy("CreatedDateTime", ascending=False)  
sign_in_locations_df.show(100, truncate=False) 

来自异常国家的登录

下面的代码示例演示如何识别不属于用户典型登录模式的国家/地区的登录。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import from_json, col
from pyspark.sql.types import StructType, StructField, StringType

data_provider = MicrosoftSentinelProvider(spark)
table_name = "signinlogs"
workspace_name = "your-workspace-name"  # Replace with your actual workspace name
df = data_provider.read_table(table_name, workspace_name)

location_schema = StructType([
    StructField("city", StringType(), True),
    StructField("state", StringType(), True),
    StructField("countryOrRegion", StringType(), True)
])

# Extract location details from JSON
df = df.withColumn("LocationDetails", from_json(col("LocationDetails"), location_schema))
df = df.select(
    "UserPrincipalName",
    "CreatedDateTime",
    "IPAddress",
    "LocationDetails.city",
    "LocationDetails.state",
    "LocationDetails.countryOrRegion"
)

sign_in_locations_df = df.orderBy("CreatedDateTime", ascending=False)
sign_in_locations_df.show(100, truncate=False)

来自多个失败登录的暴力攻击

通过分析具有大量失败登录尝试的帐户的用户登录日志来识别潜在的暴力攻击。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import col, when, count, from_json, desc
from pyspark.sql.types import StructType, StructField, StringType

data_provider = MicrosoftSentinelProvider(spark)

def process_data(table_name, workspace_name):
    df = data_provider.read_table(table_name, workspace_name)
    status_schema = StructType([StructField("errorCode", StringType(), True)])
    df = df.withColumn("Status_json", from_json(col("Status"), status_schema)) \
           .withColumn("ResultType", col("Status_json.errorCode"))
    success_codes = ["0", "50125", "50140", "70043", "70044"]
    df = df.withColumn("FailureOrSuccess", when(col("ResultType").isin(success_codes), "Success").otherwise("Failure"))
    df = df.groupBy("UserPrincipalName", "UserDisplayName", "IPAddress") \
           .agg(count(when(col("FailureOrSuccess") == "Failure", True)).alias("FailureCount"),
                count(when(col("FailureOrSuccess") == "Success", True)).alias("SuccessCount"))
    # Lower the brute force threshold to >10 failures and remove the success requirement
    df = df.filter(col("FailureCount") > 10)
    df = df.orderBy(desc("FailureCount"))
    df = df.withColumn("AccountCustomEntity", col("UserPrincipalName")) \
           .withColumn("IPCustomEntity", col("IPAddress"))
    return df
workspace_name = "your-workspace-name"  # Replace with your actual workspace name
aad_signin = process_data("SigninLogs", workspace_name)
aad_non_int = process_data("AADNonInteractiveUserSignInLogs",workspace_name)
result_df = aad_signin.unionByName(aad_non_int)
result_df.show()

检测横向移动尝试

使用 DeviceNetworkEvents 识别可能发出横向移动信号的可疑内部 IP 连接，例如终结点之间的异常 SMB/RDP 流量。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import col, count, countDistinct, desc

deviceNetworkEventTable = "DeviceNetworkEvents"
workspace_name = "<your-workspace-name>"  # Replace with your actual workspace name
data_provider = MicrosoftSentinelProvider(spark)
device_network_events = data_provider.read_table(deviceNetworkEventTable, workspace_name)

# Define internal IP address range (example: 10.x.x.x, 192.168.x.x, 172.16.x.x - 172.31.x.x)
internal_ip_regex = r"^(10\.\d{1,3}\.\d{1,3}\.\d{1,3}|192\.168\.\d{1,3}\.\d{1,3}|172\.(1[6-9]|2[0-9]|3[0-1])\.\d{1,3}\.\d{1,3})$"

# Filter for internal-to-internal connections
internal_connections = device_network_events.filter(
    col("RemoteIP").rlike(internal_ip_regex) &
    col("LocalIP").rlike(internal_ip_regex)
)

# Group by source and destination, count connections
suspicious_lateral = (
    internal_connections.groupBy("LocalIP", "RemoteIP", "InitiatingProcessAccountName")
    .agg(count("*").alias("ConnectionCount"))
    .filter(col("ConnectionCount") > 10)  # Threshold can be adjusted
    .orderBy(desc("ConnectionCount"))
)
suspicious_lateral.show()

发现凭据转储工具

查询 DeviceProcessEvents 查找 mimikatz.exe 等进程或不应执行的 lsass.exe 访问，这可能表示在进行凭据搜集。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import col, lower

workspace_id = "<your-workspace-name>"
device_process_table = "DeviceProcessEvents"

data_provider = MicrosoftSentinelProvider(spark)
process_events = data_provider.read_table(device_process_table, workspace_id)

# Look for known credential dumping tools and suspicious access to lsass.exe
suspicious_processes = process_events.filter(
    (lower(col("FileName")).rlike("mimikatz|procdump|lsassy|nanodump|sekurlsa|dumpert")) |
    (
        (lower(col("FileName")) == "lsass.exe") &
        (~lower(col("InitiatingProcessFileName")).isin(["services.exe", "wininit.exe", "taskmgr.exe"]))
    )
)

suspicious_processes.select(
    "Timestamp",
    "DeviceName",
    "AccountName",
    "FileName",
    "FolderPath",
    "InitiatingProcessFileName",
    "InitiatingProcessCommandLine"
).show(50, truncate=False)

USB 活动与敏感文件访问的关联

将 DeviceEvents 和 DeviceFileEvents 合并到笔记本中，以显示潜在的数据外泄模式。 添加可视化效果以显示涉及哪些设备、用户或文件，以及何时参与。

from sentinel_lake.providers import MicrosoftSentinelProvider
from pyspark.sql.functions import col, lower, to_timestamp, expr
import matplotlib.pyplot as plt

data_provider = MicrosoftSentinelProvider(spark)
workspace_id = “<your-workspace-id>”

# Load DeviceEvents and DeviceFileEvents tables
device_events = data_provider.read_table("DeviceEvents", workspace_id)
device_file_events = data_provider.read_table("DeviceFileEvents", workspace_id)
device_info = data_provider.read_table("DeviceInfo", workspace_id)

# Filter for USB device activity (adjust 'ActionType' or 'AdditionalFields' as needed)
usb_events = device_events.filter(
    lower(col("ActionType")).rlike("usb|removable|storage")
)

# Filter for sensitive file access (e.g., files in Documents, Desktop, or with sensitive extensions)
sensitive_file_events = device_file_events.filter(
    lower(col("FolderPath")).rlike("documents|desktop|finance|confidential|secret|sensitive") |
    lower(col("FileName")).rlike(r"\.(docx|xlsx|pdf|csv|zip|7z|rar|pst|bak)$")
)

# Convert timestamps
usb_events = usb_events.withColumn("EventTime", to_timestamp(col("Timestamp")))
sensitive_file_events = sensitive_file_events.withColumn("FileEventTime", to_timestamp(col("Timestamp")))

# Join on DeviceId and time proximity (within 10 minutes) using expr for column operations
joined = usb_events.join(
    sensitive_file_events,
    (usb_events.DeviceId == sensitive_file_events.DeviceId) &
    (expr("abs(unix_timestamp(EventTime) - unix_timestamp(FileEventTime)) <= 600")),
    "inner"
) \
.join(device_info, usb_events.DeviceId == device_info.DeviceId, "inner")


# Select relevant columns
correlated = joined.select(
    device_info.DeviceName,
    usb_events.DeviceId,
    usb_events.AccountName,
    usb_events.EventTime.alias("USBEventTime"),
    sensitive_file_events.FileName,
    sensitive_file_events.FolderPath,
    sensitive_file_events.FileEventTime
)

correlated.show(50, truncate=False)

# Visualization: Number of sensitive file accesses per device
pd_df = correlated.toPandas()
if not pd_df.empty:
    plt.figure(figsize=(12, 6))
    pd_df.groupby('DeviceName').size().sort_values(ascending=False).head(10).plot(kind='bar')
    plt.title('Top Devices with Correlated USB and Sensitive File Access Events')
    plt.xlabel('DeviceName')
    plt.ylabel('Number of Events')
    plt.tight_layout()
    plt.show()
else:
    print("No correlated USB and sensitive file access events found in the selected period.")

信标行为检测

通过对长时间内低字节量的常规出站连接进行聚类来检测潜在的命令和控制。

# Setup
from pyspark.sql.functions import col, to_timestamp, window, count, avg, stddev, hour, date_trunc
from sentinel_lake.providers import MicrosoftSentinelProvider 
import matplotlib.pyplot as plt
import pandas as pd

data_provider = MicrosoftSentinelProvider(spark)
device_net_events = "DeviceNetworkEvents"
workspace_id = "<your-workspace-id>"

network_df = data_provider.read_table(device_net_events, workspace_id)

# Add hour bucket to group by frequency
network_df = network_df.withColumn("HourBucket", date_trunc("hour", col("Timestamp")))

# Group by device and IP to count hourly traffic
hourly_traffic = network_df.groupBy("DeviceName", "RemoteIP", "HourBucket") \
    .agg(count("*").alias("ConnectionCount"))

# Count number of hours this IP talks to device
stats_df = hourly_traffic.groupBy("DeviceName", "RemoteIP") \
    .agg(
        count("*").alias("HoursSeen"),
        avg("ConnectionCount").alias("AvgConnPerHour"),
        stddev("ConnectionCount").alias("StdDevConnPerHour")
    )

# Filter beacon-like traffic: low stddev, repeated presence
beacon_candidates = stats_df.filter(
    (col("HoursSeen") > 10) &
    (col("AvgConnPerHour") < 5) &
    (col("StdDevConnPerHour") < 1.0)
)

beacon_candidates.show(truncate=False)

# Choose one Device + IP pair to plot
example = beacon_candidates.limit(1).collect()[0]
example_device = example["DeviceName"]
example_ip = example["RemoteIP"]

# Filter hourly traffic for that pair
example_df = hourly_traffic.filter(
    (col("DeviceName") == example_device) & 
    (col("RemoteIP") == example_ip)
).orderBy("HourBucket")

# Convert to Pandas and plot
example_pd = example_df.toPandas()
example_pd["HourBucket"] = pd.to_datetime(example_pd["HourBucket"])

plt.figure(figsize=(12, 5))
plt.plot(example_pd["HourBucket"], example_pd["ConnectionCount"], marker="o", linestyle="-")
plt.title(f"Outbound Connections – {example_device} to {example_ip}")
plt.xlabel("Time (Hourly)")
plt.ylabel("Connection Count")
plt.grid(True)
plt.tight_layout()
plt.show()

相关内容

• Microsoft Sentinel 提供程序类参考（预览版）
• Microsoft Sentinel data lake 概述（预览版）
• Microsoft Sentinel 数据湖权限（预览版）
• 使用 Jupyter 笔记本浏览 Microsoft Sentinel 数据湖（预览版）
• Jupyter 笔记本和 Microsoft Sentinel 数据湖（预览版）