On Thu, 6 Jul 2023 12:09:16 -0400
Steven Rostedt [off-list ref] wrote:

On Mon,  3 Jul 2023 17:23:36 +0800
Ze Gao [off-list ref] wrote:

quoted

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 <redacted>
Closes: https://lore.kernel.org/linux-trace-kernel/CALOAHbC6UpfFOOibdDiC7xFc5YFUgZnk3MZ=3Ny6we=AcrNbew@mail.gmail.com/ (local)
Signed-off-by: Ze Gao <redacted>

Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>

quoted

---
 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

quoted

 		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) [off-list ref]