kernel/bpf/arraymap.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016,2017 Facebook
   4  */
   5 #include <linux/bpf.h>
   6 #include <linux/btf.h>
   7 #include <linux/err.h>
   8 #include <linux/slab.h>
   9 #include <linux/mm.h>
  10 #include <linux/filter.h>
  11 #include <linux/perf_event.h>
  12 #include <uapi/linux/btf.h>
  13
  14 #include "map_in_map.h"
  15
  16 #define ARRAY_CREATE_FLAG_MASK \
  17         (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK)
  18
  19 static void bpf_array_free_percpu(struct bpf_array *array)
  20 {
  21         int i;
  22
  23         for (i = 0; i < array->map.max_entries; i++) {
  24                 free_percpu(array->pptrs[i]);
  25                 cond_resched();
  26         }
  27 }
  28
  29 static int bpf_array_alloc_percpu(struct bpf_array *array)
  30 {
  31         void __percpu *ptr;
  32         int i;
  33
  34         for (i = 0; i < array->map.max_entries; i++) {
  35                 ptr = __alloc_percpu_gfp(array->elem_size, 8,
  36                                          GFP_USER | __GFP_NOWARN);
  37                 if (!ptr) {
  38                         bpf_array_free_percpu(array);
  39                         return -ENOMEM;
  40                 }
  41                 array->pptrs[i] = ptr;
  42                 cond_resched();
  43         }
  44
  45         return 0;
  46 }
  47
  48 /* Called from syscall */
  49 int array_map_alloc_check(union bpf_attr *attr)
  50 {
  51         bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  52         int numa_node = bpf_map_attr_numa_node(attr);
  53
  54         /* check sanity of attributes */
  55         if (attr->max_entries == 0 || attr->key_size != 4 ||
  56             attr->value_size == 0 ||
  57             attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
  58             !bpf_map_flags_access_ok(attr->map_flags) ||
  59             (percpu && numa_node != NUMA_NO_NODE))
  60                 return -EINVAL;
  61
  62         if (attr->value_size > KMALLOC_MAX_SIZE)
  63                 /* if value_size is bigger, the user space won't be able to
  64                  * access the elements.
  65                  */
  66                 return -E2BIG;
  67
  68         return 0;
  69 }
  70
  71 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
  72 {
  73         bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  74         int ret, numa_node = bpf_map_attr_numa_node(attr);
  75         u32 elem_size, index_mask, max_entries;
  76         bool unpriv = !capable(CAP_SYS_ADMIN);
  77         u64 cost, array_size, mask64;
  78         struct bpf_array *array;
  79
  80         elem_size = round_up(attr->value_size, 8);
  81
  82         max_entries = attr->max_entries;
  83
  84         /* On 32 bit archs roundup_pow_of_two() with max_entries that has
  85          * upper most bit set in u32 space is undefined behavior due to
  86          * resulting 1U << 32, so do it manually here in u64 space.
  87          */
  88         mask64 = fls_long(max_entries - 1);
  89         mask64 = 1ULL << mask64;
  90         mask64 -= 1;
  91
  92         index_mask = mask64;
  93         if (unpriv) {
  94                 /* round up array size to nearest power of 2,
  95                  * since cpu will speculate within index_mask limits
  96                  */
  97                 max_entries = index_mask + 1;
  98                 /* Check for overflows. */
  99                 if (max_entries < attr->max_entries)
 100                         return ERR_PTR(-E2BIG);
 101         }
 102
 103         array_size = sizeof(*array);
 104         if (percpu)
 105                 array_size += (u64) max_entries * sizeof(void *);
 106         else
 107                 array_size += (u64) max_entries * elem_size;
 108
 109         /* make sure there is no u32 overflow later in round_up() */
 110         cost = array_size;
 111         if (cost >= U32_MAX - PAGE_SIZE)
 112                 return ERR_PTR(-ENOMEM);
 113         if (percpu) {
 114                 cost += (u64)attr->max_entries * elem_size * num_possible_cpus();
 115                 if (cost >= U32_MAX - PAGE_SIZE)
 116                         return ERR_PTR(-ENOMEM);
 117         }
 118         cost = round_up(cost, PAGE_SIZE) >> PAGE_SHIFT;
 119
 120         ret = bpf_map_precharge_memlock(cost);
 121         if (ret < 0)
 122                 return ERR_PTR(ret);
 123
 124         /* allocate all map elements and zero-initialize them */
 125         array = bpf_map_area_alloc(array_size, numa_node);
 126         if (!array)
 127                 return ERR_PTR(-ENOMEM);
 128         array->index_mask = index_mask;
 129         array->map.unpriv_array = unpriv;
 130
 131         /* copy mandatory map attributes */
 132         bpf_map_init_from_attr(&array->map, attr);
 133         array->map.pages = cost;
 134         array->elem_size = elem_size;
 135
 136         if (percpu && bpf_array_alloc_percpu(array)) {
 137                 bpf_map_area_free(array);
 138                 return ERR_PTR(-ENOMEM);
 139         }
 140
 141         return &array->map;
 142 }
 143
 144 /* Called from syscall or from eBPF program */
 145 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
 146 {
 147         struct bpf_array *array = container_of(map, struct bpf_array, map);
 148         u32 index = *(u32 *)key;
 149
 150         if (unlikely(index >= array->map.max_entries))
 151                 return NULL;
 152
 153         return array->value + array->elem_size * (index & array->index_mask);
 154 }
 155
 156 static int array_map_direct_value_addr(const struct bpf_map *map, u64 *imm,
 157                                        u32 off)
 158 {
 159         struct bpf_array *array = container_of(map, struct bpf_array, map);
 160
 161         if (map->max_entries != 1)
 162                 return -ENOTSUPP;
 163         if (off >= map->value_size)
 164                 return -EINVAL;
 165
 166         *imm = (unsigned long)array->value;
 167         return 0;
 168 }
 169
 170 static int array_map_direct_value_meta(const struct bpf_map *map, u64 imm,
 171                                        u32 *off)
 172 {
 173         struct bpf_array *array = container_of(map, struct bpf_array, map);
 174         u64 base = (unsigned long)array->value;
 175         u64 range = array->elem_size;
 176
 177         if (map->max_entries != 1)
 178                 return -ENOTSUPP;
 179         if (imm < base || imm >= base + range)
 180                 return -ENOENT;
 181
 182         *off = imm - base;
 183         return 0;
 184 }
 185
 186 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
 187 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 188 {
 189         struct bpf_array *array = container_of(map, struct bpf_array, map);
 190         struct bpf_insn *insn = insn_buf;
 191         u32 elem_size = round_up(map->value_size, 8);
 192         const int ret = BPF_REG_0;
 193         const int map_ptr = BPF_REG_1;
 194         const int index = BPF_REG_2;
 195
 196         *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 197         *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 198         if (map->unpriv_array) {
 199                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
 200                 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
 201         } else {
 202                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
 203         }
 204
 205         if (is_power_of_2(elem_size)) {
 206                 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
 207         } else {
 208                 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
 209         }
 210         *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
 211         *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 212         *insn++ = BPF_MOV64_IMM(ret, 0);
 213         return insn - insn_buf;
 214 }
 215
 216 /* Called from eBPF program */
 217 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
 218 {
 219         struct bpf_array *array = container_of(map, struct bpf_array, map);
 220         u32 index = *(u32 *)key;
 221
 222         if (unlikely(index >= array->map.max_entries))
 223                 return NULL;
 224
 225         return this_cpu_ptr(array->pptrs[index & array->index_mask]);
 226 }
 227
 228 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
 229 {
 230         struct bpf_array *array = container_of(map, struct bpf_array, map);
 231         u32 index = *(u32 *)key;
 232         void __percpu *pptr;
 233         int cpu, off = 0;
 234         u32 size;
 235
 236         if (unlikely(index >= array->map.max_entries))
 237                 return -ENOENT;
 238
 239         /* per_cpu areas are zero-filled and bpf programs can only
 240          * access 'value_size' of them, so copying rounded areas
 241          * will not leak any kernel data
 242          */
 243         size = round_up(map->value_size, 8);
 244         rcu_read_lock();
 245         pptr = array->pptrs[index & array->index_mask];
 246         for_each_possible_cpu(cpu) {
 247                 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
 248                 off += size;
 249         }
 250         rcu_read_unlock();
 251         return 0;
 252 }
 253
 254 /* Called from syscall */
 255 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 256 {
 257         struct bpf_array *array = container_of(map, struct bpf_array, map);
 258         u32 index = key ? *(u32 *)key : U32_MAX;
 259         u32 *next = (u32 *)next_key;
 260
 261         if (index >= array->map.max_entries) {
 262                 *next = 0;
 263                 return 0;
 264         }
 265
 266         if (index == array->map.max_entries - 1)
 267                 return -ENOENT;
 268
 269         *next = index + 1;
 270         return 0;
 271 }
 272
 273 /* Called from syscall or from eBPF program */
 274 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
 275                                  u64 map_flags)
 276 {
 277         struct bpf_array *array = container_of(map, struct bpf_array, map);
 278         u32 index = *(u32 *)key;
 279         char *val;
 280
 281         if (unlikely((map_flags & ~BPF_F_LOCK) > BPF_EXIST))
 282                 /* unknown flags */
 283                 return -EINVAL;
 284
 285         if (unlikely(index >= array->map.max_entries))
 286                 /* all elements were pre-allocated, cannot insert a new one */
 287                 return -E2BIG;
 288
 289         if (unlikely(map_flags & BPF_NOEXIST))
 290                 /* all elements already exist */
 291                 return -EEXIST;
 292
 293         if (unlikely((map_flags & BPF_F_LOCK) &&
 294                      !map_value_has_spin_lock(map)))
 295                 return -EINVAL;
 296
 297         if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 298                 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
 299                        value, map->value_size);
 300         } else {
 301                 val = array->value +
 302                         array->elem_size * (index & array->index_mask);
 303                 if (map_flags & BPF_F_LOCK)
 304                         copy_map_value_locked(map, val, value, false);
 305                 else
 306                         copy_map_value(map, val, value);
 307         }
 308         return 0;
 309 }
 310
 311 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 312                             u64 map_flags)
 313 {
 314         struct bpf_array *array = container_of(map, struct bpf_array, map);
 315         u32 index = *(u32 *)key;
 316         void __percpu *pptr;
 317         int cpu, off = 0;
 318         u32 size;
 319
 320         if (unlikely(map_flags > BPF_EXIST))
 321                 /* unknown flags */
 322                 return -EINVAL;
 323
 324         if (unlikely(index >= array->map.max_entries))
 325                 /* all elements were pre-allocated, cannot insert a new one */
 326                 return -E2BIG;
 327
 328         if (unlikely(map_flags == BPF_NOEXIST))
 329                 /* all elements already exist */
 330                 return -EEXIST;
 331
 332         /* the user space will provide round_up(value_size, 8) bytes that
 333          * will be copied into per-cpu area. bpf programs can only access
 334          * value_size of it. During lookup the same extra bytes will be
 335          * returned or zeros which were zero-filled by percpu_alloc,
 336          * so no kernel data leaks possible
 337          */
 338         size = round_up(map->value_size, 8);
 339         rcu_read_lock();
 340         pptr = array->pptrs[index & array->index_mask];
 341         for_each_possible_cpu(cpu) {
 342                 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
 343                 off += size;
 344         }
 345         rcu_read_unlock();
 346         return 0;
 347 }
 348
 349 /* Called from syscall or from eBPF program */
 350 static int array_map_delete_elem(struct bpf_map *map, void *key)
 351 {
 352         return -EINVAL;
 353 }
 354
 355 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 356 static void array_map_free(struct bpf_map *map)
 357 {
 358         struct bpf_array *array = container_of(map, struct bpf_array, map);
 359
 360         /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
 361          * so the programs (can be more than one that used this map) were
 362          * disconnected from events. Wait for outstanding programs to complete
 363          * and free the array
 364          */
 365         synchronize_rcu();
 366
 367         if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 368                 bpf_array_free_percpu(array);
 369
 370         bpf_map_area_free(array);
 371 }
 372
 373 static void array_map_seq_show_elem(struct bpf_map *map, void *key,
 374                                     struct seq_file *m)
 375 {
 376         void *value;
 377
 378         rcu_read_lock();
 379
 380         value = array_map_lookup_elem(map, key);
 381         if (!value) {
 382                 rcu_read_unlock();
 383                 return;
 384         }
 385
 386         if (map->btf_key_type_id)
 387                 seq_printf(m, "%u: ", *(u32 *)key);
 388         btf_type_seq_show(map->btf, map->btf_value_type_id, value, m);
 389         seq_puts(m, "\n");
 390
 391         rcu_read_unlock();
 392 }
 393
 394 static void percpu_array_map_seq_show_elem(struct bpf_map *map, void *key,
 395                                            struct seq_file *m)
 396 {
 397         struct bpf_array *array = container_of(map, struct bpf_array, map);
 398         u32 index = *(u32 *)key;
 399         void __percpu *pptr;
 400         int cpu;
 401
 402         rcu_read_lock();
 403
 404         seq_printf(m, "%u: {\n", *(u32 *)key);
 405         pptr = array->pptrs[index & array->index_mask];
 406         for_each_possible_cpu(cpu) {
 407                 seq_printf(m, "\tcpu%d: ", cpu);
 408                 btf_type_seq_show(map->btf, map->btf_value_type_id,
 409                                   per_cpu_ptr(pptr, cpu), m);
 410                 seq_puts(m, "\n");
 411         }
 412         seq_puts(m, "}\n");
 413
 414         rcu_read_unlock();
 415 }
 416
 417 static int array_map_check_btf(const struct bpf_map *map,
 418                                const struct btf *btf,
 419                                const struct btf_type *key_type,
 420                                const struct btf_type *value_type)
 421 {
 422         u32 int_data;
 423
 424         /* One exception for keyless BTF: .bss/.data/.rodata map */
 425         if (btf_type_is_void(key_type)) {
 426                 if (map->map_type != BPF_MAP_TYPE_ARRAY ||
 427                     map->max_entries != 1)
 428                         return -EINVAL;
 429
 430                 if (BTF_INFO_KIND(value_type->info) != BTF_KIND_DATASEC)
 431                         return -EINVAL;
 432
 433                 return 0;
 434         }
 435
 436         if (BTF_INFO_KIND(key_type->info) != BTF_KIND_INT)
 437                 return -EINVAL;
 438
 439         int_data = *(u32 *)(key_type + 1);
 440         /* bpf array can only take a u32 key. This check makes sure
 441          * that the btf matches the attr used during map_create.
 442          */
 443         if (BTF_INT_BITS(int_data) != 32 || BTF_INT_OFFSET(int_data))
 444                 return -EINVAL;
 445
 446         return 0;
 447 }
 448
 449 const struct bpf_map_ops array_map_ops = {
 450         .map_alloc_check = array_map_alloc_check,
 451         .map_alloc = array_map_alloc,
 452         .map_free = array_map_free,
 453         .map_get_next_key = array_map_get_next_key,
 454         .map_lookup_elem = array_map_lookup_elem,
 455         .map_update_elem = array_map_update_elem,
 456         .map_delete_elem = array_map_delete_elem,
 457         .map_gen_lookup = array_map_gen_lookup,
 458         .map_direct_value_addr = array_map_direct_value_addr,
 459         .map_direct_value_meta = array_map_direct_value_meta,
 460         .map_seq_show_elem = array_map_seq_show_elem,
 461         .map_check_btf = array_map_check_btf,
 462 };
 463
 464 const struct bpf_map_ops percpu_array_map_ops = {
 465         .map_alloc_check = array_map_alloc_check,
 466         .map_alloc = array_map_alloc,
 467         .map_free = array_map_free,
 468         .map_get_next_key = array_map_get_next_key,
 469         .map_lookup_elem = percpu_array_map_lookup_elem,
 470         .map_update_elem = array_map_update_elem,
 471         .map_delete_elem = array_map_delete_elem,
 472         .map_seq_show_elem = percpu_array_map_seq_show_elem,
 473         .map_check_btf = array_map_check_btf,
 474 };
 475
 476 static int fd_array_map_alloc_check(union bpf_attr *attr)
 477 {
 478         /* only file descriptors can be stored in this type of map */
 479         if (attr->value_size != sizeof(u32))
 480                 return -EINVAL;
 481         /* Program read-only/write-only not supported for special maps yet. */
 482         if (attr->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG))
 483                 return -EINVAL;
 484         return array_map_alloc_check(attr);
 485 }
 486
 487 static void fd_array_map_free(struct bpf_map *map)
 488 {
 489         struct bpf_array *array = container_of(map, struct bpf_array, map);
 490         int i;
 491
 492         synchronize_rcu();
 493
 494         /* make sure it's empty */
 495         for (i = 0; i < array->map.max_entries; i++)
 496                 BUG_ON(array->ptrs[i] != NULL);
 497
 498         bpf_map_area_free(array);
 499 }
 500
 501 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
 502 {
 503         return ERR_PTR(-EOPNOTSUPP);
 504 }
 505
 506 /* only called from syscall */
 507 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
 508 {
 509         void **elem, *ptr;
 510         int ret =  0;
 511
 512         if (!map->ops->map_fd_sys_lookup_elem)
 513                 return -ENOTSUPP;
 514
 515         rcu_read_lock();
 516         elem = array_map_lookup_elem(map, key);
 517         if (elem && (ptr = READ_ONCE(*elem)))
 518                 *value = map->ops->map_fd_sys_lookup_elem(ptr);
 519         else
 520                 ret = -ENOENT;
 521         rcu_read_unlock();
 522
 523         return ret;
 524 }
 525
 526 /* only called from syscall */
 527 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 528                                  void *key, void *value, u64 map_flags)
 529 {
 530         struct bpf_array *array = container_of(map, struct bpf_array, map);
 531         void *new_ptr, *old_ptr;
 532         u32 index = *(u32 *)key, ufd;
 533
 534         if (map_flags != BPF_ANY)
 535                 return -EINVAL;
 536
 537         if (index >= array->map.max_entries)
 538                 return -E2BIG;
 539
 540         ufd = *(u32 *)value;
 541         new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
 542         if (IS_ERR(new_ptr))
 543                 return PTR_ERR(new_ptr);
 544
 545         old_ptr = xchg(array->ptrs + index, new_ptr);
 546         if (old_ptr)
 547                 map->ops->map_fd_put_ptr(old_ptr);
 548
 549         return 0;
 550 }
 551
 552 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
 553 {
 554         struct bpf_array *array = container_of(map, struct bpf_array, map);
 555         void *old_ptr;
 556         u32 index = *(u32 *)key;
 557
 558         if (index >= array->map.max_entries)
 559                 return -E2BIG;
 560
 561         old_ptr = xchg(array->ptrs + index, NULL);
 562         if (old_ptr) {
 563                 map->ops->map_fd_put_ptr(old_ptr);
 564                 return 0;
 565         } else {
 566                 return -ENOENT;
 567         }
 568 }
 569
 570 static void *prog_fd_array_get_ptr(struct bpf_map *map,
 571                                    struct file *map_file, int fd)
 572 {
 573         struct bpf_array *array = container_of(map, struct bpf_array, map);
 574         struct bpf_prog *prog = bpf_prog_get(fd);
 575
 576         if (IS_ERR(prog))
 577                 return prog;
 578
 579         if (!bpf_prog_array_compatible(array, prog)) {
 580                 bpf_prog_put(prog);
 581                 return ERR_PTR(-EINVAL);
 582         }
 583
 584         return prog;
 585 }
 586
 587 static void prog_fd_array_put_ptr(void *ptr)
 588 {
 589         bpf_prog_put(ptr);
 590 }
 591
 592 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
 593 {
 594         return ((struct bpf_prog *)ptr)->aux->id;
 595 }
 596
 597 /* decrement refcnt of all bpf_progs that are stored in this map */
 598 static void bpf_fd_array_map_clear(struct bpf_map *map)
 599 {
 600         struct bpf_array *array = container_of(map, struct bpf_array, map);
 601         int i;
 602
 603         for (i = 0; i < array->map.max_entries; i++)
 604                 fd_array_map_delete_elem(map, &i);
 605 }
 606
 607 static void prog_array_map_seq_show_elem(struct bpf_map *map, void *key,
 608                                          struct seq_file *m)
 609 {
 610         void **elem, *ptr;
 611         u32 prog_id;
 612
 613         rcu_read_lock();
 614
 615         elem = array_map_lookup_elem(map, key);
 616         if (elem) {
 617                 ptr = READ_ONCE(*elem);
 618                 if (ptr) {
 619                         seq_printf(m, "%u: ", *(u32 *)key);
 620                         prog_id = prog_fd_array_sys_lookup_elem(ptr);
 621                         btf_type_seq_show(map->btf, map->btf_value_type_id,
 622                                           &prog_id, m);
 623                         seq_puts(m, "\n");
 624                 }
 625         }
 626
 627         rcu_read_unlock();
 628 }
 629
 630 const struct bpf_map_ops prog_array_map_ops = {
 631         .map_alloc_check = fd_array_map_alloc_check,
 632         .map_alloc = array_map_alloc,
 633         .map_free = fd_array_map_free,
 634         .map_get_next_key = array_map_get_next_key,
 635         .map_lookup_elem = fd_array_map_lookup_elem,
 636         .map_delete_elem = fd_array_map_delete_elem,
 637         .map_fd_get_ptr = prog_fd_array_get_ptr,
 638         .map_fd_put_ptr = prog_fd_array_put_ptr,
 639         .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
 640         .map_release_uref = bpf_fd_array_map_clear,
 641         .map_seq_show_elem = prog_array_map_seq_show_elem,
 642 };
 643
 644 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
 645                                                    struct file *map_file)
 646 {
 647         struct bpf_event_entry *ee;
 648
 649         ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
 650         if (ee) {
 651                 ee->event = perf_file->private_data;
 652                 ee->perf_file = perf_file;
 653                 ee->map_file = map_file;
 654         }
 655
 656         return ee;
 657 }
 658
 659 static void __bpf_event_entry_free(struct rcu_head *rcu)
 660 {
 661         struct bpf_event_entry *ee;
 662
 663         ee = container_of(rcu, struct bpf_event_entry, rcu);
 664         fput(ee->perf_file);
 665         kfree(ee);
 666 }
 667
 668 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
 669 {
 670         call_rcu(&ee->rcu, __bpf_event_entry_free);
 671 }
 672
 673 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
 674                                          struct file *map_file, int fd)
 675 {
 676         struct bpf_event_entry *ee;
 677         struct perf_event *event;
 678         struct file *perf_file;
 679         u64 value;
 680
 681         perf_file = perf_event_get(fd);
 682         if (IS_ERR(perf_file))
 683                 return perf_file;
 684
 685         ee = ERR_PTR(-EOPNOTSUPP);
 686         event = perf_file->private_data;
 687         if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
 688                 goto err_out;
 689
 690         ee = bpf_event_entry_gen(perf_file, map_file);
 691         if (ee)
 692                 return ee;
 693         ee = ERR_PTR(-ENOMEM);
 694 err_out:
 695         fput(perf_file);
 696         return ee;
 697 }
 698
 699 static void perf_event_fd_array_put_ptr(void *ptr)
 700 {
 701         bpf_event_entry_free_rcu(ptr);
 702 }
 703
 704 static void perf_event_fd_array_release(struct bpf_map *map,
 705                                         struct file *map_file)
 706 {
 707         struct bpf_array *array = container_of(map, struct bpf_array, map);
 708         struct bpf_event_entry *ee;
 709         int i;
 710
 711         rcu_read_lock();
 712         for (i = 0; i < array->map.max_entries; i++) {
 713                 ee = READ_ONCE(array->ptrs[i]);
 714                 if (ee && ee->map_file == map_file)
 715                         fd_array_map_delete_elem(map, &i);
 716         }
 717         rcu_read_unlock();
 718 }
 719
 720 const struct bpf_map_ops perf_event_array_map_ops = {
 721         .map_alloc_check = fd_array_map_alloc_check,
 722         .map_alloc = array_map_alloc,
 723         .map_free = fd_array_map_free,
 724         .map_get_next_key = array_map_get_next_key,
 725         .map_lookup_elem = fd_array_map_lookup_elem,
 726         .map_delete_elem = fd_array_map_delete_elem,
 727         .map_fd_get_ptr = perf_event_fd_array_get_ptr,
 728         .map_fd_put_ptr = perf_event_fd_array_put_ptr,
 729         .map_release = perf_event_fd_array_release,
 730         .map_check_btf = map_check_no_btf,
 731 };
 732
 733 #ifdef CONFIG_CGROUPS
 734 static void *cgroup_fd_array_get_ptr(struct bpf_map *map,
 735                                      struct file *map_file /* not used */,
 736                                      int fd)
 737 {
 738         return cgroup_get_from_fd(fd);
 739 }
 740
 741 static void cgroup_fd_array_put_ptr(void *ptr)
 742 {
 743         /* cgroup_put free cgrp after a rcu grace period */
 744         cgroup_put(ptr);
 745 }
 746
 747 static void cgroup_fd_array_free(struct bpf_map *map)
 748 {
 749         bpf_fd_array_map_clear(map);
 750         fd_array_map_free(map);
 751 }
 752
 753 const struct bpf_map_ops cgroup_array_map_ops = {
 754         .map_alloc_check = fd_array_map_alloc_check,
 755         .map_alloc = array_map_alloc,
 756         .map_free = cgroup_fd_array_free,
 757         .map_get_next_key = array_map_get_next_key,
 758         .map_lookup_elem = fd_array_map_lookup_elem,
 759         .map_delete_elem = fd_array_map_delete_elem,
 760         .map_fd_get_ptr = cgroup_fd_array_get_ptr,
 761         .map_fd_put_ptr = cgroup_fd_array_put_ptr,
 762         .map_check_btf = map_check_no_btf,
 763 };
 764 #endif
 765
 766 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
 767 {
 768         struct bpf_map *map, *inner_map_meta;
 769
 770         inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
 771         if (IS_ERR(inner_map_meta))
 772                 return inner_map_meta;
 773
 774         map = array_map_alloc(attr);
 775         if (IS_ERR(map)) {
 776                 bpf_map_meta_free(inner_map_meta);
 777                 return map;
 778         }
 779
 780         map->inner_map_meta = inner_map_meta;
 781
 782         return map;
 783 }
 784
 785 static void array_of_map_free(struct bpf_map *map)
 786 {
 787         /* map->inner_map_meta is only accessed by syscall which
 788          * is protected by fdget/fdput.
 789          */
 790         bpf_map_meta_free(map->inner_map_meta);
 791         bpf_fd_array_map_clear(map);
 792         fd_array_map_free(map);
 793 }
 794
 795 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
 796 {
 797         struct bpf_map **inner_map = array_map_lookup_elem(map, key);
 798
 799         if (!inner_map)
 800                 return NULL;
 801
 802         return READ_ONCE(*inner_map);
 803 }
 804
 805 static u32 array_of_map_gen_lookup(struct bpf_map *map,
 806                                    struct bpf_insn *insn_buf)
 807 {
 808         struct bpf_array *array = container_of(map, struct bpf_array, map);
 809         u32 elem_size = round_up(map->value_size, 8);
 810         struct bpf_insn *insn = insn_buf;
 811         const int ret = BPF_REG_0;
 812         const int map_ptr = BPF_REG_1;
 813         const int index = BPF_REG_2;
 814
 815         *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 816         *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 817         if (map->unpriv_array) {
 818                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
 819                 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
 820         } else {
 821                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
 822         }
 823         if (is_power_of_2(elem_size))
 824                 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
 825         else
 826                 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
 827         *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
 828         *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
 829         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
 830         *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 831         *insn++ = BPF_MOV64_IMM(ret, 0);
 832
 833         return insn - insn_buf;
 834 }
 835
 836 const struct bpf_map_ops array_of_maps_map_ops = {
 837         .map_alloc_check = fd_array_map_alloc_check,
 838         .map_alloc = array_of_map_alloc,
 839         .map_free = array_of_map_free,
 840         .map_get_next_key = array_map_get_next_key,
 841         .map_lookup_elem = array_of_map_lookup_elem,
 842         .map_delete_elem = fd_array_map_delete_elem,
 843         .map_fd_get_ptr = bpf_map_fd_get_ptr,
 844         .map_fd_put_ptr = bpf_map_fd_put_ptr,
 845         .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
 846         .map_gen_lookup = array_of_map_gen_lookup,
 847         .map_check_btf = map_check_no_btf,
 848 };