FROMLIST: add support for Clang's Shadow Call Stack (SCS)

This change adds generic support for Clang's Shadow Call Stack,
which uses a shadow stack to protect return addresses from being
overwritten by an attacker. Details are available here:

  https://clang.llvm.org/docs/ShadowCallStack.html

Note that security guarantees in the kernel differ from the
ones documented for user space. The kernel must store addresses
of shadow stacks used by other tasks and interrupt handlers in
memory, which means an attacker capable reading and writing
arbitrary memory may be able to locate them and hijack control
flow by modifying shadow stacks that are not currently in use.

Bug: 145210207
Change-Id: Ia5f1650593fa95da4efcf86f84830a20989f161c
(am from https://lore.kernel.org/patchwork/patch/1149054/)
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>

Makefile

@@ -860,6 +860,12 @@ DISABLE_LTO	+= $(DISABLE_CFI)
 export CFI_CFLAGS DISABLE_CFI
 endif
 
+ifdef CONFIG_SHADOW_CALL_STACK
+CC_FLAGS_SCS	:= -fsanitize=shadow-call-stack
+KBUILD_CFLAGS	+= $(CC_FLAGS_SCS)
+export CC_FLAGS_SCS
+endif
+
 # arch Makefile may override CC so keep this after arch Makefile is included
 NOSTDINC_FLAGS += -nostdinc -isystem $(shell $(CC) -print-file-name=include)
 

arch/Kconfig

@@ -554,6 +554,39 @@ config CFI_CLANG_SHADOW
 	  If you select this option, the kernel builds a fast look-up table of
 	  CFI check functions in loaded modules to reduce overhead.
 
+config ARCH_SUPPORTS_SHADOW_CALL_STACK
+	bool
+	help
+	  An architecture should select this if it supports Clang's Shadow
+	  Call Stack, has asm/scs.h, and implements runtime support for shadow
+	  stack switching.
+
+config SHADOW_CALL_STACK
+	bool "Clang Shadow Call Stack"
+	depends on ARCH_SUPPORTS_SHADOW_CALL_STACK
+	help
+	  This option enables Clang's Shadow Call Stack, which uses a
+	  shadow stack to protect function return addresses from being
+	  overwritten by an attacker. More information can be found from
+	  Clang's documentation:
+
+	    https://clang.llvm.org/docs/ShadowCallStack.html
+
+	  Note that security guarantees in the kernel differ from the ones
+	  documented for user space. The kernel must store addresses of shadow
+	  stacks used by other tasks and interrupt handlers in memory, which
+	  means an attacker capable reading and writing arbitrary memory may
+	  be able to locate them and hijack control flow by modifying shadow
+	  stacks that are not currently in use.
+
+config SHADOW_CALL_STACK_VMAP
+	bool "Use virtually mapped shadow call stacks"
+	depends on SHADOW_CALL_STACK
+	help
+	  Use virtually mapped shadow call stacks. Selecting this option
+	  provides better stack exhaustion protection, but increases per-thread
+	  memory consumption as a full page is allocated for each shadow stack.
+
 config HAVE_ARCH_WITHIN_STACK_FRAMES
 	bool
 	help

include/linux/compiler-clang.h

@@ -57,3 +57,9 @@
 
 #define __nocfi		__attribute__((no_sanitize("cfi")))
 #endif
+
+#if __has_feature(shadow_call_stack)
+# define __noscs	__attribute__((__no_sanitize__("shadow-call-stack")))
+#else
+# define __noscs
+#endif

include/linux/compiler_types.h

@@ -144,6 +144,10 @@ struct ftrace_likely_data {
 #define __visible
 #endif
 
+#ifndef __noscs
+# define __noscs
+#endif
+
 /*
  * Assume alignment of return value.
  */

include/linux/scs.h

@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Shadow Call Stack support.
+ *
+ * Copyright (C) 2019 Google LLC
+ */
+
+#ifndef _LINUX_SCS_H
+#define _LINUX_SCS_H
+
+#include <linux/gfp.h>
+#include <linux/sched.h>
+#include <asm/page.h>
+
+#ifdef CONFIG_SHADOW_CALL_STACK
+
+/*
+ * In testing, 1 KiB shadow stack size (i.e. 128 stack frames on a 64-bit
+ * architecture) provided ~40% safety margin on stack usage while keeping
+ * memory allocation overhead reasonable.
+ */
+#define SCS_SIZE	1024UL
+#define GFP_SCS		(GFP_KERNEL | __GFP_ZERO)
+
+/*
+ * A random number outside the kernel's virtual address space to mark the
+ * end of the shadow stack.
+ */
+#define SCS_END_MAGIC	0xaf0194819b1635f6UL
+
+#define task_scs(tsk)	(task_thread_info(tsk)->shadow_call_stack)
+
+static inline void task_set_scs(struct task_struct *tsk, void *s)
+{
+	task_scs(tsk) = s;
+}
+
+extern void scs_init(void);
+extern void scs_task_reset(struct task_struct *tsk);
+extern int scs_prepare(struct task_struct *tsk, int node);
+extern bool scs_corrupted(struct task_struct *tsk);
+extern void scs_release(struct task_struct *tsk);
+
+#else /* CONFIG_SHADOW_CALL_STACK */
+
+#define task_scs(tsk)	NULL
+
+static inline void task_set_scs(struct task_struct *tsk, void *s) {}
+static inline void scs_init(void) {}
+static inline void scs_task_reset(struct task_struct *tsk) {}
+static inline int scs_prepare(struct task_struct *tsk, int node) { return 0; }
+static inline bool scs_corrupted(struct task_struct *tsk) { return false; }
+static inline void scs_release(struct task_struct *tsk) {}
+
+#endif /* CONFIG_SHADOW_CALL_STACK */
+
+#endif /* _LINUX_SCS_H */

init/init_task.c

@@ -10,6 +10,7 @@
 #include <linux/fs.h>
 #include <linux/mm.h>
 #include <linux/audit.h>
+#include <linux/scs.h>
 
 #include <asm/pgtable.h>
 #include <linux/uaccess.h>
@@ -182,6 +183,13 @@ struct task_struct init_task
 };
 EXPORT_SYMBOL(init_task);
 
+#ifdef CONFIG_SHADOW_CALL_STACK
+unsigned long init_shadow_call_stack[SCS_SIZE / sizeof(long)] __init_task_data
+		__aligned(SCS_SIZE) = {
+	[(SCS_SIZE / sizeof(long)) - 1] = SCS_END_MAGIC
+};
+#endif
+
 /*
  * Initial thread structure. Alignment of this is handled by a special
  * linker map entry.

kernel/Makefile

@@ -111,6 +111,7 @@ obj-$(CONFIG_IRQ_WORK) += irq_work.o
 obj-$(CONFIG_CPU_PM) += cpu_pm.o
 obj-$(CONFIG_BPF) += bpf/
 obj-$(CONFIG_CFI_CLANG) += cfi.o
+obj-$(CONFIG_SHADOW_CALL_STACK) += scs.o
 
 obj-$(CONFIG_PERF_EVENTS) += events/
 

kernel/fork.c

@@ -94,6 +94,7 @@
 #include <linux/livepatch.h>
 #include <linux/thread_info.h>
 #include <linux/cpufreq_times.h>
+#include <linux/scs.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -398,6 +399,7 @@ void put_task_stack(struct task_struct *tsk)
 void free_task(struct task_struct *tsk)
 {
 	cpufreq_task_times_exit(tsk);
+	scs_release(tsk);
 
 #ifndef CONFIG_THREAD_INFO_IN_TASK
 	/*
@@ -777,6 +779,8 @@ void __init fork_init(void)
 			  NULL, free_vm_stack_cache);
 #endif
 
+	scs_init();
+
 	lockdep_init_task(&init_task);
 }
 
@@ -832,6 +836,10 @@ static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 	if (err)
 		goto free_stack;
 
+	err = scs_prepare(tsk, node);
+	if (err)
+		goto free_stack;
+
 #ifdef CONFIG_SECCOMP
 	/*
 	 * We must handle setting up seccomp filters once we're under

kernel/sched/core.c

@@ -10,6 +10,7 @@
 #include <linux/nospec.h>
 
 #include <linux/kcov.h>
+#include <linux/scs.h>
 
 #include <asm/switch_to.h>
 #include <asm/tlb.h>
@@ -5440,6 +5441,7 @@ void init_idle(struct task_struct *idle, int cpu)
 	idle->se.exec_start = sched_clock();
 	idle->flags |= PF_IDLE;
 
+	scs_task_reset(idle);
 	kasan_unpoison_task_stack(idle);
 
 #ifdef CONFIG_SMP

kernel/scs.c

@@ -0,0 +1,187 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Shadow Call Stack support.
+ *
+ * Copyright (C) 2019 Google LLC
+ */
+
+#include <linux/cpuhotplug.h>
+#include <linux/kasan.h>
+#include <linux/mm.h>
+#include <linux/mmzone.h>
+#include <linux/scs.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <asm/scs.h>
+
+static inline void *__scs_base(struct task_struct *tsk)
+{
+	/*
+	 * To minimize risk the of exposure, architectures may clear a
+	 * task's thread_info::shadow_call_stack while that task is
+	 * running, and only save/restore the active shadow call stack
+	 * pointer when the usual register may be clobbered (e.g. across
+	 * context switches).
+	 *
+	 * The shadow call stack is aligned to SCS_SIZE, and grows
+	 * upwards, so we can mask out the low bits to extract the base
+	 * when the task is not running.
+	 */
+	return (void *)((unsigned long)task_scs(tsk) & ~(SCS_SIZE - 1));
+}
+
+static inline unsigned long *scs_magic(void *s)
+{
+	return (unsigned long *)(s + SCS_SIZE) - 1;
+}
+
+static inline void scs_set_magic(void *s)
+{
+	*scs_magic(s) = SCS_END_MAGIC;
+}
+
+#ifdef CONFIG_SHADOW_CALL_STACK_VMAP
+
+/* Matches NR_CACHED_STACKS for VMAP_STACK */
+#define NR_CACHED_SCS 2
+static DEFINE_PER_CPU(void *, scs_cache[NR_CACHED_SCS]);
+
+static void *scs_alloc(int node)
+{
+	int i;
+	void *s;
+
+	for (i = 0; i < NR_CACHED_SCS; i++) {
+		s = this_cpu_xchg(scs_cache[i], NULL);
+		if (s) {
+			memset(s, 0, SCS_SIZE);
+			goto out;
+		}
+	}
+
+	/*
+	 * We allocate a full page for the shadow stack, which should be
+	 * more than we need. Check the assumption nevertheless.
+	 */
+	BUILD_BUG_ON(SCS_SIZE > PAGE_SIZE);
+
+	s = __vmalloc_node_range(PAGE_SIZE, SCS_SIZE,
+				 VMALLOC_START, VMALLOC_END,
+				 GFP_SCS, PAGE_KERNEL, 0,
+				 node, __builtin_return_address(0));
+
+out:
+	if (s)
+		scs_set_magic(s);
+	/* TODO: poison for KASAN, unpoison in scs_free */
+
+	return s;
+}
+
+static void scs_free(void *s)
+{
+	int i;
+
+	for (i = 0; i < NR_CACHED_SCS; i++)
+		if (this_cpu_cmpxchg(scs_cache[i], 0, s) == NULL)
+			return;
+
+	vfree_atomic(s);
+}
+
+static int scs_cleanup(unsigned int cpu)
+{
+	int i;
+	void **cache = per_cpu_ptr(scs_cache, cpu);
+
+	for (i = 0; i < NR_CACHED_SCS; i++) {
+		vfree(cache[i]);
+		cache[i] = NULL;
+	}
+
+	return 0;
+}
+
+void __init scs_init(void)
+{
+	WARN_ON(cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "scs:scs_cache", NULL,
+			scs_cleanup) < 0);
+}
+
+#else /* !CONFIG_SHADOW_CALL_STACK_VMAP */
+
+static struct kmem_cache *scs_cache;
+
+static inline void *scs_alloc(int node)
+{
+	void *s;
+
+	s = kmem_cache_alloc_node(scs_cache, GFP_SCS, node);
+	if (s) {
+		scs_set_magic(s);
+		/*
+		 * Poison the allocation to catch unintentional accesses to
+		 * the shadow stack when KASAN is enabled.
+		 */
+		kasan_poison_object_data(scs_cache, s);
+	}
+
+	return s;
+}
+
+static inline void scs_free(void *s)
+{
+	kasan_unpoison_object_data(scs_cache, s);
+	kmem_cache_free(scs_cache, s);
+}
+
+void __init scs_init(void)
+{
+	scs_cache = kmem_cache_create("scs_cache", SCS_SIZE, SCS_SIZE,
+				0, NULL);
+	WARN_ON(!scs_cache);
+}
+
+#endif /* CONFIG_SHADOW_CALL_STACK_VMAP */
+
+void scs_task_reset(struct task_struct *tsk)
+{
+	/*
+	 * Reset the shadow stack to the base address in case the task
+	 * is reused.
+	 */
+	task_set_scs(tsk, __scs_base(tsk));
+}
+
+int scs_prepare(struct task_struct *tsk, int node)
+{
+	void *s;
+
+	s = scs_alloc(node);
+	if (!s)
+		return -ENOMEM;
+
+	task_set_scs(tsk, s);
+	return 0;
+}
+
+bool scs_corrupted(struct task_struct *tsk)
+{
+	unsigned long *magic = scs_magic(__scs_base(tsk));
+
+	return READ_ONCE_NOCHECK(*magic) != SCS_END_MAGIC;
+}
+
+void scs_release(struct task_struct *tsk)
+{
+	void *s;
+
+	s = __scs_base(tsk);
+	if (!s)
+		return;
+
+	WARN_ON(scs_corrupted(tsk));
+
+	task_set_scs(tsk, NULL);
+	scs_free(s);
+}