x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 202 insertions, 0 deletions

diff --git a/drivers/misc/lkdtm/cfi.c b/drivers/misc/lkdtm/cfi.c
new file mode 100644
index 000000000..c3971f7ca
--- /dev/null
+++ b/drivers/misc/lkdtm/cfi.c
@@ -0,0 +1,202 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This is for all the tests relating directly to Control Flow Integrity.
+ */
+#include "lkdtm.h"
+#include <asm/page.h>
+
+static int called_count;
+
+/* Function taking one argument, without a return value. */
+static noinline void lkdtm_increment_void(int *counter)
+{
+	(*counter)++;
+}
+
+/* Function taking one argument, returning int. */
+static noinline int lkdtm_increment_int(int *counter)
+{
+	(*counter)++;
+
+	return *counter;
+}
+
+/* Don't allow the compiler to inline the calls. */
+static noinline void lkdtm_indirect_call(void (*func)(int *))
+{
+	func(&called_count);
+}
+
+/*
+ * This tries to call an indirect function with a mismatched prototype.
+ */
+static void lkdtm_CFI_FORWARD_PROTO(void)
+{
+	/*
+	 * Matches lkdtm_increment_void()'s prototype, but not
+	 * lkdtm_increment_int()'s prototype.
+	 */
+	pr_info("Calling matched prototype ...\n");
+	lkdtm_indirect_call(lkdtm_increment_void);
+
+	pr_info("Calling mismatched prototype ...\n");
+	lkdtm_indirect_call((void *)lkdtm_increment_int);
+
+	pr_err("FAIL: survived mismatched prototype function call!\n");
+	pr_expected_config(CONFIG_CFI);
+}
+
+/*
+ * This can stay local to LKDTM, as there should not be a production reason
+ * to disable PAC && SCS.
+ */
+#ifdef CONFIG_ARM64_PTR_AUTH_KERNEL
+# ifdef CONFIG_ARM64_BTI_KERNEL
+#  define __no_pac             "branch-protection=bti"
+# else
+#  ifdef CONFIG_CC_HAS_BRANCH_PROT_PAC_RET
+#   define __no_pac            "branch-protection=none"
+#  else
+#   define __no_pac            "sign-return-address=none"
+#  endif
+# endif
+# define __no_ret_protection   __noscs __attribute__((__target__(__no_pac)))
+#else
+# define __no_ret_protection   __noscs
+#endif
+
+#define no_pac_addr(addr)      \
+	((__force __typeof__(addr))((uintptr_t)(addr) | PAGE_OFFSET))
+
+#ifdef CONFIG_RISCV
+/* https://github.com/riscv-non-isa/riscv-elf-psabi-doc/blob/master/riscv-cc.adoc#frame-pointer-convention */
+#define FRAME_RA_OFFSET		(-1)
+#else
+#define FRAME_RA_OFFSET		1
+#endif
+
+/* The ultimate ROP gadget. */
+static noinline __no_ret_protection
+void set_return_addr_unchecked(unsigned long *expected, unsigned long *addr)
+{
+	/* Use of volatile is to make sure final write isn't seen as a dead store. */
+	unsigned long * volatile *ret_addr =
+		(unsigned long **)__builtin_frame_address(0) + FRAME_RA_OFFSET;
+
+	/* Make sure we've found the right place on the stack before writing it. */
+	if (no_pac_addr(*ret_addr) == expected)
+		*ret_addr = (addr);
+	else
+		/* Check architecture, stack layout, or compiler behavior... */
+		pr_warn("Eek: return address mismatch! %px != %px\n",
+			*ret_addr, addr);
+}
+
+static noinline
+void set_return_addr(unsigned long *expected, unsigned long *addr)
+{
+	/* Use of volatile is to make sure final write isn't seen as a dead store. */
+	unsigned long * volatile *ret_addr =
+		(unsigned long **)__builtin_frame_address(0) + FRAME_RA_OFFSET;
+
+	/* Make sure we've found the right place on the stack before writing it. */
+	if (no_pac_addr(*ret_addr) == expected)
+		*ret_addr = (addr);
+	else
+		/* Check architecture, stack layout, or compiler behavior... */
+		pr_warn("Eek: return address mismatch! %px != %px\n",
+			*ret_addr, addr);
+}
+
+static volatile int force_check;
+
+static void lkdtm_CFI_BACKWARD(void)
+{
+	/* Use calculated gotos to keep labels addressable. */
+	void *labels[] = { NULL, &&normal, &&redirected, &&check_normal, &&check_redirected };
+
+	pr_info("Attempting unchecked stack return address redirection ...\n");
+
+	/* Always false */
+	if (force_check) {
+		/*
+		 * Prepare to call with NULLs to avoid parameters being treated as
+		 * constants in -02.
+		 */
+		set_return_addr_unchecked(NULL, NULL);
+		set_return_addr(NULL, NULL);
+		if (force_check)
+			goto *labels[1];
+		if (force_check)
+			goto *labels[2];
+		if (force_check)
+			goto *labels[3];
+		if (force_check)
+			goto *labels[4];
+		return;
+	}
+
+	/*
+	 * Use fallthrough switch case to keep basic block ordering between
+	 * set_return_addr*() and the label after it.
+	 */
+	switch (force_check) {
+	case 0:
+		set_return_addr_unchecked(&&normal, &&redirected);
+		fallthrough;
+	case 1:
+normal:
+		/* Always true */
+		if (!force_check) {
+			pr_err("FAIL: stack return address manipulation failed!\n");
+			/* If we can't redirect "normally", we can't test mitigations. */
+			return;
+		}
+		break;
+	default:
+redirected:
+		pr_info("ok: redirected stack return address.\n");
+		break;
+	}
+
+	pr_info("Attempting checked stack return address redirection ...\n");
+
+	switch (force_check) {
+	case 0:
+		set_return_addr(&&check_normal, &&check_redirected);
+		fallthrough;
+	case 1:
+check_normal:
+		/* Always true */
+		if (!force_check) {
+			pr_info("ok: control flow unchanged.\n");
+			return;
+		}
+
+check_redirected:
+		pr_err("FAIL: stack return address was redirected!\n");
+		break;
+	}
+
+	if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL)) {
+		pr_expected_config(CONFIG_ARM64_PTR_AUTH_KERNEL);
+		return;
+	}
+	if (IS_ENABLED(CONFIG_SHADOW_CALL_STACK)) {
+		pr_expected_config(CONFIG_SHADOW_CALL_STACK);
+		return;
+	}
+	pr_warn("This is probably expected, since this %s was built *without* %s=y nor %s=y\n",
+		lkdtm_kernel_info,
+		"CONFIG_ARM64_PTR_AUTH_KERNEL", "CONFIG_SHADOW_CALL_STACK");
+}
+
+static struct crashtype crashtypes[] = {
+	CRASHTYPE(CFI_FORWARD_PROTO),
+	CRASHTYPE(CFI_BACKWARD),
+};
+
+struct crashtype_category cfi_crashtypes = {
+	.crashtypes = crashtypes,
+	.len	    = ARRAY_SIZE(crashtypes),
+};