kernel: Implement selective syscall userspace redirection

Introduce a mechanism to quickly disable/enable syscall handling for a
specific process and redirect to userspace via SIGSYS.  This is useful
for processes with parts that require syscall redirection and parts that
don't, but who need to perform this boundary crossing really fast,
without paying the cost of a system call to reconfigure syscall handling
on each boundary transition.  This is particularly important for Windows
games running over Wine.

The proposed interface looks like this:

  prctl(PR_SET_SYSCALL_USER_DISPATCH, <op>, <off>, <length>, [selector])

The range [<offset>,<offset>+<length>) is a part of the process memory
map that is allowed to by-pass the redirection code and dispatch
syscalls directly, such that in fast paths a process doesn't need to
disable the trap nor the kernel has to check the selector.  This is
essential to return from SIGSYS to a blocked area without triggering
another SIGSYS from rt_sigreturn.

selector is an optional pointer to a char-sized userspace memory region
that has a key switch for the mechanism. This key switch is set to
either PR_SYS_DISPATCH_ON, PR_SYS_DISPATCH_OFF to enable and disable the
redirection without calling the kernel.

The feature is meant to be set per-thread and it is disabled on
fork/clone/execv.

Internally, this doesn't add overhead to the syscall hot path, and it
requires very little per-architecture support.  I avoided using seccomp,
even though it duplicates some functionality, due to previous feedback
that maybe it shouldn't mix with seccomp since it is not a security
mechanism.  And obviously, this should never be considered a security
mechanism, since any part of the program can by-pass it by using the
syscall dispatcher.

For the sysinfo benchmark, which measures the overhead added to
executing a native syscall that doesn't require interception, the
overhead using only the direct dispatcher region to issue syscalls is
pretty much irrelevant.  The overhead of using the selector goes around
40ns for a native (unredirected) syscall in my system, and it is (as
expected) dominated by the supervisor-mode user-address access.  In
fact, with SMAP off, the overhead is consistently less than 5ns on my
test box.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andy Lutomirski <luto@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20201127193238.821364-4-krisman@collabora.com

fs/exec.c

@@ -64,6 +64,7 @@
 #include <linux/compat.h>
 #include <linux/vmalloc.h>
 #include <linux/io_uring.h>
+#include <linux/syscall_user_dispatch.h>
 
 #include <linux/uaccess.h>
 #include <asm/mmu_context.h>
@@ -1302,6 +1303,8 @@ int begin_new_exec(struct linux_binprm * bprm)
 	flush_thread();
 	me->personality &= ~bprm->per_clear;
 
+	clear_syscall_work_syscall_user_dispatch(me);
+
 	/*
 	 * We have to apply CLOEXEC before we change whether the process is
 	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace

include/linux/sched.h

@@ -28,6 +28,7 @@
 #include <linux/sched/prio.h>
 #include <linux/sched/types.h>
 #include <linux/signal_types.h>
+#include <linux/syscall_user_dispatch.h>
 #include <linux/mm_types_task.h>
 #include <linux/task_io_accounting.h>
 #include <linux/posix-timers.h>
@@ -965,6 +966,7 @@ struct task_struct {
 	unsigned int			sessionid;
 #endif
 	struct seccomp			seccomp;
+	struct syscall_user_dispatch	syscall_dispatch;
 
 	/* Thread group tracking: */
 	u64				parent_exec_id;

include/linux/syscall_user_dispatch.h

@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020 Collabora Ltd.
+ */
+#ifndef _SYSCALL_USER_DISPATCH_H
+#define _SYSCALL_USER_DISPATCH_H
+
+#include <linux/thread_info.h>
+
+#ifdef CONFIG_GENERIC_ENTRY
+
+struct syscall_user_dispatch {
+	char __user	*selector;
+	unsigned long	offset;
+	unsigned long	len;
+	bool		on_dispatch;
+};
+
+int set_syscall_user_dispatch(unsigned long mode, unsigned long offset,
+			      unsigned long len, char __user *selector);
+
+#define clear_syscall_work_syscall_user_dispatch(tsk) \
+	clear_task_syscall_work(tsk, SYSCALL_USER_DISPATCH)
+
+#else
+struct syscall_user_dispatch {};
+
+static inline int set_syscall_user_dispatch(unsigned long mode, unsigned long offset,
+					    unsigned long len, char __user *selector)
+{
+	return -EINVAL;
+}
+
+static inline void clear_syscall_work_syscall_user_dispatch(struct task_struct *tsk)
+{
+}
+
+#endif /* CONFIG_GENERIC_ENTRY */
+
+#endif /* _SYSCALL_USER_DISPATCH_H */

include/linux/thread_info.h

@@ -42,6 +42,7 @@ enum syscall_work_bit {
 	SYSCALL_WORK_BIT_SYSCALL_TRACE,
 	SYSCALL_WORK_BIT_SYSCALL_EMU,
 	SYSCALL_WORK_BIT_SYSCALL_AUDIT,
+	SYSCALL_WORK_BIT_SYSCALL_USER_DISPATCH,
 };
 
 #define SYSCALL_WORK_SECCOMP		BIT(SYSCALL_WORK_BIT_SECCOMP)
@@ -49,6 +50,7 @@ enum syscall_work_bit {
 #define SYSCALL_WORK_SYSCALL_TRACE	BIT(SYSCALL_WORK_BIT_SYSCALL_TRACE)
 #define SYSCALL_WORK_SYSCALL_EMU	BIT(SYSCALL_WORK_BIT_SYSCALL_EMU)
 #define SYSCALL_WORK_SYSCALL_AUDIT	BIT(SYSCALL_WORK_BIT_SYSCALL_AUDIT)
+#define SYSCALL_WORK_SYSCALL_USER_DISPATCH BIT(SYSCALL_WORK_BIT_SYSCALL_USER_DISPATCH)
 #endif
 
 #include <asm/thread_info.h>

include/uapi/linux/prctl.h

@@ -247,4 +247,9 @@ struct prctl_mm_map {
 #define PR_SET_IO_FLUSHER		57
 #define PR_GET_IO_FLUSHER		58
 
+/* Dispatch syscalls to a userspace handler */
+#define PR_SET_SYSCALL_USER_DISPATCH	59
+# define PR_SYS_DISPATCH_OFF		0
+# define PR_SYS_DISPATCH_ON		1
+
 #endif /* _LINUX_PRCTL_H */

kernel/entry/Makefile

@@ -9,5 +9,5 @@ KCOV_INSTRUMENT := n
 CFLAGS_REMOVE_common.o	 = -fstack-protector -fstack-protector-strong
 CFLAGS_common.o		+= -fno-stack-protector
 
-obj-$(CONFIG_GENERIC_ENTRY) 		+= common.o
+obj-$(CONFIG_GENERIC_ENTRY) 		+= common.o syscall_user_dispatch.o
 obj-$(CONFIG_KVM_XFER_TO_GUEST_WORK)	+= kvm.o

kernel/entry/common.h

@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _COMMON_H
+#define _COMMON_H
+
+bool syscall_user_dispatch(struct pt_regs *regs);
+
+#endif

kernel/entry/syscall_user_dispatch.c

@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2020 Collabora Ltd.
+ */
+#include <linux/sched.h>
+#include <linux/prctl.h>
+#include <linux/syscall_user_dispatch.h>
+#include <linux/uaccess.h>
+#include <linux/signal.h>
+#include <linux/elf.h>
+
+#include <linux/sched/signal.h>
+#include <linux/sched/task_stack.h>
+
+#include <asm/syscall.h>
+
+#include "common.h"
+
+static void trigger_sigsys(struct pt_regs *regs)
+{
+	struct kernel_siginfo info;
+
+	clear_siginfo(&info);
+	info.si_signo = SIGSYS;
+	info.si_code = SYS_USER_DISPATCH;
+	info.si_call_addr = (void __user *)KSTK_EIP(current);
+	info.si_errno = 0;
+	info.si_arch = syscall_get_arch(current);
+	info.si_syscall = syscall_get_nr(current, regs);
+
+	force_sig_info(&info);
+}
+
+bool syscall_user_dispatch(struct pt_regs *regs)
+{
+	struct syscall_user_dispatch *sd = &current->syscall_dispatch;
+	char state;
+
+	if (likely(instruction_pointer(regs) - sd->offset < sd->len))
+		return false;
+
+	if (unlikely(arch_syscall_is_vdso_sigreturn(regs)))
+		return false;
+
+	if (likely(sd->selector)) {
+		/*
+		 * access_ok() is performed once, at prctl time, when
+		 * the selector is loaded by userspace.
+		 */
+		if (unlikely(__get_user(state, sd->selector)))
+			do_exit(SIGSEGV);
+
+		if (likely(state == PR_SYS_DISPATCH_OFF))
+			return false;
+
+		if (state != PR_SYS_DISPATCH_ON)
+			do_exit(SIGSYS);
+	}
+
+	sd->on_dispatch = true;
+	syscall_rollback(current, regs);
+	trigger_sigsys(regs);
+
+	return true;
+}
+
+int set_syscall_user_dispatch(unsigned long mode, unsigned long offset,
+			      unsigned long len, char __user *selector)
+{
+	switch (mode) {
+	case PR_SYS_DISPATCH_OFF:
+		if (offset || len || selector)
+			return -EINVAL;
+		break;
+	case PR_SYS_DISPATCH_ON:
+		/*
+		 * Validate the direct dispatcher region just for basic
+		 * sanity against overflow and a 0-sized dispatcher
+		 * region.  If the user is able to submit a syscall from
+		 * an address, that address is obviously valid.
+		 */
+		if (offset && offset + len <= offset)
+			return -EINVAL;
+
+		if (selector && !access_ok(selector, sizeof(*selector)))
+			return -EFAULT;
+
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	current->syscall_dispatch.selector = selector;
+	current->syscall_dispatch.offset = offset;
+	current->syscall_dispatch.len = len;
+	current->syscall_dispatch.on_dispatch = false;
+
+	if (mode == PR_SYS_DISPATCH_ON)
+		set_syscall_work(SYSCALL_USER_DISPATCH);
+	else
+		clear_syscall_work(SYSCALL_USER_DISPATCH);
+
+	return 0;
+}

kernel/fork.c

@@ -906,6 +906,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 	clear_user_return_notifier(tsk);
 	clear_tsk_need_resched(tsk);
 	set_task_stack_end_magic(tsk);
+	clear_syscall_work_syscall_user_dispatch(tsk);
 
 #ifdef CONFIG_STACKPROTECTOR
 	tsk->stack_canary = get_random_canary();

kernel/sys.c

@@ -42,6 +42,7 @@
 #include <linux/syscore_ops.h>
 #include <linux/version.h>
 #include <linux/ctype.h>
+#include <linux/syscall_user_dispatch.h>
 
 #include <linux/compat.h>
 #include <linux/syscalls.h>
@@ -2530,6 +2531,10 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 
 		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
 		break;
+	case PR_SET_SYSCALL_USER_DISPATCH:
+		error = set_syscall_user_dispatch(arg2, arg3, arg4,
+						  (char __user *) arg5);
+		break;
 	default:
 		error = -EINVAL;
 		break;