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 }
  32
  33 static int bpf_array_alloc_percpu(struct bpf_array *array)
  34 {
  35         void __percpu *ptr;
  36         int i;
  37
  38         for (i = 0; i < array->map.max_entries; i++) {
  39                 ptr = __alloc_percpu_gfp(array->elem_size, 8,
  40                                          GFP_USER | __GFP_NOWARN);
  41                 if (!ptr) {
  42                         bpf_array_free_percpu(array);
  43                         return -ENOMEM;
  44                 }
  45                 array->pptrs[i] = ptr;
  46         }
  47
  48         return 0;
  49 }
  50
  51 /* Called from syscall */
  52 static int array_map_alloc_check(union bpf_attr *attr)
  53 {
  54         bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  55         int numa_node = bpf_map_attr_numa_node(attr);
  56
  57         /* check sanity of attributes */
  58         if (attr->max_entries == 0 || attr->key_size != 4 ||
  59             attr->value_size == 0 ||
  60             attr->map_flags & ~ARRAY_CREATE_FLAG_MASK ||
  61             (percpu && numa_node != NUMA_NO_NODE))
  62                 return -EINVAL;
  63
  64         if (attr->value_size > KMALLOC_MAX_SIZE)
  65                 /* if value_size is bigger, the user space won't be able to
  66                  * access the elements.
  67                  */
  68                 return -E2BIG;
  69
  70         return 0;
  71 }
  72
  73 static struct bpf_map *array_map_alloc(union bpf_attr *attr)
  74 {
  75         bool percpu = attr->map_type == BPF_MAP_TYPE_PERCPU_ARRAY;
  76         int numa_node = bpf_map_attr_numa_node(attr);
  77         u32 elem_size, index_mask, max_entries;
  78         bool unpriv = !capable(CAP_SYS_ADMIN);
  79         struct bpf_array *array;
  80         u64 array_size, mask64;
  81
  82         elem_size = round_up(attr->value_size, 8);
  83
  84         max_entries = attr->max_entries;
  85
  86         /* On 32 bit archs roundup_pow_of_two() with max_entries that has
  87          * upper most bit set in u32 space is undefined behavior due to
  88          * resulting 1U << 32, so do it manually here in u64 space.
  89          */
  90         mask64 = fls_long(max_entries - 1);
  91         mask64 = 1ULL << mask64;
  92         mask64 -= 1;
  93
  94         index_mask = mask64;
  95         if (unpriv) {
  96                 /* round up array size to nearest power of 2,
  97                  * since cpu will speculate within index_mask limits
  98                  */
  99                 max_entries = index_mask + 1;
 100                 /* Check for overflows. */
 101                 if (max_entries < attr->max_entries)
 102                         return ERR_PTR(-E2BIG);
 103         }
 104
 105         array_size = sizeof(*array);
 106         if (percpu)
 107                 array_size += (u64) max_entries * sizeof(void *);
 108         else
 109                 array_size += (u64) max_entries * elem_size;
 110
 111         /* make sure there is no u32 overflow later in round_up() */
 112         if (array_size >= U32_MAX - PAGE_SIZE)
 113                 return ERR_PTR(-ENOMEM);
 114
 115         /* allocate all map elements and zero-initialize them */
 116         array = bpf_map_area_alloc(array_size, numa_node);
 117         if (!array)
 118                 return ERR_PTR(-ENOMEM);
 119         array->index_mask = index_mask;
 120         array->map.unpriv_array = unpriv;
 121
 122         /* copy mandatory map attributes */
 123         bpf_map_init_from_attr(&array->map, attr);
 124         array->elem_size = elem_size;
 125
 126         if (!percpu)
 127                 goto out;
 128
 129         array_size += (u64) attr->max_entries * elem_size * num_possible_cpus();
 130
 131         if (array_size >= U32_MAX - PAGE_SIZE ||
 132             bpf_array_alloc_percpu(array)) {
 133                 bpf_map_area_free(array);
 134                 return ERR_PTR(-ENOMEM);
 135         }
 136 out:
 137         array->map.pages = round_up(array_size, PAGE_SIZE) >> PAGE_SHIFT;
 138
 139         return &array->map;
 140 }
 141
 142 /* Called from syscall or from eBPF program */
 143 static void *array_map_lookup_elem(struct bpf_map *map, void *key)
 144 {
 145         struct bpf_array *array = container_of(map, struct bpf_array, map);
 146         u32 index = *(u32 *)key;
 147
 148         if (unlikely(index >= array->map.max_entries))
 149                 return NULL;
 150
 151         return array->value + array->elem_size * (index & array->index_mask);
 152 }
 153
 154 /* emit BPF instructions equivalent to C code of array_map_lookup_elem() */
 155 static u32 array_map_gen_lookup(struct bpf_map *map, struct bpf_insn *insn_buf)
 156 {
 157         struct bpf_array *array = container_of(map, struct bpf_array, map);
 158         struct bpf_insn *insn = insn_buf;
 159         u32 elem_size = round_up(map->value_size, 8);
 160         const int ret = BPF_REG_0;
 161         const int map_ptr = BPF_REG_1;
 162         const int index = BPF_REG_2;
 163
 164         *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 165         *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 166         if (map->unpriv_array) {
 167                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 4);
 168                 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
 169         } else {
 170                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 3);
 171         }
 172
 173         if (is_power_of_2(elem_size)) {
 174                 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
 175         } else {
 176                 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
 177         }
 178         *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
 179         *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 180         *insn++ = BPF_MOV64_IMM(ret, 0);
 181         return insn - insn_buf;
 182 }
 183
 184 /* Called from eBPF program */
 185 static void *percpu_array_map_lookup_elem(struct bpf_map *map, void *key)
 186 {
 187         struct bpf_array *array = container_of(map, struct bpf_array, map);
 188         u32 index = *(u32 *)key;
 189
 190         if (unlikely(index >= array->map.max_entries))
 191                 return NULL;
 192
 193         return this_cpu_ptr(array->pptrs[index & array->index_mask]);
 194 }
 195
 196 int bpf_percpu_array_copy(struct bpf_map *map, void *key, void *value)
 197 {
 198         struct bpf_array *array = container_of(map, struct bpf_array, map);
 199         u32 index = *(u32 *)key;
 200         void __percpu *pptr;
 201         int cpu, off = 0;
 202         u32 size;
 203
 204         if (unlikely(index >= array->map.max_entries))
 205                 return -ENOENT;
 206
 207         /* per_cpu areas are zero-filled and bpf programs can only
 208          * access 'value_size' of them, so copying rounded areas
 209          * will not leak any kernel data
 210          */
 211         size = round_up(map->value_size, 8);
 212         rcu_read_lock();
 213         pptr = array->pptrs[index & array->index_mask];
 214         for_each_possible_cpu(cpu) {
 215                 bpf_long_memcpy(value + off, per_cpu_ptr(pptr, cpu), size);
 216                 off += size;
 217         }
 218         rcu_read_unlock();
 219         return 0;
 220 }
 221
 222 /* Called from syscall */
 223 static int array_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 224 {
 225         struct bpf_array *array = container_of(map, struct bpf_array, map);
 226         u32 index = key ? *(u32 *)key : U32_MAX;
 227         u32 *next = (u32 *)next_key;
 228
 229         if (index >= array->map.max_entries) {
 230                 *next = 0;
 231                 return 0;
 232         }
 233
 234         if (index == array->map.max_entries - 1)
 235                 return -ENOENT;
 236
 237         *next = index + 1;
 238         return 0;
 239 }
 240
 241 /* Called from syscall or from eBPF program */
 242 static int array_map_update_elem(struct bpf_map *map, void *key, void *value,
 243                                  u64 map_flags)
 244 {
 245         struct bpf_array *array = container_of(map, struct bpf_array, map);
 246         u32 index = *(u32 *)key;
 247
 248         if (unlikely(map_flags > BPF_EXIST))
 249                 /* unknown flags */
 250                 return -EINVAL;
 251
 252         if (unlikely(index >= array->map.max_entries))
 253                 /* all elements were pre-allocated, cannot insert a new one */
 254                 return -E2BIG;
 255
 256         if (unlikely(map_flags == BPF_NOEXIST))
 257                 /* all elements already exist */
 258                 return -EEXIST;
 259
 260         if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 261                 memcpy(this_cpu_ptr(array->pptrs[index & array->index_mask]),
 262                        value, map->value_size);
 263         else
 264                 memcpy(array->value +
 265                        array->elem_size * (index & array->index_mask),
 266                        value, map->value_size);
 267         return 0;
 268 }
 269
 270 int bpf_percpu_array_update(struct bpf_map *map, void *key, void *value,
 271                             u64 map_flags)
 272 {
 273         struct bpf_array *array = container_of(map, struct bpf_array, map);
 274         u32 index = *(u32 *)key;
 275         void __percpu *pptr;
 276         int cpu, off = 0;
 277         u32 size;
 278
 279         if (unlikely(map_flags > BPF_EXIST))
 280                 /* unknown flags */
 281                 return -EINVAL;
 282
 283         if (unlikely(index >= array->map.max_entries))
 284                 /* all elements were pre-allocated, cannot insert a new one */
 285                 return -E2BIG;
 286
 287         if (unlikely(map_flags == BPF_NOEXIST))
 288                 /* all elements already exist */
 289                 return -EEXIST;
 290
 291         /* the user space will provide round_up(value_size, 8) bytes that
 292          * will be copied into per-cpu area. bpf programs can only access
 293          * value_size of it. During lookup the same extra bytes will be
 294          * returned or zeros which were zero-filled by percpu_alloc,
 295          * so no kernel data leaks possible
 296          */
 297         size = round_up(map->value_size, 8);
 298         rcu_read_lock();
 299         pptr = array->pptrs[index & array->index_mask];
 300         for_each_possible_cpu(cpu) {
 301                 bpf_long_memcpy(per_cpu_ptr(pptr, cpu), value + off, size);
 302                 off += size;
 303         }
 304         rcu_read_unlock();
 305         return 0;
 306 }
 307
 308 /* Called from syscall or from eBPF program */
 309 static int array_map_delete_elem(struct bpf_map *map, void *key)
 310 {
 311         return -EINVAL;
 312 }
 313
 314 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 315 static void array_map_free(struct bpf_map *map)
 316 {
 317         struct bpf_array *array = container_of(map, struct bpf_array, map);
 318
 319         /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
 320          * so the programs (can be more than one that used this map) were
 321          * disconnected from events. Wait for outstanding programs to complete
 322          * and free the array
 323          */
 324         synchronize_rcu();
 325
 326         if (array->map.map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 327                 bpf_array_free_percpu(array);
 328
 329         bpf_map_area_free(array);
 330 }
 331
 332 const struct bpf_map_ops array_map_ops = {
 333         .map_alloc_check = array_map_alloc_check,
 334         .map_alloc = array_map_alloc,
 335         .map_free = array_map_free,
 336         .map_get_next_key = array_map_get_next_key,
 337         .map_lookup_elem = array_map_lookup_elem,
 338         .map_update_elem = array_map_update_elem,
 339         .map_delete_elem = array_map_delete_elem,
 340         .map_gen_lookup = array_map_gen_lookup,
 341 };
 342
 343 const struct bpf_map_ops percpu_array_map_ops = {
 344         .map_alloc_check = array_map_alloc_check,
 345         .map_alloc = array_map_alloc,
 346         .map_free = array_map_free,
 347         .map_get_next_key = array_map_get_next_key,
 348         .map_lookup_elem = percpu_array_map_lookup_elem,
 349         .map_update_elem = array_map_update_elem,
 350         .map_delete_elem = array_map_delete_elem,
 351 };
 352
 353 static int fd_array_map_alloc_check(union bpf_attr *attr)
 354 {
 355         /* only file descriptors can be stored in this type of map */
 356         if (attr->value_size != sizeof(u32))
 357                 return -EINVAL;
 358         return array_map_alloc_check(attr);
 359 }
 360
 361 static void fd_array_map_free(struct bpf_map *map)
 362 {
 363         struct bpf_array *array = container_of(map, struct bpf_array, map);
 364         int i;
 365
 366         synchronize_rcu();
 367
 368         /* make sure it's empty */
 369         for (i = 0; i < array->map.max_entries; i++)
 370                 BUG_ON(array->ptrs[i] != NULL);
 371
 372         bpf_map_area_free(array);
 373 }
 374
 375 static void *fd_array_map_lookup_elem(struct bpf_map *map, void *key)
 376 {
 377         return NULL;
 378 }
 379
 380 /* only called from syscall */
 381 int bpf_fd_array_map_lookup_elem(struct bpf_map *map, void *key, u32 *value)
 382 {
 383         void **elem, *ptr;
 384         int ret =  0;
 385
 386         if (!map->ops->map_fd_sys_lookup_elem)
 387                 return -ENOTSUPP;
 388
 389         rcu_read_lock();
 390         elem = array_map_lookup_elem(map, key);
 391         if (elem && (ptr = READ_ONCE(*elem)))
 392                 *value = map->ops->map_fd_sys_lookup_elem(ptr);
 393         else
 394                 ret = -ENOENT;
 395         rcu_read_unlock();
 396
 397         return ret;
 398 }
 399
 400 /* only called from syscall */
 401 int bpf_fd_array_map_update_elem(struct bpf_map *map, struct file *map_file,
 402                                  void *key, void *value, u64 map_flags)
 403 {
 404         struct bpf_array *array = container_of(map, struct bpf_array, map);
 405         void *new_ptr, *old_ptr;
 406         u32 index = *(u32 *)key, ufd;
 407
 408         if (map_flags != BPF_ANY)
 409                 return -EINVAL;
 410
 411         if (index >= array->map.max_entries)
 412                 return -E2BIG;
 413
 414         ufd = *(u32 *)value;
 415         new_ptr = map->ops->map_fd_get_ptr(map, map_file, ufd);
 416         if (IS_ERR(new_ptr))
 417                 return PTR_ERR(new_ptr);
 418
 419         old_ptr = xchg(array->ptrs + index, new_ptr);
 420         if (old_ptr)
 421                 map->ops->map_fd_put_ptr(old_ptr);
 422
 423         return 0;
 424 }
 425
 426 static int fd_array_map_delete_elem(struct bpf_map *map, void *key)
 427 {
 428         struct bpf_array *array = container_of(map, struct bpf_array, map);
 429         void *old_ptr;
 430         u32 index = *(u32 *)key;
 431
 432         if (index >= array->map.max_entries)
 433                 return -E2BIG;
 434
 435         old_ptr = xchg(array->ptrs + index, NULL);
 436         if (old_ptr) {
 437                 map->ops->map_fd_put_ptr(old_ptr);
 438                 return 0;
 439         } else {
 440                 return -ENOENT;
 441         }
 442 }
 443
 444 static void *prog_fd_array_get_ptr(struct bpf_map *map,
 445                                    struct file *map_file, int fd)
 446 {
 447         struct bpf_array *array = container_of(map, struct bpf_array, map);
 448         struct bpf_prog *prog = bpf_prog_get(fd);
 449
 450         if (IS_ERR(prog))
 451                 return prog;
 452
 453         if (!bpf_prog_array_compatible(array, prog)) {
 454                 bpf_prog_put(prog);
 455                 return ERR_PTR(-EINVAL);
 456         }
 457
 458         return prog;
 459 }
 460
 461 static void prog_fd_array_put_ptr(void *ptr)
 462 {
 463         bpf_prog_put(ptr);
 464 }
 465
 466 static u32 prog_fd_array_sys_lookup_elem(void *ptr)
 467 {
 468         return ((struct bpf_prog *)ptr)->aux->id;
 469 }
 470
 471 /* decrement refcnt of all bpf_progs that are stored in this map */
 472 void bpf_fd_array_map_clear(struct bpf_map *map)
 473 {
 474         struct bpf_array *array = container_of(map, struct bpf_array, map);
 475         int i;
 476
 477         for (i = 0; i < array->map.max_entries; i++)
 478                 fd_array_map_delete_elem(map, &i);
 479 }
 480
 481 const struct bpf_map_ops prog_array_map_ops = {
 482         .map_alloc_check = fd_array_map_alloc_check,
 483         .map_alloc = array_map_alloc,
 484         .map_free = fd_array_map_free,
 485         .map_get_next_key = array_map_get_next_key,
 486         .map_lookup_elem = fd_array_map_lookup_elem,
 487         .map_delete_elem = fd_array_map_delete_elem,
 488         .map_fd_get_ptr = prog_fd_array_get_ptr,
 489         .map_fd_put_ptr = prog_fd_array_put_ptr,
 490         .map_fd_sys_lookup_elem = prog_fd_array_sys_lookup_elem,
 491 };
 492
 493 static struct bpf_event_entry *bpf_event_entry_gen(struct file *perf_file,
 494                                                    struct file *map_file)
 495 {
 496         struct bpf_event_entry *ee;
 497
 498         ee = kzalloc(sizeof(*ee), GFP_ATOMIC);
 499         if (ee) {
 500                 ee->event = perf_file->private_data;
 501                 ee->perf_file = perf_file;
 502                 ee->map_file = map_file;
 503         }
 504
 505         return ee;
 506 }
 507
 508 static void __bpf_event_entry_free(struct rcu_head *rcu)
 509 {
 510         struct bpf_event_entry *ee;
 511
 512         ee = container_of(rcu, struct bpf_event_entry, rcu);
 513         fput(ee->perf_file);
 514         kfree(ee);
 515 }
 516
 517 static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
 518 {
 519         call_rcu(&ee->rcu, __bpf_event_entry_free);
 520 }
 521
 522 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
 523                                          struct file *map_file, int fd)
 524 {
 525         struct bpf_event_entry *ee;
 526         struct perf_event *event;
 527         struct file *perf_file;
 528         u64 value;
 529
 530         perf_file = perf_event_get(fd);
 531         if (IS_ERR(perf_file))
 532                 return perf_file;
 533
 534         ee = ERR_PTR(-EOPNOTSUPP);
 535         event = perf_file->private_data;
 536         if (perf_event_read_local(event, &value, NULL, NULL) == -EOPNOTSUPP)
 537                 goto err_out;
 538
 539         ee = bpf_event_entry_gen(perf_file, map_file);
 540         if (ee)
 541                 return ee;
 542         ee = ERR_PTR(-ENOMEM);
 543 err_out:
 544         fput(perf_file);
 545         return ee;
 546 }
 547
 548 static void perf_event_fd_array_put_ptr(void *ptr)
 549 {
 550         bpf_event_entry_free_rcu(ptr);
 551 }
 552
 553 static void perf_event_fd_array_release(struct bpf_map *map,
 554                                         struct file *map_file)
 555 {
 556         struct bpf_array *array = container_of(map, struct bpf_array, map);
 557         struct bpf_event_entry *ee;
 558         int i;
 559
 560         rcu_read_lock();
 561         for (i = 0; i < array->map.max_entries; i++) {
 562                 ee = READ_ONCE(array->ptrs[i]);
 563                 if (ee && ee->map_file == map_file)
 564                         fd_array_map_delete_elem(map, &i);
 565         }
 566         rcu_read_unlock();
 567 }
 568
 569 const struct bpf_map_ops perf_event_array_map_ops = {
 570         .map_alloc_check = fd_array_map_alloc_check,
 571         .map_alloc = 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_check = fd_array_map_alloc_check,
 603         .map_alloc = array_map_alloc,
 604         .map_free = cgroup_fd_array_free,
 605         .map_get_next_key = array_map_get_next_key,
 606         .map_lookup_elem = fd_array_map_lookup_elem,
 607         .map_delete_elem = fd_array_map_delete_elem,
 608         .map_fd_get_ptr = cgroup_fd_array_get_ptr,
 609         .map_fd_put_ptr = cgroup_fd_array_put_ptr,
 610 };
 611 #endif
 612
 613 static struct bpf_map *array_of_map_alloc(union bpf_attr *attr)
 614 {
 615         struct bpf_map *map, *inner_map_meta;
 616
 617         inner_map_meta = bpf_map_meta_alloc(attr->inner_map_fd);
 618         if (IS_ERR(inner_map_meta))
 619                 return inner_map_meta;
 620
 621         map = array_map_alloc(attr);
 622         if (IS_ERR(map)) {
 623                 bpf_map_meta_free(inner_map_meta);
 624                 return map;
 625         }
 626
 627         map->inner_map_meta = inner_map_meta;
 628
 629         return map;
 630 }
 631
 632 static void array_of_map_free(struct bpf_map *map)
 633 {
 634         /* map->inner_map_meta is only accessed by syscall which
 635          * is protected by fdget/fdput.
 636          */
 637         bpf_map_meta_free(map->inner_map_meta);
 638         bpf_fd_array_map_clear(map);
 639         fd_array_map_free(map);
 640 }
 641
 642 static void *array_of_map_lookup_elem(struct bpf_map *map, void *key)
 643 {
 644         struct bpf_map **inner_map = array_map_lookup_elem(map, key);
 645
 646         if (!inner_map)
 647                 return NULL;
 648
 649         return READ_ONCE(*inner_map);
 650 }
 651
 652 static u32 array_of_map_gen_lookup(struct bpf_map *map,
 653                                    struct bpf_insn *insn_buf)
 654 {
 655         struct bpf_array *array = container_of(map, struct bpf_array, map);
 656         u32 elem_size = round_up(map->value_size, 8);
 657         struct bpf_insn *insn = insn_buf;
 658         const int ret = BPF_REG_0;
 659         const int map_ptr = BPF_REG_1;
 660         const int index = BPF_REG_2;
 661
 662         *insn++ = BPF_ALU64_IMM(BPF_ADD, map_ptr, offsetof(struct bpf_array, value));
 663         *insn++ = BPF_LDX_MEM(BPF_W, ret, index, 0);
 664         if (map->unpriv_array) {
 665                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 6);
 666                 *insn++ = BPF_ALU32_IMM(BPF_AND, ret, array->index_mask);
 667         } else {
 668                 *insn++ = BPF_JMP_IMM(BPF_JGE, ret, map->max_entries, 5);
 669         }
 670         if (is_power_of_2(elem_size))
 671                 *insn++ = BPF_ALU64_IMM(BPF_LSH, ret, ilog2(elem_size));
 672         else
 673                 *insn++ = BPF_ALU64_IMM(BPF_MUL, ret, elem_size);
 674         *insn++ = BPF_ALU64_REG(BPF_ADD, ret, map_ptr);
 675         *insn++ = BPF_LDX_MEM(BPF_DW, ret, ret, 0);
 676         *insn++ = BPF_JMP_IMM(BPF_JEQ, ret, 0, 1);
 677         *insn++ = BPF_JMP_IMM(BPF_JA, 0, 0, 1);
 678         *insn++ = BPF_MOV64_IMM(ret, 0);
 679
 680         return insn - insn_buf;
 681 }
 682
 683 const struct bpf_map_ops array_of_maps_map_ops = {
 684         .map_alloc_check = fd_array_map_alloc_check,
 685         .map_alloc = array_of_map_alloc,
 686         .map_free = array_of_map_free,
 687         .map_get_next_key = array_map_get_next_key,
 688         .map_lookup_elem = array_of_map_lookup_elem,
 689         .map_delete_elem = fd_array_map_delete_elem,
 690         .map_fd_get_ptr = bpf_map_fd_get_ptr,
 691         .map_fd_put_ptr = bpf_map_fd_put_ptr,
 692         .map_fd_sys_lookup_elem = bpf_map_fd_sys_lookup_elem,
 693         .map_gen_lookup = array_of_map_gen_lookup,
 694 };