kernel/bpf/xskmap.c

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