Skip to contentSkip to conversation

kprobes: Fix to cancel optimizing/unoptimizing kprobes correctly

PATCH Applied Masami Hiramatsu 3 participants 4 messages Jan 7, 2020 – Jan 10, 2020 kprobes

lore
Share & tools
lore

Masami Hiramatsu

author maintainer
#0 lore raw

PATCH 1/1 kprobes: Fix to cancel optimizing/unoptimizing kprobes correctly

optimize_kprobe() and unoptimize_kprobe() cancels if given kprobe is on the optimizing_list or unoptimizing_list. However, since commit f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code") modified the update timing of the KPROBE_FLAG_OPTIMIZED, it doesn't work as expected anymore.

The optimized_kprobe could be following states.

- [optimizing]: Before inserting jump instruction
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [optimized]: jump inserted
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is empty.

- [unoptimizing]: Before removing jump instruction (including unused optprobe)
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [unoptimized]: jump removed
op.kp->flags doesn't have KPROBE_FLAG_OPTIMIZED and op->list is empty.

Current code mis-expects [unoptimizing] state doesn't have KPROBE_FLAG_OPTIMIZED, and that can cause wrong results.

This introduces optprobe_queued_unopt() to distinguish [optimizing] and [unoptimizing] states and fixes logics in optimize_kprobe() and unoptimize_kprobe().

Fixes f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code")
---
 kernel/kprobes.c |   65 ++++++++++++++++++++++++++++++++++--------------------
 1 file changed, 41 insertions(+), 24 deletions(-)
kernel/kprobes.c +41 -24
index 34e28b236d68..d898b633f1d6 100644
@@ -612,6 +612,16 @@ void wait_for_kprobe_optimizer(void)
 	mutex_unlock(&kprobe_mutex);
 }
 
+static bool optprobe_queued_unopt(struct optimized_kprobe *op)
+{
+	struct optimized_kprobe *_op;
+
+	list_for_each_entry(_op, &unoptimizing_list, list)
+		if (op == _op)
+			return true;
+	return false;
+}
+
 /* Optimize kprobe if p is ready to be optimized */
 static void optimize_kprobe(struct kprobe *p)
 {
@@ -633,17 +643,21 @@ static void optimize_kprobe(struct kprobe *p)
 		return;
 
 	/* Check if it is already optimized. */
-	if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
+	if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
+		if (optprobe_queued_unopt(op)) {
+			/* This is under unoptimizing. Just dequeue the probe */
+			list_del_init(&op->list);
+		}
 		return;
+	}
 	op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 
-	if (!list_empty(&op->list))
-		/* This is under unoptimizing. Just dequeue the probe */
-		list_del_init(&op->list);
-	else {
-		list_add(&op->list, &optimizing_list);
-		kick_kprobe_optimizer();
-	}
+	/* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
+	if (WARN_ON_ONCE(!list_empty(&op->list)))
+		return;
+
+	list_add(&op->list, &optimizing_list);
+	kick_kprobe_optimizer();
 }
 
 /* Short cut to direct unoptimizing */
@@ -665,30 +679,33 @@ static void unoptimize_kprobe(struct kprobe *p, bool force)
 		return; /* This is not an optprobe nor optimized */
 
 	op = container_of(p, struct optimized_kprobe, kp);
-	if (!kprobe_optimized(p)) {
-		/* Unoptimized or unoptimizing case */
-		if (force && !list_empty(&op->list)) {
-			/*
-			 * Only if this is unoptimizing kprobe and forced,
-			 * forcibly unoptimize it. (No need to unoptimize
-			 * unoptimized kprobe again :)
-			 */
-			list_del_init(&op->list);
-			force_unoptimize_kprobe(op);
-		}
+	if (!kprobe_optimized(p))
 		return;
-	}
 
 	if (!list_empty(&op->list)) {
-		/* Dequeue from the optimization queue */
-		list_del_init(&op->list);
+		if (optprobe_queued_unopt(op)) {
+			/* Queued in unoptimizing queue */
+			if (force) {
+				/*
+				 * Forcibly unoptimize probe here, and queue it
+				 * in freeing list for release probe afterwards.
+				 */
+				force_unoptimize_kprobe(op);
+				list_move(&op->list, &freeing_list);
+			}
+		} else {
+			/* Dequeue from the optimizing queue */
+			list_del_init(&op->list);
+			op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
+		}
 		return;
 	}
+
 	/* Optimized kprobe case */
-	if (force)
+	if (force) {
 		/* Forcibly update the code: this is a special case */
 		force_unoptimize_kprobe(op);
-	else {
+	} else {
 		list_add(&op->list, &unoptimizing_list);
 		kick_kprobe_optimizer();
 	}
Reviewed-by on v1 1/1
#1 lore raw
Quoted 28 lines · On Tue, 7 Jan 2020 23:42:24 +0900 Masami Hiramatsu <[email protected]> wrote:

optimize_kprobe() and unoptimize_kprobe() cancels if given kprobe is on the optimizing_list or unoptimizing_list. However, since commit f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code") modified the update timing of the KPROBE_FLAG_OPTIMIZED, it doesn't work as expected anymore.

The optimized_kprobe could be following states.

- [optimizing]: Before inserting jump instruction
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [optimized]: jump inserted
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is empty.

- [unoptimizing]: Before removing jump instruction (including unused optprobe)
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [unoptimized]: jump removed
op.kp->flags doesn't have KPROBE_FLAG_OPTIMIZED and op->list is empty.

Current code mis-expects [unoptimizing] state doesn't have KPROBE_FLAG_OPTIMIZED, and that can cause wrong results.

Quoted 6 lines

This introduces optprobe_queued_unopt() to distinguish [optimizing] and [unoptimizing] states and fixes logics in optimize_kprobe() and unoptimize_kprobe().

Fixes f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code")

Looks good.

kernel/kprobes.c static void unoptimize_kprobe(struct kprobe *p, bool force) · line 711

Quoted 13 lines
kernel/kprobes.c +3 -2
 		return;
 	}
+
 	/* Optimized kprobe case */
-	if (force)
+	if (force) {
 		/* Forcibly update the code: this is a special case */
 		force_unoptimize_kprobe(op);
-	else {
+	} else {
 		list_add(&op->list, &unoptimizing_list);
 		kick_kprobe_optimizer();
 	}

I see you added some clean up to this patch.

-- Steve

author maintainer
#2 lore raw
Quoted 35 lines · On Tue, 7 Jan 2020 18:39:07 -0500 Steven Rostedt <[email protected]> wrote:
Quoted 32 lines · On Tue, 7 Jan 2020 23:42:24 +0900 Masami Hiramatsu <[email protected]> wrote:

optimize_kprobe() and unoptimize_kprobe() cancels if given kprobe is on the optimizing_list or unoptimizing_list. However, since commit f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code") modified the update timing of the KPROBE_FLAG_OPTIMIZED, it doesn't work as expected anymore.

The optimized_kprobe could be following states.

- [optimizing]: Before inserting jump instruction
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [optimized]: jump inserted
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is empty.

- [unoptimizing]: Before removing jump instruction (including unused optprobe)
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [unoptimized]: jump removed
op.kp->flags doesn't have KPROBE_FLAG_OPTIMIZED and op->list is empty.

Current code mis-expects [unoptimizing] state doesn't have KPROBE_FLAG_OPTIMIZED, and that can cause wrong results.

This introduces optprobe_queued_unopt() to distinguish [optimizing] and [unoptimizing] states and fixes logics in optimize_kprobe() and unoptimize_kprobe().

Quoted 6 lines
Quoted 2 lines
Fixes f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code")

Looks good.

Thank you!

Quoted 17 lines
Quoted 5 lines
kernel/kprobes.c +1 -0
 		return;
 	}
+

kernel/kprobes.c static void unoptimize_kprobe(struct kprobe *p, bool force) · line 711

Quoted 10 lines
kernel/kprobes.c +2 -2
 	/* Optimized kprobe case */
-	if (force)
+	if (force) {
 		/* Forcibly update the code: this is a special case */
 		force_unoptimize_kprobe(op);
-	else {
+	} else {
 		list_add(&op->list, &unoptimizing_list);
 		kick_kprobe_optimizer();
 	}

I see you added some clean up to this patch.

Yeah, I felt somewhat uncomfortable for that.

Quoted 2 lines

-- Steve

Signature · 1 line
Masami Hiramatsu <[email protected]>
#3 lore raw

[tip: core/kprobes] kprobes: Fix optimize_kprobe()/unoptimize_kprobe() cancellation logic

In reply to Masami Hiramatsu
The following commit has been merged into the core/kprobes branch of tip:

The following commit has been merged into the core/kprobes branch of tip:

Commit-ID:     e4add247789e4ba5e08ad8256183ce2e211877d4
Gitweb:        https://git.kernel.org/tip/e4add247789e4ba5e08ad8256183ce2e211877d4
Author:        Masami Hiramatsu <[email protected]>
AuthorDate:    Tue, 07 Jan 2020 23:42:24 +09:00
Committer:     Ingo Molnar <[email protected]>
CommitterDate: Thu, 09 Jan 2020 12:40:13 +01:00

kprobes: Fix optimize_kprobe()/unoptimize_kprobe() cancellation logic

optimize_kprobe() and unoptimize_kprobe() cancels if a given kprobe is on the optimizing_list or unoptimizing_list already. However, since the following commit:

f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code")

modified the update timing of the KPROBE_FLAG_OPTIMIZED, it doesn't work as expected anymore.

The optimized_kprobe could be in the following states:

- [optimizing]: Before inserting jump instruction
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [optimized]: jump inserted
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is empty.

- [unoptimizing]: Before removing jump instruction (including unused optprobe)
op.kp->flags has KPROBE_FLAG_OPTIMIZED and
op->list is not empty.

- [unoptimized]: jump removed
op.kp->flags doesn't have KPROBE_FLAG_OPTIMIZED and op->list is empty.

Current code mis-expects [unoptimizing] state doesn't have KPROBE_FLAG_OPTIMIZED, and that can cause incorrect results.

To fix this, introduce optprobe_queued_unopt() to distinguish [optimizing] and [unoptimizing] states and fixes the logic in optimize_kprobe() and unoptimize_kprobe().

[ mingo: Cleaned up the changelog and the code a bit. ]

Cc Alexei Starovoitov <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Fixes f66c0447cca1 ("kprobes: Set unoptimized flag after unoptimizing code")
---
 kernel/kprobes.c | 67 ++++++++++++++++++++++++++++++-----------------
 1 file changed, 43 insertions(+), 24 deletions(-)
kernel/kprobes.c +43 -24
index 34e28b2..2625c24 100644
@@ -612,6 +612,18 @@ void wait_for_kprobe_optimizer(void)
 	mutex_unlock(&kprobe_mutex);
 }
 
+static bool optprobe_queued_unopt(struct optimized_kprobe *op)
+{
+	struct optimized_kprobe *_op;
+
+	list_for_each_entry(_op, &unoptimizing_list, list) {
+		if (op == _op)
+			return true;
+	}
+
+	return false;
+}
+
 /* Optimize kprobe if p is ready to be optimized */
 static void optimize_kprobe(struct kprobe *p)
 {
@@ -633,17 +645,21 @@ static void optimize_kprobe(struct kprobe *p)
 		return;
 
 	/* Check if it is already optimized. */
-	if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
+	if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
+		if (optprobe_queued_unopt(op)) {
+			/* This is under unoptimizing. Just dequeue the probe */
+			list_del_init(&op->list);
+		}
 		return;
+	}
 	op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 
-	if (!list_empty(&op->list))
-		/* This is under unoptimizing. Just dequeue the probe */
-		list_del_init(&op->list);
-	else {
-		list_add(&op->list, &optimizing_list);
-		kick_kprobe_optimizer();
-	}
+	/* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
+	if (WARN_ON_ONCE(!list_empty(&op->list)))
+		return;
+
+	list_add(&op->list, &optimizing_list);
+	kick_kprobe_optimizer();
 }
 
 /* Short cut to direct unoptimizing */
@@ -665,30 +681,33 @@ static void unoptimize_kprobe(struct kprobe *p, bool force)
 		return; /* This is not an optprobe nor optimized */
 
 	op = container_of(p, struct optimized_kprobe, kp);
-	if (!kprobe_optimized(p)) {
-		/* Unoptimized or unoptimizing case */
-		if (force && !list_empty(&op->list)) {
-			/*
-			 * Only if this is unoptimizing kprobe and forced,
-			 * forcibly unoptimize it. (No need to unoptimize
-			 * unoptimized kprobe again :)
-			 */
-			list_del_init(&op->list);
-			force_unoptimize_kprobe(op);
-		}
+	if (!kprobe_optimized(p))
 		return;
-	}
 
 	if (!list_empty(&op->list)) {
-		/* Dequeue from the optimization queue */
-		list_del_init(&op->list);
+		if (optprobe_queued_unopt(op)) {
+			/* Queued in unoptimizing queue */
+			if (force) {
+				/*
+				 * Forcibly unoptimize the kprobe here, and queue it
+				 * in the freeing list for release afterwards.
+				 */
+				force_unoptimize_kprobe(op);
+				list_move(&op->list, &freeing_list);
+			}
+		} else {
+			/* Dequeue from the optimizing queue */
+			list_del_init(&op->list);
+			op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
+		}
 		return;
 	}
+
 	/* Optimized kprobe case */
-	if (force)
+	if (force) {
 		/* Forcibly update the code: this is a special case */
 		force_unoptimize_kprobe(op);
-	else {
+	} else {
 		list_add(&op->list, &unoptimizing_list);
 		kick_kprobe_optimizer();
 	}