Merge branch 'bpf-fix-conditions-when-timer-wq-can-be-called'

Alexei Starovoitov says:

====================
bpf: Fix conditions when timer/wq can be called

From: Alexei Starovoitov <ast@kernel.org>

v2->v3:
- Add missing refcount_put
- Detect recursion of indiviual async_cb
v2: https://lore.kernel.org/bpf/20260204040834.22263-4-alexei.starovoitov@gmail.com/

v1->v2:
- Add a recursion check
v1: https://lore.kernel.org/bpf/20260204030927.171-1-alexei.starovoitov@gmail.com/
====================

Link: https://patch.msgid.link/20260204055147.54960-1-alexei.starovoitov@gmail.com
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>

2 files changed, 103 insertions, 0 deletions

diff --git a/tools/testing/selftests/bpf/prog_tests/timer_start_deadlock.c b/tools/testing/selftests/bpf/prog_tests/timer_start_deadlock.c
new file mode 100644
index 000000000000..9f1f9aec8888
--- /dev/null
+++ b/tools/testing/selftests/bpf/prog_tests/timer_start_deadlock.c
@@ -0,0 +1,33 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#include <test_progs.h>
+#include "timer_start_deadlock.skel.h"
+
+void test_timer_start_deadlock(void)
+{
+	struct timer_start_deadlock *skel;
+	int err, prog_fd;
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+
+	skel = timer_start_deadlock__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "skel_open_and_load"))
+		return;
+
+	err = timer_start_deadlock__attach(skel);
+	if (!ASSERT_OK(err, "skel_attach"))
+		goto cleanup;
+
+	prog_fd = bpf_program__fd(skel->progs.start_timer);
+
+	/*
+	 * Run the syscall program that attempts to deadlock.
+	 * If the kernel deadlocks, this call will never return.
+	 */
+	err = bpf_prog_test_run_opts(prog_fd, &opts);
+	ASSERT_OK(err, "prog_test_run");
+	ASSERT_EQ(opts.retval, 0, "prog_retval");
+
+	ASSERT_EQ(skel->bss->tp_called, 1, "tp_called");
+cleanup:
+	timer_start_deadlock__destroy(skel);
+}
diff --git a/tools/testing/selftests/bpf/progs/timer_start_deadlock.c b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
new file mode 100644
index 000000000000..019518ee18cd
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/timer_start_deadlock.c
@@ -0,0 +1,70 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2026 Meta Platforms, Inc. and affiliates. */
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+#define CLOCK_MONOTONIC 1
+
+char _license[] SEC("license") = "GPL";
+
+struct elem {
+	struct bpf_timer timer;
+};
+
+struct {
+	__uint(type, BPF_MAP_TYPE_ARRAY);
+	__uint(max_entries, 1);
+	__type(key, int);
+	__type(value, struct elem);
+} timer_map SEC(".maps");
+
+volatile int in_timer_start;
+volatile int tp_called;
+
+static int timer_cb(void *map, int *key, struct elem *value)
+{
+	return 0;
+}
+
+SEC("tp_btf/hrtimer_cancel")
+int BPF_PROG(tp_hrtimer_cancel, struct hrtimer *hrtimer)
+{
+	struct bpf_timer *timer;
+	int key = 0;
+
+	if (!in_timer_start)
+		return 0;
+
+	tp_called = 1;
+	timer = bpf_map_lookup_elem(&timer_map, &key);
+
+	/*
+	 * Call bpf_timer_start() from the tracepoint within hrtimer logic
+	 * on the same timer to make sure it doesn't deadlock.
+	 */
+	bpf_timer_start(timer, 1000000000, 0);
+	return 0;
+}
+
+SEC("syscall")
+int start_timer(void *ctx)
+{
+	struct bpf_timer *timer;
+	int key = 0;
+
+	timer = bpf_map_lookup_elem(&timer_map, &key);
+	/* claude may complain here that there is no NULL check. Ignoring it. */
+	bpf_timer_init(timer, &timer_map, CLOCK_MONOTONIC);
+	bpf_timer_set_callback(timer, timer_cb);
+
+	/*
+	 * call hrtimer_start() twice, so that 2nd call does
+	 * remove_hrtimer() and trace_hrtimer_cancel() tracepoint.
+	 */
+	in_timer_start = 1;
+	bpf_timer_start(timer, 1000000000, 0);
+	bpf_timer_start(timer, 1000000000, 0);
+	in_timer_start = 0;
+	return 0;
+}