add support for Clang CFI

This change adds support for Clang’s forward-edge Control Flow
Integrity (CFI) checking. With CONFIG_CFI_CLANG, the compiler
injects a runtime check before each indirect function call to ensure
the target is a valid function with the correct static type. This
restricts possible call targets and makes it more difficult for
an attacker to exploit bugs that allow the modification of stored
function pointers. For more details, see:

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

Clang requires CONFIG_LTO_CLANG to be enabled with CFI to gain
visibility to possible call targets. Kernel modules are supported
with Clang’s cross-DSO CFI mode, which allows checking between
independently compiled components.

With CFI enabled, the compiler injects a __cfi_check() function into
the kernel and each module for validating local call targets. For
cross-module calls that cannot be validated locally, the compiler
calls the global __cfi_slowpath_diag() function, which determines
the target module and calls the correct __cfi_check() function. This
patch includes a slowpath implementation that uses __module_address()
to resolve call targets, and with CONFIG_CFI_CLANG_SHADOW enabled, a
shadow map that speeds up module look-ups by ~3x.

Clang implements indirect call checking using jump tables and
offers two methods of generating them. With canonical jump tables,
the compiler renames each address-taken function to <function>.cfi
and points the original symbol to a jump table entry, which passes
__cfi_check() validation. This isn’t compatible with stand-alone
assembly code, which the compiler doesn’t instrument, and would
result in indirect calls to assembly code to fail. Therefore, we
default to using non-canonical jump tables instead, where the compiler
generates a local jump table entry <function>.cfi_jt for each
address-taken function, and replaces all references to the function
with the address of the jump table entry.

Note that because non-canonical jump table addresses are local
to each component, they break cross-module function address
equality. Specifically, the address of a global function will be
different in each module, as it's replaced with the address of a local
jump table entry. If this address is passed to a different module,
it won’t match the address of the same function taken there. This
may break code that relies on comparing addresses passed from other
components.

CFI checking can be disabled in a function with the __nocfi attribute.
Additionally, CFI can be disabled for an entire compilation unit by
filtering out CC_FLAGS_CFI.

By default, CFI failures result in a kernel panic to stop a potential
exploit. CONFIG_CFI_PERMISSIVE enables a permissive mode, where the
kernel prints out a rate-limited warning instead, and allows execution
to continue. This option is helpful for locating type mismatches, but
should only be enabled during development.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20210408182843.1754385-2-samitolvanen@google.com

3 files changed, 376 insertions, 0 deletions

diff --git a/kernel/Makefile b/kernel/Makefile
index 320f1f3941b7..e8a6715f38dc 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -41,6 +41,9 @@ KCSAN_SANITIZE_kcov.o := n
 UBSAN_SANITIZE_kcov.o := n
 CFLAGS_kcov.o := $(call cc-option, -fno-conserve-stack) -fno-stack-protector
 
+# Don't instrument error handlers
+CFLAGS_REMOVE_cfi.o := $(CC_FLAGS_CFI)
+
 obj-y += sched/
 obj-y += locking/
 obj-y += power/
@@ -111,6 +114,7 @@ obj-$(CONFIG_BPF) += bpf/
 obj-$(CONFIG_KCSAN) += kcsan/
 obj-$(CONFIG_SHADOW_CALL_STACK) += scs.o
 obj-$(CONFIG_HAVE_STATIC_CALL_INLINE) += static_call.o
+obj-$(CONFIG_CFI_CLANG) += cfi.o
 
 obj-$(CONFIG_PERF_EVENTS) += events/
 
diff --git a/kernel/cfi.c b/kernel/cfi.c
new file mode 100644
index 000000000000..e17a56639766
--- /dev/null
+++ b/kernel/cfi.c
@@ -0,0 +1,329 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Clang Control Flow Integrity (CFI) error and slowpath handling.
+ *
+ * Copyright (C) 2021 Google LLC
+ */
+
+#include <linux/hardirq.h>
+#include <linux/kallsyms.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/printk.h>
+#include <linux/ratelimit.h>
+#include <linux/rcupdate.h>
+#include <linux/vmalloc.h>
+#include <asm/cacheflush.h>
+#include <asm/set_memory.h>
+
+/* Compiler-defined handler names */
+#ifdef CONFIG_CFI_PERMISSIVE
+#define cfi_failure_handler	__ubsan_handle_cfi_check_fail
+#else
+#define cfi_failure_handler	__ubsan_handle_cfi_check_fail_abort
+#endif
+
+static inline void handle_cfi_failure(void *ptr)
+{
+	if (IS_ENABLED(CONFIG_CFI_PERMISSIVE))
+		WARN_RATELIMIT(1, "CFI failure (target: %pS):\n", ptr);
+	else
+		panic("CFI failure (target: %pS)\n", ptr);
+}
+
+#ifdef CONFIG_MODULES
+#ifdef CONFIG_CFI_CLANG_SHADOW
+/*
+ * Index type. A 16-bit index can address at most (2^16)-2 pages (taking
+ * into account SHADOW_INVALID), i.e. ~256M with 4k pages.
+ */
+typedef u16 shadow_t;
+#define SHADOW_INVALID		((shadow_t)~0UL)
+
+struct cfi_shadow {
+	/* Page index for the beginning of the shadow */
+	unsigned long base;
+	/* An array of __cfi_check locations (as indices to the shadow) */
+	shadow_t shadow[1];
+} __packed;
+
+/*
+ * The shadow covers ~128M from the beginning of the module region. If
+ * the region is larger, we fall back to __module_address for the rest.
+ */
+#define __SHADOW_RANGE		(_UL(SZ_128M) >> PAGE_SHIFT)
+
+/* The in-memory size of struct cfi_shadow, always at least one page */
+#define __SHADOW_PAGES		((__SHADOW_RANGE * sizeof(shadow_t)) >> PAGE_SHIFT)
+#define SHADOW_PAGES		max(1UL, __SHADOW_PAGES)
+#define SHADOW_SIZE		(SHADOW_PAGES << PAGE_SHIFT)
+
+/* The actual size of the shadow array, minus metadata */
+#define SHADOW_ARR_SIZE		(SHADOW_SIZE - offsetof(struct cfi_shadow, shadow))
+#define SHADOW_ARR_SLOTS	(SHADOW_ARR_SIZE / sizeof(shadow_t))
+
+static DEFINE_MUTEX(shadow_update_lock);
+static struct cfi_shadow __rcu *cfi_shadow __read_mostly;
+
+/* Returns the index in the shadow for the given address */
+static inline int ptr_to_shadow(const struct cfi_shadow *s, unsigned long ptr)
+{
+	unsigned long index;
+	unsigned long page = ptr >> PAGE_SHIFT;
+
+	if (unlikely(page < s->base))
+		return -1; /* Outside of module area */
+
+	index = page - s->base;
+
+	if (index >= SHADOW_ARR_SLOTS)
+		return -1; /* Cannot be addressed with shadow */
+
+	return (int)index;
+}
+
+/* Returns the page address for an index in the shadow */
+static inline unsigned long shadow_to_ptr(const struct cfi_shadow *s,
+	int index)
+{
+	if (unlikely(index < 0 || index >= SHADOW_ARR_SLOTS))
+		return 0;
+
+	return (s->base + index) << PAGE_SHIFT;
+}
+
+/* Returns the __cfi_check function address for the given shadow location */
+static inline unsigned long shadow_to_check_fn(const struct cfi_shadow *s,
+	int index)
+{
+	if (unlikely(index < 0 || index >= SHADOW_ARR_SLOTS))
+		return 0;
+
+	if (unlikely(s->shadow[index] == SHADOW_INVALID))
+		return 0;
+
+	/* __cfi_check is always page aligned */
+	return (s->base + s->shadow[index]) << PAGE_SHIFT;
+}
+
+static void prepare_next_shadow(const struct cfi_shadow __rcu *prev,
+		struct cfi_shadow *next)
+{
+	int i, index, check;
+
+	/* Mark everything invalid */
+	memset(next->shadow, 0xFF, SHADOW_ARR_SIZE);
+
+	if (!prev)
+		return; /* No previous shadow */
+
+	/* If the base address didn't change, an update is not needed */
+	if (prev->base == next->base) {
+		memcpy(next->shadow, prev->shadow, SHADOW_ARR_SIZE);
+		return;
+	}
+
+	/* Convert the previous shadow to the new address range */
+	for (i = 0; i < SHADOW_ARR_SLOTS; ++i) {
+		if (prev->shadow[i] == SHADOW_INVALID)
+			continue;
+
+		index = ptr_to_shadow(next, shadow_to_ptr(prev, i));
+		if (index < 0)
+			continue;
+
+		check = ptr_to_shadow(next,
+				shadow_to_check_fn(prev, prev->shadow[i]));
+		if (check < 0)
+			continue;
+
+		next->shadow[index] = (shadow_t)check;
+	}
+}
+
+static void add_module_to_shadow(struct cfi_shadow *s, struct module *mod,
+			unsigned long min_addr, unsigned long max_addr)
+{
+	int check_index;
+	unsigned long check = (unsigned long)mod->cfi_check;
+	unsigned long ptr;
+
+	if (unlikely(!PAGE_ALIGNED(check))) {
+		pr_warn("cfi: not using shadow for module %s\n", mod->name);
+		return;
+	}
+
+	check_index = ptr_to_shadow(s, check);
+	if (check_index < 0)
+		return; /* Module not addressable with shadow */
+
+	/* For each page, store the check function index in the shadow */
+	for (ptr = min_addr; ptr <= max_addr; ptr += PAGE_SIZE) {
+		int index = ptr_to_shadow(s, ptr);
+
+		if (index >= 0) {
+			/* Each page must only contain one module */
+			WARN_ON_ONCE(s->shadow[index] != SHADOW_INVALID);
+			s->shadow[index] = (shadow_t)check_index;
+		}
+	}
+}
+
+static void remove_module_from_shadow(struct cfi_shadow *s, struct module *mod,
+		unsigned long min_addr, unsigned long max_addr)
+{
+	unsigned long ptr;
+
+	for (ptr = min_addr; ptr <= max_addr; ptr += PAGE_SIZE) {
+		int index = ptr_to_shadow(s, ptr);
+
+		if (index >= 0)
+			s->shadow[index] = SHADOW_INVALID;
+	}
+}
+
+typedef void (*update_shadow_fn)(struct cfi_shadow *, struct module *,
+			unsigned long min_addr, unsigned long max_addr);
+
+static void update_shadow(struct module *mod, unsigned long base_addr,
+		update_shadow_fn fn)
+{
+	struct cfi_shadow *prev;
+	struct cfi_shadow *next;
+	unsigned long min_addr, max_addr;
+
+	next = vmalloc(SHADOW_SIZE);
+
+	mutex_lock(&shadow_update_lock);
+	prev = rcu_dereference_protected(cfi_shadow,
+					 mutex_is_locked(&shadow_update_lock));
+
+	if (next) {
+		next->base = base_addr >> PAGE_SHIFT;
+		prepare_next_shadow(prev, next);
+
+		min_addr = (unsigned long)mod->core_layout.base;
+		max_addr = min_addr + mod->core_layout.text_size;
+		fn(next, mod, min_addr & PAGE_MASK, max_addr & PAGE_MASK);
+
+		set_memory_ro((unsigned long)next, SHADOW_PAGES);
+	}
+
+	rcu_assign_pointer(cfi_shadow, next);
+	mutex_unlock(&shadow_update_lock);
+	synchronize_rcu();
+
+	if (prev) {
+		set_memory_rw((unsigned long)prev, SHADOW_PAGES);
+		vfree(prev);
+	}
+}
+
+void cfi_module_add(struct module *mod, unsigned long base_addr)
+{
+	update_shadow(mod, base_addr, add_module_to_shadow);
+}
+
+void cfi_module_remove(struct module *mod, unsigned long base_addr)
+{
+	update_shadow(mod, base_addr, remove_module_from_shadow);
+}
+
+static inline cfi_check_fn ptr_to_check_fn(const struct cfi_shadow __rcu *s,
+	unsigned long ptr)
+{
+	int index;
+
+	if (unlikely(!s))
+		return NULL; /* No shadow available */
+
+	index = ptr_to_shadow(s, ptr);
+	if (index < 0)
+		return NULL; /* Cannot be addressed with shadow */
+
+	return (cfi_check_fn)shadow_to_check_fn(s, index);
+}
+
+static inline cfi_check_fn find_shadow_check_fn(unsigned long ptr)
+{
+	cfi_check_fn fn;
+
+	rcu_read_lock_sched();
+	fn = ptr_to_check_fn(rcu_dereference_sched(cfi_shadow), ptr);
+	rcu_read_unlock_sched();
+
+	return fn;
+}
+
+#else /* !CONFIG_CFI_CLANG_SHADOW */
+
+static inline cfi_check_fn find_shadow_check_fn(unsigned long ptr)
+{
+	return NULL;
+}
+
+#endif /* CONFIG_CFI_CLANG_SHADOW */
+
+static inline cfi_check_fn find_module_check_fn(unsigned long ptr)
+{
+	cfi_check_fn fn = NULL;
+	struct module *mod;
+
+	rcu_read_lock_sched();
+	mod = __module_address(ptr);
+	if (mod)
+		fn = mod->cfi_check;
+	rcu_read_unlock_sched();
+
+	return fn;
+}
+
+static inline cfi_check_fn find_check_fn(unsigned long ptr)
+{
+	cfi_check_fn fn = NULL;
+
+	if (is_kernel_text(ptr))
+		return __cfi_check;
+
+	/*
+	 * Indirect call checks can happen when RCU is not watching. Both
+	 * the shadow and __module_address use RCU, so we need to wake it
+	 * up if necessary.
+	 */
+	RCU_NONIDLE({
+		if (IS_ENABLED(CONFIG_CFI_CLANG_SHADOW))
+			fn = find_shadow_check_fn(ptr);
+
+		if (!fn)
+			fn = find_module_check_fn(ptr);
+	});
+
+	return fn;
+}
+
+void __cfi_slowpath_diag(uint64_t id, void *ptr, void *diag)
+{
+	cfi_check_fn fn = find_check_fn((unsigned long)ptr);
+
+	if (likely(fn))
+		fn(id, ptr, diag);
+	else /* Don't allow unchecked modules */
+		handle_cfi_failure(ptr);
+}
+EXPORT_SYMBOL(__cfi_slowpath_diag);
+
+#else /* !CONFIG_MODULES */
+
+void __cfi_slowpath_diag(uint64_t id, void *ptr, void *diag)
+{
+	handle_cfi_failure(ptr); /* No modules */
+}
+EXPORT_SYMBOL(__cfi_slowpath_diag);
+
+#endif /* CONFIG_MODULES */
+
+void cfi_failure_handler(void *data, void *ptr, void *vtable)
+{
+	handle_cfi_failure(ptr);
+}
+EXPORT_SYMBOL(cfi_failure_handler);
diff --git a/kernel/module.c b/kernel/module.c
index 30479355ab85..20fb004e7d8d 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -2146,6 +2146,8 @@ void __weak module_arch_freeing_init(struct module *mod)
 {
 }
 
+static void cfi_cleanup(struct module *mod);
+
 /* Free a module, remove from lists, etc. */
 static void free_module(struct module *mod)
 {
@@ -2187,6 +2189,9 @@ static void free_module(struct module *mod)
 	synchronize_rcu();
 	mutex_unlock(&module_mutex);
 
+	/* Clean up CFI for the module. */
+	cfi_cleanup(mod);
+
 	/* This may be empty, but that's OK */
 	module_arch_freeing_init(mod);
 	module_memfree(mod->init_layout.base);
@@ -3866,6 +3871,8 @@ static int unknown_module_param_cb(char *param, char *val, const char *modname,
 	return 0;
 }
 
+static void cfi_init(struct module *mod);
+
 /*
  * Allocate and load the module: note that size of section 0 is always
  * zero, and we rely on this for optional sections.
@@ -3997,6 +4004,9 @@ static int load_module(struct load_info *info, const char __user *uargs,
 
 	flush_module_icache(mod);
 
+	/* Setup CFI for the module. */
+	cfi_init(mod);
+
 	/* Now copy in args */
 	mod->args = strndup_user(uargs, ~0UL >> 1);
 	if (IS_ERR(mod->args)) {
@@ -4070,6 +4080,7 @@ static int load_module(struct load_info *info, const char __user *uargs,
 	synchronize_rcu();
 	kfree(mod->args);
  free_arch_cleanup:
+	cfi_cleanup(mod);
 	module_arch_cleanup(mod);
  free_modinfo:
 	free_modinfo(mod);
@@ -4415,6 +4426,38 @@ int module_kallsyms_on_each_symbol(int (*fn)(void *, const char *,
 #endif /* CONFIG_LIVEPATCH */
 #endif /* CONFIG_KALLSYMS */
 
+static void cfi_init(struct module *mod)
+{
+#ifdef CONFIG_CFI_CLANG
+	initcall_t *init;
+	exitcall_t *exit;
+
+	rcu_read_lock_sched();
+	mod->cfi_check = (cfi_check_fn)
+		find_kallsyms_symbol_value(mod, "__cfi_check");
+	init = (initcall_t *)
+		find_kallsyms_symbol_value(mod, "__cfi_jt_init_module");
+	exit = (exitcall_t *)
+		find_kallsyms_symbol_value(mod, "__cfi_jt_cleanup_module");
+	rcu_read_unlock_sched();
+
+	/* Fix init/exit functions to point to the CFI jump table */
+	if (init)
+		mod->init = *init;
+	if (exit)
+		mod->exit = *exit;
+
+	cfi_module_add(mod, module_addr_min);
+#endif
+}
+
+static void cfi_cleanup(struct module *mod)
+{
+#ifdef CONFIG_CFI_CLANG
+	cfi_module_remove(mod, module_addr_min);
+#endif
+}
+
 /* Maximum number of characters written by module_flags() */
 #define MODULE_FLAGS_BUF_SIZE (TAINT_FLAGS_COUNT + 4)