error-injection: Fix to prohibit jump optimization

Since the kprobe which was optimized by jump can not change the execution path, the kprobe for error-injection must not be optimized. To prohibit it, set a dummy post-handler as officially stated in Documentation/kprobes.txt.

Fixes 4b1a29a7f542 ("error-injection: Support fault injection framework")

Signed-off-by Masami Hiramatsu <[email protected]>

---
 kernel/fail_function.c |   10 ++++++++++
 1 file changed, 10 insertions(+)

kernel/fail_function.c +10 -0

index 21b0122cb39c..1d5632d8bbcc 100644
@@ -14,6 +14,15 @@
 
 static int fei_kprobe_handler(struct kprobe *kp, struct pt_regs *regs);
 
+static void fei_post_handler(struct kprobe *kp, struct pt_regs *regs,
+			     unsigned long flags)
+{
+	/*
+	 * A dummy post handler is required to prohibit optimizing, because
+	 * jump optimization does not support execution path overriding.
+	 */
+}
+
 struct fei_attr {
 	struct list_head list;
 	struct kprobe kp;

@@ -56,6 +65,7 @@ static struct fei_attr *fei_attr_new(const char *sym, unsigned long addr)
 			return NULL;
 		}
 		attr->kp.pre_handler = fei_kprobe_handler;
+		attr->kp.post_handler = fei_post_handler;
 		attr->retval = adjust_error_retval(addr, 0);
 		INIT_LIST_HEAD(&attr->list);
 	}

Quoted 4 lines · On Mon, 12 Mar 2018 19:00:49 +0900 Masami Hiramatsu <[email protected]> wrote:

Since the kprobe which was optimized by jump can not change the execution path, the kprobe for error-injection must not be optimized. To prohibit it, set a dummy post-handler as officially stated in Documentation/kprobes.txt.

Note that trace-probe based BPF is not affected, because it ensures the trace-probe is based on ftrace, which is not jump optimized.

Thanks,

Quoted 36 lines

Fixes 4b1a29a7f542 ("error-injection: Support fault injection framework")

Signed-off-by Masami Hiramatsu <[email protected]>

---
 kernel/fail_function.c |   10 ++++++++++
 1 file changed, 10 insertions(+)

kernel/fail_function.c +10 -0

index 21b0122cb39c..1d5632d8bbcc 100644
@@ -14,6 +14,15 @@
 
 static int fei_kprobe_handler(struct kprobe *kp, struct pt_regs *regs);
 
+static void fei_post_handler(struct kprobe *kp, struct pt_regs *regs,
+			     unsigned long flags)
+{
+	/*
+	 * A dummy post handler is required to prohibit optimizing, because
+	 * jump optimization does not support execution path overriding.
+	 */
+}
+
 struct fei_attr {
 	struct list_head list;
 	struct kprobe kp;

@@ -56,6 +65,7 @@ static struct fei_attr *fei_attr_new(const char *sym, unsigned long addr)
 			return NULL;
 		}
 		attr->kp.pre_handler = fei_kprobe_handler;
+		attr->kp.post_handler = fei_post_handler;
 		attr->retval = adjust_error_retval(addr, 0);
 		INIT_LIST_HEAD(&attr->list);
 	}

Hi Masami,

Quoted 8 lines · Masami Hiramatsu

Quoted 4 lines · On Mon, 12 Mar 2018 19:00:49 +0900 Masami Hiramatsu <[email protected]> wrote:

Since the kprobe which was optimized by jump can not change the execution path, the kprobe for error-injection must not be optimized. To prohibit it, set a dummy post-handler as officially stated in Documentation/kprobes.txt.

Quoted 3 lines

Note that trace-probe based BPF is not affected, because it ensures the trace-probe is based on ftrace, which is not jump optimized.

Thanks for the fix! I presume this should go via bpf instead of bpf-next tree since 4b1a29a7f542 ("error-injection: Support fault injection framework") is in Linus' tree as well. Unless there are objection I would rather route it that way so it would be for 4.16.

Thanks,
Daniel

Quoted 40 lines

Thanks,

Quoted 36 lines

Fixes 4b1a29a7f542 ("error-injection: Support fault injection framework")

Signed-off-by Masami Hiramatsu <[email protected]>

---
 kernel/fail_function.c |   10 ++++++++++
 1 file changed, 10 insertions(+)

kernel/fail_function.c +10 -0

index 21b0122cb39c..1d5632d8bbcc 100644
@@ -14,6 +14,15 @@
 
 static int fei_kprobe_handler(struct kprobe *kp, struct pt_regs *regs);
 
+static void fei_post_handler(struct kprobe *kp, struct pt_regs *regs,
+			     unsigned long flags)
+{
+	/*
+	 * A dummy post handler is required to prohibit optimizing, because
+	 * jump optimization does not support execution path overriding.
+	 */
+}
+
 struct fei_attr {
 	struct list_head list;
 	struct kprobe kp;

@@ -56,6 +65,7 @@ static struct fei_attr *fei_attr_new(const char *sym, unsigned long addr)
 			return NULL;
 		}
 		attr->kp.pre_handler = fei_kprobe_handler;
+		attr->kp.post_handler = fei_post_handler;
 		attr->retval = adjust_error_retval(addr, 0);
 		INIT_LIST_HEAD(&attr->list);
 	}

Quoted 13 lines · On Mon, 12 Mar 2018 11:44:21 +0100 Daniel Borkmann <[email protected]> wrote:

Quoted 11 lines · Masami Hiramatsu

Quoted 4 lines · On Mon, 12 Mar 2018 19:00:49 +0900 Masami Hiramatsu <[email protected]> wrote:

Since the kprobe which was optimized by jump can not change the execution path, the kprobe for error-injection must not be optimized. To prohibit it, set a dummy post-handler as officially stated in Documentation/kprobes.txt.

Note that trace-probe based BPF is not affected, because it ensures the trace-probe is based on ftrace, which is not jump optimized.

Quoted 4 lines

Thanks for the fix! I presume this should go via bpf instead of bpf-next tree since 4b1a29a7f542 ("error-injection: Support fault injection framework") is in Linus' tree as well. Unless there are objection I would rather route it that way so it would be for 4.16.

Ah, right! It should go into 4.16. It should be applicable cleanly either tree since there is only the above commit on kernel/fail_function.c :)