[PATCH bpf v4 0/2] bpf: cpumap/devmap: fix per-CPU bulk queue races on PREEMPT_RT

[PATCH bpf v4 0/2] bpf: cpumap/devmap: fix per-CPU bulk queue races on PREEMPT_RT

[Jiayuan Chen]

On PREEMPT_RT kernels, local_bh_disable() only calls migrate_disable()
(when PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
preemption. This means CFS scheduling can preempt a task inside the
per-CPU bulk queue (bq) operations in cpumap and devmap, allowing
another task on the same CPU to concurrently access the same bq,
leading to use-after-free, list corruption, and kernel panics.

Patch 1 fixes the cpumap race in bq_flush_to_queue(), originally
reported by syzbot [1].

Patch 2 fixes the same class of race in devmap's bq_xmit_all(),
identified by code inspection after Sebastian Andrzej Siewior pointed
out that devmap has the same per-CPU bulk queue pattern [2].

Both patches use local_lock_nested_bh() to serialize access to the
per-CPU bq. On non-RT this is a pure lockdep annotation with no
overhead; on PREEMPT_RT it provides a per-CPU sleeping lock.

To reproduce the devmap race, insert an mdelay(100) in bq_xmit_all()
after "cnt = bq->count" and before the actual transmit loop. Then pin
two threads to the same CPU, each running BPF_PROG_TEST_RUN with an XDP
program that redirects to a DEVMAP entry (e.g. a veth pair). CFS
timeslicing during the mdelay window causes interleaving. Without the
fix, KASAN reports null-ptr-deref due to operating on freed frames:

  BUG: KASAN: null-ptr-deref in __build_skb_around+0x22d/0x340
  Write of size 32 at addr 0000000000000d50 by task devmap_race_rep/449

  CPU: 0 UID: 0 PID: 449 Comm: devmap_race_rep Not tainted 6.19.0+ #31 PREEMPT_RT
  Call Trace:
   <TASK>
   __build_skb_around+0x22d/0x340
   build_skb_around+0x25/0x260
   __xdp_build_skb_from_frame+0x103/0x860
   veth_xdp_rcv_bulk_skb.isra.0+0x162/0x320
   veth_xdp_rcv.constprop.0+0x61e/0xbb0
   veth_poll+0x280/0xb50
   __napi_poll.constprop.0+0xa5/0x590
   net_rx_action+0x4b0/0xea0
   handle_softirqs.isra.0+0x1b3/0x780
   __local_bh_enable_ip+0x12a/0x240
   xdp_test_run_batch.constprop.0+0xedd/0x1f60
   bpf_test_run_xdp_live+0x304/0x640
   bpf_prog_test_run_xdp+0xd24/0x1b70
   __sys_bpf+0x61c/0x3e00
   </TASK>

  Kernel panic - not syncing: Fatal exception in interrupt

[1] https://lore.kernel.org/all/69369331.a70a0220.38f243.009d.GAE@google.com/T/
[2] https://lore.kernel.org/bpf/20260212023634.366343-1-jiayuan.chen@linux.dev/

---
v3 -> v4: https://lore.kernel.org/all/20260213034018.284146-1-jiayuan.chen@linux.dev/
- Move panic trace to cover letter. (Sebastian Andrzej Siewior)
- Add Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> to both patches
  from cover letter.

v2 -> v3: https://lore.kernel.org/bpf/20260212023634.366343-1-jiayuan.chen@linux.dev/
- Fix commit message: remove incorrect "spin_lock() becomes rt_mutex"
claim, the per-CPU bq has no spin_lock at all. (Sebastian Andrzej Siewior)
- Fix commit message: accurately describe local_lock_nested_bh()
behavior instead of referencing local_lock(). (Sebastian Andrzej Siewior)
- Remove incomplete discussion of snapshot alternative.
(Sebastian Andrzej Siewior)
- Remove panic trace from commit message. (Sebastian Andrzej Siewior)
- Add patch 2/2 for devmap, same race pattern. (Sebastian Andrzej Siewior)

v1 -> v2: https://lore.kernel.org/bpf/20260211064417.196401-1-jiayuan.chen@linux.dev/
- Use local_lock_nested_bh()/local_unlock_nested_bh() instead of
local_lock()/local_unlock(), since these paths already run under
local_bh_disable(). (Sebastian Andrzej Siewior)
- Replace "Caller must hold bq->bq_lock" comment with
lockdep_assert_held() in bq_flush_to_queue(). (Sebastian Andrzej Siewior)
- Fix Fixes tag to 3253cb49cbad ("softirq: Allow to drop the
softirq-BKL lock on PREEMPT_RT") which is the actual commit that
makes the race possible. (Sebastian Andrzej Siewior)

Jiayuan Chen (2):
  bpf: cpumap: fix race in bq_flush_to_queue on PREEMPT_RT
  bpf: devmap: fix race in bq_xmit_all on PREEMPT_RT

 kernel/bpf/cpumap.c | 17 +++++++++++++++--
 kernel/bpf/devmap.c | 25 +++++++++++++++++++++----
 2 files changed, 36 insertions(+), 6 deletions(-)

-- 
2.43.0

[PATCH bpf v4 1/2] bpf: cpumap: fix race in bq_flush_to_queue on PREEMPT_RT

[Jiayuan Chen]

From: Jiayuan Chen <jiayuan.chen@shopee.com>

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() (when
PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
preemption, which 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.

This leads to several races:

1. Double __list_del_clearprev(): after bq->count is reset in
   bq_flush_to_queue(), a preempting task can call bq_enqueue() ->
   bq_flush_to_queue() on the same bq when bq->count reaches
   CPU_MAP_BULK_SIZE. Both tasks then call __list_del_clearprev()
   on the same bq->flush_node, the second call dereferences the
   prev pointer that was already set to NULL by the first.

2. bq->count and bq->q[] races: concurrent bq_enqueue() can corrupt
   the packet queue while bq_flush_to_queue() is processing it.

The race between task A (__cpu_map_flush -> bq_flush_to_queue) and
task B (bq_enqueue -> bq_flush_to_queue) on the same CPU:

  Task A (xdp_do_flush)          Task B (cpu_map_enqueue)
  ----------------------         ------------------------
  bq_flush_to_queue(bq)
    spin_lock(&q->producer_lock)
    /* flush bq->q[] to ptr_ring */
    bq->count = 0
    spin_unlock(&q->producer_lock)
                                   bq_enqueue(rcpu, xdpf)
    <-- CFS preempts Task A -->      bq->q[bq->count++] = xdpf
                                     /* ... more enqueues until full ... */
                                     bq_flush_to_queue(bq)
                                       spin_lock(&q->producer_lock)
                                       /* flush to ptr_ring */
                                       spin_unlock(&q->producer_lock)
                                       __list_del_clearprev(flush_node)
                                         /* sets flush_node.prev = NULL */
    <-- Task A resumes -->
    __list_del_clearprev(flush_node)
      flush_node.prev->next = ...
      /* prev is NULL -> kernel oops */

Fix this by adding a local_lock_t to xdp_bulk_queue and acquiring it
in bq_enqueue() and __cpu_map_flush(). These paths already run under
local_bh_disable(), so use local_lock_nested_bh() which on non-RT is
a pure annotation with no overhead, and on PREEMPT_RT provides a
per-CPU sleeping lock that serializes access to the bq.

To reproduce, insert an mdelay(100) between bq->count = 0 and
__list_del_clearprev() in bq_flush_to_queue(), then run reproducer
provided by syzkaller.

Fixes: 3253cb49cbad ("softirq: Allow to drop the softirq-BKL lock on PREEMPT_RT")
Reported-by: syzbot+2b3391f44313b3983e91@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/69369331.a70a0220.38f243.009d.GAE@google.com/T/
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
---
 kernel/bpf/cpumap.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 04171fbc39cb..32b43cb9061b 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -29,6 +29,7 @@
 #include <linux/sched.h>
 #include <linux/workqueue.h>
 #include <linux/kthread.h>
+#include <linux/local_lock.h>
 #include <linux/completion.h>
 #include <trace/events/xdp.h>
 #include <linux/btf_ids.h>
@@ -52,6 +53,7 @@ struct xdp_bulk_queue {
 	struct list_head flush_node;
 	struct bpf_cpu_map_entry *obj;
 	unsigned int count;
+	local_lock_t bq_lock;
 };
 
 /* Struct for every remote "destination" CPU in map */
@@ -451,6 +453,7 @@ __cpu_map_entry_alloc(struct bpf_map *map, struct bpf_cpumap_val *value,
 	for_each_possible_cpu(i) {
 		bq = per_cpu_ptr(rcpu->bulkq, i);
 		bq->obj = rcpu;
+		local_lock_init(&bq->bq_lock);
 	}
 
 	/* Alloc queue */
@@ -722,6 +725,8 @@ static void bq_flush_to_queue(struct xdp_bulk_queue *bq)
 	struct ptr_ring *q;
 	int i;
 
+	lockdep_assert_held(&bq->bq_lock);
+
 	if (unlikely(!bq->count))
 		return;
 
@@ -749,11 +754,15 @@ 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.
+ * Thus, safe percpu variable access. PREEMPT_RT relies on
+ * local_lock_nested_bh() to serialise 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_nested_bh(&rcpu->bulkq->bq_lock);
+	bq = this_cpu_ptr(rcpu->bulkq);
 
 	if (unlikely(bq->count == CPU_MAP_BULK_SIZE))
 		bq_flush_to_queue(bq);
@@ -774,6 +783,8 @@ static void bq_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf)
 
 		list_add(&bq->flush_node, flush_list);
 	}
+
+	local_unlock_nested_bh(&rcpu->bulkq->bq_lock);
 }
 
 int cpu_map_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf,
@@ -810,7 +821,9 @@ void __cpu_map_flush(struct list_head *flush_list)
 	struct xdp_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
+		local_lock_nested_bh(&bq->obj->bulkq->bq_lock);
 		bq_flush_to_queue(bq);
+		local_unlock_nested_bh(&bq->obj->bulkq->bq_lock);
 
 		/* If already running, costs spin_lock_irqsave + smb_mb */
 		wake_up_process(bq->obj->kthread);
-- 
2.43.0

[PATCH bpf v4 2/2] bpf: devmap: fix race in bq_xmit_all on PREEMPT_RT

[Jiayuan Chen]

From: Jiayuan Chen <jiayuan.chen@shopee.com>

On PREEMPT_RT kernels, the per-CPU xdp_dev_bulk_queue (bq) can be
accessed concurrently by multiple preemptible tasks on the same CPU.

The original code assumes bq_enqueue() and __dev_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() (when
PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
preemption, which allows CFS scheduling to preempt a task during
bq_xmit_all(), enabling another task on the same CPU to enter
bq_enqueue() and operate on the same per-CPU bq concurrently.

This leads to several races:

1. Double-free / use-after-free on bq->q[]: bq_xmit_all() snapshots
   cnt = bq->count, then iterates bq->q[0..cnt-1] to transmit frames.
   If preempted after the snapshot, a second task can call bq_enqueue()
   -> bq_xmit_all() on the same bq, transmitting (and freeing) the
   same frames. When the first task resumes, it operates on stale
   pointers in bq->q[], causing use-after-free.

2. bq->count and bq->q[] corruption: concurrent bq_enqueue() modifying
   bq->count and bq->q[] while bq_xmit_all() is reading them.

3. dev_rx/xdp_prog teardown race: __dev_flush() clears bq->dev_rx and
   bq->xdp_prog after bq_xmit_all(). If preempted between
   bq_xmit_all() return and bq->dev_rx = NULL, a preempting
   bq_enqueue() sees dev_rx still set (non-NULL), skips adding bq to
   the flush_list, and enqueues a frame. When __dev_flush() resumes,
   it clears dev_rx and removes bq from the flush_list, orphaning the
   newly enqueued frame.

4. __list_del_clearprev() on flush_node: similar to the cpumap race,
   both tasks can call __list_del_clearprev() on the same flush_node,
   the second dereferences the prev pointer already set to NULL.

The race between task A (__dev_flush -> bq_xmit_all) and task B
(bq_enqueue -> bq_xmit_all) on the same CPU:

  Task A (xdp_do_flush)          Task B (ndo_xdp_xmit redirect)
  ----------------------         --------------------------------
  __dev_flush(flush_list)
    bq_xmit_all(bq)
      cnt = bq->count  /* e.g. 16 */
      /* start iterating bq->q[] */
    <-- CFS preempts Task A -->
                                   bq_enqueue(dev, xdpf)
                                     bq->count == DEV_MAP_BULK_SIZE
                                     bq_xmit_all(bq, 0)
                                       cnt = bq->count  /* same 16! */
                                       ndo_xdp_xmit(bq->q[])
                                       /* frames freed by driver */
                                       bq->count = 0
    <-- Task A resumes -->
      ndo_xdp_xmit(bq->q[])
      /* use-after-free: frames already freed! */

Fix this by adding a local_lock_t to xdp_dev_bulk_queue and acquiring
it in bq_enqueue() and __dev_flush(). These paths already run under
local_bh_disable(), so use local_lock_nested_bh() which on non-RT is
a pure annotation with no overhead, and on PREEMPT_RT provides a
per-CPU sleeping lock that serializes access to the bq.

Fixes: 3253cb49cbad ("softirq: Allow to drop the softirq-BKL lock on PREEMPT_RT")
Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jiayuan Chen <jiayuan.chen@shopee.com>
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
---
 kernel/bpf/devmap.c | 25 +++++++++++++++++++++----
 1 file changed, 21 insertions(+), 4 deletions(-)

diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index 2625601de76e..10cf0731f91d 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -45,6 +45,7 @@
  * types of devmap; only the lookup and insertion is different.
  */
 #include <linux/bpf.h>
+#include <linux/local_lock.h>
 #include <net/xdp.h>
 #include <linux/filter.h>
 #include <trace/events/xdp.h>
@@ -60,6 +61,7 @@ struct xdp_dev_bulk_queue {
 	struct net_device *dev_rx;
 	struct bpf_prog *xdp_prog;
 	unsigned int count;
+	local_lock_t bq_lock;
 };
 
 struct bpf_dtab_netdev {
@@ -381,6 +383,8 @@ static void bq_xmit_all(struct xdp_dev_bulk_queue *bq, u32 flags)
 	int to_send = cnt;
 	int i;
 
+	lockdep_assert_held(&bq->bq_lock);
+
 	if (unlikely(!cnt))
 		return;
 
@@ -425,10 +429,12 @@ void __dev_flush(struct list_head *flush_list)
 	struct xdp_dev_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
+		local_lock_nested_bh(&bq->dev->xdp_bulkq->bq_lock);
 		bq_xmit_all(bq, XDP_XMIT_FLUSH);
 		bq->dev_rx = NULL;
 		bq->xdp_prog = NULL;
 		__list_del_clearprev(&bq->flush_node);
+		local_unlock_nested_bh(&bq->dev->xdp_bulkq->bq_lock);
 	}
 }
 
@@ -451,12 +457,16 @@ static void *__dev_map_lookup_elem(struct bpf_map *map, u32 key)
 
 /* Runs in NAPI, i.e., softirq under local_bh_disable(). Thus, safe percpu
  * variable access, and map elements stick around. See comment above
- * xdp_do_flush() in filter.c.
+ * xdp_do_flush() in filter.c. PREEMPT_RT relies on local_lock_nested_bh()
+ * to serialise access to the per-CPU bq.
  */
 static void bq_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
 		       struct net_device *dev_rx, struct bpf_prog *xdp_prog)
 {
-	struct xdp_dev_bulk_queue *bq = this_cpu_ptr(dev->xdp_bulkq);
+	struct xdp_dev_bulk_queue *bq;
+
+	local_lock_nested_bh(&dev->xdp_bulkq->bq_lock);
+	bq = this_cpu_ptr(dev->xdp_bulkq);
 
 	if (unlikely(bq->count == DEV_MAP_BULK_SIZE))
 		bq_xmit_all(bq, 0);
@@ -477,6 +487,8 @@ static void bq_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
 	}
 
 	bq->q[bq->count++] = xdpf;
+
+	local_unlock_nested_bh(&dev->xdp_bulkq->bq_lock);
 }
 
 static inline int __xdp_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
@@ -1115,8 +1127,13 @@ static int dev_map_notification(struct notifier_block *notifier,
 		if (!netdev->xdp_bulkq)
 			return NOTIFY_BAD;
 
-		for_each_possible_cpu(cpu)
-			per_cpu_ptr(netdev->xdp_bulkq, cpu)->dev = netdev;
+		for_each_possible_cpu(cpu) {
+			struct xdp_dev_bulk_queue *bq;
+
+			bq = per_cpu_ptr(netdev->xdp_bulkq, cpu);
+			bq->dev = netdev;
+			local_lock_init(&bq->bq_lock);
+		}
 		break;
 	case NETDEV_UNREGISTER:
 		/* This rcu_read_lock/unlock pair is needed because
-- 
2.43.0

Re: [PATCH bpf v4 0/2] bpf: cpumap/devmap: fix per-CPU bulk queue races on PREEMPT_RT

[patchwork-bot+netdevbpf]

Hello:

This series was applied to bpf/bpf.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Wed, 25 Feb 2026 20:14:54 +0800 you wrote:
> On PREEMPT_RT kernels, local_bh_disable() only calls migrate_disable()
> (when PREEMPT_RT_NEEDS_BH_LOCK is not set) and does not disable
> preemption. This means CFS scheduling can preempt a task inside the
> per-CPU bulk queue (bq) operations in cpumap and devmap, allowing
> another task on the same CPU to concurrently access the same bq,
> leading to use-after-free, list corruption, and kernel panics.
> 
> [...]

Here is the summary with links:
  - [bpf,v4,1/2] bpf: cpumap: fix race in bq_flush_to_queue on PREEMPT_RT
    https://git.kernel.org/bpf/bpf/c/869c63d5975d
  - [bpf,v4,2/2] bpf: devmap: fix race in bq_xmit_all on PREEMPT_RT
    https://git.kernel.org/bpf/bpf/c/1872e75375c4

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html