bpf: Use bpf_global_percpu_ma for per-cpu kptr in __bpf_obj_drop_impl()

The following warning was reported when running "./test_progs -t
test_bpf_ma/percpu_free_through_map_free":

  ------------[ cut here ]------------
  WARNING: CPU: 1 PID: 68 at kernel/bpf/memalloc.c:342
  CPU: 1 PID: 68 Comm: kworker/u16:2 Not tainted 6.6.0-rc2+ #222
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996)
  Workqueue: events_unbound bpf_map_free_deferred
  RIP: 0010:bpf_mem_refill+0x21c/0x2a0
  ......
  Call Trace:
   <IRQ>
   ? bpf_mem_refill+0x21c/0x2a0
   irq_work_single+0x27/0x70
   irq_work_run_list+0x2a/0x40
   irq_work_run+0x18/0x40
   __sysvec_irq_work+0x1c/0xc0
   sysvec_irq_work+0x73/0x90
   </IRQ>
   <TASK>
   asm_sysvec_irq_work+0x1b/0x20
  RIP: 0010:unit_free+0x50/0x80
   ......
   bpf_mem_free+0x46/0x60
   __bpf_obj_drop_impl+0x40/0x90
   bpf_obj_free_fields+0x17d/0x1a0
   array_map_free+0x6b/0x170
   bpf_map_free_deferred+0x54/0xa0
   process_scheduled_works+0xba/0x370
   worker_thread+0x16d/0x2e0
   kthread+0x105/0x140
   ret_from_fork+0x39/0x60
   ret_from_fork_asm+0x1b/0x30
   </TASK>
  ---[ end trace 0000000000000000 ]---

The reason is simple: __bpf_obj_drop_impl() does not know the freeing
field is a per-cpu pointer and it uses bpf_global_ma to free the
pointer. Because bpf_global_ma is not a per-cpu allocator, so ksize() is
used to select the corresponding cache. The bpf_mem_cache with 16-bytes
unit_size will always be selected to do the unmatched free and it will
trigger the warning in free_bulk() eventually.

Because per-cpu kptr doesn't support list or rb-tree now, so fix the
problem by only checking whether or not the type of kptr is per-cpu in
bpf_obj_free_fields(), and using bpf_global_percpu_ma to these kptrs.
Signed-off-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/r/20231020133202.4043247-7-houtao@huaweicloud.comSigned-off-by: Alexei Starovoitov <ast@kernel.org>

include/linux/bpf.h

@@ -2058,7 +2058,7 @@ struct btf_record *btf_record_dup(const struct btf_record *rec);
 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b);
 void bpf_obj_free_timer(const struct btf_record *rec, void *obj);
 void bpf_obj_free_fields(const struct btf_record *rec, void *obj);
-void __bpf_obj_drop_impl(void *p, const struct btf_record *rec);
+void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu);
 
 struct bpf_map *bpf_map_get(u32 ufd);
 struct bpf_map *bpf_map_get_with_uref(u32 ufd);

kernel/bpf/helpers.c

@@ -1842,7 +1842,7 @@ void bpf_list_head_free(const struct btf_field *field, void *list_head,
 		 * bpf_list_head which needs to be freed.
 		 */
 		migrate_disable();
-		__bpf_obj_drop_impl(obj, field->graph_root.value_rec);
+		__bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
 		migrate_enable();
 	}
 }
@@ -1881,7 +1881,7 @@ void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
 
 
 		migrate_disable();
-		__bpf_obj_drop_impl(obj, field->graph_root.value_rec);
+		__bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
 		migrate_enable();
 	}
 }
@@ -1913,8 +1913,10 @@ __bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
 }
 
 /* Must be called under migrate_disable(), as required by bpf_mem_free */
-void __bpf_obj_drop_impl(void *p, const struct btf_record *rec)
+void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
 {
+	struct bpf_mem_alloc *ma;
+
 	if (rec && rec->refcount_off >= 0 &&
 	    !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
 		/* Object is refcounted and refcount_dec didn't result in 0
@@ -1926,10 +1928,14 @@ void __bpf_obj_drop_impl(void *p, const struct btf_record *rec)
 	if (rec)
 		bpf_obj_free_fields(rec, p);
 
+	if (percpu)
+		ma = &bpf_global_percpu_ma;
+	else
+		ma = &bpf_global_ma;
 	if (rec && rec->refcount_off >= 0)
-		bpf_mem_free_rcu(&bpf_global_ma, p);
+		bpf_mem_free_rcu(ma, p);
 	else
-		bpf_mem_free(&bpf_global_ma, p);
+		bpf_mem_free(ma, p);
 }
 
 __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
@@ -1937,7 +1943,7 @@ __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
 	struct btf_struct_meta *meta = meta__ign;
 	void *p = p__alloc;
 
-	__bpf_obj_drop_impl(p, meta ? meta->record : NULL);
+	__bpf_obj_drop_impl(p, meta ? meta->record : NULL, false);
 }
 
 __bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
@@ -1981,7 +1987,7 @@ static int __bpf_list_add(struct bpf_list_node_kern *node,
 	 */
 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
 		/* Only called from BPF prog, no need to migrate_disable */
-		__bpf_obj_drop_impl((void *)n - off, rec);
+		__bpf_obj_drop_impl((void *)n - off, rec, false);
 		return -EINVAL;
 	}
 
@@ -2080,7 +2086,7 @@ static int __bpf_rbtree_add(struct bpf_rb_root *root,
 	 */
 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
 		/* Only called from BPF prog, no need to migrate_disable */
-		__bpf_obj_drop_impl((void *)n - off, rec);
+		__bpf_obj_drop_impl((void *)n - off, rec, false);
 		return -EINVAL;
 	}
 

kernel/bpf/syscall.c

@@ -660,8 +660,8 @@ void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 									   field->kptr.btf_id);
 				migrate_disable();
 				__bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
-								 pointee_struct_meta->record :
-								 NULL);
+								 pointee_struct_meta->record : NULL,
+								 fields[i].type == BPF_KPTR_PERCPU);
 				migrate_enable();
 			} else {
 				field->kptr.dtor(xchgd_field);