Skip to contentSkip to conversation

Fix prctl02

PATCH Superseded Martin Doucha 4 participants 8 messages Jan 23, 2020 – Jan 24, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Fix prctl02

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Fixed arguments in test case 9 (second PR_SET_NO_NEW_PRIVS)
- Dropped test case 13 (PR_CAP_AMBIENT) because the args are valid
- Dropped test case 14 (PR_GET_SPECULATION_CTRL) because the args are valid
- Fixed test call of prctl() to have all 5 arguments

---
 testcases/kernel/syscalls/prctl/prctl02.c | 14 ++++----------
 1 file changed, 4 insertions(+), 10 deletions(-)
testcases/kernel/syscalls/prctl/prctl02.c +4 -10
index 93f30b54a..7bf3684e5 100644
@@ -23,13 +23,9 @@
  * arg4, arg5 is non-zero.
  * 12) prctl() fails with EINVAL when options is PR_GET_THP_DISABLE & arg2,
  * arg3, arg4, or arg5 is nonzero.
- * 13) prctl() fails with EINVAL when options is PR_CAP_AMBIENT & an unused
- * argument such as arg4 is nonzero.
- * 14) prctl() fails with EINVAL when option is PR_GET_SPECULATION_CTRL and
- * unused arguments is nonzero.
- * 15) prctl() fails with EPERM when option is PR_SET_SECUREBITS and the
+ * 13) prctl() fails with EPERM when option is PR_SET_SECUREBITS and the
  * caller does not have the CAP_SETPCAP capability.
- * 16) prctl() fails with EPERM when option is PR_CAPBSET_DROP and the caller
+ * 14) prctl() fails with EPERM when option is PR_CAPBSET_DROP and the caller
  * does not have the CAP_SETPCAP capability.
  */
 
@@ -87,12 +83,10 @@ static struct tcase {
 	{PR_SET_SECCOMP, &num_2, &strict_addr, EACCES, "PR_SET_SECCOMP"},
 	{PR_SET_TIMING, &num_1, &num_0, EINVAL, "PR_SET_TIMING"},
 	{PR_SET_NO_NEW_PRIVS, &num_0, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
-	{PR_SET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
+	{PR_SET_NO_NEW_PRIVS, &num_1, &num_1, EINVAL, "PR_SET_NO_NEW_PRIVS"},
 	{PR_GET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_GET_NO_NEW_PRIVS"},
 	{PR_SET_THP_DISABLE, &num_0, &num_1, EINVAL, "PR_SET_THP_DISABLE"},
 	{PR_GET_THP_DISABLE, &num_1, &num_1, EINVAL, "PR_GET_THP_DISABLE"},
-	{PR_CAP_AMBIENT, &num_2, &num_1, EINVAL, "PR_CAP_AMBIENT"},
-	{PR_GET_SPECULATION_CTRL, &num_1, &num_0, EINVAL, "PR_GET_SPECULATION_CTRL"},
 	{PR_SET_SECUREBITS, &num_0, &num_0, EPERM, "PR_SET_SECUREBITS"},
 	{PR_CAPBSET_DROP, &num_1, &num_0, EPERM, "PR_CAPBSET_DROP"},
 };
@@ -140,7 +134,7 @@ static void verify_prctl(unsigned int n)
 	break;
 	}
 
-	TEST(prctl(tc->option, *tc->arg2, *tc->arg3));
+	TEST(prctl(tc->option, *tc->arg2, *tc->arg3, 0, 0));
 	if (TST_RET == 0) {
 		tst_res(TFAIL, "prctl() succeeded unexpectedly");
 		return;
Signature · 1 line
2.24.1
#1 lore raw

PATCH 1/1 Fix prctl02

----- Original Message -----

Quoted 8 lines

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Fixed arguments in test case 9 (second PR_SET_NO_NEW_PRIVS)
- Dropped test case 13 (PR_CAP_AMBIENT) because the args are valid
- Dropped test case 14 (PR_GET_SPECULATION_CTRL) because the args are valid

Can we make them invalid and keep the test cases?
For example:
PR_CAP_AMBIENT arg2 could be num_invalid
PR_GET_SPECULATION_CTRL arg3 could be num_1

The rest looks good to me.

#2 lore raw

PATCH v2 1/1 Fix prctl02

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Change num_invalid to ULONG_MAX
- Fix arguments in test case 9, 13 and 14
- Fix test call of prctl() to have all 5 arguments

---

Changes since v1:
- Change num_invalid to ULONG_MAX
- Return removed test cases and fix them instead

 testcases/kernel/syscalls/prctl/prctl02.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
testcases/kernel/syscalls/prctl/prctl02.c +6 -5
index 93f30b54a..ebc0e5060 100644
@@ -41,6 +41,7 @@
 #include <unistd.h>
 #include <stdlib.h>
 #include <stddef.h>
+#include <limits.h>
 #include "config.h"
 #include "lapi/prctl.h"
 #include "lapi/seccomp.h"
@@ -65,7 +66,7 @@ static unsigned long bad_addr;
 static unsigned long num_0;
 static unsigned long num_1 = 1;
 static unsigned long num_2 = 2;
-static unsigned long num_invalid = 999;
+static unsigned long num_invalid = ULONG_MAX;
 static int seccomp_nsup;
 static int nonewprivs_nsup;
 static int thpdisable_nsup;
@@ -87,12 +88,12 @@ static struct tcase {
 	{PR_SET_SECCOMP, &num_2, &strict_addr, EACCES, "PR_SET_SECCOMP"},
 	{PR_SET_TIMING, &num_1, &num_0, EINVAL, "PR_SET_TIMING"},
 	{PR_SET_NO_NEW_PRIVS, &num_0, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
-	{PR_SET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
+	{PR_SET_NO_NEW_PRIVS, &num_1, &num_1, EINVAL, "PR_SET_NO_NEW_PRIVS"},
 	{PR_GET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_GET_NO_NEW_PRIVS"},
 	{PR_SET_THP_DISABLE, &num_0, &num_1, EINVAL, "PR_SET_THP_DISABLE"},
 	{PR_GET_THP_DISABLE, &num_1, &num_1, EINVAL, "PR_GET_THP_DISABLE"},
-	{PR_CAP_AMBIENT, &num_2, &num_1, EINVAL, "PR_CAP_AMBIENT"},
-	{PR_GET_SPECULATION_CTRL, &num_1, &num_0, EINVAL, "PR_GET_SPECULATION_CTRL"},
+	{PR_CAP_AMBIENT, &num_invalid, &num_0, EINVAL, "PR_CAP_AMBIENT"},
+	{PR_GET_SPECULATION_CTRL, &num_0, &num_invalid, EINVAL, "PR_GET_SPECULATION_CTRL"},
 	{PR_SET_SECUREBITS, &num_0, &num_0, EPERM, "PR_SET_SECUREBITS"},
 	{PR_CAPBSET_DROP, &num_1, &num_0, EPERM, "PR_CAPBSET_DROP"},
 };
@@ -140,7 +141,7 @@ static void verify_prctl(unsigned int n)
 	break;
 	}
 
-	TEST(prctl(tc->option, *tc->arg2, *tc->arg3));
+	TEST(prctl(tc->option, *tc->arg2, *tc->arg3, 0, 0));
 	if (TST_RET == 0) {
 		tst_res(TFAIL, "prctl() succeeded unexpectedly");
 		return;
Signature · 1 line
2.24.1
#3 lore raw

PATCH v2 1/1 Fix prctl02

----- Original Message -----

Quoted 11 lines

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Change num_invalid to ULONG_MAX
- Fix arguments in test case 9, 13 and 14
- Fix test call of prctl() to have all 5 arguments

#4 lore raw

PATCH v2 1/1 Fix prctl02

In reply to Martin Doucha

Hi

Quoted 8 lines

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Change num_invalid to ULONG_MAX
- Fix arguments in test case 9, 13 and 14
- Fix test call of prctl() to have all 5 arguments

looks prctl manpages and kernel code, you are right, Thanks for the fix! Feel free to add?

Also, do we should use 5 arguments for other prctl test cases?

Best Regards
Yang Xu

Quoted 59 lines

---

Changes since v1:
- Change num_invalid to ULONG_MAX
- Return removed test cases and fix them instead

  testcases/kernel/syscalls/prctl/prctl02.c | 11 ++++++-----
  1 file changed, 6 insertions(+), 5 deletions(-)
testcases/kernel/syscalls/prctl/prctl02.c +6 -5
index 93f30b54a..ebc0e5060 100644
@@ -41,6 +41,7 @@
  #include <unistd.h>
  #include <stdlib.h>
  #include <stddef.h>
+#include <limits.h>
  #include "config.h"
  #include "lapi/prctl.h"
  #include "lapi/seccomp.h"
@@ -65,7 +66,7 @@ static unsigned long bad_addr;
  static unsigned long num_0;
  static unsigned long num_1 = 1;
  static unsigned long num_2 = 2;
-static unsigned long num_invalid = 999;
+static unsigned long num_invalid = ULONG_MAX;
  static int seccomp_nsup;
  static int nonewprivs_nsup;
  static int thpdisable_nsup;
@@ -87,12 +88,12 @@ static struct tcase {
  	{PR_SET_SECCOMP, &num_2, &strict_addr, EACCES, "PR_SET_SECCOMP"},
  	{PR_SET_TIMING, &num_1, &num_0, EINVAL, "PR_SET_TIMING"},
  	{PR_SET_NO_NEW_PRIVS, &num_0, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
-	{PR_SET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_SET_NO_NEW_PRIVS"},
+	{PR_SET_NO_NEW_PRIVS, &num_1, &num_1, EINVAL, "PR_SET_NO_NEW_PRIVS"},
  	{PR_GET_NO_NEW_PRIVS, &num_1, &num_0, EINVAL, "PR_GET_NO_NEW_PRIVS"},
  	{PR_SET_THP_DISABLE, &num_0, &num_1, EINVAL, "PR_SET_THP_DISABLE"},
  	{PR_GET_THP_DISABLE, &num_1, &num_1, EINVAL, "PR_GET_THP_DISABLE"},
-	{PR_CAP_AMBIENT, &num_2, &num_1, EINVAL, "PR_CAP_AMBIENT"},
-	{PR_GET_SPECULATION_CTRL, &num_1, &num_0, EINVAL, "PR_GET_SPECULATION_CTRL"},
+	{PR_CAP_AMBIENT, &num_invalid, &num_0, EINVAL, "PR_CAP_AMBIENT"},
+	{PR_GET_SPECULATION_CTRL, &num_0, &num_invalid, EINVAL, "PR_GET_SPECULATION_CTRL"},
  	{PR_SET_SECUREBITS, &num_0, &num_0, EPERM, "PR_SET_SECUREBITS"},
  	{PR_CAPBSET_DROP, &num_1, &num_0, EPERM, "PR_CAPBSET_DROP"},
  };
@@ -140,7 +141,7 @@ static void verify_prctl(unsigned int n)
  	break;
  	}
 
-	TEST(prctl(tc->option, *tc->arg2, *tc->arg3));
+	TEST(prctl(tc->option, *tc->arg2, *tc->arg3, 0, 0));
  	if (TST_RET == 0) {
  		tst_res(TFAIL, "prctl() succeeded unexpectedly");
  		return;
#5 lore raw

PATCH v2 1/1 Fix prctl02

Hi!

Quoted 8 lines
Quoted 8 lines

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Change num_invalid to ULONG_MAX
- Fix arguments in test case 9, 13 and 14
- Fix test call of prctl() to have all 5 arguments

Quoted 6 lines

looks prctl manpages and kernel code, you are right, Thanks for the fix! Feel free to add??

Also, do we should use 5 arguments for other prctl test cases?

It depends on the option argument, the PR_CAP_AMBIENT explicitely states that arg4 and arg5 must be set to 0 as well as the PR_GET_SPECULATION_CTRL.

Some of the options explicitely say that some arguments are ignored.

And some, including PR_SET_NO_NEW_PRIVS does not say anyting.

Signature · 2 lines
Cyril Hrubis
[email protected]
#6 lore raw

PATCH v2 1/1 Fix prctl02

Hi!

Quoted 16 lines

Hi!

Quoted 8 lines
Quoted 8 lines

The prctl() system call takes 5 integer arguments but only 3 of them were passed in the test. This means that the system call read random garbage from stack in place of the two missing arguments and failed even on some perfectly valid combinations of arguments on some platforms.

- Change num_invalid to ULONG_MAX
- Fix arguments in test case 9, 13 and 14
- Fix test call of prctl() to have all 5 arguments

Quoted 6 lines

looks prctl manpages and kernel code, you are right, Thanks for the fix! Feel free to add??

Also, do we should use 5 arguments for other prctl test cases?

Quoted 7 lines

It depends on the option argument, the PR_CAP_AMBIENT explicitely states that arg4 and arg5 must be set to 0 as well as the PR_GET_SPECULATION_CTRL.

Some of the options explicitely say that some arguments are ignored.

And some, including PR_SET_NO_NEW_PRIVS does not say anyting.

Thanks for your answer. I see.

Quoted 1 line