kernel/bpf/arraymap.c

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