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