Skip to contentSkip to conversation

syscalls/kcmp03.c: Fix CLONE_IO failure on s390

PATCH New Alexander Gordeev 3 participants 5 messages May 8, 2020 – May 11, 2020 Library code

lore
Share & tools
lore
#0 lore raw

cover syscalls/kcmp03.c: Fix CLONE_IO failure on s390

Hi all,

This series fixes the CLONE_IO flag test on s390:

	kcmp03.c:82: FAIL: clone() Failed: EBADF (9)

The reason for the failure is CLONE_IO flag is sign- extended to unsigned long and CLONE_INTO_CGROUP gets set unintentionally. The kernel does its job well and fails the syscall.

The fix is probably wrong, as the right solution needs to accomodate all archs. I however, post it as a short-term solution that might work for other archs as well.

Thanks!

Alexander Gordeev (2):
syscalls/kcmp03.c: Extend clone_type type to unsigned long lib/cloner.c: Use __clone instead of clone for s390

 lib/cloner.c                            | 5 +++++
 testcases/kernel/syscalls/kcmp/kcmp03.c | 2 +-
 2 files changed, 6 insertions(+), 1 deletion(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/2 syscalls/kcmp03.c: Extend clone_type type to unsigned long

Member clone_type is defined as signed integer type. That is too small to accomodate CLONE_IO (which conflicts with the sign bit) and newer clone flags (i.e CLONE_INTO_CGROUP).

---
 testcases/kernel/syscalls/kcmp/kcmp03.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/kcmp/kcmp03.c +1 -1
index 30ac3ec..f36515f 100644
@@ -31,7 +31,7 @@ static int pid2;
 static void *stack;
 
 static struct tcase {
-	int clone_type;
+	unsigned long clone_type;
 	int kcmp_type;
 } tcases[] = {
 	{CLONE_VM, KCMP_VM},
Signature · 1 line
1.8.3.1
#2 lore raw

PATCH 2/2 lib/cloner.c: Use __clone instead of clone for s390

In reply to Alexander Gordeev

The kernel sys_clone and s390 glibc entry __clone assume the flags parameter type as unsigned long. However, the glibc clone wrapper used by ltp_clone helper defines flags as signed int. That leads to gcc extending CLONE_IO flag of 0x80000000 value to 0xffffffff80000000.

A quick solution for s390 is to skip clone weak alias and use __clone entry point directly.

---
 lib/cloner.c | 5 +++++
 1 file changed, 5 insertions(+)
lib/cloner.c +5 -0
index 11401f2..1192e4b 100644
@@ -49,6 +49,11 @@ extern int __clone2(int (*fn) (void *arg), void *child_stack_base,
                     size_t child_stack_size, int flags, void *arg,
                     pid_t *parent_tid, void *tls, pid_t *child_tid);
 #endif
+#if defined(__s390__)
+# define clone __clone
+extern int __clone(int (*fn)(void *arg), void *child_stack, unsigned long flags,
+		   void *arg, pid_t *parent_tid, void *tls, pid_t *child_tid);
+#endif
 
 #ifndef CLONE_SUPPORTS_7_ARGS
 # define clone(fn, stack, flags, arg, ptid, tls, ctid) \
Signature · 1 line
1.8.3.1
#3 lore raw

PATCH 2/2 lib/cloner.c: Use __clone instead of clone for s390

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

Quoted 5 lines

The kernel sys_clone and s390 glibc entry __clone assume the flags parameter type as unsigned long. However, the glibc clone wrapper used by ltp_clone helper defines flags as signed int. That leads to gcc extending CLONE_IO flag of 0x80000000 value to 0xffffffff80000000.

Same problem exists on ppc64le. There's a patch proposed to fix this on kernel side:
https://lore.kernel.org/lkml/[email protected]/

Earlier discussion is here:
https://sourceware.org/pipermail/libc-alpha/2020-May/113596.html

#4 lore raw

PATCH 2/2 lib/cloner.c: Use __clone instead of clone for s390

Hi,

Quoted 7 lines

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

Quoted 5 lines

The kernel sys_clone and s390 glibc entry __clone assume the flags parameter type as unsigned long. However, the glibc clone wrapper used by ltp_clone helper defines flags as signed int. That leads to gcc extending CLONE_IO flag of 0x80000000 value to 0xffffffff80000000.

Quoted 6 lines

Same problem exists on ppc64le. There's a patch proposed to fix this on kernel side:
https://lore.kernel.org/lkml/[email protected]/

Earlier discussion is here:
https://sourceware.org/pipermail/libc-alpha/2020-May/113596.html

OK instead of merging whole patchset we should add git commit hash once this fix is merged (although no reply since v2 post).

Kind regards,
Petr