1 // SPDX-License-Identifier: GPL-2.0
2 /* XSKMAP used for AF_XDP sockets
3 * Copyright(c) 2018 Intel Corporation.
7 #include <linux/filter.h>
8 #include <linux/capability.h>
9 #include <net/xdp_sock.h>
10 #include <linux/slab.h>
11 #include <linux/sched.h>
12 #include <linux/btf_ids.h>
16 static struct xsk_map_node *xsk_map_node_alloc(struct xsk_map *map,
17 struct xdp_sock __rcu **map_entry)
19 struct xsk_map_node *node;
21 node = bpf_map_kzalloc(&map->map, sizeof(*node),
22 GFP_ATOMIC | __GFP_NOWARN);
24 return ERR_PTR(-ENOMEM);
26 bpf_map_inc(&map->map);
29 node->map_entry = map_entry;
33 static void xsk_map_node_free(struct xsk_map_node *node)
35 bpf_map_put(&node->map->map);
39 static void xsk_map_sock_add(struct xdp_sock *xs, struct xsk_map_node *node)
41 spin_lock_bh(&xs->map_list_lock);
42 list_add_tail(&node->node, &xs->map_list);
43 spin_unlock_bh(&xs->map_list_lock);
46 static void xsk_map_sock_delete(struct xdp_sock *xs,
47 struct xdp_sock __rcu **map_entry)
49 struct xsk_map_node *n, *tmp;
51 spin_lock_bh(&xs->map_list_lock);
52 list_for_each_entry_safe(n, tmp, &xs->map_list, node) {
53 if (map_entry == n->map_entry) {
58 spin_unlock_bh(&xs->map_list_lock);
61 static struct bpf_map *xsk_map_alloc(union bpf_attr *attr)
67 if (!capable(CAP_NET_ADMIN))
68 return ERR_PTR(-EPERM);
70 if (attr->max_entries == 0 || attr->key_size != 4 ||
71 attr->value_size != 4 ||
72 attr->map_flags & ~(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY))
73 return ERR_PTR(-EINVAL);
75 numa_node = bpf_map_attr_numa_node(attr);
76 size = struct_size(m, xsk_map, attr->max_entries);
78 m = bpf_map_area_alloc(size, numa_node);
80 return ERR_PTR(-ENOMEM);
82 bpf_map_init_from_attr(&m->map, attr);
83 spin_lock_init(&m->lock);
88 static void xsk_map_free(struct bpf_map *map)
90 struct xsk_map *m = container_of(map, struct xsk_map, map);
96 static int xsk_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
98 struct xsk_map *m = container_of(map, struct xsk_map, map);
99 u32 index = key ? *(u32 *)key : U32_MAX;
100 u32 *next = next_key;
102 if (index >= m->map.max_entries) {
107 if (index == m->map.max_entries - 1)
113 static int xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
115 const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2;
116 struct bpf_insn *insn = insn_buf;
118 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
119 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
120 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *)));
121 *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map));
122 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp);
123 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(struct xsk_sock *), ret, ret, 0);
124 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
125 *insn++ = BPF_MOV64_IMM(ret, 0);
126 return insn - insn_buf;
129 /* Elements are kept alive by RCU; either by rcu_read_lock() (from syscall) or
130 * by local_bh_disable() (from XDP calls inside NAPI). The
131 * rcu_read_lock_bh_held() below makes lockdep accept both.
133 static void *__xsk_map_lookup_elem(struct bpf_map *map, u32 key)
135 struct xsk_map *m = container_of(map, struct xsk_map, map);
137 if (key >= map->max_entries)
140 return rcu_dereference_check(m->xsk_map[key], rcu_read_lock_bh_held());
143 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
145 return __xsk_map_lookup_elem(map, *(u32 *)key);
148 static void *xsk_map_lookup_elem_sys_only(struct bpf_map *map, void *key)
150 return ERR_PTR(-EOPNOTSUPP);
153 static int xsk_map_update_elem(struct bpf_map *map, void *key, void *value,
156 struct xsk_map *m = container_of(map, struct xsk_map, map);
157 struct xdp_sock __rcu **map_entry;
158 struct xdp_sock *xs, *old_xs;
159 u32 i = *(u32 *)key, fd = *(u32 *)value;
160 struct xsk_map_node *node;
164 if (unlikely(map_flags > BPF_EXIST))
166 if (unlikely(i >= m->map.max_entries))
169 sock = sockfd_lookup(fd, &err);
173 if (sock->sk->sk_family != PF_XDP) {
178 xs = (struct xdp_sock *)sock->sk;
180 map_entry = &m->xsk_map[i];
181 node = xsk_map_node_alloc(m, map_entry);
184 return PTR_ERR(node);
187 spin_lock_bh(&m->lock);
188 old_xs = rcu_dereference_protected(*map_entry, lockdep_is_held(&m->lock));
192 } else if (old_xs && map_flags == BPF_NOEXIST) {
195 } else if (!old_xs && map_flags == BPF_EXIST) {
199 xsk_map_sock_add(xs, node);
200 rcu_assign_pointer(*map_entry, xs);
202 xsk_map_sock_delete(old_xs, map_entry);
203 spin_unlock_bh(&m->lock);
208 spin_unlock_bh(&m->lock);
210 xsk_map_node_free(node);
214 static int xsk_map_delete_elem(struct bpf_map *map, void *key)
216 struct xsk_map *m = container_of(map, struct xsk_map, map);
217 struct xdp_sock __rcu **map_entry;
218 struct xdp_sock *old_xs;
221 if (k >= map->max_entries)
224 spin_lock_bh(&m->lock);
225 map_entry = &m->xsk_map[k];
226 old_xs = unrcu_pointer(xchg(map_entry, NULL));
228 xsk_map_sock_delete(old_xs, map_entry);
229 spin_unlock_bh(&m->lock);
234 static int xsk_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
236 return __bpf_xdp_redirect_map(map, ifindex, flags, 0,
237 __xsk_map_lookup_elem);
240 void xsk_map_try_sock_delete(struct xsk_map *map, struct xdp_sock *xs,
241 struct xdp_sock __rcu **map_entry)
243 spin_lock_bh(&map->lock);
244 if (rcu_access_pointer(*map_entry) == xs) {
245 rcu_assign_pointer(*map_entry, NULL);
246 xsk_map_sock_delete(xs, map_entry);
248 spin_unlock_bh(&map->lock);
251 static bool xsk_map_meta_equal(const struct bpf_map *meta0,
252 const struct bpf_map *meta1)
254 return meta0->max_entries == meta1->max_entries &&
255 bpf_map_meta_equal(meta0, meta1);
258 BTF_ID_LIST_SINGLE(xsk_map_btf_ids, struct, xsk_map)
259 const struct bpf_map_ops xsk_map_ops = {
260 .map_meta_equal = xsk_map_meta_equal,
261 .map_alloc = xsk_map_alloc,
262 .map_free = xsk_map_free,
263 .map_get_next_key = xsk_map_get_next_key,
264 .map_lookup_elem = xsk_map_lookup_elem,
265 .map_gen_lookup = xsk_map_gen_lookup,
266 .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only,
267 .map_update_elem = xsk_map_update_elem,
268 .map_delete_elem = xsk_map_delete_elem,
269 .map_check_btf = map_check_no_btf,
270 .map_btf_id = &xsk_map_btf_ids[0],
271 .map_redirect = xsk_map_redirect,