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  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/bpf.h>
  16 #include <linux/capability.h>
  17 #include <net/xdp_sock.h>
  18 #include <linux/slab.h>
  19 #include <linux/sched.h>
  20
  21 struct xsk_map {
  22         struct bpf_map map;
  23         struct xdp_sock **xsk_map;
  24         struct list_head __percpu *flush_list;
  25 };
  26
  27 static struct bpf_map *xsk_map_alloc(union bpf_attr *attr)
  28 {
  29         int cpu, err = -EINVAL;
  30         struct xsk_map *m;
  31         u64 cost;
  32
  33         if (!capable(CAP_NET_ADMIN))
  34                 return ERR_PTR(-EPERM);
  35
  36         if (attr->max_entries == 0 || attr->key_size != 4 ||
  37             attr->value_size != 4 ||
  38             attr->map_flags & ~(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY))
  39                 return ERR_PTR(-EINVAL);
  40
  41         m = kzalloc(sizeof(*m), GFP_USER);
  42         if (!m)
  43                 return ERR_PTR(-ENOMEM);
  44
  45         bpf_map_init_from_attr(&m->map, attr);
  46
  47         cost = (u64)m->map.max_entries * sizeof(struct xdp_sock *);
  48         cost += sizeof(struct list_head) * num_possible_cpus();
  49         if (cost >= U32_MAX - PAGE_SIZE)
  50                 goto free_m;
  51
  52         m->map.pages = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
  53
  54         /* Notice returns -EPERM on if map size is larger than memlock limit */
  55         err = bpf_map_precharge_memlock(m->map.pages);
  56         if (err)
  57                 goto free_m;
  58
  59         err = -ENOMEM;
  60
  61         m->flush_list = alloc_percpu(struct list_head);
  62         if (!m->flush_list)
  63                 goto free_m;
  64
  65         for_each_possible_cpu(cpu)
  66                 INIT_LIST_HEAD(per_cpu_ptr(m->flush_list, cpu));
  67
  68         m->xsk_map = bpf_map_area_alloc(m->map.max_entries *
  69                                         sizeof(struct xdp_sock *),
  70                                         m->map.numa_node);
  71         if (!m->xsk_map)
  72                 goto free_percpu;
  73         return &m->map;
  74
  75 free_percpu:
  76         free_percpu(m->flush_list);
  77 free_m:
  78         kfree(m);
  79         return ERR_PTR(err);
  80 }
  81
  82 static void xsk_map_free(struct bpf_map *map)
  83 {
  84         struct xsk_map *m = container_of(map, struct xsk_map, map);
  85         int i;
  86
  87         synchronize_net();
  88
  89         for (i = 0; i < map->max_entries; i++) {
  90                 struct xdp_sock *xs;
  91
  92                 xs = m->xsk_map[i];
  93                 if (!xs)
  94                         continue;
  95
  96                 sock_put((struct sock *)xs);
  97         }
  98
  99         free_percpu(m->flush_list);
 100         bpf_map_area_free(m->xsk_map);
 101         kfree(m);
 102 }
 103
 104 static int xsk_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 105 {
 106         struct xsk_map *m = container_of(map, struct xsk_map, map);
 107         u32 index = key ? *(u32 *)key : U32_MAX;
 108         u32 *next = next_key;
 109
 110         if (index >= m->map.max_entries) {
 111                 *next = 0;
 112                 return 0;
 113         }
 114
 115         if (index == m->map.max_entries - 1)
 116                 return -ENOENT;
 117         *next = index + 1;
 118         return 0;
 119 }
 120
 121 struct xdp_sock *__xsk_map_lookup_elem(struct bpf_map *map, u32 key)
 122 {
 123         struct xsk_map *m = container_of(map, struct xsk_map, map);
 124         struct xdp_sock *xs;
 125
 126         if (key >= map->max_entries)
 127                 return NULL;
 128
 129         xs = READ_ONCE(m->xsk_map[key]);
 130         return xs;
 131 }
 132
 133 int __xsk_map_redirect(struct bpf_map *map, struct xdp_buff *xdp,
 134                        struct xdp_sock *xs)
 135 {
 136         struct xsk_map *m = container_of(map, struct xsk_map, map);
 137         struct list_head *flush_list = this_cpu_ptr(m->flush_list);
 138         int err;
 139
 140         err = xsk_rcv(xs, xdp);
 141         if (err)
 142                 return err;
 143
 144         if (!xs->flush_node.prev)
 145                 list_add(&xs->flush_node, flush_list);
 146
 147         return 0;
 148 }
 149
 150 void __xsk_map_flush(struct bpf_map *map)
 151 {
 152         struct xsk_map *m = container_of(map, struct xsk_map, map);
 153         struct list_head *flush_list = this_cpu_ptr(m->flush_list);
 154         struct xdp_sock *xs, *tmp;
 155
 156         list_for_each_entry_safe(xs, tmp, flush_list, flush_node) {
 157                 xsk_flush(xs);
 158                 __list_del(xs->flush_node.prev, xs->flush_node.next);
 159                 xs->flush_node.prev = NULL;
 160         }
 161 }
 162
 163 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
 164 {
 165         return NULL;
 166 }
 167
 168 static int xsk_map_update_elem(struct bpf_map *map, void *key, void *value,
 169                                u64 map_flags)
 170 {
 171         struct xsk_map *m = container_of(map, struct xsk_map, map);
 172         u32 i = *(u32 *)key, fd = *(u32 *)value;
 173         struct xdp_sock *xs, *old_xs;
 174         struct socket *sock;
 175         int err;
 176
 177         if (unlikely(map_flags > BPF_EXIST))
 178                 return -EINVAL;
 179         if (unlikely(i >= m->map.max_entries))
 180                 return -E2BIG;
 181         if (unlikely(map_flags == BPF_NOEXIST))
 182                 return -EEXIST;
 183
 184         sock = sockfd_lookup(fd, &err);
 185         if (!sock)
 186                 return err;
 187
 188         if (sock->sk->sk_family != PF_XDP) {
 189                 sockfd_put(sock);
 190                 return -EOPNOTSUPP;
 191         }
 192
 193         xs = (struct xdp_sock *)sock->sk;
 194
 195         if (!xsk_is_setup_for_bpf_map(xs)) {
 196                 sockfd_put(sock);
 197                 return -EOPNOTSUPP;
 198         }
 199
 200         sock_hold(sock->sk);
 201
 202         old_xs = xchg(&m->xsk_map[i], xs);
 203         if (old_xs) {
 204                 /* Make sure we've flushed everything. */
 205                 synchronize_net();
 206                 sock_put((struct sock *)old_xs);
 207         }
 208
 209         sockfd_put(sock);
 210         return 0;
 211 }
 212
 213 static int xsk_map_delete_elem(struct bpf_map *map, void *key)
 214 {
 215         struct xsk_map *m = container_of(map, struct xsk_map, map);
 216         struct xdp_sock *old_xs;
 217         int k = *(u32 *)key;
 218
 219         if (k >= map->max_entries)
 220                 return -EINVAL;
 221
 222         old_xs = xchg(&m->xsk_map[k], NULL);
 223         if (old_xs) {
 224                 /* Make sure we've flushed everything. */
 225                 synchronize_net();
 226                 sock_put((struct sock *)old_xs);
 227         }
 228
 229         return 0;
 230 }
 231
 232 const struct bpf_map_ops xsk_map_ops = {
 233         .map_alloc = xsk_map_alloc,
 234         .map_free = xsk_map_free,
 235         .map_get_next_key = xsk_map_get_next_key,
 236         .map_lookup_elem = xsk_map_lookup_elem,
 237         .map_update_elem = xsk_map_update_elem,
 238         .map_delete_elem = xsk_map_delete_elem,
 239 };
 240
 241