s390: fix syscall_get_error for compat processes

If both the tracer and the tracee are compat processes, and gprs[2]
is assigned a value by __poke_user_compat, then the higher 32 bits
of gprs[2] are cleared, IS_ERR_VALUE() always returns false, and
syscall_get_error() always returns 0.

Fix the implementation by sign-extending the value for compat processes
the same way as x86 implementation does.

The bug was exposed to user space by commit 201766a20e ("ptrace: add
PTRACE_GET_SYSCALL_INFO request") and detected by strace test suite.

This change fixes strace syscall tampering on s390.

Link: https://lkml.kernel.org/r/20200602180051.GA2427@altlinux.org
Fixes: 753c4dd6a2 ("[S390] ptrace changes")
Cc: Elvira Khabirova <lineprinter@altlinux.org>
Cc: stable@vger.kernel.org # v2.6.28+
Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
This commit is contained in:
Дмитрий Левин 2020-06-02 21:00:51 +03:00 committed by Vasily Gorbik
parent c920c54528
commit b3583fca5f

View File

@ -33,7 +33,17 @@ static inline void syscall_rollback(struct task_struct *task,
static inline long syscall_get_error(struct task_struct *task,
struct pt_regs *regs)
{
return IS_ERR_VALUE(regs->gprs[2]) ? regs->gprs[2] : 0;
unsigned long error = regs->gprs[2];
#ifdef CONFIG_COMPAT
if (test_tsk_thread_flag(task, TIF_31BIT)) {
/*
* Sign-extend the value so (int)-EFOO becomes (long)-EFOO
* and will match correctly in comparisons.
*/
error = (long)(int)error;
}
#endif
return IS_ERR_VALUE(error) ? error : 0;
}
static inline long syscall_get_return_value(struct task_struct *task,