Skip to contentSkip to conversation

kprobes/x86: While list ftrace locations in kprobe blacklist areas

RFC PATCH In review Steven Rostedt 2 participants 4 messages Oct 10, 2019 – Oct 11, 2019 kprobes

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 kprobes/x86: While list ftrace locations in kprobe blacklist areas

From: Steven Rostedt (VMware) <[email protected]>

I noticed some of my old tests failing on kprobes, and realized that this was due to black listing irq_entry functions on x86 from being used by kprobes. IIRC, this was due to the cr2 being corrupted and such, and I believe other things were to cause. But black listing all irq_entry code is a big hammer to this.

(See commit 0eae81dc9f026 "x86/kprobes: Prohibit probing on IRQ handlers directly" for more details)

Anyway, if kprobes is using ftrace as a hook, there shouldn't be any problems here. If we white list ftrace locations in the range of kprobe_add_area_blacklist(), it should be safe.

---

kernel/kprobes.c +5 -0
index d9770a5393c8..9d28a279282c 100644
@@ -2124,6 +2124,11 @@ int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
 	int ret = 0;
 
 	for (entry = start; entry < end; entry += ret) {
+#ifdef CONFIG_KPROBES_ON_FTRACE
+		/* We are safe if using ftrace */
+		if (ftrace_location(entry))
+			continue;
+#endif
 		ret = kprobe_add_ksym_blacklist(entry);
 		if (ret < 0)
 			return ret;
#1 lore raw
Quoted 8 lines · On Thu, 10 Oct 2019 17:52:16 -0400 Steven Rostedt <[email protected]> wrote:

I noticed some of my old tests failing on kprobes, and realized that this was due to black listing irq_entry functions on x86 from being used by kprobes. IIRC, this was due to the cr2 being corrupted and such, and I believe other things were to cause. But black listing all irq_entry code is a big hammer to this.

(See commit 0eae81dc9f026 "x86/kprobes: Prohibit probing on IRQ handlers directly" for more details)

BTW, I noticed this recently (again) when running my tests by hand. I forgot that I have my automated tests revert the above commit before compiling the kernel it is about to test (because it tests kprobes on irq entry locations!). My tests never had issues with kprobes on irq entry locations.

-- Steve

maintainer
#2 lore raw
In reply to Steven Rostedt

Hi Steve,

Quoted 7 lines · On Thu, 10 Oct 2019 17:52:16 -0400 Steven Rostedt <[email protected]> wrote:

From: Steven Rostedt (VMware) <[email protected]>

I noticed some of my old tests failing on kprobes, and realized that this was due to black listing irq_entry functions on x86 from being used by kprobes. IIRC, this was due to the cr2 being corrupted and such, and I believe other things were to cause. But black listing all irq_entry code is a big hammer to this.

OK, I think if we can use ftrace for hooking, probing on "that address" is good, but the function body still can not be probed.

Quoted 7 lines

(See commit 0eae81dc9f026 "x86/kprobes: Prohibit probing on IRQ handlers directly" for more details)

Anyway, if kprobes is using ftrace as a hook, there shouldn't be any problems here. If we white list ftrace locations in the range of kprobe_add_area_blacklist(), it should be safe.

Agreed.

Quoted 8 lines

---

kernel/kprobes.c +0 -0
index d9770a5393c8..9d28a279282c 100644

kernel/kprobes.c int kprobe_add_area_blacklist(unsigned long start, unsigned long end) · line 2131

Quoted 8 lines
kernel/kprobes.c +5 -0
@@ -2124,6 +2124,11 @@ int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
 	int ret = 0;
 
 	for (entry = start; entry < end; entry += ret) {
+#ifdef CONFIG_KPROBES_ON_FTRACE
+		/* We are safe if using ftrace */
+		if (ftrace_location(entry))
+			continue;
+#endif

Have you tested the patch? it doesn't measure the entry function's size. (kprobe_add_ksym_blacklist(entry) returns the function size)

Could you do this in kprobe_add_ksym_blacklist()?
Instead of continue, increment ent->start_addr by MCOUNT size(?) will be OK. (Note that since each blacklist symbol is managed independently, you can make
a "gap" between them as a safe area)

Thank you,

Quoted 3 lines
kernel/kprobes.c +0 -0
 		ret = kprobe_add_ksym_blacklist(entry);
 		if (ret < 0)
 			return ret;
Signature · 1 line
Masami Hiramatsu <[email protected]>
#3 lore raw
Quoted 36 lines · On Fri, 11 Oct 2019 17:39:29 +0900 Masami Hiramatsu <[email protected]> wrote:

Hi Steve,

Quoted 7 lines · On Thu, 10 Oct 2019 17:52:16 -0400 Steven Rostedt <[email protected]> wrote:

From: Steven Rostedt (VMware) <[email protected]>

I noticed some of my old tests failing on kprobes, and realized that this was due to black listing irq_entry functions on x86 from being used by kprobes. IIRC, this was due to the cr2 being corrupted and such, and I believe other things were to cause. But black listing all irq_entry code is a big hammer to this.

OK, I think if we can use ftrace for hooking, probing on "that address" is good, but the function body still can not be probed.

Quoted 7 lines

(See commit 0eae81dc9f026 "x86/kprobes: Prohibit probing on IRQ handlers directly" for more details)

Anyway, if kprobes is using ftrace as a hook, there shouldn't be any problems here. If we white list ftrace locations in the range of kprobe_add_area_blacklist(), it should be safe.

Agreed.

Quoted 10 lines

---

kernel/kprobes.c +0 -0
index d9770a5393c8..9d28a279282c 100644
@@ -2124,6 +2124,11 @@ int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
 	int ret = 0;
 
Quoted 9 lines

kernel/kprobes.c int kprobe_add_area_blacklist(unsigned long start, unsigned long end) · line 2131

Quoted 6 lines
kernel/kprobes.c +5 -0
 	for (entry = start; entry < end; entry += ret) {
+#ifdef CONFIG_KPROBES_ON_FTRACE
+		/* We are safe if using ftrace */
+		if (ftrace_location(entry))
+			continue;
+#endif  

Have you tested the patch? it doesn't measure the entry function's size. (kprobe_add_ksym_blacklist(entry) returns the function size)

Yeah, I tested the patch, but it is obviously buggy. It fixed the issue I was having (was able to test against do_IRQ ;-)

Quoted 5 lines

Could you do this in kprobe_add_ksym_blacklist()?
Instead of continue, increment ent->start_addr by MCOUNT size(?) will be OK. (Note that since each blacklist symbol is managed independently, you can make
a "gap" between them as a safe area)

OK, I'll move it to kprobe_add_ksym_blacklist.

-- Steve

Quoted 8 lines

Thank you,

Quoted 3 lines
kernel/kprobes.c +0 -0
 		ret = kprobe_add_ksym_blacklist(entry);
 		if (ret < 0)
 			return ret;