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 struct xsk_map {
  13         struct bpf_map map;
  14         struct xdp_sock **xsk_map;
  15         struct list_head __percpu *flush_list;
  16 };
  17
  18 static struct bpf_map *xsk_map_alloc(union bpf_attr *attr)
  19 {
  20         struct xsk_map *m;
  21         int cpu, err;
  22         u64 cost;
  23
  24         if (!capable(CAP_NET_ADMIN))
  25                 return ERR_PTR(-EPERM);
  26
  27         if (attr->max_entries == 0 || attr->key_size != 4 ||
  28             attr->value_size != 4 ||
  29             attr->map_flags & ~(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY))
  30                 return ERR_PTR(-EINVAL);
  31
  32         m = kzalloc(sizeof(*m), GFP_USER);
  33         if (!m)
  34                 return ERR_PTR(-ENOMEM);
  35
  36         bpf_map_init_from_attr(&m->map, attr);
  37
  38         cost = (u64)m->map.max_entries * sizeof(struct xdp_sock *);
  39         cost += sizeof(struct list_head) * num_possible_cpus();
  40
  41         /* Notice returns -EPERM on if map size is larger than memlock limit */
  42         err = bpf_map_charge_init(&m->map.memory, cost);
  43         if (err)
  44                 goto free_m;
  45
  46         err = -ENOMEM;
  47
  48         m->flush_list = alloc_percpu(struct list_head);
  49         if (!m->flush_list)
  50                 goto free_charge;
  51
  52         for_each_possible_cpu(cpu)
  53                 INIT_LIST_HEAD(per_cpu_ptr(m->flush_list, cpu));
  54
  55         m->xsk_map = bpf_map_area_alloc(m->map.max_entries *
  56                                         sizeof(struct xdp_sock *),
  57                                         m->map.numa_node);
  58         if (!m->xsk_map)
  59                 goto free_percpu;
  60         return &m->map;
  61
  62 free_percpu:
  63         free_percpu(m->flush_list);
  64 free_charge:
  65         bpf_map_charge_finish(&m->map.memory);
  66 free_m:
  67         kfree(m);
  68         return ERR_PTR(err);
  69 }
  70
  71 static void xsk_map_free(struct bpf_map *map)
  72 {
  73         struct xsk_map *m = container_of(map, struct xsk_map, map);
  74         int i;
  75
  76         bpf_clear_redirect_map(map);
  77         synchronize_net();
  78
  79         for (i = 0; i < map->max_entries; i++) {
  80                 struct xdp_sock *xs;
  81
  82                 xs = m->xsk_map[i];
  83                 if (!xs)
  84                         continue;
  85
  86                 sock_put((struct sock *)xs);
  87         }
  88
  89         free_percpu(m->flush_list);
  90         bpf_map_area_free(m->xsk_map);
  91         kfree(m);
  92 }
  93
  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
 111 struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, u32 key)
 112 {
 113         struct xsk_map *m = container_of(map, struct xsk_map, map);
 114         struct xdp_sock *xs;
 115
 116         if (key >= map->max_entries)
 117                 return NULL;
 118
 119         xs = READ_ONCE(m->xsk_map[key]);
 120         return xs;
 121 }
 122
 123 int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp,
 124                        struct xdp_sock *xs)
 125 {
 126         struct xsk_map *m = container_of(map, struct xsk_map, map);
 127         struct list_head *flush_list = this_cpu_ptr(m->flush_list);
 128         int err;
 129
 130         err = xsk_rcv(xs, xdp);
 131         if (err)
 132                 return err;
 133
 134         if (!xs->flush_node.prev)
 135                 list_add(&xs->flush_node, flush_list);
 136
 137         return 0;
 138 }
 139
 140 void __xsk_map_flush(struct bpf_map *map)
 141 {
 142         struct xsk_map *m = container_of(map, struct xsk_map, map);
 143         struct list_head *flush_list = this_cpu_ptr(m->flush_list);
 144         struct xdp_sock *xs, *tmp;
 145
 146         list_for_each_entry_safe(xs, tmp, flush_list, flush_node) {
 147                 xsk_flush(xs);
 148                 __list_del_clearprev(&xs->flush_node);
 149         }
 150 }
 151
 152 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
 153 {
 154         WARN_ON_ONCE(!rcu_read_lock_held());
 155         return __xsk_map_lookup_elem(map, *(u32 *)key);
 156 }
 157
 158 static void *xsk_map_lookup_elem_sys_only(struct bpf_map *map, void *key)
 159 {
 160         return ERR_PTR(-EOPNOTSUPP);
 161 }
 162
 163 static int xsk_map_update_elem(struct bpf_map *map, void *key, void *value,
 164                                u64 map_flags)
 165 {
 166         struct xsk_map *m = container_of(map, struct xsk_map, map);
 167         u32 i = *(u32 *)key, fd = *(u32 *)value;
 168         struct xdp_sock *xs, *old_xs;
 169         struct socket *sock;
 170         int err;
 171
 172         if (unlikely(map_flags > BPF_EXIST))
 173                 return -EINVAL;
 174         if (unlikely(i >= m->map.max_entries))
 175                 return -E2BIG;
 176         if (unlikely(map_flags == BPF_NOEXIST))
 177                 return -EEXIST;
 178
 179         sock = sockfd_lookup(fd, &err);
 180         if (!sock)
 181                 return err;
 182
 183         if (sock->sk->sk_family != PF_XDP) {
 184                 sockfd_put(sock);
 185                 return -EOPNOTSUPP;
 186         }
 187
 188         xs = (struct xdp_sock *)sock->sk;
 189
 190         if (!xsk_is_setup_for_bpf_map(xs)) {
 191                 sockfd_put(sock);
 192                 return -EOPNOTSUPP;
 193         }
 194
 195         sock_hold(sock->sk);
 196
 197         old_xs = xchg(&m->xsk_map[i], xs);
 198         if (old_xs)
 199                 sock_put((struct sock *)old_xs);
 200
 201         sockfd_put(sock);
 202         return 0;
 203 }
 204
 205 static int xsk_map_delete_elem(struct bpf_map *map, void *key)
 206 {
 207         struct xsk_map *m = container_of(map, struct xsk_map, map);
 208         struct xdp_sock *old_xs;
 209         int k = *(u32 *)key;
 210
 211         if (k >= map->max_entries)
 212                 return -EINVAL;
 213
 214         old_xs = xchg(&m->xsk_map[k], NULL);
 215         if (old_xs)
 216                 sock_put((struct sock *)old_xs);
 217
 218         return 0;
 219 }
 220
 221 const struct bpf_map_ops xsk_map_ops = {
 222         .map_alloc = xsk_map_alloc,
 223         .map_free = xsk_map_free,
 224         .map_get_next_key = xsk_map_get_next_key,
 225         .map_lookup_elem = xsk_map_lookup_elem,
 226         .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only,
 227         .map_update_elem = xsk_map_update_elem,
 228         .map_delete_elem = xsk_map_delete_elem,
 229         .map_check_btf = map_check_no_btf,
 230 };