xref: /linux/kernel/bpf/map_in_map.c (revision 8e3ed5440b0c305dcd1d5fa7419bd8066d22ef42)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
3  */
4 #include <linux/slab.h>
5 #include <linux/bpf.h>
6 #include <linux/btf.h>
7 
8 #include "map_in_map.h"
9 
10 struct bpf_map *bpf_map_meta_alloc(int inner_map_ufd)
11 {
12 	struct bpf_map *inner_map, *inner_map_meta;
13 	u32 inner_map_meta_size;
14 	struct fd f;
15 	int ret;
16 
17 	f = fdget(inner_map_ufd);
18 	inner_map = __bpf_map_get(f);
19 	if (IS_ERR(inner_map))
20 		return inner_map;
21 
22 	/* Does not support >1 level map-in-map */
23 	if (inner_map->inner_map_meta) {
24 		ret = -EINVAL;
25 		goto put;
26 	}
27 
28 	if (!inner_map->ops->map_meta_equal) {
29 		ret = -ENOTSUPP;
30 		goto put;
31 	}
32 
33 	inner_map_meta_size = sizeof(*inner_map_meta);
34 	/* In some cases verifier needs to access beyond just base map. */
35 	if (inner_map->ops == &array_map_ops)
36 		inner_map_meta_size = sizeof(struct bpf_array);
37 
38 	inner_map_meta = kzalloc(inner_map_meta_size, GFP_USER);
39 	if (!inner_map_meta) {
40 		ret = -ENOMEM;
41 		goto put;
42 	}
43 
44 	inner_map_meta->map_type = inner_map->map_type;
45 	inner_map_meta->key_size = inner_map->key_size;
46 	inner_map_meta->value_size = inner_map->value_size;
47 	inner_map_meta->map_flags = inner_map->map_flags;
48 	inner_map_meta->max_entries = inner_map->max_entries;
49 
50 	inner_map_meta->record = btf_record_dup(inner_map->record);
51 	if (IS_ERR(inner_map_meta->record)) {
52 		/* btf_record_dup returns NULL or valid pointer in case of
53 		 * invalid/empty/valid, but ERR_PTR in case of errors. During
54 		 * equality NULL or IS_ERR is equivalent.
55 		 */
56 		ret = PTR_ERR(inner_map_meta->record);
57 		goto free;
58 	}
59 	/* Note: We must use the same BTF, as we also used btf_record_dup above
60 	 * which relies on BTF being same for both maps, as some members like
61 	 * record->fields.list_head have pointers like value_rec pointing into
62 	 * inner_map->btf.
63 	 */
64 	if (inner_map->btf) {
65 		btf_get(inner_map->btf);
66 		inner_map_meta->btf = inner_map->btf;
67 	}
68 
69 	/* Misc members not needed in bpf_map_meta_equal() check. */
70 	inner_map_meta->ops = inner_map->ops;
71 	if (inner_map->ops == &array_map_ops) {
72 		struct bpf_array *inner_array_meta =
73 			container_of(inner_map_meta, struct bpf_array, map);
74 		struct bpf_array *inner_array = container_of(inner_map, struct bpf_array, map);
75 
76 		inner_array_meta->index_mask = inner_array->index_mask;
77 		inner_array_meta->elem_size = inner_array->elem_size;
78 		inner_map_meta->bypass_spec_v1 = inner_map->bypass_spec_v1;
79 	}
80 
81 	fdput(f);
82 	return inner_map_meta;
83 free:
84 	kfree(inner_map_meta);
85 put:
86 	fdput(f);
87 	return ERR_PTR(ret);
88 }
89 
90 void bpf_map_meta_free(struct bpf_map *map_meta)
91 {
92 	bpf_map_free_record(map_meta);
93 	btf_put(map_meta->btf);
94 	kfree(map_meta);
95 }
96 
97 bool bpf_map_meta_equal(const struct bpf_map *meta0,
98 			const struct bpf_map *meta1)
99 {
100 	/* No need to compare ops because it is covered by map_type */
101 	return meta0->map_type == meta1->map_type &&
102 		meta0->key_size == meta1->key_size &&
103 		meta0->value_size == meta1->value_size &&
104 		meta0->map_flags == meta1->map_flags &&
105 		btf_record_equal(meta0->record, meta1->record);
106 }
107 
108 void *bpf_map_fd_get_ptr(struct bpf_map *map,
109 			 struct file *map_file /* not used */,
110 			 int ufd)
111 {
112 	struct bpf_map *inner_map, *inner_map_meta;
113 	struct fd f;
114 
115 	f = fdget(ufd);
116 	inner_map = __bpf_map_get(f);
117 	if (IS_ERR(inner_map))
118 		return inner_map;
119 
120 	inner_map_meta = map->inner_map_meta;
121 	if (inner_map_meta->ops->map_meta_equal(inner_map_meta, inner_map))
122 		bpf_map_inc(inner_map);
123 	else
124 		inner_map = ERR_PTR(-EINVAL);
125 
126 	fdput(f);
127 	return inner_map;
128 }
129 
130 void bpf_map_fd_put_ptr(struct bpf_map *map, void *ptr, bool need_defer)
131 {
132 	struct bpf_map *inner_map = ptr;
133 
134 	/* Defer the freeing of inner map according to the sleepable attribute
135 	 * of bpf program which owns the outer map, so unnecessary waiting for
136 	 * RCU tasks trace grace period can be avoided.
137 	 */
138 	if (need_defer) {
139 		if (atomic64_read(&map->sleepable_refcnt))
140 			WRITE_ONCE(inner_map->free_after_mult_rcu_gp, true);
141 		else
142 			WRITE_ONCE(inner_map->free_after_rcu_gp, true);
143 	}
144 	bpf_map_put(inner_map);
145 }
146 
147 u32 bpf_map_fd_sys_lookup_elem(void *ptr)
148 {
149 	return ((struct bpf_map *)ptr)->id;
150 }
151