kunit/fortify: Validate __alloc_size attribute results
PATCH 1/1 kunit/fortify: Validate __alloc_size attribute results
Validate the effect of the __alloc_size attribute on allocators. If the compiler doesn't support __builtin_dynamic_object_size(), skip the test.
---
To pass this depends on the following patches:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
To not be skipped, either GCC 12 or Clang is needed.
--- lib/fortify_kunit.c | 92 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 92 insertions(+)
lib/fortify_kunit.c +92 -0
@@ -16,7 +16,10 @@#define pr_fmt (fmt) KBUILD_MODNAME ": " fmt#include <kunit/test.h> +#include <linux/device.h> #include <linux/string.h> +#include <linux/slab.h> +#include <linux/vmalloc.h> static const char array_of_10[] ="this is 10" ;static const char *ptr_of_11 ="this is 11!" ;
@@ -60,9 +63,98 @@ static void control_flow_split_test(struct kunit *test)KUNIT_EXPECT_EQ (test,want_minus_one (pick),SIZE_MAX ); } +#define check_alloc (alloc, free) do { \ + size_t expected = size; \ + void *p = alloc; \ + KUNIT_EXPECT_TRUE_MSG(test, p != NULL, #alloc " failed?!\n"); \ + KUNIT_EXPECT_EQ_MSG(test, __builtin_dynamic_object_size(p, 1), \ + expected, \ + "__alloc_size() not working with " #alloc "\n"); \ + free; \ +} while (0) + +static volatile size_t unknown_size =50 ; + +static void alloc_size_test (struct kunit *test) +{ +#if !__has_builtin (__builtin_dynamic_object_size) +kunit_skip (test,"Compiler is missing __builtin_dynamic_object_size() support\n" ); +#else +const char device_name[] ="fortify-test" ; +struct device *dev; +gfp_t gfp =GFP_KERNEL | __GFP_NOWARN; +size_t size = unknown_size, prev_size; +void *orig; + +/* kmalloc()-family */ +check_alloc (kmalloc (size++, gfp),kfree (p)); +check_alloc (kmalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kzalloc (size++, gfp),kfree (p)); +check_alloc (kzalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kcalloc (1 , size++, gfp),kfree (p)); +check_alloc (kcalloc_node (1 , size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kmalloc_array (1 , size++, gfp),kfree (p)); +check_alloc (kmalloc_array_node (1 , size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (__kmalloc (size++, gfp),kfree (p)); +check_alloc (__kmalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); + +/* kmemdup() */ + prev_size = size; + size =11 ; +check_alloc (kmemdup ("hello there" , size, gfp),kfree (p)); + size = prev_size +1 ; + +/* krealloc()-family */ + orig =kmalloc (size++, gfp); +check_alloc (krealloc (orig, size++, gfp),kfree (p)); + orig =kmalloc (size++, gfp); +check_alloc (krealloc_array (orig,1 , size++, gfp),kfree (p)); + +/* vmalloc()-family */ +check_alloc (vmalloc (size++),vfree (p)); +check_alloc (vzalloc (size++),vfree (p)); +check_alloc (__vmalloc (size++, gfp),vfree (p)); + +/* kvalloc()-family */ +check_alloc (kvmalloc (size++, gfp),kvfree (p)); +check_alloc (kvmalloc_node (size++, gfp,NUMA_NO_NODE ),kvfree (p)); +check_alloc (kvzalloc (size++, gfp),kvfree (p)); +check_alloc (kvzalloc_node (size++, gfp,NUMA_NO_NODE ),kvfree (p)); +check_alloc (kvcalloc (1 , size++, gfp),kvfree (p)); +check_alloc (kvmalloc_array (1 , size++, gfp),kvfree (p)); + +/* kvrealloc() */ + prev_size = size++; + orig =kvmalloc (prev_size, gfp); +check_alloc (kvrealloc (orig, prev_size, size++, gfp),kfree (p)); + +/* Create dummy device for devm_kmalloc()-family tests. */ + dev =root_device_register (device_name); +KUNIT_ASSERT_FALSE_MSG (test,IS_ERR (dev), +"Cannot register test device\n" ); + +/* devm_kmalloc()-family */ +check_alloc (devm_kmalloc (dev, size++, gfp),devm_kfree (dev, p)); +check_alloc (devm_kzalloc (dev, size++, gfp),devm_kfree (dev, p)); + +/* devm_kmemdup() */ + prev_size = size; + size =4 ; +check_alloc (devm_kmemdup (dev,"Ohai" , size, gfp),devm_kfree (dev, p)); + size = prev_size +1 ; + +/* devm_kremalloc() */ + orig =devm_kmalloc (dev, size++, gfp); +check_alloc (devm_krealloc (dev, orig, size++, gfp),devm_kfree (dev, p)); + +device_unregister (dev); +#endif +} +static struct kunit_case fortify_test_cases[] = {KUNIT_CASE (known_sizes_test),KUNIT_CASE (control_flow_split_test), +KUNIT_CASE (alloc_size_test), {} };
Signature · 1 line
2.34.1
Quoted 4 lines · Kees Cook
Validate the effect of the
__alloc_sizeattribute on allocators. If the compiler doesn't support__builtin_dynamic_object_size(), skip the test.
Quoted 8 lines
Signed-off-by Kees Cook <[email protected]>---
To pass this depends on the following patches:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
To not be skipped, either GCC 12 or Clang is needed.
---
While this _looks_ good, I can't actually get the tests to pass on my machine, with the following all having a __builtin_dynamic_object_size() of -1:
- kmalloc_node(size++, gfp, NUMA_NO_NODE)
- kzalloc(size++, gfp)
- kzalloc_node(size++, gfp, NUMA_NO_NODE)
- kcalloc(1, size++, gfp)
- kcalloc_node(1, size++, gfp, NUMA_NO_NODE)
- kmalloc_array(1, size++, gfp)
- kmalloc_array_node(1, size++, gfp, NUMA_NO_NODE)
I've been using the following command to run the tests:./tools/testing/kunit/kunit.py run --kconfig_add CONFIG_FORTIFY_SOURCE=y
And I've also tried it on x86_64 and arm64 under qemu, with both gcc 12.2.0 and clang 14.0.6-2, with the same failures.
Is there a dependency somewhere I've missed? (I've tried it on the ksefltest/kunit branch, with the mentioned dependencies applied, and also on your for-next/hardening branch, with the missing patches applied.)
Cheers,
-- David
Quoted 120 lines
lib/fortify_kunit.c | 92 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 92 insertions(+)lib/fortify_kunit.c +92
-1@@ -16,7 +16,10 @@#define pr_fmt (fmt) KBUILD_MODNAME ": " fmt#include <kunit/test.h> +#include <linux/device.h> #include <linux/string.h> +#include <linux/slab.h> +#include <linux/vmalloc.h> static const char array_of_10[] ="this is 10" ;static const char *ptr_of_11 ="this is 11!" ;@@ -60,9 +63,98 @@ static void control_flow_split_test(struct kunit *test)KUNIT_EXPECT_EQ (test,want_minus_one (pick),SIZE_MAX ); } +#define check_alloc (alloc, free) do { \ + size_t expected = size; \ + void *p = alloc; \ + KUNIT_EXPECT_TRUE_MSG(test, p != NULL, #alloc " failed?!\n"); \ + KUNIT_EXPECT_EQ_MSG(test, __builtin_dynamic_object_size(p, 1), \ + expected, \ + "__alloc_size() not working with " #alloc "\n"); \ + free; \ +} while (0) + +static volatile size_t unknown_size =50 ; + +static void alloc_size_test (struct kunit *test) +{ +#if !__has_builtin (__builtin_dynamic_object_size) +kunit_skip (test,"Compiler is missing __builtin_dynamic_object_size() support\n" ); +#else +const char device_name[] ="fortify-test" ; +struct device *dev; +gfp_t gfp =GFP_KERNEL | __GFP_NOWARN; +size_t size = unknown_size, prev_size; +void *orig; + +/* kmalloc()-family */ +check_alloc (kmalloc (size++, gfp),kfree (p)); +check_alloc (kmalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kzalloc (size++, gfp),kfree (p)); +check_alloc (kzalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kcalloc (1 , size++, gfp),kfree (p)); +check_alloc (kcalloc_node (1 , size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (kmalloc_array (1 , size++, gfp),kfree (p)); +check_alloc (kmalloc_array_node (1 , size++, gfp,NUMA_NO_NODE ),kfree (p)); +check_alloc (__kmalloc (size++, gfp),kfree (p)); +check_alloc (__kmalloc_node (size++, gfp,NUMA_NO_NODE ),kfree (p)); + +/* kmemdup() */ + prev_size = size; + size =11 ; +check_alloc (kmemdup ("hello there" , size, gfp),kfree (p)); + size = prev_size +1 ; + +/* krealloc()-family */ + orig =kmalloc (size++, gfp); +check_alloc (krealloc (orig, size++, gfp),kfree (p)); + orig =kmalloc (size++, gfp); +check_alloc (krealloc_array (orig,1 , size++, gfp),kfree (p)); + +/* vmalloc()-family */ +check_alloc (vmalloc (size++),vfree (p)); +check_alloc (vzalloc (size++),vfree (p)); +check_alloc (__vmalloc (size++, gfp),vfree (p)); + +/* kvalloc()-family */ +check_alloc (kvmalloc (size++, gfp),kvfree (p)); +check_alloc (kvmalloc_node (size++, gfp,NUMA_NO_NODE ),kvfree (p)); +check_alloc (kvzalloc (size++, gfp),kvfree (p)); +check_alloc (kvzalloc_node (size++, gfp,NUMA_NO_NODE ),kvfree (p)); +check_alloc (kvcalloc (1 , size++, gfp),kvfree (p)); +check_alloc (kvmalloc_array (1 , size++, gfp),kvfree (p)); + +/* kvrealloc() */ + prev_size = size++; + orig =kvmalloc (prev_size, gfp); +check_alloc (kvrealloc (orig, prev_size, size++, gfp),kfree (p)); + +/* Create dummy device for devm_kmalloc()-family tests. */ + dev =root_device_register (device_name); +KUNIT_ASSERT_FALSE_MSG (test,IS_ERR (dev), +"Cannot register test device\n" ); + +/* devm_kmalloc()-family */ +check_alloc (devm_kmalloc (dev, size++, gfp),devm_kfree (dev, p)); +check_alloc (devm_kzalloc (dev, size++, gfp),devm_kfree (dev, p)); + +/* devm_kmemdup() */ + prev_size = size; + size =4 ; +check_alloc (devm_kmemdup (dev,"Ohai" , size, gfp),devm_kfree (dev, p)); + size = prev_size +1 ; + +/* devm_kremalloc() */ + orig =devm_kmalloc (dev, size++, gfp); +check_alloc (devm_krealloc (dev, orig, size++, gfp),devm_kfree (dev, p)); + +device_unregister (dev); +#endif +} +static struct kunit_case fortify_test_cases[] = {KUNIT_CASE (known_sizes_test),KUNIT_CASE (control_flow_split_test), +KUNIT_CASE (alloc_size_test), {} }; --2.34.1
Quoted 28 lines · David Gow
Quoted 4 lines · Kees Cook
Validate the effect of the
__alloc_sizeattribute on allocators. If the compiler doesn't support__builtin_dynamic_object_size(), skip the test.Quoted 8 lines
Signed-off-by Kees Cook <[email protected]>---
To pass this depends on the following patches:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
To not be skipped, either GCC 12 or Clang is needed.
---While this _looks_ good, I can't actually get the tests to pass on my machine, with the following all having a
__builtin_dynamic_object_size()of -1:
-kmalloc_node(size++, gfp, NUMA_NO_NODE)
-kzalloc(size++, gfp)
-kzalloc_node(size++, gfp, NUMA_NO_NODE)
-kcalloc(1, size++, gfp)
-kcalloc_node(1, size++, gfp, NUMA_NO_NODE)
-kmalloc_array(1, size++, gfp)
-kmalloc_array_node(1, size++, gfp, NUMA_NO_NODE)I've been using the following command to run the tests:
./tools/testing/kunit/kunit.pyrun --kconfig_addCONFIG_FORTIFY_SOURCE=y
Quoted 7 lines
And I've also tried it on
x86_64and arm64 under qemu, with both gcc 12.2.0 and clang 14.0.6-2, with the same failures.Is there a dependency somewhere I've missed? (I've tried it on the ksefltest/kunit branch, with the mentioned dependencies applied, and also on your for-next/hardening branch, with the missing patches applied.)
I would expect this to pass with v6.1-rc1 when used with the above two patches added, but it seems those _did_ pass, but not the k*alloc() helpers for you? That is curious. Here's my testing:
$ ./tools/testing/kunit/kunit.py run --arch x86_64 \ --kconfig_add CONFIG_FORTIFY_SOURCE=y --make_options LLVM=1 fortify ... [22:43:32] =================== fortify (3 subtests) =================== [22:43:32] [PASSED] known_sizes_test [22:43:32] [PASSED] control_flow_split_test [22:43:32] [PASSED] alloc_size_test [22:43:32] ===================== [PASSED] fortify ===================== [22:43:32] ============================================================ [22:43:32] Testing complete. Ran 3 tests: passed: 3 [22:43:32] Elapsed time: 33.210s total, 3.369s configuring, 28.367s building, 0.799s running
$ clang --version ClangBuiltLinux clang version 16.0.0 (https://github.com/llvm/llvm-project.git 3291eac12340f465084f347720d99352241f621c)
Signature · 1 line
Kees Cook
Quoted 55 lines · Kees Cook
Quoted 28 lines · David Gow
Quoted 4 lines · Kees Cook
Validate the effect of the
__alloc_sizeattribute on allocators. If the compiler doesn't support__builtin_dynamic_object_size(), skip the test.Quoted 8 lines
Signed-off-by Kees Cook <[email protected]>---
To pass this depends on the following patches:
https://lore.kernel.org/lkml/[email protected]/
https://lore.kernel.org/lkml/[email protected]/
To not be skipped, either GCC 12 or Clang is needed.
---While this _looks_ good, I can't actually get the tests to pass on my machine, with the following all having a
__builtin_dynamic_object_size()of -1:
-kmalloc_node(size++, gfp, NUMA_NO_NODE)
-kzalloc(size++, gfp)
-kzalloc_node(size++, gfp, NUMA_NO_NODE)
-kcalloc(1, size++, gfp)
-kcalloc_node(1, size++, gfp, NUMA_NO_NODE)
-kmalloc_array(1, size++, gfp)
-kmalloc_array_node(1, size++, gfp, NUMA_NO_NODE)I've been using the following command to run the tests:
./tools/testing/kunit/kunit.pyrun --kconfig_addCONFIG_FORTIFY_SOURCE=yQuoted 7 lines
And I've also tried it on
x86_64and arm64 under qemu, with both gcc 12.2.0 and clang 14.0.6-2, with the same failures.Is there a dependency somewhere I've missed? (I've tried it on the ksefltest/kunit branch, with the mentioned dependencies applied, and also on your for-next/hardening branch, with the missing patches applied.)
I would expect this to pass with v6.1-rc1 when used with the above two patches added, but it seems those _did_ pass, but not the k*alloc() helpers for you? That is curious. Here's my testing:
$ ./tools/testing/kunit/kunit.py run --arch x86_64 \ --kconfig_add CONFIG_FORTIFY_SOURCE=y --make_options LLVM=1 fortify ... [22:43:32] =================== fortify (3 subtests) =================== [22:43:32] [PASSED] known_sizes_test [22:43:32] [PASSED] control_flow_split_test [22:43:32] [PASSED] alloc_size_test [22:43:32] ===================== [PASSED] fortify ===================== [22:43:32] ============================================================ [22:43:32] Testing complete. Ran 3 tests: passed: 3 [22:43:32] Elapsed time: 33.210s total, 3.369s configuring, 28.367s building, 0.799s running
Quoted 4 lines
$ clang --version ClangBuiltLinux clang version 16.0.0 (https://github.com/llvm/llvm-project.git 3291eac12340f465084f347720d99352241f621c)
Running the exact same command here gives the following output (spam incoming):
[13:55:34] Configuring KUnit Kernel ...
[13:55:34] Building KUnit Kernel ...
Populating config with:
$ make ARCH=x86_64 O=.kunit olddefconfig LLVM=1 Building with: $ make ARCH=x86_64 O=.kunit --jobs=48 LLVM=1 [13:55:43] Starting KUnit Kernel (1/1)... [13:55:43] ============================================================ Running tests with: $ qemu-system-x86_64 -nodefaults -m 1024 -kernel .kunit/arch/x86/boot/bzImage -append 'kunit.filter_glob=fortify kunit.enable=1 console=ttyS0 kunit_shutdown=reboot' -no-reboot -nographic -serial stdio [13:55:44] =================== fortify (3 subtests) =================== [13:55:44] [PASSED] known_sizes_test [13:55:44] [PASSED] control_flow_split_test [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:91 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 51 [13:55:44] __alloc_size() not working with kmalloc_node(size++, gfp, NUMA_NO_NODE) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:92 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 52 [13:55:44] __alloc_size() not working with kzalloc(size++, gfp) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:93 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 53 [13:55:44] __alloc_size() not working with kzalloc_node(size++, gfp, NUMA_NO_NODE) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:94 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 54 [13:55:44] __alloc_size() not working with kcalloc(1, size++, gfp) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:95 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 55 [13:55:44] __alloc_size() not working with kcalloc_node(1, size++, gfp, NUMA_NO_NODE) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:96 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 56 [13:55:44] __alloc_size() not working with kmalloc_array(1, size++, gfp) [13:55:44] # alloc_size_test: EXPECTATION FAILED at lib/fortify_kunit.c:97 [13:55:44] Expected __builtin_dynamic_object_size(p, 1) == expected, but [13:55:44] __builtin_dynamic_object_size(p, 1) == -1 [13:55:44] expected == 57 [13:55:44] __alloc_size() not working with kmalloc_array_node(1, size++, gfp, NUMA_NO_NODE) [13:55:44] not ok 3 - alloc_size_test [13:55:44] [FAILED] alloc_size_test [13:55:44] # Subtest: fortify [13:55:44] 1..3 [13:55:44] # fortify: pass:2 fail:1 skip:0 total:3 [13:55:44] # Totals: pass:2 fail:1 skip:0 total:3 [13:55:44] not ok 1 - fortify [13:55:44] ===================== [FAILED] fortify ===================== [13:55:44] ============================================================ [13:55:44] Testing complete. Ran 3 tests: passed: 2, failed: 1 [13:55:45] Elapsed time: 10.424s total, 0.002s configuring, 8.950s building, 0.835s running
With:
clang --version
Debian clang version 14.0.6-2
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
Same thing with gcc:
gcc (Debian 12.2.0-1) 12.2.0
Copyright (C) 2022 Free Software Foundation, Inc.
I can also reproduce it on a different machine, running openSUSE Tumbleweed's gcc 12.2 and clang 15.0.2.
It also fails the same way with just the mentioned patches, applied to torvalds/master at aae803b02f92.
Do you have a specific working tree somewhere public I should try?
-- David
Quoted 6 lines · David Gow
[13:55:44] #
alloc_size_test: EXPECTATION FAILED atlib/fortify_kunit.c:91[13:55:44] Expected__builtin_dynamic_object_size(p, 1)== expected, but [13:55:44]__builtin_dynamic_object_size(p, 1)== -1 [13:55:44] expected == 51
[13:55:44]__alloc_size()not working withkmalloc_node(size++, gfp, NUMA_NO_NODE)
So, this turned out to be a rat-hole de-ja-vu. The short version is "I was using Clang 16 where this doesn't manifest", and the long version is "some inline attributes are broke on all versions of GCC[1] and on Clang until version 16". :(
I will send the work-around series I've put together to address it. At the end of the day I now have a WAY more robust set of __alloc_size KUnit tests. :P
-Kees
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96503
Signature · 1 line
Kees Cook