fix: harden import discovery file handling

ingestion/guided_imports.py

@@ -1,5 +1,7 @@
 """Safe discovery of local source documents for the guided import workflow."""
 
+import os
+import stat
 from dataclasses import dataclass
 from hashlib import sha256
 from pathlib import Path
@@ -36,12 +38,27 @@ def _profile_and_sections(path: Path, data: bytes) -> tuple[str | None, tuple[st
     return None, ()
 
 
-def _read_discovered_file(path: Path) -> DiscoveredFile:
+def _unreadable(path: Path) -> DiscoveredFile:
+    return DiscoveredFile(path.name, "", None, (), "UNREADABLE_FILE")
+
+
+def _read_discovered_file(path: Path, expected: os.stat_result) -> DiscoveredFile:
+    descriptor = None
     try:
-        with path.open("rb") as stream:
+        flags = os.O_RDONLY | getattr(os, "O_BINARY", 0)
+        if no_follow := getattr(os, "O_NOFOLLOW", 0):
+            flags |= no_follow
+        descriptor = os.open(path, flags)
+        opened = os.fstat(descriptor)
+        if not stat.S_ISREG(opened.st_mode) or not os.path.samestat(expected, opened):
+            return _unreadable(path)
+        with os.fdopen(descriptor, "rb", closefd=False) as stream:
             data = stream.read(MAX_BYTES + 1)
     except OSError:
-        return DiscoveredFile(path.name, "", None, (), "UNREADABLE_FILE")
+        return _unreadable(path)
+    finally:
+        if descriptor is not None:
+            os.close(descriptor)
     if len(data) > MAX_BYTES:
         return DiscoveredFile(path.name, "", None, (), "INPUT_TOO_LARGE")
     document_hash = sha256(data).hexdigest()
@@ -55,17 +72,19 @@ def _read_discovered_file(path: Path) -> DiscoveredFile:
 
 def discover_imports() -> tuple[DiscoveredFile, ...]:
     """Return safe metadata for direct regular files under ``IMPORT_ROOT``."""
-    root = Path(settings.IMPORT_ROOT).resolve()
     try:
-        entries = root.iterdir()
+        root = Path(settings.IMPORT_ROOT).resolve()
-    except OSError:
+        entries = tuple(root.iterdir())
+    except (OSError, RuntimeError):
         return ()
     discovered = []
     for path in entries:
         try:
-            if path.is_symlink() or not path.is_file():
+            expected = path.lstat()
+            if not stat.S_ISREG(expected.st_mode):
                 continue
         except OSError:
+            discovered.append(_unreadable(path))
             continue
-        discovered.append(_read_discovered_file(path))
+        discovered.append(_read_discovered_file(path, expected))
     return tuple(sorted(discovered, key=lambda row: row.name))

tests/test_guided_import_discovery.py

@@ -1,9 +1,12 @@
 """Synthetic checks for source-safe local import discovery."""
 
+import os
 from hashlib import sha256
+from pathlib import Path
 
 from django.test import override_settings
 
+from ingestion import guided_imports
 from ingestion.guided_imports import discover_imports
 
 
@@ -59,3 +62,66 @@ def test_discovery_ignores_symlinks_and_non_files(tmp_path):
         rows = discover_imports()
 
     assert [row.name for row in rows] == ["target.csv"]
+
+
+def test_discovery_marks_a_post_listing_symlink_replacement_unreadable(tmp_path, monkeypatch):
+    bank = HEADER + b"01/02/2026,,SYNTHETIC BANK ROW,,,12.34\n"
+    candidate = tmp_path / "candidate.csv"
+    replacement = tmp_path / "replacement.csv"
+    candidate.write_bytes(bank)
+    replacement.write_bytes(bank)
+    original_open = os.open
+    replaced = False
+
+    def replace_then_open(value, flags, *args, **kwargs):
+        nonlocal replaced
+        if not replaced and Path(value) == candidate:
+            replaced = True
+            candidate.unlink()
+            candidate.symlink_to(replacement)
+        return original_open(value, flags, *args, **kwargs)
+
+    monkeypatch.setattr(guided_imports, "os", os, raising=False)
+    monkeypatch.setattr(guided_imports.os, "open", replace_then_open)
+
+    with override_settings(IMPORT_ROOT=tmp_path):
+        rows = discover_imports()
+
+    assert [(row.name, row.profile, row.sections, row.code) for row in rows] == [
+        ("candidate.csv", None, (), "UNREADABLE_FILE"),
+        ("replacement.csv", "BANK_CSV_V1", ("1",), None),
+    ]
+
+
+def test_discovery_returns_empty_results_when_listing_the_import_root_fails(tmp_path, monkeypatch):
+    root = tmp_path / "unavailable"
+    root.mkdir()
+    original_iterdir = Path.iterdir
+
+    def delayed_failure(self):
+        if self == root.resolve():
+            def entries():
+                raise OSError("SYNTHETIC LISTING FAILURE")
+                yield
+            return entries()
+        return original_iterdir(self)
+
+    monkeypatch.setattr(Path, "iterdir", delayed_failure)
+
+    with override_settings(IMPORT_ROOT=root):
+        assert discover_imports() == ()
+
+
+def test_discovery_returns_empty_results_when_resolving_the_import_root_fails(tmp_path, monkeypatch):
+    root = tmp_path / "unresolvable"
+    original_resolve = Path.resolve
+
+    def resolution_failure(self, *args, **kwargs):
+        if self == root:
+            raise RuntimeError("SYNTHETIC RESOLUTION FAILURE")
+        return original_resolve(self, *args, **kwargs)
+
+    monkeypatch.setattr(Path, "resolve", resolution_failure)
+
+    with override_settings(IMPORT_ROOT=root):
+        assert discover_imports() == ()