fix: harden path validation in reselect_static_kernel_with

Mtest/_inductor/test_npu_static_kernel.py+24-11

	@@ -1,13 +1,10 @@
1	-import os
2	-import subprocess
3	-import datetime
4	-from pathlib import Path
5	import stat	1	import stat
		2	+import unittest
		3	+from pathlib import Path
		4	+
6	import torch_npu	5	import torch_npu
7	-from torch_npu._inductor.config import log	6	+from torch_npu._inductor.npu_static_kernel import StaticKernelCompiler, safe_resolve_output_dir
8	-from torch_npu._inductor.npu_static_kernel import StaticKernelCompiler
9	from torch_npu.testing.testcase import TestCase, run_tests	7	from torch_npu.testing.testcase import TestCase, run_tests
10	-from torch_npu._inductor.npu_static_kernel import safe_resolve_output_dir
11		8
12		9
13	class TestNpuStaticKernel(TestCase):	10	class TestNpuStaticKernel(TestCase):
	@@ -63,9 +60,9 @@ class TestNpuStaticKernel(TestCase):
63	safe_resolve_output_dir(build_path)	60	safe_resolve_output_dir(build_path)
64		61
65	def test_uninstall_static_kernel_no_path(self):	62	def test_uninstall_static_kernel_no_path(self):
66	- from torch_npu._inductor.npu_static_kernel import _uninstall_path, uninstall_static_kernel	63	+ from torch_npu._inductor.npu_static_kernel import uninstall_static_kernel
67	- _uninstall_path = None	64	+ with unittest.mock.patch("torch_npu._inductor.npu_static_kernel._uninstall_path", None):
68	- uninstall_static_kernel()	65	+ uninstall_static_kernel()
69		66
70	def test_static_kernel_compiler_context_manager(self):	67	def test_static_kernel_compiler_context_manager(self):
71	with StaticKernelCompiler() as compiler:	68	with StaticKernelCompiler() as compiler:
	@@ -80,7 +77,6 @@ class TestNpuStaticKernel(TestCase):
80		77
81	def test_safe_resolve_output_dir_absolute_path(self):	78	def test_safe_resolve_output_dir_absolute_path(self):
82	import tempfile	79	import tempfile
83	- import shutil
84	with tempfile.TemporaryDirectory() as tmpdir:	80	with tempfile.TemporaryDirectory() as tmpdir:
85	abs_dir = Path(tmpdir) / "test_build"	81	abs_dir = Path(tmpdir) / "test_build"
86	abs_dir.mkdir()	82	abs_dir.mkdir()
	@@ -88,6 +84,23 @@ class TestNpuStaticKernel(TestCase):
88	self.assertTrue(result.exists())	84	self.assertTrue(result.exists())
89	self.assertIn("kernel_aot_optimization_build_outputs", str(result))	85	self.assertIn("kernel_aot_optimization_build_outputs", str(result))
90		86
		87	+ def test_reselect_static_kernel_with_path_not_string(self):
		88	+ with self.assertRaisesRegex(RuntimeError, "path must be a string"):
		89	+ torch_npu._C._aclnn_reselect_static_kernel_with_path(123)
		90	+
		91	+ def test_reselect_static_kernel_with_path_null_byte(self):
		92	+ with self.assertRaisesRegex(RuntimeError, "null byte"):
		93	+ torch_npu._C._aclnn_reselect_static_kernel_with_path("test\x00dir")
		94	+
		95	+ def test_reselect_static_kernel_with_path_non_existent(self):
		96	+ with self.assertRaisesRegex(RuntimeError, "failed to resolve path"):
		97	+ torch_npu._C._aclnn_reselect_static_kernel_with_path("/nonexistent/path")
		98	+
		99	+ def test_reselect_static_kernel_with_path_is_file(self):
		100	+ import tempfile
		101	+ with tempfile.NamedTemporaryFile() as f:
		102	+ with self.assertRaisesRegex(RuntimeError, "must be a directory"):
		103	+ torch_npu._C._aclnn_reselect_static_kernel_with_path(f.name)
91		104
92	if __name__ == "__main__":	105	if __name__ == "__main__":
93	run_tests()	106	run_tests()

Mtorch_npu/csrc/core/npu/interface/OpInterface.cpp+3-3

	@@ -47,10 +47,10 @@ aclnnStatus ReselectStaticKernelWithPath(const std::string &path)
47	if (aclnnReselectStaticKernelWithPathFunc == nullptr) {	47	if (aclnnReselectStaticKernelWithPathFunc == nullptr) {
48	aclnnReselectStaticKernelWithPathFunc =	48	aclnnReselectStaticKernelWithPathFunc =
49	(AclnnApiFunc)TORCH_NPU_GET_FUNC(aclnnReselectStaticKernelWithPath);	49	(AclnnApiFunc)TORCH_NPU_GET_FUNC(aclnnReselectStaticKernelWithPath);
		50	+ TORCH_CHECK(aclnnReselectStaticKernelWithPathFunc,
		51	+ "Failed to find function ", "aclnnReselectStaticKernelWithPath",
		52	+ PTA_ERROR(ErrCode::NOT_FOUND));
50	}	53	}
51	- TORCH_CHECK(aclnnReselectStaticKernelWithPathFunc,
52	- "Failed to find function ", "aclnnReselectStaticKernelWithPath",
53	- PTA_ERROR(ErrCode::NOT_FOUND));
54	auto ret = aclnnReselectStaticKernelWithPathFunc(path.c_str());	54	auto ret = aclnnReselectStaticKernelWithPathFunc(path.c_str());
55	return ret;	55	return ret;
56	}	56	}

Mtorch_npu/csrc/core/npu/interface/OpInterface.h+1-2

	@@ -18,8 +18,7 @@ bool IsExistAclnnSilentCheck();
18	aclnnStatus ReselectStaticKernel();	18	aclnnStatus ReselectStaticKernel();
19		19
20	/**	20	/**
21	- This Api is used to reselect static kernel with a specified path,	21	+ This Api is used to reselect static kernel with a specified path.
22	- it need to be called once at process.
23	*/	22	*/
24	aclnnStatus ReselectStaticKernelWithPath(const std::string &path);	23	aclnnStatus ReselectStaticKernelWithPath(const std::string &path);
25		24

Mtorch_npu/csrc/npu/Module.cpp+12-8

	@@ -1,5 +1,7 @@
1	#include <chrono>	1	#include <chrono>
		2	+#include <cstdlib>
2	#include <future>	3	#include <future>
		4	+#include <linux/limits.h>
3	#include <sstream>	5	#include <sstream>
4	#include <sys/stat.h>	6	#include <sys/stat.h>
5	#include <thread>	7	#include <thread>
	@@ -2093,13 +2095,15 @@ PyObject* THNPModule_aclnn_reselect_static_kernel_with_path(
2093	TORCH_CHECK(path.find('\0') == std::string::npos,	2095	TORCH_CHECK(path.find('\0') == std::string::npos,
2094	"path must not contain null byte",	2096	"path must not contain null byte",
2095	PTA_ERROR(ErrCode::PARAM));	2097	PTA_ERROR(ErrCode::PARAM));
2096	- struct stat st;	2098	+ char abs_path[PATH_MAX] = {'\0'};
2097	- TORCH_CHECK(stat(path.c_str(), &st) == 0,	2099	+ TORCH_CHECK(realpath(path.c_str(), abs_path) != nullptr,
2098	- "path does not exist: ", path,	2100	+ "failed to resolve path: ", path,
2099	PTA_ERROR(ErrCode::NOT_FOUND));	2101	PTA_ERROR(ErrCode::NOT_FOUND));
2100	- TORCH_CHECK(S_ISDIR(st.st_mode),	2102	+ struct stat st;
2101	- "path must be a directory: ", path,	2103	+ TORCH_CHECK(stat(abs_path, &st) == 0 && S_ISDIR(st.st_mode),
		2104	+ "path must be a directory: ", abs_path,
2102	PTA_ERROR(ErrCode::PARAM));	2105	PTA_ERROR(ErrCode::PARAM));
		2106	+ std::string resolved_path(abs_path);
2103		2107
2104	NPUStatus ret = c10_npu::emptyAllNPUStream();	2108	NPUStatus ret = c10_npu::emptyAllNPUStream();
2105	TORCH_CHECK(	2109	TORCH_CHECK(
	@@ -2111,8 +2115,8 @@ PyObject* THNPModule_aclnn_reselect_static_kernel_with_path(
2111	static const auto task_queue_enable =	2115	static const auto task_queue_enable =
2112	c10_npu::option::OptionsManager::GetTaskQueueEnable();	2116	c10_npu::option::OptionsManager::GetTaskQueueEnable();
2113	if (task_queue_enable == 2) {	2117	if (task_queue_enable == 2) {
2114	- auto acl_call = [path]() -> int {	2118	+ auto acl_call = [resolved_path]() -> int {
2115	- return c10_npu::opapi::ReselectStaticKernelWithPath(path);	2119	+ return c10_npu::opapi::ReselectStaticKernelWithPath(resolved_path);
2116	};	2120	};
2117	at_npu::native::OpCommand::RunOpApiV2("reselect_static_kernel_with_path", acl_call);	2121	at_npu::native::OpCommand::RunOpApiV2("reselect_static_kernel_with_path", acl_call);
2118	NPUStatus ret = c10_npu::emptyAllNPUStream();	2122	NPUStatus ret = c10_npu::emptyAllNPUStream();
	@@ -2122,7 +2126,7 @@ PyObject* THNPModule_aclnn_reselect_static_kernel_with_path(
2122	ret,	2126	ret,
2123	PTA_ERROR(ErrCode::INTERNAL));	2127	PTA_ERROR(ErrCode::INTERNAL));
2124	} else {	2128	} else {
2125	- NPU_CHECK_ERROR(c10_npu::opapi::ReselectStaticKernelWithPath(path));	2129	+ NPU_CHECK_ERROR(c10_npu::opapi::ReselectStaticKernelWithPath(resolved_path));
2126	}	2130	}
2127		2131
2128	Py_RETURN_NONE;	2132	Py_RETURN_NONE;