2023-07-03 09:46:56

by Ze Gao

Subject: [PATCH v2] fprobe: add unlock to match a succeeded ftrace_test_recursion_trylock

Unlock ftrace recursion lock when fprobe_kprobe_handler() is failed
because of some running kprobe.

Fixes: 3cc4e2c5fbae ("fprobe: make fprobe_kprobe_handler recursion free")
Reported-by: Yafang <[email protected]>
Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/
Signed-off-by: Ze Gao <[email protected]>
---
kernel/trace/fprobe.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
index 18d36842faf5..93b3e361bb97 100644
--- a/kernel/trace/fprobe.c
+++ b/kernel/trace/fprobe.c
@@ -102,12 +102,14 @@ static void fprobe_kprobe_handler(unsigned long ip, unsigned long parent_ip,

if (unlikely(kprobe_running())) {
fp->nmissed++;
- return;
+ goto recursion_unlock;
}

kprobe_busy_begin();
__fprobe_handler(ip, parent_ip, ops, fregs);
kprobe_busy_end();
+
+recursion_unlock:
ftrace_test_recursion_unlock(bit);
}

--
2.40.1

2023-07-03 10:19:23

by Masami Hiramatsu

Subject: Re: [PATCH v2] fprobe: add unlock to match a succeeded ftrace_test_recursion_trylock

On Mon, 3 Jul 2023 17:23:36 +0800
Ze Gao <[email protected]> wrote:

> Unlock ftrace recursion lock when fprobe_kprobe_handler() is failed
> because of some running kprobe.
>

Acked-by: Masami Hiramatsu (Google) <[email protected]>

Thanks!

> Fixes: 3cc4e2c5fbae ("fprobe: make fprobe_kprobe_handler recursion free")
> Reported-by: Yafang <[email protected]>
> Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/
> Signed-off-by: Ze Gao <[email protected]>
> ---
> kernel/trace/fprobe.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
> index 18d36842faf5..93b3e361bb97 100644
> --- a/kernel/trace/fprobe.c
> +++ b/kernel/trace/fprobe.c
> @@ -102,12 +102,14 @@ static void fprobe_kprobe_handler(unsigned long ip, unsigned long parent_ip,
>
> if (unlikely(kprobe_running())) {
> fp->nmissed++;
> - return;
> + goto recursion_unlock;
> }
>
> kprobe_busy_begin();
> __fprobe_handler(ip, parent_ip, ops, fregs);
> kprobe_busy_end();
> +
> +recursion_unlock:
> ftrace_test_recursion_unlock(bit);
> }
>
> --
> 2.40.1
>


--
Masami Hiramatsu (Google) <[email protected]>

2023-07-03 14:15:40

by Yafang Shao

Subject: Re: [PATCH v2] fprobe: add unlock to match a succeeded ftrace_test_recursion_trylock

On Mon, Jul 3, 2023 at 5:23 PM Ze Gao <[email protected]> wrote:
>
> Unlock ftrace recursion lock when fprobe_kprobe_handler() is failed
> because of some running kprobe.
>
> Fixes: 3cc4e2c5fbae ("fprobe: make fprobe_kprobe_handler recursion free")
> Reported-by: Yafang <[email protected]>
> Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/
> Signed-off-by: Ze Gao <[email protected]>

Acked-by: Yafang Shao <[email protected]>

> ---
> kernel/trace/fprobe.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
> index 18d36842faf5..93b3e361bb97 100644
> --- a/kernel/trace/fprobe.c
> +++ b/kernel/trace/fprobe.c
> @@ -102,12 +102,14 @@ static void fprobe_kprobe_handler(unsigned long ip, unsigned long parent_ip,
>
> if (unlikely(kprobe_running())) {
> fp->nmissed++;
> - return;
> + goto recursion_unlock;
> }
>
> kprobe_busy_begin();
> __fprobe_handler(ip, parent_ip, ops, fregs);
> kprobe_busy_end();
> +
> +recursion_unlock:
> ftrace_test_recursion_unlock(bit);
> }
>
> --
> 2.40.1
>


--
Regards
Yafang

2023-07-06 16:25:07

by Steven Rostedt

Subject: Re: [PATCH v2] fprobe: add unlock to match a succeeded ftrace_test_recursion_trylock

On Mon, 3 Jul 2023 17:23:36 +0800
Ze Gao <[email protected]> wrote:

> Unlock ftrace recursion lock when fprobe_kprobe_handler() is failed
> because of some running kprobe.
>
> Fixes: 3cc4e2c5fbae ("fprobe: make fprobe_kprobe_handler recursion free")
> Reported-by: Yafang <[email protected]>
> Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/
> Signed-off-by: Ze Gao <[email protected]>

Reviewed-by: Steven Rostedt (Google) <[email protected]>

> ---
> kernel/trace/fprobe.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
> index 18d36842faf5..93b3e361bb97 100644
> --- a/kernel/trace/fprobe.c
> +++ b/kernel/trace/fprobe.c
> @@ -102,12 +102,14 @@ static void fprobe_kprobe_handler(unsigned long ip, unsigned long parent_ip,
>
> if (unlikely(kprobe_running())) {

Off topic for this patch, but Masami, what's the purpose of not calling the
fprobe when a kprobe is running? Does that mean it has probed another kprobe?

Probably could add a comment here to explain the issue.

-- Steve


> fp->nmissed++;
> - return;
> + goto recursion_unlock;
> }
>
> kprobe_busy_begin();
> __fprobe_handler(ip, parent_ip, ops, fregs);
> kprobe_busy_end();
> +
> +recursion_unlock:
> ftrace_test_recursion_unlock(bit);
> }
>

2023-07-07 00:39:02

by Masami Hiramatsu

Subject: Re: [PATCH v2] fprobe: add unlock to match a succeeded ftrace_test_recursion_trylock

On Thu, 6 Jul 2023 12:09:16 -0400
Steven Rostedt <[email protected]> wrote:

> On Mon, 3 Jul 2023 17:23:36 +0800
> Ze Gao <[email protected]> wrote:
>
> > Unlock ftrace recursion lock when fprobe_kprobe_handler() is failed
> > because of some running kprobe.
> >
> > Fixes: 3cc4e2c5fbae ("fprobe: make fprobe_kprobe_handler recursion free")
> > Reported-by: Yafang <[email protected]>
> > Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/
> > Signed-off-by: Ze Gao <[email protected]>
>
> Reviewed-by: Steven Rostedt (Google) <[email protected]>
>
> > ---
> > kernel/trace/fprobe.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/trace/fprobe.c b/kernel/trace/fprobe.c
> > index 18d36842faf5..93b3e361bb97 100644
> > --- a/kernel/trace/fprobe.c
> > +++ b/kernel/trace/fprobe.c
> > @@ -102,12 +102,14 @@ static void fprobe_kprobe_handler(unsigned long ip, unsigned long parent_ip,
> >
> > if (unlikely(kprobe_running())) {
>
> Off topic for this patch, but Masami, what's the purpose of not calling the
> fprobe when a kprobe is running? Does that mean it has probed another kprobe?

This is for the user who is sharing their handler with kprobes (like eBPF),
which may expect that the handler is not called recursively. (e.g. an interrupt
happens while kprobe handler is running and that interrupt calls a function
which is fprobed)

>
> Probably could add a comment here to explain the issue.

OK, it is also documented in Documentation/trace/fprobe.rst, but it is better
to comment in the code too.

Thanks,

>
> -- Steve
>
>
> > fp->nmissed++;
> > - return;
> > + goto recursion_unlock;
> > }
> >
> > kprobe_busy_begin();
> > __fprobe_handler(ip, parent_ip, ops, fregs);
> > kprobe_busy_end();
> > +
> > +recursion_unlock:
> > ftrace_test_recursion_unlock(bit);
> > }
> >
>


--
Masami Hiramatsu (Google) <[email protected]>