From: Chuang Wang <[email protected]>
In aggrprobe scenes, if arm_kprobe() returns an error(e.g. livepatch and
kprobe are using the same function X), kprobe flags, while has been
modified to ~KPROBE_FLAG_DISABLED, is not rollled back.
Then, __disable_kprobe() will be failed in __unregister_kprobe_top(),
the kprobe list will be not removed from aggrprobe, memory leaks or
illegal pointers will be caused.
WARN disarm_kprobe:
Failed to disarm kprobe-ftrace at 00000000c729fdbc (-2)
RIP: 0010:disarm_kprobe+0xcc/0x110
Call Trace:
__disable_kprobe+0x78/0x90
__unregister_kprobe_top+0x13/0x1b0
? _cond_resched+0x15/0x30
unregister_kprobes+0x32/0x80
unregister_kprobe+0x1a/0x20
Illegal Pointers:
BUG: unable to handle kernel paging request at 0000000000656369
RIP: 0010:__get_valid_kprobe+0x69/0x90
Call Trace:
register_kprobe+0x30/0x60
__register_trace_kprobe.part.7+0x8b/0xc0
create_local_trace_kprobe+0xd2/0x130
perf_kprobe_init+0x83/0xd0
Signed-off-by: Chuang Wang <[email protected]>
Signed-off-by: Jingren Zhou <[email protected]>
---
kernel/kprobes.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index f214f8c088ed..96c75e23113c 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -2422,8 +2422,10 @@ int enable_kprobe(struct kprobe *kp)
if (!kprobes_all_disarmed && kprobe_disabled(p)) {
p->flags &= ~KPROBE_FLAG_DISABLED;
ret = arm_kprobe(p);
- if (ret)
+ if (ret) {
p->flags |= KPROBE_FLAG_DISABLED;
+ kp->flags |= KPROBE_FLAG_DISABLED;
+ }
}
out:
mutex_unlock(&kprobe_mutex);
--
2.34.1
Hi Chuang,
On Thu, 2 Jun 2022 15:32:59 +0800
Chuang <[email protected]> wrote:
> From: Chuang Wang <[email protected]>
>
> In aggrprobe scenes, if arm_kprobe() returns an error(e.g. livepatch and
> kprobe are using the same function X), kprobe flags, while has been
> modified to ~KPROBE_FLAG_DISABLED, is not rollled back.
>
> Then, __disable_kprobe() will be failed in __unregister_kprobe_top(),
> the kprobe list will be not removed from aggrprobe, memory leaks or
> illegal pointers will be caused.
>
> WARN disarm_kprobe:
> Failed to disarm kprobe-ftrace at 00000000c729fdbc (-2)
> RIP: 0010:disarm_kprobe+0xcc/0x110
> Call Trace:
> __disable_kprobe+0x78/0x90
> __unregister_kprobe_top+0x13/0x1b0
> ? _cond_resched+0x15/0x30
> unregister_kprobes+0x32/0x80
> unregister_kprobe+0x1a/0x20
>
> Illegal Pointers:
> BUG: unable to handle kernel paging request at 0000000000656369
> RIP: 0010:__get_valid_kprobe+0x69/0x90
> Call Trace:
> register_kprobe+0x30/0x60
> __register_trace_kprobe.part.7+0x8b/0xc0
> create_local_trace_kprobe+0xd2/0x130
> perf_kprobe_init+0x83/0xd0
Oops, thanks for reporting!
> Signed-off-by: Chuang Wang <[email protected]>
> Signed-off-by: Jingren Zhou <[email protected]>
This should go to stable, so add below tag. (No need to CC to stable)
Fixes: 12310e343755 ("kprobes: Propagate error from arm_kprobe_ftrace()")
Cc: [email protected]
And could you also update this patch as below?
> ---
> kernel/kprobes.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> index f214f8c088ed..96c75e23113c 100644
> --- a/kernel/kprobes.c
> +++ b/kernel/kprobes.c
> @@ -2422,8 +2422,10 @@ int enable_kprobe(struct kprobe *kp)
> if (!kprobes_all_disarmed && kprobe_disabled(p)) {
> p->flags &= ~KPROBE_FLAG_DISABLED;
> ret = arm_kprobe(p);
> - if (ret)
> + if (ret) {
> p->flags |= KPROBE_FLAG_DISABLED;
Here, can you add a check?
if (p != kp)
> + kp->flags |= KPROBE_FLAG_DISABLED;
Thus is is clear that this is corresponding to
---
if (p != kp)
kp->flags &= ~KPROBE_FLAG_DISABLED;
---
Thank you,
> + }
> }
> out:
> mutex_unlock(&kprobe_mutex);
> --
> 2.34.1
>
--
Masami Hiramatsu (Google) <[email protected]>
Thanks for your quick reply.
I'm very, very sorry for sending multiple emails. I am submitting a
patch for the first time.
On Fri, Jun 3, 2022 at 11:03 PM Masami Hiramatsu <[email protected]> wrote:
>
> This should go to stable, so add below tag. (No need to CC to stable)
>
> Fixes: 12310e343755 ("kprobes: Propagate error from arm_kprobe_ftrace()")
> Cc: [email protected]
Thanks for your kind reminder.
>
> And could you also update this patch as below?
>
> > ---
> > kernel/kprobes.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> > index f214f8c088ed..96c75e23113c 100644
> > --- a/kernel/kprobes.c
> > +++ b/kernel/kprobes.c
> > @@ -2422,8 +2422,10 @@ int enable_kprobe(struct kprobe *kp)
> > if (!kprobes_all_disarmed && kprobe_disabled(p)) {
> > p->flags &= ~KPROBE_FLAG_DISABLED;
> > ret = arm_kprobe(p);
> > - if (ret)
> > + if (ret) {
> > p->flags |= KPROBE_FLAG_DISABLED;
>
> Here, can you add a check?
>
> if (p != kp)
>
> > + kp->flags |= KPROBE_FLAG_DISABLED;
Well, I also thought about it. This already covers conditions `p ==
kp` and `p != kp`:
---
kp->flags |= KPROBE_FLAG_DISABLED;
---
When p is equal to kp, `kp->flags` is assigned to KPROBE_FLAG_DISABLED twice.
However, if you add a check that p is not equal to kp, this makes the
code read more clearly. Anything is OK, what is your suggestion?
>
> Thus is is clear that this is corresponding to
> ---
> if (p != kp)
> kp->flags &= ~KPROBE_FLAG_DISABLED;
> ---
>
> Thank you,
>
> > + }
> > }
> > out:
> > mutex_unlock(&kprobe_mutex);
> > --
> > 2.34.1
> >
>
>
> --
> Masami Hiramatsu (Google) <[email protected]>
Have a wonderful day!