kernel/bpf/reuseport_array.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2018 Facebook
   4  */
   5 #include <linux/bpf.h>
   6 #include <linux/err.h>
   7 #include <linux/sock_diag.h>
   8 #include <net/sock_reuseport.h>
   9
  10 struct reuseport_array {
  11         struct bpf_map map;
  12         struct sock __rcu *ptrs[];
  13 };
  14
  15 static struct reuseport_array *reuseport_array(struct bpf_map *map)
  16 {
  17         return (struct reuseport_array *)map;
  18 }
  19
  20 /* The caller must hold the reuseport_lock */
  21 void bpf_sk_reuseport_detach(struct sock *sk)
  22 {
  23         struct sock __rcu **socks;
  24
  25         write_lock_bh(&sk->sk_callback_lock);
  26         socks = sk->sk_user_data;
  27         if (socks) {
  28                 WRITE_ONCE(sk->sk_user_data, NULL);
  29                 /*
  30                  * Do not move this NULL assignment outside of
  31                  * sk->sk_callback_lock because there is
  32                  * a race with reuseport_array_free()
  33                  * which does not hold the reuseport_lock.
  34                  */
  35                 RCU_INIT_POINTER(*socks, NULL);
  36         }
  37         write_unlock_bh(&sk->sk_callback_lock);
  38 }
  39
  40 static int reuseport_array_alloc_check(union bpf_attr *attr)
  41 {
  42         if (attr->value_size != sizeof(u32) &&
  43             attr->value_size != sizeof(u64))
  44                 return -EINVAL;
  45
  46         return array_map_alloc_check(attr);
  47 }
  48
  49 static void *reuseport_array_lookup_elem(struct bpf_map *map, void *key)
  50 {
  51         struct reuseport_array *array = reuseport_array(map);
  52         u32 index = *(u32 *)key;
  53
  54         if (unlikely(index >= array->map.max_entries))
  55                 return NULL;
  56
  57         return rcu_dereference(array->ptrs[index]);
  58 }
  59
  60 /* Called from syscall only */
  61 static int reuseport_array_delete_elem(struct bpf_map *map, void *key)
  62 {
  63         struct reuseport_array *array = reuseport_array(map);
  64         u32 index = *(u32 *)key;
  65         struct sock *sk;
  66         int err;
  67
  68         if (index >= map->max_entries)
  69                 return -E2BIG;
  70
  71         if (!rcu_access_pointer(array->ptrs[index]))
  72                 return -ENOENT;
  73
  74         spin_lock_bh(&reuseport_lock);
  75
  76         sk = rcu_dereference_protected(array->ptrs[index],
  77                                        lockdep_is_held(&reuseport_lock));
  78         if (sk) {
  79                 write_lock_bh(&sk->sk_callback_lock);
  80                 WRITE_ONCE(sk->sk_user_data, NULL);
  81                 RCU_INIT_POINTER(array->ptrs[index], NULL);
  82                 write_unlock_bh(&sk->sk_callback_lock);
  83                 err = 0;
  84         } else {
  85                 err = -ENOENT;
  86         }
  87
  88         spin_unlock_bh(&reuseport_lock);
  89
  90         return err;
  91 }
  92
  93 static void reuseport_array_free(struct bpf_map *map)
  94 {
  95         struct reuseport_array *array = reuseport_array(map);
  96         struct sock *sk;
  97         u32 i;
  98
  99         synchronize_rcu();
 100
 101         /*
 102          * ops->map_*_elem() will not be able to access this
 103          * array now. Hence, this function only races with
 104          * bpf_sk_reuseport_detach() which was triggerred by
 105          * close() or disconnect().
 106          *
 107          * This function and bpf_sk_reuseport_detach() are
 108          * both removing sk from "array".  Who removes it
 109          * first does not matter.
 110          *
 111          * The only concern here is bpf_sk_reuseport_detach()
 112          * may access "array" which is being freed here.
 113          * bpf_sk_reuseport_detach() access this "array"
 114          * through sk->sk_user_data _and_ with sk->sk_callback_lock
 115          * held which is enough because this "array" is not freed
 116          * until all sk->sk_user_data has stopped referencing this "array".
 117          *
 118          * Hence, due to the above, taking "reuseport_lock" is not
 119          * needed here.
 120          */
 121
 122         /*
 123          * Since reuseport_lock is not taken, sk is accessed under
 124          * rcu_read_lock()
 125          */
 126         rcu_read_lock();
 127         for (i = 0; i < map->max_entries; i++) {
 128                 sk = rcu_dereference(array->ptrs[i]);
 129                 if (sk) {
 130                         write_lock_bh(&sk->sk_callback_lock);
 131                         /*
 132                          * No need for WRITE_ONCE(). At this point,
 133                          * no one is reading it without taking the
 134                          * sk->sk_callback_lock.
 135                          */
 136                         sk->sk_user_data = NULL;
 137                         write_unlock_bh(&sk->sk_callback_lock);
 138                         RCU_INIT_POINTER(array->ptrs[i], NULL);
 139                 }
 140         }
 141         rcu_read_unlock();
 142
 143         /*
 144          * Once reaching here, all sk->sk_user_data is not
 145          * referenceing this "array".  "array" can be freed now.
 146          */
 147         bpf_map_area_free(array);
 148 }
 149
 150 static struct bpf_map *reuseport_array_alloc(union bpf_attr *attr)
 151 {
 152         int err, numa_node = bpf_map_attr_numa_node(attr);
 153         struct reuseport_array *array;
 154         u64 cost, array_size;
 155
 156         if (!capable(CAP_SYS_ADMIN))
 157                 return ERR_PTR(-EPERM);
 158
 159         array_size = sizeof(*array);
 160         array_size += (u64)attr->max_entries * sizeof(struct sock *);
 161
 162         /* make sure there is no u32 overflow later in round_up() */
 163         cost = array_size;
 164         if (cost >= U32_MAX - PAGE_SIZE)
 165                 return ERR_PTR(-ENOMEM);
 166         cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
 167
 168         err = bpf_map_precharge_memlock(cost);
 169         if (err)
 170                 return ERR_PTR(err);
 171
 172         /* allocate all map elements and zero-initialize them */
 173         array = bpf_map_area_alloc(array_size, numa_node);
 174         if (!array)
 175                 return ERR_PTR(-ENOMEM);
 176
 177         /* copy mandatory map attributes */
 178         bpf_map_init_from_attr(&array->map, attr);
 179         array->map.pages = cost;
 180
 181         return &array->map;
 182 }
 183
 184 int bpf_fd_reuseport_array_lookup_elem(struct bpf_map *map, void *key,
 185                                        void *value)
 186 {
 187         struct sock *sk;
 188         int err;
 189
 190         if (map->value_size != sizeof(u64))
 191                 return -ENOSPC;
 192
 193         rcu_read_lock();
 194         sk = reuseport_array_lookup_elem(map, key);
 195         if (sk) {
 196                 *(u64 *)value = sock_gen_cookie(sk);
 197                 err = 0;
 198         } else {
 199                 err = -ENOENT;
 200         }
 201         rcu_read_unlock();
 202
 203         return err;
 204 }
 205
 206 static int
 207 reuseport_array_update_check(const struct reuseport_array *array,
 208                              const struct sock *nsk,
 209                              const struct sock *osk,
 210                              const struct sock_reuseport *nsk_reuse,
 211                              u32 map_flags)
 212 {
 213         if (osk && map_flags == BPF_NOEXIST)
 214                 return -EEXIST;
 215
 216         if (!osk && map_flags == BPF_EXIST)
 217                 return -ENOENT;
 218
 219         if (nsk->sk_protocol != IPPROTO_UDP && nsk->sk_protocol != IPPROTO_TCP)
 220                 return -ENOTSUPP;
 221
 222         if (nsk->sk_family != AF_INET && nsk->sk_family != AF_INET6)
 223                 return -ENOTSUPP;
 224
 225         if (nsk->sk_type != SOCK_STREAM && nsk->sk_type != SOCK_DGRAM)
 226                 return -ENOTSUPP;
 227
 228         /*
 229          * sk must be hashed (i.e. listening in the TCP case or binded
 230          * in the UDP case) and
 231          * it must also be a SO_REUSEPORT sk (i.e. reuse cannot be NULL).
 232          *
 233          * Also, sk will be used in bpf helper that is protected by
 234          * rcu_read_lock().
 235          */
 236         if (!sock_flag(nsk, SOCK_RCU_FREE) || !sk_hashed(nsk) || !nsk_reuse)
 237                 return -EINVAL;
 238
 239         /* READ_ONCE because the sk->sk_callback_lock may not be held here */
 240         if (READ_ONCE(nsk->sk_user_data))
 241                 return -EBUSY;
 242
 243         return 0;
 244 }
 245
 246 /*
 247  * Called from syscall only.
 248  * The "nsk" in the fd refcnt.
 249  * The "osk" and "reuse" are protected by reuseport_lock.
 250  */
 251 int bpf_fd_reuseport_array_update_elem(struct bpf_map *map, void *key,
 252                                        void *value, u64 map_flags)
 253 {
 254         struct reuseport_array *array = reuseport_array(map);
 255         struct sock *free_osk = NULL, *osk, *nsk;
 256         struct sock_reuseport *reuse;
 257         u32 index = *(u32 *)key;
 258         struct socket *socket;
 259         int err, fd;
 260
 261         if (map_flags > BPF_EXIST)
 262                 return -EINVAL;
 263
 264         if (index >= map->max_entries)
 265                 return -E2BIG;
 266
 267         if (map->value_size == sizeof(u64)) {
 268                 u64 fd64 = *(u64 *)value;
 269
 270                 if (fd64 > S32_MAX)
 271                         return -EINVAL;
 272                 fd = fd64;
 273         } else {
 274                 fd = *(int *)value;
 275         }
 276
 277         socket = sockfd_lookup(fd, &err);
 278         if (!socket)
 279                 return err;
 280
 281         nsk = socket->sk;
 282         if (!nsk) {
 283                 err = -EINVAL;
 284                 goto put_file;
 285         }
 286
 287         /* Quick checks before taking reuseport_lock */
 288         err = reuseport_array_update_check(array, nsk,
 289                                            rcu_access_pointer(array->ptrs[index]),
 290                                            rcu_access_pointer(nsk->sk_reuseport_cb),
 291                                            map_flags);
 292         if (err)
 293                 goto put_file;
 294
 295         spin_lock_bh(&reuseport_lock);
 296         /*
 297          * Some of the checks only need reuseport_lock
 298          * but it is done under sk_callback_lock also
 299          * for simplicity reason.
 300          */
 301         write_lock_bh(&nsk->sk_callback_lock);
 302
 303         osk = rcu_dereference_protected(array->ptrs[index],
 304                                         lockdep_is_held(&reuseport_lock));
 305         reuse = rcu_dereference_protected(nsk->sk_reuseport_cb,
 306                                           lockdep_is_held(&reuseport_lock));
 307         err = reuseport_array_update_check(array, nsk, osk, reuse, map_flags);
 308         if (err)
 309                 goto put_file_unlock;
 310
 311         /* Ensure reuse->reuseport_id is set */
 312         err = reuseport_get_id(reuse);
 313         if (err < 0)
 314                 goto put_file_unlock;
 315
 316         WRITE_ONCE(nsk->sk_user_data, &array->ptrs[index]);
 317         rcu_assign_pointer(array->ptrs[index], nsk);
 318         free_osk = osk;
 319         err = 0;
 320
 321 put_file_unlock:
 322         write_unlock_bh(&nsk->sk_callback_lock);
 323
 324         if (free_osk) {
 325                 write_lock_bh(&free_osk->sk_callback_lock);
 326                 WRITE_ONCE(free_osk->sk_user_data, NULL);
 327                 write_unlock_bh(&free_osk->sk_callback_lock);
 328         }
 329
 330         spin_unlock_bh(&reuseport_lock);
 331 put_file:
 332         fput(socket->file);
 333         return err;
 334 }
 335
 336 /* Called from syscall */
 337 static int reuseport_array_get_next_key(struct bpf_map *map, void *key,
 338                                         void *next_key)
 339 {
 340         struct reuseport_array *array = reuseport_array(map);
 341         u32 index = key ? *(u32 *)key : U32_MAX;
 342         u32 *next = (u32 *)next_key;
 343
 344         if (index >= array->map.max_entries) {
 345                 *next = 0;
 346                 return 0;
 347         }
 348
 349         if (index == array->map.max_entries - 1)
 350                 return -ENOENT;
 351
 352         *next = index + 1;
 353         return 0;
 354 }
 355
 356 const struct bpf_map_ops reuseport_array_ops = {
 357         .map_alloc_check = reuseport_array_alloc_check,
 358         .map_alloc = reuseport_array_alloc,
 359         .map_free = reuseport_array_free,
 360         .map_lookup_elem = reuseport_array_lookup_elem,
 361         .map_get_next_key = reuseport_array_get_next_key,
 362         .map_delete_elem = reuseport_array_delete_elem,
 363 };