kprobes: improve error handling when arming/disarming kprobes

Improve error handling when arming ftrace-based kprobes. Specifically, if we fail to arm a ftrace-based kprobe, register_kprobe()/enable_kprobe() should report an error instead of success. Previously, this has lead to confusing situations where register_kprobe() would return 0 indicating success, but the kprobe would not be functional if ftrace registration during the kprobe arming process had failed. We should therefore take any errors returned by ftrace into account and propagate this error so that we do not register/enable kprobes that cannot be armed. This can happen if, for example, register_ftrace_function() finds an IPMODIFY conflict (since kprobe_ftrace_ops has this flag set) and returns an error. Such a conflict is possible since livepatches also set the IPMODIFY flag for their ftrace_ops.

arm_all_kprobes() keeps its current behavior and attempts to arm all kprobes. It returns the last encountered error and gives a warning if not all kprobes could be armed.

This patch is based on Petr Mladek's original patchset (patches 2 and 3) back in 2015, which improved kprobes error handling, found here:

   https://lkml.org/lkml/2015/2/26/452

However, further work on this had been paused since then and the patches were not upstreamed.

Based-on-patches-by Petr Mladek <[email protected]>

Signed-off-by Jessica Yu <[email protected]>

---
 kernel/kprobes.c | 87 +++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 61 insertions(+), 26 deletions(-)

kernel/kprobes.c +22 -12

index 2d28377a0e32..6e889be0d93c 100644
@@ -979,18 +979,27 @@ static int prepare_kprobe(struct kprobe *p)
 }
 
 /* Caller must lock kprobe_mutex */
-static void arm_kprobe_ftrace(struct kprobe *p)
+static int arm_kprobe_ftrace(struct kprobe *p)
 {
-	int ret;
+	int ret = 0;
 
 	ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 				   (unsigned long)p->addr, 0, 0);
-	WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
-	kprobe_ftrace_enabled++;
-	if (kprobe_ftrace_enabled == 1) {
+	if (WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret))
+		return ret;
+
+	if (kprobe_ftrace_enabled == 0) {
 		ret = register_ftrace_function(&kprobe_ftrace_ops);
-		WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
+		if (WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret))
+			goto err_ftrace;
 	}
+
+	kprobe_ftrace_enabled++;
+	return ret;
+
+err_ftrace:
+	ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0);
+	return ret;
 }
 
 /* Caller must lock kprobe_mutex */

@@ -1009,22 +1018,23 @@ static void disarm_kprobe_ftrace(struct kprobe *p)
 }
 #else	/* !CONFIG_KPROBES_ON_FTRACE */
 #define prepare_kprobe(p)	arch_prepare_kprobe(p)
-#define arm_kprobe_ftrace(p)	do {} while (0)
+#define arm_kprobe_ftrace(p)	(0)
 #define disarm_kprobe_ftrace(p)	do {} while (0)
 #endif
 
 /* Arm a kprobe with text_mutex */
-static void arm_kprobe(struct kprobe *kp)
+static int arm_kprobe(struct kprobe *kp)
 {
-	if (unlikely(kprobe_ftrace(kp))) {
-		arm_kprobe_ftrace(kp);
-		return;
-	}
+	if (unlikely(kprobe_ftrace(kp)))
+		return arm_kprobe_ftrace(kp);
+
 	cpus_read_lock();
 	mutex_lock(&text_mutex);
 	__arm_kprobe(kp);
 	mutex_unlock(&text_mutex);
 	cpus_read_unlock();
+
+	return 0;
 }
 
 /* Disarm a kprobe with text_mutex */

@@ -1363,9 +1373,14 @@ static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
 

kernel/kprobes.c +2 -0

+			}
+		}
 	}
 	return ret;
 }

Hi Jessica,

Quoted 16 lines · On Wed, 4 Oct 2017 21:14:13 +0200 Jessica Yu <[email protected]> wrote:

Improve error handling when arming ftrace-based kprobes. Specifically, if we fail to arm a ftrace-based kprobe, register_kprobe()/enable_kprobe() should report an error instead of success. Previously, this has lead to confusing situations where register_kprobe() would return 0 indicating success, but the kprobe would not be functional if ftrace registration during the kprobe arming process had failed. We should therefore take any errors returned by ftrace into account and propagate this error so that we do not register/enable kprobes that cannot be armed. This can happen if, for example, register_ftrace_function() finds an IPMODIFY conflict (since kprobe_ftrace_ops has this flag set) and returns an error. Such a conflict is possible since livepatches also set the IPMODIFY flag for their ftrace_ops.

arm_all_kprobes() keeps its current behavior and attempts to arm all kprobes. It returns the last encountered error and gives a warning if not all kprobes could be armed.

Quoted 7 lines

This patch is based on Petr Mladek's original patchset (patches 2 and 3) back in 2015, which improved kprobes error handling, found here:

   https://lkml.org/lkml/2015/2/26/452

However, further work on this had been paused since then and the patches were not upstreamed.

Ok, I have some comment. See below.

Quoted 77 lines

Based-on-patches-by Petr Mladek <[email protected]>

Signed-off-by Jessica Yu <[email protected]>

---
 kernel/kprobes.c | 87 +++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 61 insertions(+), 26 deletions(-)

kernel/kprobes.c +22 -12

index 2d28377a0e32..6e889be0d93c 100644
@@ -979,18 +979,27 @@ static int prepare_kprobe(struct kprobe *p)
 }
 
 /* Caller must lock kprobe_mutex */
-static void arm_kprobe_ftrace(struct kprobe *p)
+static int arm_kprobe_ftrace(struct kprobe *p)
 {
-	int ret;
+	int ret = 0;
 
 	ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 				   (unsigned long)p->addr, 0, 0);
-	WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
-	kprobe_ftrace_enabled++;
-	if (kprobe_ftrace_enabled == 1) {
+	if (WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret))
+		return ret;
+
+	if (kprobe_ftrace_enabled == 0) {
 		ret = register_ftrace_function(&kprobe_ftrace_ops);
-		WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
+		if (WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret))
+			goto err_ftrace;
 	}
+
+	kprobe_ftrace_enabled++;
+	return ret;
+
+err_ftrace:
+	ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0);
+	return ret;
 }
 
 /* Caller must lock kprobe_mutex */

@@ -1009,22 +1018,23 @@ static void disarm_kprobe_ftrace(struct kprobe *p)
 }
 #else	/* !CONFIG_KPROBES_ON_FTRACE */
 #define prepare_kprobe(p)	arch_prepare_kprobe(p)
-#define arm_kprobe_ftrace(p)	do {} while (0)
+#define arm_kprobe_ftrace(p)	(0)
 #define disarm_kprobe_ftrace(p)	do {} while (0)
 #endif
 
 /* Arm a kprobe with text_mutex */
-static void arm_kprobe(struct kprobe *kp)
+static int arm_kprobe(struct kprobe *kp)
 {
-	if (unlikely(kprobe_ftrace(kp))) {
-		arm_kprobe_ftrace(kp);
-		return;
-	}
+	if (unlikely(kprobe_ftrace(kp)))
+		return arm_kprobe_ftrace(kp);
+
 	cpus_read_lock();
 	mutex_lock(&text_mutex);
 	__arm_kprobe(kp);
 	mutex_unlock(&text_mutex);
 	cpus_read_unlock();
+
+	return 0;
 }
 
 /* Disarm a kprobe with text_mutex */

@@ -1363,9 +1373,14 @@ static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
 

kernel/kprobes.c static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p) · line 1381

Quoted 10 lines

kernel/kprobes.c +5 -2

 	if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
 		ap->flags &= ~KPROBE_FLAG_DISABLED;
-		if (!kprobes_all_disarmed)
+		if (!kprobes_all_disarmed) {
 			/* Arm the breakpoint again. */
-			arm_kprobe(ap);
+			ret = arm_kprobe(ap);
+			if (ret) {
+				ap->flags |= KPROBE_FLAG_DISABLED;
+				list_del_rcu(&p->list);

Nice catch :) this list_del_rcu() is important to keep error case behavior sane.

Quoted 9 lines

kernel/kprobes.c +2 -0

+			}
+		}
 	}
 	return ret;
 }

@@ -1570,13 +1585,16 @@ int register_kprobe(struct kprobe *p)
 	if (ret)
 		goto out;
 

kernel/kprobes.c int register_kprobe(struct kprobe *p) · line 1593

Quoted 6 lines

kernel/kprobes.c +6 -0

+	if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
+		ret = arm_kprobe(p);
+		if (ret)
+			goto out;
+	}
+

No, this is no good. It is a small chance to hit kprobe on other CPUs before adding it to kprobe_table hashlist. In that case, we will see a stray breakpoint instruction.

Quoted 4 lines

kernel/kprobes.c +0 -0

 	INIT_HLIST_NODE(&p->hlist);
 	hlist_add_head_rcu(&p->hlist,
 		       &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 

kernel/kprobes.c int register_kprobe(struct kprobe *p) · line 1579

Quoted 3 lines

kernel/kprobes.c +0 -3

-	if (!kprobes_all_disarmed && !kprobe_disabled(p))
-		arm_kprobe(p);
-

So, you'll have to rollback by hlist_del_rcu() here.
Hmm, by the way, in this case, you also have to add a synchronize_rcu() in the end of error path, so that user can release kprobe right after error return of register_kprobe... (I think that's OK because it is not a hot path)

Improve error handling when disarming ftrace-based kprobes. Like with arm_kprobe_ftrace(), propagate any errors from disarm_kprobe_ftrace() so that we do not disable/unregister kprobes that are still armed. In other words, unregister_kprobe() and disable_kprobe() should not report success if the kprobe could not be disarmed.

disarm_all_kprobes() keeps its current behavior and attempts to disarm all kprobes. It returns the last encountered error and gives a warning if not all kprobes could be disarmed.

This patch is based on Petr Mladek's original patchset (patches 2 and 3) back in 2015, which improved kprobes error handling, found here:

   https://lkml.org/lkml/2015/2/26/452

However, further work on this had been paused since then and the patches were not upstreamed.

Based-on-patches-by Petr Mladek <[email protected]>

Signed-off-by Jessica Yu <[email protected]>

---
 kernel/kprobes.c | 76 +++++++++++++++++++++++++++++++++++++-------------------
 1 file changed, 51 insertions(+), 25 deletions(-)

kernel/kprobes.c +51 -25

index 6e889be0d93c..2e9edf646ae3 100644
@@ -1003,23 +1003,27 @@ static int arm_kprobe_ftrace(struct kprobe *p)
 }
 
 /* Caller must lock kprobe_mutex */
-static void disarm_kprobe_ftrace(struct kprobe *p)
+static int disarm_kprobe_ftrace(struct kprobe *p)
 {
-	int ret;
+	int ret = 0;
 
-	kprobe_ftrace_enabled--;
-	if (kprobe_ftrace_enabled == 0) {
+	if (kprobe_ftrace_enabled == 1) {
 		ret = unregister_ftrace_function(&kprobe_ftrace_ops);
-		WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
+		if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
+			return ret;
 	}
+
+	kprobe_ftrace_enabled--;
+
 	ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 			   (unsigned long)p->addr, 1, 0);
 	WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
+	return ret;
 }
 #else	/* !CONFIG_KPROBES_ON_FTRACE */
 #define prepare_kprobe(p)	arch_prepare_kprobe(p)
 #define arm_kprobe_ftrace(p)	(0)
-#define disarm_kprobe_ftrace(p)	do {} while (0)
+#define disarm_kprobe_ftrace(p)	(0)
 #endif
 
 /* Arm a kprobe with text_mutex */

@@ -1038,18 +1042,18 @@ static int arm_kprobe(struct kprobe *kp)
 }
 
 /* Disarm a kprobe with text_mutex */
-static void disarm_kprobe(struct kprobe *kp, bool reopt)
+static int disarm_kprobe(struct kprobe *kp, bool reopt)
 {
-	if (unlikely(kprobe_ftrace(kp))) {
-		disarm_kprobe_ftrace(kp);
-		return;
-	}
+	if (unlikely(kprobe_ftrace(kp)))
+		return disarm_kprobe_ftrace(kp);
 
 	cpus_read_lock();
 	mutex_lock(&text_mutex);
 	__disarm_kprobe(kp, reopt);
 	mutex_unlock(&text_mutex);
 	cpus_read_unlock();
+
+	return 0;
 }
 
 /*

@@ -1627,11 +1631,12 @@ static int aggr_kprobe_disabled(struct kprobe *ap)
 static struct kprobe *__disable_kprobe(struct kprobe *p)
 {
 	struct kprobe *orig_p;
+	int ret;
 
 	/* Get an original kprobe for return */
 	orig_p = __get_valid_kprobe(p);
 	if (unlikely(orig_p == NULL))
-		return NULL;
+		return ERR_PTR(-EINVAL);
 
 	if (!kprobe_disabled(p)) {
 		/* Disable probe if it is a child probe */

@@ -1645,8 +1650,13 @@ static struct kprobe *__disable_kprobe(struct kprobe *p)
 			 * should have already been disarmed, so
 			 * skip unneed disarming process.
 			 */
-			if (!kprobes_all_disarmed)
-				disarm_kprobe(orig_p, true);
+			if (!kprobes_all_disarmed) {
+				ret = disarm_kprobe(orig_p, true);
+				if (ret) {
+					p->flags &= ~KPROBE_FLAG_DISABLED;
+					return ERR_PTR(ret);
+				}
+			}
 			orig_p->flags |= KPROBE_FLAG_DISABLED;
 		}
 	}

@@ -1663,8 +1673,8 @@ static int __unregister_kprobe_top(struct kprobe *p)
 
 	/* Disable kprobe. This will disarm it if needed. */
 	ap = __disable_kprobe(p);
-	if (ap == NULL)
-		return -EINVAL;
+	if (IS_ERR(ap))
+		return PTR_ERR(ap);
 
 	if (ap == p)
 		/*

@@ -2095,12 +2105,14 @@ static void kill_kprobe(struct kprobe *p)
 int disable_kprobe(struct kprobe *kp)
 {
 	int ret = 0;
+	struct kprobe *p;
 
 	mutex_lock(&kprobe_mutex);
 
 	/* Disable this kprobe */
-	if (__disable_kprobe(kp) == NULL)
-		ret = -EINVAL;
+	p = __disable_kprobe(kp);
+	if (IS_ERR(p))
+		ret = PTR_ERR(p);
 
 	mutex_unlock(&kprobe_mutex);
 	return ret;

@@ -2470,34 +2482,48 @@ static int arm_all_kprobes(void)
 	return ret;
 }
 
-static void disarm_all_kprobes(void)
+static int disarm_all_kprobes(void)
 {
 	struct hlist_head *head;
 	struct kprobe *p;
-	unsigned int i;
+	unsigned int i, errors = 0;
+	int err, ret = 0;
 
 	mutex_lock(&kprobe_mutex);
 
 	/* If kprobes are already disarmed, just return */
 	if (kprobes_all_disarmed) {
 		mutex_unlock(&kprobe_mutex);
-		return;
+		return 0;
 	}
 
 	kprobes_all_disarmed = true;
-	printk(KERN_INFO "Kprobes globally disabled\n");
 
 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 		head = &kprobe_table[i];
+		/* Disarm all kprobes on a best-effort basis */
 		hlist_for_each_entry_rcu(p, head, hlist) {
-			if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
-				disarm_kprobe(p, false);
+			if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
+				err = disarm_kprobe(p, false);
+				if (err) {
+					errors++;
+					ret = err;
+				}
+			}
 		}
 	}
+
+	if (errors)
+		pr_warn("Kprobes globally disabled, but failed to disarm %d kprobes\n", errors);
+	else
+		pr_info("Kprobes globally disabled\n");
+
 	mutex_unlock(&kprobe_mutex);
 
 	/* Wait for disarming all kprobes by optimizer */
 	wait_for_kprobe_optimizer();
+
+	return ret;
 }
 
 /*

@@ -2540,7 +2566,7 @@ static ssize_t write_enabled_file_bool(struct file *file,
 	case 'n':
 	case 'N':
 	case '0':
-		disarm_all_kprobes();
+		ret = disarm_all_kprobes();
 		break;
 	default:
 		return -EINVAL;

Signature · 1 line

2.13.6

Hi,

This patchset attempts to improve error handling when arming or disarming ftrace-based kprobes. The current behavior is to simply WARN when ftrace (un-)registration fails, without propagating the error code. This can lead to confusing situations where, for example, register_kprobe()/enable_kprobe() would return 0 indicating success even if arming via ftrace had failed. In this scenario we'd end up with a non-functioning kprobe even though kprobe registration (or enablement) returned success. In this patchset, we take errors from ftrace into account and propagate the error when we cannot arm or disarm a kprobe.

Below is an example that illustrates the problem using livepatch and systemtap (which uses kprobes underneath). Both livepatch and kprobes use ftrace ops with the IPMODIFY flag set, so registration at the same function entry is limited to only one ftrace user.

Before
------

# modprobe livepatch-sample 	# patches cmdline_proc_show, ftrace ops has IPMODIFY set
# stap -e 'probe kernel.function("cmdline_proc_show").call { printf ("cmdline_proc_show\n"); }'

.. (nothing prints after reading /proc/cmdline) ..

The systemtap handler doesn't execute due to a kprobe arming failure caused by a ftrace IPMODIFY conflict with livepatch, and there isn't an obvious indication of error from systemtap (because register_kprobe() returned success) unless the user inspects dmesg.

After
-----

# modprobe livepatch-sample 
# stap -e 'probe kernel.function("cmdline_proc_show").call { printf ("cmdline_proc_show\n"); }'
WARNING: probe kernel.function("cmdline_proc_show@/home/jeyu/work/linux-next/fs/proc/cmdline.c:6").call (address 0xffffffffa82fe910) registration error (rc -16)

Although the systemtap handler doesn't execute (as it shouldn't), the ftrace error is propagated and now systemtap prints a visible error message stating that (kprobe) registration had failed (because register_kprobe() returned an error), along with the propagated error code.

This patchset was based on Petr Mladek's original patchset (patches 2 and 3) back in 2015, which improved kprobes error handling, found here:

   https://lkml.org/lkml/2015/2/26/452

However, further work on this had been paused since then and the patches were not upstreamed.

This patchset has been lightly sanity-tested (on linux-next) with kprobes, kretprobes, jprobes, and optimized kprobes. It passes the kprobes smoke test, but more testing is greatly appreciated.

Thanks,

Jessica

---
Jessica Yu (2):
kprobes: propagate error from arm_kprobe_ftrace()
kprobes: propagate error from disarm_kprobe_ftrace()

 kernel/kprobes.c | 163 ++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 112 insertions(+), 51 deletions(-)

Masami,

Can you review these patches?

-- Steve

Quoted 28 lines · On Wed, 4 Oct 2017 21:14:12 +0200 Jessica Yu <[email protected]> wrote:

Hi,

This patchset attempts to improve error handling when arming or disarming ftrace-based kprobes. The current behavior is to simply WARN when ftrace (un-)registration fails, without propagating the error code. This can lead to confusing situations where, for example, register_kprobe()/enable_kprobe() would return 0 indicating success even if arming via ftrace had failed. In this scenario we'd end up with a non-functioning kprobe even though kprobe registration (or enablement) returned success. In this patchset, we take errors from ftrace into account and propagate the error when we cannot arm or disarm a kprobe.

Below is an example that illustrates the problem using livepatch and systemtap (which uses kprobes underneath). Both livepatch and kprobes use ftrace ops with the IPMODIFY flag set, so registration at the same function entry is limited to only one ftrace user.

Before
------

# modprobe livepatch-sample 	# patches cmdline_proc_show, ftrace ops has IPMODIFY set
# stap -e 'probe kernel.function("cmdline_proc_show").call { printf ("cmdline_proc_show\n"); }'

.. (nothing prints after reading /proc/cmdline) ..

The systemtap handler doesn't execute due to a kprobe arming failure caused by a ftrace IPMODIFY conflict with livepatch, and there isn't an obvious indication of error from systemtap (because register_kprobe() returned success) unless the user inspects dmesg.

Quoted 11 lines

After
-----

# modprobe livepatch-sample 
# stap -e 'probe kernel.function("cmdline_proc_show").call { printf ("cmdline_proc_show\n"); }'
WARNING: probe kernel.function("cmdline_proc_show@/home/jeyu/work/linux-next/fs/proc/cmdline.c:6").call (address 0xffffffffa82fe910) registration error (rc -16)

Although the systemtap handler doesn't execute (as it shouldn't), the ftrace error is propagated and now systemtap prints a visible error message stating that (kprobe) registration had failed (because register_kprobe() returned an error), along with the propagated error code.

Quoted 25 lines

This patchset was based on Petr Mladek's original patchset (patches 2 and 3) back in 2015, which improved kprobes error handling, found here:

   https://lkml.org/lkml/2015/2/26/452

However, further work on this had been paused since then and the patches were not upstreamed.

This patchset has been lightly sanity-tested (on linux-next) with kprobes, kretprobes, jprobes, and optimized kprobes. It passes the kprobes smoke test, but more testing is greatly appreciated.

Thanks,

Jessica

---
Jessica Yu (2):
kprobes: propagate error from arm_kprobe_ftrace()
kprobes: propagate error from disarm_kprobe_ftrace()

 kernel/kprobes.c | 163 ++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 112 insertions(+), 51 deletions(-)