This test checks that orig_a0 allows a syscall argument to be modified,
and that changing a0 does not change the syscall argument.

Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
---
 tools/testing/selftests/riscv/abi/.gitignore |   1 +
 tools/testing/selftests/riscv/abi/Makefile   |   5 +-
 tools/testing/selftests/riscv/abi/ptrace.c   | 151 +++++++++++++++++++++++++++
 3 files changed, 156 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/riscv/abi/.gitignore b/tools/testing/selftests/riscv/abi/.gitignore
index b38358f91c4d2240ae64892871d9ca98bda1ae58..378c605919a3b3d58eec2701eb7af430cfe315d6 100644
--- a/tools/testing/selftests/riscv/abi/.gitignore
+++ b/tools/testing/selftests/riscv/abi/.gitignore
@@ -1 +1,2 @@
 pointer_masking
+ptrace
diff --git a/tools/testing/selftests/riscv/abi/Makefile b/tools/testing/selftests/riscv/abi/Makefile
index ed82ff9c664e7eb3f760cbab81fb957ff72579c5..3f74d059dfdcbce4d45d8ff618781ccea1419061 100644
--- a/tools/testing/selftests/riscv/abi/Makefile
+++ b/tools/testing/selftests/riscv/abi/Makefile
@@ -2,9 +2,12 @@
 
 CFLAGS += -I$(top_srcdir)/tools/include
 
-TEST_GEN_PROGS := pointer_masking
+TEST_GEN_PROGS := pointer_masking ptrace
 
 include ../../lib.mk
 
 $(OUTPUT)/pointer_masking: pointer_masking.c
 	$(CC) -static -o$@ $(CFLAGS) $(LDFLAGS) $^
+
+$(OUTPUT)/ptrace: ptrace.c
+	$(CC) -static -o$@ $(CFLAGS) $(LDFLAGS) $^
diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
new file mode 100644
index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
--- /dev/null
+++ b/tools/testing/selftests/riscv/abi/ptrace.c
@@ -0,0 +1,151 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <signal.h>
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/ptrace.h>
+#include <sys/stat.h>
+#include <sys/user.h>
+#include <sys/wait.h>
+#include <sys/uio.h>
+#include <linux/elf.h>
+#include <linux/unistd.h>
+#include <linux/ptrace.h>
+#include <asm/ptrace.h>
+
+#include "../../kselftest_harness.h"
+
+#define ORIG_A0_MODIFY      0x01
+#define A0_MODIFY           0x02
+#define A0_OLD              0x03
+#define A0_NEW              0x04
+
+struct a0_regs {
+	__s64 orig_a0;
+	__u64 a0;
+};
+
+#define perr_and_exit(fmt, ...)						\
+	({								\
+		char buf[256];						\
+		snprintf(buf, sizeof(buf), "%s:%d:" fmt ": %m\n",	\
+			__func__, __LINE__, ##__VA_ARGS__);		\
+		ksft_exit_fail_perror(buf);				\
+	})
+
+static inline void resume_and_wait_tracee(pid_t pid, int flag)
+{
+	int status;
+
+	if (ptrace(flag, pid, 0, 0))
+		perr_and_exit("failed to resume the tracee %d\n", pid);
+
+	if (waitpid(pid, &status, 0) != pid)
+		perr_and_exit("failed to wait for the tracee %d\n", pid);
+}
+
+static void ptrace_test(int opt, struct a0_regs *result)
+{
+	int status;
+	pid_t pid;
+	struct user_regs_struct regs;
+	struct iovec iov = {
+		.iov_base = &regs,
+		.iov_len = sizeof(regs),
+	};
+
+	unsigned long orig_a0;
+	struct iovec a0_iov = {
+		.iov_base = &orig_a0,
+		.iov_len = sizeof(orig_a0),
+	};
+	struct ptrace_syscall_info syscall_info_entry, syscall_info_exit;
+
+	pid = fork();
+	if (pid == 0) {
+		/* Mark oneself being traced */
+		long val = ptrace(PTRACE_TRACEME, 0, 0, 0);
+
+		if (val)
+			perr_and_exit("failed to request for tracer to trace me: %ld\n", val);
+
+		kill(getpid(), SIGSTOP);
+
+		/* Perform exit syscall that will be intercepted */
+		exit(A0_OLD);
+	}
+
+	if (pid < 0)
+		ksft_exit_fail_perror("failed to fork");
+
+	if (waitpid(pid, &status, 0) != pid)
+		perr_and_exit("failed to wait for the tracee %d\n", pid);
+
+	/* Stop at the entry point of the syscall */
+	resume_and_wait_tracee(pid, PTRACE_SYSCALL);
+
+	/* Check tracee regs before the syscall */
+	if (ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &iov))
+		perr_and_exit("failed to get tracee registers\n");
+	if (ptrace(PTRACE_GETREGSET, pid, NT_RISCV_ORIG_A0, &a0_iov))
+		perr_and_exit("failed to get tracee registers\n");
+	if (orig_a0 != A0_OLD)
+		perr_and_exit("unexpected orig_a0: 0x%lx\n", orig_a0);
+
+	/* Modify a0/orig_a0 for the syscall */
+	switch (opt) {
+	case A0_MODIFY:
+		regs.a0 = A0_NEW;
+		break;
+	case ORIG_A0_MODIFY:
+		orig_a0 = A0_NEW;
+		break;
+	}
+
+	if (ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &iov))
+		perr_and_exit("failed to set tracee registers\n");
+	if (ptrace(PTRACE_SETREGSET, pid, NT_RISCV_ORIG_A0, &a0_iov))
+		perr_and_exit("failed to set tracee registers\n");
+
+	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
+		perr_and_exit("failed to get syscall info of entry\n");
+	result->orig_a0 = syscall_info_entry->entry.args[0];
+	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
+		perr_and_exit("failed to get syscall info of exit\n");
+	result->a0 = syscall_info_exit->exit.rval;
+
+	/* Resume the tracee */
+	ptrace(PTRACE_CONT, pid, 0, 0);
+	if (waitpid(pid, &status, 0) != pid)
+		perr_and_exit("failed to wait for the tracee\n");
+
+}
+
+TEST(ptrace_modify_a0)
+{
+	struct a0_regs result;
+
+	ptrace_test(A0_MODIFY, &result);
+
+	/* The modification of a0 cannot affect the first argument of the syscall */
+	EXPECT_EQ(A0_OLD, result.orig_a0);
+	EXPECT_EQ(A0_NEW, result.a0);
+}
+
+TEST(ptrace_modify_orig_a0)
+{
+	struct a0_regs result;
+
+	ptrace_test(ORIG_A0_MODIFY, &result);
+
+	/* Only modify orig_a0 to change the first argument of the syscall */
+	EXPECT_EQ(A0_NEW, result.orig_a0);
+	/* a0 will keep default value, orig_a0 or -ENOSYS, depends on internal. */
+	EXPECT_NE(A0_NEW, result.a0);
+}
+
+TEST_HARNESS_MAIN

-- 
2.47.1

On Thu, Dec 26, 2024 at 06:52:52PM +0800, Celeste Liu wrote:
> This test checks that orig_a0 allows a syscall argument to be modified,
> and that changing a0 does not change the syscall argument.
> 
> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
> Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
> Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
> Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
[...]
> diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
> --- /dev/null
> +++ b/tools/testing/selftests/riscv/abi/ptrace.c
[...]
> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
> +		perr_and_exit("failed to get syscall info of entry\n");
> +	result->orig_a0 = syscall_info_entry->entry.args[0];
> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
> +		perr_and_exit("failed to get syscall info of exit\n");
> +	result->a0 = syscall_info_exit->exit.rval;

I'm sorry but this is not how PTRACE_GET_SYSCALL_INFO should be used.

PTRACE_GET_SYSCALL_INFO operation takes a pointer and a size,
and in this example instead of size you pass constants 1 and 2, which
essentially means that both syscall_info_entry->entry.args[0] and
syscall_info_exit->exit.rval are not going to be assigned
and would just contain some garbage from the stack.

Also, PTRACE_GET_SYSCALL_INFO operation returns the number of bytes
available to be written by the kernel, which is always nonzero on any
PTRACE_GET_SYSCALL_INFO-capable kernel.  In other words, this example
will always end up with perr_and_exit() call.

I wonder how this test was tested before the submission.


-- 
ldv

On 2024-12-26 21:35, Dmitry V. Levin wrote:
> On Thu, Dec 26, 2024 at 06:52:52PM +0800, Celeste Liu wrote:
>> This test checks that orig_a0 allows a syscall argument to be modified,
>> and that changing a0 does not change the syscall argument.
>>
>> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
>> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
>> Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
>> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
>> Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
>> Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
> [...]
>> diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
>> new file mode 100644
>> index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
>> --- /dev/null
>> +++ b/tools/testing/selftests/riscv/abi/ptrace.c
> [...]
>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
>> +		perr_and_exit("failed to get syscall info of entry\n");
>> +	result->orig_a0 = syscall_info_entry->entry.args[0];
>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
>> +		perr_and_exit("failed to get syscall info of exit\n");
>> +	result->a0 = syscall_info_exit->exit.rval;
> 
> I'm sorry but this is not how PTRACE_GET_SYSCALL_INFO should be used.
> 
> PTRACE_GET_SYSCALL_INFO operation takes a pointer and a size,
> and in this example instead of size you pass constants 1 and 2, which
> essentially means that both syscall_info_entry->entry.args[0] and
> syscall_info_exit->exit.rval are not going to be assigned
> and would just contain some garbage from the stack.
> 
> Also, PTRACE_GET_SYSCALL_INFO operation returns the number of bytes
> available to be written by the kernel, which is always nonzero on any
> PTRACE_GET_SYSCALL_INFO-capable kernel.  In other words, this example
> will always end up with perr_and_exit() call.
> 
> I wonder how this test was tested before the submission.

Oops... It seems I forget sync the code to test board so it runs with the old code...
The code is completely not tested...
I'm so sorry for my mistake.

I will correct it and test it carefully later...

> 
> 

On Thu, Dec 26, 2024 at 11:21:59PM +0800, Celeste Liu wrote:
> 
> On 2024-12-26 21:35, Dmitry V. Levin wrote:
> > On Thu, Dec 26, 2024 at 06:52:52PM +0800, Celeste Liu wrote:
> >> This test checks that orig_a0 allows a syscall argument to be modified,
> >> and that changing a0 does not change the syscall argument.
> >>
> >> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
> >> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
> >> Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
> >> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
> >> Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
> >> Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
> > [...]
> >> diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
> >> new file mode 100644
> >> index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
> >> --- /dev/null
> >> +++ b/tools/testing/selftests/riscv/abi/ptrace.c
> > [...]
> >> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
> >> +		perr_and_exit("failed to get syscall info of entry\n");
> >> +	result->orig_a0 = syscall_info_entry->entry.args[0];
> >> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
> >> +		perr_and_exit("failed to get syscall info of exit\n");
> >> +	result->a0 = syscall_info_exit->exit.rval;
> > 
> > I'm sorry but this is not how PTRACE_GET_SYSCALL_INFO should be used.
> > 
> > PTRACE_GET_SYSCALL_INFO operation takes a pointer and a size,
> > and in this example instead of size you pass constants 1 and 2, which
> > essentially means that both syscall_info_entry->entry.args[0] and
> > syscall_info_exit->exit.rval are not going to be assigned
> > and would just contain some garbage from the stack.
> > 
> > Also, PTRACE_GET_SYSCALL_INFO operation returns the number of bytes
> > available to be written by the kernel, which is always nonzero on any
> > PTRACE_GET_SYSCALL_INFO-capable kernel.  In other words, this example
> > will always end up with perr_and_exit() call.
> > 
> > I wonder how this test was tested before the submission.
> 
> Oops... It seems I forget sync the code to test board so it runs with the old code...
> The code is completely not tested...
> I'm so sorry for my mistake.
> 
> I will correct it and test it carefully later...

It would be great to get this into 6.14. Let me know if you would like
any help!

- Charlie

> 
> > 
> > 
> 

On 2025-01-10 11:34, Charlie Jenkins wrote:
> On Thu, Dec 26, 2024 at 11:21:59PM +0800, Celeste Liu wrote:
>>
>> On 2024-12-26 21:35, Dmitry V. Levin wrote:
>>> On Thu, Dec 26, 2024 at 06:52:52PM +0800, Celeste Liu wrote:
>>>> This test checks that orig_a0 allows a syscall argument to be modified,
>>>> and that changing a0 does not change the syscall argument.
>>>>
>>>> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
>>>> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
>>>> Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
>>>> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
>>>> Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
>>>> Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
>>> [...]
>>>> diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
>>>> new file mode 100644
>>>> index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
>>>> --- /dev/null
>>>> +++ b/tools/testing/selftests/riscv/abi/ptrace.c
>>> [...]
>>>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
>>>> +		perr_and_exit("failed to get syscall info of entry\n");
>>>> +	result->orig_a0 = syscall_info_entry->entry.args[0];
>>>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
>>>> +		perr_and_exit("failed to get syscall info of exit\n");
>>>> +	result->a0 = syscall_info_exit->exit.rval;
>>>
>>> I'm sorry but this is not how PTRACE_GET_SYSCALL_INFO should be used.
>>>
>>> PTRACE_GET_SYSCALL_INFO operation takes a pointer and a size,
>>> and in this example instead of size you pass constants 1 and 2, which
>>> essentially means that both syscall_info_entry->entry.args[0] and
>>> syscall_info_exit->exit.rval are not going to be assigned
>>> and would just contain some garbage from the stack.
>>>
>>> Also, PTRACE_GET_SYSCALL_INFO operation returns the number of bytes
>>> available to be written by the kernel, which is always nonzero on any
>>> PTRACE_GET_SYSCALL_INFO-capable kernel.  In other words, this example
>>> will always end up with perr_and_exit() call.
>>>
>>> I wonder how this test was tested before the submission.
>>
>> Oops... It seems I forget sync the code to test board so it runs with the old code...
>> The code is completely not tested...
>> I'm so sorry for my mistake.
>>
>> I will correct it and test it carefully later...
> 
> It would be great to get this into 6.14. Let me know if you would like
> any help!

v5 has been sent.

Sorry for delay. My test environment was broken yesterday so I have to
spend time to repair it first...

https://lore.kernel.org/lkml/20250115-riscv-new-regset-v5-0-d0e6ec031a23@coelacanthus.name/T/

> 
> - Charlie
> 
>>
>>>
>>>
>>

On 2025-01-10 11:34, Charlie Jenkins wrote:
> On Thu, Dec 26, 2024 at 11:21:59PM +0800, Celeste Liu wrote:
>>
>> On 2024-12-26 21:35, Dmitry V. Levin wrote:
>>> On Thu, Dec 26, 2024 at 06:52:52PM +0800, Celeste Liu wrote:
>>>> This test checks that orig_a0 allows a syscall argument to be modified,
>>>> and that changing a0 does not change the syscall argument.
>>>>
>>>> Co-developed-by: Quan Zhou <zhouquan@iscas.ac.cn>
>>>> Signed-off-by: Quan Zhou <zhouquan@iscas.ac.cn>
>>>> Co-developed-by: Charlie Jenkins <charlie@rivosinc.com>
>>>> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
>>>> Reviewed-by: Björn Töpel <bjorn@rivosinc.com>
>>>> Signed-off-by: Celeste Liu <uwu@coelacanthus.name>
>>> [...]
>>>> diff --git a/tools/testing/selftests/riscv/abi/ptrace.c b/tools/testing/selftests/riscv/abi/ptrace.c
>>>> new file mode 100644
>>>> index 0000000000000000000000000000000000000000..023695352215bb5de3f91c1a6f5ea3b4f9373ff9
>>>> --- /dev/null
>>>> +++ b/tools/testing/selftests/riscv/abi/ptrace.c
>>> [...]
>>>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_ENTRY, &syscall_info_entry))
>>>> +		perr_and_exit("failed to get syscall info of entry\n");
>>>> +	result->orig_a0 = syscall_info_entry->entry.args[0];
>>>> +	if (ptrace(PTRACE_GET_SYSCALL_INFO, pid, PTRACE_SYSCALL_INFO_EXIT, &syscall_info_exit))
>>>> +		perr_and_exit("failed to get syscall info of exit\n");
>>>> +	result->a0 = syscall_info_exit->exit.rval;
>>>
>>> I'm sorry but this is not how PTRACE_GET_SYSCALL_INFO should be used.
>>>
>>> PTRACE_GET_SYSCALL_INFO operation takes a pointer and a size,
>>> and in this example instead of size you pass constants 1 and 2, which
>>> essentially means that both syscall_info_entry->entry.args[0] and
>>> syscall_info_exit->exit.rval are not going to be assigned
>>> and would just contain some garbage from the stack.
>>>
>>> Also, PTRACE_GET_SYSCALL_INFO operation returns the number of bytes
>>> available to be written by the kernel, which is always nonzero on any
>>> PTRACE_GET_SYSCALL_INFO-capable kernel.  In other words, this example
>>> will always end up with perr_and_exit() call.
>>>
>>> I wonder how this test was tested before the submission.
>>
>> Oops... It seems I forget sync the code to test board so it runs with the old code...
>> The code is completely not tested...
>> I'm so sorry for my mistake.
>>
>> I will correct it and test it carefully later...
> 
> It would be great to get this into 6.14. Let me know if you would like
> any help!

Sorry, delayed due to personal reasons. I have returned to this job recently.
The update patchset will be sent tomorrow.

> 
> - Charlie
> 
>>
>>>
>>>
>>