net/core/bpf_sk_storage.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2019 Facebook  */
   3 #include <linux/rculist.h>
   4 #include <linux/list.h>
   5 #include <linux/hash.h>
   6 #include <linux/types.h>
   7 #include <linux/spinlock.h>
   8 #include <linux/bpf.h>
   9 #include <linux/btf.h>
  10 #include <linux/btf_ids.h>
  11 #include <linux/bpf_local_storage.h>
  12 #include <net/bpf_sk_storage.h>
  13 #include <net/sock.h>
  14 #include <uapi/linux/sock_diag.h>
  15 #include <uapi/linux/btf.h>
  16 #include <linux/rcupdate_trace.h>
  17
  18 DEFINE_BPF_STORAGE_CACHE(sk_cache);
  19
  20 static struct bpf_local_storage_data *
  21 bpf_sk_storage_lookup(struct sock *sk, struct bpf_map *map, bool cacheit_lockit)
  22 {
  23         struct bpf_local_storage *sk_storage;
  24         struct bpf_local_storage_map *smap;
  25
  26         sk_storage =
  27                 rcu_dereference_check(sk->sk_bpf_storage, bpf_rcu_lock_held());
  28         if (!sk_storage)
  29                 return NULL;
  30
  31         smap = (struct bpf_local_storage_map *)map;
  32         return bpf_local_storage_lookup(sk_storage, smap, cacheit_lockit);
  33 }
  34
  35 static int bpf_sk_storage_del(struct sock *sk, struct bpf_map *map)
  36 {
  37         struct bpf_local_storage_data *sdata;
  38
  39         sdata = bpf_sk_storage_lookup(sk, map, false);
  40         if (!sdata)
  41                 return -ENOENT;
  42
  43         bpf_selem_unlink(SELEM(sdata));
  44
  45         return 0;
  46 }
  47
  48 /* Called by __sk_destruct() & bpf_sk_storage_clone() */
  49 void bpf_sk_storage_free(struct sock *sk)
  50 {
  51         struct bpf_local_storage_elem *selem;
  52         struct bpf_local_storage *sk_storage;
  53         bool free_sk_storage = false;
  54         struct hlist_node *n;
  55
  56         rcu_read_lock();
  57         sk_storage = rcu_dereference(sk->sk_bpf_storage);
  58         if (!sk_storage) {
  59                 rcu_read_unlock();
  60                 return;
  61         }
  62
  63         /* Netiher the bpf_prog nor the bpf-map's syscall
  64          * could be modifying the sk_storage->list now.
  65          * Thus, no elem can be added-to or deleted-from the
  66          * sk_storage->list by the bpf_prog or by the bpf-map's syscall.
  67          *
  68          * It is racing with bpf_local_storage_map_free() alone
  69          * when unlinking elem from the sk_storage->list and
  70          * the map's bucket->list.
  71          */
  72         raw_spin_lock_bh(&sk_storage->lock);
  73         hlist_for_each_entry_safe(selem, n, &sk_storage->list, snode) {
  74                 /* Always unlink from map before unlinking from
  75                  * sk_storage.
  76                  */
  77                 bpf_selem_unlink_map(selem);
  78                 free_sk_storage = bpf_selem_unlink_storage_nolock(sk_storage,
  79                                                                   selem, true);
  80         }
  81         raw_spin_unlock_bh(&sk_storage->lock);
  82         rcu_read_unlock();
  83
  84         if (free_sk_storage)
  85                 kfree_rcu(sk_storage, rcu);
  86 }
  87
  88 static void bpf_sk_storage_map_free(struct bpf_map *map)
  89 {
  90         struct bpf_local_storage_map *smap;
  91
  92         smap = (struct bpf_local_storage_map *)map;
  93         bpf_local_storage_cache_idx_free(&sk_cache, smap->cache_idx);
  94         bpf_local_storage_map_free(smap, NULL);
  95 }
  96
  97 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
  98 {
  99         struct bpf_local_storage_map *smap;
 100
 101         smap = bpf_local_storage_map_alloc(attr);
 102         if (IS_ERR(smap))
 103                 return ERR_CAST(smap);
 104
 105         smap->cache_idx = bpf_local_storage_cache_idx_get(&sk_cache);
 106         return &smap->map;
 107 }
 108
 109 static int notsupp_get_next_key(struct bpf_map *map, void *key,
 110                                 void *next_key)
 111 {
 112         return -ENOTSUPP;
 113 }
 114
 115 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
 116 {
 117         struct bpf_local_storage_data *sdata;
 118         struct socket *sock;
 119         int fd, err;
 120
 121         fd = *(int *)key;
 122         sock = sockfd_lookup(fd, &err);
 123         if (sock) {
 124                 sdata = bpf_sk_storage_lookup(sock->sk, map, true);
 125                 sockfd_put(sock);
 126                 return sdata ? sdata->data : NULL;
 127         }
 128
 129         return ERR_PTR(err);
 130 }
 131
 132 static int bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
 133                                          void *value, u64 map_flags)
 134 {
 135         struct bpf_local_storage_data *sdata;
 136         struct socket *sock;
 137         int fd, err;
 138
 139         fd = *(int *)key;
 140         sock = sockfd_lookup(fd, &err);
 141         if (sock) {
 142                 sdata = bpf_local_storage_update(
 143                         sock->sk, (struct bpf_local_storage_map *)map, value,
 144                         map_flags);
 145                 sockfd_put(sock);
 146                 return PTR_ERR_OR_ZERO(sdata);
 147         }
 148
 149         return err;
 150 }
 151
 152 static int bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
 153 {
 154         struct socket *sock;
 155         int fd, err;
 156
 157         fd = *(int *)key;
 158         sock = sockfd_lookup(fd, &err);
 159         if (sock) {
 160                 err = bpf_sk_storage_del(sock->sk, map);
 161                 sockfd_put(sock);
 162                 return err;
 163         }
 164
 165         return err;
 166 }
 167
 168 static struct bpf_local_storage_elem *
 169 bpf_sk_storage_clone_elem(struct sock *newsk,
 170                           struct bpf_local_storage_map *smap,
 171                           struct bpf_local_storage_elem *selem)
 172 {
 173         struct bpf_local_storage_elem *copy_selem;
 174
 175         copy_selem = bpf_selem_alloc(smap, newsk, NULL, true);
 176         if (!copy_selem)
 177                 return NULL;
 178
 179         if (map_value_has_spin_lock(&smap->map))
 180                 copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
 181                                       SDATA(selem)->data, true);
 182         else
 183                 copy_map_value(&smap->map, SDATA(copy_selem)->data,
 184                                SDATA(selem)->data);
 185
 186         return copy_selem;
 187 }
 188
 189 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
 190 {
 191         struct bpf_local_storage *new_sk_storage = NULL;
 192         struct bpf_local_storage *sk_storage;
 193         struct bpf_local_storage_elem *selem;
 194         int ret = 0;
 195
 196         RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
 197
 198         rcu_read_lock();
 199         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 200
 201         if (!sk_storage || hlist_empty(&sk_storage->list))
 202                 goto out;
 203
 204         hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
 205                 struct bpf_local_storage_elem *copy_selem;
 206                 struct bpf_local_storage_map *smap;
 207                 struct bpf_map *map;
 208
 209                 smap = rcu_dereference(SDATA(selem)->smap);
 210                 if (!(smap->map.map_flags & BPF_F_CLONE))
 211                         continue;
 212
 213                 /* Note that for lockless listeners adding new element
 214                  * here can race with cleanup in bpf_local_storage_map_free.
 215                  * Try to grab map refcnt to make sure that it's still
 216                  * alive and prevent concurrent removal.
 217                  */
 218                 map = bpf_map_inc_not_zero(&smap->map);
 219                 if (IS_ERR(map))
 220                         continue;
 221
 222                 copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
 223                 if (!copy_selem) {
 224                         ret = -ENOMEM;
 225                         bpf_map_put(map);
 226                         goto out;
 227                 }
 228
 229                 if (new_sk_storage) {
 230                         bpf_selem_link_map(smap, copy_selem);
 231                         bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
 232                 } else {
 233                         ret = bpf_local_storage_alloc(newsk, smap, copy_selem);
 234                         if (ret) {
 235                                 kfree(copy_selem);
 236                                 atomic_sub(smap->elem_size,
 237                                            &newsk->sk_omem_alloc);
 238                                 bpf_map_put(map);
 239                                 goto out;
 240                         }
 241
 242                         new_sk_storage =
 243                                 rcu_dereference(copy_selem->local_storage);
 244                 }
 245                 bpf_map_put(map);
 246         }
 247
 248 out:
 249         rcu_read_unlock();
 250
 251         /* In case of an error, don't free anything explicitly here, the
 252          * caller is responsible to call bpf_sk_storage_free.
 253          */
 254
 255         return ret;
 256 }
 257
 258 BPF_CALL_4(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
 259            void *, value, u64, flags)
 260 {
 261         struct bpf_local_storage_data *sdata;
 262
 263         WARN_ON_ONCE(!bpf_rcu_lock_held());
 264         if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
 265                 return (unsigned long)NULL;
 266
 267         sdata = bpf_sk_storage_lookup(sk, map, true);
 268         if (sdata)
 269                 return (unsigned long)sdata->data;
 270
 271         if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
 272             /* Cannot add new elem to a going away sk.
 273              * Otherwise, the new elem may become a leak
 274              * (and also other memory issues during map
 275              *  destruction).
 276              */
 277             refcount_inc_not_zero(&sk->sk_refcnt)) {
 278                 sdata = bpf_local_storage_update(
 279                         sk, (struct bpf_local_storage_map *)map, value,
 280                         BPF_NOEXIST);
 281                 /* sk must be a fullsock (guaranteed by verifier),
 282                  * so sock_gen_put() is unnecessary.
 283                  */
 284                 sock_put(sk);
 285                 return IS_ERR(sdata) ?
 286                         (unsigned long)NULL : (unsigned long)sdata->data;
 287         }
 288
 289         return (unsigned long)NULL;
 290 }
 291
 292 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
 293 {
 294         WARN_ON_ONCE(!bpf_rcu_lock_held());
 295         if (!sk || !sk_fullsock(sk))
 296                 return -EINVAL;
 297
 298         if (refcount_inc_not_zero(&sk->sk_refcnt)) {
 299                 int err;
 300
 301                 err = bpf_sk_storage_del(sk, map);
 302                 sock_put(sk);
 303                 return err;
 304         }
 305
 306         return -ENOENT;
 307 }
 308
 309 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
 310                                  void *owner, u32 size)
 311 {
 312         struct sock *sk = (struct sock *)owner;
 313
 314         /* same check as in sock_kmalloc() */
 315         if (size <= sysctl_optmem_max &&
 316             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 317                 atomic_add(size, &sk->sk_omem_alloc);
 318                 return 0;
 319         }
 320
 321         return -ENOMEM;
 322 }
 323
 324 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
 325                                     void *owner, u32 size)
 326 {
 327         struct sock *sk = owner;
 328
 329         atomic_sub(size, &sk->sk_omem_alloc);
 330 }
 331
 332 static struct bpf_local_storage __rcu **
 333 bpf_sk_storage_ptr(void *owner)
 334 {
 335         struct sock *sk = owner;
 336
 337         return &sk->sk_bpf_storage;
 338 }
 339
 340 static int sk_storage_map_btf_id;
 341 const struct bpf_map_ops sk_storage_map_ops = {
 342         .map_meta_equal = bpf_map_meta_equal,
 343         .map_alloc_check = bpf_local_storage_map_alloc_check,
 344         .map_alloc = bpf_sk_storage_map_alloc,
 345         .map_free = bpf_sk_storage_map_free,
 346         .map_get_next_key = notsupp_get_next_key,
 347         .map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
 348         .map_update_elem = bpf_fd_sk_storage_update_elem,
 349         .map_delete_elem = bpf_fd_sk_storage_delete_elem,
 350         .map_check_btf = bpf_local_storage_map_check_btf,
 351         .map_btf_name = "bpf_local_storage_map",
 352         .map_btf_id = &sk_storage_map_btf_id,
 353         .map_local_storage_charge = bpf_sk_storage_charge,
 354         .map_local_storage_uncharge = bpf_sk_storage_uncharge,
 355         .map_owner_storage_ptr = bpf_sk_storage_ptr,
 356 };
 357
 358 const struct bpf_func_proto bpf_sk_storage_get_proto = {
 359         .func           = bpf_sk_storage_get,
 360         .gpl_only       = false,
 361         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 362         .arg1_type      = ARG_CONST_MAP_PTR,
 363         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
 364         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 365         .arg4_type      = ARG_ANYTHING,
 366 };
 367
 368 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
 369         .func           = bpf_sk_storage_get,
 370         .gpl_only       = false,
 371         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 372         .arg1_type      = ARG_CONST_MAP_PTR,
 373         .arg2_type      = ARG_PTR_TO_CTX, /* context is 'struct sock' */
 374         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 375         .arg4_type      = ARG_ANYTHING,
 376 };
 377
 378 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
 379         .func           = bpf_sk_storage_delete,
 380         .gpl_only       = false,
 381         .ret_type       = RET_INTEGER,
 382         .arg1_type      = ARG_CONST_MAP_PTR,
 383         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
 384 };
 385
 386 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
 387 {
 388         const struct btf *btf_vmlinux;
 389         const struct btf_type *t;
 390         const char *tname;
 391         u32 btf_id;
 392
 393         if (prog->aux->dst_prog)
 394                 return false;
 395
 396         /* Ensure the tracing program is not tracing
 397          * any bpf_sk_storage*() function and also
 398          * use the bpf_sk_storage_(get|delete) helper.
 399          */
 400         switch (prog->expected_attach_type) {
 401         case BPF_TRACE_ITER:
 402         case BPF_TRACE_RAW_TP:
 403                 /* bpf_sk_storage has no trace point */
 404                 return true;
 405         case BPF_TRACE_FENTRY:
 406         case BPF_TRACE_FEXIT:
 407                 btf_vmlinux = bpf_get_btf_vmlinux();
 408                 btf_id = prog->aux->attach_btf_id;
 409                 t = btf_type_by_id(btf_vmlinux, btf_id);
 410                 tname = btf_name_by_offset(btf_vmlinux, t->name_off);
 411                 return !!strncmp(tname, "bpf_sk_storage",
 412                                  strlen("bpf_sk_storage"));
 413         default:
 414                 return false;
 415         }
 416
 417         return false;
 418 }
 419
 420 BPF_CALL_4(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
 421            void *, value, u64, flags)
 422 {
 423         WARN_ON_ONCE(!bpf_rcu_lock_held());
 424         if (in_hardirq() || in_nmi())
 425                 return (unsigned long)NULL;
 426
 427         return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags);
 428 }
 429
 430 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
 431            struct sock *, sk)
 432 {
 433         WARN_ON_ONCE(!bpf_rcu_lock_held());
 434         if (in_hardirq() || in_nmi())
 435                 return -EPERM;
 436
 437         return ____bpf_sk_storage_delete(map, sk);
 438 }
 439
 440 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
 441         .func           = bpf_sk_storage_get_tracing,
 442         .gpl_only       = false,
 443         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 444         .arg1_type      = ARG_CONST_MAP_PTR,
 445         .arg2_type      = ARG_PTR_TO_BTF_ID,
 446         .arg2_btf_id    = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
 447         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 448         .arg4_type      = ARG_ANYTHING,
 449         .allowed        = bpf_sk_storage_tracing_allowed,
 450 };
 451
 452 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
 453         .func           = bpf_sk_storage_delete_tracing,
 454         .gpl_only       = false,
 455         .ret_type       = RET_INTEGER,
 456         .arg1_type      = ARG_CONST_MAP_PTR,
 457         .arg2_type      = ARG_PTR_TO_BTF_ID,
 458         .arg2_btf_id    = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
 459         .allowed        = bpf_sk_storage_tracing_allowed,
 460 };
 461
 462 struct bpf_sk_storage_diag {
 463         u32 nr_maps;
 464         struct bpf_map *maps[];
 465 };
 466
 467 /* The reply will be like:
 468  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
 469  *      SK_DIAG_BPF_STORAGE (nla_nest)
 470  *              SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 471  *              SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 472  *      SK_DIAG_BPF_STORAGE (nla_nest)
 473  *              SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 474  *              SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 475  *      ....
 476  */
 477 static int nla_value_size(u32 value_size)
 478 {
 479         /* SK_DIAG_BPF_STORAGE (nla_nest)
 480          *      SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 481          *      SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 482          */
 483         return nla_total_size(0) + nla_total_size(sizeof(u32)) +
 484                 nla_total_size_64bit(value_size);
 485 }
 486
 487 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
 488 {
 489         u32 i;
 490
 491         if (!diag)
 492                 return;
 493
 494         for (i = 0; i < diag->nr_maps; i++)
 495                 bpf_map_put(diag->maps[i]);
 496
 497         kfree(diag);
 498 }
 499 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
 500
 501 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
 502                            const struct bpf_map *map)
 503 {
 504         u32 i;
 505
 506         for (i = 0; i < diag->nr_maps; i++) {
 507                 if (diag->maps[i] == map)
 508                         return true;
 509         }
 510
 511         return false;
 512 }
 513
 514 struct bpf_sk_storage_diag *
 515 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
 516 {
 517         struct bpf_sk_storage_diag *diag;
 518         struct nlattr *nla;
 519         u32 nr_maps = 0;
 520         int rem, err;
 521
 522         /* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
 523          * the map_alloc_check() side also does.
 524          */
 525         if (!bpf_capable())
 526                 return ERR_PTR(-EPERM);
 527
 528         nla_for_each_nested(nla, nla_stgs, rem) {
 529                 if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
 530                         nr_maps++;
 531         }
 532
 533         diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL);
 534         if (!diag)
 535                 return ERR_PTR(-ENOMEM);
 536
 537         nla_for_each_nested(nla, nla_stgs, rem) {
 538                 struct bpf_map *map;
 539                 int map_fd;
 540
 541                 if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
 542                         continue;
 543
 544                 map_fd = nla_get_u32(nla);
 545                 map = bpf_map_get(map_fd);
 546                 if (IS_ERR(map)) {
 547                         err = PTR_ERR(map);
 548                         goto err_free;
 549                 }
 550                 if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
 551                         bpf_map_put(map);
 552                         err = -EINVAL;
 553                         goto err_free;
 554                 }
 555                 if (diag_check_dup(diag, map)) {
 556                         bpf_map_put(map);
 557                         err = -EEXIST;
 558                         goto err_free;
 559                 }
 560                 diag->maps[diag->nr_maps++] = map;
 561         }
 562
 563         return diag;
 564
 565 err_free:
 566         bpf_sk_storage_diag_free(diag);
 567         return ERR_PTR(err);
 568 }
 569 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
 570
 571 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
 572 {
 573         struct nlattr *nla_stg, *nla_value;
 574         struct bpf_local_storage_map *smap;
 575
 576         /* It cannot exceed max nlattr's payload */
 577         BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
 578
 579         nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
 580         if (!nla_stg)
 581                 return -EMSGSIZE;
 582
 583         smap = rcu_dereference(sdata->smap);
 584         if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
 585                 goto errout;
 586
 587         nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
 588                                       smap->map.value_size,
 589                                       SK_DIAG_BPF_STORAGE_PAD);
 590         if (!nla_value)
 591                 goto errout;
 592
 593         if (map_value_has_spin_lock(&smap->map))
 594                 copy_map_value_locked(&smap->map, nla_data(nla_value),
 595                                       sdata->data, true);
 596         else
 597                 copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
 598
 599         nla_nest_end(skb, nla_stg);
 600         return 0;
 601
 602 errout:
 603         nla_nest_cancel(skb, nla_stg);
 604         return -EMSGSIZE;
 605 }
 606
 607 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
 608                                        int stg_array_type,
 609                                        unsigned int *res_diag_size)
 610 {
 611         /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
 612         unsigned int diag_size = nla_total_size(0);
 613         struct bpf_local_storage *sk_storage;
 614         struct bpf_local_storage_elem *selem;
 615         struct bpf_local_storage_map *smap;
 616         struct nlattr *nla_stgs;
 617         unsigned int saved_len;
 618         int err = 0;
 619
 620         rcu_read_lock();
 621
 622         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 623         if (!sk_storage || hlist_empty(&sk_storage->list)) {
 624                 rcu_read_unlock();
 625                 return 0;
 626         }
 627
 628         nla_stgs = nla_nest_start(skb, stg_array_type);
 629         if (!nla_stgs)
 630                 /* Continue to learn diag_size */
 631                 err = -EMSGSIZE;
 632
 633         saved_len = skb->len;
 634         hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
 635                 smap = rcu_dereference(SDATA(selem)->smap);
 636                 diag_size += nla_value_size(smap->map.value_size);
 637
 638                 if (nla_stgs && diag_get(SDATA(selem), skb))
 639                         /* Continue to learn diag_size */
 640                         err = -EMSGSIZE;
 641         }
 642
 643         rcu_read_unlock();
 644
 645         if (nla_stgs) {
 646                 if (saved_len == skb->len)
 647                         nla_nest_cancel(skb, nla_stgs);
 648                 else
 649                         nla_nest_end(skb, nla_stgs);
 650         }
 651
 652         if (diag_size == nla_total_size(0)) {
 653                 *res_diag_size = 0;
 654                 return 0;
 655         }
 656
 657         *res_diag_size = diag_size;
 658         return err;
 659 }
 660
 661 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
 662                             struct sock *sk, struct sk_buff *skb,
 663                             int stg_array_type,
 664                             unsigned int *res_diag_size)
 665 {
 666         /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
 667         unsigned int diag_size = nla_total_size(0);
 668         struct bpf_local_storage *sk_storage;
 669         struct bpf_local_storage_data *sdata;
 670         struct nlattr *nla_stgs;
 671         unsigned int saved_len;
 672         int err = 0;
 673         u32 i;
 674
 675         *res_diag_size = 0;
 676
 677         /* No map has been specified.  Dump all. */
 678         if (!diag->nr_maps)
 679                 return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
 680                                                    res_diag_size);
 681
 682         rcu_read_lock();
 683         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 684         if (!sk_storage || hlist_empty(&sk_storage->list)) {
 685                 rcu_read_unlock();
 686                 return 0;
 687         }
 688
 689         nla_stgs = nla_nest_start(skb, stg_array_type);
 690         if (!nla_stgs)
 691                 /* Continue to learn diag_size */
 692                 err = -EMSGSIZE;
 693
 694         saved_len = skb->len;
 695         for (i = 0; i < diag->nr_maps; i++) {
 696                 sdata = bpf_local_storage_lookup(sk_storage,
 697                                 (struct bpf_local_storage_map *)diag->maps[i],
 698                                 false);
 699
 700                 if (!sdata)
 701                         continue;
 702
 703                 diag_size += nla_value_size(diag->maps[i]->value_size);
 704
 705                 if (nla_stgs && diag_get(sdata, skb))
 706                         /* Continue to learn diag_size */
 707                         err = -EMSGSIZE;
 708         }
 709         rcu_read_unlock();
 710
 711         if (nla_stgs) {
 712                 if (saved_len == skb->len)
 713                         nla_nest_cancel(skb, nla_stgs);
 714                 else
 715                         nla_nest_end(skb, nla_stgs);
 716         }
 717
 718         if (diag_size == nla_total_size(0)) {
 719                 *res_diag_size = 0;
 720                 return 0;
 721         }
 722
 723         *res_diag_size = diag_size;
 724         return err;
 725 }
 726 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
 727
 728 struct bpf_iter_seq_sk_storage_map_info {
 729         struct bpf_map *map;
 730         unsigned int bucket_id;
 731         unsigned skip_elems;
 732 };
 733
 734 static struct bpf_local_storage_elem *
 735 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
 736                                  struct bpf_local_storage_elem *prev_selem)
 737         __acquires(RCU) __releases(RCU)
 738 {
 739         struct bpf_local_storage *sk_storage;
 740         struct bpf_local_storage_elem *selem;
 741         u32 skip_elems = info->skip_elems;
 742         struct bpf_local_storage_map *smap;
 743         u32 bucket_id = info->bucket_id;
 744         u32 i, count, n_buckets;
 745         struct bpf_local_storage_map_bucket *b;
 746
 747         smap = (struct bpf_local_storage_map *)info->map;
 748         n_buckets = 1U << smap->bucket_log;
 749         if (bucket_id >= n_buckets)
 750                 return NULL;
 751
 752         /* try to find next selem in the same bucket */
 753         selem = prev_selem;
 754         count = 0;
 755         while (selem) {
 756                 selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
 757                                          struct bpf_local_storage_elem, map_node);
 758                 if (!selem) {
 759                         /* not found, unlock and go to the next bucket */
 760                         b = &smap->buckets[bucket_id++];
 761                         rcu_read_unlock();
 762                         skip_elems = 0;
 763                         break;
 764                 }
 765                 sk_storage = rcu_dereference(selem->local_storage);
 766                 if (sk_storage) {
 767                         info->skip_elems = skip_elems + count;
 768                         return selem;
 769                 }
 770                 count++;
 771         }
 772
 773         for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
 774                 b = &smap->buckets[i];
 775                 rcu_read_lock();
 776                 count = 0;
 777                 hlist_for_each_entry_rcu(selem, &b->list, map_node) {
 778                         sk_storage = rcu_dereference(selem->local_storage);
 779                         if (sk_storage && count >= skip_elems) {
 780                                 info->bucket_id = i;
 781                                 info->skip_elems = count;
 782                                 return selem;
 783                         }
 784                         count++;
 785                 }
 786                 rcu_read_unlock();
 787                 skip_elems = 0;
 788         }
 789
 790         info->bucket_id = i;
 791         info->skip_elems = 0;
 792         return NULL;
 793 }
 794
 795 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
 796 {
 797         struct bpf_local_storage_elem *selem;
 798
 799         selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
 800         if (!selem)
 801                 return NULL;
 802
 803         if (*pos == 0)
 804                 ++*pos;
 805         return selem;
 806 }
 807
 808 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
 809                                          loff_t *pos)
 810 {
 811         struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
 812
 813         ++*pos;
 814         ++info->skip_elems;
 815         return bpf_sk_storage_map_seq_find_next(seq->private, v);
 816 }
 817
 818 struct bpf_iter__bpf_sk_storage_map {
 819         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 820         __bpf_md_ptr(struct bpf_map *, map);
 821         __bpf_md_ptr(struct sock *, sk);
 822         __bpf_md_ptr(void *, value);
 823 };
 824
 825 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
 826                      struct bpf_map *map, struct sock *sk,
 827                      void *value)
 828
 829 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
 830                                          struct bpf_local_storage_elem *selem)
 831 {
 832         struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
 833         struct bpf_iter__bpf_sk_storage_map ctx = {};
 834         struct bpf_local_storage *sk_storage;
 835         struct bpf_iter_meta meta;
 836         struct bpf_prog *prog;
 837         int ret = 0;
 838
 839         meta.seq = seq;
 840         prog = bpf_iter_get_info(&meta, selem == NULL);
 841         if (prog) {
 842                 ctx.meta = &meta;
 843                 ctx.map = info->map;
 844                 if (selem) {
 845                         sk_storage = rcu_dereference(selem->local_storage);
 846                         ctx.sk = sk_storage->owner;
 847                         ctx.value = SDATA(selem)->data;
 848                 }
 849                 ret = bpf_iter_run_prog(prog, &ctx);
 850         }
 851
 852         return ret;
 853 }
 854
 855 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
 856 {
 857         return __bpf_sk_storage_map_seq_show(seq, v);
 858 }
 859
 860 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
 861         __releases(RCU)
 862 {
 863         if (!v)
 864                 (void)__bpf_sk_storage_map_seq_show(seq, v);
 865         else
 866                 rcu_read_unlock();
 867 }
 868
 869 static int bpf_iter_init_sk_storage_map(void *priv_data,
 870                                         struct bpf_iter_aux_info *aux)
 871 {
 872         struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
 873
 874         seq_info->map = aux->map;
 875         return 0;
 876 }
 877
 878 static int bpf_iter_attach_map(struct bpf_prog *prog,
 879                                union bpf_iter_link_info *linfo,
 880                                struct bpf_iter_aux_info *aux)
 881 {
 882         struct bpf_map *map;
 883         int err = -EINVAL;
 884
 885         if (!linfo->map.map_fd)
 886                 return -EBADF;
 887
 888         map = bpf_map_get_with_uref(linfo->map.map_fd);
 889         if (IS_ERR(map))
 890                 return PTR_ERR(map);
 891
 892         if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 893                 goto put_map;
 894
 895         if (prog->aux->max_rdonly_access > map->value_size) {
 896                 err = -EACCES;
 897                 goto put_map;
 898         }
 899
 900         aux->map = map;
 901         return 0;
 902
 903 put_map:
 904         bpf_map_put_with_uref(map);
 905         return err;
 906 }
 907
 908 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
 909 {
 910         bpf_map_put_with_uref(aux->map);
 911 }
 912
 913 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
 914         .start  = bpf_sk_storage_map_seq_start,
 915         .next   = bpf_sk_storage_map_seq_next,
 916         .stop   = bpf_sk_storage_map_seq_stop,
 917         .show   = bpf_sk_storage_map_seq_show,
 918 };
 919
 920 static const struct bpf_iter_seq_info iter_seq_info = {
 921         .seq_ops                = &bpf_sk_storage_map_seq_ops,
 922         .init_seq_private       = bpf_iter_init_sk_storage_map,
 923         .fini_seq_private       = NULL,
 924         .seq_priv_size          = sizeof(struct bpf_iter_seq_sk_storage_map_info),
 925 };
 926
 927 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
 928         .target                 = "bpf_sk_storage_map",
 929         .attach_target          = bpf_iter_attach_map,
 930         .detach_target          = bpf_iter_detach_map,
 931         .show_fdinfo            = bpf_iter_map_show_fdinfo,
 932         .fill_link_info         = bpf_iter_map_fill_link_info,
 933         .ctx_arg_info_size      = 2,
 934         .ctx_arg_info           = {
 935                 { offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
 936                   PTR_TO_BTF_ID_OR_NULL },
 937                 { offsetof(struct bpf_iter__bpf_sk_storage_map, value),
 938                   PTR_TO_BUF | PTR_MAYBE_NULL },
 939         },
 940         .seq_info               = &iter_seq_info,
 941 };
 942
 943 static int __init bpf_sk_storage_map_iter_init(void)
 944 {
 945         bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
 946                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
 947         return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
 948 }
 949 late_initcall(bpf_sk_storage_map_iter_init);