[RFC PATCH] powerpc: Wire up three syscalls

I see that the three syscalls seccomp, getrandom and memfd_create are not wired
because of which we get a warning while compilation.

So I wired them up in this patch. What else needs to be done? I tried the
memfd_test after compiling this kernel, but it is failing. What am I missing for
this to work? Any advice is really appreciated! :)

Signed-off-by: Pranith Kumar <redacted>
---
 arch/powerpc/include/asm/systbl.h      | 3 +++
 arch/powerpc/include/asm/unistd.h      | 2 +-
 arch/powerpc/include/uapi/asm/unistd.h | 3 +++
 3 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/systbl.h b/arch/powerpc/include/asm/systbl.h
index 542bc0f..7d8a600 100644
--- a/arch/powerpc/include/asm/systbl.h
+++ b/arch/powerpc/include/asm/systbl.h

@@ -362,3 +362,6 @@ SYSCALL(ni_syscall) /* sys_kcmp */
 SYSCALL_SPU(sched_setattr)
 SYSCALL_SPU(sched_getattr)
 SYSCALL_SPU(renameat2)
+SYSCALL_SPU(seccomp)
+SYSCALL_SPU(getrandom)
+SYSCALL_SPU(memfd_create)

diff --git a/arch/powerpc/include/asm/unistd.h b/arch/powerpc/include/asm/unistd.h
index 5ce5552..4e9af3f 100644
--- a/arch/powerpc/include/asm/unistd.h
+++ b/arch/powerpc/include/asm/unistd.h

@@ -12,7 +12,7 @@
 #include <uapi/asm/unistd.h>
 
 
-#define __NR_syscalls		358
+#define __NR_syscalls		361
 
 #define __NR__exit __NR_exit
 #define NR_syscalls	__NR_syscalls

diff --git a/arch/powerpc/include/uapi/asm/unistd.h b/arch/powerpc/include/uapi/asm/unistd.h
index 2d526f7..0688fc0 100644
--- a/arch/powerpc/include/uapi/asm/unistd.h
+++ b/arch/powerpc/include/uapi/asm/unistd.h

@@ -380,5 +380,8 @@
 #define __NR_sched_setattr	355
 #define __NR_sched_getattr	356
 #define __NR_renameat2		357
+#define __NR_seccomp		358
+#define __NR_getrandom		359
+#define __NR_memfd_create	360
 
 #endif /* _UAPI_ASM_POWERPC_UNISTD_H_ */

-- 
2.1.0

Hi Pranith,

On Sat, Aug 30, 2014 at 5:36 AM, Pranith Kumar [off-list ref] wrote:

I see that the three syscalls seccomp, getrandom and memfd_create are not wired
because of which we get a warning while compilation.

So I wired them up in this patch. What else needs to be done? I tried the
memfd_test after compiling this kernel, but it is failing. What am I missing for
this to work? Any advice is really appreciated! :)

Did it fail due to the (silly) "ifeq ($(ARCH),X86)" checks in
tools/testing/selftests/memfd/Makefile?

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Hi Geert,

On Sun, Aug 31, 2014 at 4:53 AM, Geert Uytterhoeven
[off-list ref] wrote:

Hi Pranith,

On Sat, Aug 30, 2014 at 5:36 AM, Pranith Kumar [off-list ref] wrote:

quoted

I see that the three syscalls seccomp, getrandom and memfd_create are not wired
because of which we get a warning while compilation.

So I wired them up in this patch. What else needs to be done? I tried the
memfd_test after compiling this kernel, but it is failing. What am I missing for
this to work? Any advice is really appreciated! :)

Did it fail due to the (silly) "ifeq ($(ARCH),X86)" checks in
tools/testing/selftests/memfd/Makefile?

I removed that check and compiled memfd_test.c by hand. This is the
following error which I get when I run the test:

$ ./memfd_test
memfd: CREATE
memfd: BASIC
10 != 0 = GET_SEALS(3)
Aborted

This is basically when checking the seals which we already added. It
should return 10 (F_SEAL_SHRINK | F_SEAL_WRITE), instead it is returning 0.

What else needs to be done for this to properly work? I see that for
m68k, you just wired it up like in this patch. Did it work after that?

-- 
Pranith

Hi Pranith,

On Sun, Aug 31, 2014 at 2:52 PM, Pranith Kumar [off-list ref] wrote:

On Sun, Aug 31, 2014 at 4:53 AM, Geert Uytterhoeven
[off-list ref] wrote:

quoted

On Sat, Aug 30, 2014 at 5:36 AM, Pranith Kumar [off-list ref] wrote:

quoted

I see that the three syscalls seccomp, getrandom and memfd_create are not wired
because of which we get a warning while compilation.

So I wired them up in this patch. What else needs to be done? I tried the
memfd_test after compiling this kernel, but it is failing. What am I missing for
this to work? Any advice is really appreciated! :)

Did it fail due to the (silly) "ifeq ($(ARCH),X86)" checks in
tools/testing/selftests/memfd/Makefile?

I removed that check and compiled memfd_test.c by hand. This is the
following error which I get when I run the test:

$ ./memfd_test
memfd: CREATE
memfd: BASIC
10 != 0 = GET_SEALS(3)
Aborted

This is basically when checking the seals which we already added. It
should return 10 (F_SEAL_SHRINK | F_SEAL_WRITE), instead it is returning 0.

So it does fail.

What else needs to be done for this to properly work? I see that for
m68k, you just wired it up like in this patch. Did it work after that?

To be honest, I didn't run any tests.

I had a few spare minutes, so I wanted to give it a try, but "make kselftest"
doesn't work with building in a separate directory (O=), doesn't support
CROSS_COMPILE=, etc... So I gave up (for now).

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Hi

On Sun, Aug 31, 2014 at 2:52 PM, Pranith Kumar [off-list ref] wrote:

Hi Geert,

On Sun, Aug 31, 2014 at 4:53 AM, Geert Uytterhoeven
[off-list ref] wrote:

quoted

Hi Pranith,

On Sat, Aug 30, 2014 at 5:36 AM, Pranith Kumar [off-list ref] wrote:

quoted

I see that the three syscalls seccomp, getrandom and memfd_create are not wired
because of which we get a warning while compilation.

So I wired them up in this patch. What else needs to be done? I tried the
memfd_test after compiling this kernel, but it is failing. What am I missing for
this to work? Any advice is really appreciated! :)

Did it fail due to the (silly) "ifeq ($(ARCH),X86)" checks in
tools/testing/selftests/memfd/Makefile?

I removed that check and compiled memfd_test.c by hand. This is the
following error which I get when I run the test:

$ ./memfd_test
memfd: CREATE
memfd: BASIC
10 != 0 = GET_SEALS(3)
Aborted

This is basically when checking the seals which we already added. It
should return 10 (F_SEAL_SHRINK | F_SEAL_WRITE), instead it is returning 0.

What else needs to be done for this to properly work? I see that for
m68k, you just wired it up like in this patch. Did it work after that?

The only arch-dependent code for memfd_test.c is the syscall invocation:
    memfd_create(const char *name, unsigned int flags);
via glibc as:
    syscall(__NR_memfd_create, name, flags);

Can you debug your test-run (maybe via simple printk() in mm/shmem.c
memfd_create()) and see what's going wrong there?

Thanks
David

On 08/31/2014 10:34 AM, David Herrmann wrote:

The only arch-dependent code for memfd_test.c is the syscall invocation:
    memfd_create(const char *name, unsigned int flags);
via glibc as:
    syscall(__NR_memfd_create, name, flags);

Can you debug your test-run (maybe via simple printk() in mm/shmem.c
memfd_create()) and see what's going wrong there?

Hi David,

I figured out the problem. I am on a 32-bit system and using u64 for flags in fcntl() is the cause of the problem. Will you accept a patch making the test work on 32-bit systems as below?

Thanks!
--
Pranith

From: Pranith Kumar <redacted>
Date: Sun, 31 Aug 2014 13:38:07 -0400
Subject: [PATCH] memfd_test: Make it work on 32-bit systems

This test currently fails on 32-bit systems since we use u64 type to pass the
flags to fcntl.

This commit changes this to use u32 type for flags to fcntl making it work on
32-bit systems.

Signed-off-by: Pranith Kumar <redacted>
---
 tools/testing/selftests/memfd/memfd_test.c | 32 +++++++++++++++---------------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/tools/testing/selftests/memfd/memfd_test.c b/tools/testing/selftests/memfd/memfd_test.c
index 3634c90..77e56ff 100644
--- a/tools/testing/selftests/memfd/memfd_test.c
+++ b/tools/testing/selftests/memfd/memfd_test.c

@@ -59,9 +59,9 @@ static void mfd_fail_new(const char *name, unsigned int flags)
     }
 }
 
-static __u64 mfd_assert_get_seals(int fd)
+static __u32 mfd_assert_get_seals(int fd)
 {
-    long r;
+    int r;
 
     r = fcntl(fd, F_GET_SEALS);
     if (r < 0) {

@@ -72,36 +72,36 @@ static __u64 mfd_assert_get_seals(int fd)
     return r;
 }
 
-static void mfd_assert_has_seals(int fd, __u64 seals)
+static void mfd_assert_has_seals(int fd, __u32 seals)
 {
-    __u64 s;
+    __u32 s;
 
     s = mfd_assert_get_seals(fd);
     if (s != seals) {
-        printf("%llu != %llu = GET_SEALS(%d)\n",
-               (unsigned long long)seals, (unsigned long long)s, fd);
+        printf("%lu != %lu = GET_SEALS(%d)\n",
+               (unsigned long)seals, (unsigned long)s, fd);
         abort();
     }
 }
 
-static void mfd_assert_add_seals(int fd, __u64 seals)
+static void mfd_assert_add_seals(int fd, __u32 seals)
 {
-    long r;
-    __u64 s;
+    int r;
+    __u32 s;
 
     s = mfd_assert_get_seals(fd);
     r = fcntl(fd, F_ADD_SEALS, seals);
     if (r < 0) {
-        printf("ADD_SEALS(%d, %llu -> %llu) failed: %m\n",
-               fd, (unsigned long long)s, (unsigned long long)seals);
+        printf("ADD_SEALS(%d, %lu -> %lu) failed: %m\n",
+               fd, (unsigned long)s, (unsigned long)seals);
         abort();
     }
 }
 
-static void mfd_fail_add_seals(int fd, __u64 seals)
+static void mfd_fail_add_seals(int fd, __u32 seals)
 {
-    long r;
-    __u64 s;
+    int r;
+    __u32 s;
 
     r = fcntl(fd, F_GET_SEALS);
     if (r < 0)

@@ -111,8 +111,8 @@ static void mfd_fail_add_seals(int fd, __u64 seals)
 
     r = fcntl(fd, F_ADD_SEALS, seals);
     if (r >= 0) {
-        printf("ADD_SEALS(%d, %llu -> %llu) didn't fail as expected\n",
-               fd, (unsigned long long)s, (unsigned long long)seals);
+        printf("ADD_SEALS(%d, %lu -> %lu) didn't fail as expected\n",
+               fd, (unsigned long)s, (unsigned long)seals);
         abort();
     }
 }

-- 
2.1.0

Hi

On Sun, Aug 31, 2014 at 7:51 PM, Pranith Kumar [off-list ref] wrote:

On 08/31/2014 10:34 AM, David Herrmann wrote:

quoted

The only arch-dependent code for memfd_test.c is the syscall invocation:
    memfd_create(const char *name, unsigned int flags);
via glibc as:
    syscall(__NR_memfd_create, name, flags);

Can you debug your test-run (maybe via simple printk() in mm/shmem.c
memfd_create()) and see what's going wrong there?

Hi David,

I figured out the problem. I am on a 32-bit system and using u64 for flags in fcntl() is the cause of the problem. Will you accept a patch making the test work on 32-bit systems as below?

Thanks!
--
Pranith

From: Pranith Kumar <redacted>
Date: Sun, 31 Aug 2014 13:38:07 -0400
Subject: [PATCH] memfd_test: Make it work on 32-bit systems

This test currently fails on 32-bit systems since we use u64 type to pass the
flags to fcntl.

This commit changes this to use u32 type for flags to fcntl making it work on
32-bit systems.

Nice catch. We changed 'flags' from u64 to "unsigned int" in the last
revision of the series. Patch looks good, but I'd prefer using
"unsigned int" as type, instead of __u32. Just to be consistent with
the syscall interface. The return type of F_GET_SEALS is actually
"int" and the MSB is reserved for signed error codes, so you can
savely use "int r = fcntl(fd, F_GET_SEALS, 0)" in
mfd_assert_get_seals().

Thanks
David

Hi David,

On Mon, Sep 1, 2014 at 7:33 AM, David Herrmann [off-list ref] wrote:

Nice catch. We changed 'flags' from u64 to "unsigned int" in the last
revision of the series. Patch looks good, but I'd prefer using
"unsigned int" as type, instead of __u32. Just to be consistent with
the syscall interface. The return type of F_GET_SEALS is actually
"int" and the MSB is reserved for signed error codes, so you can
savely use "int r = fcntl(fd, F_GET_SEALS, 0)" in
mfd_assert_get_seals().

OK. Should I send a new patch with these changes or do you have one
line up already?

-- 
Pranith

Hi

On Mon, Sep 1, 2014 at 5:21 PM, Pranith Kumar [off-list ref] wrote:

Hi David,

On Mon, Sep 1, 2014 at 7:33 AM, David Herrmann [off-list ref] wrote:

quoted

Nice catch. We changed 'flags' from u64 to "unsigned int" in the last
revision of the series. Patch looks good, but I'd prefer using
"unsigned int" as type, instead of __u32. Just to be consistent with
the syscall interface. The return type of F_GET_SEALS is actually
"int" and the MSB is reserved for signed error codes, so you can
savely use "int r = fcntl(fd, F_GET_SEALS, 0)" in
mfd_assert_get_seals().

OK. Should I send a new patch with these changes or do you have one
line up already?

I'd appreciate if you can resend it.

Btw., the original patch (wire up syscalls) can be applied unchanged.

Thanks
David

On Mon, Sep 1, 2014 at 11:31 AM, David Herrmann [off-list ref] wrote:

Btw., the original patch (wire up syscalls) can be applied unchanged.

Great! Can I use that as an Ack-by? I will send in the patch with
updated changelog.


-- 
Pranith

Hi

On Mon, Sep 1, 2014 at 7:16 PM, Pranith Kumar [off-list ref] wrote:

On Mon, Sep 1, 2014 at 11:31 AM, David Herrmann [off-list ref] wrote:

quoted

Btw., the original patch (wire up syscalls) can be applied unchanged.

Great! Can I use that as an Ack-by? I will send in the patch with
updated changelog.

Sure, go ahead! This syscall patch is:

Reviewed-by: David Herrmann <redacted>

Thanks
David