io_ring_exit_work() hangs are hard to debug partly because there is not
much information of who created the ctx by the time it's exiting, and
the function is running in a wq context, so the task name tells us
nothing. Save creator's task comm and print it when it hangs.
Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
Just for discussion, I hope there are better ways of doing it.
It leaves out the second wait_for_completion() in io_ring_exit_work(),
which is of interest, so would be great to cover the case as well.
fs/io_uring.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 73135c5c6168..db0065637549 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -447,6 +447,8 @@ struct io_ring_ctx {
struct work_struct exit_work;
struct list_head tctx_list;
struct completion ref_comp;
+ /* save owner thread's comm for debugging purposes */
+ char owner_comm[TASK_COMM_LEN];
};
};
@@ -9344,7 +9346,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)
io_req_caches_free(ctx);
- if (WARN_ON_ONCE(time_after(jiffies, timeout))) {
+ if (WARN_ONCE(time_after(jiffies, timeout), "comm %s\n",
+ ctx->owner_comm)) {
/* there is little hope left, don't run it too often */
interval = HZ * 60;
}
@@ -10266,6 +10269,8 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
if (!capable(CAP_IPC_LOCK))
ctx->user = get_uid(current_user());
+ get_task_comm(ctx->owner_comm, current);
+
/*
* This is just grabbed for accounting purposes. When a process exits,
* the mm is exited and dropped before the files, hence we need to hang
--
2.33.0
