feat: judge SSRF flows on the destination their URL's constant text fixes

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## Unreleased
+
+### Precision
+
+- `ssrf` judges a flow on how `+` and f-strings build its URL. A fixed host is a limited proof: when constant text fixes `scheme://host` and ends the authority with `/`, `?` or `#` before any input, as in `"https://api.example.com/users/" + id`, the flow is a hotspot, because the input still chooses the path and a redirect may lead elsewhere. It is refuted only when the constant text fixes the path too, the input reaching the query or fragment only, and the call disables redirects (`allow_redirects=False`, `follow_redirects=False`). A module-level name counts as known text when the module binds it once to a string literal and never rebinds it. Everything else stays a vulnerability, including a base of unknown value such as an imported `BASE_URL`, which may leave the authority open. The advisory correlation keeps its own verdict. The regression corpus is unchanged: its six `ssrf` findings build their URL from values of unknown content (#165).
+
+### Plugins
+
+- `TaintDetector.judge(ctx, function, flow, verdict)` lets a rule refine the refutation's verdict on a flow for its own findings, without changing the verdict other consumers of the flow read (#165).
+
 ## 0.13.0 (2026-09-26)
 
 ### Precision

docs/plugins.md

@@ -115,6 +115,12 @@ ordinary input reaching a database write is not a plaintext credential; `LOG`, f
 values that must not be written to a log as they are; and `PII`, for personal data
 that must not leave the application through a log, a third party or plain storage.
 
+A rule may know more about its own sinks than the refutation does: `judge(ctx, function,
+flow, verdict)` returns the rule's verdict on a flow, given the refutation's, and by
+default returns it unchanged. `ssrf` reads there how the URL is built. The refutation's
+verdict, which other consumers of the flow such as the advisory correlation read, does
+not change; a rule reading more analyses declares them in its `requires`.
+
 ### Security models: `ModelPlugin`
 
 Declares sources, sinks and sanitizers for a library or a framework. A model plugin

docs/usage.md

@@ -114,7 +114,7 @@ variable is not treated that way yet.
 | `sql-injection` | A database statement (`cursor.execute`, SQLAlchemy `text`, Django `raw`, …). Parameters of a parameterised query are not statements. |
 | `nosql-injection` | A NoSQL query, such as a MongoDB filter, given attacker input that may hold a structure: a JSON body giving `{"$ne": null}` matches any value. The input that may hold a structure is JSON, the raw bodies a `json.loads` decodes, files, WebSocket messages and standard input. Text cannot hold an operator: form fields, query parameters, headers, cookies, URL parameters, the command line, the environment, and an entry-point parameter annotated with scalars (`str`, `int`, `Optional[str]`, `Annotated[str, Query()]`, `list[str]`, …), which FastAPI validates. A Django, REST framework, aiohttp or tornado view reads a request object: its text attributes (`GET`, `POST`, `headers`, `query_params`, `query`, `match_info`, `arguments`, …) and the URL parameters after it are text, its body, files and `data` are not. Limits: a helper function the request is passed to reads it as a whole; the keys of a dict are not followed, so a string concatenated into `$where` is not reported. The analyzer ships no NoSQL sink: model plugins declare them. |
 | `path-traversal` | A file system path (`open`, `send_file`, `os.remove`, …). |
-| `ssrf` | The URL of an HTTP client request (Requests, httpx, `urllib`), by position or as `url=`; the body, headers and query parameters are not the destination. |
+| `ssrf` | The URL of an HTTP client request (Requests, httpx, `urllib`), by position or as `url=`; the body, headers and query parameters are not the destination. How `+` and f-strings build the URL can prove part of its destination. When constant text fixes `scheme://host` and ends the authority with `/`, `?` or `#` before any input (`"https://api.example.com/users/" + id`, or a module-level string bound once and never rebound), the flow is a hotspot: the input still chooses the path, which may reach a sensitive resource, and a redirect may lead elsewhere. With the path fixed too, the input reaching the query or fragment only, and the call disabling redirects (`allow_redirects=False`, `follow_redirects=False`), the destination is fixed and the flow is refuted. Anything unproven stays a vulnerability: a base of unknown value (imported, a call result), a constant leaving the authority open (`"https:/"`, no `/` after the host, a variable port), `urljoin`, `%`, `format`, a URL built in a helper function. |
 | `xss` | An HTTP response body built without escaping. A Django template rendered by `render_to_string` escapes, where the analyzer can establish it (below). |
 | `insecure-deserialization` | `pickle`, `marshal`, `dill`, `jsonpickle`, `shelve`, and the YAML loaders (`yaml.load`, `load_all`, `full_load`, `unsafe_load` and their several-document versions); `yaml.load` and `load_all` given a safe loader (`SafeLoader`, `BaseLoader`) are not. |
 | `open-redirect` | An HTTP redirect target. |

src/coretrace_python/abstract/strings.py

@@ -0,0 +1,98 @@
+"""The constant text a string starts with (architecture §18): what ``+`` and f-strings
+build from string literals and from module-level names bound once to one. Anything
+else, a parameter, a call result, an import, ``%`` or ``format``, is text of unknown
+value."""
+
+from __future__ import annotations
+
+from collections.abc import Mapping
+from types import MappingProxyType
+from typing import ClassVar
+
+from coretrace_python.analysis import Analysis, AnalysisContext, AnyAnalysis
+from coretrace_python.cfg import CFGError
+from coretrace_python.hir import nodes
+from coretrace_python.ir.lowering import (
+    MODULE_BODY,
+    LoweringError,
+    PyIRAnalysis,
+    analyzable_functions,
+)
+from coretrace_python.ir.model import (
+    BinaryOp,
+    BuildString,
+    Constant,
+    ForNext,
+    Global,
+    Instruction,
+    SetGlobal,
+    StoreLocal,
+    Value,
+)
+
+
+def leading_text(value: Value, defs: Mapping[Value, Instruction], strings: Mapping[str, str]) -> tuple[str, bool]:
+    """The constant text ``value`` starts with, and whether that text is all of it, given
+    the definitions of the function's values and the module's ``strings``."""
+
+    made = defs.get(value)
+    if isinstance(made, Constant):
+        return (made.value, True) if isinstance(made.value, str) else ("", False)
+    if isinstance(made, Global):
+        known = strings.get(made.name)
+        return ("", False) if known is None else (known, True)
+    if isinstance(made, BinaryOp) and made.operator == "add":
+        left, whole = leading_text(made.left, defs, strings)
+        if not whole:
+            return left, False
+        right, whole = leading_text(made.right, defs, strings)
+        return left + right, whole
+    if isinstance(made, BuildString):
+        text = ""
+        for part in made.parts:
+            piece, whole = leading_text(part, defs, strings)
+            text += piece
+            if not whole:
+                return text, False
+        return text, True
+    return "", False
+
+
+class ModuleStringsAnalysis(Analysis[Mapping[str, str]]):
+    """The module-level names bound once, to a string literal, and never rebound: by no
+    other statement of the module body, no function assigning them as ``global``, no
+    definition of that name. A module that does not lower entirely has none."""
+
+    name: ClassVar[str] = "abstract.module_strings"
+    requires: ClassVar[frozenset[AnyAnalysis]] = frozenset({PyIRAnalysis})
+
+    @classmethod
+    def compute(cls, ctx: AnalysisContext) -> Mapping[str, str]:
+        try:
+            functions = [ctx.get(PyIRAnalysis, function) for function in analyzable_functions(ctx.module)]
+        except (LoweringError, CFGError):
+            return MappingProxyType({})
+        body = next((function for function in functions if function.name == MODULE_BODY), None)
+        if body is None:
+            return MappingProxyType({})
+        defs = {i.result: i for block in body.blocks for i in block.instructions if i.result is not None}
+        stored: dict[str, list[Value | None]] = {}
+        for block in body.blocks:
+            for instruction in block.instructions:
+                if isinstance(instruction, StoreLocal):
+                    stored.setdefault(instruction.name, []).append(instruction.value)
+            if isinstance(block.terminator, ForNext):
+                # A ``for`` binds its target to each item.
+                stored.setdefault(block.terminator.target, []).append(None)
+        rebound = {
+            i.name for function in functions for block in function.blocks for i in block.instructions
+            if isinstance(i, SetGlobal)
+        }
+        rebound |= {s.name for s in ctx.module.body if isinstance(s, nodes.Function | nodes.Class)}
+        found: dict[str, str] = {}
+        for name, values in stored.items():
+            value = values[0] if len(values) == 1 and name not in rebound else None
+            made = defs.get(value) if value is not None else None
+            if isinstance(made, Constant) and isinstance(made.value, str):
+                found[name] = made.value
+        return MappingProxyType(found)

src/coretrace_python/bundled/security/ssrf/plugin.toml

@@ -1,7 +1,7 @@
 name = "ssrf"
 version = "1.0.0"
 plugin_api = ">=1,<2"
-requires = ["taint.flows", "findings.refutation"]
+requires = ["taint.flows", "findings.refutation", "ir.ssa", "abstract.module_strings"]
 provides = ["vulnerability.ssrf"]
 
 [entrypoint]

src/coretrace_python/bundled/security/ssrf/ssrf.py

@@ -1,12 +1,37 @@
-"""Server-side request forgery: attacker-controlled input reaching a SSRF sink."""
+"""Server-side request forgery: attacker-controlled input reaching a SSRF sink.
+
+How the URL is built, with ``+`` and f-strings, can prove part of its destination. Constant
+text fixing ``scheme://host`` and ending the authority with ``/``, ``?`` or ``#`` before
+any input keeps the input from choosing the host; the input still chooses the path, which
+may reach a sensitive resource of that host, and a redirect may lead elsewhere, so the
+flow is a hotspot. With the path fixed too, the input reaching the query or fragment
+only, and redirects disabled in the call, the destination is fixed and the flow is
+refuted. Anything unproven, a base of unknown value or a URL built in a helper function,
+stays a vulnerability.
+"""
 
 from __future__ import annotations
 
+import re
 from typing import ClassVar
 
+from coretrace_python.abstract.strings import ModuleStringsAnalysis, leading_text
+from coretrace_python.analysis import AnyAnalysis
 from coretrace_python.findings import Severity
-from coretrace_python.plugins import TaintDetector
-from coretrace_python.taint import TaintKind
+from coretrace_python.findings.refutation import Status, Verdict
+from coretrace_python.hir import nodes
+from coretrace_python.ir.ssa import SSAAnalysis
+from coretrace_python.plugins import PluginContext, TaintDetector
+from coretrace_python.taint import TaintFlow, TaintKind
+
+# ``scheme://`` and an authority ended by ``/``, ``?`` or ``#``. A backslash, which some
+# URL parsers take for a slash, does not end it here.
+_FIXED_HOST = re.compile(r"[A-Za-z][A-Za-z0-9+.-]*://[^/?#\\]+[/?#]")
+# The same with the path fixed too: what follows is query or fragment.
+_FIXED_PATH = re.compile(r"[A-Za-z][A-Za-z0-9+.-]*://[^/?#\\]+(?:/[^?#]*)?[?#]")
+# A client honours the one of these keywords it knows and rejects the other with a
+# ``TypeError``: either way, the call follows no redirect.
+_NO_REDIRECTS = ("allow_redirects", "follow_redirects")
 
 
 class SsrfPlugin(TaintDetector):
@@ -15,3 +40,25 @@ class SsrfPlugin(TaintDetector):
     kind: ClassVar[TaintKind] = TaintKind.SSRF
     severity: ClassVar[Severity] = Severity.HIGH
     title: ClassVar[str] = "Server-side request forgery"
+    requires: ClassVar[frozenset[AnyAnalysis]] = TaintDetector.requires | {SSAAnalysis, ModuleStringsAnalysis}
+
+    def judge(self, ctx: PluginContext, function: nodes.Function, flow: TaintFlow, verdict: Verdict) -> Verdict:
+        if flow.through is not None:
+            return verdict  # the URL is built in the callee
+        ssa = ctx.get(SSAAnalysis, function)
+        defs = {i.result: i for block in ssa.blocks for i in block.instructions if i.result is not None}
+        text, _ = leading_text(flow.argument, defs, ctx.get(ModuleStringsAnalysis))
+        host = _FIXED_HOST.match(text)
+        if host is None:
+            return verdict
+        arguments = flow.sink_arguments
+        no_redirects = arguments is not None and any(arguments.given(name) == (True, "False") for name in _NO_REDIRECTS)
+        if _FIXED_PATH.match(text) is not None:
+            if no_redirects:
+                return Verdict(flow, Status.REFUTED, f"the destination is fixed by {text!r} and redirects are disabled")
+            left = "a redirect may lead elsewhere"
+        else:
+            left = "the input chooses the path" + ("" if no_redirects else ", and a redirect may lead elsewhere")
+        if verdict.status is not Status.VULNERABILITY:
+            return verdict
+        return Verdict(flow, Status.HOTSPOT, f"the host is fixed by {host.group(0)!r}, but {left}")

src/coretrace_python/engine.py

@@ -17,6 +17,7 @@
 
 from coretrace_python import __version__
 from coretrace_python.abstract import ConstantPropagation, HeapAnalysis, RangeAnalysis
+from coretrace_python.abstract.strings import ModuleStringsAnalysis
 from coretrace_python.analysis import (
     AnalysisContext,
     AnalysisManager,
@@ -143,6 +144,7 @@
     DefUseAnalysis,
     ConstantPropagation,
     RangeAnalysis,
+    ModuleStringsAnalysis,
     HeapAnalysis,
     CallGraphAnalysis,
     SummaryAnalysis,

src/coretrace_python/plugins/detectors.py

@@ -13,13 +13,14 @@
 
 from coretrace_python.analysis import AnyAnalysis
 from coretrace_python.findings import Confidence, Finding, Severity
-from coretrace_python.findings.refutation import RefutationAnalysis, Status
+from coretrace_python.findings.refutation import RefutationAnalysis, Status, Verdict
+from coretrace_python.hir import nodes
 from coretrace_python.interprocedural import CallGraphAnalysis
 from coretrace_python.ir.model import Call, Instruction, Symbol, Value
 from coretrace_python.ir.ssa import SSAAnalysis
 from coretrace_python.plugins.api import Plugin, PluginContext
 from coretrace_python.semantic.symbols import SymbolId
-from coretrace_python.taint import TaintAnalysis, TaintKind
+from coretrace_python.taint import TaintAnalysis, TaintFlow, TaintKind
 
 
 class TaintDetector(Plugin):
@@ -40,6 +41,9 @@ def analyze(self, ctx: PluginContext) -> Sequence[Finding]:
                 if not flow.kinds & self.kind:
                     continue
                 verdict = verdicts.verdict(flow)
+                if verdict.status is Status.REFUTED:
+                    continue
+                verdict = self.judge(ctx, function, flow, verdict)
                 if verdict.status is Status.REFUTED:
                     continue
                 message = f"{self.title}: {flow.source.label} input reaches {flow.sink.symbol}"
@@ -70,6 +74,13 @@ def analyze(self, ctx: PluginContext) -> Sequence[Finding]:
                 )
         return findings
 
+    def judge(self, ctx: PluginContext, function: nodes.Function, flow: TaintFlow, verdict: Verdict) -> Verdict:
+        """This rule's verdict on ``flow``, given the refutation's: the same, unless the rule
+        knows more about its own sinks. The refutation's verdict, which other consumers
+        of the flow read, does not change."""
+
+        return verdict
+
 
 def _lower(confidence: Confidence) -> Confidence:
     return {Confidence.HIGH: Confidence.MEDIUM}.get(confidence, Confidence.LOW)

tests/test_detectors.py

@@ -201,8 +201,11 @@ def test_shipped_plugins_load_with_their_manifests() -> None:
         "xss",
     }
     for rule in ("sql-injection", "command-injection", "code-injection", "nosql-injection", "path-traversal", "ssrf", "xss"):
-        assert by_name[rule].requires == ("taint.flows", "findings.refutation")
         assert by_name[rule].provides == (f"vulnerability.{rule}",)
+    for rule in ("sql-injection", "command-injection", "code-injection", "nosql-injection", "path-traversal", "xss"):
+        assert by_name[rule].requires == ("taint.flows", "findings.refutation")
+    # ``ssrf`` also reads how the URL of each flow is built.
+    assert by_name["ssrf"].requires == ("taint.flows", "findings.refutation", "ir.ssa", "abstract.module_strings")
     assert by_name["python-stdlib-models"].provides == ("model.python-stdlib",)
     assert by_name["weak-crypto"].provides == ("vulnerability.weak-crypto",)