Re: WARNING in debug_mutex_unlock

From: Yonghong Song
Date: Tue Jan 17 2023 - 01:03:24 EST




On 1/15/23 6:24 PM, Hao Sun wrote:


On 13 Jan 2023, at 2:45 PM, Yonghong Song <yhs@xxxxxxxx> wrote:



On 1/8/23 11:44 PM, Hao Sun wrote:
Hi,
The following warning can be triggered with the C reproducer in
the link. The repro starts 32 threads, each attaches a tracepoint
into `ext4_mark_inode_dirty`. The prog loads the following insns
that simply sends signal to current proc, and then wait.
Seems issues in queued irq_work with `do_bpf_send_signal`, also
I'm wondering what if the task in `send_signal_irq_work` exited,
at the time the callback invoked.

Somehow, I cannot reproduce the issue in my qemu environment
with below kernel config and C reproducer.

But could you try the following patch to see whether it
fixed the issue in your environment?

Tested the below patch on my local machine, seems fixed the issue.

Before applying the patch, the reproducer can still trigger the
reported issue on a latest bpf-next build; After applying the
patch, the warning no longer appears.

The test is conducted on: dfff86f8eb6a (“Merge branch 'samples/bpf:
modernize BPF functionality test programs'")

Thanks for testing. I will submit a patch shortly with your
Reported-by and Tested-by.




diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 23ce498bca97..1b26d51caf31 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -831,6 +831,7 @@ static void do_bpf_send_signal(struct irq_work *entry)

work = container_of(entry, struct send_signal_irq_work, irq_work);
group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
+ put_task_struct(work->task);
}

static int bpf_send_signal_common(u32 sig, enum pid_type type)
@@ -862,7 +863,7 @@ static int bpf_send_signal_common(u32 sig, enum pid_type type)
* to the irq_work. The current task may change when queued
* irq works get executed.
*/
- work->task = current;
+ work->task = get_task_struct(current);
work->sig = sig;
work->type = type;
irq_work_queue(&work->irq_work);