Re: [PATCH bpf v1] bpf: cpumap: fix race in bq_flush_to_queue on PREEMPT_RT
From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: 2026-02-11 11:44:21
Also in:
bpf, linux-hardening, linux-rt-devel, lkml
On 2026-02-11 14:44:16 [+0800], Jiayuan Chen wrote:
From: Jiayuan Chen <redacted> On PREEMPT_RT kernels, the per-CPU xdp_bulk_queue (bq) can be accessed concurrently by multiple preemptible tasks on the same CPU. The original code assumes bq_enqueue() and __cpu_map_flush() run atomically with respect to each other on the same CPU, relying on local_bh_disable() to prevent preemption. However, on PREEMPT_RT, local_bh_disable() only calls migrate_disable() and does not disable preemption. spin_lock() also becomes a sleeping rt_mutex. Together, this allows CFS scheduling to preempt a task during bq_flush_to_queue(), enabling another task on the same CPU to enter bq_enqueue() and operate on the same per-CPU bq concurrently.
…
Fixes: d2d6422f8bd1 ("x86: Allow to enable PREEMPT_RT.")
Can you reproduce this? It should not trigger with the commit above.
It should trigger starting with
3253cb49cbad4 ("softirq: Allow to drop the softirq-BKL lock on PREEMPT_RT")
quoted hunk ↗ jump to hunk
Reported-by: syzbot+2b3391f44313b3983e91@syzkaller.appspotmail.com Closes: https://lore.kernel.org/all/69369331.a70a0220.38f243.009d.GAE@google.com/T/ (local) Signed-off-by: Jiayuan Chen <redacted> Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev> --- kernel/bpf/cpumap.c | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-)diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c index 04171fbc39cb..7fda8421ec40 100644 --- a/kernel/bpf/cpumap.c +++ b/kernel/bpf/cpumap.c@@ -714,6 +717,7 @@ const struct bpf_map_ops cpu_map_ops = { .map_redirect = cpu_map_redirect, }; +/* Caller must hold bq->bq_lock */
If this information is important please use lockdep_assert_held() in the function below. This can be used by lockdep and is understood by humans while the comment is only visible to humans.
quoted hunk ↗ jump to hunk
static void bq_flush_to_queue(struct xdp_bulk_queue *bq) { struct bpf_cpu_map_entry *rcpu = bq->obj;@@ -750,10 +754,16 @@ static void bq_flush_to_queue(struct xdp_bulk_queue *bq) /* Runs under RCU-read-side, plus in softirq under NAPI protection. * Thus, safe percpu variable access.
+ PREEMPT_RT relies on local_lock_nested_bh().
+ *
+ * On PREEMPT_RT, local_bh_disable() does not disable preemption,
+ * so we use local_lock to serialize access to the per-CPU bq.
*/
static void bq_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf)
{
- struct xdp_bulk_queue *bq = this_cpu_ptr(rcpu->bulkq);
+ struct xdp_bulk_queue *bq;
+
+ local_lock(&rcpu->bulkq->bq_lock);local_lock_nested_bh() & the matching unlock here and in the other places, please. Sebastian