[PATCH] kprobes: Skip disarmed probes when checking optkprobe overlap

[PATCH] kprobes: Skip disarmed probes when checking optkprobe overlap

[Leon Hwang]

On x86, an optkprobe at A replaces five bytes with a jump. If a disabled
probe B is at A+2, get_optimized_kprobe() stops at B when arming a new
probe C at A+4. It leaves A optimized:

              A    A+1  A+2  A+3  A+4
  A's jump  | e9 | d0 | d1 | d2 | d3 |
  after C   | e9 | d0 | d1 | d2 | cc |

The INT3 for C overwrites the last byte of A's jump displacement, so
execution can jump to the wrong address. B can have prepared optinsns
while disarmed, but has no jump to unoptimize.

Continue past disarmed and unprepared probes to find the active optimized
probe before arming a probe in its jump.

Fixes: afd66255b9a4 ("kprobes: Introduce kprobes jump optimization")
Cc: stable@vger.kernel.org
Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
---
 kernel/kprobes.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 6337da5cab9e..e6f6e18fa1b5 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -495,14 +495,16 @@ static bool kprobe_queued(struct kprobe *p)
 static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
 {
 	int i;
-	struct kprobe *p = NULL;
+	struct kprobe *p;
 	struct optimized_kprobe *op;
 
 	/* Don't check i == 0, since that is a breakpoint case. */
-	for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
+	for (i = 1; i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++) {
 		p = get_kprobe(addr - i);
+		/* A disabled probe can have prepared, but inactive, optinsns. */
+		if (!p || !kprobe_optready(p) || kprobe_disarmed(p))
+			continue;
 
-	if (p && kprobe_optready(p)) {
 		op = container_of(p, struct optimized_kprobe, kp);
 		if (arch_within_optimized_kprobe(op, addr))
 			return p;
-- 
2.55.0

Re: [PATCH] kprobes: Skip disarmed probes when checking optkprobe overlap

[Masami Hiramatsu]

On Wed, 30 Sep 2026 13:36:18 +0800
Leon Hwang <leon.hwang@linux.dev> wrote:

> On x86, an optkprobe at A replaces five bytes with a jump. If a disabled
> probe B is at A+2, get_optimized_kprobe() stops at B when arming a new
> probe C at A+4. It leaves A optimized:
> 
>               A    A+1  A+2  A+3  A+4
>   A's jump  | e9 | d0 | d1 | d2 | d3 |
>   after C   | e9 | d0 | d1 | d2 | cc |
> 
> The INT3 for C overwrites the last byte of A's jump displacement, so
> execution can jump to the wrong address. B can have prepared optinsns
> while disarmed, but has no jump to unoptimize.

Oops, good catch! get_optimized_kprobe() only checks the first disabled
one and do not unoptimize the A.

> 
> Continue past disarmed and unprepared probes to find the active optimized
> probe before arming a probe in its jump.

Yeah, this looks good to me. Let me send via probes/fixes.

Thank you!!

> 
> Fixes: afd66255b9a4 ("kprobes: Introduce kprobes jump optimization")
> Cc: stable@vger.kernel.org
> Signed-off-by: Leon Hwang <leon.hwang@linux.dev>
> ---
>  kernel/kprobes.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> index 6337da5cab9e..e6f6e18fa1b5 100644
> --- a/kernel/kprobes.c
> +++ b/kernel/kprobes.c
> @@ -495,14 +495,16 @@ static bool kprobe_queued(struct kprobe *p)
>  static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
>  {
>  	int i;
> -	struct kprobe *p = NULL;
> +	struct kprobe *p;
>  	struct optimized_kprobe *op;
>  
>  	/* Don't check i == 0, since that is a breakpoint case. */
> -	for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
> +	for (i = 1; i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++) {
>  		p = get_kprobe(addr - i);
> +		/* A disabled probe can have prepared, but inactive, optinsns. */
> +		if (!p || !kprobe_optready(p) || kprobe_disarmed(p))
> +			continue;
>  
> -	if (p && kprobe_optready(p)) {
>  		op = container_of(p, struct optimized_kprobe, kp);
>  		if (arch_within_optimized_kprobe(op, addr))
>  			return p;
> -- 
> 2.55.0
> 


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>