[PATCH v7 3/3] x86/refcount: Implement fast refcount overflow protection

[Kees Cook]

This implements refcount_t overflow protection on x86 without a noticeable
performance impact, though without the fuller checking of REFCOUNT_FULL.
This is done by duplicating the existing atomic_t refcount implementation
but with normally a single instruction added to detect if the refcount
has gone negative (e.g. wrapped past INT_MAX or below zero). When detected,
the handler saturates the refcount_t to INT_MIN / 2. With this overflow
protection, the erroneous reference release that would follow a wrap back
to zero is blocked from happening, avoiding the class of refcount-overflow
use-after-free vulnerabilities entirely.

Only the overflow case of refcounting can be perfectly protected, since
it can be detected and stopped before the reference is freed and left to
be abused by an attacker. There isn't a way to block early decrements,
and while REFCOUNT_FULL stops increment-from-zero cases (which would
be the state _after_ an early decrement and stops potential double-free
conditions), this fast implementation does not, since it would require
the more expensive cmpxchg loops. Since the overflow case is much more
common (e.g. missing a "put" during an error path), this protection
provides real-world protection. For example, the two public refcount
overflow use-after-free exploits published in 2016 would have been
rendered unexploitable:

http://perception-point.io/2016/01/14/analysis-and-exploitation-of-a-linux-kernel-vulnerability-cve-2016-0728/

http://cyseclabs.com/page?n=02012016

This implementation does, however, notice an unchecked decement to zero
(i.e. caller used refcount_dec() instead of refcount_dec_and_test() and it
resulted in a zero). Decrements under zero are noticed (since they will
have resulted in a negative value), though this only indicates that a
use-after-free may have already happened. Such notifications are likely
avoidable by an attacker that has already exploited a use-after-free
vulnerability, but it's better to have them reported than allow such
conditions to remain universally silent.

On first overflow detection, the refcount value is reset to INT_MIN / 2
(which serves as a saturation value) and a report and stack trace are
produced. When operations detect only negative value results (such as
changing an already saturated value), saturation still happens but no
notification is performed (since the value was already saturated).

On the matter of races, since the entire range beyond INT_MAX but before
0 is negative, every operation at INT_MIN / 2 will trap, leaving no
overflow-only race condition.

As for performance, this implementation adds a single "js" instruction
to the regular execution flow of a copy of the standard atomic_t refcount
operations. (The non-"and_test" refcount_dec() function, which is uncommon
in regular refcount design patterns, has an additional "jz" instruction
to detect reaching exactly zero.) Since this is a forward jump, it is by
default the non-predicted path, which will be reinforced by dynamic branch
prediction. The result is this protection having virtually no measurable
change in performance over standard atomic_t operations. The error path,
located in .text.unlikely, saves the refcount location and then uses UD0
to fire a refcount exception handler, which resets the refcount, handles
reporting, and returns to regular execution. This keeps the changes to
.text size minimal, avoiding return jumps and open-coded calls to the
error reporting routine.

Example assembly comparison:

refcount_inc before
.text:
ffffffff81546149:       f0 ff 45 f4             lock incl -0xc(%rbp)

refcount_inc after
.text:
ffffffff81546149:       f0 ff 45 f4             lock incl -0xc(%rbp)
ffffffff8154614d:       0f 88 80 d5 17 00       js     ffffffff816c36d3
...
.text.unlikely:
ffffffff816c36d3:       48 8d 4d f4             lea    -0xc(%rbp),%rcx
ffffffff816c36d7:       0f ff                   (bad)

These are the cycle counts comparing a loop of refcount_inc() from 1
to INT_MAX and back down to 0 (via refcount_dec_and_test()), between
unprotected refcount_t (atomic_t), fully protected REFCOUNT_FULL
(refcount_t-full), and this overflow-protected refcount (refcount_t-fast):

2147483646 refcount_inc()s and 2147483647 refcount_dec_and_test()s:
			cycles	protections
atomic_t	   82249267387	none
refcount_t-fast    82211446892	overflow, untested dec-to-zero
refcount_t-full   144814735193	overflow, untested dec-to-zero, inc-from-zero

This code is a modified version of the x86 PAX_REFCOUNT atomic_t
overflow defense from the last public patch of PaX/grsecurity, based
on my understanding of the code. Changes or omissions from the original
code are mine and don't reflect the original grsecurity/PaX code. Thanks
to PaX Team for various suggestions for improvement for repurposing this
code to be a refcount-only protection.

Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
Reviewed-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
---
 arch/Kconfig                    |   9 ++++
 arch/x86/Kconfig                |   1 +
 arch/x86/include/asm/asm.h      |   6 +++
 arch/x86/include/asm/refcount.h | 113 ++++++++++++++++++++++++++++++++++++++++
 arch/x86/mm/extable.c           |  42 +++++++++++++++
 include/linux/kernel.h          |   7 +++
 include/linux/refcount.h        |   4 ++
 kernel/panic.c                  |  12 +++++
 8 files changed, 194 insertions(+)
 create mode 100644 arch/x86/include/asm/refcount.h

diff --git a/arch/Kconfig b/arch/Kconfig
index 21d0089117fe..349185f951cd 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -931,6 +931,15 @@ config STRICT_MODULE_RWX
 config ARCH_WANT_RELAX_ORDER
 	bool
 
+config ARCH_HAS_REFCOUNT
+	bool
+	help
+	  An architecture selects this when it has implemented refcount_t
+	  using primitizes that provide a faster runtime at the expense
+	  of some full refcount state checks. The refcount overflow condition,
+	  however, must be retained. Catching overflows is the primary
+	  security concern for protecting against bugs in reference counts.
+
 config REFCOUNT_FULL
 	bool "Perform full reference count validation at the expense of speed"
 	help
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 781521b7cf9e..73574c91e857 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -55,6 +55,7 @@ config X86
 	select ARCH_HAS_KCOV			if X86_64
 	select ARCH_HAS_MMIO_FLUSH
 	select ARCH_HAS_PMEM_API		if X86_64
+	select ARCH_HAS_REFCOUNT
 	select ARCH_HAS_UACCESS_FLUSHCACHE	if X86_64
 	select ARCH_HAS_SET_MEMORY
 	select ARCH_HAS_SG_CHAIN
diff --git a/arch/x86/include/asm/asm.h b/arch/x86/include/asm/asm.h
index 7a9df3beb89b..676ee5807d86 100644
--- a/arch/x86/include/asm/asm.h
+++ b/arch/x86/include/asm/asm.h
@@ -74,6 +74,9 @@
 # define _ASM_EXTABLE_EX(from, to)				\
 	_ASM_EXTABLE_HANDLE(from, to, ex_handler_ext)
 
+# define _ASM_EXTABLE_REFCOUNT(from, to)			\
+	_ASM_EXTABLE_HANDLE(from, to, ex_handler_refcount)
+
 # define _ASM_NOKPROBE(entry)					\
 	.pushsection "_kprobe_blacklist","aw" ;			\
 	_ASM_ALIGN ;						\
@@ -123,6 +126,9 @@
 # define _ASM_EXTABLE_EX(from, to)				\
 	_ASM_EXTABLE_HANDLE(from, to, ex_handler_ext)
 
+# define _ASM_EXTABLE_REFCOUNT(from, to)			\
+	_ASM_EXTABLE_HANDLE(from, to, ex_handler_refcount)
+
 /* For C file, we already have NOKPROBE_SYMBOL macro */
 #endif
 
diff --git a/arch/x86/include/asm/refcount.h b/arch/x86/include/asm/refcount.h
new file mode 100644
index 000000000000..3c4c0576c048
--- /dev/null
+++ b/arch/x86/include/asm/refcount.h
@@ -0,0 +1,113 @@
+#ifndef __ASM_X86_REFCOUNT_H
+#define __ASM_X86_REFCOUNT_H
+/*
+ * x86-specific implementation of refcount_t. Based on PAX_REFCOUNT from
+ * PaX/grsecurity.
+ */
+#include <linux/refcount.h>
+
+/*
+ * Body of refcount error handling: in .text.unlikely, saved into CX the
+ * address of the refcount that has entered a bad state, and trigger an
+ * exception. Fixup address is back in regular execution flow in .text.
+ */
+#define _REFCOUNT_EXCEPTION				\
+	".pushsection .text.unlikely\n"			\
+	"111:\tlea %[counter], %%" _ASM_CX "\n"		\
+	"112:\t" ASM_UD0 "\n"				\
+	ASM_UNREACHABLE					\
+	".popsection\n"					\
+	"113:\n"					\
+	_ASM_EXTABLE_REFCOUNT(112b, 113b)
+
+/* Trigger refcount exception if refcount result is negative. */
+#define REFCOUNT_CHECK_LT_ZERO				\
+	"js 111f\n\t"					\
+	_REFCOUNT_EXCEPTION
+
+/* Trigger refcount exception if refcount result is zero or negative. */
+#define REFCOUNT_CHECK_LE_ZERO				\
+	"jz 111f\n\t"					\
+	REFCOUNT_CHECK_LT_ZERO
+
+static __always_inline void refcount_add(unsigned int i, refcount_t *r)
+{
+	asm volatile(LOCK_PREFIX "addl %1,%0\n\t"
+		REFCOUNT_CHECK_LT_ZERO
+		: [counter] "+m" (r->refs.counter)
+		: "ir" (i)
+		: "cc", "cx");
+}
+
+static __always_inline void refcount_inc(refcount_t *r)
+{
+	asm volatile(LOCK_PREFIX "incl %0\n\t"
+		REFCOUNT_CHECK_LT_ZERO
+		: [counter] "+m" (r->refs.counter)
+		: : "cc", "cx");
+}
+
+static __always_inline void refcount_dec(refcount_t *r)
+{
+	asm volatile(LOCK_PREFIX "decl %0\n\t"
+		REFCOUNT_CHECK_LE_ZERO
+		: [counter] "+m" (r->refs.counter)
+		: : "cc", "cx");
+}
+
+static __always_inline __must_check
+bool refcount_sub_and_test(unsigned int i, refcount_t *r)
+{
+	GEN_BINARY_SUFFIXED_RMWcc(LOCK_PREFIX "subl", REFCOUNT_CHECK_LT_ZERO,
+				  r->refs.counter, "er", i, "%0", e);
+}
+
+static __always_inline __must_check bool refcount_dec_and_test(refcount_t *r)
+{
+	GEN_UNARY_SUFFIXED_RMWcc(LOCK_PREFIX "decl", REFCOUNT_CHECK_LT_ZERO,
+				 r->refs.counter, "%0", e);
+}
+
+/**
+ * __refcount_add_unless - add unless the number is already a given value
+ * @r: pointer of type refcount_t
+ * @a: the amount to add to v...
+ * @u: ...unless v is equal to u.
+ *
+ * Atomically adds @a to @r, so long as @r was not already @u.
+ * Returns the old value of @r.
+ */
+static __always_inline __must_check
+int __refcount_add_unless(refcount_t *r, int a, int u)
+{
+	int c, new;
+
+	c = atomic_read(&(r->refs));
+	do {
+		if (unlikely(c == u))
+			break;
+
+		asm volatile("addl %2,%0\n\t"
+			REFCOUNT_CHECK_LT_ZERO
+			: "=r" (new)
+			: "0" (c), "ir" (a),
+			  [counter] "m" (r->refs.counter)
+			: "cc", "cx");
+
+	} while (!atomic_try_cmpxchg(&(r->refs), &c, new));
+
+	return c;
+}
+
+static __always_inline __must_check
+bool refcount_add_not_zero(unsigned int i, refcount_t *r)
+{
+	return __refcount_add_unless(r, i, 0) != 0;
+}
+
+static __always_inline __must_check bool refcount_inc_not_zero(refcount_t *r)
+{
+	return refcount_add_not_zero(1, r);
+}
+
+#endif
diff --git a/arch/x86/mm/extable.c b/arch/x86/mm/extable.c
index 0ea8afcb929c..956075fb3d59 100644
--- a/arch/x86/mm/extable.c
+++ b/arch/x86/mm/extable.c
@@ -36,6 +36,48 @@ bool ex_handler_fault(const struct exception_table_entry *fixup,
 }
 EXPORT_SYMBOL_GPL(ex_handler_fault);
 
+/*
+ * Handler for UD0 exception following a failed test against the
+ * result of a refcount inc/dec/add/sub.
+ */
+bool ex_handler_refcount(const struct exception_table_entry *fixup,
+			 struct pt_regs *regs, int trapnr)
+{
+	/* First unconditionally saturate the refcount. */
+	*(int *)regs->cx = INT_MIN / 2;
+
+	/*
+	 * Strictly speaking, this reports the fixup destination, not
+	 * the fault location, and not the actually overflowing
+	 * instruction, which is the instruction before the "js", but
+	 * since that instruction could be a variety of lengths, just
+	 * report the location after the overflow, which should be close
+	 * enough for finding the overflow, as it's at least back in
+	 * the function, having returned from .text.unlikely.
+	 */
+	regs->ip = ex_fixup_addr(fixup);
+
+	/*
+	 * This function has been called because either a negative refcount
+	 * value was seen by any of the refcount functions, or a zero
+	 * refcount value was seen by refcount_dec().
+	 *
+	 * If we crossed from INT_MAX to INT_MIN, the OF flag (result
+	 * wrapped around) will be set. Additionally, seeing the refcount
+	 * reach 0 will set the ZF flag. In each of these cases we want a
+	 * report, since it's a boundary condition.
+	 *
+	 */
+	if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) {
+		bool zero = regs->flags & X86_EFLAGS_ZF;
+
+		refcount_error_report(regs, zero ? "hit zero" : "overflow");
+	}
+
+	return true;
+}
+EXPORT_SYMBOL_GPL(ex_handler_refcount);
+
 bool ex_handler_ext(const struct exception_table_entry *fixup,
 		   struct pt_regs *regs, int trapnr)
 {
diff --git a/include/linux/kernel.h b/include/linux/kernel.h
index bd6d96cf80b1..87de25965a92 100644
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -277,6 +277,13 @@ extern int oops_may_print(void);
 void do_exit(long error_code) __noreturn;
 void complete_and_exit(struct completion *, long) __noreturn;
 
+#ifdef CONFIG_ARCH_HAS_REFCOUNT
+void refcount_error_report(struct pt_regs *regs, const char *msg);
+#else
+static inline void refcount_error_report(struct pt_regs *regs,
+					 const char *msg) { }
+#endif
+
 /* Internal, do not use. */
 int __must_check _kstrtoul(const char *s, unsigned int base, unsigned long *res);
 int __must_check _kstrtol(const char *s, unsigned int base, long *res);
diff --git a/include/linux/refcount.h b/include/linux/refcount.h
index 591792c8e5b0..48b7c9c68c4d 100644
--- a/include/linux/refcount.h
+++ b/include/linux/refcount.h
@@ -53,6 +53,9 @@ extern __must_check bool refcount_sub_and_test(unsigned int i, refcount_t *r);
 extern __must_check bool refcount_dec_and_test(refcount_t *r);
 extern void refcount_dec(refcount_t *r);
 #else
+# ifdef CONFIG_ARCH_HAS_REFCOUNT
+#  include <asm/refcount.h>
+# else
 static inline __must_check bool refcount_add_not_zero(unsigned int i, refcount_t *r)
 {
 	return atomic_add_unless(&r->refs, i, 0);
@@ -87,6 +90,7 @@ static inline void refcount_dec(refcount_t *r)
 {
 	atomic_dec(&r->refs);
 }
+# endif /* !CONFIG_ARCH_HAS_REFCOUNT */
 #endif /* CONFIG_REFCOUNT_FULL */
 
 extern __must_check bool refcount_dec_if_one(refcount_t *r);
diff --git a/kernel/panic.c b/kernel/panic.c
index a58932b41700..a33d6bc4b0ce 100644
--- a/kernel/panic.c
+++ b/kernel/panic.c
@@ -26,6 +26,7 @@
 #include <linux/nmi.h>
 #include <linux/console.h>
 #include <linux/bug.h>
+#include <linux/ratelimit.h>
 
 #define PANIC_TIMER_STEP 100
 #define PANIC_BLINK_SPD 18
@@ -601,6 +602,17 @@ EXPORT_SYMBOL(__stack_chk_fail);
 
 #endif
 
+#ifdef CONFIG_ARCH_HAS_REFCOUNT
+void refcount_error_report(struct pt_regs *regs, const char *msg)
+{
+	WARN_RATELIMIT(1, "refcount %s detected at %pB in %s[%d], uid/euid: %u/%u\n",
+		msg, (void *)instruction_pointer(regs),
+		current->comm, task_pid_nr(current),
+		from_kuid_munged(&init_user_ns, current_uid()),
+		from_kuid_munged(&init_user_ns, current_euid()));
+}
+#endif
+
 core_param(panic, panic_timeout, int, 0644);
 core_param(pause_on_oops, pause_on_oops, int, 0644);
 core_param(panic_on_warn, panic_on_warn, int, 0644);
-- 
2.7.4