1 // SPDX-License-Identifier: GPL-2.0
2 /* XSKMAP used for AF_XDP sockets
3 * Copyright(c) 2018 Intel Corporation.
4 */
5
6 #include <linux/bpf.h>
7 #include <linux/capability.h>
8 #include <net/xdp_sock.h>
9 #include <linux/slab.h>
10 #include <linux/sched.h>
11
12 #include "xsk.h"
13
xsk_map_node_alloc(struct xsk_map * map,struct xdp_sock ** map_entry)14 static struct xsk_map_node *xsk_map_node_alloc(struct xsk_map *map,
15 struct xdp_sock **map_entry)
16 {
17 struct xsk_map_node *node;
18
19 node = bpf_map_kzalloc(&map->map, sizeof(*node),
20 GFP_ATOMIC | __GFP_NOWARN);
21 if (!node)
22 return ERR_PTR(-ENOMEM);
23
24 bpf_map_inc(&map->map);
25
26 node->map = map;
27 node->map_entry = map_entry;
28 return node;
29 }
30
xsk_map_node_free(struct xsk_map_node * node)31 static void xsk_map_node_free(struct xsk_map_node *node)
32 {
33 bpf_map_put(&node->map->map);
34 kfree(node);
35 }
36
xsk_map_sock_add(struct xdp_sock * xs,struct xsk_map_node * node)37 static void xsk_map_sock_add(struct xdp_sock *xs, struct xsk_map_node *node)
38 {
39 spin_lock_bh(&xs->map_list_lock);
40 list_add_tail(&node->node, &xs->map_list);
41 spin_unlock_bh(&xs->map_list_lock);
42 }
43
xsk_map_sock_delete(struct xdp_sock * xs,struct xdp_sock ** map_entry)44 static void xsk_map_sock_delete(struct xdp_sock *xs,
45 struct xdp_sock **map_entry)
46 {
47 struct xsk_map_node *n, *tmp;
48
49 spin_lock_bh(&xs->map_list_lock);
50 list_for_each_entry_safe(n, tmp, &xs->map_list, node) {
51 if (map_entry == n->map_entry) {
52 list_del(&n->node);
53 xsk_map_node_free(n);
54 }
55 }
56 spin_unlock_bh(&xs->map_list_lock);
57 }
58
xsk_map_alloc(union bpf_attr * attr)59 static struct bpf_map *xsk_map_alloc(union bpf_attr *attr)
60 {
61 struct xsk_map *m;
62 int numa_node;
63 u64 size;
64
65 if (!capable(CAP_NET_ADMIN))
66 return ERR_PTR(-EPERM);
67
68 if (attr->max_entries == 0 || attr->key_size != 4 ||
69 attr->value_size != 4 ||
70 attr->map_flags & ~(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY))
71 return ERR_PTR(-EINVAL);
72
73 numa_node = bpf_map_attr_numa_node(attr);
74 size = struct_size(m, xsk_map, attr->max_entries);
75
76 m = bpf_map_area_alloc(size, numa_node);
77 if (!m)
78 return ERR_PTR(-ENOMEM);
79
80 bpf_map_init_from_attr(&m->map, attr);
81 spin_lock_init(&m->lock);
82
83 return &m->map;
84 }
85
xsk_map_free(struct bpf_map * map)86 static void xsk_map_free(struct bpf_map *map)
87 {
88 struct xsk_map *m = container_of(map, struct xsk_map, map);
89
90 synchronize_net();
91 bpf_map_area_free(m);
92 }
93
xsk_map_get_next_key(struct bpf_map * map,void * key,void * next_key)94 static int xsk_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
95 {
96 struct xsk_map *m = container_of(map, struct xsk_map, map);
97 u32 index = key ? *(u32 *)key : U32_MAX;
98 u32 *next = next_key;
99
100 if (index >= m->map.max_entries) {
101 *next = 0;
102 return 0;
103 }
104
105 if (index == m->map.max_entries - 1)
106 return -ENOENT;
107 *next = index + 1;
108 return 0;
109 }
110
xsk_map_gen_lookup(struct bpf_map * map,struct bpf_insn * insn_buf)111 static int xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
112 {
113 const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2;
114 struct bpf_insn *insn = insn_buf;
115
116 *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
117 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
118 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *)));
119 *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map));
120 *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp);
121 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(struct xsk_sock *), ret, ret, 0);
122 *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
123 *insn++ = BPF_MOV64_IMM(ret, 0);
124 return insn - insn_buf;
125 }
126
__xsk_map_lookup_elem(struct bpf_map * map,u32 key)127 static void *__xsk_map_lookup_elem(struct bpf_map *map, u32 key)
128 {
129 struct xsk_map *m = container_of(map, struct xsk_map, map);
130
131 if (key >= map->max_entries)
132 return NULL;
133
134 return READ_ONCE(m->xsk_map[key]);
135 }
136
xsk_map_lookup_elem(struct bpf_map * map,void * key)137 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
138 {
139 WARN_ON_ONCE(!rcu_read_lock_held());
140 return __xsk_map_lookup_elem(map, *(u32 *)key);
141 }
142
xsk_map_lookup_elem_sys_only(struct bpf_map * map,void * key)143 static void *xsk_map_lookup_elem_sys_only(struct bpf_map *map, void *key)
144 {
145 return ERR_PTR(-EOPNOTSUPP);
146 }
147
xsk_map_update_elem(struct bpf_map * map,void * key,void * value,u64 map_flags)148 static int xsk_map_update_elem(struct bpf_map *map, void *key, void *value,
149 u64 map_flags)
150 {
151 struct xsk_map *m = container_of(map, struct xsk_map, map);
152 struct xdp_sock *xs, *old_xs, **map_entry;
153 u32 i = *(u32 *)key, fd = *(u32 *)value;
154 struct xsk_map_node *node;
155 struct socket *sock;
156 int err;
157
158 if (unlikely(map_flags > BPF_EXIST))
159 return -EINVAL;
160 if (unlikely(i >= m->map.max_entries))
161 return -E2BIG;
162
163 sock = sockfd_lookup(fd, &err);
164 if (!sock)
165 return err;
166
167 if (sock->sk->sk_family != PF_XDP) {
168 sockfd_put(sock);
169 return -EOPNOTSUPP;
170 }
171
172 xs = (struct xdp_sock *)sock->sk;
173
174 map_entry = &m->xsk_map[i];
175 node = xsk_map_node_alloc(m, map_entry);
176 if (IS_ERR(node)) {
177 sockfd_put(sock);
178 return PTR_ERR(node);
179 }
180
181 spin_lock_bh(&m->lock);
182 old_xs = READ_ONCE(*map_entry);
183 if (old_xs == xs) {
184 err = 0;
185 goto out;
186 } else if (old_xs && map_flags == BPF_NOEXIST) {
187 err = -EEXIST;
188 goto out;
189 } else if (!old_xs && map_flags == BPF_EXIST) {
190 err = -ENOENT;
191 goto out;
192 }
193 xsk_map_sock_add(xs, node);
194 WRITE_ONCE(*map_entry, xs);
195 if (old_xs)
196 xsk_map_sock_delete(old_xs, map_entry);
197 spin_unlock_bh(&m->lock);
198 sockfd_put(sock);
199 return 0;
200
201 out:
202 spin_unlock_bh(&m->lock);
203 sockfd_put(sock);
204 xsk_map_node_free(node);
205 return err;
206 }
207
xsk_map_delete_elem(struct bpf_map * map,void * key)208 static int xsk_map_delete_elem(struct bpf_map *map, void *key)
209 {
210 struct xsk_map *m = container_of(map, struct xsk_map, map);
211 struct xdp_sock *old_xs, **map_entry;
212 int k = *(u32 *)key;
213
214 if (k >= map->max_entries)
215 return -EINVAL;
216
217 spin_lock_bh(&m->lock);
218 map_entry = &m->xsk_map[k];
219 old_xs = xchg(map_entry, NULL);
220 if (old_xs)
221 xsk_map_sock_delete(old_xs, map_entry);
222 spin_unlock_bh(&m->lock);
223
224 return 0;
225 }
226
xsk_map_redirect(struct bpf_map * map,u32 ifindex,u64 flags)227 static int xsk_map_redirect(struct bpf_map *map, u32 ifindex, u64 flags)
228 {
229 return __bpf_xdp_redirect_map(map, ifindex, flags, __xsk_map_lookup_elem);
230 }
231
xsk_map_try_sock_delete(struct xsk_map * map,struct xdp_sock * xs,struct xdp_sock ** map_entry)232 void xsk_map_try_sock_delete(struct xsk_map *map, struct xdp_sock *xs,
233 struct xdp_sock **map_entry)
234 {
235 spin_lock_bh(&map->lock);
236 if (READ_ONCE(*map_entry) == xs) {
237 WRITE_ONCE(*map_entry, NULL);
238 xsk_map_sock_delete(xs, map_entry);
239 }
240 spin_unlock_bh(&map->lock);
241 }
242
xsk_map_meta_equal(const struct bpf_map * meta0,const struct bpf_map * meta1)243 static bool xsk_map_meta_equal(const struct bpf_map *meta0,
244 const struct bpf_map *meta1)
245 {
246 return meta0->max_entries == meta1->max_entries &&
247 bpf_map_meta_equal(meta0, meta1);
248 }
249
250 static int xsk_map_btf_id;
251 const struct bpf_map_ops xsk_map_ops = {
252 .map_meta_equal = xsk_map_meta_equal,
253 .map_alloc = xsk_map_alloc,
254 .map_free = xsk_map_free,
255 .map_get_next_key = xsk_map_get_next_key,
256 .map_lookup_elem = xsk_map_lookup_elem,
257 .map_gen_lookup = xsk_map_gen_lookup,
258 .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only,
259 .map_update_elem = xsk_map_update_elem,
260 .map_delete_elem = xsk_map_delete_elem,
261 .map_check_btf = map_check_no_btf,
262 .map_btf_name = "xsk_map",
263 .map_btf_id = &xsk_map_btf_id,
264 .map_redirect = xsk_map_redirect,
265 };
266