bpf: Replace bpf_map_kmalloc_node() with kmalloc_nolock() to allocate bpf_async_cb structures.

The following kmemleak splat:

[ 8.105530] kmemleak: Trying to color unknown object at 0xff11000100e918c0 as Black
[ 8.106521] Call Trace:
[ 8.106521] <TASK>
[ 8.106521] dump_stack_lvl+0x4b/0x70
[ 8.106521] kvfree_call_rcu+0xcb/0x3b0
[ 8.106521] ? hrtimer_cancel+0x21/0x40
[ 8.106521] bpf_obj_free_fields+0x193/0x200
[ 8.106521] htab_map_update_elem+0x29c/0x410
[ 8.106521] bpf_prog_cfc8cd0f42c04044_overwrite_cb+0x47/0x4b
[ 8.106521] bpf_prog_8c30cd7c4db2e963_overwrite_timer+0x65/0x86
[ 8.106521] bpf_prog_test_run_syscall+0xe1/0x2a0

happens due to the combination of features and fixes, but mainly due to
commit 6d78b4473cdb ("bpf: Tell memcg to use allow_spinning=false path in bpf_timer_init()")
It's using __GFP_HIGH, which instructs slub/kmemleak internals to skip
kmemleak_alloc_recursive() on allocation, so subsequent kfree_rcu()->
kvfree_call_rcu()->kmemleak_ignore() complains with the above splat.

To fix this imbalance, replace bpf_map_kmalloc_node() with
kmalloc_nolock() and kfree_rcu() with call_rcu() + kfree_nolock() to
make sure that the objects allocated with kmalloc_nolock() are freed
with kfree_nolock() rather than the implicit kfree() that kfree_rcu()
uses internally.

Note, the kmalloc_nolock() happens under bpf_spin_lock_irqsave(), so
it will always fail in PREEMPT_RT. This is not an issue at the moment,
since bpf_timers are disabled in PREEMPT_RT. In the future
bpf_spin_lock will be replaced with state machine similar to
bpf_task_work.

Fixes: 6d78b4473cdb ("bpf: Tell memcg to use allow_spinning=false path in bpf_timer_init()")
Signed-off-by: Alexei Starovoitov <[e-mail]>
Signed-off-by: Daniel Borkmann <[e-mail]>
Reviewed-by: Shakeel Butt <[e-mail]>
Acked-by: Harry Yoo <[e-mail]>
Acked-by: Vlastimil Babka <[e-mail]>
Cc: [e-mail]
Link: https://lore.kernel.org/bpf/[e-mail]

include/linux/bpf.h

···
 #ifdef CONFIG_MEMCG
 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 			   int node);
+void *bpf_map_kmalloc_nolock(const struct bpf_map *map, size_t size, gfp_t flags,
+			     int node);
 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags);
 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 		       gfp_t flags);
···
  */
 #define bpf_map_kmalloc_node(_map, _size, _flags, _node)	\
 		kmalloc_node(_size, _flags, _node)
+#define bpf_map_kmalloc_nolock(_map, _size, _flags, _node)	\
+		kmalloc_nolock(_size, _flags, _node)
 #define bpf_map_kzalloc(_map, _size, _flags)			\
 		kzalloc(_size, _flags)
 #define bpf_map_kvcalloc(_map, _n, _size, _flags)		\

kernel/bpf/helpers.c

···
 	rcu_read_unlock_trace();
 }
 
+static void bpf_async_cb_rcu_free(struct rcu_head *rcu)
+{
+	struct bpf_async_cb *cb = container_of(rcu, struct bpf_async_cb, rcu);
+
+	kfree_nolock(cb);
+}
+
 static void bpf_wq_delete_work(struct work_struct *work)
 {
 	struct bpf_work *w = container_of(work, struct bpf_work, delete_work);
 
 	cancel_work_sync(&w->work);
 
-	kfree_rcu(w, cb.rcu);
+	call_rcu(&w->cb.rcu, bpf_async_cb_rcu_free);
 }
 
 static void bpf_timer_delete_work(struct work_struct *work)
···
 
 	/* Cancel the timer and wait for callback to complete if it was running.
 	 * If hrtimer_cancel() can be safely called it's safe to call
-	 * kfree_rcu(t) right after for both preallocated and non-preallocated
+	 * call_rcu() right after for both preallocated and non-preallocated
 	 * maps.  The async->cb = NULL was already done and no code path can see
 	 * address 't' anymore. Timer if armed for existing bpf_hrtimer before
 	 * bpf_timer_cancel_and_free will have been cancelled.
 	 */
 	hrtimer_cancel(&t->timer);
-	kfree_rcu(t, cb.rcu);
+	call_rcu(&t->cb.rcu, bpf_async_cb_rcu_free);
 }
 
 static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u64 flags,
···
 		goto out;
 	}
 
-	/* Allocate via bpf_map_kmalloc_node() for memcg accounting. Until
-	 * kmalloc_nolock() is available, avoid locking issues by using
-	 * __GFP_HIGH (GFP_ATOMIC & ~__GFP_RECLAIM).
-	 */
-	cb = bpf_map_kmalloc_node(map, size, __GFP_HIGH, map->numa_node);
+	cb = bpf_map_kmalloc_nolock(map, size, 0, map->numa_node);
 	if (!cb) {
 		ret = -ENOMEM;
 		goto out;
···
 		 * or pinned in bpffs.
 		 */
 		WRITE_ONCE(async->cb, NULL);
-		kfree(cb);
+		kfree_nolock(cb);
 		ret = -EPERM;
 	}
 out:
···
 	 * timer _before_ calling us, such that failing to cancel it here will
 	 * cause it to possibly use struct hrtimer after freeing bpf_hrtimer.
 	 * Therefore, we _need_ to cancel any outstanding timers before we do
-	 * kfree_rcu, even though no more timers can be armed.
+	 * call_rcu, even though no more timers can be armed.
 	 *
 	 * Moreover, we need to schedule work even if timer does not belong to
 	 * the calling callback_fn, as on two different CPUs, we can end up in a
···
 		 * completion.
 		 */
 		if (hrtimer_try_to_cancel(&t->timer) >= 0)
-			kfree_rcu(t, cb.rcu);
+			call_rcu(&t->cb.rcu, bpf_async_cb_rcu_free);
 		else
 			queue_work(system_dfl_wq, &t->cb.delete_work);
 	} else {

kernel/bpf/syscall.c

···
 	return ptr;
 }
 
+void *bpf_map_kmalloc_nolock(const struct bpf_map *map, size_t size, gfp_t flags,
+			     int node)
+{
+	struct mem_cgroup *memcg, *old_memcg;
+	void *ptr;
+
+	memcg = bpf_map_get_memcg(map);
+	old_memcg = set_active_memcg(memcg);
+	ptr = kmalloc_nolock(size, flags | __GFP_ACCOUNT, node);
+	set_active_memcg(old_memcg);
+	mem_cgroup_put(memcg);
+
+	return ptr;
+}
+
 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 {
 	struct mem_cgroup *memcg, *old_memcg;