diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 8d95f4c66275317099a8eeb06ea95a5ae04cf2fb..3586f15e58e4a6ac51cbf327043c64d59ddf2d51 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -82,7 +82,12 @@ struct bpf_map_ops {
 	/* funcs called by prog_array and perf_event_array map */
 	void *(*map_fd_get_ptr)(struct bpf_map *map, struct file *map_file,
 				int fd);
-	void (*map_fd_put_ptr)(void *ptr);
+	/* If need_defer is true, the implementation should guarantee that
+	 * the to-be-put element is still alive before the bpf program, which
+	 * may manipulate it, exists.
+	 */
+	KABI_BROKEN_REPLACE(void (*map_fd_put_ptr)(void *ptr),
+			    void (*map_fd_put_ptr)(struct bpf_map *map, void *ptr, bool need_defer))
 	int (*map_gen_lookup)(struct bpf_map *map, struct bpf_insn *insn_buf);
 	u32 (*map_fd_sys_lookup_elem)(void *ptr);
 	void (*map_seq_show_elem)(struct bpf_map *map, void *key,
@@ -164,14 +169,21 @@ struct bpf_map {
 	u32 btf_vmlinux_value_type_id;
 	bool bypass_spec_v1;
 	bool frozen; /* write-once; write-protected by freeze_mutex */
-	/* 22 bytes hole */
+	KABI_EXTEND(bool free_after_mult_rcu_gp)
+	/* 17 bytes hole */
 
 	/* The 3rd and 4th cacheline with misc members to avoid false sharing
 	 * particularly with refcounting.
 	 */
 	atomic64_t refcnt ____cacheline_aligned;
 	atomic64_t usercnt;
-	struct work_struct work;
+	/* rcu is used before freeing and work is only used during freeing */
+	KABI_BROKEN_REPLACE(
+	struct work_struct work,
+	union {
+		struct work_struct work;
+		struct rcu_head rcu;
+	})
 	struct mutex freeze_mutex;
 	atomic64_t writecnt;
 };
diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index f241bda2679d43d765c08eec6a6287bd6a8e4fa5..b5b8b57dc212072e5e491cc8677f7324697a3098 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -764,11 +764,11 @@ int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 	}
 
 	if (old_ptr)
-		map->ops->map_fd_put_ptr(old_ptr);
+		map->ops->map_fd_put_ptr(map, old_ptr, true);
 	return 0;
 }
 
-static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
+static int __fd_array_map_delete_elem(struct bpf_map *map, void *key, bool need_defer)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	void *old_ptr;
@@ -787,13 +787,18 @@ static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
 	}
 
 	if (old_ptr) {
-		map->ops->map_fd_put_ptr(old_ptr);
+		map->ops->map_fd_put_ptr(map, old_ptr, need_defer);
 		return 0;
 	} else {
 		return -ENOENT;
 	}
 }
 
+static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
+{
+	return __fd_array_map_delete_elem(map, key, true);
+}
+
 static void *prog_fd_array_get_ptr(struct bpf_map *map,
 				   struct file *map_file, int fd)
 {
@@ -811,8 +816,9 @@ static void *prog_fd_array_get_ptr(struct bpf_map *map,
 	return prog;
 }
 
-static void prog_fd_array_put_ptr(void *ptr)
+static void prog_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
 {
+	/* bpf_prog is freed after one RCU or tasks trace grace period */
 	bpf_prog_put(ptr);
 }
 
@@ -822,13 +828,13 @@ static u32 prog_fd_array_sys_lookup_elem(void *ptr)
 }
 
 /* decrement refcnt of all bpf_progs that are stored in this map */
-static void bpf_fd_array_map_clear(struct bpf_map *map)
+static void bpf_fd_array_map_clear(struct bpf_map *map, bool need_defer)
 {
 	struct bpf_array *array = container_of(map, struct bpf_array, map);
 	int i;
 
 	for (i = 0; i < array->map.max_entries; i++)
-		fd_array_map_delete_elem(map, &i);
+		__fd_array_map_delete_elem(map, &i, need_defer);
 }
 
 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
@@ -1007,7 +1013,7 @@ static void prog_array_map_clear_deferred(struct work_struct *work)
 {
 	struct bpf_map *map = container_of(work, struct bpf_array_aux,
 					   work)->map;
-	bpf_fd_array_map_clear(map);
+	bpf_fd_array_map_clear(map, true);
 	bpf_map_put(map);
 }
 
@@ -1139,8 +1145,9 @@ static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
 	return ee;
 }
 
-static void perf_event_fd_array_put_ptr(void *ptr)
+static void perf_event_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
 {
+	/* bpf_perf_event is freed after one RCU grace period */
 	bpf_event_entry_free_rcu(ptr);
 }
 
@@ -1158,7 +1165,7 @@ static void perf_event_fd_array_release(struct bpf_map *map,
 	for (i = 0; i < array->map.max_entries; i++) {
 		ee = READ_ONCE(array->ptrs[i]);
 		if (ee && ee->map_file == map_file)
-			fd_array_map_delete_elem(map, &i);
+			__fd_array_map_delete_elem(map, &i, true);
 	}
 	rcu_read_unlock();
 }
@@ -1166,7 +1173,7 @@ static void perf_event_fd_array_release(struct bpf_map *map,
 static void perf_event_fd_array_map_free(struct bpf_map *map)
 {
 	if (map->map_flags & BPF_F_PRESERVE_ELEMS)
-		bpf_fd_array_map_clear(map);
+		bpf_fd_array_map_clear(map, false);
 	fd_array_map_free(map);
 }
 
@@ -1195,7 +1202,7 @@ static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
 	return cgroup_get_from_fd(fd);
 }
 
-static void cgroup_fd_array_put_ptr(void *ptr)
+static void cgroup_fd_array_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
 {
 	/* cgroup_put free cgrp after a rcu grace period */
 	cgroup_put(ptr);
@@ -1203,7 +1210,7 @@ static void cgroup_fd_array_put_ptr(void *ptr)
 
 static void cgroup_fd_array_free(struct bpf_map *map)
 {
-	bpf_fd_array_map_clear(map);
+	bpf_fd_array_map_clear(map, false);
 	fd_array_map_free(map);
 }
 
@@ -1249,7 +1256,7 @@ static void array_of_map_free(struct bpf_map *map)
 	 * is protected by fdget/fdput.
 	 */
 	bpf_map_meta_free(map->inner_map_meta);
-	bpf_fd_array_map_clear(map);
+	bpf_fd_array_map_clear(map, false);
 	fd_array_map_free(map);
 }
 
diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index 0ce445aadfdfb19e3b7032dead3e73a36339c169..ec849731427251a119d5267d2f2efd3e2e5cfd93 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -786,7 +786,7 @@ static void htab_put_fd_value(struct bpf_htab *htab, struct htab_elem *l)
 
 	if (map->ops->map_fd_put_ptr) {
 		ptr = fd_htab_map_get_ptr(map, l);
-		map->ops->map_fd_put_ptr(ptr);
+		map->ops->map_fd_put_ptr(map, ptr, true);
 	}
 }
 
@@ -2023,7 +2023,7 @@ static void fd_htab_map_free(struct bpf_map *map)
 		hlist_nulls_for_each_entry_safe(l, n, head, hash_node) {
 			void *ptr = fd_htab_map_get_ptr(map, l);
 
-			map->ops->map_fd_put_ptr(ptr);
+			map->ops->map_fd_put_ptr(map, ptr, false);
 		}
 	}
 
@@ -2064,7 +2064,7 @@ int bpf_fd_htab_map_update_elem(struct bpf_map *map, struct file *map_file,
 
 	ret = htab_map_update_elem(map, key, &ptr, map_flags);
 	if (ret)
-		map->ops->map_fd_put_ptr(ptr);
+		map->ops->map_fd_put_ptr(map, ptr, false);
 
 	return ret;
 }
diff --git a/kernel/bpf/map_in_map.c b/kernel/bpf/map_in_map.c
index 39ab0b68cade574acb7a8226566e30aa91307ee1..caa1a17cbae1533cfdd5be89e4baef3edfc549d1 100644
--- a/kernel/bpf/map_in_map.c
+++ b/kernel/bpf/map_in_map.c
@@ -100,12 +100,17 @@ void *bpf_map_fd_get_ptr(struct bpf_map *map,
 	return inner_map;
 }
 
-void bpf_map_fd_put_ptr(void *ptr)
+void bpf_map_fd_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
 {
-	/* ptr->ops->map_free() has to go through one
-	 * rcu grace period by itself.
+	struct bpf_map *inner_map = ptr;
+
+	/* The inner map may still be used by both non-sleepable and sleepable
+	 * bpf program, so free it after one RCU grace period and one tasks
+	 * trace RCU grace period.
 	 */
-	bpf_map_put(ptr);
+	if (need_defer)
+		WRITE_ONCE(inner_map->free_after_mult_rcu_gp, true);
+	bpf_map_put(inner_map);
 }
 
 u32 bpf_map_fd_sys_lookup_elem(void *ptr)
diff --git a/kernel/bpf/map_in_map.h b/kernel/bpf/map_in_map.h
index bcb7534afb3c0d073222a20bc7d38538e986ee27..7d61602354de8074126944aeea4bb1d116b939e1 100644
--- a/kernel/bpf/map_in_map.h
+++ b/kernel/bpf/map_in_map.h
@@ -13,7 +13,7 @@ struct bpf_map *bpf_map_meta_alloc(int inner_map_ufd);
 void bpf_map_meta_free(struct bpf_map *map_meta);
 void *bpf_map_fd_get_ptr(struct bpf_map *map, struct file *map_file,
 			 int ufd);
-void bpf_map_fd_put_ptr(void *ptr);
+void bpf_map_fd_put_ptr(struct bpf_map *map, void *ptr, bool need_defer);
 u32 bpf_map_fd_sys_lookup_elem(void *ptr);
 
 #endif
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 419dbc3d060ee1fea6835a59291bf36b9c570d7e..cc20df2c3465fb908a2628ffe96e06d2759cf9c7 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -493,6 +493,22 @@ static void bpf_map_put_uref(struct bpf_map *map)
 	}
 }
 
+static void bpf_map_free_in_work(struct bpf_map *map)
+{
+	INIT_WORK(&map->work, bpf_map_free_deferred);
+	schedule_work(&map->work);
+}
+
+static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
+{
+	bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
+}
+
+static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
+{
+	call_rcu(rcu, bpf_map_free_rcu_gp);
+}
+
 /* decrement map refcnt and schedule it for freeing via workqueue
  * (unrelying map implementation ops->map_free() might sleep)
  */
@@ -502,8 +518,11 @@ static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 		/* bpf_map_free_id() must be called first */
 		bpf_map_free_id(map, do_idr_lock);
 		btf_put(map->btf);
-		INIT_WORK(&map->work, bpf_map_free_deferred);
-		schedule_work(&map->work);
+
+		if (READ_ONCE(map->free_after_mult_rcu_gp))
+			call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
+		else
+			bpf_map_free_in_work(map);
 	}
 }