[PATCH v1 1/4] syscalls: Restore address limit after a syscall

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

For example, it would mitigation this bug:

- https://bugs.chromium.org/p/project-zero/issues/detail?id=990

By default, this change warns if the segment is incorrect while
returning to user-mode and fix it. The
CONFIG_VERIFY_PRE_USERMODE_STATE_BUG option can be enabled to halt
instead if needed.

The CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE option is also
added so each architecture can optimize how the
verify_pre_usermode_state function is called.

Signed-off-by: Thomas Garnier <redacted>
---
Based on next-20170308
---
 include/linux/syscalls.h | 19 +++++++++++++++++++
 init/Kconfig             | 16 ++++++++++++++++
 kernel/sys.c             | 11 +++++++++++
 3 files changed, 46 insertions(+)

diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
index 980c3c9b06f8..78a2268ecd6e 100644
--- a/include/linux/syscalls.h
+++ b/include/linux/syscalls.h

@@ -191,6 +191,22 @@ extern struct trace_event_functions exit_syscall_print_funcs;
 	SYSCALL_METADATA(sname, x, __VA_ARGS__)			\
 	__SYSCALL_DEFINEx(x, sname, __VA_ARGS__)
 
+asmlinkage void verify_pre_usermode_state(void);
+
+#ifndef CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
+static inline bool has_user_ds(void) {
+	bool ret = segment_eq(get_fs(), USER_DS);
+	// Prevent re-ordering the call
+	barrier();
+	return ret;
+}
+#else
+static inline bool has_user_ds(void) {
+	return false;
+}
+#endif
+
+
 #define __PROTECT(...) asmlinkage_protect(__VA_ARGS__)
 #define __SYSCALL_DEFINEx(x, name, ...)					\
 	asmlinkage long sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))	\

@@ -199,7 +215,10 @@ extern struct trace_event_functions exit_syscall_print_funcs;
 	asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__));	\
 	asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__))	\
 	{								\
+		bool user_caller = has_user_ds();			\
 		long ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));	\
+		if (user_caller)					\
+			verify_pre_usermode_state();			\
 		__MAP(x,__SC_TEST,__VA_ARGS__);				\
 		__PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__));	\
 		return ret;						\

diff --git a/init/Kconfig b/init/Kconfig
index c859c993c26f..ab958b59063f 100644
--- a/init/Kconfig
+++ b/init/Kconfig

@@ -1929,6 +1929,22 @@ config PROFILING
 config TRACEPOINTS
 	bool
 
+#
+# Set by each architecture that want to optimize how verify_pre_usermode_state
+# is called.
+#
+config ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
+	bool
+
+config VERIFY_PRE_USERMODE_STATE_BUG
+	bool "Halt on incorrect state on returning to user-mode"
+	default n
+	help
+	  By default a warning is logged and the state is fixed. This option
+	  crashes the kernel instead.
+
+	  If unsure, say Y.
+
 source "arch/Kconfig"
 
 endmenu		# General setup

diff --git a/kernel/sys.c b/kernel/sys.c
index 196c7134bee6..cc2ebf7fae55 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c

@@ -2459,3 +2459,14 @@ COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
 	return 0;
 }
 #endif /* CONFIG_COMPAT */
+
+/* Called before coming back to user-mode */
+asmlinkage void verify_pre_usermode_state(void)
+{
+#ifdef CONFIG_VERIFY_PRE_USERMODE_STATE_BUG
+	BUG_ON(!segment_eq(get_fs(), USER_DS));
+#else
+	if (WARN_ON(!segment_eq(get_fs(), USER_DS)))
+		set_fs(USER_DS);
+#endif
+}

-- 
2.12.0.246.ga2ecc84866-goog

On Wed, Mar 08, 2017 at 01:38:43PM -0800, Thomas Garnier wrote:

This really makes the fast exit utterly pointless, and we might as well
rip all that out.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

On Wed, Mar 8, 2017 at 1:38 PM, Thomas Garnier [off-list ref] wrote:

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

For example, it would mitigation this bug:

- https://bugs.chromium.org/p/project-zero/issues/detail?id=990

By default, this change warns if the segment is incorrect while
returning to user-mode and fix it. The
CONFIG_VERIFY_PRE_USERMODE_STATE_BUG option can be enabled to halt
instead if needed.

Instead of this new config, please reuse the CHECK_DATA_CORRUPTION
test instead, which already controls very similar WARN vs BUG
behavior. Example below...

I would just make this:

if (CHECK_DATA_CORRUPTION(!segment_eq(get_fs(), USER_DS))
    set_fs(USER_DS);

-Kees


-- 
Kees Cook
Pixel Security

On Wed, Mar 08, 2017 at 01:38:41PM -0800, Thomas Garnier wrote:

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

I would much rather architectures were given the opportunity to code up
checks like this efficiently (iow, inline in the exit path assembly),
rather than having to unconditionally call an additional function on
every syscall, with its register saving overheads.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

On Wed, Mar 8, 2017 at 1:58 PM, Russell King - ARM Linux
[off-list ref] wrote:

On Wed, Mar 08, 2017 at 01:38:41PM -0800, Thomas Garnier wrote:

quoted

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

I would much rather architectures were given the opportunity to code up
checks like this efficiently (iow, inline in the exit path assembly),
rather than having to unconditionally call an additional function on
every syscall, with its register saving overheads.

Me too.  I think the two config choices should be:

(a) BUG_ON(!segment_eq(...));

(b) No generic check at all -- arch code will handle it

--Andy

That make sense. I will optimize each architecture to not require a call.

On Wed, Mar 8, 2017 at 2:20 PM, Andy Lutomirski [off-list ref] wrote:

On Wed, Mar 8, 2017 at 1:58 PM, Russell King - ARM Linux
[off-list ref] wrote:

quoted

On Wed, Mar 08, 2017 at 01:38:41PM -0800, Thomas Garnier wrote:

quoted

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

I would much rather architectures were given the opportunity to code up
checks like this efficiently (iow, inline in the exit path assembly),
rather than having to unconditionally call an additional function on
every syscall, with its register saving overheads.

Me too.  I think the two config choices should be:

(a) BUG_ON(!segment_eq(...));

(b) No generic check at all -- arch code will handle it

--Andy

-- 
Thomas

Make sense, as discussed on patch number one, I will write custom
assembly for each architecture when BUG_ON is not required. I will try
to use macros to make it easier to read.

On Wed, Mar 8, 2017 at 2:05 PM, Nicolas Pitre [off-list ref] wrote:

On Wed, 8 Mar 2017, Thomas Garnier wrote:

quoted

Implement specific usage of verify_pre_usermode_state for user-mode
returns for arm.
---
Based on next-20170308
---
 arch/arm/Kconfig               | 1 +
 arch/arm/kernel/entry-common.S | 5 +++++
 2 files changed, 6 insertions(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 0d4e71b42c77..704fd8f197fa 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -12,6 +12,7 @@ config ARM
      select ARCH_HAVE_CUSTOM_GPIO_H
      select ARCH_HAS_GCOV_PROFILE_ALL
      select ARCH_MIGHT_HAVE_PC_PARPORT
+     select ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
      select ARCH_OPTIONAL_KERNEL_RWX if ARCH_HAS_STRICT_KERNEL_RWX
      select ARCH_OPTIONAL_KERNEL_RWX_DEFAULT if CPU_V7
      select ARCH_SUPPORTS_ATOMIC_RMW

diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index eb5cd77bf1d8..80cfdc7fabde 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S

@@ -39,6 +39,9 @@
 ret_fast_syscall:
  UNWIND(.fnstart     )
  UNWIND(.cantunwind  )
+     push    {r0}                            @ save returned r0
+     bl      verify_pre_usermode_state
+     pop     {r0}                            @ restore r0
      disable_irq_notrace                     @ disable interrupts
      ldr     r1, [tsk, #TI_FLAGS]            @ re-check for syscall tracing
      tst     r1, #_TIF_SYSCALL_WORK | _TIF_WORK_MASK

This feature is configurable, right?

Here the branch overhead is imposed even if the feature is configured
out. You should consider conditionally defining a macro like some other
features do.

Furthermore I think we still support old toolchains that don't know what
push and pop mean. You should use the legacy syntax instead.


Nicolas

-- 
Thomas

On Wed, 8 Mar 2017, Thomas Garnier wrote:

This feature is configurable, right?

Here the branch overhead is imposed even if the feature is configured 
out. You should consider conditionally defining a macro like some other 
features do.

Furthermore I think we still support old toolchains that don't know what 
push and pop mean. You should use the legacy syntax instead.


Nicolas

Implement specific usage of verify_pre_usermode_state for user-mode
returns for arm64.
---
Based on next-20170308
---
 arch/arm64/Kconfig        | 1 +
 arch/arm64/kernel/entry.S | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 896eba61e5ed..da54774838d8 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig

@@ -24,6 +24,7 @@ config ARM64
 	select ARCH_WANT_COMPAT_IPC_PARSE_VERSION
 	select ARCH_WANT_FRAME_POINTERS
 	select ARCH_HAS_UBSAN_SANITIZE_ALL
+	select ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
 	select ARM_AMBA
 	select ARM_ARCH_TIMER
 	select ARM_GIC

diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
index 43512d4d7df2..b3d1d9e5c7c1 100644
--- a/arch/arm64/kernel/entry.S
+++ b/arch/arm64/kernel/entry.S

@@ -744,6 +744,7 @@ ENDPROC(cpu_switch_to)
 ret_fast_syscall:
 	disable_irq				// disable interrupts
 	str	x0, [sp, #S_X0]			// returned x0
+	bl 	verify_pre_usermode_state
 	ldr	x1, [tsk, #TSK_TI_FLAGS]	// re-check for syscall tracing
 	and	x2, x1, #_TIF_SYSCALL_WORK
 	cbnz	x2, ret_fast_syscall_trace

@@ -771,6 +772,7 @@ work_pending:
  */
 ret_to_user:
 	disable_irq				// disable interrupts
+	bl 	verify_pre_usermode_state
 	ldr	x1, [tsk, #TSK_TI_FLAGS]
 	and	x2, x1, #_TIF_WORK_MASK
 	cbnz	x2, work_pending

-- 
2.12.0.246.ga2ecc84866-goog

Implement specific usage of verify_pre_usermode_state for user-mode
returns for x86.
---
Based on next-20170308
---
 arch/x86/Kconfig          | 1 +
 arch/x86/entry/common.c   | 3 +++
 arch/x86/entry/entry_64.S | 6 ++++++
 3 files changed, 10 insertions(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 005df7c825f5..6d48e18e6f09 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig

@@ -63,6 +63,7 @@ config X86
 	select ARCH_MIGHT_HAVE_ACPI_PDC		if ACPI
 	select ARCH_MIGHT_HAVE_PC_PARPORT
 	select ARCH_MIGHT_HAVE_PC_SERIO
+	select ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
 	select ARCH_SUPPORTS_ATOMIC_RMW
 	select ARCH_SUPPORTS_DEFERRED_STRUCT_PAGE_INIT
 	select ARCH_SUPPORTS_NUMA_BALANCING	if X86_64

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 370c42c7f046..525edbb77f03 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c

@@ -22,6 +22,7 @@
 #include <linux/context_tracking.h>
 #include <linux/user-return-notifier.h>
 #include <linux/uprobes.h>
+#include <linux/syscalls.h>
 
 #include <asm/desc.h>
 #include <asm/traps.h>

@@ -180,6 +181,8 @@ __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
 	struct thread_info *ti = current_thread_info();
 	u32 cached_flags;
 
+	verify_pre_usermode_state();
+
 	if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
 		local_irq_disable();
 

diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index d2b2a2948ffe..795c31eebd09 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S

@@ -201,6 +201,7 @@ entry_SYSCALL_64_fastpath:
 	 * It might end up jumping to the slow path.  If it jumps, RAX
 	 * and all argument registers are clobbered.
 	 */
+
 	call	*sys_call_table(, %rax, 8)
 .Lentry_SYSCALL_64_after_fastpath_call:
 

@@ -218,6 +219,11 @@ entry_SYSCALL_64_fastpath:
 	testl	$_TIF_ALLWORK_MASK, TASK_TI_flags(%r11)
 	jnz	1f
 
+	/*
+	 * Check user-mode state on fast path return, the same check is done
+	 * under the slow path through syscall_return_slowpath.
+	 */
+	call	verify_pre_usermode_state
 	LOCKDEP_SYS_EXIT
 	TRACE_IRQS_ON		/* user mode is traced as IRQs on */
 	movq	RIP(%rsp), %rcx

-- 
2.12.0.246.ga2ecc84866-goog

Implement specific usage of verify_pre_usermode_state for user-mode
returns for arm.
---
Based on next-20170308
---
 arch/arm/Kconfig               | 1 +
 arch/arm/kernel/entry-common.S | 5 +++++
 2 files changed, 6 insertions(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 0d4e71b42c77..704fd8f197fa 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -12,6 +12,7 @@ config ARM
 	select ARCH_HAVE_CUSTOM_GPIO_H
 	select ARCH_HAS_GCOV_PROFILE_ALL
 	select ARCH_MIGHT_HAVE_PC_PARPORT
+	select ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
 	select ARCH_OPTIONAL_KERNEL_RWX if ARCH_HAS_STRICT_KERNEL_RWX
 	select ARCH_OPTIONAL_KERNEL_RWX_DEFAULT if CPU_V7
 	select ARCH_SUPPORTS_ATOMIC_RMW

diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index eb5cd77bf1d8..80cfdc7fabde 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S

@@ -39,6 +39,9 @@
 ret_fast_syscall:
  UNWIND(.fnstart	)
  UNWIND(.cantunwind	)
+	push	{r0}				@ save returned r0
+	bl	verify_pre_usermode_state
+	pop	{r0}				@ restore r0
 	disable_irq_notrace			@ disable interrupts
 	ldr	r1, [tsk, #TI_FLAGS]		@ re-check for syscall tracing
 	tst	r1, #_TIF_SYSCALL_WORK | _TIF_WORK_MASK

@@ -65,6 +68,7 @@ ret_fast_syscall:
  UNWIND(.fnstart	)
  UNWIND(.cantunwind	)
 	str	r0, [sp, #S_R0 + S_OFF]!	@ save returned r0
+	bl	verify_pre_usermode_state
 	disable_irq_notrace			@ disable interrupts
 	ldr	r1, [tsk, #TI_FLAGS]		@ re-check for syscall tracing
 	tst	r1, #_TIF_SYSCALL_WORK | _TIF_WORK_MASK

@@ -263,6 +267,7 @@ __sys_trace:
 
 __sys_trace_return:
 	str	r0, [sp, #S_R0 + S_OFF]!	@ save returned r0
+	bl	verify_pre_usermode_state
 	mov	r0, sp
 	bl	syscall_trace_exit
 	b	ret_slow_syscall

-- 
2.12.0.246.ga2ecc84866-goog

On Wed, Mar 8, 2017 at 1:57 PM, Kees Cook [off-list ref] wrote:

On Wed, Mar 8, 2017 at 1:38 PM, Thomas Garnier [off-list ref] wrote:

quoted

This patch prevents a syscall to modify the address limit of the
caller. The address limit is kept by the syscall wrapper and restored
just after the syscall ends.

For example, it would mitigation this bug:

- https://bugs.chromium.org/p/project-zero/issues/detail?id=990

By default, this change warns if the segment is incorrect while
returning to user-mode and fix it. The
CONFIG_VERIFY_PRE_USERMODE_STATE_BUG option can be enabled to halt
instead if needed.

Instead of this new config, please reuse the CHECK_DATA_CORRUPTION
test instead, which already controls very similar WARN vs BUG
behavior. Example below...

quoted

The CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE option is also
added so each architecture can optimize how the
verify_pre_usermode_state function is called.

Signed-off-by: Thomas Garnier <redacted>
---
Based on next-20170308
---
 include/linux/syscalls.h | 19 +++++++++++++++++++
 init/Kconfig             | 16 ++++++++++++++++
 kernel/sys.c             | 11 +++++++++++
 3 files changed, 46 insertions(+)

diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
index 980c3c9b06f8..78a2268ecd6e 100644
--- a/include/linux/syscalls.h
+++ b/include/linux/syscalls.h

@@ -191,6 +191,22 @@ extern struct trace_event_functions exit_syscall_print_funcs;
        SYSCALL_METADATA(sname, x, __VA_ARGS__)                 \
        __SYSCALL_DEFINEx(x, sname, __VA_ARGS__)

+asmlinkage void verify_pre_usermode_state(void);
+
+#ifndef CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
+static inline bool has_user_ds(void) {
+       bool ret = segment_eq(get_fs(), USER_DS);
+       // Prevent re-ordering the call
+       barrier();
+       return ret;
+}
+#else
+static inline bool has_user_ds(void) {
+       return false;
+}
+#endif
+
+
 #define __PROTECT(...) asmlinkage_protect(__VA_ARGS__)
 #define __SYSCALL_DEFINEx(x, name, ...)                                        \
        asmlinkage long sys##name(__MAP(x,__SC_DECL,__VA_ARGS__))       \

@@ -199,7 +215,10 @@ extern struct trace_event_functions exit_syscall_print_funcs;
        asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__));      \
        asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__))       \
        {                                                               \
+               bool user_caller = has_user_ds();                       \
                long ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));  \
+               if (user_caller)                                        \
+                       verify_pre_usermode_state();                    \
                __MAP(x,__SC_TEST,__VA_ARGS__);                         \
                __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__));       \
                return ret;                                             \

diff --git a/init/Kconfig b/init/Kconfig
index c859c993c26f..ab958b59063f 100644
--- a/init/Kconfig
+++ b/init/Kconfig

@@ -1929,6 +1929,22 @@ config PROFILING
 config TRACEPOINTS
        bool

+#
+# Set by each architecture that want to optimize how verify_pre_usermode_state
+# is called.
+#
+config ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
+       bool
+
+config VERIFY_PRE_USERMODE_STATE_BUG
+       bool "Halt on incorrect state on returning to user-mode"
+       default n
+       help
+         By default a warning is logged and the state is fixed. This option
+         crashes the kernel instead.
+
+         If unsure, say Y.
+
 source "arch/Kconfig"

 endmenu                # General setup

diff --git a/kernel/sys.c b/kernel/sys.c
index 196c7134bee6..cc2ebf7fae55 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c

@@ -2459,3 +2459,14 @@ COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
        return 0;
 }
 #endif /* CONFIG_COMPAT */
+
+/* Called before coming back to user-mode */
+asmlinkage void verify_pre_usermode_state(void)
+{
+#ifdef CONFIG_VERIFY_PRE_USERMODE_STATE_BUG
+       BUG_ON(!segment_eq(get_fs(), USER_DS));
+#else
+       if (WARN_ON(!segment_eq(get_fs(), USER_DS)))
+               set_fs(USER_DS);
+#endif

I would just make this:

if (CHECK_DATA_CORRUPTION(!segment_eq(get_fs(), USER_DS))
    set_fs(USER_DS);

Make sense, I will remove my custom CONFIG and use that one instead
(still doing inline assembly if not set).

-Kees


--
Kees Cook
Pixel Security

-- 
Thomas