kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26 #include <uapi/linux/btf.h>
  27
  28 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  29                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  30                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  31 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  32 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  33 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  34                         IS_FD_HASH(map))
  35
  36 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  37
  38 DEFINE_PER_CPU(int, bpf_prog_active);
  39 static DEFINE_IDR(prog_idr);
  40 static DEFINE_SPINLOCK(prog_idr_lock);
  41 static DEFINE_IDR(map_idr);
  42 static DEFINE_SPINLOCK(map_idr_lock);
  43
  44 int sysctl_unprivileged_bpf_disabled __read_mostly;
  45
  46 static const struct bpf_map_ops * const bpf_map_types[] = {
  47 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  48 #define BPF_MAP_TYPE(_id, _ops) \
  49         [_id] = &_ops,
  50 #include <linux/bpf_types.h>
  51 #undef BPF_PROG_TYPE
  52 #undef BPF_MAP_TYPE
  53 };
  54
  55 /*
  56  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  57  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  58  * we don't know about yet.
  59  *
  60  * There is a ToCToU between this function call and the following
  61  * copy_from_user() call. However, this is not a concern since this function is
  62  * meant to be a future-proofing of bits.
  63  */
  64 int bpf_check_uarg_tail_zero(void __user *uaddr,
  65                              size_t expected_size,
  66                              size_t actual_size)
  67 {
  68         unsigned char __user *addr;
  69         unsigned char __user *end;
  70         unsigned char val;
  71         int err;
  72
  73         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  74                 return -E2BIG;
  75
  76         if (unlikely(!access_ok(uaddr, actual_size)))
  77                 return -EFAULT;
  78
  79         if (actual_size <= expected_size)
  80                 return 0;
  81
  82         addr = uaddr + expected_size;
  83         end  = uaddr + actual_size;
  84
  85         for (; addr < end; addr++) {
  86                 err = get_user(val, addr);
  87                 if (err)
  88                         return err;
  89                 if (val)
  90                         return -E2BIG;
  91         }
  92
  93         return 0;
  94 }
  95
  96 const struct bpf_map_ops bpf_map_offload_ops = {
  97         .map_alloc = bpf_map_offload_map_alloc,
  98         .map_free = bpf_map_offload_map_free,
  99         .map_check_btf = map_check_no_btf,
 100 };
 101
 102 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 103 {
 104         const struct bpf_map_ops *ops;
 105         u32 type = attr->map_type;
 106         struct bpf_map *map;
 107         int err;
 108
 109         if (type >= ARRAY_SIZE(bpf_map_types))
 110                 return ERR_PTR(-EINVAL);
 111         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 112         ops = bpf_map_types[type];
 113         if (!ops)
 114                 return ERR_PTR(-EINVAL);
 115
 116         if (ops->map_alloc_check) {
 117                 err = ops->map_alloc_check(attr);
 118                 if (err)
 119                         return ERR_PTR(err);
 120         }
 121         if (attr->map_ifindex)
 122                 ops = &bpf_map_offload_ops;
 123         map = ops->map_alloc(attr);
 124         if (IS_ERR(map))
 125                 return map;
 126         map->ops = ops;
 127         map->map_type = type;
 128         return map;
 129 }
 130
 131 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 132 {
 133         /* We really just want to fail instead of triggering OOM killer
 134          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 135          * which is used for lower order allocation requests.
 136          *
 137          * It has been observed that higher order allocation requests done by
 138          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 139          * to reclaim memory from the page cache, thus we set
 140          * __GFP_RETRY_MAYFAIL to avoid such situations.
 141          */
 142
 143         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 144         void *area;
 145
 146         if (size >= SIZE_MAX)
 147                 return NULL;
 148
 149         /* kmalloc()'ed memory can't be mmap()'ed */
 150         if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 151                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 152                                     numa_node);
 153                 if (area != NULL)
 154                         return area;
 155         }
 156         if (mmapable) {
 157                 BUG_ON(!PAGE_ALIGNED(size));
 158                 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
 159                                                __GFP_RETRY_MAYFAIL | flags);
 160         }
 161         return __vmalloc_node_flags_caller(size, numa_node,
 162                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 163                                            flags, __builtin_return_address(0));
 164 }
 165
 166 void *bpf_map_area_alloc(u64 size, int numa_node)
 167 {
 168         return __bpf_map_area_alloc(size, numa_node, false);
 169 }
 170
 171 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 172 {
 173         return __bpf_map_area_alloc(size, numa_node, true);
 174 }
 175
 176 void bpf_map_area_free(void *area)
 177 {
 178         kvfree(area);
 179 }
 180
 181 static u32 bpf_map_flags_retain_permanent(u32 flags)
 182 {
 183         /* Some map creation flags are not tied to the map object but
 184          * rather to the map fd instead, so they have no meaning upon
 185          * map object inspection since multiple file descriptors with
 186          * different (access) properties can exist here. Thus, given
 187          * this has zero meaning for the map itself, lets clear these
 188          * from here.
 189          */
 190         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 191 }
 192
 193 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 194 {
 195         map->map_type = attr->map_type;
 196         map->key_size = attr->key_size;
 197         map->value_size = attr->value_size;
 198         map->max_entries = attr->max_entries;
 199         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 200         map->numa_node = bpf_map_attr_numa_node(attr);
 201 }
 202
 203 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 204 {
 205         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 206
 207         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 208                 atomic_long_sub(pages, &user->locked_vm);
 209                 return -EPERM;
 210         }
 211         return 0;
 212 }
 213
 214 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 215 {
 216         if (user)
 217                 atomic_long_sub(pages, &user->locked_vm);
 218 }
 219
 220 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
 221 {
 222         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 223         struct user_struct *user;
 224         int ret;
 225
 226         if (size >= U32_MAX - PAGE_SIZE)
 227                 return -E2BIG;
 228
 229         user = get_current_user();
 230         ret = bpf_charge_memlock(user, pages);
 231         if (ret) {
 232                 free_uid(user);
 233                 return ret;
 234         }
 235
 236         mem->pages = pages;
 237         mem->user = user;
 238
 239         return 0;
 240 }
 241
 242 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 243 {
 244         bpf_uncharge_memlock(mem->user, mem->pages);
 245         free_uid(mem->user);
 246 }
 247
 248 void bpf_map_charge_move(struct bpf_map_memory *dst,
 249                          struct bpf_map_memory *src)
 250 {
 251         *dst = *src;
 252
 253         /* Make sure src will not be used for the redundant uncharging. */
 254         memset(src, 0, sizeof(struct bpf_map_memory));
 255 }
 256
 257 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 258 {
 259         int ret;
 260
 261         ret = bpf_charge_memlock(map->memory.user, pages);
 262         if (ret)
 263                 return ret;
 264         map->memory.pages += pages;
 265         return ret;
 266 }
 267
 268 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 269 {
 270         bpf_uncharge_memlock(map->memory.user, pages);
 271         map->memory.pages -= pages;
 272 }
 273
 274 static int bpf_map_alloc_id(struct bpf_map *map)
 275 {
 276         int id;
 277
 278         idr_preload(GFP_KERNEL);
 279         spin_lock_bh(&map_idr_lock);
 280         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 281         if (id > 0)
 282                 map->id = id;
 283         spin_unlock_bh(&map_idr_lock);
 284         idr_preload_end();
 285
 286         if (WARN_ON_ONCE(!id))
 287                 return -ENOSPC;
 288
 289         return id > 0 ? 0 : id;
 290 }
 291
 292 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 293 {
 294         unsigned long flags;
 295
 296         /* Offloaded maps are removed from the IDR store when their device
 297          * disappears - even if someone holds an fd to them they are unusable,
 298          * the memory is gone, all ops will fail; they are simply waiting for
 299          * refcnt to drop to be freed.
 300          */
 301         if (!map->id)
 302                 return;
 303
 304         if (do_idr_lock)
 305                 spin_lock_irqsave(&map_idr_lock, flags);
 306         else
 307                 __acquire(&map_idr_lock);
 308
 309         idr_remove(&map_idr, map->id);
 310         map->id = 0;
 311
 312         if (do_idr_lock)
 313                 spin_unlock_irqrestore(&map_idr_lock, flags);
 314         else
 315                 __release(&map_idr_lock);
 316 }
 317
 318 /* called from workqueue */
 319 static void bpf_map_free_deferred(struct work_struct *work)
 320 {
 321         struct bpf_map *map = container_of(work, struct bpf_map, work);
 322         struct bpf_map_memory mem;
 323
 324         bpf_map_charge_move(&mem, &map->memory);
 325         security_bpf_map_free(map);
 326         /* implementation dependent freeing */
 327         map->ops->map_free(map);
 328         bpf_map_charge_finish(&mem);
 329 }
 330
 331 static void bpf_map_put_uref(struct bpf_map *map)
 332 {
 333         if (atomic64_dec_and_test(&map->usercnt)) {
 334                 if (map->ops->map_release_uref)
 335                         map->ops->map_release_uref(map);
 336         }
 337 }
 338
 339 /* decrement map refcnt and schedule it for freeing via workqueue
 340  * (unrelying map implementation ops->map_free() might sleep)
 341  */
 342 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 343 {
 344         if (atomic64_dec_and_test(&map->refcnt)) {
 345                 /* bpf_map_free_id() must be called first */
 346                 bpf_map_free_id(map, do_idr_lock);
 347                 btf_put(map->btf);
 348                 INIT_WORK(&map->work, bpf_map_free_deferred);
 349                 schedule_work(&map->work);
 350         }
 351 }
 352
 353 void bpf_map_put(struct bpf_map *map)
 354 {
 355         __bpf_map_put(map, true);
 356 }
 357 EXPORT_SYMBOL_GPL(bpf_map_put);
 358
 359 void bpf_map_put_with_uref(struct bpf_map *map)
 360 {
 361         bpf_map_put_uref(map);
 362         bpf_map_put(map);
 363 }
 364
 365 static int bpf_map_release(struct inode *inode, struct file *filp)
 366 {
 367         struct bpf_map *map = filp->private_data;
 368
 369         if (map->ops->map_release)
 370                 map->ops->map_release(map, filp);
 371
 372         bpf_map_put_with_uref(map);
 373         return 0;
 374 }
 375
 376 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 377 {
 378         fmode_t mode = f.file->f_mode;
 379
 380         /* Our file permissions may have been overridden by global
 381          * map permissions facing syscall side.
 382          */
 383         if (READ_ONCE(map->frozen))
 384                 mode &= ~FMODE_CAN_WRITE;
 385         return mode;
 386 }
 387
 388 #ifdef CONFIG_PROC_FS
 389 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 390 {
 391         const struct bpf_map *map = filp->private_data;
 392         const struct bpf_array *array;
 393         u32 type = 0, jited = 0;
 394
 395         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 396                 array = container_of(map, struct bpf_array, map);
 397                 type  = array->aux->type;
 398                 jited = array->aux->jited;
 399         }
 400
 401         seq_printf(m,
 402                    "map_type:\t%u\n"
 403                    "key_size:\t%u\n"
 404                    "value_size:\t%u\n"
 405                    "max_entries:\t%u\n"
 406                    "map_flags:\t%#x\n"
 407                    "memlock:\t%llu\n"
 408                    "map_id:\t%u\n"
 409                    "frozen:\t%u\n",
 410                    map->map_type,
 411                    map->key_size,
 412                    map->value_size,
 413                    map->max_entries,
 414                    map->map_flags,
 415                    map->memory.pages * 1ULL << PAGE_SHIFT,
 416                    map->id,
 417                    READ_ONCE(map->frozen));
 418         if (type) {
 419                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 420                 seq_printf(m, "owner_jited:\t%u\n", jited);
 421         }
 422 }
 423 #endif
 424
 425 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 426                               loff_t *ppos)
 427 {
 428         /* We need this handler such that alloc_file() enables
 429          * f_mode with FMODE_CAN_READ.
 430          */
 431         return -EINVAL;
 432 }
 433
 434 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 435                                size_t siz, loff_t *ppos)
 436 {
 437         /* We need this handler such that alloc_file() enables
 438          * f_mode with FMODE_CAN_WRITE.
 439          */
 440         return -EINVAL;
 441 }
 442
 443 /* called for any extra memory-mapped regions (except initial) */
 444 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 445 {
 446         struct bpf_map *map = vma->vm_file->private_data;
 447
 448         bpf_map_inc_with_uref(map);
 449
 450         if (vma->vm_flags & VM_WRITE) {
 451                 mutex_lock(&map->freeze_mutex);
 452                 map->writecnt++;
 453                 mutex_unlock(&map->freeze_mutex);
 454         }
 455 }
 456
 457 /* called for all unmapped memory region (including initial) */
 458 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 459 {
 460         struct bpf_map *map = vma->vm_file->private_data;
 461
 462         if (vma->vm_flags & VM_WRITE) {
 463                 mutex_lock(&map->freeze_mutex);
 464                 map->writecnt--;
 465                 mutex_unlock(&map->freeze_mutex);
 466         }
 467
 468         bpf_map_put_with_uref(map);
 469 }
 470
 471 static const struct vm_operations_struct bpf_map_default_vmops = {
 472         .open           = bpf_map_mmap_open,
 473         .close          = bpf_map_mmap_close,
 474 };
 475
 476 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 477 {
 478         struct bpf_map *map = filp->private_data;
 479         int err;
 480
 481         if (!map->ops->map_mmap || map_value_has_spin_lock(map))
 482                 return -ENOTSUPP;
 483
 484         if (!(vma->vm_flags & VM_SHARED))
 485                 return -EINVAL;
 486
 487         mutex_lock(&map->freeze_mutex);
 488
 489         if ((vma->vm_flags & VM_WRITE) && map->frozen) {
 490                 err = -EPERM;
 491                 goto out;
 492         }
 493
 494         /* set default open/close callbacks */
 495         vma->vm_ops = &bpf_map_default_vmops;
 496         vma->vm_private_data = map;
 497
 498         err = map->ops->map_mmap(map, vma);
 499         if (err)
 500                 goto out;
 501
 502         bpf_map_inc_with_uref(map);
 503
 504         if (vma->vm_flags & VM_WRITE)
 505                 map->writecnt++;
 506 out:
 507         mutex_unlock(&map->freeze_mutex);
 508         return err;
 509 }
 510
 511 const struct file_operations bpf_map_fops = {
 512 #ifdef CONFIG_PROC_FS
 513         .show_fdinfo    = bpf_map_show_fdinfo,
 514 #endif
 515         .release        = bpf_map_release,
 516         .read           = bpf_dummy_read,
 517         .write          = bpf_dummy_write,
 518         .mmap           = bpf_map_mmap,
 519 };
 520
 521 int bpf_map_new_fd(struct bpf_map *map, int flags)
 522 {
 523         int ret;
 524
 525         ret = security_bpf_map(map, OPEN_FMODE(flags));
 526         if (ret < 0)
 527                 return ret;
 528
 529         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 530                                 flags | O_CLOEXEC);
 531 }
 532
 533 int bpf_get_file_flag(int flags)
 534 {
 535         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 536                 return -EINVAL;
 537         if (flags & BPF_F_RDONLY)
 538                 return O_RDONLY;
 539         if (flags & BPF_F_WRONLY)
 540                 return O_WRONLY;
 541         return O_RDWR;
 542 }
 543
 544 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 545 #define CHECK_ATTR(CMD) \
 546         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 547                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 548                    sizeof(*attr) - \
 549                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 550                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 551
 552 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 553  * Return 0 on success and < 0 on error.
 554  */
 555 static int bpf_obj_name_cpy(char *dst, const char *src)
 556 {
 557         const char *end = src + BPF_OBJ_NAME_LEN;
 558
 559         memset(dst, 0, BPF_OBJ_NAME_LEN);
 560         /* Copy all isalnum(), '_' and '.' chars. */
 561         while (src < end && *src) {
 562                 if (!isalnum(*src) &&
 563                     *src != '_' && *src != '.')
 564                         return -EINVAL;
 565                 *dst++ = *src++;
 566         }
 567
 568         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 569         if (src == end)
 570                 return -EINVAL;
 571
 572         return 0;
 573 }
 574
 575 int map_check_no_btf(const struct bpf_map *map,
 576                      const struct btf *btf,
 577                      const struct btf_type *key_type,
 578                      const struct btf_type *value_type)
 579 {
 580         return -ENOTSUPP;
 581 }
 582
 583 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 584                          u32 btf_key_id, u32 btf_value_id)
 585 {
 586         const struct btf_type *key_type, *value_type;
 587         u32 key_size, value_size;
 588         int ret = 0;
 589
 590         /* Some maps allow key to be unspecified. */
 591         if (btf_key_id) {
 592                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 593                 if (!key_type || key_size != map->key_size)
 594                         return -EINVAL;
 595         } else {
 596                 key_type = btf_type_by_id(btf, 0);
 597                 if (!map->ops->map_check_btf)
 598                         return -EINVAL;
 599         }
 600
 601         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 602         if (!value_type || value_size != map->value_size)
 603                 return -EINVAL;
 604
 605         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 606
 607         if (map_value_has_spin_lock(map)) {
 608                 if (map->map_flags & BPF_F_RDONLY_PROG)
 609                         return -EACCES;
 610                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 611                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 612                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 613                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 614                         return -ENOTSUPP;
 615                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 616                     map->value_size) {
 617                         WARN_ONCE(1,
 618                                   "verifier bug spin_lock_off %d value_size %d\n",
 619                                   map->spin_lock_off, map->value_size);
 620                         return -EFAULT;
 621                 }
 622         }
 623
 624         if (map->ops->map_check_btf)
 625                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 626
 627         return ret;
 628 }
 629
 630 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 631 /* called via syscall */
 632 static int map_create(union bpf_attr *attr)
 633 {
 634         int numa_node = bpf_map_attr_numa_node(attr);
 635         struct bpf_map_memory mem;
 636         struct bpf_map *map;
 637         int f_flags;
 638         int err;
 639
 640         err = CHECK_ATTR(BPF_MAP_CREATE);
 641         if (err)
 642                 return -EINVAL;
 643
 644         f_flags = bpf_get_file_flag(attr->map_flags);
 645         if (f_flags < 0)
 646                 return f_flags;
 647
 648         if (numa_node != NUMA_NO_NODE &&
 649             ((unsigned int)numa_node >= nr_node_ids ||
 650              !node_online(numa_node)))
 651                 return -EINVAL;
 652
 653         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 654         map = find_and_alloc_map(attr);
 655         if (IS_ERR(map))
 656                 return PTR_ERR(map);
 657
 658         err = bpf_obj_name_cpy(map->name, attr->map_name);
 659         if (err)
 660                 goto free_map;
 661
 662         atomic64_set(&map->refcnt, 1);
 663         atomic64_set(&map->usercnt, 1);
 664         mutex_init(&map->freeze_mutex);
 665
 666         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 667                 struct btf *btf;
 668
 669                 if (!attr->btf_value_type_id) {
 670                         err = -EINVAL;
 671                         goto free_map;
 672                 }
 673
 674                 btf = btf_get_by_fd(attr->btf_fd);
 675                 if (IS_ERR(btf)) {
 676                         err = PTR_ERR(btf);
 677                         goto free_map;
 678                 }
 679
 680                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 681                                     attr->btf_value_type_id);
 682                 if (err) {
 683                         btf_put(btf);
 684                         goto free_map;
 685                 }
 686
 687                 map->btf = btf;
 688                 map->btf_key_type_id = attr->btf_key_type_id;
 689                 map->btf_value_type_id = attr->btf_value_type_id;
 690         } else {
 691                 map->spin_lock_off = -EINVAL;
 692         }
 693
 694         err = security_bpf_map_alloc(map);
 695         if (err)
 696                 goto free_map;
 697
 698         err = bpf_map_alloc_id(map);
 699         if (err)
 700                 goto free_map_sec;
 701
 702         err = bpf_map_new_fd(map, f_flags);
 703         if (err < 0) {
 704                 /* failed to allocate fd.
 705                  * bpf_map_put_with_uref() is needed because the above
 706                  * bpf_map_alloc_id() has published the map
 707                  * to the userspace and the userspace may
 708                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 709                  */
 710                 bpf_map_put_with_uref(map);
 711                 return err;
 712         }
 713
 714         return err;
 715
 716 free_map_sec:
 717         security_bpf_map_free(map);
 718 free_map:
 719         btf_put(map->btf);
 720         bpf_map_charge_move(&mem, &map->memory);
 721         map->ops->map_free(map);
 722         bpf_map_charge_finish(&mem);
 723         return err;
 724 }
 725
 726 /* if error is returned, fd is released.
 727  * On success caller should complete fd access with matching fdput()
 728  */
 729 struct bpf_map *__bpf_map_get(struct fd f)
 730 {
 731         if (!f.file)
 732                 return ERR_PTR(-EBADF);
 733         if (f.file->f_op != &bpf_map_fops) {
 734                 fdput(f);
 735                 return ERR_PTR(-EINVAL);
 736         }
 737
 738         return f.file->private_data;
 739 }
 740
 741 void bpf_map_inc(struct bpf_map *map)
 742 {
 743         atomic64_inc(&map->refcnt);
 744 }
 745 EXPORT_SYMBOL_GPL(bpf_map_inc);
 746
 747 void bpf_map_inc_with_uref(struct bpf_map *map)
 748 {
 749         atomic64_inc(&map->refcnt);
 750         atomic64_inc(&map->usercnt);
 751 }
 752 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
 753
 754 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 755 {
 756         struct fd f = fdget(ufd);
 757         struct bpf_map *map;
 758
 759         map = __bpf_map_get(f);
 760         if (IS_ERR(map))
 761                 return map;
 762
 763         bpf_map_inc_with_uref(map);
 764         fdput(f);
 765
 766         return map;
 767 }
 768
 769 /* map_idr_lock should have been held */
 770 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 771 {
 772         int refold;
 773
 774         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
 775         if (!refold)
 776                 return ERR_PTR(-ENOENT);
 777         if (uref)
 778                 atomic64_inc(&map->usercnt);
 779
 780         return map;
 781 }
 782
 783 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
 784 {
 785         spin_lock_bh(&map_idr_lock);
 786         map = __bpf_map_inc_not_zero(map, false);
 787         spin_unlock_bh(&map_idr_lock);
 788
 789         return map;
 790 }
 791 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 792
 793 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 794 {
 795         return -ENOTSUPP;
 796 }
 797
 798 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 799 {
 800         if (key_size)
 801                 return memdup_user(ukey, key_size);
 802
 803         if (ukey)
 804                 return ERR_PTR(-EINVAL);
 805
 806         return NULL;
 807 }
 808
 809 /* last field in 'union bpf_attr' used by this command */
 810 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 811
 812 static int map_lookup_elem(union bpf_attr *attr)
 813 {
 814         void __user *ukey = u64_to_user_ptr(attr->key);
 815         void __user *uvalue = u64_to_user_ptr(attr->value);
 816         int ufd = attr->map_fd;
 817         struct bpf_map *map;
 818         void *key, *value, *ptr;
 819         u32 value_size;
 820         struct fd f;
 821         int err;
 822
 823         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 824                 return -EINVAL;
 825
 826         if (attr->flags & ~BPF_F_LOCK)
 827                 return -EINVAL;
 828
 829         f = fdget(ufd);
 830         map = __bpf_map_get(f);
 831         if (IS_ERR(map))
 832                 return PTR_ERR(map);
 833         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 834                 err = -EPERM;
 835                 goto err_put;
 836         }
 837
 838         if ((attr->flags & BPF_F_LOCK) &&
 839             !map_value_has_spin_lock(map)) {
 840                 err = -EINVAL;
 841                 goto err_put;
 842         }
 843
 844         key = __bpf_copy_key(ukey, map->key_size);
 845         if (IS_ERR(key)) {
 846                 err = PTR_ERR(key);
 847                 goto err_put;
 848         }
 849
 850         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 851             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 852             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 853             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 854                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 855         else if (IS_FD_MAP(map))
 856                 value_size = sizeof(u32);
 857         else
 858                 value_size = map->value_size;
 859
 860         err = -ENOMEM;
 861         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 862         if (!value)
 863                 goto free_key;
 864
 865         if (bpf_map_is_dev_bound(map)) {
 866                 err = bpf_map_offload_lookup_elem(map, key, value);
 867                 goto done;
 868         }
 869
 870         preempt_disable();
 871         this_cpu_inc(bpf_prog_active);
 872         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 873             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 874                 err = bpf_percpu_hash_copy(map, key, value);
 875         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 876                 err = bpf_percpu_array_copy(map, key, value);
 877         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 878                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 879         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 880                 err = bpf_stackmap_copy(map, key, value);
 881         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 882                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 883         } else if (IS_FD_HASH(map)) {
 884                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 885         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 886                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 887         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 888                    map->map_type == BPF_MAP_TYPE_STACK) {
 889                 err = map->ops->map_peek_elem(map, value);
 890         } else {
 891                 rcu_read_lock();
 892                 if (map->ops->map_lookup_elem_sys_only)
 893                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 894                 else
 895                         ptr = map->ops->map_lookup_elem(map, key);
 896                 if (IS_ERR(ptr)) {
 897                         err = PTR_ERR(ptr);
 898                 } else if (!ptr) {
 899                         err = -ENOENT;
 900                 } else {
 901                         err = 0;
 902                         if (attr->flags & BPF_F_LOCK)
 903                                 /* lock 'ptr' and copy everything but lock */
 904                                 copy_map_value_locked(map, value, ptr, true);
 905                         else
 906                                 copy_map_value(map, value, ptr);
 907                         /* mask lock, since value wasn't zero inited */
 908                         check_and_init_map_lock(map, value);
 909                 }
 910                 rcu_read_unlock();
 911         }
 912         this_cpu_dec(bpf_prog_active);
 913         preempt_enable();
 914
 915 done:
 916         if (err)
 917                 goto free_value;
 918
 919         err = -EFAULT;
 920         if (copy_to_user(uvalue, value, value_size) != 0)
 921                 goto free_value;
 922
 923         err = 0;
 924
 925 free_value:
 926         kfree(value);
 927 free_key:
 928         kfree(key);
 929 err_put:
 930         fdput(f);
 931         return err;
 932 }
 933
 934 static void maybe_wait_bpf_programs(struct bpf_map *map)
 935 {
 936         /* Wait for any running BPF programs to complete so that
 937          * userspace, when we return to it, knows that all programs
 938          * that could be running use the new map value.
 939          */
 940         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 941             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 942                 synchronize_rcu();
 943 }
 944
 945 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 946
 947 static int map_update_elem(union bpf_attr *attr)
 948 {
 949         void __user *ukey = u64_to_user_ptr(attr->key);
 950         void __user *uvalue = u64_to_user_ptr(attr->value);
 951         int ufd = attr->map_fd;
 952         struct bpf_map *map;
 953         void *key, *value;
 954         u32 value_size;
 955         struct fd f;
 956         int err;
 957
 958         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 959                 return -EINVAL;
 960
 961         f = fdget(ufd);
 962         map = __bpf_map_get(f);
 963         if (IS_ERR(map))
 964                 return PTR_ERR(map);
 965         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 966                 err = -EPERM;
 967                 goto err_put;
 968         }
 969
 970         if ((attr->flags & BPF_F_LOCK) &&
 971             !map_value_has_spin_lock(map)) {
 972                 err = -EINVAL;
 973                 goto err_put;
 974         }
 975
 976         key = __bpf_copy_key(ukey, map->key_size);
 977         if (IS_ERR(key)) {
 978                 err = PTR_ERR(key);
 979                 goto err_put;
 980         }
 981
 982         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 983             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 984             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 985             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 986                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 987         else
 988                 value_size = map->value_size;
 989
 990         err = -ENOMEM;
 991         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 992         if (!value)
 993                 goto free_key;
 994
 995         err = -EFAULT;
 996         if (copy_from_user(value, uvalue, value_size) != 0)
 997                 goto free_value;
 998
 999         /* Need to create a kthread, thus must support schedule */
1000         if (bpf_map_is_dev_bound(map)) {
1001                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
1002                 goto out;
1003         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
1004                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
1005                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
1006                 err = map->ops->map_update_elem(map, key, value, attr->flags);
1007                 goto out;
1008         } else if (IS_FD_PROG_ARRAY(map)) {
1009                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1010                                                    attr->flags);
1011                 goto out;
1012         }
1013
1014         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
1015          * inside bpf map update or delete otherwise deadlocks are possible
1016          */
1017         preempt_disable();
1018         __this_cpu_inc(bpf_prog_active);
1019         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1020             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1021                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
1022         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
1023                 err = bpf_percpu_array_update(map, key, value, attr->flags);
1024         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
1025                 err = bpf_percpu_cgroup_storage_update(map, key, value,
1026                                                        attr->flags);
1027         } else if (IS_FD_ARRAY(map)) {
1028                 rcu_read_lock();
1029                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1030                                                    attr->flags);
1031                 rcu_read_unlock();
1032         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1033                 rcu_read_lock();
1034                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
1035                                                   attr->flags);
1036                 rcu_read_unlock();
1037         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
1038                 /* rcu_read_lock() is not needed */
1039                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
1040                                                          attr->flags);
1041         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1042                    map->map_type == BPF_MAP_TYPE_STACK) {
1043                 err = map->ops->map_push_elem(map, value, attr->flags);
1044         } else {
1045                 rcu_read_lock();
1046                 err = map->ops->map_update_elem(map, key, value, attr->flags);
1047                 rcu_read_unlock();
1048         }
1049         __this_cpu_dec(bpf_prog_active);
1050         preempt_enable();
1051         maybe_wait_bpf_programs(map);
1052 out:
1053 free_value:
1054         kfree(value);
1055 free_key:
1056         kfree(key);
1057 err_put:
1058         fdput(f);
1059         return err;
1060 }
1061
1062 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1063
1064 static int map_delete_elem(union bpf_attr *attr)
1065 {
1066         void __user *ukey = u64_to_user_ptr(attr->key);
1067         int ufd = attr->map_fd;
1068         struct bpf_map *map;
1069         struct fd f;
1070         void *key;
1071         int err;
1072
1073         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1074                 return -EINVAL;
1075
1076         f = fdget(ufd);
1077         map = __bpf_map_get(f);
1078         if (IS_ERR(map))
1079                 return PTR_ERR(map);
1080         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1081                 err = -EPERM;
1082                 goto err_put;
1083         }
1084
1085         key = __bpf_copy_key(ukey, map->key_size);
1086         if (IS_ERR(key)) {
1087                 err = PTR_ERR(key);
1088                 goto err_put;
1089         }
1090
1091         if (bpf_map_is_dev_bound(map)) {
1092                 err = bpf_map_offload_delete_elem(map, key);
1093                 goto out;
1094         } else if (IS_FD_PROG_ARRAY(map)) {
1095                 err = map->ops->map_delete_elem(map, key);
1096                 goto out;
1097         }
1098
1099         preempt_disable();
1100         __this_cpu_inc(bpf_prog_active);
1101         rcu_read_lock();
1102         err = map->ops->map_delete_elem(map, key);
1103         rcu_read_unlock();
1104         __this_cpu_dec(bpf_prog_active);
1105         preempt_enable();
1106         maybe_wait_bpf_programs(map);
1107 out:
1108         kfree(key);
1109 err_put:
1110         fdput(f);
1111         return err;
1112 }
1113
1114 /* last field in 'union bpf_attr' used by this command */
1115 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1116
1117 static int map_get_next_key(union bpf_attr *attr)
1118 {
1119         void __user *ukey = u64_to_user_ptr(attr->key);
1120         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1121         int ufd = attr->map_fd;
1122         struct bpf_map *map;
1123         void *key, *next_key;
1124         struct fd f;
1125         int err;
1126
1127         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1128                 return -EINVAL;
1129
1130         f = fdget(ufd);
1131         map = __bpf_map_get(f);
1132         if (IS_ERR(map))
1133                 return PTR_ERR(map);
1134         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1135                 err = -EPERM;
1136                 goto err_put;
1137         }
1138
1139         if (ukey) {
1140                 key = __bpf_copy_key(ukey, map->key_size);
1141                 if (IS_ERR(key)) {
1142                         err = PTR_ERR(key);
1143                         goto err_put;
1144                 }
1145         } else {
1146                 key = NULL;
1147         }
1148
1149         err = -ENOMEM;
1150         next_key = kmalloc(map->key_size, GFP_USER);
1151         if (!next_key)
1152                 goto free_key;
1153
1154         if (bpf_map_is_dev_bound(map)) {
1155                 err = bpf_map_offload_get_next_key(map, key, next_key);
1156                 goto out;
1157         }
1158
1159         rcu_read_lock();
1160         err = map->ops->map_get_next_key(map, key, next_key);
1161         rcu_read_unlock();
1162 out:
1163         if (err)
1164                 goto free_next_key;
1165
1166         err = -EFAULT;
1167         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1168                 goto free_next_key;
1169
1170         err = 0;
1171
1172 free_next_key:
1173         kfree(next_key);
1174 free_key:
1175         kfree(key);
1176 err_put:
1177         fdput(f);
1178         return err;
1179 }
1180
1181 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1182
1183 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1184 {
1185         void __user *ukey = u64_to_user_ptr(attr->key);
1186         void __user *uvalue = u64_to_user_ptr(attr->value);
1187         int ufd = attr->map_fd;
1188         struct bpf_map *map;
1189         void *key, *value;
1190         u32 value_size;
1191         struct fd f;
1192         int err;
1193
1194         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1195                 return -EINVAL;
1196
1197         f = fdget(ufd);
1198         map = __bpf_map_get(f);
1199         if (IS_ERR(map))
1200                 return PTR_ERR(map);
1201         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1202                 err = -EPERM;
1203                 goto err_put;
1204         }
1205
1206         key = __bpf_copy_key(ukey, map->key_size);
1207         if (IS_ERR(key)) {
1208                 err = PTR_ERR(key);
1209                 goto err_put;
1210         }
1211
1212         value_size = map->value_size;
1213
1214         err = -ENOMEM;
1215         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1216         if (!value)
1217                 goto free_key;
1218
1219         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1220             map->map_type == BPF_MAP_TYPE_STACK) {
1221                 err = map->ops->map_pop_elem(map, value);
1222         } else {
1223                 err = -ENOTSUPP;
1224         }
1225
1226         if (err)
1227                 goto free_value;
1228
1229         if (copy_to_user(uvalue, value, value_size) != 0)
1230                 goto free_value;
1231
1232         err = 0;
1233
1234 free_value:
1235         kfree(value);
1236 free_key:
1237         kfree(key);
1238 err_put:
1239         fdput(f);
1240         return err;
1241 }
1242
1243 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1244
1245 static int map_freeze(const union bpf_attr *attr)
1246 {
1247         int err = 0, ufd = attr->map_fd;
1248         struct bpf_map *map;
1249         struct fd f;
1250
1251         if (CHECK_ATTR(BPF_MAP_FREEZE))
1252                 return -EINVAL;
1253
1254         f = fdget(ufd);
1255         map = __bpf_map_get(f);
1256         if (IS_ERR(map))
1257                 return PTR_ERR(map);
1258
1259         mutex_lock(&map->freeze_mutex);
1260
1261         if (map->writecnt) {
1262                 err = -EBUSY;
1263                 goto err_put;
1264         }
1265         if (READ_ONCE(map->frozen)) {
1266                 err = -EBUSY;
1267                 goto err_put;
1268         }
1269         if (!capable(CAP_SYS_ADMIN)) {
1270                 err = -EPERM;
1271                 goto err_put;
1272         }
1273
1274         WRITE_ONCE(map->frozen, true);
1275 err_put:
1276         mutex_unlock(&map->freeze_mutex);
1277         fdput(f);
1278         return err;
1279 }
1280
1281 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1282 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1283         [_id] = & _name ## _prog_ops,
1284 #define BPF_MAP_TYPE(_id, _ops)
1285 #include <linux/bpf_types.h>
1286 #undef BPF_PROG_TYPE
1287 #undef BPF_MAP_TYPE
1288 };
1289
1290 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1291 {
1292         const struct bpf_prog_ops *ops;
1293
1294         if (type >= ARRAY_SIZE(bpf_prog_types))
1295                 return -EINVAL;
1296         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1297         ops = bpf_prog_types[type];
1298         if (!ops)
1299                 return -EINVAL;
1300
1301         if (!bpf_prog_is_dev_bound(prog->aux))
1302                 prog->aux->ops = ops;
1303         else
1304                 prog->aux->ops = &bpf_offload_prog_ops;
1305         prog->type = type;
1306         return 0;
1307 }
1308
1309 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1310 {
1311         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1312         unsigned long user_bufs;
1313
1314         if (user) {
1315                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1316                 if (user_bufs > memlock_limit) {
1317                         atomic_long_sub(pages, &user->locked_vm);
1318                         return -EPERM;
1319                 }
1320         }
1321
1322         return 0;
1323 }
1324
1325 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1326 {
1327         if (user)
1328                 atomic_long_sub(pages, &user->locked_vm);
1329 }
1330
1331 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1332 {
1333         struct user_struct *user = get_current_user();
1334         int ret;
1335
1336         ret = __bpf_prog_charge(user, prog->pages);
1337         if (ret) {
1338                 free_uid(user);
1339                 return ret;
1340         }
1341
1342         prog->aux->user = user;
1343         return 0;
1344 }
1345
1346 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1347 {
1348         struct user_struct *user = prog->aux->user;
1349
1350         __bpf_prog_uncharge(user, prog->pages);
1351         free_uid(user);
1352 }
1353
1354 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1355 {
1356         int id;
1357
1358         idr_preload(GFP_KERNEL);
1359         spin_lock_bh(&prog_idr_lock);
1360         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1361         if (id > 0)
1362                 prog->aux->id = id;
1363         spin_unlock_bh(&prog_idr_lock);
1364         idr_preload_end();
1365
1366         /* id is in [1, INT_MAX) */
1367         if (WARN_ON_ONCE(!id))
1368                 return -ENOSPC;
1369
1370         return id > 0 ? 0 : id;
1371 }
1372
1373 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1374 {
1375         /* cBPF to eBPF migrations are currently not in the idr store.
1376          * Offloaded programs are removed from the store when their device
1377          * disappears - even if someone grabs an fd to them they are unusable,
1378          * simply waiting for refcnt to drop to be freed.
1379          */
1380         if (!prog->aux->id)
1381                 return;
1382
1383         if (do_idr_lock)
1384                 spin_lock_bh(&prog_idr_lock);
1385         else
1386                 __acquire(&prog_idr_lock);
1387
1388         idr_remove(&prog_idr, prog->aux->id);
1389         prog->aux->id = 0;
1390
1391         if (do_idr_lock)
1392                 spin_unlock_bh(&prog_idr_lock);
1393         else
1394                 __release(&prog_idr_lock);
1395 }
1396
1397 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1398 {
1399         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1400
1401         kvfree(aux->func_info);
1402         kfree(aux->func_info_aux);
1403         bpf_prog_uncharge_memlock(aux->prog);
1404         security_bpf_prog_free(aux);
1405         bpf_prog_free(aux->prog);
1406 }
1407
1408 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1409 {
1410         bpf_prog_kallsyms_del_all(prog);
1411         btf_put(prog->aux->btf);
1412         bpf_prog_free_linfo(prog);
1413
1414         if (deferred)
1415                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1416         else
1417                 __bpf_prog_put_rcu(&prog->aux->rcu);
1418 }
1419
1420 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1421 {
1422         if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1423                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1424                 /* bpf_prog_free_id() must be called first */
1425                 bpf_prog_free_id(prog, do_idr_lock);
1426                 __bpf_prog_put_noref(prog, true);
1427         }
1428 }
1429
1430 void bpf_prog_put(struct bpf_prog *prog)
1431 {
1432         __bpf_prog_put(prog, true);
1433 }
1434 EXPORT_SYMBOL_GPL(bpf_prog_put);
1435
1436 static int bpf_prog_release(struct inode *inode, struct file *filp)
1437 {
1438         struct bpf_prog *prog = filp->private_data;
1439
1440         bpf_prog_put(prog);
1441         return 0;
1442 }
1443
1444 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1445                                struct bpf_prog_stats *stats)
1446 {
1447         u64 nsecs = 0, cnt = 0;
1448         int cpu;
1449
1450         for_each_possible_cpu(cpu) {
1451                 const struct bpf_prog_stats *st;
1452                 unsigned int start;
1453                 u64 tnsecs, tcnt;
1454
1455                 st = per_cpu_ptr(prog->aux->stats, cpu);
1456                 do {
1457                         start = u64_stats_fetch_begin_irq(&st->syncp);
1458                         tnsecs = st->nsecs;
1459                         tcnt = st->cnt;
1460                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1461                 nsecs += tnsecs;
1462                 cnt += tcnt;
1463         }
1464         stats->nsecs = nsecs;
1465         stats->cnt = cnt;
1466 }
1467
1468 #ifdef CONFIG_PROC_FS
1469 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1470 {
1471         const struct bpf_prog *prog = filp->private_data;
1472         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1473         struct bpf_prog_stats stats;
1474
1475         bpf_prog_get_stats(prog, &stats);
1476         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1477         seq_printf(m,
1478                    "prog_type:\t%u\n"
1479                    "prog_jited:\t%u\n"
1480                    "prog_tag:\t%s\n"
1481                    "memlock:\t%llu\n"
1482                    "prog_id:\t%u\n"
1483                    "run_time_ns:\t%llu\n"
1484                    "run_cnt:\t%llu\n",
1485                    prog->type,
1486                    prog->jited,
1487                    prog_tag,
1488                    prog->pages * 1ULL << PAGE_SHIFT,
1489                    prog->aux->id,
1490                    stats.nsecs,
1491                    stats.cnt);
1492 }
1493 #endif
1494
1495 const struct file_operations bpf_prog_fops = {
1496 #ifdef CONFIG_PROC_FS
1497         .show_fdinfo    = bpf_prog_show_fdinfo,
1498 #endif
1499         .release        = bpf_prog_release,
1500         .read           = bpf_dummy_read,
1501         .write          = bpf_dummy_write,
1502 };
1503
1504 int bpf_prog_new_fd(struct bpf_prog *prog)
1505 {
1506         int ret;
1507
1508         ret = security_bpf_prog(prog);
1509         if (ret < 0)
1510                 return ret;
1511
1512         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1513                                 O_RDWR | O_CLOEXEC);
1514 }
1515
1516 static struct bpf_prog *____bpf_prog_get(struct fd f)
1517 {
1518         if (!f.file)
1519                 return ERR_PTR(-EBADF);
1520         if (f.file->f_op != &bpf_prog_fops) {
1521                 fdput(f);
1522                 return ERR_PTR(-EINVAL);
1523         }
1524
1525         return f.file->private_data;
1526 }
1527
1528 void bpf_prog_add(struct bpf_prog *prog, int i)
1529 {
1530         atomic64_add(i, &prog->aux->refcnt);
1531 }
1532 EXPORT_SYMBOL_GPL(bpf_prog_add);
1533
1534 void bpf_prog_sub(struct bpf_prog *prog, int i)
1535 {
1536         /* Only to be used for undoing previous bpf_prog_add() in some
1537          * error path. We still know that another entity in our call
1538          * path holds a reference to the program, thus atomic_sub() can
1539          * be safely used in such cases!
1540          */
1541         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1542 }
1543 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1544
1545 void bpf_prog_inc(struct bpf_prog *prog)
1546 {
1547         atomic64_inc(&prog->aux->refcnt);
1548 }
1549 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1550
1551 /* prog_idr_lock should have been held */
1552 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1553 {
1554         int refold;
1555
1556         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1557
1558         if (!refold)
1559                 return ERR_PTR(-ENOENT);
1560
1561         return prog;
1562 }
1563 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1564
1565 bool bpf_prog_get_ok(struct bpf_prog *prog,
1566                             enum bpf_prog_type *attach_type, bool attach_drv)
1567 {
1568         /* not an attachment, just a refcount inc, always allow */
1569         if (!attach_type)
1570                 return true;
1571
1572         if (prog->type != *attach_type)
1573                 return false;
1574         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1575                 return false;
1576
1577         return true;
1578 }
1579
1580 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1581                                        bool attach_drv)
1582 {
1583         struct fd f = fdget(ufd);
1584         struct bpf_prog *prog;
1585
1586         prog = ____bpf_prog_get(f);
1587         if (IS_ERR(prog))
1588                 return prog;
1589         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1590                 prog = ERR_PTR(-EINVAL);
1591                 goto out;
1592         }
1593
1594         bpf_prog_inc(prog);
1595 out:
1596         fdput(f);
1597         return prog;
1598 }
1599
1600 struct bpf_prog *bpf_prog_get(u32 ufd)
1601 {
1602         return __bpf_prog_get(ufd, NULL, false);
1603 }
1604
1605 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1606                                        bool attach_drv)
1607 {
1608         return __bpf_prog_get(ufd, &type, attach_drv);
1609 }
1610 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1611
1612 /* Initially all BPF programs could be loaded w/o specifying
1613  * expected_attach_type. Later for some of them specifying expected_attach_type
1614  * at load time became required so that program could be validated properly.
1615  * Programs of types that are allowed to be loaded both w/ and w/o (for
1616  * backward compatibility) expected_attach_type, should have the default attach
1617  * type assigned to expected_attach_type for the latter case, so that it can be
1618  * validated later at attach time.
1619  *
1620  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1621  * prog type requires it but has some attach types that have to be backward
1622  * compatible.
1623  */
1624 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1625 {
1626         switch (attr->prog_type) {
1627         case BPF_PROG_TYPE_CGROUP_SOCK:
1628                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1629                  * exist so checking for non-zero is the way to go here.
1630                  */
1631                 if (!attr->expected_attach_type)
1632                         attr->expected_attach_type =
1633                                 BPF_CGROUP_INET_SOCK_CREATE;
1634                 break;
1635         }
1636 }
1637
1638 static int
1639 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1640                            enum bpf_attach_type expected_attach_type,
1641                            u32 btf_id, u32 prog_fd)
1642 {
1643         switch (prog_type) {
1644         case BPF_PROG_TYPE_TRACING:
1645                 if (btf_id > BTF_MAX_TYPE)
1646                         return -EINVAL;
1647                 break;
1648         default:
1649                 if (btf_id || prog_fd)
1650                         return -EINVAL;
1651                 break;
1652         }
1653
1654         switch (prog_type) {
1655         case BPF_PROG_TYPE_CGROUP_SOCK:
1656                 switch (expected_attach_type) {
1657                 case BPF_CGROUP_INET_SOCK_CREATE:
1658                 case BPF_CGROUP_INET4_POST_BIND:
1659                 case BPF_CGROUP_INET6_POST_BIND:
1660                         return 0;
1661                 default:
1662                         return -EINVAL;
1663                 }
1664         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1665                 switch (expected_attach_type) {
1666                 case BPF_CGROUP_INET4_BIND:
1667                 case BPF_CGROUP_INET6_BIND:
1668                 case BPF_CGROUP_INET4_CONNECT:
1669                 case BPF_CGROUP_INET6_CONNECT:
1670                 case BPF_CGROUP_UDP4_SENDMSG:
1671                 case BPF_CGROUP_UDP6_SENDMSG:
1672                 case BPF_CGROUP_UDP4_RECVMSG:
1673                 case BPF_CGROUP_UDP6_RECVMSG:
1674                         return 0;
1675                 default:
1676                         return -EINVAL;
1677                 }
1678         case BPF_PROG_TYPE_CGROUP_SKB:
1679                 switch (expected_attach_type) {
1680                 case BPF_CGROUP_INET_INGRESS:
1681                 case BPF_CGROUP_INET_EGRESS:
1682                         return 0;
1683                 default:
1684                         return -EINVAL;
1685                 }
1686         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1687                 switch (expected_attach_type) {
1688                 case BPF_CGROUP_SETSOCKOPT:
1689                 case BPF_CGROUP_GETSOCKOPT:
1690                         return 0;
1691                 default:
1692                         return -EINVAL;
1693                 }
1694         default:
1695                 return 0;
1696         }
1697 }
1698
1699 /* last field in 'union bpf_attr' used by this command */
1700 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
1701
1702 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1703 {
1704         enum bpf_prog_type type = attr->prog_type;
1705         struct bpf_prog *prog;
1706         int err;
1707         char license[128];
1708         bool is_gpl;
1709
1710         if (CHECK_ATTR(BPF_PROG_LOAD))
1711                 return -EINVAL;
1712
1713         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1714                                  BPF_F_ANY_ALIGNMENT |
1715                                  BPF_F_TEST_STATE_FREQ |
1716                                  BPF_F_TEST_RND_HI32))
1717                 return -EINVAL;
1718
1719         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1720             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1721             !capable(CAP_SYS_ADMIN))
1722                 return -EPERM;
1723
1724         /* copy eBPF program license from user space */
1725         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1726                               sizeof(license) - 1) < 0)
1727                 return -EFAULT;
1728         license[sizeof(license) - 1] = 0;
1729
1730         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1731         is_gpl = license_is_gpl_compatible(license);
1732
1733         if (attr->insn_cnt == 0 ||
1734             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1735                 return -E2BIG;
1736         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1737             type != BPF_PROG_TYPE_CGROUP_SKB &&
1738             !capable(CAP_SYS_ADMIN))
1739                 return -EPERM;
1740
1741         bpf_prog_load_fixup_attach_type(attr);
1742         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1743                                        attr->attach_btf_id,
1744                                        attr->attach_prog_fd))
1745                 return -EINVAL;
1746
1747         /* plain bpf_prog allocation */
1748         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1749         if (!prog)
1750                 return -ENOMEM;
1751
1752         prog->expected_attach_type = attr->expected_attach_type;
1753         prog->aux->attach_btf_id = attr->attach_btf_id;
1754         if (attr->attach_prog_fd) {
1755                 struct bpf_prog *tgt_prog;
1756
1757                 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
1758                 if (IS_ERR(tgt_prog)) {
1759                         err = PTR_ERR(tgt_prog);
1760                         goto free_prog_nouncharge;
1761                 }
1762                 prog->aux->linked_prog = tgt_prog;
1763         }
1764
1765         prog->aux->offload_requested = !!attr->prog_ifindex;
1766
1767         err = security_bpf_prog_alloc(prog->aux);
1768         if (err)
1769                 goto free_prog_nouncharge;
1770
1771         err = bpf_prog_charge_memlock(prog);
1772         if (err)
1773                 goto free_prog_sec;
1774
1775         prog->len = attr->insn_cnt;
1776
1777         err = -EFAULT;
1778         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1779                            bpf_prog_insn_size(prog)) != 0)
1780                 goto free_prog;
1781
1782         prog->orig_prog = NULL;
1783         prog->jited = 0;
1784
1785         atomic64_set(&prog->aux->refcnt, 1);
1786         prog->gpl_compatible = is_gpl ? 1 : 0;
1787
1788         if (bpf_prog_is_dev_bound(prog->aux)) {
1789                 err = bpf_prog_offload_init(prog, attr);
1790                 if (err)
1791                         goto free_prog;
1792         }
1793
1794         /* find program type: socket_filter vs tracing_filter */
1795         err = find_prog_type(type, prog);
1796         if (err < 0)
1797                 goto free_prog;
1798
1799         prog->aux->load_time = ktime_get_boottime_ns();
1800         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1801         if (err)
1802                 goto free_prog;
1803
1804         /* run eBPF verifier */
1805         err = bpf_check(&prog, attr, uattr);
1806         if (err < 0)
1807                 goto free_used_maps;
1808
1809         prog = bpf_prog_select_runtime(prog, &err);
1810         if (err < 0)
1811                 goto free_used_maps;
1812
1813         err = bpf_prog_alloc_id(prog);
1814         if (err)
1815                 goto free_used_maps;
1816
1817         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1818          * effectively publicly exposed. However, retrieving via
1819          * bpf_prog_get_fd_by_id() will take another reference,
1820          * therefore it cannot be gone underneath us.
1821          *
1822          * Only for the time /after/ successful bpf_prog_new_fd()
1823          * and before returning to userspace, we might just hold
1824          * one reference and any parallel close on that fd could
1825          * rip everything out. Hence, below notifications must
1826          * happen before bpf_prog_new_fd().
1827          *
1828          * Also, any failure handling from this point onwards must
1829          * be using bpf_prog_put() given the program is exposed.
1830          */
1831         bpf_prog_kallsyms_add(prog);
1832         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1833
1834         err = bpf_prog_new_fd(prog);
1835         if (err < 0)
1836                 bpf_prog_put(prog);
1837         return err;
1838
1839 free_used_maps:
1840         /* In case we have subprogs, we need to wait for a grace
1841          * period before we can tear down JIT memory since symbols
1842          * are already exposed under kallsyms.
1843          */
1844         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1845         return err;
1846 free_prog:
1847         bpf_prog_uncharge_memlock(prog);
1848 free_prog_sec:
1849         security_bpf_prog_free(prog->aux);
1850 free_prog_nouncharge:
1851         bpf_prog_free(prog);
1852         return err;
1853 }
1854
1855 #define BPF_OBJ_LAST_FIELD file_flags
1856
1857 static int bpf_obj_pin(const union bpf_attr *attr)
1858 {
1859         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1860                 return -EINVAL;
1861
1862         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1863 }
1864
1865 static int bpf_obj_get(const union bpf_attr *attr)
1866 {
1867         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1868             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1869                 return -EINVAL;
1870
1871         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1872                                 attr->file_flags);
1873 }
1874
1875 static int bpf_tracing_prog_release(struct inode *inode, struct file *filp)
1876 {
1877         struct bpf_prog *prog = filp->private_data;
1878
1879         WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1880         bpf_prog_put(prog);
1881         return 0;
1882 }
1883
1884 static const struct file_operations bpf_tracing_prog_fops = {
1885         .release        = bpf_tracing_prog_release,
1886         .read           = bpf_dummy_read,
1887         .write          = bpf_dummy_write,
1888 };
1889
1890 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
1891 {
1892         int tr_fd, err;
1893
1894         if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
1895             prog->expected_attach_type != BPF_TRACE_FEXIT) {
1896                 err = -EINVAL;
1897                 goto out_put_prog;
1898         }
1899
1900         err = bpf_trampoline_link_prog(prog);
1901         if (err)
1902                 goto out_put_prog;
1903
1904         tr_fd = anon_inode_getfd("bpf-tracing-prog", &bpf_tracing_prog_fops,
1905                                  prog, O_CLOEXEC);
1906         if (tr_fd < 0) {
1907                 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1908                 err = tr_fd;
1909                 goto out_put_prog;
1910         }
1911         return tr_fd;
1912
1913 out_put_prog:
1914         bpf_prog_put(prog);
1915         return err;
1916 }
1917
1918 struct bpf_raw_tracepoint {
1919         struct bpf_raw_event_map *btp;
1920         struct bpf_prog *prog;
1921 };
1922
1923 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1924 {
1925         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1926
1927         if (raw_tp->prog) {
1928                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1929                 bpf_prog_put(raw_tp->prog);
1930         }
1931         bpf_put_raw_tracepoint(raw_tp->btp);
1932         kfree(raw_tp);
1933         return 0;
1934 }
1935
1936 static const struct file_operations bpf_raw_tp_fops = {
1937         .release        = bpf_raw_tracepoint_release,
1938         .read           = bpf_dummy_read,
1939         .write          = bpf_dummy_write,
1940 };
1941
1942 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1943
1944 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1945 {
1946         struct bpf_raw_tracepoint *raw_tp;
1947         struct bpf_raw_event_map *btp;
1948         struct bpf_prog *prog;
1949         const char *tp_name;
1950         char buf[128];
1951         int tp_fd, err;
1952
1953         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1954                 return -EINVAL;
1955
1956         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1957         if (IS_ERR(prog))
1958                 return PTR_ERR(prog);
1959
1960         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1961             prog->type != BPF_PROG_TYPE_TRACING &&
1962             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1963                 err = -EINVAL;
1964                 goto out_put_prog;
1965         }
1966
1967         if (prog->type == BPF_PROG_TYPE_TRACING) {
1968                 if (attr->raw_tracepoint.name) {
1969                         /* The attach point for this category of programs
1970                          * should be specified via btf_id during program load.
1971                          */
1972                         err = -EINVAL;
1973                         goto out_put_prog;
1974                 }
1975                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
1976                         tp_name = prog->aux->attach_func_name;
1977                 else
1978                         return bpf_tracing_prog_attach(prog);
1979         } else {
1980                 if (strncpy_from_user(buf,
1981                                       u64_to_user_ptr(attr->raw_tracepoint.name),
1982                                       sizeof(buf) - 1) < 0) {
1983                         err = -EFAULT;
1984                         goto out_put_prog;
1985                 }
1986                 buf[sizeof(buf) - 1] = 0;
1987                 tp_name = buf;
1988         }
1989
1990         btp = bpf_get_raw_tracepoint(tp_name);
1991         if (!btp) {
1992                 err = -ENOENT;
1993                 goto out_put_prog;
1994         }
1995
1996         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1997         if (!raw_tp) {
1998                 err = -ENOMEM;
1999                 goto out_put_btp;
2000         }
2001         raw_tp->btp = btp;
2002         raw_tp->prog = prog;
2003
2004         err = bpf_probe_register(raw_tp->btp, prog);
2005         if (err)
2006                 goto out_free_tp;
2007
2008         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
2009                                  O_CLOEXEC);
2010         if (tp_fd < 0) {
2011                 bpf_probe_unregister(raw_tp->btp, prog);
2012                 err = tp_fd;
2013                 goto out_free_tp;
2014         }
2015         return tp_fd;
2016
2017 out_free_tp:
2018         kfree(raw_tp);
2019 out_put_btp:
2020         bpf_put_raw_tracepoint(btp);
2021 out_put_prog:
2022         bpf_prog_put(prog);
2023         return err;
2024 }
2025
2026 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2027                                              enum bpf_attach_type attach_type)
2028 {
2029         switch (prog->type) {
2030         case BPF_PROG_TYPE_CGROUP_SOCK:
2031         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2032         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2033                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2034         case BPF_PROG_TYPE_CGROUP_SKB:
2035                 return prog->enforce_expected_attach_type &&
2036                         prog->expected_attach_type != attach_type ?
2037                         -EINVAL : 0;
2038         default:
2039                 return 0;
2040         }
2041 }
2042
2043 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
2044
2045 #define BPF_F_ATTACH_MASK \
2046         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
2047
2048 static int bpf_prog_attach(const union bpf_attr *attr)
2049 {
2050         enum bpf_prog_type ptype;
2051         struct bpf_prog *prog;
2052         int ret;
2053
2054         if (!capable(CAP_NET_ADMIN))
2055                 return -EPERM;
2056
2057         if (CHECK_ATTR(BPF_PROG_ATTACH))
2058                 return -EINVAL;
2059
2060         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2061                 return -EINVAL;
2062
2063         switch (attr->attach_type) {
2064         case BPF_CGROUP_INET_INGRESS:
2065         case BPF_CGROUP_INET_EGRESS:
2066                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2067                 break;
2068         case BPF_CGROUP_INET_SOCK_CREATE:
2069         case BPF_CGROUP_INET4_POST_BIND:
2070         case BPF_CGROUP_INET6_POST_BIND:
2071                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2072                 break;
2073         case BPF_CGROUP_INET4_BIND:
2074         case BPF_CGROUP_INET6_BIND:
2075         case BPF_CGROUP_INET4_CONNECT:
2076         case BPF_CGROUP_INET6_CONNECT:
2077         case BPF_CGROUP_UDP4_SENDMSG:
2078         case BPF_CGROUP_UDP6_SENDMSG:
2079         case BPF_CGROUP_UDP4_RECVMSG:
2080         case BPF_CGROUP_UDP6_RECVMSG:
2081                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2082                 break;
2083         case BPF_CGROUP_SOCK_OPS:
2084                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2085                 break;
2086         case BPF_CGROUP_DEVICE:
2087                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2088                 break;
2089         case BPF_SK_MSG_VERDICT:
2090                 ptype = BPF_PROG_TYPE_SK_MSG;
2091                 break;
2092         case BPF_SK_SKB_STREAM_PARSER:
2093         case BPF_SK_SKB_STREAM_VERDICT:
2094                 ptype = BPF_PROG_TYPE_SK_SKB;
2095                 break;
2096         case BPF_LIRC_MODE2:
2097                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
2098                 break;
2099         case BPF_FLOW_DISSECTOR:
2100                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2101                 break;
2102         case BPF_CGROUP_SYSCTL:
2103                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2104                 break;
2105         case BPF_CGROUP_GETSOCKOPT:
2106         case BPF_CGROUP_SETSOCKOPT:
2107                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2108                 break;
2109         default:
2110                 return -EINVAL;
2111         }
2112
2113         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2114         if (IS_ERR(prog))
2115                 return PTR_ERR(prog);
2116
2117         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2118                 bpf_prog_put(prog);
2119                 return -EINVAL;
2120         }
2121
2122         switch (ptype) {
2123         case BPF_PROG_TYPE_SK_SKB:
2124         case BPF_PROG_TYPE_SK_MSG:
2125                 ret = sock_map_get_from_fd(attr, prog);
2126                 break;
2127         case BPF_PROG_TYPE_LIRC_MODE2:
2128                 ret = lirc_prog_attach(attr, prog);
2129                 break;
2130         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2131                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2132                 break;
2133         default:
2134                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2135         }
2136
2137         if (ret)
2138                 bpf_prog_put(prog);
2139         return ret;
2140 }
2141
2142 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2143
2144 static int bpf_prog_detach(const union bpf_attr *attr)
2145 {
2146         enum bpf_prog_type ptype;
2147
2148         if (!capable(CAP_NET_ADMIN))
2149                 return -EPERM;
2150
2151         if (CHECK_ATTR(BPF_PROG_DETACH))
2152                 return -EINVAL;
2153
2154         switch (attr->attach_type) {
2155         case BPF_CGROUP_INET_INGRESS:
2156         case BPF_CGROUP_INET_EGRESS:
2157                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2158                 break;
2159         case BPF_CGROUP_INET_SOCK_CREATE:
2160         case BPF_CGROUP_INET4_POST_BIND:
2161         case BPF_CGROUP_INET6_POST_BIND:
2162                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2163                 break;
2164         case BPF_CGROUP_INET4_BIND:
2165         case BPF_CGROUP_INET6_BIND:
2166         case BPF_CGROUP_INET4_CONNECT:
2167         case BPF_CGROUP_INET6_CONNECT:
2168         case BPF_CGROUP_UDP4_SENDMSG:
2169         case BPF_CGROUP_UDP6_SENDMSG:
2170         case BPF_CGROUP_UDP4_RECVMSG:
2171         case BPF_CGROUP_UDP6_RECVMSG:
2172                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2173                 break;
2174         case BPF_CGROUP_SOCK_OPS:
2175                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2176                 break;
2177         case BPF_CGROUP_DEVICE:
2178                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2179                 break;
2180         case BPF_SK_MSG_VERDICT:
2181                 return sock_map_get_from_fd(attr, NULL);
2182         case BPF_SK_SKB_STREAM_PARSER:
2183         case BPF_SK_SKB_STREAM_VERDICT:
2184                 return sock_map_get_from_fd(attr, NULL);
2185         case BPF_LIRC_MODE2:
2186                 return lirc_prog_detach(attr);
2187         case BPF_FLOW_DISSECTOR:
2188                 return skb_flow_dissector_bpf_prog_detach(attr);
2189         case BPF_CGROUP_SYSCTL:
2190                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2191                 break;
2192         case BPF_CGROUP_GETSOCKOPT:
2193         case BPF_CGROUP_SETSOCKOPT:
2194                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2195                 break;
2196         default:
2197                 return -EINVAL;
2198         }
2199
2200         return cgroup_bpf_prog_detach(attr, ptype);
2201 }
2202
2203 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2204
2205 static int bpf_prog_query(const union bpf_attr *attr,
2206                           union bpf_attr __user *uattr)
2207 {
2208         if (!capable(CAP_NET_ADMIN))
2209                 return -EPERM;
2210         if (CHECK_ATTR(BPF_PROG_QUERY))
2211                 return -EINVAL;
2212         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2213                 return -EINVAL;
2214
2215         switch (attr->query.attach_type) {
2216         case BPF_CGROUP_INET_INGRESS:
2217         case BPF_CGROUP_INET_EGRESS:
2218         case BPF_CGROUP_INET_SOCK_CREATE:
2219         case BPF_CGROUP_INET4_BIND:
2220         case BPF_CGROUP_INET6_BIND:
2221         case BPF_CGROUP_INET4_POST_BIND:
2222         case BPF_CGROUP_INET6_POST_BIND:
2223         case BPF_CGROUP_INET4_CONNECT:
2224         case BPF_CGROUP_INET6_CONNECT:
2225         case BPF_CGROUP_UDP4_SENDMSG:
2226         case BPF_CGROUP_UDP6_SENDMSG:
2227         case BPF_CGROUP_UDP4_RECVMSG:
2228         case BPF_CGROUP_UDP6_RECVMSG:
2229         case BPF_CGROUP_SOCK_OPS:
2230         case BPF_CGROUP_DEVICE:
2231         case BPF_CGROUP_SYSCTL:
2232         case BPF_CGROUP_GETSOCKOPT:
2233         case BPF_CGROUP_SETSOCKOPT:
2234                 break;
2235         case BPF_LIRC_MODE2:
2236                 return lirc_prog_query(attr, uattr);
2237         case BPF_FLOW_DISSECTOR:
2238                 return skb_flow_dissector_prog_query(attr, uattr);
2239         default:
2240                 return -EINVAL;
2241         }
2242
2243         return cgroup_bpf_prog_query(attr, uattr);
2244 }
2245
2246 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2247
2248 static int bpf_prog_test_run(const union bpf_attr *attr,
2249                              union bpf_attr __user *uattr)
2250 {
2251         struct bpf_prog *prog;
2252         int ret = -ENOTSUPP;
2253
2254         if (!capable(CAP_SYS_ADMIN))
2255                 return -EPERM;
2256         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2257                 return -EINVAL;
2258
2259         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2260             (!attr->test.ctx_size_in && attr->test.ctx_in))
2261                 return -EINVAL;
2262
2263         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2264             (!attr->test.ctx_size_out && attr->test.ctx_out))
2265                 return -EINVAL;
2266
2267         prog = bpf_prog_get(attr->test.prog_fd);
2268         if (IS_ERR(prog))
2269                 return PTR_ERR(prog);
2270
2271         if (prog->aux->ops->test_run)
2272                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2273
2274         bpf_prog_put(prog);
2275         return ret;
2276 }
2277
2278 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2279
2280 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2281                                union bpf_attr __user *uattr,
2282                                struct idr *idr,
2283                                spinlock_t *lock)
2284 {
2285         u32 next_id = attr->start_id;
2286         int err = 0;
2287
2288         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2289                 return -EINVAL;
2290
2291         if (!capable(CAP_SYS_ADMIN))
2292                 return -EPERM;
2293
2294         next_id++;
2295         spin_lock_bh(lock);
2296         if (!idr_get_next(idr, &next_id))
2297                 err = -ENOENT;
2298         spin_unlock_bh(lock);
2299
2300         if (!err)
2301                 err = put_user(next_id, &uattr->next_id);
2302
2303         return err;
2304 }
2305
2306 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2307
2308 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2309 {
2310         struct bpf_prog *prog;
2311         u32 id = attr->prog_id;
2312         int fd;
2313
2314         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2315                 return -EINVAL;
2316
2317         if (!capable(CAP_SYS_ADMIN))
2318                 return -EPERM;
2319
2320         spin_lock_bh(&prog_idr_lock);
2321         prog = idr_find(&prog_idr, id);
2322         if (prog)
2323                 prog = bpf_prog_inc_not_zero(prog);
2324         else
2325                 prog = ERR_PTR(-ENOENT);
2326         spin_unlock_bh(&prog_idr_lock);
2327
2328         if (IS_ERR(prog))
2329                 return PTR_ERR(prog);
2330
2331         fd = bpf_prog_new_fd(prog);
2332         if (fd < 0)
2333                 bpf_prog_put(prog);
2334
2335         return fd;
2336 }
2337
2338 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2339
2340 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2341 {
2342         struct bpf_map *map;
2343         u32 id = attr->map_id;
2344         int f_flags;
2345         int fd;
2346
2347         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2348             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2349                 return -EINVAL;
2350
2351         if (!capable(CAP_SYS_ADMIN))
2352                 return -EPERM;
2353
2354         f_flags = bpf_get_file_flag(attr->open_flags);
2355         if (f_flags < 0)
2356                 return f_flags;
2357
2358         spin_lock_bh(&map_idr_lock);
2359         map = idr_find(&map_idr, id);
2360         if (map)
2361                 map = __bpf_map_inc_not_zero(map, true);
2362         else
2363                 map = ERR_PTR(-ENOENT);
2364         spin_unlock_bh(&map_idr_lock);
2365
2366         if (IS_ERR(map))
2367                 return PTR_ERR(map);
2368
2369         fd = bpf_map_new_fd(map, f_flags);
2370         if (fd < 0)
2371                 bpf_map_put_with_uref(map);
2372
2373         return fd;
2374 }
2375
2376 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2377                                               unsigned long addr, u32 *off,
2378                                               u32 *type)
2379 {
2380         const struct bpf_map *map;
2381         int i;
2382
2383         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2384                 map = prog->aux->used_maps[i];
2385                 if (map == (void *)addr) {
2386                         *type = BPF_PSEUDO_MAP_FD;
2387                         return map;
2388                 }
2389                 if (!map->ops->map_direct_value_meta)
2390                         continue;
2391                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2392                         *type = BPF_PSEUDO_MAP_VALUE;
2393                         return map;
2394                 }
2395         }
2396
2397         return NULL;
2398 }
2399
2400 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2401 {
2402         const struct bpf_map *map;
2403         struct bpf_insn *insns;
2404         u32 off, type;
2405         u64 imm;
2406         int i;
2407
2408         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2409                         GFP_USER);
2410         if (!insns)
2411                 return insns;
2412
2413         for (i = 0; i < prog->len; i++) {
2414                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2415                         insns[i].code = BPF_JMP | BPF_CALL;
2416                         insns[i].imm = BPF_FUNC_tail_call;
2417                         /* fall-through */
2418                 }
2419                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2420                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2421                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2422                                 insns[i].code = BPF_JMP | BPF_CALL;
2423                         if (!bpf_dump_raw_ok())
2424                                 insns[i].imm = 0;
2425                         continue;
2426                 }
2427
2428                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2429                         continue;
2430
2431                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2432                 map = bpf_map_from_imm(prog, imm, &off, &type);
2433                 if (map) {
2434                         insns[i].src_reg = type;
2435                         insns[i].imm = map->id;
2436                         insns[i + 1].imm = off;
2437                         continue;
2438                 }
2439         }
2440
2441         return insns;
2442 }
2443
2444 static int set_info_rec_size(struct bpf_prog_info *info)
2445 {
2446         /*
2447          * Ensure info.*_rec_size is the same as kernel expected size
2448          *
2449          * or
2450          *
2451          * Only allow zero *_rec_size if both _rec_size and _cnt are
2452          * zero.  In this case, the kernel will set the expected
2453          * _rec_size back to the info.
2454          */
2455
2456         if ((info->nr_func_info || info->func_info_rec_size) &&
2457             info->func_info_rec_size != sizeof(struct bpf_func_info))
2458                 return -EINVAL;
2459
2460         if ((info->nr_line_info || info->line_info_rec_size) &&
2461             info->line_info_rec_size != sizeof(struct bpf_line_info))
2462                 return -EINVAL;
2463
2464         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2465             info->jited_line_info_rec_size != sizeof(__u64))
2466                 return -EINVAL;
2467
2468         info->func_info_rec_size = sizeof(struct bpf_func_info);
2469         info->line_info_rec_size = sizeof(struct bpf_line_info);
2470         info->jited_line_info_rec_size = sizeof(__u64);
2471
2472         return 0;
2473 }
2474
2475 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2476                                    const union bpf_attr *attr,
2477                                    union bpf_attr __user *uattr)
2478 {
2479         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2480         struct bpf_prog_info info = {};
2481         u32 info_len = attr->info.info_len;
2482         struct bpf_prog_stats stats;
2483         char __user *uinsns;
2484         u32 ulen;
2485         int err;
2486
2487         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2488         if (err)
2489                 return err;
2490         info_len = min_t(u32, sizeof(info), info_len);
2491
2492         if (copy_from_user(&info, uinfo, info_len))
2493                 return -EFAULT;
2494
2495         info.type = prog->type;
2496         info.id = prog->aux->id;
2497         info.load_time = prog->aux->load_time;
2498         info.created_by_uid = from_kuid_munged(current_user_ns(),
2499                                                prog->aux->user->uid);
2500         info.gpl_compatible = prog->gpl_compatible;
2501
2502         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2503         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2504
2505         ulen = info.nr_map_ids;
2506         info.nr_map_ids = prog->aux->used_map_cnt;
2507         ulen = min_t(u32, info.nr_map_ids, ulen);
2508         if (ulen) {
2509                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2510                 u32 i;
2511
2512                 for (i = 0; i < ulen; i++)
2513                         if (put_user(prog->aux->used_maps[i]->id,
2514                                      &user_map_ids[i]))
2515                                 return -EFAULT;
2516         }
2517
2518         err = set_info_rec_size(&info);
2519         if (err)
2520                 return err;
2521
2522         bpf_prog_get_stats(prog, &stats);
2523         info.run_time_ns = stats.nsecs;
2524         info.run_cnt = stats.cnt;
2525
2526         if (!capable(CAP_SYS_ADMIN)) {
2527                 info.jited_prog_len = 0;
2528                 info.xlated_prog_len = 0;
2529                 info.nr_jited_ksyms = 0;
2530                 info.nr_jited_func_lens = 0;
2531                 info.nr_func_info = 0;
2532                 info.nr_line_info = 0;
2533                 info.nr_jited_line_info = 0;
2534                 goto done;
2535         }
2536
2537         ulen = info.xlated_prog_len;
2538         info.xlated_prog_len = bpf_prog_insn_size(prog);
2539         if (info.xlated_prog_len && ulen) {
2540                 struct bpf_insn *insns_sanitized;
2541                 bool fault;
2542
2543                 if (prog->blinded && !bpf_dump_raw_ok()) {
2544                         info.xlated_prog_insns = 0;
2545                         goto done;
2546                 }
2547                 insns_sanitized = bpf_insn_prepare_dump(prog);
2548                 if (!insns_sanitized)
2549                         return -ENOMEM;
2550                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2551                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2552                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2553                 kfree(insns_sanitized);
2554                 if (fault)
2555                         return -EFAULT;
2556         }
2557
2558         if (bpf_prog_is_dev_bound(prog->aux)) {
2559                 err = bpf_prog_offload_info_fill(&info, prog);
2560                 if (err)
2561                         return err;
2562                 goto done;
2563         }
2564
2565         /* NOTE: the following code is supposed to be skipped for offload.
2566          * bpf_prog_offload_info_fill() is the place to fill similar fields
2567          * for offload.
2568          */
2569         ulen = info.jited_prog_len;
2570         if (prog->aux->func_cnt) {
2571                 u32 i;
2572
2573                 info.jited_prog_len = 0;
2574                 for (i = 0; i < prog->aux->func_cnt; i++)
2575                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2576         } else {
2577                 info.jited_prog_len = prog->jited_len;
2578         }
2579
2580         if (info.jited_prog_len && ulen) {
2581                 if (bpf_dump_raw_ok()) {
2582                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2583                         ulen = min_t(u32, info.jited_prog_len, ulen);
2584
2585                         /* for multi-function programs, copy the JITed
2586                          * instructions for all the functions
2587                          */
2588                         if (prog->aux->func_cnt) {
2589                                 u32 len, free, i;
2590                                 u8 *img;
2591
2592                                 free = ulen;
2593                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2594                                         len = prog->aux->func[i]->jited_len;
2595                                         len = min_t(u32, len, free);
2596                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2597                                         if (copy_to_user(uinsns, img, len))
2598                                                 return -EFAULT;
2599                                         uinsns += len;
2600                                         free -= len;
2601                                         if (!free)
2602                                                 break;
2603                                 }
2604                         } else {
2605                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2606                                         return -EFAULT;
2607                         }
2608                 } else {
2609                         info.jited_prog_insns = 0;
2610                 }
2611         }
2612
2613         ulen = info.nr_jited_ksyms;
2614         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2615         if (ulen) {
2616                 if (bpf_dump_raw_ok()) {
2617                         unsigned long ksym_addr;
2618                         u64 __user *user_ksyms;
2619                         u32 i;
2620
2621                         /* copy the address of the kernel symbol
2622                          * corresponding to each function
2623                          */
2624                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2625                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2626                         if (prog->aux->func_cnt) {
2627                                 for (i = 0; i < ulen; i++) {
2628                                         ksym_addr = (unsigned long)
2629                                                 prog->aux->func[i]->bpf_func;
2630                                         if (put_user((u64) ksym_addr,
2631                                                      &user_ksyms[i]))
2632                                                 return -EFAULT;
2633                                 }
2634                         } else {
2635                                 ksym_addr = (unsigned long) prog->bpf_func;
2636                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2637                                         return -EFAULT;
2638                         }
2639                 } else {
2640                         info.jited_ksyms = 0;
2641                 }
2642         }
2643
2644         ulen = info.nr_jited_func_lens;
2645         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2646         if (ulen) {
2647                 if (bpf_dump_raw_ok()) {
2648                         u32 __user *user_lens;
2649                         u32 func_len, i;
2650
2651                         /* copy the JITed image lengths for each function */
2652                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2653                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2654                         if (prog->aux->func_cnt) {
2655                                 for (i = 0; i < ulen; i++) {
2656                                         func_len =
2657                                                 prog->aux->func[i]->jited_len;
2658                                         if (put_user(func_len, &user_lens[i]))
2659                                                 return -EFAULT;
2660                                 }
2661                         } else {
2662                                 func_len = prog->jited_len;
2663                                 if (put_user(func_len, &user_lens[0]))
2664                                         return -EFAULT;
2665                         }
2666                 } else {
2667                         info.jited_func_lens = 0;
2668                 }
2669         }
2670
2671         if (prog->aux->btf)
2672                 info.btf_id = btf_id(prog->aux->btf);
2673
2674         ulen = info.nr_func_info;
2675         info.nr_func_info = prog->aux->func_info_cnt;
2676         if (info.nr_func_info && ulen) {
2677                 char __user *user_finfo;
2678
2679                 user_finfo = u64_to_user_ptr(info.func_info);
2680                 ulen = min_t(u32, info.nr_func_info, ulen);
2681                 if (copy_to_user(user_finfo, prog->aux->func_info,
2682                                  info.func_info_rec_size * ulen))
2683                         return -EFAULT;
2684         }
2685
2686         ulen = info.nr_line_info;
2687         info.nr_line_info = prog->aux->nr_linfo;
2688         if (info.nr_line_info && ulen) {
2689                 __u8 __user *user_linfo;
2690
2691                 user_linfo = u64_to_user_ptr(info.line_info);
2692                 ulen = min_t(u32, info.nr_line_info, ulen);
2693                 if (copy_to_user(user_linfo, prog->aux->linfo,
2694                                  info.line_info_rec_size * ulen))
2695                         return -EFAULT;
2696         }
2697
2698         ulen = info.nr_jited_line_info;
2699         if (prog->aux->jited_linfo)
2700                 info.nr_jited_line_info = prog->aux->nr_linfo;
2701         else
2702                 info.nr_jited_line_info = 0;
2703         if (info.nr_jited_line_info && ulen) {
2704                 if (bpf_dump_raw_ok()) {
2705                         __u64 __user *user_linfo;
2706                         u32 i;
2707
2708                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2709                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2710                         for (i = 0; i < ulen; i++) {
2711                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2712                                              &user_linfo[i]))
2713                                         return -EFAULT;
2714                         }
2715                 } else {
2716                         info.jited_line_info = 0;
2717                 }
2718         }
2719
2720         ulen = info.nr_prog_tags;
2721         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2722         if (ulen) {
2723                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2724                 u32 i;
2725
2726                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2727                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2728                 if (prog->aux->func_cnt) {
2729                         for (i = 0; i < ulen; i++) {
2730                                 if (copy_to_user(user_prog_tags[i],
2731                                                  prog->aux->func[i]->tag,
2732                                                  BPF_TAG_SIZE))
2733                                         return -EFAULT;
2734                         }
2735                 } else {
2736                         if (copy_to_user(user_prog_tags[0],
2737                                          prog->tag, BPF_TAG_SIZE))
2738                                 return -EFAULT;
2739                 }
2740         }
2741
2742 done:
2743         if (copy_to_user(uinfo, &info, info_len) ||
2744             put_user(info_len, &uattr->info.info_len))
2745                 return -EFAULT;
2746
2747         return 0;
2748 }
2749
2750 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2751                                   const union bpf_attr *attr,
2752                                   union bpf_attr __user *uattr)
2753 {
2754         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2755         struct bpf_map_info info = {};
2756         u32 info_len = attr->info.info_len;
2757         int err;
2758
2759         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2760         if (err)
2761                 return err;
2762         info_len = min_t(u32, sizeof(info), info_len);
2763
2764         info.type = map->map_type;
2765         info.id = map->id;
2766         info.key_size = map->key_size;
2767         info.value_size = map->value_size;
2768         info.max_entries = map->max_entries;
2769         info.map_flags = map->map_flags;
2770         memcpy(info.name, map->name, sizeof(map->name));
2771
2772         if (map->btf) {
2773                 info.btf_id = btf_id(map->btf);
2774                 info.btf_key_type_id = map->btf_key_type_id;
2775                 info.btf_value_type_id = map->btf_value_type_id;
2776         }
2777
2778         if (bpf_map_is_dev_bound(map)) {
2779                 err = bpf_map_offload_info_fill(&info, map);
2780                 if (err)
2781                         return err;
2782         }
2783
2784         if (copy_to_user(uinfo, &info, info_len) ||
2785             put_user(info_len, &uattr->info.info_len))
2786                 return -EFAULT;
2787
2788         return 0;
2789 }
2790
2791 static int bpf_btf_get_info_by_fd(struct btf *btf,
2792                                   const union bpf_attr *attr,
2793                                   union bpf_attr __user *uattr)
2794 {
2795         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2796         u32 info_len = attr->info.info_len;
2797         int err;
2798
2799         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2800         if (err)
2801                 return err;
2802
2803         return btf_get_info_by_fd(btf, attr, uattr);
2804 }
2805
2806 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2807
2808 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2809                                   union bpf_attr __user *uattr)
2810 {
2811         int ufd = attr->info.bpf_fd;
2812         struct fd f;
2813         int err;
2814
2815         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2816                 return -EINVAL;
2817
2818         f = fdget(ufd);
2819         if (!f.file)
2820                 return -EBADFD;
2821
2822         if (f.file->f_op == &bpf_prog_fops)
2823                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2824                                               uattr);
2825         else if (f.file->f_op == &bpf_map_fops)
2826                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2827                                              uattr);
2828         else if (f.file->f_op == &btf_fops)
2829                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2830         else
2831                 err = -EINVAL;
2832
2833         fdput(f);
2834         return err;
2835 }
2836
2837 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2838
2839 static int bpf_btf_load(const union bpf_attr *attr)
2840 {
2841         if (CHECK_ATTR(BPF_BTF_LOAD))
2842                 return -EINVAL;
2843
2844         if (!capable(CAP_SYS_ADMIN))
2845                 return -EPERM;
2846
2847         return btf_new_fd(attr);
2848 }
2849
2850 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2851
2852 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2853 {
2854         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2855                 return -EINVAL;
2856
2857         if (!capable(CAP_SYS_ADMIN))
2858                 return -EPERM;
2859
2860         return btf_get_fd_by_id(attr->btf_id);
2861 }
2862
2863 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2864                                     union bpf_attr __user *uattr,
2865                                     u32 prog_id, u32 fd_type,
2866                                     const char *buf, u64 probe_offset,
2867                                     u64 probe_addr)
2868 {
2869         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2870         u32 len = buf ? strlen(buf) : 0, input_len;
2871         int err = 0;
2872
2873         if (put_user(len, &uattr->task_fd_query.buf_len))
2874                 return -EFAULT;
2875         input_len = attr->task_fd_query.buf_len;
2876         if (input_len && ubuf) {
2877                 if (!len) {
2878                         /* nothing to copy, just make ubuf NULL terminated */
2879                         char zero = '\0';
2880
2881                         if (put_user(zero, ubuf))
2882                                 return -EFAULT;
2883                 } else if (input_len >= len + 1) {
2884                         /* ubuf can hold the string with NULL terminator */
2885                         if (copy_to_user(ubuf, buf, len + 1))
2886                                 return -EFAULT;
2887                 } else {
2888                         /* ubuf cannot hold the string with NULL terminator,
2889                          * do a partial copy with NULL terminator.
2890                          */
2891                         char zero = '\0';
2892
2893                         err = -ENOSPC;
2894                         if (copy_to_user(ubuf, buf, input_len - 1))
2895                                 return -EFAULT;
2896                         if (put_user(zero, ubuf + input_len - 1))
2897                                 return -EFAULT;
2898                 }
2899         }
2900
2901         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2902             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2903             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2904             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2905                 return -EFAULT;
2906
2907         return err;
2908 }
2909
2910 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2911
2912 static int bpf_task_fd_query(const union bpf_attr *attr,
2913                              union bpf_attr __user *uattr)
2914 {
2915         pid_t pid = attr->task_fd_query.pid;
2916         u32 fd = attr->task_fd_query.fd;
2917         const struct perf_event *event;
2918         struct files_struct *files;
2919         struct task_struct *task;
2920         struct file *file;
2921         int err;
2922
2923         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2924                 return -EINVAL;
2925
2926         if (!capable(CAP_SYS_ADMIN))
2927                 return -EPERM;
2928
2929         if (attr->task_fd_query.flags != 0)
2930                 return -EINVAL;
2931
2932         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2933         if (!task)
2934                 return -ENOENT;
2935
2936         files = get_files_struct(task);
2937         put_task_struct(task);
2938         if (!files)
2939                 return -ENOENT;
2940
2941         err = 0;
2942         spin_lock(&files->file_lock);
2943         file = fcheck_files(files, fd);
2944         if (!file)
2945                 err = -EBADF;
2946         else
2947                 get_file(file);
2948         spin_unlock(&files->file_lock);
2949         put_files_struct(files);
2950
2951         if (err)
2952                 goto out;
2953
2954         if (file->f_op == &bpf_raw_tp_fops) {
2955                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2956                 struct bpf_raw_event_map *btp = raw_tp->btp;
2957
2958                 err = bpf_task_fd_query_copy(attr, uattr,
2959                                              raw_tp->prog->aux->id,
2960                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2961                                              btp->tp->name, 0, 0);
2962                 goto put_file;
2963         }
2964
2965         event = perf_get_event(file);
2966         if (!IS_ERR(event)) {
2967                 u64 probe_offset, probe_addr;
2968                 u32 prog_id, fd_type;
2969                 const char *buf;
2970
2971                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2972                                               &buf, &probe_offset,
2973                                               &probe_addr);
2974                 if (!err)
2975                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2976                                                      fd_type, buf,
2977                                                      probe_offset,
2978                                                      probe_addr);
2979                 goto put_file;
2980         }
2981
2982         err = -ENOTSUPP;
2983 put_file:
2984         fput(file);
2985 out:
2986         return err;
2987 }
2988
2989 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2990 {
2991         union bpf_attr attr = {};
2992         int err;
2993
2994         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2995                 return -EPERM;
2996
2997         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2998         if (err)
2999                 return err;
3000         size = min_t(u32, size, sizeof(attr));
3001
3002         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3003         if (copy_from_user(&attr, uattr, size) != 0)
3004                 return -EFAULT;
3005
3006         err = security_bpf(cmd, &attr, size);
3007         if (err < 0)
3008                 return err;
3009
3010         switch (cmd) {
3011         case BPF_MAP_CREATE:
3012                 err = map_create(&attr);
3013                 break;
3014         case BPF_MAP_LOOKUP_ELEM:
3015                 err = map_lookup_elem(&attr);
3016                 break;
3017         case BPF_MAP_UPDATE_ELEM:
3018                 err = map_update_elem(&attr);
3019                 break;
3020         case BPF_MAP_DELETE_ELEM:
3021                 err = map_delete_elem(&attr);
3022                 break;
3023         case BPF_MAP_GET_NEXT_KEY:
3024                 err = map_get_next_key(&attr);
3025                 break;
3026         case BPF_MAP_FREEZE:
3027                 err = map_freeze(&attr);
3028                 break;
3029         case BPF_PROG_LOAD:
3030                 err = bpf_prog_load(&attr, uattr);
3031                 break;
3032         case BPF_OBJ_PIN:
3033                 err = bpf_obj_pin(&attr);
3034                 break;
3035         case BPF_OBJ_GET:
3036                 err = bpf_obj_get(&attr);
3037                 break;
3038         case BPF_PROG_ATTACH:
3039                 err = bpf_prog_attach(&attr);
3040                 break;
3041         case BPF_PROG_DETACH:
3042                 err = bpf_prog_detach(&attr);
3043                 break;
3044         case BPF_PROG_QUERY:
3045                 err = bpf_prog_query(&attr, uattr);
3046                 break;
3047         case BPF_PROG_TEST_RUN:
3048                 err = bpf_prog_test_run(&attr, uattr);
3049                 break;
3050         case BPF_PROG_GET_NEXT_ID:
3051                 err = bpf_obj_get_next_id(&attr, uattr,
3052                                           &prog_idr, &prog_idr_lock);
3053                 break;
3054         case BPF_MAP_GET_NEXT_ID:
3055                 err = bpf_obj_get_next_id(&attr, uattr,
3056                                           &map_idr, &map_idr_lock);
3057                 break;
3058         case BPF_BTF_GET_NEXT_ID:
3059                 err = bpf_obj_get_next_id(&attr, uattr,
3060                                           &btf_idr, &btf_idr_lock);
3061                 break;
3062         case BPF_PROG_GET_FD_BY_ID:
3063                 err = bpf_prog_get_fd_by_id(&attr);
3064                 break;
3065         case BPF_MAP_GET_FD_BY_ID:
3066                 err = bpf_map_get_fd_by_id(&attr);
3067                 break;
3068         case BPF_OBJ_GET_INFO_BY_FD:
3069                 err = bpf_obj_get_info_by_fd(&attr, uattr);
3070                 break;
3071         case BPF_RAW_TRACEPOINT_OPEN:
3072                 err = bpf_raw_tracepoint_open(&attr);
3073                 break;
3074         case BPF_BTF_LOAD:
3075                 err = bpf_btf_load(&attr);
3076                 break;
3077         case BPF_BTF_GET_FD_BY_ID:
3078                 err = bpf_btf_get_fd_by_id(&attr);
3079                 break;
3080         case BPF_TASK_FD_QUERY:
3081                 err = bpf_task_fd_query(&attr, uattr);
3082                 break;
3083         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3084                 err = map_lookup_and_delete_elem(&attr);
3085                 break;
3086         default:
3087                 err = -EINVAL;
3088                 break;
3089         }
3090
3091         return err;
3092 }