net/xdp/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 #include "xsk.h"
  13
  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
  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
  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
  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
  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
  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         bpf_clear_redirect_map(map);
  91         synchronize_net();
  92         bpf_map_area_free(m);
  93 }
  94
  95 static int xsk_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
  96 {
  97         struct xsk_map *m = container_of(map, struct xsk_map, map);
  98         u32 index = key ? *(u32 *)key : U32_MAX;
  99         u32 *next = next_key;
 100
 101         if (index >= m->map.max_entries) {
 102                 *next = 0;
 103                 return 0;
 104         }
 105
 106         if (index == m->map.max_entries - 1)
 107                 return -ENOENT;
 108         *next = index + 1;
 109         return 0;
 110 }
 111
 112 static int xsk_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 113 {
 114         const int ret = BPF_REG_0, mp = BPF_REG_1, index = BPF_REG_2;
 115         struct bpf_insn *insn = insn_buf;
 116
 117         *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 118         *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
 119         *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(sizeof(struct xsk_sock *)));
 120         *insn++ = BPF_ALU64_IMM(BPF_ADD, mp, offsetof(struct xsk_map, xsk_map));
 121         *insn++ = BPF_ALU64_REG(BPF_ADD, ret, mp);
 122         *insn++ = BPF_LDX_MEM(BPF_SIZEOF(struct xsk_sock *), ret, ret, 0);
 123         *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 124         *insn++ = BPF_MOV64_IMM(ret, 0);
 125         return insn - insn_buf;
 126 }
 127
 128 static void *xsk_map_lookup_elem(struct bpf_map *map, void *key)
 129 {
 130         WARN_ON_ONCE(!rcu_read_lock_held());
 131         return __xsk_map_lookup_elem(map, *(u32 *)key);
 132 }
 133
 134 static void *xsk_map_lookup_elem_sys_only(struct bpf_map *map, void *key)
 135 {
 136         return ERR_PTR(-EOPNOTSUPP);
 137 }
 138
 139 static int xsk_map_update_elem(struct bpf_map *map, void *key, void *value,
 140                                u64 map_flags)
 141 {
 142         struct xsk_map *m = container_of(map, struct xsk_map, map);
 143         struct xdp_sock *xs, *old_xs, **map_entry;
 144         u32 i = *(u32 *)key, fd = *(u32 *)value;
 145         struct xsk_map_node *node;
 146         struct socket *sock;
 147         int err;
 148
 149         if (unlikely(map_flags > BPF_EXIST))
 150                 return -EINVAL;
 151         if (unlikely(i >= m->map.max_entries))
 152                 return -E2BIG;
 153
 154         sock = sockfd_lookup(fd, &err);
 155         if (!sock)
 156                 return err;
 157
 158         if (sock->sk->sk_family != PF_XDP) {
 159                 sockfd_put(sock);
 160                 return -EOPNOTSUPP;
 161         }
 162
 163         xs = (struct xdp_sock *)sock->sk;
 164
 165         map_entry = &m->xsk_map[i];
 166         node = xsk_map_node_alloc(m, map_entry);
 167         if (IS_ERR(node)) {
 168                 sockfd_put(sock);
 169                 return PTR_ERR(node);
 170         }
 171
 172         spin_lock_bh(&m->lock);
 173         old_xs = READ_ONCE(*map_entry);
 174         if (old_xs == xs) {
 175                 err = 0;
 176                 goto out;
 177         } else if (old_xs && map_flags == BPF_NOEXIST) {
 178                 err = -EEXIST;
 179                 goto out;
 180         } else if (!old_xs && map_flags == BPF_EXIST) {
 181                 err = -ENOENT;
 182                 goto out;
 183         }
 184         xsk_map_sock_add(xs, node);
 185         WRITE_ONCE(*map_entry, xs);
 186         if (old_xs)
 187                 xsk_map_sock_delete(old_xs, map_entry);
 188         spin_unlock_bh(&m->lock);
 189         sockfd_put(sock);
 190         return 0;
 191
 192 out:
 193         spin_unlock_bh(&m->lock);
 194         sockfd_put(sock);
 195         xsk_map_node_free(node);
 196         return err;
 197 }
 198
 199 static int xsk_map_delete_elem(struct bpf_map *map, void *key)
 200 {
 201         struct xsk_map *m = container_of(map, struct xsk_map, map);
 202         struct xdp_sock *old_xs, **map_entry;
 203         int k = *(u32 *)key;
 204
 205         if (k >= map->max_entries)
 206                 return -EINVAL;
 207
 208         spin_lock_bh(&m->lock);
 209         map_entry = &m->xsk_map[k];
 210         old_xs = xchg(map_entry, NULL);
 211         if (old_xs)
 212                 xsk_map_sock_delete(old_xs, map_entry);
 213         spin_unlock_bh(&m->lock);
 214
 215         return 0;
 216 }
 217
 218 void xsk_map_try_sock_delete(struct xsk_map *map, struct xdp_sock *xs,
 219                              struct xdp_sock **map_entry)
 220 {
 221         spin_lock_bh(&map->lock);
 222         if (READ_ONCE(*map_entry) == xs) {
 223                 WRITE_ONCE(*map_entry, NULL);
 224                 xsk_map_sock_delete(xs, map_entry);
 225         }
 226         spin_unlock_bh(&map->lock);
 227 }
 228
 229 static bool xsk_map_meta_equal(const struct bpf_map *meta0,
 230                                const struct bpf_map *meta1)
 231 {
 232         return meta0->max_entries == meta1->max_entries &&
 233                 bpf_map_meta_equal(meta0, meta1);
 234 }
 235
 236 static int xsk_map_btf_id;
 237 const struct bpf_map_ops xsk_map_ops = {
 238         .map_meta_equal = xsk_map_meta_equal,
 239         .map_alloc = xsk_map_alloc,
 240         .map_free = xsk_map_free,
 241         .map_get_next_key = xsk_map_get_next_key,
 242         .map_lookup_elem = xsk_map_lookup_elem,
 243         .map_gen_lookup = xsk_map_gen_lookup,
 244         .map_lookup_elem_sys_only = xsk_map_lookup_elem_sys_only,
 245         .map_update_elem = xsk_map_update_elem,
 246         .map_delete_elem = xsk_map_delete_elem,
 247         .map_check_btf = map_check_no_btf,
 248         .map_btf_name = "xsk_map",
 249         .map_btf_id = &xsk_map_btf_id,
 250 };