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), false);
  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 *sk_storage;
  52
  53         rcu_read_lock();
  54         sk_storage = rcu_dereference(sk->sk_bpf_storage);
  55         if (!sk_storage) {
  56                 rcu_read_unlock();
  57                 return;
  58         }
  59
  60         bpf_local_storage_destroy(sk_storage);
  61         rcu_read_unlock();
  62 }
  63
  64 static void bpf_sk_storage_map_free(struct bpf_map *map)
  65 {
  66         bpf_local_storage_map_free(map, &sk_cache, NULL);
  67 }
  68
  69 static struct bpf_map *bpf_sk_storage_map_alloc(union bpf_attr *attr)
  70 {
  71         return bpf_local_storage_map_alloc(attr, &sk_cache, false);
  72 }
  73
  74 static int notsupp_get_next_key(struct bpf_map *map, void *key,
  75                                 void *next_key)
  76 {
  77         return -ENOTSUPP;
  78 }
  79
  80 static void *bpf_fd_sk_storage_lookup_elem(struct bpf_map *map, void *key)
  81 {
  82         struct bpf_local_storage_data *sdata;
  83         struct socket *sock;
  84         int fd, err;
  85
  86         fd = *(int *)key;
  87         sock = sockfd_lookup(fd, &err);
  88         if (sock) {
  89                 sdata = bpf_sk_storage_lookup(sock->sk, map, true);
  90                 sockfd_put(sock);
  91                 return sdata ? sdata->data : NULL;
  92         }
  93
  94         return ERR_PTR(err);
  95 }
  96
  97 static long bpf_fd_sk_storage_update_elem(struct bpf_map *map, void *key,
  98                                           void *value, u64 map_flags)
  99 {
 100         struct bpf_local_storage_data *sdata;
 101         struct socket *sock;
 102         int fd, err;
 103
 104         fd = *(int *)key;
 105         sock = sockfd_lookup(fd, &err);
 106         if (sock) {
 107                 sdata = bpf_local_storage_update(
 108                         sock->sk, (struct bpf_local_storage_map *)map, value,
 109                         map_flags, GFP_ATOMIC);
 110                 sockfd_put(sock);
 111                 return PTR_ERR_OR_ZERO(sdata);
 112         }
 113
 114         return err;
 115 }
 116
 117 static long bpf_fd_sk_storage_delete_elem(struct bpf_map *map, void *key)
 118 {
 119         struct socket *sock;
 120         int fd, err;
 121
 122         fd = *(int *)key;
 123         sock = sockfd_lookup(fd, &err);
 124         if (sock) {
 125                 err = bpf_sk_storage_del(sock->sk, map);
 126                 sockfd_put(sock);
 127                 return err;
 128         }
 129
 130         return err;
 131 }
 132
 133 static struct bpf_local_storage_elem *
 134 bpf_sk_storage_clone_elem(struct sock *newsk,
 135                           struct bpf_local_storage_map *smap,
 136                           struct bpf_local_storage_elem *selem)
 137 {
 138         struct bpf_local_storage_elem *copy_selem;
 139
 140         copy_selem = bpf_selem_alloc(smap, newsk, NULL, true, GFP_ATOMIC);
 141         if (!copy_selem)
 142                 return NULL;
 143
 144         if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
 145                 copy_map_value_locked(&smap->map, SDATA(copy_selem)->data,
 146                                       SDATA(selem)->data, true);
 147         else
 148                 copy_map_value(&smap->map, SDATA(copy_selem)->data,
 149                                SDATA(selem)->data);
 150
 151         return copy_selem;
 152 }
 153
 154 int bpf_sk_storage_clone(const struct sock *sk, struct sock *newsk)
 155 {
 156         struct bpf_local_storage *new_sk_storage = NULL;
 157         struct bpf_local_storage *sk_storage;
 158         struct bpf_local_storage_elem *selem;
 159         int ret = 0;
 160
 161         RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL);
 162
 163         rcu_read_lock();
 164         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 165
 166         if (!sk_storage || hlist_empty(&sk_storage->list))
 167                 goto out;
 168
 169         hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
 170                 struct bpf_local_storage_elem *copy_selem;
 171                 struct bpf_local_storage_map *smap;
 172                 struct bpf_map *map;
 173
 174                 smap = rcu_dereference(SDATA(selem)->smap);
 175                 if (!(smap->map.map_flags & BPF_F_CLONE))
 176                         continue;
 177
 178                 /* Note that for lockless listeners adding new element
 179                  * here can race with cleanup in bpf_local_storage_map_free.
 180                  * Try to grab map refcnt to make sure that it's still
 181                  * alive and prevent concurrent removal.
 182                  */
 183                 map = bpf_map_inc_not_zero(&smap->map);
 184                 if (IS_ERR(map))
 185                         continue;
 186
 187                 copy_selem = bpf_sk_storage_clone_elem(newsk, smap, selem);
 188                 if (!copy_selem) {
 189                         ret = -ENOMEM;
 190                         bpf_map_put(map);
 191                         goto out;
 192                 }
 193
 194                 if (new_sk_storage) {
 195                         bpf_selem_link_map(smap, copy_selem);
 196                         bpf_selem_link_storage_nolock(new_sk_storage, copy_selem);
 197                 } else {
 198                         ret = bpf_local_storage_alloc(newsk, smap, copy_selem, GFP_ATOMIC);
 199                         if (ret) {
 200                                 bpf_selem_free(copy_selem, smap, true);
 201                                 atomic_sub(smap->elem_size,
 202                                            &newsk->sk_omem_alloc);
 203                                 bpf_map_put(map);
 204                                 goto out;
 205                         }
 206
 207                         new_sk_storage =
 208                                 rcu_dereference(copy_selem->local_storage);
 209                 }
 210                 bpf_map_put(map);
 211         }
 212
 213 out:
 214         rcu_read_unlock();
 215
 216         /* In case of an error, don't free anything explicitly here, the
 217          * caller is responsible to call bpf_sk_storage_free.
 218          */
 219
 220         return ret;
 221 }
 222
 223 /* *gfp_flags* is a hidden argument provided by the verifier */
 224 BPF_CALL_5(bpf_sk_storage_get, struct bpf_map *, map, struct sock *, sk,
 225            void *, value, u64, flags, gfp_t, gfp_flags)
 226 {
 227         struct bpf_local_storage_data *sdata;
 228
 229         WARN_ON_ONCE(!bpf_rcu_lock_held());
 230         if (!sk || !sk_fullsock(sk) || flags > BPF_SK_STORAGE_GET_F_CREATE)
 231                 return (unsigned long)NULL;
 232
 233         sdata = bpf_sk_storage_lookup(sk, map, true);
 234         if (sdata)
 235                 return (unsigned long)sdata->data;
 236
 237         if (flags == BPF_SK_STORAGE_GET_F_CREATE &&
 238             /* Cannot add new elem to a going away sk.
 239              * Otherwise, the new elem may become a leak
 240              * (and also other memory issues during map
 241              *  destruction).
 242              */
 243             refcount_inc_not_zero(&sk->sk_refcnt)) {
 244                 sdata = bpf_local_storage_update(
 245                         sk, (struct bpf_local_storage_map *)map, value,
 246                         BPF_NOEXIST, gfp_flags);
 247                 /* sk must be a fullsock (guaranteed by verifier),
 248                  * so sock_gen_put() is unnecessary.
 249                  */
 250                 sock_put(sk);
 251                 return IS_ERR(sdata) ?
 252                         (unsigned long)NULL : (unsigned long)sdata->data;
 253         }
 254
 255         return (unsigned long)NULL;
 256 }
 257
 258 BPF_CALL_2(bpf_sk_storage_delete, struct bpf_map *, map, struct sock *, sk)
 259 {
 260         WARN_ON_ONCE(!bpf_rcu_lock_held());
 261         if (!sk || !sk_fullsock(sk))
 262                 return -EINVAL;
 263
 264         if (refcount_inc_not_zero(&sk->sk_refcnt)) {
 265                 int err;
 266
 267                 err = bpf_sk_storage_del(sk, map);
 268                 sock_put(sk);
 269                 return err;
 270         }
 271
 272         return -ENOENT;
 273 }
 274
 275 static int bpf_sk_storage_charge(struct bpf_local_storage_map *smap,
 276                                  void *owner, u32 size)
 277 {
 278         int optmem_max = READ_ONCE(sysctl_optmem_max);
 279         struct sock *sk = (struct sock *)owner;
 280
 281         /* same check as in sock_kmalloc() */
 282         if (size <= optmem_max &&
 283             atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
 284                 atomic_add(size, &sk->sk_omem_alloc);
 285                 return 0;
 286         }
 287
 288         return -ENOMEM;
 289 }
 290
 291 static void bpf_sk_storage_uncharge(struct bpf_local_storage_map *smap,
 292                                     void *owner, u32 size)
 293 {
 294         struct sock *sk = owner;
 295
 296         atomic_sub(size, &sk->sk_omem_alloc);
 297 }
 298
 299 static struct bpf_local_storage __rcu **
 300 bpf_sk_storage_ptr(void *owner)
 301 {
 302         struct sock *sk = owner;
 303
 304         return &sk->sk_bpf_storage;
 305 }
 306
 307 const struct bpf_map_ops sk_storage_map_ops = {
 308         .map_meta_equal = bpf_map_meta_equal,
 309         .map_alloc_check = bpf_local_storage_map_alloc_check,
 310         .map_alloc = bpf_sk_storage_map_alloc,
 311         .map_free = bpf_sk_storage_map_free,
 312         .map_get_next_key = notsupp_get_next_key,
 313         .map_lookup_elem = bpf_fd_sk_storage_lookup_elem,
 314         .map_update_elem = bpf_fd_sk_storage_update_elem,
 315         .map_delete_elem = bpf_fd_sk_storage_delete_elem,
 316         .map_check_btf = bpf_local_storage_map_check_btf,
 317         .map_btf_id = &bpf_local_storage_map_btf_id[0],
 318         .map_local_storage_charge = bpf_sk_storage_charge,
 319         .map_local_storage_uncharge = bpf_sk_storage_uncharge,
 320         .map_owner_storage_ptr = bpf_sk_storage_ptr,
 321         .map_mem_usage = bpf_local_storage_map_mem_usage,
 322 };
 323
 324 const struct bpf_func_proto bpf_sk_storage_get_proto = {
 325         .func           = bpf_sk_storage_get,
 326         .gpl_only       = false,
 327         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 328         .arg1_type      = ARG_CONST_MAP_PTR,
 329         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
 330         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 331         .arg4_type      = ARG_ANYTHING,
 332 };
 333
 334 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
 335         .func           = bpf_sk_storage_get,
 336         .gpl_only       = false,
 337         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 338         .arg1_type      = ARG_CONST_MAP_PTR,
 339         .arg2_type      = ARG_PTR_TO_CTX, /* context is 'struct sock' */
 340         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 341         .arg4_type      = ARG_ANYTHING,
 342 };
 343
 344 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
 345         .func           = bpf_sk_storage_delete,
 346         .gpl_only       = false,
 347         .ret_type       = RET_INTEGER,
 348         .arg1_type      = ARG_CONST_MAP_PTR,
 349         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
 350 };
 351
 352 static bool bpf_sk_storage_tracing_allowed(const struct bpf_prog *prog)
 353 {
 354         const struct btf *btf_vmlinux;
 355         const struct btf_type *t;
 356         const char *tname;
 357         u32 btf_id;
 358
 359         if (prog->aux->dst_prog)
 360                 return false;
 361
 362         /* Ensure the tracing program is not tracing
 363          * any bpf_sk_storage*() function and also
 364          * use the bpf_sk_storage_(get|delete) helper.
 365          */
 366         switch (prog->expected_attach_type) {
 367         case BPF_TRACE_ITER:
 368         case BPF_TRACE_RAW_TP:
 369                 /* bpf_sk_storage has no trace point */
 370                 return true;
 371         case BPF_TRACE_FENTRY:
 372         case BPF_TRACE_FEXIT:
 373                 btf_vmlinux = bpf_get_btf_vmlinux();
 374                 if (IS_ERR_OR_NULL(btf_vmlinux))
 375                         return false;
 376                 btf_id = prog->aux->attach_btf_id;
 377                 t = btf_type_by_id(btf_vmlinux, btf_id);
 378                 tname = btf_name_by_offset(btf_vmlinux, t->name_off);
 379                 return !!strncmp(tname, "bpf_sk_storage",
 380                                  strlen("bpf_sk_storage"));
 381         default:
 382                 return false;
 383         }
 384
 385         return false;
 386 }
 387
 388 /* *gfp_flags* is a hidden argument provided by the verifier */
 389 BPF_CALL_5(bpf_sk_storage_get_tracing, struct bpf_map *, map, struct sock *, sk,
 390            void *, value, u64, flags, gfp_t, gfp_flags)
 391 {
 392         WARN_ON_ONCE(!bpf_rcu_lock_held());
 393         if (in_hardirq() || in_nmi())
 394                 return (unsigned long)NULL;
 395
 396         return (unsigned long)____bpf_sk_storage_get(map, sk, value, flags,
 397                                                      gfp_flags);
 398 }
 399
 400 BPF_CALL_2(bpf_sk_storage_delete_tracing, struct bpf_map *, map,
 401            struct sock *, sk)
 402 {
 403         WARN_ON_ONCE(!bpf_rcu_lock_held());
 404         if (in_hardirq() || in_nmi())
 405                 return -EPERM;
 406
 407         return ____bpf_sk_storage_delete(map, sk);
 408 }
 409
 410 const struct bpf_func_proto bpf_sk_storage_get_tracing_proto = {
 411         .func           = bpf_sk_storage_get_tracing,
 412         .gpl_only       = false,
 413         .ret_type       = RET_PTR_TO_MAP_VALUE_OR_NULL,
 414         .arg1_type      = ARG_CONST_MAP_PTR,
 415         .arg2_type      = ARG_PTR_TO_BTF_ID_OR_NULL,
 416         .arg2_btf_id    = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
 417         .arg3_type      = ARG_PTR_TO_MAP_VALUE_OR_NULL,
 418         .arg4_type      = ARG_ANYTHING,
 419         .allowed        = bpf_sk_storage_tracing_allowed,
 420 };
 421
 422 const struct bpf_func_proto bpf_sk_storage_delete_tracing_proto = {
 423         .func           = bpf_sk_storage_delete_tracing,
 424         .gpl_only       = false,
 425         .ret_type       = RET_INTEGER,
 426         .arg1_type      = ARG_CONST_MAP_PTR,
 427         .arg2_type      = ARG_PTR_TO_BTF_ID_OR_NULL,
 428         .arg2_btf_id    = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
 429         .allowed        = bpf_sk_storage_tracing_allowed,
 430 };
 431
 432 struct bpf_sk_storage_diag {
 433         u32 nr_maps;
 434         struct bpf_map *maps[];
 435 };
 436
 437 /* The reply will be like:
 438  * INET_DIAG_BPF_SK_STORAGES (nla_nest)
 439  *      SK_DIAG_BPF_STORAGE (nla_nest)
 440  *              SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 441  *              SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 442  *      SK_DIAG_BPF_STORAGE (nla_nest)
 443  *              SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 444  *              SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 445  *      ....
 446  */
 447 static int nla_value_size(u32 value_size)
 448 {
 449         /* SK_DIAG_BPF_STORAGE (nla_nest)
 450          *      SK_DIAG_BPF_STORAGE_MAP_ID (nla_put_u32)
 451          *      SK_DIAG_BPF_STORAGE_MAP_VALUE (nla_reserve_64bit)
 452          */
 453         return nla_total_size(0) + nla_total_size(sizeof(u32)) +
 454                 nla_total_size_64bit(value_size);
 455 }
 456
 457 void bpf_sk_storage_diag_free(struct bpf_sk_storage_diag *diag)
 458 {
 459         u32 i;
 460
 461         if (!diag)
 462                 return;
 463
 464         for (i = 0; i < diag->nr_maps; i++)
 465                 bpf_map_put(diag->maps[i]);
 466
 467         kfree(diag);
 468 }
 469 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_free);
 470
 471 static bool diag_check_dup(const struct bpf_sk_storage_diag *diag,
 472                            const struct bpf_map *map)
 473 {
 474         u32 i;
 475
 476         for (i = 0; i < diag->nr_maps; i++) {
 477                 if (diag->maps[i] == map)
 478                         return true;
 479         }
 480
 481         return false;
 482 }
 483
 484 struct bpf_sk_storage_diag *
 485 bpf_sk_storage_diag_alloc(const struct nlattr *nla_stgs)
 486 {
 487         struct bpf_sk_storage_diag *diag;
 488         struct nlattr *nla;
 489         u32 nr_maps = 0;
 490         int rem, err;
 491
 492         /* bpf_local_storage_map is currently limited to CAP_SYS_ADMIN as
 493          * the map_alloc_check() side also does.
 494          */
 495         if (!bpf_capable())
 496                 return ERR_PTR(-EPERM);
 497
 498         nla_for_each_nested(nla, nla_stgs, rem) {
 499                 if (nla_type(nla) == SK_DIAG_BPF_STORAGE_REQ_MAP_FD) {
 500                         if (nla_len(nla) != sizeof(u32))
 501                                 return ERR_PTR(-EINVAL);
 502                         nr_maps++;
 503                 }
 504         }
 505
 506         diag = kzalloc(struct_size(diag, maps, nr_maps), GFP_KERNEL);
 507         if (!diag)
 508                 return ERR_PTR(-ENOMEM);
 509
 510         nla_for_each_nested(nla, nla_stgs, rem) {
 511                 struct bpf_map *map;
 512                 int map_fd;
 513
 514                 if (nla_type(nla) != SK_DIAG_BPF_STORAGE_REQ_MAP_FD)
 515                         continue;
 516
 517                 map_fd = nla_get_u32(nla);
 518                 map = bpf_map_get(map_fd);
 519                 if (IS_ERR(map)) {
 520                         err = PTR_ERR(map);
 521                         goto err_free;
 522                 }
 523                 if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) {
 524                         bpf_map_put(map);
 525                         err = -EINVAL;
 526                         goto err_free;
 527                 }
 528                 if (diag_check_dup(diag, map)) {
 529                         bpf_map_put(map);
 530                         err = -EEXIST;
 531                         goto err_free;
 532                 }
 533                 diag->maps[diag->nr_maps++] = map;
 534         }
 535
 536         return diag;
 537
 538 err_free:
 539         bpf_sk_storage_diag_free(diag);
 540         return ERR_PTR(err);
 541 }
 542 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_alloc);
 543
 544 static int diag_get(struct bpf_local_storage_data *sdata, struct sk_buff *skb)
 545 {
 546         struct nlattr *nla_stg, *nla_value;
 547         struct bpf_local_storage_map *smap;
 548
 549         /* It cannot exceed max nlattr's payload */
 550         BUILD_BUG_ON(U16_MAX - NLA_HDRLEN < BPF_LOCAL_STORAGE_MAX_VALUE_SIZE);
 551
 552         nla_stg = nla_nest_start(skb, SK_DIAG_BPF_STORAGE);
 553         if (!nla_stg)
 554                 return -EMSGSIZE;
 555
 556         smap = rcu_dereference(sdata->smap);
 557         if (nla_put_u32(skb, SK_DIAG_BPF_STORAGE_MAP_ID, smap->map.id))
 558                 goto errout;
 559
 560         nla_value = nla_reserve_64bit(skb, SK_DIAG_BPF_STORAGE_MAP_VALUE,
 561                                       smap->map.value_size,
 562                                       SK_DIAG_BPF_STORAGE_PAD);
 563         if (!nla_value)
 564                 goto errout;
 565
 566         if (btf_record_has_field(smap->map.record, BPF_SPIN_LOCK))
 567                 copy_map_value_locked(&smap->map, nla_data(nla_value),
 568                                       sdata->data, true);
 569         else
 570                 copy_map_value(&smap->map, nla_data(nla_value), sdata->data);
 571
 572         nla_nest_end(skb, nla_stg);
 573         return 0;
 574
 575 errout:
 576         nla_nest_cancel(skb, nla_stg);
 577         return -EMSGSIZE;
 578 }
 579
 580 static int bpf_sk_storage_diag_put_all(struct sock *sk, struct sk_buff *skb,
 581                                        int stg_array_type,
 582                                        unsigned int *res_diag_size)
 583 {
 584         /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
 585         unsigned int diag_size = nla_total_size(0);
 586         struct bpf_local_storage *sk_storage;
 587         struct bpf_local_storage_elem *selem;
 588         struct bpf_local_storage_map *smap;
 589         struct nlattr *nla_stgs;
 590         unsigned int saved_len;
 591         int err = 0;
 592
 593         rcu_read_lock();
 594
 595         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 596         if (!sk_storage || hlist_empty(&sk_storage->list)) {
 597                 rcu_read_unlock();
 598                 return 0;
 599         }
 600
 601         nla_stgs = nla_nest_start(skb, stg_array_type);
 602         if (!nla_stgs)
 603                 /* Continue to learn diag_size */
 604                 err = -EMSGSIZE;
 605
 606         saved_len = skb->len;
 607         hlist_for_each_entry_rcu(selem, &sk_storage->list, snode) {
 608                 smap = rcu_dereference(SDATA(selem)->smap);
 609                 diag_size += nla_value_size(smap->map.value_size);
 610
 611                 if (nla_stgs && diag_get(SDATA(selem), skb))
 612                         /* Continue to learn diag_size */
 613                         err = -EMSGSIZE;
 614         }
 615
 616         rcu_read_unlock();
 617
 618         if (nla_stgs) {
 619                 if (saved_len == skb->len)
 620                         nla_nest_cancel(skb, nla_stgs);
 621                 else
 622                         nla_nest_end(skb, nla_stgs);
 623         }
 624
 625         if (diag_size == nla_total_size(0)) {
 626                 *res_diag_size = 0;
 627                 return 0;
 628         }
 629
 630         *res_diag_size = diag_size;
 631         return err;
 632 }
 633
 634 int bpf_sk_storage_diag_put(struct bpf_sk_storage_diag *diag,
 635                             struct sock *sk, struct sk_buff *skb,
 636                             int stg_array_type,
 637                             unsigned int *res_diag_size)
 638 {
 639         /* stg_array_type (e.g. INET_DIAG_BPF_SK_STORAGES) */
 640         unsigned int diag_size = nla_total_size(0);
 641         struct bpf_local_storage *sk_storage;
 642         struct bpf_local_storage_data *sdata;
 643         struct nlattr *nla_stgs;
 644         unsigned int saved_len;
 645         int err = 0;
 646         u32 i;
 647
 648         *res_diag_size = 0;
 649
 650         /* No map has been specified.  Dump all. */
 651         if (!diag->nr_maps)
 652                 return bpf_sk_storage_diag_put_all(sk, skb, stg_array_type,
 653                                                    res_diag_size);
 654
 655         rcu_read_lock();
 656         sk_storage = rcu_dereference(sk->sk_bpf_storage);
 657         if (!sk_storage || hlist_empty(&sk_storage->list)) {
 658                 rcu_read_unlock();
 659                 return 0;
 660         }
 661
 662         nla_stgs = nla_nest_start(skb, stg_array_type);
 663         if (!nla_stgs)
 664                 /* Continue to learn diag_size */
 665                 err = -EMSGSIZE;
 666
 667         saved_len = skb->len;
 668         for (i = 0; i < diag->nr_maps; i++) {
 669                 sdata = bpf_local_storage_lookup(sk_storage,
 670                                 (struct bpf_local_storage_map *)diag->maps[i],
 671                                 false);
 672
 673                 if (!sdata)
 674                         continue;
 675
 676                 diag_size += nla_value_size(diag->maps[i]->value_size);
 677
 678                 if (nla_stgs && diag_get(sdata, skb))
 679                         /* Continue to learn diag_size */
 680                         err = -EMSGSIZE;
 681         }
 682         rcu_read_unlock();
 683
 684         if (nla_stgs) {
 685                 if (saved_len == skb->len)
 686                         nla_nest_cancel(skb, nla_stgs);
 687                 else
 688                         nla_nest_end(skb, nla_stgs);
 689         }
 690
 691         if (diag_size == nla_total_size(0)) {
 692                 *res_diag_size = 0;
 693                 return 0;
 694         }
 695
 696         *res_diag_size = diag_size;
 697         return err;
 698 }
 699 EXPORT_SYMBOL_GPL(bpf_sk_storage_diag_put);
 700
 701 struct bpf_iter_seq_sk_storage_map_info {
 702         struct bpf_map *map;
 703         unsigned int bucket_id;
 704         unsigned skip_elems;
 705 };
 706
 707 static struct bpf_local_storage_elem *
 708 bpf_sk_storage_map_seq_find_next(struct bpf_iter_seq_sk_storage_map_info *info,
 709                                  struct bpf_local_storage_elem *prev_selem)
 710         __acquires(RCU) __releases(RCU)
 711 {
 712         struct bpf_local_storage *sk_storage;
 713         struct bpf_local_storage_elem *selem;
 714         u32 skip_elems = info->skip_elems;
 715         struct bpf_local_storage_map *smap;
 716         u32 bucket_id = info->bucket_id;
 717         u32 i, count, n_buckets;
 718         struct bpf_local_storage_map_bucket *b;
 719
 720         smap = (struct bpf_local_storage_map *)info->map;
 721         n_buckets = 1U << smap->bucket_log;
 722         if (bucket_id >= n_buckets)
 723                 return NULL;
 724
 725         /* try to find next selem in the same bucket */
 726         selem = prev_selem;
 727         count = 0;
 728         while (selem) {
 729                 selem = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&selem->map_node)),
 730                                          struct bpf_local_storage_elem, map_node);
 731                 if (!selem) {
 732                         /* not found, unlock and go to the next bucket */
 733                         b = &smap->buckets[bucket_id++];
 734                         rcu_read_unlock();
 735                         skip_elems = 0;
 736                         break;
 737                 }
 738                 sk_storage = rcu_dereference(selem->local_storage);
 739                 if (sk_storage) {
 740                         info->skip_elems = skip_elems + count;
 741                         return selem;
 742                 }
 743                 count++;
 744         }
 745
 746         for (i = bucket_id; i < (1U << smap->bucket_log); i++) {
 747                 b = &smap->buckets[i];
 748                 rcu_read_lock();
 749                 count = 0;
 750                 hlist_for_each_entry_rcu(selem, &b->list, map_node) {
 751                         sk_storage = rcu_dereference(selem->local_storage);
 752                         if (sk_storage && count >= skip_elems) {
 753                                 info->bucket_id = i;
 754                                 info->skip_elems = count;
 755                                 return selem;
 756                         }
 757                         count++;
 758                 }
 759                 rcu_read_unlock();
 760                 skip_elems = 0;
 761         }
 762
 763         info->bucket_id = i;
 764         info->skip_elems = 0;
 765         return NULL;
 766 }
 767
 768 static void *bpf_sk_storage_map_seq_start(struct seq_file *seq, loff_t *pos)
 769 {
 770         struct bpf_local_storage_elem *selem;
 771
 772         selem = bpf_sk_storage_map_seq_find_next(seq->private, NULL);
 773         if (!selem)
 774                 return NULL;
 775
 776         if (*pos == 0)
 777                 ++*pos;
 778         return selem;
 779 }
 780
 781 static void *bpf_sk_storage_map_seq_next(struct seq_file *seq, void *v,
 782                                          loff_t *pos)
 783 {
 784         struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
 785
 786         ++*pos;
 787         ++info->skip_elems;
 788         return bpf_sk_storage_map_seq_find_next(seq->private, v);
 789 }
 790
 791 struct bpf_iter__bpf_sk_storage_map {
 792         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 793         __bpf_md_ptr(struct bpf_map *, map);
 794         __bpf_md_ptr(struct sock *, sk);
 795         __bpf_md_ptr(void *, value);
 796 };
 797
 798 DEFINE_BPF_ITER_FUNC(bpf_sk_storage_map, struct bpf_iter_meta *meta,
 799                      struct bpf_map *map, struct sock *sk,
 800                      void *value)
 801
 802 static int __bpf_sk_storage_map_seq_show(struct seq_file *seq,
 803                                          struct bpf_local_storage_elem *selem)
 804 {
 805         struct bpf_iter_seq_sk_storage_map_info *info = seq->private;
 806         struct bpf_iter__bpf_sk_storage_map ctx = {};
 807         struct bpf_local_storage *sk_storage;
 808         struct bpf_iter_meta meta;
 809         struct bpf_prog *prog;
 810         int ret = 0;
 811
 812         meta.seq = seq;
 813         prog = bpf_iter_get_info(&meta, selem == NULL);
 814         if (prog) {
 815                 ctx.meta = &meta;
 816                 ctx.map = info->map;
 817                 if (selem) {
 818                         sk_storage = rcu_dereference(selem->local_storage);
 819                         ctx.sk = sk_storage->owner;
 820                         ctx.value = SDATA(selem)->data;
 821                 }
 822                 ret = bpf_iter_run_prog(prog, &ctx);
 823         }
 824
 825         return ret;
 826 }
 827
 828 static int bpf_sk_storage_map_seq_show(struct seq_file *seq, void *v)
 829 {
 830         return __bpf_sk_storage_map_seq_show(seq, v);
 831 }
 832
 833 static void bpf_sk_storage_map_seq_stop(struct seq_file *seq, void *v)
 834         __releases(RCU)
 835 {
 836         if (!v)
 837                 (void)__bpf_sk_storage_map_seq_show(seq, v);
 838         else
 839                 rcu_read_unlock();
 840 }
 841
 842 static int bpf_iter_init_sk_storage_map(void *priv_data,
 843                                         struct bpf_iter_aux_info *aux)
 844 {
 845         struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
 846
 847         bpf_map_inc_with_uref(aux->map);
 848         seq_info->map = aux->map;
 849         return 0;
 850 }
 851
 852 static void bpf_iter_fini_sk_storage_map(void *priv_data)
 853 {
 854         struct bpf_iter_seq_sk_storage_map_info *seq_info = priv_data;
 855
 856         bpf_map_put_with_uref(seq_info->map);
 857 }
 858
 859 static int bpf_iter_attach_map(struct bpf_prog *prog,
 860                                union bpf_iter_link_info *linfo,
 861                                struct bpf_iter_aux_info *aux)
 862 {
 863         struct bpf_map *map;
 864         int err = -EINVAL;
 865
 866         if (!linfo->map.map_fd)
 867                 return -EBADF;
 868
 869         map = bpf_map_get_with_uref(linfo->map.map_fd);
 870         if (IS_ERR(map))
 871                 return PTR_ERR(map);
 872
 873         if (map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 874                 goto put_map;
 875
 876         if (prog->aux->max_rdwr_access > map->value_size) {
 877                 err = -EACCES;
 878                 goto put_map;
 879         }
 880
 881         aux->map = map;
 882         return 0;
 883
 884 put_map:
 885         bpf_map_put_with_uref(map);
 886         return err;
 887 }
 888
 889 static void bpf_iter_detach_map(struct bpf_iter_aux_info *aux)
 890 {
 891         bpf_map_put_with_uref(aux->map);
 892 }
 893
 894 static const struct seq_operations bpf_sk_storage_map_seq_ops = {
 895         .start  = bpf_sk_storage_map_seq_start,
 896         .next   = bpf_sk_storage_map_seq_next,
 897         .stop   = bpf_sk_storage_map_seq_stop,
 898         .show   = bpf_sk_storage_map_seq_show,
 899 };
 900
 901 static const struct bpf_iter_seq_info iter_seq_info = {
 902         .seq_ops                = &bpf_sk_storage_map_seq_ops,
 903         .init_seq_private       = bpf_iter_init_sk_storage_map,
 904         .fini_seq_private       = bpf_iter_fini_sk_storage_map,
 905         .seq_priv_size          = sizeof(struct bpf_iter_seq_sk_storage_map_info),
 906 };
 907
 908 static struct bpf_iter_reg bpf_sk_storage_map_reg_info = {
 909         .target                 = "bpf_sk_storage_map",
 910         .attach_target          = bpf_iter_attach_map,
 911         .detach_target          = bpf_iter_detach_map,
 912         .show_fdinfo            = bpf_iter_map_show_fdinfo,
 913         .fill_link_info         = bpf_iter_map_fill_link_info,
 914         .ctx_arg_info_size      = 2,
 915         .ctx_arg_info           = {
 916                 { offsetof(struct bpf_iter__bpf_sk_storage_map, sk),
 917                   PTR_TO_BTF_ID_OR_NULL },
 918                 { offsetof(struct bpf_iter__bpf_sk_storage_map, value),
 919                   PTR_TO_BUF | PTR_MAYBE_NULL },
 920         },
 921         .seq_info               = &iter_seq_info,
 922 };
 923
 924 static int __init bpf_sk_storage_map_iter_init(void)
 925 {
 926         bpf_sk_storage_map_reg_info.ctx_arg_info[0].btf_id =
 927                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
 928         return bpf_iter_reg_target(&bpf_sk_storage_map_reg_info);
 929 }
 930 late_initcall(bpf_sk_storage_map_iter_init);