syscall.c - kernel/bpf/syscall.c - Linux diff v6.2 - Bootlin Elixir Cross Referencer

   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-cgroup.h>
   6#include <linux/bpf_trace.h>
   7#include <linux/bpf_lirc.h>
   8#include <linux/bpf_verifier.h>
   9#include <linux/bsearch.h>
  10#include <linux/btf.h>
  11#include <linux/syscalls.h>
  12#include <linux/slab.h>
  13#include <linux/sched/signal.h>
  14#include <linux/vmalloc.h>
  15#include <linux/mmzone.h>
  16#include <linux/anon_inodes.h>
  17#include <linux/fdtable.h>
  18#include <linux/file.h>
  19#include <linux/fs.h>
  20#include <linux/license.h>
  21#include <linux/filter.h>
  22#include <linux/kernel.h>
  23#include <linux/idr.h>
  24#include <linux/cred.h>
  25#include <linux/timekeeping.h>
  26#include <linux/ctype.h>
  27#include <linux/nospec.h>
  28#include <linux/audit.h>
  29#include <uapi/linux/btf.h>
  30#include <linux/pgtable.h>
  31#include <linux/bpf_lsm.h>
  32#include <linux/poll.h>
  33#include <linux/sort.h>
  34#include <linux/bpf-netns.h>
  35#include <linux/rcupdate_trace.h>
  36#include <linux/memcontrol.h>
  37#include <linux/trace_events.h>
  38
 
 
 
 
  39#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  40			  (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  41			  (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  42#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  43#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  44#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  45			IS_FD_HASH(map))
  46
  47#define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  48
  49DEFINE_PER_CPU(int, bpf_prog_active);
  50static DEFINE_IDR(prog_idr);
  51static DEFINE_SPINLOCK(prog_idr_lock);
  52static DEFINE_IDR(map_idr);
  53static DEFINE_SPINLOCK(map_idr_lock);
  54static DEFINE_IDR(link_idr);
  55static DEFINE_SPINLOCK(link_idr_lock);
  56
  57int sysctl_unprivileged_bpf_disabled __read_mostly =
  58	IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  59
  60static const struct bpf_map_ops * const bpf_map_types[] = {
  61#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  62#define BPF_MAP_TYPE(_id, _ops) \
  63	[_id] = &_ops,
  64#define BPF_LINK_TYPE(_id, _name)
  65#include <linux/bpf_types.h>
  66#undef BPF_PROG_TYPE
  67#undef BPF_MAP_TYPE
  68#undef BPF_LINK_TYPE
  69};
  70
  71/*
  72 * If we're handed a bigger struct than we know of, ensure all the unknown bits
  73 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  74 * we don't know about yet.
  75 *
  76 * There is a ToCToU between this function call and the following
  77 * copy_from_user() call. However, this is not a concern since this function is
  78 * meant to be a future-proofing of bits.
  79 */
  80int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  81			     size_t expected_size,
  82			     size_t actual_size)
  83{
  84	int res;
  85
  86	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
  87		return -E2BIG;
  88
  89	if (actual_size <= expected_size)
  90		return 0;
  91
  92	if (uaddr.is_kernel)
  93		res = memchr_inv(uaddr.kernel + expected_size, 0,
  94				 actual_size - expected_size) == NULL;
  95	else
  96		res = check_zeroed_user(uaddr.user + expected_size,
  97					actual_size - expected_size);
  98	if (res < 0)
  99		return res;
 100	return res ? 0 : -E2BIG;
 101}
 102
 103const struct bpf_map_ops bpf_map_offload_ops = {
 104	.map_meta_equal = bpf_map_meta_equal,
 105	.map_alloc = bpf_map_offload_map_alloc,
 106	.map_free = bpf_map_offload_map_free,
 107	.map_check_btf = map_check_no_btf,
 
 108};
 109
 110static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 111{
 112	const struct bpf_map_ops *ops;
 113	u32 type = attr->map_type;
 114	struct bpf_map *map;
 115	int err;
 116
 117	if (type >= ARRAY_SIZE(bpf_map_types))
 118		return ERR_PTR(-EINVAL);
 119	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 120	ops = bpf_map_types[type];
 121	if (!ops)
 122		return ERR_PTR(-EINVAL);
 123
 124	if (ops->map_alloc_check) {
 125		err = ops->map_alloc_check(attr);
 126		if (err)
 127			return ERR_PTR(err);
 128	}
 129	if (attr->map_ifindex)
 130		ops = &bpf_map_offload_ops;
 131	map = ops->map_alloc(attr);
 132	if (IS_ERR(map))
 133		return map;
 134	map->ops = ops;
 135	map->map_type = type;
 136	return map;
 137}
 138
 139static void bpf_map_write_active_inc(struct bpf_map *map)
 140{
 141	atomic64_inc(&map->writecnt);
 142}
 143
 144static void bpf_map_write_active_dec(struct bpf_map *map)
 145{
 146	atomic64_dec(&map->writecnt);
 147}
 148
 149bool bpf_map_write_active(const struct bpf_map *map)
 150{
 151	return atomic64_read(&map->writecnt) != 0;
 152}
 153
 154static u32 bpf_map_value_size(const struct bpf_map *map)
 155{
 156	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 157	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 158	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 159	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 160		return round_up(map->value_size, 8) * num_possible_cpus();
 161	else if (IS_FD_MAP(map))
 162		return sizeof(u32);
 163	else
 164		return  map->value_size;
 165}
 166
 167static void maybe_wait_bpf_programs(struct bpf_map *map)
 168{
 169	/* Wait for any running BPF programs to complete so that
 170	 * userspace, when we return to it, knows that all programs
 171	 * that could be running use the new map value.
 
 
 
 
 172	 */
 173	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 174	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 175		synchronize_rcu();
 176}
 177
 178static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 179				void *key, void *value, __u64 flags)
 180{
 181	int err;
 182
 183	/* Need to create a kthread, thus must support schedule */
 184	if (bpf_map_is_dev_bound(map)) {
 185		return bpf_map_offload_update_elem(map, key, value, flags);
 186	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 
 187		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 188		return map->ops->map_update_elem(map, key, value, flags);
 189	} else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 190		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 191		return sock_map_update_elem_sys(map, key, value, flags);
 192	} else if (IS_FD_PROG_ARRAY(map)) {
 193		return bpf_fd_array_map_update_elem(map, map_file, key, value,
 194						    flags);
 195	}
 196
 197	bpf_disable_instrumentation();
 198	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 199	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 200		err = bpf_percpu_hash_update(map, key, value, flags);
 201	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 202		err = bpf_percpu_array_update(map, key, value, flags);
 203	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 204		err = bpf_percpu_cgroup_storage_update(map, key, value,
 205						       flags);
 206	} else if (IS_FD_ARRAY(map)) {
 207		rcu_read_lock();
 208		err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 209						   flags);
 210		rcu_read_unlock();
 211	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 212		rcu_read_lock();
 213		err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 214						  flags);
 215		rcu_read_unlock();
 216	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 217		/* rcu_read_lock() is not needed */
 218		err = bpf_fd_reuseport_array_update_elem(map, key, value,
 219							 flags);
 220	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 221		   map->map_type == BPF_MAP_TYPE_STACK ||
 222		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 223		err = map->ops->map_push_elem(map, value, flags);
 224	} else {
 225		rcu_read_lock();
 226		err = map->ops->map_update_elem(map, key, value, flags);
 227		rcu_read_unlock();
 228	}
 229	bpf_enable_instrumentation();
 230	maybe_wait_bpf_programs(map);
 231
 232	return err;
 233}
 234
 235static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 236			      __u64 flags)
 237{
 238	void *ptr;
 239	int err;
 240
 241	if (bpf_map_is_dev_bound(map))
 242		return bpf_map_offload_lookup_elem(map, key, value);
 243
 244	bpf_disable_instrumentation();
 245	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 246	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 247		err = bpf_percpu_hash_copy(map, key, value);
 248	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 249		err = bpf_percpu_array_copy(map, key, value);
 250	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 251		err = bpf_percpu_cgroup_storage_copy(map, key, value);
 252	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 253		err = bpf_stackmap_copy(map, key, value);
 254	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 255		err = bpf_fd_array_map_lookup_elem(map, key, value);
 256	} else if (IS_FD_HASH(map)) {
 257		err = bpf_fd_htab_map_lookup_elem(map, key, value);
 258	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 259		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 260	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 261		   map->map_type == BPF_MAP_TYPE_STACK ||
 262		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 263		err = map->ops->map_peek_elem(map, value);
 264	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 265		/* struct_ops map requires directly updating "value" */
 266		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 267	} else {
 268		rcu_read_lock();
 269		if (map->ops->map_lookup_elem_sys_only)
 270			ptr = map->ops->map_lookup_elem_sys_only(map, key);
 271		else
 272			ptr = map->ops->map_lookup_elem(map, key);
 273		if (IS_ERR(ptr)) {
 274			err = PTR_ERR(ptr);
 275		} else if (!ptr) {
 276			err = -ENOENT;
 277		} else {
 278			err = 0;
 279			if (flags & BPF_F_LOCK)
 280				/* lock 'ptr' and copy everything but lock */
 281				copy_map_value_locked(map, value, ptr, true);
 282			else
 283				copy_map_value(map, value, ptr);
 284			/* mask lock and timer, since value wasn't zero inited */
 285			check_and_init_map_value(map, value);
 286		}
 287		rcu_read_unlock();
 288	}
 289
 290	bpf_enable_instrumentation();
 291	maybe_wait_bpf_programs(map);
 292
 293	return err;
 294}
 295
 296/* Please, do not use this function outside from the map creation path
 297 * (e.g. in map update path) without taking care of setting the active
 298 * memory cgroup (see at bpf_map_kmalloc_node() for example).
 299 */
 300static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 301{
 302	/* We really just want to fail instead of triggering OOM killer
 303	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 304	 * which is used for lower order allocation requests.
 305	 *
 306	 * It has been observed that higher order allocation requests done by
 307	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
 308	 * to reclaim memory from the page cache, thus we set
 309	 * __GFP_RETRY_MAYFAIL to avoid such situations.
 310	 */
 311
 312	const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
 313	unsigned int flags = 0;
 314	unsigned long align = 1;
 315	void *area;
 316
 317	if (size >= SIZE_MAX)
 318		return NULL;
 319
 320	/* kmalloc()'ed memory can't be mmap()'ed */
 321	if (mmapable) {
 322		BUG_ON(!PAGE_ALIGNED(size));
 323		align = SHMLBA;
 324		flags = VM_USERMAP;
 325	} else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 326		area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 327				    numa_node);
 328		if (area != NULL)
 329			return area;
 330	}
 331
 332	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 333			gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 334			flags, numa_node, __builtin_return_address(0));
 335}
 336
 337void *bpf_map_area_alloc(u64 size, int numa_node)
 338{
 339	return __bpf_map_area_alloc(size, numa_node, false);
 340}
 341
 342void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 343{
 344	return __bpf_map_area_alloc(size, numa_node, true);
 345}
 346
 347void bpf_map_area_free(void *area)
 348{
 349	kvfree(area);
 350}
 351
 352static u32 bpf_map_flags_retain_permanent(u32 flags)
 353{
 354	/* Some map creation flags are not tied to the map object but
 355	 * rather to the map fd instead, so they have no meaning upon
 356	 * map object inspection since multiple file descriptors with
 357	 * different (access) properties can exist here. Thus, given
 358	 * this has zero meaning for the map itself, lets clear these
 359	 * from here.
 360	 */
 361	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 362}
 363
 364void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 365{
 366	map->map_type = attr->map_type;
 367	map->key_size = attr->key_size;
 368	map->value_size = attr->value_size;
 369	map->max_entries = attr->max_entries;
 370	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 371	map->numa_node = bpf_map_attr_numa_node(attr);
 372	map->map_extra = attr->map_extra;
 373}
 374
 375static int bpf_map_alloc_id(struct bpf_map *map)
 376{
 377	int id;
 378
 379	idr_preload(GFP_KERNEL);
 380	spin_lock_bh(&map_idr_lock);
 381	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 382	if (id > 0)
 383		map->id = id;
 384	spin_unlock_bh(&map_idr_lock);
 385	idr_preload_end();
 386
 387	if (WARN_ON_ONCE(!id))
 388		return -ENOSPC;
 389
 390	return id > 0 ? 0 : id;
 391}
 392
 393void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 394{
 395	unsigned long flags;
 396
 397	/* Offloaded maps are removed from the IDR store when their device
 398	 * disappears - even if someone holds an fd to them they are unusable,
 399	 * the memory is gone, all ops will fail; they are simply waiting for
 400	 * refcnt to drop to be freed.
 401	 */
 402	if (!map->id)
 403		return;
 404
 405	if (do_idr_lock)
 406		spin_lock_irqsave(&map_idr_lock, flags);
 407	else
 408		__acquire(&map_idr_lock);
 409
 410	idr_remove(&map_idr, map->id);
 411	map->id = 0;
 412
 413	if (do_idr_lock)
 414		spin_unlock_irqrestore(&map_idr_lock, flags);
 415	else
 416		__release(&map_idr_lock);
 417}
 418
 419#ifdef CONFIG_MEMCG_KMEM
 420static void bpf_map_save_memcg(struct bpf_map *map)
 421{
 422	/* Currently if a map is created by a process belonging to the root
 423	 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 424	 * So we have to check map->objcg for being NULL each time it's
 425	 * being used.
 426	 */
 427	map->objcg = get_obj_cgroup_from_current();
 
 428}
 429
 430static void bpf_map_release_memcg(struct bpf_map *map)
 431{
 432	if (map->objcg)
 433		obj_cgroup_put(map->objcg);
 434}
 435
 436static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 437{
 438	if (map->objcg)
 439		return get_mem_cgroup_from_objcg(map->objcg);
 440
 441	return root_mem_cgroup;
 442}
 443
 444void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 445			   int node)
 446{
 447	struct mem_cgroup *memcg, *old_memcg;
 448	void *ptr;
 449
 450	memcg = bpf_map_get_memcg(map);
 451	old_memcg = set_active_memcg(memcg);
 452	ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 453	set_active_memcg(old_memcg);
 454	mem_cgroup_put(memcg);
 455
 456	return ptr;
 457}
 458
 459void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 460{
 461	struct mem_cgroup *memcg, *old_memcg;
 462	void *ptr;
 463
 464	memcg = bpf_map_get_memcg(map);
 465	old_memcg = set_active_memcg(memcg);
 466	ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 467	set_active_memcg(old_memcg);
 468	mem_cgroup_put(memcg);
 469
 470	return ptr;
 471}
 472
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 473void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 474				    size_t align, gfp_t flags)
 475{
 476	struct mem_cgroup *memcg, *old_memcg;
 477	void __percpu *ptr;
 478
 479	memcg = bpf_map_get_memcg(map);
 480	old_memcg = set_active_memcg(memcg);
 481	ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 482	set_active_memcg(old_memcg);
 483	mem_cgroup_put(memcg);
 484
 485	return ptr;
 486}
 487
 488#else
 489static void bpf_map_save_memcg(struct bpf_map *map)
 490{
 491}
 492
 493static void bpf_map_release_memcg(struct bpf_map *map)
 494{
 495}
 496#endif
 497
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 498static int btf_field_cmp(const void *a, const void *b)
 499{
 500	const struct btf_field *f1 = a, *f2 = b;
 501
 502	if (f1->offset < f2->offset)
 503		return -1;
 504	else if (f1->offset > f2->offset)
 505		return 1;
 506	return 0;
 507}
 508
 509struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 510				  enum btf_field_type type)
 511{
 512	struct btf_field *field;
 513
 514	if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & type))
 515		return NULL;
 516	field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 517	if (!field || !(field->type & type))
 518		return NULL;
 519	return field;
 520}
 521
 522void btf_record_free(struct btf_record *rec)
 523{
 524	int i;
 525
 526	if (IS_ERR_OR_NULL(rec))
 527		return;
 528	for (i = 0; i < rec->cnt; i++) {
 529		switch (rec->fields[i].type) {
 530		case BPF_SPIN_LOCK:
 531		case BPF_TIMER:
 532			break;
 533		case BPF_KPTR_UNREF:
 534		case BPF_KPTR_REF:
 
 535			if (rec->fields[i].kptr.module)
 536				module_put(rec->fields[i].kptr.module);
 537			btf_put(rec->fields[i].kptr.btf);
 538			break;
 539		case BPF_LIST_HEAD:
 540		case BPF_LIST_NODE:
 541			/* Nothing to release for bpf_list_head */
 
 
 
 
 
 542			break;
 543		default:
 544			WARN_ON_ONCE(1);
 545			continue;
 546		}
 547	}
 548	kfree(rec);
 549}
 550
 551void bpf_map_free_record(struct bpf_map *map)
 552{
 553	btf_record_free(map->record);
 554	map->record = NULL;
 555}
 556
 557struct btf_record *btf_record_dup(const struct btf_record *rec)
 558{
 559	const struct btf_field *fields;
 560	struct btf_record *new_rec;
 561	int ret, size, i;
 562
 563	if (IS_ERR_OR_NULL(rec))
 564		return NULL;
 565	size = offsetof(struct btf_record, fields[rec->cnt]);
 566	new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 567	if (!new_rec)
 568		return ERR_PTR(-ENOMEM);
 569	/* Do a deep copy of the btf_record */
 570	fields = rec->fields;
 571	new_rec->cnt = 0;
 572	for (i = 0; i < rec->cnt; i++) {
 573		switch (fields[i].type) {
 574		case BPF_SPIN_LOCK:
 575		case BPF_TIMER:
 576			break;
 577		case BPF_KPTR_UNREF:
 578		case BPF_KPTR_REF:
 
 579			btf_get(fields[i].kptr.btf);
 580			if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 581				ret = -ENXIO;
 582				goto free;
 583			}
 584			break;
 585		case BPF_LIST_HEAD:
 586		case BPF_LIST_NODE:
 587			/* Nothing to acquire for bpf_list_head */
 
 
 
 
 
 588			break;
 589		default:
 590			ret = -EFAULT;
 591			WARN_ON_ONCE(1);
 592			goto free;
 593		}
 594		new_rec->cnt++;
 595	}
 596	return new_rec;
 597free:
 598	btf_record_free(new_rec);
 599	return ERR_PTR(ret);
 600}
 601
 602bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 603{
 604	bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 605	int size;
 606
 607	if (!a_has_fields && !b_has_fields)
 608		return true;
 609	if (a_has_fields != b_has_fields)
 610		return false;
 611	if (rec_a->cnt != rec_b->cnt)
 612		return false;
 613	size = offsetof(struct btf_record, fields[rec_a->cnt]);
 614	/* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 615	 * members are zeroed out. So memcmp is safe to do without worrying
 616	 * about padding/unused fields.
 617	 *
 618	 * While spin_lock, timer, and kptr have no relation to map BTF,
 619	 * list_head metadata is specific to map BTF, the btf and value_rec
 620	 * members in particular. btf is the map BTF, while value_rec points to
 621	 * btf_record in that map BTF.
 622	 *
 623	 * So while by default, we don't rely on the map BTF (which the records
 624	 * were parsed from) matching for both records, which is not backwards
 625	 * compatible, in case list_head is part of it, we implicitly rely on
 626	 * that by way of depending on memcmp succeeding for it.
 627	 */
 628	return !memcmp(rec_a, rec_b, size);
 629}
 630
 631void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 632{
 633	if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 634		return;
 635	bpf_timer_cancel_and_free(obj + rec->timer_off);
 636}
 637
 638void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 639{
 640	const struct btf_field *fields;
 641	int i;
 642
 643	if (IS_ERR_OR_NULL(rec))
 644		return;
 645	fields = rec->fields;
 646	for (i = 0; i < rec->cnt; i++) {
 
 647		const struct btf_field *field = &fields[i];
 648		void *field_ptr = obj + field->offset;
 
 649
 650		switch (fields[i].type) {
 651		case BPF_SPIN_LOCK:
 652			break;
 653		case BPF_TIMER:
 654			bpf_timer_cancel_and_free(field_ptr);
 655			break;
 656		case BPF_KPTR_UNREF:
 657			WRITE_ONCE(*(u64 *)field_ptr, 0);
 658			break;
 659		case BPF_KPTR_REF:
 660			field->kptr.dtor((void *)xchg((unsigned long *)field_ptr, 0));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 661			break;
 662		case BPF_LIST_HEAD:
 663			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 664				continue;
 665			bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 666			break;
 
 
 
 
 
 667		case BPF_LIST_NODE:
 
 
 668			break;
 669		default:
 670			WARN_ON_ONCE(1);
 671			continue;
 672		}
 673	}
 674}
 675
 676/* called from workqueue */
 677static void bpf_map_free_deferred(struct work_struct *work)
 678{
 679	struct bpf_map *map = container_of(work, struct bpf_map, work);
 680	struct btf_field_offs *foffs = map->field_offs;
 681	struct btf_record *rec = map->record;
 
 682
 683	security_bpf_map_free(map);
 684	bpf_map_release_memcg(map);
 685	/* implementation dependent freeing */
 686	map->ops->map_free(map);
 687	/* Delay freeing of field_offs and btf_record for maps, as map_free
 688	 * callback usually needs access to them. It is better to do it here
 689	 * than require each callback to do the free itself manually.
 690	 *
 691	 * Note that the btf_record stashed in map->inner_map_meta->record was
 692	 * already freed using the map_free callback for map in map case which
 693	 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 694	 * template bpf_map struct used during verification.
 695	 */
 696	kfree(foffs);
 697	btf_record_free(rec);
 
 
 
 
 698}
 699
 700static void bpf_map_put_uref(struct bpf_map *map)
 701{
 702	if (atomic64_dec_and_test(&map->usercnt)) {
 703		if (map->ops->map_release_uref)
 704			map->ops->map_release_uref(map);
 705	}
 706}
 707
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 708/* decrement map refcnt and schedule it for freeing via workqueue
 709 * (unrelying map implementation ops->map_free() might sleep)
 710 */
 711static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 712{
 713	if (atomic64_dec_and_test(&map->refcnt)) {
 714		/* bpf_map_free_id() must be called first */
 715		bpf_map_free_id(map, do_idr_lock);
 716		btf_put(map->btf);
 717		INIT_WORK(&map->work, bpf_map_free_deferred);
 718		/* Avoid spawning kworkers, since they all might contend
 719		 * for the same mutex like slab_mutex.
 720		 */
 721		queue_work(system_unbound_wq, &map->work);
 722	}
 723}
 724
 725void bpf_map_put(struct bpf_map *map)
 726{
 727	__bpf_map_put(map, true);
 
 
 
 
 
 728}
 729EXPORT_SYMBOL_GPL(bpf_map_put);
 730
 731void bpf_map_put_with_uref(struct bpf_map *map)
 732{
 733	bpf_map_put_uref(map);
 734	bpf_map_put(map);
 735}
 736
 737static int bpf_map_release(struct inode *inode, struct file *filp)
 738{
 739	struct bpf_map *map = filp->private_data;
 740
 741	if (map->ops->map_release)
 742		map->ops->map_release(map, filp);
 743
 744	bpf_map_put_with_uref(map);
 745	return 0;
 746}
 747
 748static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 749{
 750	fmode_t mode = f.file->f_mode;
 751
 752	/* Our file permissions may have been overridden by global
 753	 * map permissions facing syscall side.
 754	 */
 755	if (READ_ONCE(map->frozen))
 756		mode &= ~FMODE_CAN_WRITE;
 757	return mode;
 758}
 759
 760#ifdef CONFIG_PROC_FS
 761/* Provides an approximation of the map's memory footprint.
 762 * Used only to provide a backward compatibility and display
 763 * a reasonable "memlock" info.
 764 */
 765static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
 766{
 767	unsigned long size;
 768
 769	size = round_up(map->key_size + bpf_map_value_size(map), 8);
 770
 771	return round_up(map->max_entries * size, PAGE_SIZE);
 772}
 773
 774static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 775{
 776	struct bpf_map *map = filp->private_data;
 777	u32 type = 0, jited = 0;
 778
 779	if (map_type_contains_progs(map)) {
 780		spin_lock(&map->owner.lock);
 781		type  = map->owner.type;
 782		jited = map->owner.jited;
 783		spin_unlock(&map->owner.lock);
 784	}
 785
 786	seq_printf(m,
 787		   "map_type:\t%u\n"
 788		   "key_size:\t%u\n"
 789		   "value_size:\t%u\n"
 790		   "max_entries:\t%u\n"
 791		   "map_flags:\t%#x\n"
 792		   "map_extra:\t%#llx\n"
 793		   "memlock:\t%lu\n"
 794		   "map_id:\t%u\n"
 795		   "frozen:\t%u\n",
 796		   map->map_type,
 797		   map->key_size,
 798		   map->value_size,
 799		   map->max_entries,
 800		   map->map_flags,
 801		   (unsigned long long)map->map_extra,
 802		   bpf_map_memory_footprint(map),
 803		   map->id,
 804		   READ_ONCE(map->frozen));
 805	if (type) {
 806		seq_printf(m, "owner_prog_type:\t%u\n", type);
 807		seq_printf(m, "owner_jited:\t%u\n", jited);
 808	}
 809}
 810#endif
 811
 812static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 813			      loff_t *ppos)
 814{
 815	/* We need this handler such that alloc_file() enables
 816	 * f_mode with FMODE_CAN_READ.
 817	 */
 818	return -EINVAL;
 819}
 820
 821static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 822			       size_t siz, loff_t *ppos)
 823{
 824	/* We need this handler such that alloc_file() enables
 825	 * f_mode with FMODE_CAN_WRITE.
 826	 */
 827	return -EINVAL;
 828}
 829
 830/* called for any extra memory-mapped regions (except initial) */
 831static void bpf_map_mmap_open(struct vm_area_struct *vma)
 832{
 833	struct bpf_map *map = vma->vm_file->private_data;
 834
 835	if (vma->vm_flags & VM_MAYWRITE)
 836		bpf_map_write_active_inc(map);
 837}
 838
 839/* called for all unmapped memory region (including initial) */
 840static void bpf_map_mmap_close(struct vm_area_struct *vma)
 841{
 842	struct bpf_map *map = vma->vm_file->private_data;
 843
 844	if (vma->vm_flags & VM_MAYWRITE)
 845		bpf_map_write_active_dec(map);
 846}
 847
 848static const struct vm_operations_struct bpf_map_default_vmops = {
 849	.open		= bpf_map_mmap_open,
 850	.close		= bpf_map_mmap_close,
 851};
 852
 853static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 854{
 855	struct bpf_map *map = filp->private_data;
 856	int err;
 857
 858	if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 859		return -ENOTSUPP;
 860
 861	if (!(vma->vm_flags & VM_SHARED))
 862		return -EINVAL;
 863
 864	mutex_lock(&map->freeze_mutex);
 865
 866	if (vma->vm_flags & VM_WRITE) {
 867		if (map->frozen) {
 868			err = -EPERM;
 869			goto out;
 870		}
 871		/* map is meant to be read-only, so do not allow mapping as
 872		 * writable, because it's possible to leak a writable page
 873		 * reference and allows user-space to still modify it after
 874		 * freezing, while verifier will assume contents do not change
 875		 */
 876		if (map->map_flags & BPF_F_RDONLY_PROG) {
 877			err = -EACCES;
 878			goto out;
 879		}
 880	}
 881
 882	/* set default open/close callbacks */
 883	vma->vm_ops = &bpf_map_default_vmops;
 884	vma->vm_private_data = map;
 885	vma->vm_flags &= ~VM_MAYEXEC;
 886	if (!(vma->vm_flags & VM_WRITE))
 887		/* disallow re-mapping with PROT_WRITE */
 888		vma->vm_flags &= ~VM_MAYWRITE;
 889
 890	err = map->ops->map_mmap(map, vma);
 891	if (err)
 892		goto out;
 893
 894	if (vma->vm_flags & VM_MAYWRITE)
 895		bpf_map_write_active_inc(map);
 896out:
 897	mutex_unlock(&map->freeze_mutex);
 898	return err;
 899}
 900
 901static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 902{
 903	struct bpf_map *map = filp->private_data;
 904
 905	if (map->ops->map_poll)
 906		return map->ops->map_poll(map, filp, pts);
 907
 908	return EPOLLERR;
 909}
 910
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 911const struct file_operations bpf_map_fops = {
 912#ifdef CONFIG_PROC_FS
 913	.show_fdinfo	= bpf_map_show_fdinfo,
 914#endif
 915	.release	= bpf_map_release,
 916	.read		= bpf_dummy_read,
 917	.write		= bpf_dummy_write,
 918	.mmap		= bpf_map_mmap,
 919	.poll		= bpf_map_poll,
 
 920};
 921
 922int bpf_map_new_fd(struct bpf_map *map, int flags)
 923{
 924	int ret;
 925
 926	ret = security_bpf_map(map, OPEN_FMODE(flags));
 927	if (ret < 0)
 928		return ret;
 929
 930	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 931				flags | O_CLOEXEC);
 932}
 933
 934int bpf_get_file_flag(int flags)
 935{
 936	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 937		return -EINVAL;
 938	if (flags & BPF_F_RDONLY)
 939		return O_RDONLY;
 940	if (flags & BPF_F_WRONLY)
 941		return O_WRONLY;
 942	return O_RDWR;
 943}
 944
 945/* helper macro to check that unused fields 'union bpf_attr' are zero */
 946#define CHECK_ATTR(CMD) \
 947	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 948		   sizeof(attr->CMD##_LAST_FIELD), 0, \
 949		   sizeof(*attr) - \
 950		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 951		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
 952
 953/* dst and src must have at least "size" number of bytes.
 954 * Return strlen on success and < 0 on error.
 955 */
 956int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 957{
 958	const char *end = src + size;
 959	const char *orig_src = src;
 960
 961	memset(dst, 0, size);
 962	/* Copy all isalnum(), '_' and '.' chars. */
 963	while (src < end && *src) {
 964		if (!isalnum(*src) &&
 965		    *src != '_' && *src != '.')
 966			return -EINVAL;
 967		*dst++ = *src++;
 968	}
 969
 970	/* No '\0' found in "size" number of bytes */
 971	if (src == end)
 972		return -EINVAL;
 973
 974	return src - orig_src;
 975}
 976
 977int map_check_no_btf(const struct bpf_map *map,
 978		     const struct btf *btf,
 979		     const struct btf_type *key_type,
 980		     const struct btf_type *value_type)
 981{
 982	return -ENOTSUPP;
 983}
 984
 985static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 986			 u32 btf_key_id, u32 btf_value_id)
 987{
 988	const struct btf_type *key_type, *value_type;
 989	u32 key_size, value_size;
 990	int ret = 0;
 991
 992	/* Some maps allow key to be unspecified. */
 993	if (btf_key_id) {
 994		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 995		if (!key_type || key_size != map->key_size)
 996			return -EINVAL;
 997	} else {
 998		key_type = btf_type_by_id(btf, 0);
 999		if (!map->ops->map_check_btf)
1000			return -EINVAL;
1001	}
1002
1003	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1004	if (!value_type || value_size != map->value_size)
1005		return -EINVAL;
1006
1007	map->record = btf_parse_fields(btf, value_type,
1008				       BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD,
 
1009				       map->value_size);
1010	if (!IS_ERR_OR_NULL(map->record)) {
1011		int i;
1012
1013		if (!bpf_capable()) {
1014			ret = -EPERM;
1015			goto free_map_tab;
1016		}
1017		if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1018			ret = -EACCES;
1019			goto free_map_tab;
1020		}
1021		for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1022			switch (map->record->field_mask & (1 << i)) {
1023			case 0:
1024				continue;
1025			case BPF_SPIN_LOCK:
1026				if (map->map_type != BPF_MAP_TYPE_HASH &&
1027				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1028				    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1029				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1030				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1031				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1032				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1033					ret = -EOPNOTSUPP;
1034					goto free_map_tab;
1035				}
1036				break;
1037			case BPF_TIMER:
1038				if (map->map_type != BPF_MAP_TYPE_HASH &&
1039				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1040				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1041					ret = -EOPNOTSUPP;
1042					goto free_map_tab;
1043				}
1044				break;
1045			case BPF_KPTR_UNREF:
1046			case BPF_KPTR_REF:
 
 
1047				if (map->map_type != BPF_MAP_TYPE_HASH &&
 
1048				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
 
1049				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1050				    map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) {
 
 
 
 
1051					ret = -EOPNOTSUPP;
1052					goto free_map_tab;
1053				}
1054				break;
1055			case BPF_LIST_HEAD:
 
1056				if (map->map_type != BPF_MAP_TYPE_HASH &&
1057				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1058				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1059					ret = -EOPNOTSUPP;
1060					goto free_map_tab;
1061				}
1062				break;
1063			default:
1064				/* Fail if map_type checks are missing for a field type */
1065				ret = -EOPNOTSUPP;
1066				goto free_map_tab;
1067			}
1068		}
1069	}
1070
1071	ret = btf_check_and_fixup_fields(btf, map->record);
1072	if (ret < 0)
1073		goto free_map_tab;
1074
1075	if (map->ops->map_check_btf) {
1076		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1077		if (ret < 0)
1078			goto free_map_tab;
1079	}
1080
1081	return ret;
1082free_map_tab:
1083	bpf_map_free_record(map);
1084	return ret;
1085}
1086
1087#define BPF_MAP_CREATE_LAST_FIELD map_extra
 
 
 
 
 
1088/* called via syscall */
1089static int map_create(union bpf_attr *attr)
1090{
 
 
1091	int numa_node = bpf_map_attr_numa_node(attr);
1092	struct btf_field_offs *foffs;
1093	struct bpf_map *map;
 
1094	int f_flags;
1095	int err;
1096
1097	err = CHECK_ATTR(BPF_MAP_CREATE);
1098	if (err)
1099		return -EINVAL;
1100
 
 
 
 
 
 
1101	if (attr->btf_vmlinux_value_type_id) {
1102		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1103		    attr->btf_key_type_id || attr->btf_value_type_id)
1104			return -EINVAL;
1105	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1106		return -EINVAL;
1107	}
1108
1109	if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
 
1110	    attr->map_extra != 0)
1111		return -EINVAL;
1112
1113	f_flags = bpf_get_file_flag(attr->map_flags);
1114	if (f_flags < 0)
1115		return f_flags;
1116
1117	if (numa_node != NUMA_NO_NODE &&
1118	    ((unsigned int)numa_node >= nr_node_ids ||
1119	     !node_online(numa_node)))
1120		return -EINVAL;
1121
1122	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1123	map = find_and_alloc_map(attr);
1124	if (IS_ERR(map))
1125		return PTR_ERR(map);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1126
1127	err = bpf_obj_name_cpy(map->name, attr->map_name,
1128			       sizeof(attr->map_name));
1129	if (err < 0)
1130		goto free_map;
1131
1132	atomic64_set(&map->refcnt, 1);
1133	atomic64_set(&map->usercnt, 1);
1134	mutex_init(&map->freeze_mutex);
1135	spin_lock_init(&map->owner.lock);
1136
1137	if (attr->btf_key_type_id || attr->btf_value_type_id ||
1138	    /* Even the map's value is a kernel's struct,
1139	     * the bpf_prog.o must have BTF to begin with
1140	     * to figure out the corresponding kernel's
1141	     * counter part.  Thus, attr->btf_fd has
1142	     * to be valid also.
1143	     */
1144	    attr->btf_vmlinux_value_type_id) {
1145		struct btf *btf;
1146
1147		btf = btf_get_by_fd(attr->btf_fd);
1148		if (IS_ERR(btf)) {
1149			err = PTR_ERR(btf);
1150			goto free_map;
1151		}
1152		if (btf_is_kernel(btf)) {
1153			btf_put(btf);
1154			err = -EACCES;
1155			goto free_map;
1156		}
1157		map->btf = btf;
1158
1159		if (attr->btf_value_type_id) {
1160			err = map_check_btf(map, btf, attr->btf_key_type_id,
1161					    attr->btf_value_type_id);
1162			if (err)
1163				goto free_map;
1164		}
1165
1166		map->btf_key_type_id = attr->btf_key_type_id;
1167		map->btf_value_type_id = attr->btf_value_type_id;
1168		map->btf_vmlinux_value_type_id =
1169			attr->btf_vmlinux_value_type_id;
1170	}
1171
1172
1173	foffs = btf_parse_field_offs(map->record);
1174	if (IS_ERR(foffs)) {
1175		err = PTR_ERR(foffs);
1176		goto free_map;
1177	}
1178	map->field_offs = foffs;
1179
1180	err = security_bpf_map_alloc(map);
1181	if (err)
1182		goto free_map_field_offs;
1183
1184	err = bpf_map_alloc_id(map);
1185	if (err)
1186		goto free_map_sec;
1187
1188	bpf_map_save_memcg(map);
 
1189
1190	err = bpf_map_new_fd(map, f_flags);
1191	if (err < 0) {
1192		/* failed to allocate fd.
1193		 * bpf_map_put_with_uref() is needed because the above
1194		 * bpf_map_alloc_id() has published the map
1195		 * to the userspace and the userspace may
1196		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1197		 */
1198		bpf_map_put_with_uref(map);
1199		return err;
1200	}
1201
1202	return err;
1203
1204free_map_sec:
1205	security_bpf_map_free(map);
1206free_map_field_offs:
1207	kfree(map->field_offs);
1208free_map:
1209	btf_put(map->btf);
1210	map->ops->map_free(map);
 
 
1211	return err;
1212}
1213
1214/* if error is returned, fd is released.
1215 * On success caller should complete fd access with matching fdput()
1216 */
1217struct bpf_map *__bpf_map_get(struct fd f)
1218{
1219	if (!f.file)
1220		return ERR_PTR(-EBADF);
1221	if (f.file->f_op != &bpf_map_fops) {
1222		fdput(f);
1223		return ERR_PTR(-EINVAL);
1224	}
1225
1226	return f.file->private_data;
1227}
1228
1229void bpf_map_inc(struct bpf_map *map)
1230{
1231	atomic64_inc(&map->refcnt);
1232}
1233EXPORT_SYMBOL_GPL(bpf_map_inc);
1234
1235void bpf_map_inc_with_uref(struct bpf_map *map)
1236{
1237	atomic64_inc(&map->refcnt);
1238	atomic64_inc(&map->usercnt);
1239}
1240EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1241
1242struct bpf_map *bpf_map_get(u32 ufd)
1243{
1244	struct fd f = fdget(ufd);
1245	struct bpf_map *map;
1246
1247	map = __bpf_map_get(f);
1248	if (IS_ERR(map))
1249		return map;
1250
1251	bpf_map_inc(map);
1252	fdput(f);
1253
1254	return map;
1255}
1256EXPORT_SYMBOL(bpf_map_get);
1257
1258struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1259{
1260	struct fd f = fdget(ufd);
1261	struct bpf_map *map;
1262
1263	map = __bpf_map_get(f);
1264	if (IS_ERR(map))
1265		return map;
1266
1267	bpf_map_inc_with_uref(map);
1268	fdput(f);
1269
1270	return map;
1271}
1272
1273/* map_idr_lock should have been held */
1274static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 
 
1275{
1276	int refold;
1277
1278	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1279	if (!refold)
1280		return ERR_PTR(-ENOENT);
1281	if (uref)
1282		atomic64_inc(&map->usercnt);
1283
1284	return map;
1285}
1286
1287struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1288{
1289	spin_lock_bh(&map_idr_lock);
1290	map = __bpf_map_inc_not_zero(map, false);
1291	spin_unlock_bh(&map_idr_lock);
1292
1293	return map;
1294}
1295EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1296
1297int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1298{
1299	return -ENOTSUPP;
1300}
1301
1302static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1303{
1304	if (key_size)
1305		return vmemdup_user(ukey, key_size);
1306
1307	if (ukey)
1308		return ERR_PTR(-EINVAL);
1309
1310	return NULL;
1311}
1312
1313static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1314{
1315	if (key_size)
1316		return kvmemdup_bpfptr(ukey, key_size);
1317
1318	if (!bpfptr_is_null(ukey))
1319		return ERR_PTR(-EINVAL);
1320
1321	return NULL;
1322}
1323
1324/* last field in 'union bpf_attr' used by this command */
1325#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1326
1327static int map_lookup_elem(union bpf_attr *attr)
1328{
1329	void __user *ukey = u64_to_user_ptr(attr->key);
1330	void __user *uvalue = u64_to_user_ptr(attr->value);
1331	int ufd = attr->map_fd;
1332	struct bpf_map *map;
1333	void *key, *value;
1334	u32 value_size;
1335	struct fd f;
1336	int err;
1337
1338	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1339		return -EINVAL;
1340
1341	if (attr->flags & ~BPF_F_LOCK)
1342		return -EINVAL;
1343
1344	f = fdget(ufd);
1345	map = __bpf_map_get(f);
1346	if (IS_ERR(map))
1347		return PTR_ERR(map);
1348	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1349		err = -EPERM;
1350		goto err_put;
1351	}
1352
1353	if ((attr->flags & BPF_F_LOCK) &&
1354	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1355		err = -EINVAL;
1356		goto err_put;
1357	}
1358
1359	key = __bpf_copy_key(ukey, map->key_size);
1360	if (IS_ERR(key)) {
1361		err = PTR_ERR(key);
1362		goto err_put;
1363	}
1364
1365	value_size = bpf_map_value_size(map);
1366
1367	err = -ENOMEM;
1368	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1369	if (!value)
1370		goto free_key;
1371
1372	if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1373		if (copy_from_user(value, uvalue, value_size))
1374			err = -EFAULT;
1375		else
1376			err = bpf_map_copy_value(map, key, value, attr->flags);
1377		goto free_value;
1378	}
1379
1380	err = bpf_map_copy_value(map, key, value, attr->flags);
1381	if (err)
1382		goto free_value;
1383
1384	err = -EFAULT;
1385	if (copy_to_user(uvalue, value, value_size) != 0)
1386		goto free_value;
1387
1388	err = 0;
1389
1390free_value:
1391	kvfree(value);
1392free_key:
1393	kvfree(key);
1394err_put:
1395	fdput(f);
1396	return err;
1397}
1398
1399
1400#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1401
1402static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1403{
1404	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1405	bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1406	int ufd = attr->map_fd;
1407	struct bpf_map *map;
1408	void *key, *value;
1409	u32 value_size;
1410	struct fd f;
1411	int err;
1412
1413	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1414		return -EINVAL;
1415
1416	f = fdget(ufd);
1417	map = __bpf_map_get(f);
1418	if (IS_ERR(map))
1419		return PTR_ERR(map);
1420	bpf_map_write_active_inc(map);
1421	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1422		err = -EPERM;
1423		goto err_put;
1424	}
1425
1426	if ((attr->flags & BPF_F_LOCK) &&
1427	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1428		err = -EINVAL;
1429		goto err_put;
1430	}
1431
1432	key = ___bpf_copy_key(ukey, map->key_size);
1433	if (IS_ERR(key)) {
1434		err = PTR_ERR(key);
1435		goto err_put;
1436	}
1437
1438	value_size = bpf_map_value_size(map);
1439	value = kvmemdup_bpfptr(uvalue, value_size);
1440	if (IS_ERR(value)) {
1441		err = PTR_ERR(value);
1442		goto free_key;
1443	}
1444
1445	err = bpf_map_update_value(map, f.file, key, value, attr->flags);
 
 
1446
1447	kvfree(value);
1448free_key:
1449	kvfree(key);
1450err_put:
1451	bpf_map_write_active_dec(map);
1452	fdput(f);
1453	return err;
1454}
1455
1456#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1457
1458static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1459{
1460	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1461	int ufd = attr->map_fd;
1462	struct bpf_map *map;
1463	struct fd f;
1464	void *key;
1465	int err;
1466
1467	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1468		return -EINVAL;
1469
1470	f = fdget(ufd);
1471	map = __bpf_map_get(f);
1472	if (IS_ERR(map))
1473		return PTR_ERR(map);
1474	bpf_map_write_active_inc(map);
1475	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1476		err = -EPERM;
1477		goto err_put;
1478	}
1479
1480	key = ___bpf_copy_key(ukey, map->key_size);
1481	if (IS_ERR(key)) {
1482		err = PTR_ERR(key);
1483		goto err_put;
1484	}
1485
1486	if (bpf_map_is_dev_bound(map)) {
1487		err = bpf_map_offload_delete_elem(map, key);
1488		goto out;
1489	} else if (IS_FD_PROG_ARRAY(map) ||
1490		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1491		/* These maps require sleepable context */
1492		err = map->ops->map_delete_elem(map, key);
1493		goto out;
1494	}
1495
1496	bpf_disable_instrumentation();
1497	rcu_read_lock();
1498	err = map->ops->map_delete_elem(map, key);
1499	rcu_read_unlock();
1500	bpf_enable_instrumentation();
1501	maybe_wait_bpf_programs(map);
 
1502out:
1503	kvfree(key);
1504err_put:
1505	bpf_map_write_active_dec(map);
1506	fdput(f);
1507	return err;
1508}
1509
1510/* last field in 'union bpf_attr' used by this command */
1511#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1512
1513static int map_get_next_key(union bpf_attr *attr)
1514{
1515	void __user *ukey = u64_to_user_ptr(attr->key);
1516	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1517	int ufd = attr->map_fd;
1518	struct bpf_map *map;
1519	void *key, *next_key;
1520	struct fd f;
1521	int err;
1522
1523	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1524		return -EINVAL;
1525
1526	f = fdget(ufd);
1527	map = __bpf_map_get(f);
1528	if (IS_ERR(map))
1529		return PTR_ERR(map);
1530	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1531		err = -EPERM;
1532		goto err_put;
1533	}
1534
1535	if (ukey) {
1536		key = __bpf_copy_key(ukey, map->key_size);
1537		if (IS_ERR(key)) {
1538			err = PTR_ERR(key);
1539			goto err_put;
1540		}
1541	} else {
1542		key = NULL;
1543	}
1544
1545	err = -ENOMEM;
1546	next_key = kvmalloc(map->key_size, GFP_USER);
1547	if (!next_key)
1548		goto free_key;
1549
1550	if (bpf_map_is_dev_bound(map)) {
1551		err = bpf_map_offload_get_next_key(map, key, next_key);
1552		goto out;
1553	}
1554
1555	rcu_read_lock();
1556	err = map->ops->map_get_next_key(map, key, next_key);
1557	rcu_read_unlock();
1558out:
1559	if (err)
1560		goto free_next_key;
1561
1562	err = -EFAULT;
1563	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1564		goto free_next_key;
1565
1566	err = 0;
1567
1568free_next_key:
1569	kvfree(next_key);
1570free_key:
1571	kvfree(key);
1572err_put:
1573	fdput(f);
1574	return err;
1575}
1576
1577int generic_map_delete_batch(struct bpf_map *map,
1578			     const union bpf_attr *attr,
1579			     union bpf_attr __user *uattr)
1580{
1581	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1582	u32 cp, max_count;
1583	int err = 0;
1584	void *key;
1585
1586	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1587		return -EINVAL;
1588
1589	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1590	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1591		return -EINVAL;
1592	}
1593
1594	max_count = attr->batch.count;
1595	if (!max_count)
1596		return 0;
1597
 
 
 
1598	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1599	if (!key)
1600		return -ENOMEM;
1601
1602	for (cp = 0; cp < max_count; cp++) {
1603		err = -EFAULT;
1604		if (copy_from_user(key, keys + cp * map->key_size,
1605				   map->key_size))
1606			break;
1607
1608		if (bpf_map_is_dev_bound(map)) {
1609			err = bpf_map_offload_delete_elem(map, key);
1610			break;
1611		}
1612
1613		bpf_disable_instrumentation();
1614		rcu_read_lock();
1615		err = map->ops->map_delete_elem(map, key);
1616		rcu_read_unlock();
1617		bpf_enable_instrumentation();
1618		if (err)
1619			break;
1620		cond_resched();
1621	}
1622	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1623		err = -EFAULT;
1624
1625	kvfree(key);
1626
1627	maybe_wait_bpf_programs(map);
1628	return err;
1629}
1630
1631int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1632			     const union bpf_attr *attr,
1633			     union bpf_attr __user *uattr)
1634{
1635	void __user *values = u64_to_user_ptr(attr->batch.values);
1636	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1637	u32 value_size, cp, max_count;
1638	void *key, *value;
1639	int err = 0;
1640
1641	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1642		return -EINVAL;
1643
1644	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1645	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1646		return -EINVAL;
1647	}
1648
1649	value_size = bpf_map_value_size(map);
1650
1651	max_count = attr->batch.count;
1652	if (!max_count)
1653		return 0;
1654
 
 
 
1655	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1656	if (!key)
1657		return -ENOMEM;
1658
1659	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1660	if (!value) {
1661		kvfree(key);
1662		return -ENOMEM;
1663	}
1664
1665	for (cp = 0; cp < max_count; cp++) {
1666		err = -EFAULT;
1667		if (copy_from_user(key, keys + cp * map->key_size,
1668		    map->key_size) ||
1669		    copy_from_user(value, values + cp * value_size, value_size))
1670			break;
1671
1672		err = bpf_map_update_value(map, map_file, key, value,
1673					   attr->batch.elem_flags);
1674
1675		if (err)
1676			break;
1677		cond_resched();
1678	}
1679
1680	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1681		err = -EFAULT;
1682
1683	kvfree(value);
1684	kvfree(key);
 
1685	return err;
1686}
1687
1688#define MAP_LOOKUP_RETRIES 3
1689
1690int generic_map_lookup_batch(struct bpf_map *map,
1691				    const union bpf_attr *attr,
1692				    union bpf_attr __user *uattr)
1693{
1694	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1695	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1696	void __user *values = u64_to_user_ptr(attr->batch.values);
1697	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1698	void *buf, *buf_prevkey, *prev_key, *key, *value;
1699	int err, retry = MAP_LOOKUP_RETRIES;
1700	u32 value_size, cp, max_count;
1701
1702	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1703		return -EINVAL;
1704
1705	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1706	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1707		return -EINVAL;
1708
1709	value_size = bpf_map_value_size(map);
1710
1711	max_count = attr->batch.count;
1712	if (!max_count)
1713		return 0;
1714
1715	if (put_user(0, &uattr->batch.count))
1716		return -EFAULT;
1717
1718	buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1719	if (!buf_prevkey)
1720		return -ENOMEM;
1721
1722	buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1723	if (!buf) {
1724		kvfree(buf_prevkey);
1725		return -ENOMEM;
1726	}
1727
1728	err = -EFAULT;
1729	prev_key = NULL;
1730	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1731		goto free_buf;
1732	key = buf;
1733	value = key + map->key_size;
1734	if (ubatch)
1735		prev_key = buf_prevkey;
1736
1737	for (cp = 0; cp < max_count;) {
1738		rcu_read_lock();
1739		err = map->ops->map_get_next_key(map, prev_key, key);
1740		rcu_read_unlock();
1741		if (err)
1742			break;
1743		err = bpf_map_copy_value(map, key, value,
1744					 attr->batch.elem_flags);
1745
1746		if (err == -ENOENT) {
1747			if (retry) {
1748				retry--;
1749				continue;
1750			}
1751			err = -EINTR;
1752			break;
1753		}
1754
1755		if (err)
1756			goto free_buf;
1757
1758		if (copy_to_user(keys + cp * map->key_size, key,
1759				 map->key_size)) {
1760			err = -EFAULT;
1761			goto free_buf;
1762		}
1763		if (copy_to_user(values + cp * value_size, value, value_size)) {
1764			err = -EFAULT;
1765			goto free_buf;
1766		}
1767
1768		if (!prev_key)
1769			prev_key = buf_prevkey;
1770
1771		swap(prev_key, key);
1772		retry = MAP_LOOKUP_RETRIES;
1773		cp++;
1774		cond_resched();
1775	}
1776
1777	if (err == -EFAULT)
1778		goto free_buf;
1779
1780	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1781		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1782		err = -EFAULT;
1783
1784free_buf:
1785	kvfree(buf_prevkey);
1786	kvfree(buf);
1787	return err;
1788}
1789
1790#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1791
1792static int map_lookup_and_delete_elem(union bpf_attr *attr)
1793{
1794	void __user *ukey = u64_to_user_ptr(attr->key);
1795	void __user *uvalue = u64_to_user_ptr(attr->value);
1796	int ufd = attr->map_fd;
1797	struct bpf_map *map;
1798	void *key, *value;
1799	u32 value_size;
1800	struct fd f;
1801	int err;
1802
1803	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1804		return -EINVAL;
1805
1806	if (attr->flags & ~BPF_F_LOCK)
1807		return -EINVAL;
1808
1809	f = fdget(ufd);
1810	map = __bpf_map_get(f);
1811	if (IS_ERR(map))
1812		return PTR_ERR(map);
1813	bpf_map_write_active_inc(map);
1814	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1815	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1816		err = -EPERM;
1817		goto err_put;
1818	}
1819
1820	if (attr->flags &&
1821	    (map->map_type == BPF_MAP_TYPE_QUEUE ||
1822	     map->map_type == BPF_MAP_TYPE_STACK)) {
1823		err = -EINVAL;
1824		goto err_put;
1825	}
1826
1827	if ((attr->flags & BPF_F_LOCK) &&
1828	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1829		err = -EINVAL;
1830		goto err_put;
1831	}
1832
1833	key = __bpf_copy_key(ukey, map->key_size);
1834	if (IS_ERR(key)) {
1835		err = PTR_ERR(key);
1836		goto err_put;
1837	}
1838
1839	value_size = bpf_map_value_size(map);
1840
1841	err = -ENOMEM;
1842	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1843	if (!value)
1844		goto free_key;
1845
1846	err = -ENOTSUPP;
1847	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1848	    map->map_type == BPF_MAP_TYPE_STACK) {
1849		err = map->ops->map_pop_elem(map, value);
1850	} else if (map->map_type == BPF_MAP_TYPE_HASH ||
1851		   map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1852		   map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1853		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1854		if (!bpf_map_is_dev_bound(map)) {
1855			bpf_disable_instrumentation();
1856			rcu_read_lock();
1857			err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1858			rcu_read_unlock();
1859			bpf_enable_instrumentation();
1860		}
1861	}
1862
1863	if (err)
1864		goto free_value;
1865
1866	if (copy_to_user(uvalue, value, value_size) != 0) {
1867		err = -EFAULT;
1868		goto free_value;
1869	}
1870
1871	err = 0;
1872
1873free_value:
1874	kvfree(value);
1875free_key:
1876	kvfree(key);
1877err_put:
1878	bpf_map_write_active_dec(map);
1879	fdput(f);
1880	return err;
1881}
1882
1883#define BPF_MAP_FREEZE_LAST_FIELD map_fd
1884
1885static int map_freeze(const union bpf_attr *attr)
1886{
1887	int err = 0, ufd = attr->map_fd;
1888	struct bpf_map *map;
1889	struct fd f;
1890
1891	if (CHECK_ATTR(BPF_MAP_FREEZE))
1892		return -EINVAL;
1893
1894	f = fdget(ufd);
1895	map = __bpf_map_get(f);
1896	if (IS_ERR(map))
1897		return PTR_ERR(map);
1898
1899	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1900		fdput(f);
1901		return -ENOTSUPP;
1902	}
1903
 
 
 
 
 
1904	mutex_lock(&map->freeze_mutex);
1905	if (bpf_map_write_active(map)) {
1906		err = -EBUSY;
1907		goto err_put;
1908	}
1909	if (READ_ONCE(map->frozen)) {
1910		err = -EBUSY;
1911		goto err_put;
1912	}
1913	if (!bpf_capable()) {
1914		err = -EPERM;
1915		goto err_put;
1916	}
1917
1918	WRITE_ONCE(map->frozen, true);
1919err_put:
1920	mutex_unlock(&map->freeze_mutex);
1921	fdput(f);
1922	return err;
1923}
1924
1925static const struct bpf_prog_ops * const bpf_prog_types[] = {
1926#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1927	[_id] = & _name ## _prog_ops,
1928#define BPF_MAP_TYPE(_id, _ops)
1929#define BPF_LINK_TYPE(_id, _name)
1930#include <linux/bpf_types.h>
1931#undef BPF_PROG_TYPE
1932#undef BPF_MAP_TYPE
1933#undef BPF_LINK_TYPE
1934};
1935
1936static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1937{
1938	const struct bpf_prog_ops *ops;
1939
1940	if (type >= ARRAY_SIZE(bpf_prog_types))
1941		return -EINVAL;
1942	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1943	ops = bpf_prog_types[type];
1944	if (!ops)
1945		return -EINVAL;
1946
1947	if (!bpf_prog_is_dev_bound(prog->aux))
1948		prog->aux->ops = ops;
1949	else
1950		prog->aux->ops = &bpf_offload_prog_ops;
1951	prog->type = type;
1952	return 0;
1953}
1954
1955enum bpf_audit {
1956	BPF_AUDIT_LOAD,
1957	BPF_AUDIT_UNLOAD,
1958	BPF_AUDIT_MAX,
1959};
1960
1961static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1962	[BPF_AUDIT_LOAD]   = "LOAD",
1963	[BPF_AUDIT_UNLOAD] = "UNLOAD",
1964};
1965
1966static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1967{
1968	struct audit_context *ctx = NULL;
1969	struct audit_buffer *ab;
1970
1971	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1972		return;
1973	if (audit_enabled == AUDIT_OFF)
1974		return;
1975	if (!in_irq() && !irqs_disabled())
1976		ctx = audit_context();
1977	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1978	if (unlikely(!ab))
1979		return;
1980	audit_log_format(ab, "prog-id=%u op=%s",
1981			 prog->aux->id, bpf_audit_str[op]);
1982	audit_log_end(ab);
1983}
1984
1985static int bpf_prog_alloc_id(struct bpf_prog *prog)
1986{
1987	int id;
1988
1989	idr_preload(GFP_KERNEL);
1990	spin_lock_bh(&prog_idr_lock);
1991	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1992	if (id > 0)
1993		prog->aux->id = id;
1994	spin_unlock_bh(&prog_idr_lock);
1995	idr_preload_end();
1996
1997	/* id is in [1, INT_MAX) */
1998	if (WARN_ON_ONCE(!id))
1999		return -ENOSPC;
2000
2001	return id > 0 ? 0 : id;
2002}
2003
2004void bpf_prog_free_id(struct bpf_prog *prog)
2005{
2006	unsigned long flags;
2007
2008	/* cBPF to eBPF migrations are currently not in the idr store.
2009	 * Offloaded programs are removed from the store when their device
2010	 * disappears - even if someone grabs an fd to them they are unusable,
2011	 * simply waiting for refcnt to drop to be freed.
2012	 */
2013	if (!prog->aux->id)
2014		return;
2015
2016	spin_lock_irqsave(&prog_idr_lock, flags);
2017	idr_remove(&prog_idr, prog->aux->id);
2018	prog->aux->id = 0;
2019	spin_unlock_irqrestore(&prog_idr_lock, flags);
2020}
2021
2022static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2023{
2024	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2025
2026	kvfree(aux->func_info);
2027	kfree(aux->func_info_aux);
2028	free_uid(aux->user);
2029	security_bpf_prog_free(aux);
2030	bpf_prog_free(aux->prog);
2031}
2032
2033static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2034{
2035	bpf_prog_kallsyms_del_all(prog);
2036	btf_put(prog->aux->btf);
 
2037	kvfree(prog->aux->jited_linfo);
2038	kvfree(prog->aux->linfo);
2039	kfree(prog->aux->kfunc_tab);
2040	if (prog->aux->attach_btf)
2041		btf_put(prog->aux->attach_btf);
2042
2043	if (deferred) {
2044		if (prog->aux->sleepable)
2045			call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2046		else
2047			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2048	} else {
2049		__bpf_prog_put_rcu(&prog->aux->rcu);
2050	}
2051}
2052
2053static void bpf_prog_put_deferred(struct work_struct *work)
2054{
2055	struct bpf_prog_aux *aux;
2056	struct bpf_prog *prog;
2057
2058	aux = container_of(work, struct bpf_prog_aux, work);
2059	prog = aux->prog;
2060	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2061	bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2062	bpf_prog_free_id(prog);
2063	__bpf_prog_put_noref(prog, true);
2064}
2065
2066static void __bpf_prog_put(struct bpf_prog *prog)
2067{
2068	struct bpf_prog_aux *aux = prog->aux;
2069
2070	if (atomic64_dec_and_test(&aux->refcnt)) {
2071		if (in_irq() || irqs_disabled()) {
2072			INIT_WORK(&aux->work, bpf_prog_put_deferred);
2073			schedule_work(&aux->work);
2074		} else {
2075			bpf_prog_put_deferred(&aux->work);
2076		}
2077	}
2078}
2079
2080void bpf_prog_put(struct bpf_prog *prog)
2081{
2082	__bpf_prog_put(prog);
2083}
2084EXPORT_SYMBOL_GPL(bpf_prog_put);
2085
2086static int bpf_prog_release(struct inode *inode, struct file *filp)
2087{
2088	struct bpf_prog *prog = filp->private_data;
2089
2090	bpf_prog_put(prog);
2091	return 0;
2092}
2093
2094struct bpf_prog_kstats {
2095	u64 nsecs;
2096	u64 cnt;
2097	u64 misses;
2098};
2099
2100void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2101{
2102	struct bpf_prog_stats *stats;
2103	unsigned int flags;
2104
2105	stats = this_cpu_ptr(prog->stats);
2106	flags = u64_stats_update_begin_irqsave(&stats->syncp);
2107	u64_stats_inc(&stats->misses);
2108	u64_stats_update_end_irqrestore(&stats->syncp, flags);
2109}
2110
2111static void bpf_prog_get_stats(const struct bpf_prog *prog,
2112			       struct bpf_prog_kstats *stats)
2113{
2114	u64 nsecs = 0, cnt = 0, misses = 0;
2115	int cpu;
2116
2117	for_each_possible_cpu(cpu) {
2118		const struct bpf_prog_stats *st;
2119		unsigned int start;
2120		u64 tnsecs, tcnt, tmisses;
2121
2122		st = per_cpu_ptr(prog->stats, cpu);
2123		do {
2124			start = u64_stats_fetch_begin(&st->syncp);
2125			tnsecs = u64_stats_read(&st->nsecs);
2126			tcnt = u64_stats_read(&st->cnt);
2127			tmisses = u64_stats_read(&st->misses);
2128		} while (u64_stats_fetch_retry(&st->syncp, start));
2129		nsecs += tnsecs;
2130		cnt += tcnt;
2131		misses += tmisses;
2132	}
2133	stats->nsecs = nsecs;
2134	stats->cnt = cnt;
2135	stats->misses = misses;
2136}
2137
2138#ifdef CONFIG_PROC_FS
2139static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2140{
2141	const struct bpf_prog *prog = filp->private_data;
2142	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2143	struct bpf_prog_kstats stats;
2144
2145	bpf_prog_get_stats(prog, &stats);
2146	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2147	seq_printf(m,
2148		   "prog_type:\t%u\n"
2149		   "prog_jited:\t%u\n"
2150		   "prog_tag:\t%s\n"
2151		   "memlock:\t%llu\n"
2152		   "prog_id:\t%u\n"
2153		   "run_time_ns:\t%llu\n"
2154		   "run_cnt:\t%llu\n"
2155		   "recursion_misses:\t%llu\n"
2156		   "verified_insns:\t%u\n",
2157		   prog->type,
2158		   prog->jited,
2159		   prog_tag,
2160		   prog->pages * 1ULL << PAGE_SHIFT,
2161		   prog->aux->id,
2162		   stats.nsecs,
2163		   stats.cnt,
2164		   stats.misses,
2165		   prog->aux->verified_insns);
2166}
2167#endif
2168
2169const struct file_operations bpf_prog_fops = {
2170#ifdef CONFIG_PROC_FS
2171	.show_fdinfo	= bpf_prog_show_fdinfo,
2172#endif
2173	.release	= bpf_prog_release,
2174	.read		= bpf_dummy_read,
2175	.write		= bpf_dummy_write,
2176};
2177
2178int bpf_prog_new_fd(struct bpf_prog *prog)
2179{
2180	int ret;
2181
2182	ret = security_bpf_prog(prog);
2183	if (ret < 0)
2184		return ret;
2185
2186	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2187				O_RDWR | O_CLOEXEC);
2188}
2189
2190static struct bpf_prog *____bpf_prog_get(struct fd f)
2191{
2192	if (!f.file)
2193		return ERR_PTR(-EBADF);
2194	if (f.file->f_op != &bpf_prog_fops) {
2195		fdput(f);
2196		return ERR_PTR(-EINVAL);
2197	}
2198
2199	return f.file->private_data;
2200}
2201
2202void bpf_prog_add(struct bpf_prog *prog, int i)
2203{
2204	atomic64_add(i, &prog->aux->refcnt);
2205}
2206EXPORT_SYMBOL_GPL(bpf_prog_add);
2207
2208void bpf_prog_sub(struct bpf_prog *prog, int i)
2209{
2210	/* Only to be used for undoing previous bpf_prog_add() in some
2211	 * error path. We still know that another entity in our call
2212	 * path holds a reference to the program, thus atomic_sub() can
2213	 * be safely used in such cases!
2214	 */
2215	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2216}
2217EXPORT_SYMBOL_GPL(bpf_prog_sub);
2218
2219void bpf_prog_inc(struct bpf_prog *prog)
2220{
2221	atomic64_inc(&prog->aux->refcnt);
2222}
2223EXPORT_SYMBOL_GPL(bpf_prog_inc);
2224
2225/* prog_idr_lock should have been held */
2226struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2227{
2228	int refold;
2229
2230	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2231
2232	if (!refold)
2233		return ERR_PTR(-ENOENT);
2234
2235	return prog;
2236}
2237EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2238
2239bool bpf_prog_get_ok(struct bpf_prog *prog,
2240			    enum bpf_prog_type *attach_type, bool attach_drv)
2241{
2242	/* not an attachment, just a refcount inc, always allow */
2243	if (!attach_type)
2244		return true;
2245
2246	if (prog->type != *attach_type)
2247		return false;
2248	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
2249		return false;
2250
2251	return true;
2252}
2253
2254static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2255				       bool attach_drv)
2256{
2257	struct fd f = fdget(ufd);
2258	struct bpf_prog *prog;
2259
2260	prog = ____bpf_prog_get(f);
2261	if (IS_ERR(prog))
2262		return prog;
2263	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2264		prog = ERR_PTR(-EINVAL);
2265		goto out;
2266	}
2267
2268	bpf_prog_inc(prog);
2269out:
2270	fdput(f);
2271	return prog;
2272}
2273
2274struct bpf_prog *bpf_prog_get(u32 ufd)
2275{
2276	return __bpf_prog_get(ufd, NULL, false);
2277}
2278
2279struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2280				       bool attach_drv)
2281{
2282	return __bpf_prog_get(ufd, &type, attach_drv);
2283}
2284EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2285
2286/* Initially all BPF programs could be loaded w/o specifying
2287 * expected_attach_type. Later for some of them specifying expected_attach_type
2288 * at load time became required so that program could be validated properly.
2289 * Programs of types that are allowed to be loaded both w/ and w/o (for
2290 * backward compatibility) expected_attach_type, should have the default attach
2291 * type assigned to expected_attach_type for the latter case, so that it can be
2292 * validated later at attach time.
2293 *
2294 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2295 * prog type requires it but has some attach types that have to be backward
2296 * compatible.
2297 */
2298static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2299{
2300	switch (attr->prog_type) {
2301	case BPF_PROG_TYPE_CGROUP_SOCK:
2302		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2303		 * exist so checking for non-zero is the way to go here.
2304		 */
2305		if (!attr->expected_attach_type)
2306			attr->expected_attach_type =
2307				BPF_CGROUP_INET_SOCK_CREATE;
2308		break;
2309	case BPF_PROG_TYPE_SK_REUSEPORT:
2310		if (!attr->expected_attach_type)
2311			attr->expected_attach_type =
2312				BPF_SK_REUSEPORT_SELECT;
2313		break;
2314	}
2315}
2316
2317static int
2318bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2319			   enum bpf_attach_type expected_attach_type,
2320			   struct btf *attach_btf, u32 btf_id,
2321			   struct bpf_prog *dst_prog)
2322{
2323	if (btf_id) {
2324		if (btf_id > BTF_MAX_TYPE)
2325			return -EINVAL;
2326
2327		if (!attach_btf && !dst_prog)
2328			return -EINVAL;
2329
2330		switch (prog_type) {
2331		case BPF_PROG_TYPE_TRACING:
2332		case BPF_PROG_TYPE_LSM:
2333		case BPF_PROG_TYPE_STRUCT_OPS:
2334		case BPF_PROG_TYPE_EXT:
2335			break;
2336		default:
2337			return -EINVAL;
2338		}
2339	}
2340
2341	if (attach_btf && (!btf_id || dst_prog))
2342		return -EINVAL;
2343
2344	if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2345	    prog_type != BPF_PROG_TYPE_EXT)
2346		return -EINVAL;
2347
2348	switch (prog_type) {
2349	case BPF_PROG_TYPE_CGROUP_SOCK:
2350		switch (expected_attach_type) {
2351		case BPF_CGROUP_INET_SOCK_CREATE:
2352		case BPF_CGROUP_INET_SOCK_RELEASE:
2353		case BPF_CGROUP_INET4_POST_BIND:
2354		case BPF_CGROUP_INET6_POST_BIND:
2355			return 0;
2356		default:
2357			return -EINVAL;
2358		}
2359	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2360		switch (expected_attach_type) {
2361		case BPF_CGROUP_INET4_BIND:
2362		case BPF_CGROUP_INET6_BIND:
2363		case BPF_CGROUP_INET4_CONNECT:
2364		case BPF_CGROUP_INET6_CONNECT:
 
2365		case BPF_CGROUP_INET4_GETPEERNAME:
2366		case BPF_CGROUP_INET6_GETPEERNAME:
 
2367		case BPF_CGROUP_INET4_GETSOCKNAME:
2368		case BPF_CGROUP_INET6_GETSOCKNAME:
 
2369		case BPF_CGROUP_UDP4_SENDMSG:
2370		case BPF_CGROUP_UDP6_SENDMSG:
 
2371		case BPF_CGROUP_UDP4_RECVMSG:
2372		case BPF_CGROUP_UDP6_RECVMSG:
 
2373			return 0;
2374		default:
2375			return -EINVAL;
2376		}
2377	case BPF_PROG_TYPE_CGROUP_SKB:
2378		switch (expected_attach_type) {
2379		case BPF_CGROUP_INET_INGRESS:
2380		case BPF_CGROUP_INET_EGRESS:
2381			return 0;
2382		default:
2383			return -EINVAL;
2384		}
2385	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2386		switch (expected_attach_type) {
2387		case BPF_CGROUP_SETSOCKOPT:
2388		case BPF_CGROUP_GETSOCKOPT:
2389			return 0;
2390		default:
2391			return -EINVAL;
2392		}
2393	case BPF_PROG_TYPE_SK_LOOKUP:
2394		if (expected_attach_type == BPF_SK_LOOKUP)
2395			return 0;
2396		return -EINVAL;
2397	case BPF_PROG_TYPE_SK_REUSEPORT:
2398		switch (expected_attach_type) {
2399		case BPF_SK_REUSEPORT_SELECT:
2400		case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2401			return 0;
2402		default:
2403			return -EINVAL;
2404		}
 
 
 
 
2405	case BPF_PROG_TYPE_SYSCALL:
2406	case BPF_PROG_TYPE_EXT:
2407		if (expected_attach_type)
2408			return -EINVAL;
2409		fallthrough;
2410	default:
2411		return 0;
2412	}
2413}
2414
2415static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2416{
2417	switch (prog_type) {
2418	case BPF_PROG_TYPE_SCHED_CLS:
2419	case BPF_PROG_TYPE_SCHED_ACT:
2420	case BPF_PROG_TYPE_XDP:
2421	case BPF_PROG_TYPE_LWT_IN:
2422	case BPF_PROG_TYPE_LWT_OUT:
2423	case BPF_PROG_TYPE_LWT_XMIT:
2424	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2425	case BPF_PROG_TYPE_SK_SKB:
2426	case BPF_PROG_TYPE_SK_MSG:
2427	case BPF_PROG_TYPE_LIRC_MODE2:
2428	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2429	case BPF_PROG_TYPE_CGROUP_DEVICE:
2430	case BPF_PROG_TYPE_CGROUP_SOCK:
2431	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2432	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2433	case BPF_PROG_TYPE_CGROUP_SYSCTL:
2434	case BPF_PROG_TYPE_SOCK_OPS:
2435	case BPF_PROG_TYPE_EXT: /* extends any prog */
 
2436		return true;
2437	case BPF_PROG_TYPE_CGROUP_SKB:
2438		/* always unpriv */
2439	case BPF_PROG_TYPE_SK_REUSEPORT:
2440		/* equivalent to SOCKET_FILTER. need CAP_BPF only */
2441	default:
2442		return false;
2443	}
2444}
2445
2446static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2447{
2448	switch (prog_type) {
2449	case BPF_PROG_TYPE_KPROBE:
2450	case BPF_PROG_TYPE_TRACEPOINT:
2451	case BPF_PROG_TYPE_PERF_EVENT:
2452	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2453	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2454	case BPF_PROG_TYPE_TRACING:
2455	case BPF_PROG_TYPE_LSM:
2456	case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2457	case BPF_PROG_TYPE_EXT: /* extends any prog */
2458		return true;
2459	default:
2460		return false;
2461	}
2462}
2463
2464/* last field in 'union bpf_attr' used by this command */
2465#define	BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
2466
2467static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2468{
2469	enum bpf_prog_type type = attr->prog_type;
2470	struct bpf_prog *prog, *dst_prog = NULL;
2471	struct btf *attach_btf = NULL;
 
 
2472	int err;
2473	char license[128];
2474	bool is_gpl;
2475
2476	if (CHECK_ATTR(BPF_PROG_LOAD))
2477		return -EINVAL;
2478
2479	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2480				 BPF_F_ANY_ALIGNMENT |
2481				 BPF_F_TEST_STATE_FREQ |
2482				 BPF_F_SLEEPABLE |
2483				 BPF_F_TEST_RND_HI32 |
2484				 BPF_F_XDP_HAS_FRAGS))
 
 
 
2485		return -EINVAL;
2486
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2487	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2488	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2489	    !bpf_capable())
2490		return -EPERM;
2491
2492	/* copy eBPF program license from user space */
2493	if (strncpy_from_bpfptr(license,
2494				make_bpfptr(attr->license, uattr.is_kernel),
2495				sizeof(license) - 1) < 0)
2496		return -EFAULT;
2497	license[sizeof(license) - 1] = 0;
2498
2499	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2500	is_gpl = license_is_gpl_compatible(license);
 
 
 
 
 
 
 
2501
2502	if (attr->insn_cnt == 0 ||
2503	    attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2504		return -E2BIG;
 
 
2505	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2506	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2507	    !bpf_capable())
2508		return -EPERM;
2509
2510	if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2511		return -EPERM;
2512	if (is_perfmon_prog_type(type) && !perfmon_capable())
2513		return -EPERM;
2514
2515	/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2516	 * or btf, we need to check which one it is
2517	 */
2518	if (attr->attach_prog_fd) {
2519		dst_prog = bpf_prog_get(attr->attach_prog_fd);
2520		if (IS_ERR(dst_prog)) {
2521			dst_prog = NULL;
2522			attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2523			if (IS_ERR(attach_btf))
2524				return -EINVAL;
 
 
2525			if (!btf_is_kernel(attach_btf)) {
2526				/* attaching through specifying bpf_prog's BTF
2527				 * objects directly might be supported eventually
2528				 */
2529				btf_put(attach_btf);
2530				return -ENOTSUPP;
 
2531			}
2532		}
2533	} else if (attr->attach_btf_id) {
2534		/* fall back to vmlinux BTF, if BTF type ID is specified */
2535		attach_btf = bpf_get_btf_vmlinux();
2536		if (IS_ERR(attach_btf))
2537			return PTR_ERR(attach_btf);
2538		if (!attach_btf)
2539			return -EINVAL;
 
 
 
 
2540		btf_get(attach_btf);
2541	}
2542
2543	bpf_prog_load_fixup_attach_type(attr);
2544	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2545				       attach_btf, attr->attach_btf_id,
2546				       dst_prog)) {
2547		if (dst_prog)
2548			bpf_prog_put(dst_prog);
2549		if (attach_btf)
2550			btf_put(attach_btf);
2551		return -EINVAL;
 
2552	}
2553
2554	/* plain bpf_prog allocation */
2555	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2556	if (!prog) {
2557		if (dst_prog)
2558			bpf_prog_put(dst_prog);
2559		if (attach_btf)
2560			btf_put(attach_btf);
2561		return -ENOMEM;
 
2562	}
2563
2564	prog->expected_attach_type = attr->expected_attach_type;
 
2565	prog->aux->attach_btf = attach_btf;
2566	prog->aux->attach_btf_id = attr->attach_btf_id;
2567	prog->aux->dst_prog = dst_prog;
2568	prog->aux->offload_requested = !!attr->prog_ifindex;
2569	prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2570	prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2571
2572	err = security_bpf_prog_alloc(prog->aux);
2573	if (err)
2574		goto free_prog;
2575
2576	prog->aux->user = get_current_user();
2577	prog->len = attr->insn_cnt;
2578
2579	err = -EFAULT;
2580	if (copy_from_bpfptr(prog->insns,
2581			     make_bpfptr(attr->insns, uattr.is_kernel),
2582			     bpf_prog_insn_size(prog)) != 0)
2583		goto free_prog_sec;
 
 
 
 
 
 
 
 
 
2584
2585	prog->orig_prog = NULL;
2586	prog->jited = 0;
2587
2588	atomic64_set(&prog->aux->refcnt, 1);
2589	prog->gpl_compatible = is_gpl ? 1 : 0;
2590
2591	if (bpf_prog_is_dev_bound(prog->aux)) {
2592		err = bpf_prog_offload_init(prog, attr);
2593		if (err)
2594			goto free_prog_sec;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2595	}
2596
2597	/* find program type: socket_filter vs tracing_filter */
2598	err = find_prog_type(type, prog);
2599	if (err < 0)
2600		goto free_prog_sec;
2601
2602	prog->aux->load_time = ktime_get_boottime_ns();
2603	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2604			       sizeof(attr->prog_name));
2605	if (err < 0)
 
 
 
 
2606		goto free_prog_sec;
2607
2608	/* run eBPF verifier */
2609	err = bpf_check(&prog, attr, uattr);
2610	if (err < 0)
2611		goto free_used_maps;
2612
2613	prog = bpf_prog_select_runtime(prog, &err);
2614	if (err < 0)
2615		goto free_used_maps;
2616
2617	err = bpf_prog_alloc_id(prog);
2618	if (err)
2619		goto free_used_maps;
2620
2621	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2622	 * effectively publicly exposed. However, retrieving via
2623	 * bpf_prog_get_fd_by_id() will take another reference,
2624	 * therefore it cannot be gone underneath us.
2625	 *
2626	 * Only for the time /after/ successful bpf_prog_new_fd()
2627	 * and before returning to userspace, we might just hold
2628	 * one reference and any parallel close on that fd could
2629	 * rip everything out. Hence, below notifications must
2630	 * happen before bpf_prog_new_fd().
2631	 *
2632	 * Also, any failure handling from this point onwards must
2633	 * be using bpf_prog_put() given the program is exposed.
2634	 */
2635	bpf_prog_kallsyms_add(prog);
2636	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2637	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2638
2639	err = bpf_prog_new_fd(prog);
2640	if (err < 0)
2641		bpf_prog_put(prog);
2642	return err;
2643
2644free_used_maps:
2645	/* In case we have subprogs, we need to wait for a grace
2646	 * period before we can tear down JIT memory since symbols
2647	 * are already exposed under kallsyms.
2648	 */
2649	__bpf_prog_put_noref(prog, prog->aux->func_cnt);
2650	return err;
 
2651free_prog_sec:
2652	free_uid(prog->aux->user);
2653	security_bpf_prog_free(prog->aux);
2654free_prog:
 
2655	if (prog->aux->attach_btf)
2656		btf_put(prog->aux->attach_btf);
2657	bpf_prog_free(prog);
 
 
2658	return err;
2659}
2660
2661#define BPF_OBJ_LAST_FIELD file_flags
2662
2663static int bpf_obj_pin(const union bpf_attr *attr)
2664{
2665	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
 
 
2666		return -EINVAL;
2667
2668	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
 
 
 
 
 
 
2669}
2670
2671static int bpf_obj_get(const union bpf_attr *attr)
2672{
 
 
2673	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2674	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
 
 
 
 
2675		return -EINVAL;
2676
2677	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
 
2678				attr->file_flags);
2679}
2680
2681void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2682		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
2683{
2684	atomic64_set(&link->refcnt, 1);
2685	link->type = type;
2686	link->id = 0;
2687	link->ops = ops;
2688	link->prog = prog;
2689}
2690
2691static void bpf_link_free_id(int id)
2692{
2693	if (!id)
2694		return;
2695
2696	spin_lock_bh(&link_idr_lock);
2697	idr_remove(&link_idr, id);
2698	spin_unlock_bh(&link_idr_lock);
2699}
2700
2701/* Clean up bpf_link and corresponding anon_inode file and FD. After
2702 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2703 * anon_inode's release() call. This helper marksbpf_link as
2704 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2705 * is not decremented, it's the responsibility of a calling code that failed
2706 * to complete bpf_link initialization.
 
 
2707 */
2708void bpf_link_cleanup(struct bpf_link_primer *primer)
2709{
2710	primer->link->prog = NULL;
2711	bpf_link_free_id(primer->id);
2712	fput(primer->file);
2713	put_unused_fd(primer->fd);
2714}
2715
2716void bpf_link_inc(struct bpf_link *link)
2717{
2718	atomic64_inc(&link->refcnt);
2719}
2720
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2721/* bpf_link_free is guaranteed to be called from process context */
2722static void bpf_link_free(struct bpf_link *link)
2723{
 
 
2724	bpf_link_free_id(link->id);
2725	if (link->prog) {
 
2726		/* detach BPF program, clean up used resources */
2727		link->ops->release(link);
2728		bpf_prog_put(link->prog);
2729	}
2730	/* free bpf_link and its containing memory */
2731	link->ops->dealloc(link);
 
 
 
 
 
 
 
 
 
 
2732}
2733
2734static void bpf_link_put_deferred(struct work_struct *work)
2735{
2736	struct bpf_link *link = container_of(work, struct bpf_link, work);
2737
2738	bpf_link_free(link);
2739}
2740
2741/* bpf_link_put can be called from atomic context, but ensures that resources
2742 * are freed from process context
2743 */
2744void bpf_link_put(struct bpf_link *link)
2745{
2746	if (!atomic64_dec_and_test(&link->refcnt))
2747		return;
2748
2749	if (in_atomic()) {
2750		INIT_WORK(&link->work, bpf_link_put_deferred);
2751		schedule_work(&link->work);
2752	} else {
2753		bpf_link_free(link);
2754	}
2755}
2756EXPORT_SYMBOL(bpf_link_put);
2757
 
 
 
 
 
 
 
2758static int bpf_link_release(struct inode *inode, struct file *filp)
2759{
2760	struct bpf_link *link = filp->private_data;
2761
2762	bpf_link_put(link);
2763	return 0;
2764}
2765
2766#ifdef CONFIG_PROC_FS
2767#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2768#define BPF_MAP_TYPE(_id, _ops)
2769#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2770static const char *bpf_link_type_strs[] = {
2771	[BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2772#include <linux/bpf_types.h>
2773};
2774#undef BPF_PROG_TYPE
2775#undef BPF_MAP_TYPE
2776#undef BPF_LINK_TYPE
2777
2778static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2779{
2780	const struct bpf_link *link = filp->private_data;
2781	const struct bpf_prog *prog = link->prog;
2782	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2783
2784	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2785	seq_printf(m,
2786		   "link_type:\t%s\n"
2787		   "link_id:\t%u\n"
2788		   "prog_tag:\t%s\n"
2789		   "prog_id:\t%u\n",
2790		   bpf_link_type_strs[link->type],
2791		   link->id,
2792		   prog_tag,
2793		   prog->aux->id);
 
 
 
 
 
 
2794	if (link->ops->show_fdinfo)
2795		link->ops->show_fdinfo(link, m);
2796}
2797#endif
2798
2799static const struct file_operations bpf_link_fops = {
2800#ifdef CONFIG_PROC_FS
2801	.show_fdinfo	= bpf_link_show_fdinfo,
2802#endif
2803	.release	= bpf_link_release,
2804	.read		= bpf_dummy_read,
2805	.write		= bpf_dummy_write,
2806};
2807
2808static int bpf_link_alloc_id(struct bpf_link *link)
2809{
2810	int id;
2811
2812	idr_preload(GFP_KERNEL);
2813	spin_lock_bh(&link_idr_lock);
2814	id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2815	spin_unlock_bh(&link_idr_lock);
2816	idr_preload_end();
2817
2818	return id;
2819}
2820
2821/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2822 * reserving unused FD and allocating ID from link_idr. This is to be paired
2823 * with bpf_link_settle() to install FD and ID and expose bpf_link to
2824 * user-space, if bpf_link is successfully attached. If not, bpf_link and
2825 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2826 * transient state is passed around in struct bpf_link_primer.
2827 * This is preferred way to create and initialize bpf_link, especially when
2828 * there are complicated and expensive operations in between creating bpf_link
2829 * itself and attaching it to BPF hook. By using bpf_link_prime() and
2830 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2831 * expensive (and potentially failing) roll back operations in a rare case
2832 * that file, FD, or ID can't be allocated.
2833 */
2834int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2835{
2836	struct file *file;
2837	int fd, id;
2838
2839	fd = get_unused_fd_flags(O_CLOEXEC);
2840	if (fd < 0)
2841		return fd;
2842
2843
2844	id = bpf_link_alloc_id(link);
2845	if (id < 0) {
2846		put_unused_fd(fd);
2847		return id;
2848	}
2849
2850	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2851	if (IS_ERR(file)) {
2852		bpf_link_free_id(id);
2853		put_unused_fd(fd);
2854		return PTR_ERR(file);
2855	}
2856
2857	primer->link = link;
2858	primer->file = file;
2859	primer->fd = fd;
2860	primer->id = id;
2861	return 0;
2862}
2863
2864int bpf_link_settle(struct bpf_link_primer *primer)
2865{
2866	/* make bpf_link fetchable by ID */
2867	spin_lock_bh(&link_idr_lock);
2868	primer->link->id = primer->id;
2869	spin_unlock_bh(&link_idr_lock);
2870	/* make bpf_link fetchable by FD */
2871	fd_install(primer->fd, primer->file);
2872	/* pass through installed FD */
2873	return primer->fd;
2874}
2875
2876int bpf_link_new_fd(struct bpf_link *link)
2877{
2878	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2879}
2880
2881struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2882{
2883	struct fd f = fdget(ufd);
2884	struct bpf_link *link;
2885
2886	if (!f.file)
2887		return ERR_PTR(-EBADF);
2888	if (f.file->f_op != &bpf_link_fops) {
2889		fdput(f);
2890		return ERR_PTR(-EINVAL);
2891	}
2892
2893	link = f.file->private_data;
2894	bpf_link_inc(link);
2895	fdput(f);
2896
2897	return link;
2898}
2899EXPORT_SYMBOL(bpf_link_get_from_fd);
2900
2901static void bpf_tracing_link_release(struct bpf_link *link)
2902{
2903	struct bpf_tracing_link *tr_link =
2904		container_of(link, struct bpf_tracing_link, link.link);
2905
2906	WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2907						tr_link->trampoline));
2908
2909	bpf_trampoline_put(tr_link->trampoline);
2910
2911	/* tgt_prog is NULL if target is a kernel function */
2912	if (tr_link->tgt_prog)
2913		bpf_prog_put(tr_link->tgt_prog);
2914}
2915
2916static void bpf_tracing_link_dealloc(struct bpf_link *link)
2917{
2918	struct bpf_tracing_link *tr_link =
2919		container_of(link, struct bpf_tracing_link, link.link);
2920
2921	kfree(tr_link);
2922}
2923
2924static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2925					 struct seq_file *seq)
2926{
2927	struct bpf_tracing_link *tr_link =
2928		container_of(link, struct bpf_tracing_link, link.link);
 
2929
 
 
2930	seq_printf(seq,
2931		   "attach_type:\t%d\n",
2932		   tr_link->attach_type);
 
 
 
 
2933}
2934
2935static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2936					   struct bpf_link_info *info)
2937{
2938	struct bpf_tracing_link *tr_link =
2939		container_of(link, struct bpf_tracing_link, link.link);
2940
2941	info->tracing.attach_type = tr_link->attach_type;
2942	bpf_trampoline_unpack_key(tr_link->trampoline->key,
2943				  &info->tracing.target_obj_id,
2944				  &info->tracing.target_btf_id);
2945
2946	return 0;
2947}
2948
2949static const struct bpf_link_ops bpf_tracing_link_lops = {
2950	.release = bpf_tracing_link_release,
2951	.dealloc = bpf_tracing_link_dealloc,
2952	.show_fdinfo = bpf_tracing_link_show_fdinfo,
2953	.fill_link_info = bpf_tracing_link_fill_link_info,
2954};
2955
2956static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2957				   int tgt_prog_fd,
2958				   u32 btf_id,
2959				   u64 bpf_cookie)
2960{
2961	struct bpf_link_primer link_primer;
2962	struct bpf_prog *tgt_prog = NULL;
2963	struct bpf_trampoline *tr = NULL;
2964	struct bpf_tracing_link *link;
2965	u64 key = 0;
2966	int err;
2967
2968	switch (prog->type) {
2969	case BPF_PROG_TYPE_TRACING:
2970		if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2971		    prog->expected_attach_type != BPF_TRACE_FEXIT &&
2972		    prog->expected_attach_type != BPF_MODIFY_RETURN) {
2973			err = -EINVAL;
2974			goto out_put_prog;
2975		}
2976		break;
2977	case BPF_PROG_TYPE_EXT:
2978		if (prog->expected_attach_type != 0) {
2979			err = -EINVAL;
2980			goto out_put_prog;
2981		}
2982		break;
2983	case BPF_PROG_TYPE_LSM:
2984		if (prog->expected_attach_type != BPF_LSM_MAC) {
2985			err = -EINVAL;
2986			goto out_put_prog;
2987		}
2988		break;
2989	default:
2990		err = -EINVAL;
2991		goto out_put_prog;
2992	}
2993
2994	if (!!tgt_prog_fd != !!btf_id) {
2995		err = -EINVAL;
2996		goto out_put_prog;
2997	}
2998
2999	if (tgt_prog_fd) {
3000		/* For now we only allow new targets for BPF_PROG_TYPE_EXT */
 
 
 
 
 
3001		if (prog->type != BPF_PROG_TYPE_EXT) {
3002			err = -EINVAL;
3003			goto out_put_prog;
3004		}
3005
3006		tgt_prog = bpf_prog_get(tgt_prog_fd);
3007		if (IS_ERR(tgt_prog)) {
3008			err = PTR_ERR(tgt_prog);
3009			tgt_prog = NULL;
3010			goto out_put_prog;
3011		}
3012
3013		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3014	}
3015
3016	link = kzalloc(sizeof(*link), GFP_USER);
3017	if (!link) {
3018		err = -ENOMEM;
3019		goto out_put_prog;
3020	}
3021	bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3022		      &bpf_tracing_link_lops, prog);
3023	link->attach_type = prog->expected_attach_type;
3024	link->link.cookie = bpf_cookie;
3025
3026	mutex_lock(&prog->aux->dst_mutex);
3027
3028	/* There are a few possible cases here:
3029	 *
3030	 * - if prog->aux->dst_trampoline is set, the program was just loaded
3031	 *   and not yet attached to anything, so we can use the values stored
3032	 *   in prog->aux
3033	 *
3034	 * - if prog->aux->dst_trampoline is NULL, the program has already been
3035         *   attached to a target and its initial target was cleared (below)
3036	 *
3037	 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3038	 *   target_btf_id using the link_create API.
3039	 *
3040	 * - if tgt_prog == NULL when this function was called using the old
3041	 *   raw_tracepoint_open API, and we need a target from prog->aux
3042	 *
3043	 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3044	 *   was detached and is going for re-attachment.
 
 
 
 
3045	 */
3046	if (!prog->aux->dst_trampoline && !tgt_prog) {
3047		/*
3048		 * Allow re-attach for TRACING and LSM programs. If it's
3049		 * currently linked, bpf_trampoline_link_prog will fail.
3050		 * EXT programs need to specify tgt_prog_fd, so they
3051		 * re-attach in separate code path.
3052		 */
3053		if (prog->type != BPF_PROG_TYPE_TRACING &&
3054		    prog->type != BPF_PROG_TYPE_LSM) {
3055			err = -EINVAL;
3056			goto out_unlock;
3057		}
 
 
 
 
 
3058		btf_id = prog->aux->attach_btf_id;
3059		key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3060	}
3061
3062	if (!prog->aux->dst_trampoline ||
3063	    (key && key != prog->aux->dst_trampoline->key)) {
3064		/* If there is no saved target, or the specified target is
3065		 * different from the destination specified at load time, we
3066		 * need a new trampoline and a check for compatibility
3067		 */
3068		struct bpf_attach_target_info tgt_info = {};
3069
3070		err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3071					      &tgt_info);
3072		if (err)
3073			goto out_unlock;
3074
 
 
 
 
 
3075		tr = bpf_trampoline_get(key, &tgt_info);
3076		if (!tr) {
3077			err = -ENOMEM;
3078			goto out_unlock;
3079		}
3080	} else {
3081		/* The caller didn't specify a target, or the target was the
3082		 * same as the destination supplied during program load. This
3083		 * means we can reuse the trampoline and reference from program
3084		 * load time, and there is no need to allocate a new one. This
3085		 * can only happen once for any program, as the saved values in
3086		 * prog->aux are cleared below.
3087		 */
3088		tr = prog->aux->dst_trampoline;
3089		tgt_prog = prog->aux->dst_prog;
3090	}
3091
3092	err = bpf_link_prime(&link->link.link, &link_primer);
3093	if (err)
3094		goto out_unlock;
3095
3096	err = bpf_trampoline_link_prog(&link->link, tr);
3097	if (err) {
3098		bpf_link_cleanup(&link_primer);
3099		link = NULL;
3100		goto out_unlock;
3101	}
3102
3103	link->tgt_prog = tgt_prog;
3104	link->trampoline = tr;
3105
3106	/* Always clear the trampoline and target prog from prog->aux to make
3107	 * sure the original attach destination is not kept alive after a
3108	 * program is (re-)attached to another target.
3109	 */
3110	if (prog->aux->dst_prog &&
3111	    (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3112		/* got extra prog ref from syscall, or attaching to different prog */
3113		bpf_prog_put(prog->aux->dst_prog);
3114	if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3115		/* we allocated a new trampoline, so free the old one */
3116		bpf_trampoline_put(prog->aux->dst_trampoline);
3117
3118	prog->aux->dst_prog = NULL;
3119	prog->aux->dst_trampoline = NULL;
3120	mutex_unlock(&prog->aux->dst_mutex);
3121
3122	return bpf_link_settle(&link_primer);
3123out_unlock:
3124	if (tr && tr != prog->aux->dst_trampoline)
3125		bpf_trampoline_put(tr);
3126	mutex_unlock(&prog->aux->dst_mutex);
3127	kfree(link);
3128out_put_prog:
3129	if (tgt_prog_fd && tgt_prog)
3130		bpf_prog_put(tgt_prog);
3131	return err;
3132}
3133
3134struct bpf_raw_tp_link {
3135	struct bpf_link link;
3136	struct bpf_raw_event_map *btp;
3137};
3138
3139static void bpf_raw_tp_link_release(struct bpf_link *link)
3140{
3141	struct bpf_raw_tp_link *raw_tp =
3142		container_of(link, struct bpf_raw_tp_link, link);
3143
3144	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3145	bpf_put_raw_tracepoint(raw_tp->btp);
3146}
3147
3148static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3149{
3150	struct bpf_raw_tp_link *raw_tp =
3151		container_of(link, struct bpf_raw_tp_link, link);
3152
3153	kfree(raw_tp);
3154}
3155
3156static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3157					struct seq_file *seq)
3158{
3159	struct bpf_raw_tp_link *raw_tp_link =
3160		container_of(link, struct bpf_raw_tp_link, link);
3161
3162	seq_printf(seq,
3163		   "tp_name:\t%s\n",
3164		   raw_tp_link->btp->tp->name);
3165}
3166
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3167static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3168					  struct bpf_link_info *info)
3169{
3170	struct bpf_raw_tp_link *raw_tp_link =
3171		container_of(link, struct bpf_raw_tp_link, link);
3172	char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3173	const char *tp_name = raw_tp_link->btp->tp->name;
3174	u32 ulen = info->raw_tracepoint.tp_name_len;
3175	size_t tp_len = strlen(tp_name);
3176
3177	if (!ulen ^ !ubuf)
3178		return -EINVAL;
3179
3180	info->raw_tracepoint.tp_name_len = tp_len + 1;
3181
3182	if (!ubuf)
3183		return 0;
3184
3185	if (ulen >= tp_len + 1) {
3186		if (copy_to_user(ubuf, tp_name, tp_len + 1))
3187			return -EFAULT;
3188	} else {
3189		char zero = '\0';
3190
3191		if (copy_to_user(ubuf, tp_name, ulen - 1))
3192			return -EFAULT;
3193		if (put_user(zero, ubuf + ulen - 1))
3194			return -EFAULT;
3195		return -ENOSPC;
3196	}
3197
3198	return 0;
3199}
3200
3201static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3202	.release = bpf_raw_tp_link_release,
3203	.dealloc = bpf_raw_tp_link_dealloc,
3204	.show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3205	.fill_link_info = bpf_raw_tp_link_fill_link_info,
3206};
3207
3208#ifdef CONFIG_PERF_EVENTS
3209struct bpf_perf_link {
3210	struct bpf_link link;
3211	struct file *perf_file;
3212};
3213
3214static void bpf_perf_link_release(struct bpf_link *link)
3215{
3216	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3217	struct perf_event *event = perf_link->perf_file->private_data;
3218
3219	perf_event_free_bpf_prog(event);
3220	fput(perf_link->perf_file);
3221}
3222
3223static void bpf_perf_link_dealloc(struct bpf_link *link)
3224{
3225	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3226
3227	kfree(perf_link);
3228}
3229
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3230static const struct bpf_link_ops bpf_perf_link_lops = {
3231	.release = bpf_perf_link_release,
3232	.dealloc = bpf_perf_link_dealloc,
 
3233};
3234
3235static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3236{
3237	struct bpf_link_primer link_primer;
3238	struct bpf_perf_link *link;
3239	struct perf_event *event;
3240	struct file *perf_file;
3241	int err;
3242
3243	if (attr->link_create.flags)
3244		return -EINVAL;
3245
3246	perf_file = perf_event_get(attr->link_create.target_fd);
3247	if (IS_ERR(perf_file))
3248		return PTR_ERR(perf_file);
3249
3250	link = kzalloc(sizeof(*link), GFP_USER);
3251	if (!link) {
3252		err = -ENOMEM;
3253		goto out_put_file;
3254	}
3255	bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3256	link->perf_file = perf_file;
3257
3258	err = bpf_link_prime(&link->link, &link_primer);
3259	if (err) {
3260		kfree(link);
3261		goto out_put_file;
3262	}
3263
3264	event = perf_file->private_data;
3265	err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3266	if (err) {
3267		bpf_link_cleanup(&link_primer);
3268		goto out_put_file;
3269	}
3270	/* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3271	bpf_prog_inc(prog);
3272
3273	return bpf_link_settle(&link_primer);
3274
3275out_put_file:
3276	fput(perf_file);
3277	return err;
3278}
3279#else
3280static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3281{
3282	return -EOPNOTSUPP;
3283}
3284#endif /* CONFIG_PERF_EVENTS */
3285
3286static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3287				  const char __user *user_tp_name)
3288{
3289	struct bpf_link_primer link_primer;
3290	struct bpf_raw_tp_link *link;
3291	struct bpf_raw_event_map *btp;
3292	const char *tp_name;
3293	char buf[128];
3294	int err;
3295
3296	switch (prog->type) {
3297	case BPF_PROG_TYPE_TRACING:
3298	case BPF_PROG_TYPE_EXT:
3299	case BPF_PROG_TYPE_LSM:
3300		if (user_tp_name)
3301			/* The attach point for this category of programs
3302			 * should be specified via btf_id during program load.
3303			 */
3304			return -EINVAL;
3305		if (prog->type == BPF_PROG_TYPE_TRACING &&
3306		    prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3307			tp_name = prog->aux->attach_func_name;
3308			break;
3309		}
3310		return bpf_tracing_prog_attach(prog, 0, 0, 0);
3311	case BPF_PROG_TYPE_RAW_TRACEPOINT:
3312	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3313		if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3314			return -EFAULT;
3315		buf[sizeof(buf) - 1] = 0;
3316		tp_name = buf;
3317		break;
3318	default:
3319		return -EINVAL;
3320	}
3321
3322	btp = bpf_get_raw_tracepoint(tp_name);
3323	if (!btp)
3324		return -ENOENT;
3325
3326	link = kzalloc(sizeof(*link), GFP_USER);
3327	if (!link) {
3328		err = -ENOMEM;
3329		goto out_put_btp;
3330	}
3331	bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3332		      &bpf_raw_tp_link_lops, prog);
3333	link->btp = btp;
3334
3335	err = bpf_link_prime(&link->link, &link_primer);
3336	if (err) {
3337		kfree(link);
3338		goto out_put_btp;
3339	}
3340
3341	err = bpf_probe_register(link->btp, prog);
3342	if (err) {
3343		bpf_link_cleanup(&link_primer);
3344		goto out_put_btp;
3345	}
3346
3347	return bpf_link_settle(&link_primer);
3348
3349out_put_btp:
3350	bpf_put_raw_tracepoint(btp);
3351	return err;
3352}
3353
3354#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3355
3356static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3357{
3358	struct bpf_prog *prog;
3359	int fd;
3360
3361	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3362		return -EINVAL;
3363
3364	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3365	if (IS_ERR(prog))
3366		return PTR_ERR(prog);
3367
3368	fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3369	if (fd < 0)
3370		bpf_prog_put(prog);
3371	return fd;
3372}
3373
3374static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3375					     enum bpf_attach_type attach_type)
3376{
3377	switch (prog->type) {
3378	case BPF_PROG_TYPE_CGROUP_SOCK:
3379	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3380	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3381	case BPF_PROG_TYPE_SK_LOOKUP:
3382		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3383	case BPF_PROG_TYPE_CGROUP_SKB:
3384		if (!capable(CAP_NET_ADMIN))
3385			/* cg-skb progs can be loaded by unpriv user.
3386			 * check permissions at attach time.
3387			 */
3388			return -EPERM;
3389		return prog->enforce_expected_attach_type &&
3390			prog->expected_attach_type != attach_type ?
3391			-EINVAL : 0;
3392	default:
3393		return 0;
3394	}
3395}
3396
3397static enum bpf_prog_type
3398attach_type_to_prog_type(enum bpf_attach_type attach_type)
3399{
3400	switch (attach_type) {
3401	case BPF_CGROUP_INET_INGRESS:
3402	case BPF_CGROUP_INET_EGRESS:
3403		return BPF_PROG_TYPE_CGROUP_SKB;
3404	case BPF_CGROUP_INET_SOCK_CREATE:
3405	case BPF_CGROUP_INET_SOCK_RELEASE:
3406	case BPF_CGROUP_INET4_POST_BIND:
3407	case BPF_CGROUP_INET6_POST_BIND:
3408		return BPF_PROG_TYPE_CGROUP_SOCK;
3409	case BPF_CGROUP_INET4_BIND:
3410	case BPF_CGROUP_INET6_BIND:
3411	case BPF_CGROUP_INET4_CONNECT:
3412	case BPF_CGROUP_INET6_CONNECT:
 
3413	case BPF_CGROUP_INET4_GETPEERNAME:
3414	case BPF_CGROUP_INET6_GETPEERNAME:
 
3415	case BPF_CGROUP_INET4_GETSOCKNAME:
3416	case BPF_CGROUP_INET6_GETSOCKNAME:
 
3417	case BPF_CGROUP_UDP4_SENDMSG:
3418	case BPF_CGROUP_UDP6_SENDMSG:
 
3419	case BPF_CGROUP_UDP4_RECVMSG:
3420	case BPF_CGROUP_UDP6_RECVMSG:
 
3421		return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3422	case BPF_CGROUP_SOCK_OPS:
3423		return BPF_PROG_TYPE_SOCK_OPS;
3424	case BPF_CGROUP_DEVICE:
3425		return BPF_PROG_TYPE_CGROUP_DEVICE;
3426	case BPF_SK_MSG_VERDICT:
3427		return BPF_PROG_TYPE_SK_MSG;
3428	case BPF_SK_SKB_STREAM_PARSER:
3429	case BPF_SK_SKB_STREAM_VERDICT:
3430	case BPF_SK_SKB_VERDICT:
3431		return BPF_PROG_TYPE_SK_SKB;
3432	case BPF_LIRC_MODE2:
3433		return BPF_PROG_TYPE_LIRC_MODE2;
3434	case BPF_FLOW_DISSECTOR:
3435		return BPF_PROG_TYPE_FLOW_DISSECTOR;
3436	case BPF_CGROUP_SYSCTL:
3437		return BPF_PROG_TYPE_CGROUP_SYSCTL;
3438	case BPF_CGROUP_GETSOCKOPT:
3439	case BPF_CGROUP_SETSOCKOPT:
3440		return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3441	case BPF_TRACE_ITER:
3442	case BPF_TRACE_RAW_TP:
3443	case BPF_TRACE_FENTRY:
3444	case BPF_TRACE_FEXIT:
3445	case BPF_MODIFY_RETURN:
3446		return BPF_PROG_TYPE_TRACING;
3447	case BPF_LSM_MAC:
3448		return BPF_PROG_TYPE_LSM;
3449	case BPF_SK_LOOKUP:
3450		return BPF_PROG_TYPE_SK_LOOKUP;
3451	case BPF_XDP:
3452		return BPF_PROG_TYPE_XDP;
3453	case BPF_LSM_CGROUP:
3454		return BPF_PROG_TYPE_LSM;
 
 
 
 
 
3455	default:
3456		return BPF_PROG_TYPE_UNSPEC;
3457	}
3458}
3459
3460#define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3461
3462#define BPF_F_ATTACH_MASK \
3463	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
 
 
 
 
 
 
 
 
 
3464
3465static int bpf_prog_attach(const union bpf_attr *attr)
3466{
3467	enum bpf_prog_type ptype;
3468	struct bpf_prog *prog;
3469	int ret;
3470
3471	if (CHECK_ATTR(BPF_PROG_ATTACH))
3472		return -EINVAL;
3473
3474	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3475		return -EINVAL;
3476
3477	ptype = attach_type_to_prog_type(attr->attach_type);
3478	if (ptype == BPF_PROG_TYPE_UNSPEC)
3479		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
3480
3481	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3482	if (IS_ERR(prog))
3483		return PTR_ERR(prog);
3484
3485	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3486		bpf_prog_put(prog);
3487		return -EINVAL;
3488	}
3489
3490	switch (ptype) {
3491	case BPF_PROG_TYPE_SK_SKB:
3492	case BPF_PROG_TYPE_SK_MSG:
3493		ret = sock_map_get_from_fd(attr, prog);
3494		break;
3495	case BPF_PROG_TYPE_LIRC_MODE2:
3496		ret = lirc_prog_attach(attr, prog);
3497		break;
3498	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3499		ret = netns_bpf_prog_attach(attr, prog);
3500		break;
3501	case BPF_PROG_TYPE_CGROUP_DEVICE:
3502	case BPF_PROG_TYPE_CGROUP_SKB:
3503	case BPF_PROG_TYPE_CGROUP_SOCK:
3504	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3505	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3506	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3507	case BPF_PROG_TYPE_SOCK_OPS:
3508	case BPF_PROG_TYPE_LSM:
3509		if (ptype == BPF_PROG_TYPE_LSM &&
3510		    prog->expected_attach_type != BPF_LSM_CGROUP)
3511			ret = -EINVAL;
3512		else
3513			ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3514		break;
 
 
 
 
 
 
 
3515	default:
3516		ret = -EINVAL;
3517	}
3518
3519	if (ret)
3520		bpf_prog_put(prog);
3521	return ret;
3522}
3523
3524#define BPF_PROG_DETACH_LAST_FIELD attach_type
3525
3526static int bpf_prog_detach(const union bpf_attr *attr)
3527{
 
3528	enum bpf_prog_type ptype;
 
3529
3530	if (CHECK_ATTR(BPF_PROG_DETACH))
3531		return -EINVAL;
3532
3533	ptype = attach_type_to_prog_type(attr->attach_type);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3534
3535	switch (ptype) {
3536	case BPF_PROG_TYPE_SK_MSG:
3537	case BPF_PROG_TYPE_SK_SKB:
3538		return sock_map_prog_detach(attr, ptype);
 
3539	case BPF_PROG_TYPE_LIRC_MODE2:
3540		return lirc_prog_detach(attr);
 
3541	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3542		return netns_bpf_prog_detach(attr, ptype);
 
3543	case BPF_PROG_TYPE_CGROUP_DEVICE:
3544	case BPF_PROG_TYPE_CGROUP_SKB:
3545	case BPF_PROG_TYPE_CGROUP_SOCK:
3546	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3547	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3548	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3549	case BPF_PROG_TYPE_SOCK_OPS:
3550	case BPF_PROG_TYPE_LSM:
3551		return cgroup_bpf_prog_detach(attr, ptype);
 
 
 
 
 
 
 
 
3552	default:
3553		return -EINVAL;
3554	}
 
 
 
 
3555}
3556
3557#define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3558
3559static int bpf_prog_query(const union bpf_attr *attr,
3560			  union bpf_attr __user *uattr)
3561{
3562	if (!capable(CAP_NET_ADMIN))
3563		return -EPERM;
3564	if (CHECK_ATTR(BPF_PROG_QUERY))
3565		return -EINVAL;
3566	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3567		return -EINVAL;
3568
3569	switch (attr->query.attach_type) {
3570	case BPF_CGROUP_INET_INGRESS:
3571	case BPF_CGROUP_INET_EGRESS:
3572	case BPF_CGROUP_INET_SOCK_CREATE:
3573	case BPF_CGROUP_INET_SOCK_RELEASE:
3574	case BPF_CGROUP_INET4_BIND:
3575	case BPF_CGROUP_INET6_BIND:
3576	case BPF_CGROUP_INET4_POST_BIND:
3577	case BPF_CGROUP_INET6_POST_BIND:
3578	case BPF_CGROUP_INET4_CONNECT:
3579	case BPF_CGROUP_INET6_CONNECT:
 
3580	case BPF_CGROUP_INET4_GETPEERNAME:
3581	case BPF_CGROUP_INET6_GETPEERNAME:
 
3582	case BPF_CGROUP_INET4_GETSOCKNAME:
3583	case BPF_CGROUP_INET6_GETSOCKNAME:
 
3584	case BPF_CGROUP_UDP4_SENDMSG:
3585	case BPF_CGROUP_UDP6_SENDMSG:
 
3586	case BPF_CGROUP_UDP4_RECVMSG:
3587	case BPF_CGROUP_UDP6_RECVMSG:
 
3588	case BPF_CGROUP_SOCK_OPS:
3589	case BPF_CGROUP_DEVICE:
3590	case BPF_CGROUP_SYSCTL:
3591	case BPF_CGROUP_GETSOCKOPT:
3592	case BPF_CGROUP_SETSOCKOPT:
3593	case BPF_LSM_CGROUP:
3594		return cgroup_bpf_prog_query(attr, uattr);
3595	case BPF_LIRC_MODE2:
3596		return lirc_prog_query(attr, uattr);
3597	case BPF_FLOW_DISSECTOR:
3598	case BPF_SK_LOOKUP:
3599		return netns_bpf_prog_query(attr, uattr);
3600	case BPF_SK_SKB_STREAM_PARSER:
3601	case BPF_SK_SKB_STREAM_VERDICT:
3602	case BPF_SK_MSG_VERDICT:
3603	case BPF_SK_SKB_VERDICT:
3604		return sock_map_bpf_prog_query(attr, uattr);
 
 
 
 
 
 
3605	default:
3606		return -EINVAL;
3607	}
3608}
3609
3610#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3611
3612static int bpf_prog_test_run(const union bpf_attr *attr,
3613			     union bpf_attr __user *uattr)
3614{
3615	struct bpf_prog *prog;
3616	int ret = -ENOTSUPP;
3617
3618	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3619		return -EINVAL;
3620
3621	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3622	    (!attr->test.ctx_size_in && attr->test.ctx_in))
3623		return -EINVAL;
3624
3625	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3626	    (!attr->test.ctx_size_out && attr->test.ctx_out))
3627		return -EINVAL;
3628
3629	prog = bpf_prog_get(attr->test.prog_fd);
3630	if (IS_ERR(prog))
3631		return PTR_ERR(prog);
3632
3633	if (prog->aux->ops->test_run)
3634		ret = prog->aux->ops->test_run(prog, attr, uattr);
3635
3636	bpf_prog_put(prog);
3637	return ret;
3638}
3639
3640#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3641
3642static int bpf_obj_get_next_id(const union bpf_attr *attr,
3643			       union bpf_attr __user *uattr,
3644			       struct idr *idr,
3645			       spinlock_t *lock)
3646{
3647	u32 next_id = attr->start_id;
3648	int err = 0;
3649
3650	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3651		return -EINVAL;
3652
3653	if (!capable(CAP_SYS_ADMIN))
3654		return -EPERM;
3655
3656	next_id++;
3657	spin_lock_bh(lock);
3658	if (!idr_get_next(idr, &next_id))
3659		err = -ENOENT;
3660	spin_unlock_bh(lock);
3661
3662	if (!err)
3663		err = put_user(next_id, &uattr->next_id);
3664
3665	return err;
3666}
3667
3668struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3669{
3670	struct bpf_map *map;
3671
3672	spin_lock_bh(&map_idr_lock);
3673again:
3674	map = idr_get_next(&map_idr, id);
3675	if (map) {
3676		map = __bpf_map_inc_not_zero(map, false);
3677		if (IS_ERR(map)) {
3678			(*id)++;
3679			goto again;
3680		}
3681	}
3682	spin_unlock_bh(&map_idr_lock);
3683
3684	return map;
3685}
3686
3687struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3688{
3689	struct bpf_prog *prog;
3690
3691	spin_lock_bh(&prog_idr_lock);
3692again:
3693	prog = idr_get_next(&prog_idr, id);
3694	if (prog) {
3695		prog = bpf_prog_inc_not_zero(prog);
3696		if (IS_ERR(prog)) {
3697			(*id)++;
3698			goto again;
3699		}
3700	}
3701	spin_unlock_bh(&prog_idr_lock);
3702
3703	return prog;
3704}
3705
3706#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3707
3708struct bpf_prog *bpf_prog_by_id(u32 id)
3709{
3710	struct bpf_prog *prog;
3711
3712	if (!id)
3713		return ERR_PTR(-ENOENT);
3714
3715	spin_lock_bh(&prog_idr_lock);
3716	prog = idr_find(&prog_idr, id);
3717	if (prog)
3718		prog = bpf_prog_inc_not_zero(prog);
3719	else
3720		prog = ERR_PTR(-ENOENT);
3721	spin_unlock_bh(&prog_idr_lock);
3722	return prog;
3723}
3724
3725static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3726{
3727	struct bpf_prog *prog;
3728	u32 id = attr->prog_id;
3729	int fd;
3730
3731	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3732		return -EINVAL;
3733
3734	if (!capable(CAP_SYS_ADMIN))
3735		return -EPERM;
3736
3737	prog = bpf_prog_by_id(id);
3738	if (IS_ERR(prog))
3739		return PTR_ERR(prog);
3740
3741	fd = bpf_prog_new_fd(prog);
3742	if (fd < 0)
3743		bpf_prog_put(prog);
3744
3745	return fd;
3746}
3747
3748#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3749
3750static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3751{
3752	struct bpf_map *map;
3753	u32 id = attr->map_id;
3754	int f_flags;
3755	int fd;
3756
3757	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3758	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3759		return -EINVAL;
3760
3761	if (!capable(CAP_SYS_ADMIN))
3762		return -EPERM;
3763
3764	f_flags = bpf_get_file_flag(attr->open_flags);
3765	if (f_flags < 0)
3766		return f_flags;
3767
3768	spin_lock_bh(&map_idr_lock);
3769	map = idr_find(&map_idr, id);
3770	if (map)
3771		map = __bpf_map_inc_not_zero(map, true);
3772	else
3773		map = ERR_PTR(-ENOENT);
3774	spin_unlock_bh(&map_idr_lock);
3775
3776	if (IS_ERR(map))
3777		return PTR_ERR(map);
3778
3779	fd = bpf_map_new_fd(map, f_flags);
3780	if (fd < 0)
3781		bpf_map_put_with_uref(map);
3782
3783	return fd;
3784}
3785
3786static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3787					      unsigned long addr, u32 *off,
3788					      u32 *type)
3789{
3790	const struct bpf_map *map;
3791	int i;
3792
3793	mutex_lock(&prog->aux->used_maps_mutex);
3794	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3795		map = prog->aux->used_maps[i];
3796		if (map == (void *)addr) {
3797			*type = BPF_PSEUDO_MAP_FD;
3798			goto out;
3799		}
3800		if (!map->ops->map_direct_value_meta)
3801			continue;
3802		if (!map->ops->map_direct_value_meta(map, addr, off)) {
3803			*type = BPF_PSEUDO_MAP_VALUE;
3804			goto out;
3805		}
3806	}
3807	map = NULL;
3808
3809out:
3810	mutex_unlock(&prog->aux->used_maps_mutex);
3811	return map;
3812}
3813
3814static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3815					      const struct cred *f_cred)
3816{
3817	const struct bpf_map *map;
3818	struct bpf_insn *insns;
3819	u32 off, type;
3820	u64 imm;
3821	u8 code;
3822	int i;
3823
3824	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3825			GFP_USER);
3826	if (!insns)
3827		return insns;
3828
3829	for (i = 0; i < prog->len; i++) {
3830		code = insns[i].code;
3831
3832		if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3833			insns[i].code = BPF_JMP | BPF_CALL;
3834			insns[i].imm = BPF_FUNC_tail_call;
3835			/* fall-through */
3836		}
3837		if (code == (BPF_JMP | BPF_CALL) ||
3838		    code == (BPF_JMP | BPF_CALL_ARGS)) {
3839			if (code == (BPF_JMP | BPF_CALL_ARGS))
3840				insns[i].code = BPF_JMP | BPF_CALL;
3841			if (!bpf_dump_raw_ok(f_cred))
3842				insns[i].imm = 0;
3843			continue;
3844		}
3845		if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3846			insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3847			continue;
3848		}
3849
 
 
 
 
 
 
3850		if (code != (BPF_LD | BPF_IMM | BPF_DW))
3851			continue;
3852
3853		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3854		map = bpf_map_from_imm(prog, imm, &off, &type);
3855		if (map) {
3856			insns[i].src_reg = type;
3857			insns[i].imm = map->id;
3858			insns[i + 1].imm = off;
3859			continue;
3860		}
3861	}
3862
3863	return insns;
3864}
3865
3866static int set_info_rec_size(struct bpf_prog_info *info)
3867{
3868	/*
3869	 * Ensure info.*_rec_size is the same as kernel expected size
3870	 *
3871	 * or
3872	 *
3873	 * Only allow zero *_rec_size if both _rec_size and _cnt are
3874	 * zero.  In this case, the kernel will set the expected
3875	 * _rec_size back to the info.
3876	 */
3877
3878	if ((info->nr_func_info || info->func_info_rec_size) &&
3879	    info->func_info_rec_size != sizeof(struct bpf_func_info))
3880		return -EINVAL;
3881
3882	if ((info->nr_line_info || info->line_info_rec_size) &&
3883	    info->line_info_rec_size != sizeof(struct bpf_line_info))
3884		return -EINVAL;
3885
3886	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3887	    info->jited_line_info_rec_size != sizeof(__u64))
3888		return -EINVAL;
3889
3890	info->func_info_rec_size = sizeof(struct bpf_func_info);
3891	info->line_info_rec_size = sizeof(struct bpf_line_info);
3892	info->jited_line_info_rec_size = sizeof(__u64);
3893
3894	return 0;
3895}
3896
3897static int bpf_prog_get_info_by_fd(struct file *file,
3898				   struct bpf_prog *prog,
3899				   const union bpf_attr *attr,
3900				   union bpf_attr __user *uattr)
3901{
3902	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3903	struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3904	struct bpf_prog_info info;
3905	u32 info_len = attr->info.info_len;
3906	struct bpf_prog_kstats stats;
3907	char __user *uinsns;
3908	u32 ulen;
3909	int err;
3910
3911	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3912	if (err)
3913		return err;
3914	info_len = min_t(u32, sizeof(info), info_len);
3915
3916	memset(&info, 0, sizeof(info));
3917	if (copy_from_user(&info, uinfo, info_len))
3918		return -EFAULT;
3919
3920	info.type = prog->type;
3921	info.id = prog->aux->id;
3922	info.load_time = prog->aux->load_time;
3923	info.created_by_uid = from_kuid_munged(current_user_ns(),
3924					       prog->aux->user->uid);
3925	info.gpl_compatible = prog->gpl_compatible;
3926
3927	memcpy(info.tag, prog->tag, sizeof(prog->tag));
3928	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3929
3930	mutex_lock(&prog->aux->used_maps_mutex);
3931	ulen = info.nr_map_ids;
3932	info.nr_map_ids = prog->aux->used_map_cnt;
3933	ulen = min_t(u32, info.nr_map_ids, ulen);
3934	if (ulen) {
3935		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3936		u32 i;
3937
3938		for (i = 0; i < ulen; i++)
3939			if (put_user(prog->aux->used_maps[i]->id,
3940				     &user_map_ids[i])) {
3941				mutex_unlock(&prog->aux->used_maps_mutex);
3942				return -EFAULT;
3943			}
3944	}
3945	mutex_unlock(&prog->aux->used_maps_mutex);
3946
3947	err = set_info_rec_size(&info);
3948	if (err)
3949		return err;
3950
3951	bpf_prog_get_stats(prog, &stats);
3952	info.run_time_ns = stats.nsecs;
3953	info.run_cnt = stats.cnt;
3954	info.recursion_misses = stats.misses;
3955
3956	info.verified_insns = prog->aux->verified_insns;
3957
3958	if (!bpf_capable()) {
3959		info.jited_prog_len = 0;
3960		info.xlated_prog_len = 0;
3961		info.nr_jited_ksyms = 0;
3962		info.nr_jited_func_lens = 0;
3963		info.nr_func_info = 0;
3964		info.nr_line_info = 0;
3965		info.nr_jited_line_info = 0;
3966		goto done;
3967	}
3968
3969	ulen = info.xlated_prog_len;
3970	info.xlated_prog_len = bpf_prog_insn_size(prog);
3971	if (info.xlated_prog_len && ulen) {
3972		struct bpf_insn *insns_sanitized;
3973		bool fault;
3974
3975		if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3976			info.xlated_prog_insns = 0;
3977			goto done;
3978		}
3979		insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3980		if (!insns_sanitized)
3981			return -ENOMEM;
3982		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3983		ulen = min_t(u32, info.xlated_prog_len, ulen);
3984		fault = copy_to_user(uinsns, insns_sanitized, ulen);
3985		kfree(insns_sanitized);
3986		if (fault)
3987			return -EFAULT;
3988	}
3989
3990	if (bpf_prog_is_dev_bound(prog->aux)) {
3991		err = bpf_prog_offload_info_fill(&info, prog);
3992		if (err)
3993			return err;
3994		goto done;
3995	}
3996
3997	/* NOTE: the following code is supposed to be skipped for offload.
3998	 * bpf_prog_offload_info_fill() is the place to fill similar fields
3999	 * for offload.
4000	 */
4001	ulen = info.jited_prog_len;
4002	if (prog->aux->func_cnt) {
4003		u32 i;
4004
4005		info.jited_prog_len = 0;
4006		for (i = 0; i < prog->aux->func_cnt; i++)
4007			info.jited_prog_len += prog->aux->func[i]->jited_len;
4008	} else {
4009		info.jited_prog_len = prog->jited_len;
4010	}
4011
4012	if (info.jited_prog_len && ulen) {
4013		if (bpf_dump_raw_ok(file->f_cred)) {
4014			uinsns = u64_to_user_ptr(info.jited_prog_insns);
4015			ulen = min_t(u32, info.jited_prog_len, ulen);
4016
4017			/* for multi-function programs, copy the JITed
4018			 * instructions for all the functions
4019			 */
4020			if (prog->aux->func_cnt) {
4021				u32 len, free, i;
4022				u8 *img;
4023
4024				free = ulen;
4025				for (i = 0; i < prog->aux->func_cnt; i++) {
4026					len = prog->aux->func[i]->jited_len;
4027					len = min_t(u32, len, free);
4028					img = (u8 *) prog->aux->func[i]->bpf_func;
4029					if (copy_to_user(uinsns, img, len))
4030						return -EFAULT;
4031					uinsns += len;
4032					free -= len;
4033					if (!free)
4034						break;
4035				}
4036			} else {
4037				if (copy_to_user(uinsns, prog->bpf_func, ulen))
4038					return -EFAULT;
4039			}
4040		} else {
4041			info.jited_prog_insns = 0;
4042		}
4043	}
4044
4045	ulen = info.nr_jited_ksyms;
4046	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4047	if (ulen) {
4048		if (bpf_dump_raw_ok(file->f_cred)) {
4049			unsigned long ksym_addr;
4050			u64 __user *user_ksyms;
4051			u32 i;
4052
4053			/* copy the address of the kernel symbol
4054			 * corresponding to each function
4055			 */
4056			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4057			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4058			if (prog->aux->func_cnt) {
4059				for (i = 0; i < ulen; i++) {
4060					ksym_addr = (unsigned long)
4061						prog->aux->func[i]->bpf_func;
4062					if (put_user((u64) ksym_addr,
4063						     &user_ksyms[i]))
4064						return -EFAULT;
4065				}
4066			} else {
4067				ksym_addr = (unsigned long) prog->bpf_func;
4068				if (put_user((u64) ksym_addr, &user_ksyms[0]))
4069					return -EFAULT;
4070			}
4071		} else {
4072			info.jited_ksyms = 0;
4073		}
4074	}
4075
4076	ulen = info.nr_jited_func_lens;
4077	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4078	if (ulen) {
4079		if (bpf_dump_raw_ok(file->f_cred)) {
4080			u32 __user *user_lens;
4081			u32 func_len, i;
4082
4083			/* copy the JITed image lengths for each function */
4084			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4085			user_lens = u64_to_user_ptr(info.jited_func_lens);
4086			if (prog->aux->func_cnt) {
4087				for (i = 0; i < ulen; i++) {
4088					func_len =
4089						prog->aux->func[i]->jited_len;
4090					if (put_user(func_len, &user_lens[i]))
4091						return -EFAULT;
4092				}
4093			} else {
4094				func_len = prog->jited_len;
4095				if (put_user(func_len, &user_lens[0]))
4096					return -EFAULT;
4097			}
4098		} else {
4099			info.jited_func_lens = 0;
4100		}
4101	}
4102
4103	if (prog->aux->btf)
4104		info.btf_id = btf_obj_id(prog->aux->btf);
4105	info.attach_btf_id = prog->aux->attach_btf_id;
4106	if (attach_btf)
4107		info.attach_btf_obj_id = btf_obj_id(attach_btf);
4108
4109	ulen = info.nr_func_info;
4110	info.nr_func_info = prog->aux->func_info_cnt;
4111	if (info.nr_func_info && ulen) {
4112		char __user *user_finfo;
4113
4114		user_finfo = u64_to_user_ptr(info.func_info);
4115		ulen = min_t(u32, info.nr_func_info, ulen);
4116		if (copy_to_user(user_finfo, prog->aux->func_info,
4117				 info.func_info_rec_size * ulen))
4118			return -EFAULT;
4119	}
4120
4121	ulen = info.nr_line_info;
4122	info.nr_line_info = prog->aux->nr_linfo;
4123	if (info.nr_line_info && ulen) {
4124		__u8 __user *user_linfo;
4125
4126		user_linfo = u64_to_user_ptr(info.line_info);
4127		ulen = min_t(u32, info.nr_line_info, ulen);
4128		if (copy_to_user(user_linfo, prog->aux->linfo,
4129				 info.line_info_rec_size * ulen))
4130			return -EFAULT;
4131	}
4132
4133	ulen = info.nr_jited_line_info;
4134	if (prog->aux->jited_linfo)
4135		info.nr_jited_line_info = prog->aux->nr_linfo;
4136	else
4137		info.nr_jited_line_info = 0;
4138	if (info.nr_jited_line_info && ulen) {
4139		if (bpf_dump_raw_ok(file->f_cred)) {
4140			unsigned long line_addr;
4141			__u64 __user *user_linfo;
4142			u32 i;
4143
4144			user_linfo = u64_to_user_ptr(info.jited_line_info);
4145			ulen = min_t(u32, info.nr_jited_line_info, ulen);
4146			for (i = 0; i < ulen; i++) {
4147				line_addr = (unsigned long)prog->aux->jited_linfo[i];
4148				if (put_user((__u64)line_addr, &user_linfo[i]))
4149					return -EFAULT;
4150			}
4151		} else {
4152			info.jited_line_info = 0;
4153		}
4154	}
4155
4156	ulen = info.nr_prog_tags;
4157	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4158	if (ulen) {
4159		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4160		u32 i;
4161
4162		user_prog_tags = u64_to_user_ptr(info.prog_tags);
4163		ulen = min_t(u32, info.nr_prog_tags, ulen);
4164		if (prog->aux->func_cnt) {
4165			for (i = 0; i < ulen; i++) {
4166				if (copy_to_user(user_prog_tags[i],
4167						 prog->aux->func[i]->tag,
4168						 BPF_TAG_SIZE))
4169					return -EFAULT;
4170			}
4171		} else {
4172			if (copy_to_user(user_prog_tags[0],
4173					 prog->tag, BPF_TAG_SIZE))
4174				return -EFAULT;
4175		}
4176	}
4177
4178done:
4179	if (copy_to_user(uinfo, &info, info_len) ||
4180	    put_user(info_len, &uattr->info.info_len))
4181		return -EFAULT;
4182
4183	return 0;
4184}
4185
4186static int bpf_map_get_info_by_fd(struct file *file,
4187				  struct bpf_map *map,
4188				  const union bpf_attr *attr,
4189				  union bpf_attr __user *uattr)
4190{
4191	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4192	struct bpf_map_info info;
4193	u32 info_len = attr->info.info_len;
4194	int err;
4195
4196	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4197	if (err)
4198		return err;
4199	info_len = min_t(u32, sizeof(info), info_len);
4200
4201	memset(&info, 0, sizeof(info));
4202	info.type = map->map_type;
4203	info.id = map->id;
4204	info.key_size = map->key_size;
4205	info.value_size = map->value_size;
4206	info.max_entries = map->max_entries;
4207	info.map_flags = map->map_flags;
4208	info.map_extra = map->map_extra;
4209	memcpy(info.name, map->name, sizeof(map->name));
4210
4211	if (map->btf) {
4212		info.btf_id = btf_obj_id(map->btf);
4213		info.btf_key_type_id = map->btf_key_type_id;
4214		info.btf_value_type_id = map->btf_value_type_id;
4215	}
4216	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
 
 
4217
4218	if (bpf_map_is_dev_bound(map)) {
4219		err = bpf_map_offload_info_fill(&info, map);
4220		if (err)
4221			return err;
4222	}
4223
4224	if (copy_to_user(uinfo, &info, info_len) ||
4225	    put_user(info_len, &uattr->info.info_len))
4226		return -EFAULT;
4227
4228	return 0;
4229}
4230
4231static int bpf_btf_get_info_by_fd(struct file *file,
4232				  struct btf *btf,
4233				  const union bpf_attr *attr,
4234				  union bpf_attr __user *uattr)
4235{
4236	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4237	u32 info_len = attr->info.info_len;
4238	int err;
4239
4240	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4241	if (err)
4242		return err;
4243
4244	return btf_get_info_by_fd(btf, attr, uattr);
4245}
4246
4247static int bpf_link_get_info_by_fd(struct file *file,
4248				  struct bpf_link *link,
4249				  const union bpf_attr *attr,
4250				  union bpf_attr __user *uattr)
4251{
4252	struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4253	struct bpf_link_info info;
4254	u32 info_len = attr->info.info_len;
4255	int err;
4256
4257	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4258	if (err)
4259		return err;
4260	info_len = min_t(u32, sizeof(info), info_len);
4261
4262	memset(&info, 0, sizeof(info));
4263	if (copy_from_user(&info, uinfo, info_len))
4264		return -EFAULT;
4265
4266	info.type = link->type;
4267	info.id = link->id;
4268	info.prog_id = link->prog->aux->id;
 
4269
4270	if (link->ops->fill_link_info) {
4271		err = link->ops->fill_link_info(link, &info);
4272		if (err)
4273			return err;
4274	}
4275
4276	if (copy_to_user(uinfo, &info, info_len) ||
4277	    put_user(info_len, &uattr->info.info_len))
4278		return -EFAULT;
4279
4280	return 0;
4281}
4282
4283
4284#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4285
4286static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4287				  union bpf_attr __user *uattr)
4288{
4289	int ufd = attr->info.bpf_fd;
4290	struct fd f;
4291	int err;
4292
4293	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4294		return -EINVAL;
4295
4296	f = fdget(ufd);
4297	if (!f.file)
4298		return -EBADFD;
4299
4300	if (f.file->f_op == &bpf_prog_fops)
4301		err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4302					      uattr);
4303	else if (f.file->f_op == &bpf_map_fops)
4304		err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4305					     uattr);
4306	else if (f.file->f_op == &btf_fops)
4307		err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4308	else if (f.file->f_op == &bpf_link_fops)
4309		err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4310					      attr, uattr);
4311	else
4312		err = -EINVAL;
4313
4314	fdput(f);
4315	return err;
4316}
4317
4318#define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4319
4320static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
4321{
 
 
4322	if (CHECK_ATTR(BPF_BTF_LOAD))
4323		return -EINVAL;
4324
4325	if (!bpf_capable())
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4326		return -EPERM;
 
4327
4328	return btf_new_fd(attr, uattr);
 
 
4329}
4330
4331#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4332
4333static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4334{
4335	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4336		return -EINVAL;
4337
4338	if (!capable(CAP_SYS_ADMIN))
4339		return -EPERM;
4340
4341	return btf_get_fd_by_id(attr->btf_id);
4342}
4343
4344static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4345				    union bpf_attr __user *uattr,
4346				    u32 prog_id, u32 fd_type,
4347				    const char *buf, u64 probe_offset,
4348				    u64 probe_addr)
4349{
4350	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4351	u32 len = buf ? strlen(buf) : 0, input_len;
4352	int err = 0;
4353
4354	if (put_user(len, &uattr->task_fd_query.buf_len))
4355		return -EFAULT;
4356	input_len = attr->task_fd_query.buf_len;
4357	if (input_len && ubuf) {
4358		if (!len) {
4359			/* nothing to copy, just make ubuf NULL terminated */
4360			char zero = '\0';
4361
4362			if (put_user(zero, ubuf))
4363				return -EFAULT;
4364		} else if (input_len >= len + 1) {
4365			/* ubuf can hold the string with NULL terminator */
4366			if (copy_to_user(ubuf, buf, len + 1))
4367				return -EFAULT;
4368		} else {
4369			/* ubuf cannot hold the string with NULL terminator,
4370			 * do a partial copy with NULL terminator.
4371			 */
4372			char zero = '\0';
4373
4374			err = -ENOSPC;
4375			if (copy_to_user(ubuf, buf, input_len - 1))
4376				return -EFAULT;
4377			if (put_user(zero, ubuf + input_len - 1))
4378				return -EFAULT;
4379		}
4380	}
4381
4382	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4383	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4384	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4385	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4386		return -EFAULT;
4387
4388	return err;
4389}
4390
4391#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4392
4393static int bpf_task_fd_query(const union bpf_attr *attr,
4394			     union bpf_attr __user *uattr)
4395{
4396	pid_t pid = attr->task_fd_query.pid;
4397	u32 fd = attr->task_fd_query.fd;
4398	const struct perf_event *event;
4399	struct task_struct *task;
4400	struct file *file;
4401	int err;
4402
4403	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4404		return -EINVAL;
4405
4406	if (!capable(CAP_SYS_ADMIN))
4407		return -EPERM;
4408
4409	if (attr->task_fd_query.flags != 0)
4410		return -EINVAL;
4411
4412	rcu_read_lock();
4413	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4414	rcu_read_unlock();
4415	if (!task)
4416		return -ENOENT;
4417
4418	err = 0;
4419	file = fget_task(task, fd);
4420	put_task_struct(task);
4421	if (!file)
4422		return -EBADF;
4423
4424	if (file->f_op == &bpf_link_fops) {
4425		struct bpf_link *link = file->private_data;
4426
4427		if (link->ops == &bpf_raw_tp_link_lops) {
4428			struct bpf_raw_tp_link *raw_tp =
4429				container_of(link, struct bpf_raw_tp_link, link);
4430			struct bpf_raw_event_map *btp = raw_tp->btp;
4431
4432			err = bpf_task_fd_query_copy(attr, uattr,
4433						     raw_tp->link.prog->aux->id,
4434						     BPF_FD_TYPE_RAW_TRACEPOINT,
4435						     btp->tp->name, 0, 0);
4436			goto put_file;
4437		}
4438		goto out_not_supp;
4439	}
4440
4441	event = perf_get_event(file);
4442	if (!IS_ERR(event)) {
4443		u64 probe_offset, probe_addr;
4444		u32 prog_id, fd_type;
4445		const char *buf;
4446
4447		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4448					      &buf, &probe_offset,
4449					      &probe_addr);
4450		if (!err)
4451			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4452						     fd_type, buf,
4453						     probe_offset,
4454						     probe_addr);
4455		goto put_file;
4456	}
4457
4458out_not_supp:
4459	err = -ENOTSUPP;
4460put_file:
4461	fput(file);
4462	return err;
4463}
4464
4465#define BPF_MAP_BATCH_LAST_FIELD batch.flags
4466
4467#define BPF_DO_BATCH(fn, ...)			\
4468	do {					\
4469		if (!fn) {			\
4470			err = -ENOTSUPP;	\
4471			goto err_put;		\
4472		}				\
4473		err = fn(__VA_ARGS__);		\
4474	} while (0)
4475
4476static int bpf_map_do_batch(const union bpf_attr *attr,
4477			    union bpf_attr __user *uattr,
4478			    int cmd)
4479{
4480	bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4481			 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4482	bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4483	struct bpf_map *map;
4484	int err, ufd;
4485	struct fd f;
4486
4487	if (CHECK_ATTR(BPF_MAP_BATCH))
4488		return -EINVAL;
4489
4490	ufd = attr->batch.map_fd;
4491	f = fdget(ufd);
4492	map = __bpf_map_get(f);
4493	if (IS_ERR(map))
4494		return PTR_ERR(map);
4495	if (has_write)
4496		bpf_map_write_active_inc(map);
4497	if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4498		err = -EPERM;
4499		goto err_put;
4500	}
4501	if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4502		err = -EPERM;
4503		goto err_put;
4504	}
4505
4506	if (cmd == BPF_MAP_LOOKUP_BATCH)
4507		BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4508	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4509		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4510	else if (cmd == BPF_MAP_UPDATE_BATCH)
4511		BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4512	else
4513		BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4514err_put:
4515	if (has_write)
 
4516		bpf_map_write_active_dec(map);
 
4517	fdput(f);
4518	return err;
4519}
4520
4521#define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4522static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4523{
4524	enum bpf_prog_type ptype;
4525	struct bpf_prog *prog;
4526	int ret;
4527
4528	if (CHECK_ATTR(BPF_LINK_CREATE))
4529		return -EINVAL;
4530
 
 
 
4531	prog = bpf_prog_get(attr->link_create.prog_fd);
4532	if (IS_ERR(prog))
4533		return PTR_ERR(prog);
4534
4535	ret = bpf_prog_attach_check_attach_type(prog,
4536						attr->link_create.attach_type);
4537	if (ret)
4538		goto out;
4539
4540	switch (prog->type) {
4541	case BPF_PROG_TYPE_EXT:
4542		break;
4543	case BPF_PROG_TYPE_PERF_EVENT:
4544	case BPF_PROG_TYPE_TRACEPOINT:
4545		if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4546			ret = -EINVAL;
4547			goto out;
4548		}
4549		break;
4550	case BPF_PROG_TYPE_KPROBE:
4551		if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4552		    attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4553			ret = -EINVAL;
4554			goto out;
4555		}
4556		break;
4557	default:
4558		ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4559		if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4560			ret = -EINVAL;
4561			goto out;
4562		}
4563		break;
4564	}
4565
4566	switch (prog->type) {
4567	case BPF_PROG_TYPE_CGROUP_SKB:
4568	case BPF_PROG_TYPE_CGROUP_SOCK:
4569	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4570	case BPF_PROG_TYPE_SOCK_OPS:
4571	case BPF_PROG_TYPE_CGROUP_DEVICE:
4572	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4573	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4574		ret = cgroup_bpf_link_attach(attr, prog);
4575		break;
4576	case BPF_PROG_TYPE_EXT:
4577		ret = bpf_tracing_prog_attach(prog,
4578					      attr->link_create.target_fd,
4579					      attr->link_create.target_btf_id,
4580					      attr->link_create.tracing.cookie);
4581		break;
4582	case BPF_PROG_TYPE_LSM:
4583	case BPF_PROG_TYPE_TRACING:
4584		if (attr->link_create.attach_type != prog->expected_attach_type) {
4585			ret = -EINVAL;
4586			goto out;
4587		}
4588		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4589			ret = bpf_raw_tp_link_attach(prog, NULL);
4590		else if (prog->expected_attach_type == BPF_TRACE_ITER)
4591			ret = bpf_iter_link_attach(attr, uattr, prog);
4592		else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4593			ret = cgroup_bpf_link_attach(attr, prog);
4594		else
4595			ret = bpf_tracing_prog_attach(prog,
4596						      attr->link_create.target_fd,
4597						      attr->link_create.target_btf_id,
4598						      attr->link_create.tracing.cookie);
4599		break;
4600	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4601	case BPF_PROG_TYPE_SK_LOOKUP:
4602		ret = netns_bpf_link_create(attr, prog);
4603		break;
4604#ifdef CONFIG_NET
4605	case BPF_PROG_TYPE_XDP:
4606		ret = bpf_xdp_link_attach(attr, prog);
4607		break;
 
 
 
 
 
 
 
 
 
 
4608#endif
4609	case BPF_PROG_TYPE_PERF_EVENT:
4610	case BPF_PROG_TYPE_TRACEPOINT:
4611		ret = bpf_perf_link_attach(attr, prog);
4612		break;
4613	case BPF_PROG_TYPE_KPROBE:
4614		if (attr->link_create.attach_type == BPF_PERF_EVENT)
4615			ret = bpf_perf_link_attach(attr, prog);
4616		else
4617			ret = bpf_kprobe_multi_link_attach(attr, prog);
 
 
4618		break;
4619	default:
4620		ret = -EINVAL;
4621	}
4622
4623out:
4624	if (ret < 0)
4625		bpf_prog_put(prog);
4626	return ret;
4627}
4628
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4629#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4630
4631static int link_update(union bpf_attr *attr)
4632{
4633	struct bpf_prog *old_prog = NULL, *new_prog;
4634	struct bpf_link *link;
4635	u32 flags;
4636	int ret;
4637
4638	if (CHECK_ATTR(BPF_LINK_UPDATE))
4639		return -EINVAL;
4640
4641	flags = attr->link_update.flags;
4642	if (flags & ~BPF_F_REPLACE)
4643		return -EINVAL;
4644
4645	link = bpf_link_get_from_fd(attr->link_update.link_fd);
4646	if (IS_ERR(link))
4647		return PTR_ERR(link);
4648
 
 
 
 
 
4649	new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4650	if (IS_ERR(new_prog)) {
4651		ret = PTR_ERR(new_prog);
4652		goto out_put_link;
4653	}
4654
4655	if (flags & BPF_F_REPLACE) {
4656		old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4657		if (IS_ERR(old_prog)) {
4658			ret = PTR_ERR(old_prog);
4659			old_prog = NULL;
4660			goto out_put_progs;
4661		}
4662	} else if (attr->link_update.old_prog_fd) {
4663		ret = -EINVAL;
4664		goto out_put_progs;
4665	}
4666
4667	if (link->ops->update_prog)
4668		ret = link->ops->update_prog(link, new_prog, old_prog);
4669	else
4670		ret = -EINVAL;
4671
4672out_put_progs:
4673	if (old_prog)
4674		bpf_prog_put(old_prog);
4675	if (ret)
4676		bpf_prog_put(new_prog);
4677out_put_link:
4678	bpf_link_put(link);
4679	return ret;
4680}
4681
4682#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4683
4684static int link_detach(union bpf_attr *attr)
4685{
4686	struct bpf_link *link;
4687	int ret;
4688
4689	if (CHECK_ATTR(BPF_LINK_DETACH))
4690		return -EINVAL;
4691
4692	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4693	if (IS_ERR(link))
4694		return PTR_ERR(link);
4695
4696	if (link->ops->detach)
4697		ret = link->ops->detach(link);
4698	else
4699		ret = -EOPNOTSUPP;
4700
4701	bpf_link_put(link);
4702	return ret;
4703}
4704
4705static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4706{
4707	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4708}
4709
4710struct bpf_link *bpf_link_by_id(u32 id)
4711{
4712	struct bpf_link *link;
4713
4714	if (!id)
4715		return ERR_PTR(-ENOENT);
4716
4717	spin_lock_bh(&link_idr_lock);
4718	/* before link is "settled", ID is 0, pretend it doesn't exist yet */
4719	link = idr_find(&link_idr, id);
4720	if (link) {
4721		if (link->id)
4722			link = bpf_link_inc_not_zero(link);
4723		else
4724			link = ERR_PTR(-EAGAIN);
4725	} else {
4726		link = ERR_PTR(-ENOENT);
4727	}
4728	spin_unlock_bh(&link_idr_lock);
4729	return link;
4730}
4731
4732struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4733{
4734	struct bpf_link *link;
4735
4736	spin_lock_bh(&link_idr_lock);
4737again:
4738	link = idr_get_next(&link_idr, id);
4739	if (link) {
4740		link = bpf_link_inc_not_zero(link);
4741		if (IS_ERR(link)) {
4742			(*id)++;
4743			goto again;
4744		}
4745	}
4746	spin_unlock_bh(&link_idr_lock);
4747
4748	return link;
4749}
4750
4751#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4752
4753static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4754{
4755	struct bpf_link *link;
4756	u32 id = attr->link_id;
4757	int fd;
4758
4759	if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4760		return -EINVAL;
4761
4762	if (!capable(CAP_SYS_ADMIN))
4763		return -EPERM;
4764
4765	link = bpf_link_by_id(id);
4766	if (IS_ERR(link))
4767		return PTR_ERR(link);
4768
4769	fd = bpf_link_new_fd(link);
4770	if (fd < 0)
4771		bpf_link_put(link);
4772
4773	return fd;
4774}
4775
4776DEFINE_MUTEX(bpf_stats_enabled_mutex);
4777
4778static int bpf_stats_release(struct inode *inode, struct file *file)
4779{
4780	mutex_lock(&bpf_stats_enabled_mutex);
4781	static_key_slow_dec(&bpf_stats_enabled_key.key);
4782	mutex_unlock(&bpf_stats_enabled_mutex);
4783	return 0;
4784}
4785
4786static const struct file_operations bpf_stats_fops = {
4787	.release = bpf_stats_release,
4788};
4789
4790static int bpf_enable_runtime_stats(void)
4791{
4792	int fd;
4793
4794	mutex_lock(&bpf_stats_enabled_mutex);
4795
4796	/* Set a very high limit to avoid overflow */
4797	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4798		mutex_unlock(&bpf_stats_enabled_mutex);
4799		return -EBUSY;
4800	}
4801
4802	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4803	if (fd >= 0)
4804		static_key_slow_inc(&bpf_stats_enabled_key.key);
4805
4806	mutex_unlock(&bpf_stats_enabled_mutex);
4807	return fd;
4808}
4809
4810#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4811
4812static int bpf_enable_stats(union bpf_attr *attr)
4813{
4814
4815	if (CHECK_ATTR(BPF_ENABLE_STATS))
4816		return -EINVAL;
4817
4818	if (!capable(CAP_SYS_ADMIN))
4819		return -EPERM;
4820
4821	switch (attr->enable_stats.type) {
4822	case BPF_STATS_RUN_TIME:
4823		return bpf_enable_runtime_stats();
4824	default:
4825		break;
4826	}
4827	return -EINVAL;
4828}
4829
4830#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4831
4832static int bpf_iter_create(union bpf_attr *attr)
4833{
4834	struct bpf_link *link;
4835	int err;
4836
4837	if (CHECK_ATTR(BPF_ITER_CREATE))
4838		return -EINVAL;
4839
4840	if (attr->iter_create.flags)
4841		return -EINVAL;
4842
4843	link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4844	if (IS_ERR(link))
4845		return PTR_ERR(link);
4846
4847	err = bpf_iter_new_fd(link);
4848	bpf_link_put(link);
4849
4850	return err;
4851}
4852
4853#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4854
4855static int bpf_prog_bind_map(union bpf_attr *attr)
4856{
4857	struct bpf_prog *prog;
4858	struct bpf_map *map;
4859	struct bpf_map **used_maps_old, **used_maps_new;
4860	int i, ret = 0;
4861
4862	if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4863		return -EINVAL;
4864
4865	if (attr->prog_bind_map.flags)
4866		return -EINVAL;
4867
4868	prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4869	if (IS_ERR(prog))
4870		return PTR_ERR(prog);
4871
4872	map = bpf_map_get(attr->prog_bind_map.map_fd);
4873	if (IS_ERR(map)) {
4874		ret = PTR_ERR(map);
4875		goto out_prog_put;
4876	}
4877
4878	mutex_lock(&prog->aux->used_maps_mutex);
4879
4880	used_maps_old = prog->aux->used_maps;
4881
4882	for (i = 0; i < prog->aux->used_map_cnt; i++)
4883		if (used_maps_old[i] == map) {
4884			bpf_map_put(map);
4885			goto out_unlock;
4886		}
4887
4888	used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4889				      sizeof(used_maps_new[0]),
4890				      GFP_KERNEL);
4891	if (!used_maps_new) {
4892		ret = -ENOMEM;
4893		goto out_unlock;
4894	}
4895
 
 
 
 
 
4896	memcpy(used_maps_new, used_maps_old,
4897	       sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4898	used_maps_new[prog->aux->used_map_cnt] = map;
4899
4900	prog->aux->used_map_cnt++;
4901	prog->aux->used_maps = used_maps_new;
4902
4903	kfree(used_maps_old);
4904
4905out_unlock:
4906	mutex_unlock(&prog->aux->used_maps_mutex);
4907
4908	if (ret)
4909		bpf_map_put(map);
4910out_prog_put:
4911	bpf_prog_put(prog);
4912	return ret;
4913}
4914
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4915static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4916{
4917	union bpf_attr attr;
4918	bool capable;
4919	int err;
4920
4921	capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
4922
4923	/* Intent here is for unprivileged_bpf_disabled to block key object
4924	 * creation commands for unprivileged users; other actions depend
4925	 * of fd availability and access to bpffs, so are dependent on
4926	 * object creation success.  Capabilities are later verified for
4927	 * operations such as load and map create, so even with unprivileged
4928	 * BPF disabled, capability checks are still carried out for these
4929	 * and other operations.
4930	 */
4931	if (!capable &&
4932	    (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
4933		return -EPERM;
4934
4935	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4936	if (err)
4937		return err;
4938	size = min_t(u32, size, sizeof(attr));
4939
4940	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
4941	memset(&attr, 0, sizeof(attr));
4942	if (copy_from_bpfptr(&attr, uattr, size) != 0)
4943		return -EFAULT;
4944
4945	err = security_bpf(cmd, &attr, size);
4946	if (err < 0)
4947		return err;
4948
4949	switch (cmd) {
4950	case BPF_MAP_CREATE:
4951		err = map_create(&attr);
4952		break;
4953	case BPF_MAP_LOOKUP_ELEM:
4954		err = map_lookup_elem(&attr);
4955		break;
4956	case BPF_MAP_UPDATE_ELEM:
4957		err = map_update_elem(&attr, uattr);
4958		break;
4959	case BPF_MAP_DELETE_ELEM:
4960		err = map_delete_elem(&attr, uattr);
4961		break;
4962	case BPF_MAP_GET_NEXT_KEY:
4963		err = map_get_next_key(&attr);
4964		break;
4965	case BPF_MAP_FREEZE:
4966		err = map_freeze(&attr);
4967		break;
4968	case BPF_PROG_LOAD:
4969		err = bpf_prog_load(&attr, uattr);
4970		break;
4971	case BPF_OBJ_PIN:
4972		err = bpf_obj_pin(&attr);
4973		break;
4974	case BPF_OBJ_GET:
4975		err = bpf_obj_get(&attr);
4976		break;
4977	case BPF_PROG_ATTACH:
4978		err = bpf_prog_attach(&attr);
4979		break;
4980	case BPF_PROG_DETACH:
4981		err = bpf_prog_detach(&attr);
4982		break;
4983	case BPF_PROG_QUERY:
4984		err = bpf_prog_query(&attr, uattr.user);
4985		break;
4986	case BPF_PROG_TEST_RUN:
4987		err = bpf_prog_test_run(&attr, uattr.user);
4988		break;
4989	case BPF_PROG_GET_NEXT_ID:
4990		err = bpf_obj_get_next_id(&attr, uattr.user,
4991					  &prog_idr, &prog_idr_lock);
4992		break;
4993	case BPF_MAP_GET_NEXT_ID:
4994		err = bpf_obj_get_next_id(&attr, uattr.user,
4995					  &map_idr, &map_idr_lock);
4996		break;
4997	case BPF_BTF_GET_NEXT_ID:
4998		err = bpf_obj_get_next_id(&attr, uattr.user,
4999					  &btf_idr, &btf_idr_lock);
5000		break;
5001	case BPF_PROG_GET_FD_BY_ID:
5002		err = bpf_prog_get_fd_by_id(&attr);
5003		break;
5004	case BPF_MAP_GET_FD_BY_ID:
5005		err = bpf_map_get_fd_by_id(&attr);
5006		break;
5007	case BPF_OBJ_GET_INFO_BY_FD:
5008		err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5009		break;
5010	case BPF_RAW_TRACEPOINT_OPEN:
5011		err = bpf_raw_tracepoint_open(&attr);
5012		break;
5013	case BPF_BTF_LOAD:
5014		err = bpf_btf_load(&attr, uattr);
5015		break;
5016	case BPF_BTF_GET_FD_BY_ID:
5017		err = bpf_btf_get_fd_by_id(&attr);
5018		break;
5019	case BPF_TASK_FD_QUERY:
5020		err = bpf_task_fd_query(&attr, uattr.user);
5021		break;
5022	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5023		err = map_lookup_and_delete_elem(&attr);
5024		break;
5025	case BPF_MAP_LOOKUP_BATCH:
5026		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5027		break;
5028	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5029		err = bpf_map_do_batch(&attr, uattr.user,
5030				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5031		break;
5032	case BPF_MAP_UPDATE_BATCH:
5033		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5034		break;
5035	case BPF_MAP_DELETE_BATCH:
5036		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5037		break;
5038	case BPF_LINK_CREATE:
5039		err = link_create(&attr, uattr);
5040		break;
5041	case BPF_LINK_UPDATE:
5042		err = link_update(&attr);
5043		break;
5044	case BPF_LINK_GET_FD_BY_ID:
5045		err = bpf_link_get_fd_by_id(&attr);
5046		break;
5047	case BPF_LINK_GET_NEXT_ID:
5048		err = bpf_obj_get_next_id(&attr, uattr.user,
5049					  &link_idr, &link_idr_lock);
5050		break;
5051	case BPF_ENABLE_STATS:
5052		err = bpf_enable_stats(&attr);
5053		break;
5054	case BPF_ITER_CREATE:
5055		err = bpf_iter_create(&attr);
5056		break;
5057	case BPF_LINK_DETACH:
5058		err = link_detach(&attr);
5059		break;
5060	case BPF_PROG_BIND_MAP:
5061		err = bpf_prog_bind_map(&attr);
5062		break;
 
 
 
5063	default:
5064		err = -EINVAL;
5065		break;
5066	}
5067
5068	return err;
5069}
5070
5071SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5072{
5073	return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5074}
5075
5076static bool syscall_prog_is_valid_access(int off, int size,
5077					 enum bpf_access_type type,
5078					 const struct bpf_prog *prog,
5079					 struct bpf_insn_access_aux *info)
5080{
5081	if (off < 0 || off >= U16_MAX)
5082		return false;
5083	if (off % size != 0)
5084		return false;
5085	return true;
5086}
5087
5088BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5089{
5090	switch (cmd) {
5091	case BPF_MAP_CREATE:
5092	case BPF_MAP_DELETE_ELEM:
5093	case BPF_MAP_UPDATE_ELEM:
5094	case BPF_MAP_FREEZE:
5095	case BPF_MAP_GET_FD_BY_ID:
5096	case BPF_PROG_LOAD:
5097	case BPF_BTF_LOAD:
5098	case BPF_LINK_CREATE:
5099	case BPF_RAW_TRACEPOINT_OPEN:
5100		break;
5101	default:
5102		return -EINVAL;
5103	}
5104	return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5105}
5106
5107
5108/* To shut up -Wmissing-prototypes.
5109 * This function is used by the kernel light skeleton
5110 * to load bpf programs when modules are loaded or during kernel boot.
5111 * See tools/lib/bpf/skel_internal.h
5112 */
5113int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5114
5115int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5116{
5117	struct bpf_prog * __maybe_unused prog;
5118	struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5119
5120	switch (cmd) {
5121#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5122	case BPF_PROG_TEST_RUN:
5123		if (attr->test.data_in || attr->test.data_out ||
5124		    attr->test.ctx_out || attr->test.duration ||
5125		    attr->test.repeat || attr->test.flags)
5126			return -EINVAL;
5127
5128		prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5129		if (IS_ERR(prog))
5130			return PTR_ERR(prog);
5131
5132		if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5133		    attr->test.ctx_size_in > U16_MAX) {
5134			bpf_prog_put(prog);
5135			return -EINVAL;
5136		}
5137
5138		run_ctx.bpf_cookie = 0;
5139		run_ctx.saved_run_ctx = NULL;
5140		if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5141			/* recursion detected */
 
5142			bpf_prog_put(prog);
5143			return -EBUSY;
5144		}
5145		attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5146		__bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5147						&run_ctx);
5148		bpf_prog_put(prog);
5149		return 0;
5150#endif
5151	default:
5152		return ____bpf_sys_bpf(cmd, attr, size);
5153	}
5154}
5155EXPORT_SYMBOL(kern_sys_bpf);
5156
5157static const struct bpf_func_proto bpf_sys_bpf_proto = {
5158	.func		= bpf_sys_bpf,
5159	.gpl_only	= false,
5160	.ret_type	= RET_INTEGER,
5161	.arg1_type	= ARG_ANYTHING,
5162	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
5163	.arg3_type	= ARG_CONST_SIZE,
5164};
5165
5166const struct bpf_func_proto * __weak
5167tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5168{
5169	return bpf_base_func_proto(func_id);
5170}
5171
5172BPF_CALL_1(bpf_sys_close, u32, fd)
5173{
5174	/* When bpf program calls this helper there should not be
5175	 * an fdget() without matching completed fdput().
5176	 * This helper is allowed in the following callchain only:
5177	 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5178	 */
5179	return close_fd(fd);
5180}
5181
5182static const struct bpf_func_proto bpf_sys_close_proto = {
5183	.func		= bpf_sys_close,
5184	.gpl_only	= false,
5185	.ret_type	= RET_INTEGER,
5186	.arg1_type	= ARG_ANYTHING,
5187};
5188
5189BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5190{
5191	if (flags)
5192		return -EINVAL;
5193
5194	if (name_sz <= 1 || name[name_sz - 1])
5195		return -EINVAL;
5196
5197	if (!bpf_dump_raw_ok(current_cred()))
5198		return -EPERM;
5199
5200	*res = kallsyms_lookup_name(name);
5201	return *res ? 0 : -ENOENT;
5202}
5203
5204static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5205	.func		= bpf_kallsyms_lookup_name,
5206	.gpl_only	= false,
5207	.ret_type	= RET_INTEGER,
5208	.arg1_type	= ARG_PTR_TO_MEM,
5209	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
5210	.arg3_type	= ARG_ANYTHING,
5211	.arg4_type	= ARG_PTR_TO_LONG,
5212};
5213
5214static const struct bpf_func_proto *
5215syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5216{
5217	switch (func_id) {
5218	case BPF_FUNC_sys_bpf:
5219		return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
 
5220	case BPF_FUNC_btf_find_by_name_kind:
5221		return &bpf_btf_find_by_name_kind_proto;
5222	case BPF_FUNC_sys_close:
5223		return &bpf_sys_close_proto;
5224	case BPF_FUNC_kallsyms_lookup_name:
5225		return &bpf_kallsyms_lookup_name_proto;
5226	default:
5227		return tracing_prog_func_proto(func_id, prog);
5228	}
5229}
5230
5231const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5232	.get_func_proto  = syscall_prog_func_proto,
5233	.is_valid_access = syscall_prog_is_valid_access,
5234};
5235
5236const struct bpf_prog_ops bpf_syscall_prog_ops = {
5237	.test_run = bpf_prog_test_run_syscall,
5238};
5239
5240#ifdef CONFIG_SYSCTL
5241static int bpf_stats_handler(struct ctl_table *table, int write,
5242			     void *buffer, size_t *lenp, loff_t *ppos)
5243{
5244	struct static_key *key = (struct static_key *)table->data;
5245	static int saved_val;
5246	int val, ret;
5247	struct ctl_table tmp = {
5248		.data   = &val,
5249		.maxlen = sizeof(val),
5250		.mode   = table->mode,
5251		.extra1 = SYSCTL_ZERO,
5252		.extra2 = SYSCTL_ONE,
5253	};
5254
5255	if (write && !capable(CAP_SYS_ADMIN))
5256		return -EPERM;
5257
5258	mutex_lock(&bpf_stats_enabled_mutex);
5259	val = saved_val;
5260	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5261	if (write && !ret && val != saved_val) {
5262		if (val)
5263			static_key_slow_inc(key);
5264		else
5265			static_key_slow_dec(key);
5266		saved_val = val;
5267	}
5268	mutex_unlock(&bpf_stats_enabled_mutex);
5269	return ret;
5270}
5271
5272void __weak unpriv_ebpf_notify(int new_state)
5273{
5274}
5275
5276static int bpf_unpriv_handler(struct ctl_table *table, int write,
5277			      void *buffer, size_t *lenp, loff_t *ppos)
5278{
5279	int ret, unpriv_enable = *(int *)table->data;
5280	bool locked_state = unpriv_enable == 1;
5281	struct ctl_table tmp = *table;
5282
5283	if (write && !capable(CAP_SYS_ADMIN))
5284		return -EPERM;
5285
5286	tmp.data = &unpriv_enable;
5287	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5288	if (write && !ret) {
5289		if (locked_state && unpriv_enable != 1)
5290			return -EPERM;
5291		*(int *)table->data = unpriv_enable;
5292	}
5293
5294	unpriv_ebpf_notify(unpriv_enable);
 
5295
5296	return ret;
5297}
5298
5299static struct ctl_table bpf_syscall_table[] = {
5300	{
5301		.procname	= "unprivileged_bpf_disabled",
5302		.data		= &sysctl_unprivileged_bpf_disabled,
5303		.maxlen		= sizeof(sysctl_unprivileged_bpf_disabled),
5304		.mode		= 0644,
5305		.proc_handler	= bpf_unpriv_handler,
5306		.extra1		= SYSCTL_ZERO,
5307		.extra2		= SYSCTL_TWO,
5308	},
5309	{
5310		.procname	= "bpf_stats_enabled",
5311		.data		= &bpf_stats_enabled_key.key,
5312		.maxlen		= sizeof(bpf_stats_enabled_key),
5313		.mode		= 0644,
5314		.proc_handler	= bpf_stats_handler,
5315	},
5316	{ }
5317};
5318
5319static int __init bpf_syscall_sysctl_init(void)
5320{
5321	register_sysctl_init("kernel", bpf_syscall_table);
5322	return 0;
5323}
5324late_initcall(bpf_syscall_sysctl_init);
5325#endif /* CONFIG_SYSCTL */

   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-cgroup.h>
   6#include <linux/bpf_trace.h>
   7#include <linux/bpf_lirc.h>
   8#include <linux/bpf_verifier.h>
   9#include <linux/bsearch.h>
  10#include <linux/btf.h>
  11#include <linux/syscalls.h>
  12#include <linux/slab.h>
  13#include <linux/sched/signal.h>
  14#include <linux/vmalloc.h>
  15#include <linux/mmzone.h>
  16#include <linux/anon_inodes.h>
  17#include <linux/fdtable.h>
  18#include <linux/file.h>
  19#include <linux/fs.h>
  20#include <linux/license.h>
  21#include <linux/filter.h>
  22#include <linux/kernel.h>
  23#include <linux/idr.h>
  24#include <linux/cred.h>
  25#include <linux/timekeeping.h>
  26#include <linux/ctype.h>
  27#include <linux/nospec.h>
  28#include <linux/audit.h>
  29#include <uapi/linux/btf.h>
  30#include <linux/pgtable.h>
  31#include <linux/bpf_lsm.h>
  32#include <linux/poll.h>
  33#include <linux/sort.h>
  34#include <linux/bpf-netns.h>
  35#include <linux/rcupdate_trace.h>
  36#include <linux/memcontrol.h>
  37#include <linux/trace_events.h>
  38
  39#include <net/netfilter/nf_bpf_link.h>
  40#include <net/netkit.h>
  41#include <net/tcx.h>
  42
  43#define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  44			  (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  45			  (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  46#define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  47#define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  48#define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  49			IS_FD_HASH(map))
  50
  51#define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  52
  53DEFINE_PER_CPU(int, bpf_prog_active);
  54static DEFINE_IDR(prog_idr);
  55static DEFINE_SPINLOCK(prog_idr_lock);
  56static DEFINE_IDR(map_idr);
  57static DEFINE_SPINLOCK(map_idr_lock);
  58static DEFINE_IDR(link_idr);
  59static DEFINE_SPINLOCK(link_idr_lock);
  60
  61int sysctl_unprivileged_bpf_disabled __read_mostly =
  62	IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  63
  64static const struct bpf_map_ops * const bpf_map_types[] = {
  65#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  66#define BPF_MAP_TYPE(_id, _ops) \
  67	[_id] = &_ops,
  68#define BPF_LINK_TYPE(_id, _name)
  69#include <linux/bpf_types.h>
  70#undef BPF_PROG_TYPE
  71#undef BPF_MAP_TYPE
  72#undef BPF_LINK_TYPE
  73};
  74
  75/*
  76 * If we're handed a bigger struct than we know of, ensure all the unknown bits
  77 * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  78 * we don't know about yet.
  79 *
  80 * There is a ToCToU between this function call and the following
  81 * copy_from_user() call. However, this is not a concern since this function is
  82 * meant to be a future-proofing of bits.
  83 */
  84int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  85			     size_t expected_size,
  86			     size_t actual_size)
  87{
  88	int res;
  89
  90	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
  91		return -E2BIG;
  92
  93	if (actual_size <= expected_size)
  94		return 0;
  95
  96	if (uaddr.is_kernel)
  97		res = memchr_inv(uaddr.kernel + expected_size, 0,
  98				 actual_size - expected_size) == NULL;
  99	else
 100		res = check_zeroed_user(uaddr.user + expected_size,
 101					actual_size - expected_size);
 102	if (res < 0)
 103		return res;
 104	return res ? 0 : -E2BIG;
 105}
 106
 107const struct bpf_map_ops bpf_map_offload_ops = {
 108	.map_meta_equal = bpf_map_meta_equal,
 109	.map_alloc = bpf_map_offload_map_alloc,
 110	.map_free = bpf_map_offload_map_free,
 111	.map_check_btf = map_check_no_btf,
 112	.map_mem_usage = bpf_map_offload_map_mem_usage,
 113};
 114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 115static void bpf_map_write_active_inc(struct bpf_map *map)
 116{
 117	atomic64_inc(&map->writecnt);
 118}
 119
 120static void bpf_map_write_active_dec(struct bpf_map *map)
 121{
 122	atomic64_dec(&map->writecnt);
 123}
 124
 125bool bpf_map_write_active(const struct bpf_map *map)
 126{
 127	return atomic64_read(&map->writecnt) != 0;
 128}
 129
 130static u32 bpf_map_value_size(const struct bpf_map *map)
 131{
 132	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 133	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 134	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 135	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 136		return round_up(map->value_size, 8) * num_possible_cpus();
 137	else if (IS_FD_MAP(map))
 138		return sizeof(u32);
 139	else
 140		return  map->value_size;
 141}
 142
 143static void maybe_wait_bpf_programs(struct bpf_map *map)
 144{
 145	/* Wait for any running non-sleepable BPF programs to complete so that
 146	 * userspace, when we return to it, knows that all non-sleepable
 147	 * programs that could be running use the new map value. For sleepable
 148	 * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
 149	 * for the completions of these programs, but considering the waiting
 150	 * time can be very long and userspace may think it will hang forever,
 151	 * so don't handle sleepable BPF programs now.
 152	 */
 153	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 154	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 155		synchronize_rcu();
 156}
 157
 158static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 159				void *key, void *value, __u64 flags)
 160{
 161	int err;
 162
 163	/* Need to create a kthread, thus must support schedule */
 164	if (bpf_map_is_offloaded(map)) {
 165		return bpf_map_offload_update_elem(map, key, value, flags);
 166	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 167		   map->map_type == BPF_MAP_TYPE_ARENA ||
 168		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 169		return map->ops->map_update_elem(map, key, value, flags);
 170	} else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 171		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 172		return sock_map_update_elem_sys(map, key, value, flags);
 173	} else if (IS_FD_PROG_ARRAY(map)) {
 174		return bpf_fd_array_map_update_elem(map, map_file, key, value,
 175						    flags);
 176	}
 177
 178	bpf_disable_instrumentation();
 179	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 180	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 181		err = bpf_percpu_hash_update(map, key, value, flags);
 182	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 183		err = bpf_percpu_array_update(map, key, value, flags);
 184	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 185		err = bpf_percpu_cgroup_storage_update(map, key, value,
 186						       flags);
 187	} else if (IS_FD_ARRAY(map)) {
 
 188		err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 189						   flags);
 
 190	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 
 191		err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 192						  flags);
 
 193	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 194		/* rcu_read_lock() is not needed */
 195		err = bpf_fd_reuseport_array_update_elem(map, key, value,
 196							 flags);
 197	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 198		   map->map_type == BPF_MAP_TYPE_STACK ||
 199		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 200		err = map->ops->map_push_elem(map, value, flags);
 201	} else {
 202		rcu_read_lock();
 203		err = map->ops->map_update_elem(map, key, value, flags);
 204		rcu_read_unlock();
 205	}
 206	bpf_enable_instrumentation();
 
 207
 208	return err;
 209}
 210
 211static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 212			      __u64 flags)
 213{
 214	void *ptr;
 215	int err;
 216
 217	if (bpf_map_is_offloaded(map))
 218		return bpf_map_offload_lookup_elem(map, key, value);
 219
 220	bpf_disable_instrumentation();
 221	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 222	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 223		err = bpf_percpu_hash_copy(map, key, value);
 224	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 225		err = bpf_percpu_array_copy(map, key, value);
 226	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 227		err = bpf_percpu_cgroup_storage_copy(map, key, value);
 228	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 229		err = bpf_stackmap_copy(map, key, value);
 230	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 231		err = bpf_fd_array_map_lookup_elem(map, key, value);
 232	} else if (IS_FD_HASH(map)) {
 233		err = bpf_fd_htab_map_lookup_elem(map, key, value);
 234	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 235		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 236	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 237		   map->map_type == BPF_MAP_TYPE_STACK ||
 238		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 239		err = map->ops->map_peek_elem(map, value);
 240	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 241		/* struct_ops map requires directly updating "value" */
 242		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 243	} else {
 244		rcu_read_lock();
 245		if (map->ops->map_lookup_elem_sys_only)
 246			ptr = map->ops->map_lookup_elem_sys_only(map, key);
 247		else
 248			ptr = map->ops->map_lookup_elem(map, key);
 249		if (IS_ERR(ptr)) {
 250			err = PTR_ERR(ptr);
 251		} else if (!ptr) {
 252			err = -ENOENT;
 253		} else {
 254			err = 0;
 255			if (flags & BPF_F_LOCK)
 256				/* lock 'ptr' and copy everything but lock */
 257				copy_map_value_locked(map, value, ptr, true);
 258			else
 259				copy_map_value(map, value, ptr);
 260			/* mask lock and timer, since value wasn't zero inited */
 261			check_and_init_map_value(map, value);
 262		}
 263		rcu_read_unlock();
 264	}
 265
 266	bpf_enable_instrumentation();
 
 267
 268	return err;
 269}
 270
 271/* Please, do not use this function outside from the map creation path
 272 * (e.g. in map update path) without taking care of setting the active
 273 * memory cgroup (see at bpf_map_kmalloc_node() for example).
 274 */
 275static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 276{
 277	/* We really just want to fail instead of triggering OOM killer
 278	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 279	 * which is used for lower order allocation requests.
 280	 *
 281	 * It has been observed that higher order allocation requests done by
 282	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
 283	 * to reclaim memory from the page cache, thus we set
 284	 * __GFP_RETRY_MAYFAIL to avoid such situations.
 285	 */
 286
 287	gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 288	unsigned int flags = 0;
 289	unsigned long align = 1;
 290	void *area;
 291
 292	if (size >= SIZE_MAX)
 293		return NULL;
 294
 295	/* kmalloc()'ed memory can't be mmap()'ed */
 296	if (mmapable) {
 297		BUG_ON(!PAGE_ALIGNED(size));
 298		align = SHMLBA;
 299		flags = VM_USERMAP;
 300	} else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 301		area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 302				    numa_node);
 303		if (area != NULL)
 304			return area;
 305	}
 306
 307	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 308			gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 309			flags, numa_node, __builtin_return_address(0));
 310}
 311
 312void *bpf_map_area_alloc(u64 size, int numa_node)
 313{
 314	return __bpf_map_area_alloc(size, numa_node, false);
 315}
 316
 317void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 318{
 319	return __bpf_map_area_alloc(size, numa_node, true);
 320}
 321
 322void bpf_map_area_free(void *area)
 323{
 324	kvfree(area);
 325}
 326
 327static u32 bpf_map_flags_retain_permanent(u32 flags)
 328{
 329	/* Some map creation flags are not tied to the map object but
 330	 * rather to the map fd instead, so they have no meaning upon
 331	 * map object inspection since multiple file descriptors with
 332	 * different (access) properties can exist here. Thus, given
 333	 * this has zero meaning for the map itself, lets clear these
 334	 * from here.
 335	 */
 336	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 337}
 338
 339void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 340{
 341	map->map_type = attr->map_type;
 342	map->key_size = attr->key_size;
 343	map->value_size = attr->value_size;
 344	map->max_entries = attr->max_entries;
 345	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 346	map->numa_node = bpf_map_attr_numa_node(attr);
 347	map->map_extra = attr->map_extra;
 348}
 349
 350static int bpf_map_alloc_id(struct bpf_map *map)
 351{
 352	int id;
 353
 354	idr_preload(GFP_KERNEL);
 355	spin_lock_bh(&map_idr_lock);
 356	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 357	if (id > 0)
 358		map->id = id;
 359	spin_unlock_bh(&map_idr_lock);
 360	idr_preload_end();
 361
 362	if (WARN_ON_ONCE(!id))
 363		return -ENOSPC;
 364
 365	return id > 0 ? 0 : id;
 366}
 367
 368void bpf_map_free_id(struct bpf_map *map)
 369{
 370	unsigned long flags;
 371
 372	/* Offloaded maps are removed from the IDR store when their device
 373	 * disappears - even if someone holds an fd to them they are unusable,
 374	 * the memory is gone, all ops will fail; they are simply waiting for
 375	 * refcnt to drop to be freed.
 376	 */
 377	if (!map->id)
 378		return;
 379
 380	spin_lock_irqsave(&map_idr_lock, flags);
 
 
 
 381
 382	idr_remove(&map_idr, map->id);
 383	map->id = 0;
 384
 385	spin_unlock_irqrestore(&map_idr_lock, flags);
 
 
 
 386}
 387
 388#ifdef CONFIG_MEMCG_KMEM
 389static void bpf_map_save_memcg(struct bpf_map *map)
 390{
 391	/* Currently if a map is created by a process belonging to the root
 392	 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 393	 * So we have to check map->objcg for being NULL each time it's
 394	 * being used.
 395	 */
 396	if (memcg_bpf_enabled())
 397		map->objcg = get_obj_cgroup_from_current();
 398}
 399
 400static void bpf_map_release_memcg(struct bpf_map *map)
 401{
 402	if (map->objcg)
 403		obj_cgroup_put(map->objcg);
 404}
 405
 406static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 407{
 408	if (map->objcg)
 409		return get_mem_cgroup_from_objcg(map->objcg);
 410
 411	return root_mem_cgroup;
 412}
 413
 414void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 415			   int node)
 416{
 417	struct mem_cgroup *memcg, *old_memcg;
 418	void *ptr;
 419
 420	memcg = bpf_map_get_memcg(map);
 421	old_memcg = set_active_memcg(memcg);
 422	ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 423	set_active_memcg(old_memcg);
 424	mem_cgroup_put(memcg);
 425
 426	return ptr;
 427}
 428
 429void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 430{
 431	struct mem_cgroup *memcg, *old_memcg;
 432	void *ptr;
 433
 434	memcg = bpf_map_get_memcg(map);
 435	old_memcg = set_active_memcg(memcg);
 436	ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 437	set_active_memcg(old_memcg);
 438	mem_cgroup_put(memcg);
 439
 440	return ptr;
 441}
 442
 443void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 444		       gfp_t flags)
 445{
 446	struct mem_cgroup *memcg, *old_memcg;
 447	void *ptr;
 448
 449	memcg = bpf_map_get_memcg(map);
 450	old_memcg = set_active_memcg(memcg);
 451	ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 452	set_active_memcg(old_memcg);
 453	mem_cgroup_put(memcg);
 454
 455	return ptr;
 456}
 457
 458void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 459				    size_t align, gfp_t flags)
 460{
 461	struct mem_cgroup *memcg, *old_memcg;
 462	void __percpu *ptr;
 463
 464	memcg = bpf_map_get_memcg(map);
 465	old_memcg = set_active_memcg(memcg);
 466	ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 467	set_active_memcg(old_memcg);
 468	mem_cgroup_put(memcg);
 469
 470	return ptr;
 471}
 472
 473#else
 474static void bpf_map_save_memcg(struct bpf_map *map)
 475{
 476}
 477
 478static void bpf_map_release_memcg(struct bpf_map *map)
 479{
 480}
 481#endif
 482
 483int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
 484			unsigned long nr_pages, struct page **pages)
 485{
 486	unsigned long i, j;
 487	struct page *pg;
 488	int ret = 0;
 489#ifdef CONFIG_MEMCG_KMEM
 490	struct mem_cgroup *memcg, *old_memcg;
 491
 492	memcg = bpf_map_get_memcg(map);
 493	old_memcg = set_active_memcg(memcg);
 494#endif
 495	for (i = 0; i < nr_pages; i++) {
 496		pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
 497
 498		if (pg) {
 499			pages[i] = pg;
 500			continue;
 501		}
 502		for (j = 0; j < i; j++)
 503			__free_page(pages[j]);
 504		ret = -ENOMEM;
 505		break;
 506	}
 507
 508#ifdef CONFIG_MEMCG_KMEM
 509	set_active_memcg(old_memcg);
 510	mem_cgroup_put(memcg);
 511#endif
 512	return ret;
 513}
 514
 515
 516static int btf_field_cmp(const void *a, const void *b)
 517{
 518	const struct btf_field *f1 = a, *f2 = b;
 519
 520	if (f1->offset < f2->offset)
 521		return -1;
 522	else if (f1->offset > f2->offset)
 523		return 1;
 524	return 0;
 525}
 526
 527struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 528				  u32 field_mask)
 529{
 530	struct btf_field *field;
 531
 532	if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 533		return NULL;
 534	field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 535	if (!field || !(field->type & field_mask))
 536		return NULL;
 537	return field;
 538}
 539
 540void btf_record_free(struct btf_record *rec)
 541{
 542	int i;
 543
 544	if (IS_ERR_OR_NULL(rec))
 545		return;
 546	for (i = 0; i < rec->cnt; i++) {
 547		switch (rec->fields[i].type) {
 
 
 
 548		case BPF_KPTR_UNREF:
 549		case BPF_KPTR_REF:
 550		case BPF_KPTR_PERCPU:
 551			if (rec->fields[i].kptr.module)
 552				module_put(rec->fields[i].kptr.module);
 553			btf_put(rec->fields[i].kptr.btf);
 554			break;
 555		case BPF_LIST_HEAD:
 556		case BPF_LIST_NODE:
 557		case BPF_RB_ROOT:
 558		case BPF_RB_NODE:
 559		case BPF_SPIN_LOCK:
 560		case BPF_TIMER:
 561		case BPF_REFCOUNT:
 562			/* Nothing to release */
 563			break;
 564		default:
 565			WARN_ON_ONCE(1);
 566			continue;
 567		}
 568	}
 569	kfree(rec);
 570}
 571
 572void bpf_map_free_record(struct bpf_map *map)
 573{
 574	btf_record_free(map->record);
 575	map->record = NULL;
 576}
 577
 578struct btf_record *btf_record_dup(const struct btf_record *rec)
 579{
 580	const struct btf_field *fields;
 581	struct btf_record *new_rec;
 582	int ret, size, i;
 583
 584	if (IS_ERR_OR_NULL(rec))
 585		return NULL;
 586	size = offsetof(struct btf_record, fields[rec->cnt]);
 587	new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 588	if (!new_rec)
 589		return ERR_PTR(-ENOMEM);
 590	/* Do a deep copy of the btf_record */
 591	fields = rec->fields;
 592	new_rec->cnt = 0;
 593	for (i = 0; i < rec->cnt; i++) {
 594		switch (fields[i].type) {
 
 
 
 595		case BPF_KPTR_UNREF:
 596		case BPF_KPTR_REF:
 597		case BPF_KPTR_PERCPU:
 598			btf_get(fields[i].kptr.btf);
 599			if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 600				ret = -ENXIO;
 601				goto free;
 602			}
 603			break;
 604		case BPF_LIST_HEAD:
 605		case BPF_LIST_NODE:
 606		case BPF_RB_ROOT:
 607		case BPF_RB_NODE:
 608		case BPF_SPIN_LOCK:
 609		case BPF_TIMER:
 610		case BPF_REFCOUNT:
 611			/* Nothing to acquire */
 612			break;
 613		default:
 614			ret = -EFAULT;
 615			WARN_ON_ONCE(1);
 616			goto free;
 617		}
 618		new_rec->cnt++;
 619	}
 620	return new_rec;
 621free:
 622	btf_record_free(new_rec);
 623	return ERR_PTR(ret);
 624}
 625
 626bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 627{
 628	bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 629	int size;
 630
 631	if (!a_has_fields && !b_has_fields)
 632		return true;
 633	if (a_has_fields != b_has_fields)
 634		return false;
 635	if (rec_a->cnt != rec_b->cnt)
 636		return false;
 637	size = offsetof(struct btf_record, fields[rec_a->cnt]);
 638	/* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 639	 * members are zeroed out. So memcmp is safe to do without worrying
 640	 * about padding/unused fields.
 641	 *
 642	 * While spin_lock, timer, and kptr have no relation to map BTF,
 643	 * list_head metadata is specific to map BTF, the btf and value_rec
 644	 * members in particular. btf is the map BTF, while value_rec points to
 645	 * btf_record in that map BTF.
 646	 *
 647	 * So while by default, we don't rely on the map BTF (which the records
 648	 * were parsed from) matching for both records, which is not backwards
 649	 * compatible, in case list_head is part of it, we implicitly rely on
 650	 * that by way of depending on memcmp succeeding for it.
 651	 */
 652	return !memcmp(rec_a, rec_b, size);
 653}
 654
 655void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 656{
 657	if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 658		return;
 659	bpf_timer_cancel_and_free(obj + rec->timer_off);
 660}
 661
 662void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 663{
 664	const struct btf_field *fields;
 665	int i;
 666
 667	if (IS_ERR_OR_NULL(rec))
 668		return;
 669	fields = rec->fields;
 670	for (i = 0; i < rec->cnt; i++) {
 671		struct btf_struct_meta *pointee_struct_meta;
 672		const struct btf_field *field = &fields[i];
 673		void *field_ptr = obj + field->offset;
 674		void *xchgd_field;
 675
 676		switch (fields[i].type) {
 677		case BPF_SPIN_LOCK:
 678			break;
 679		case BPF_TIMER:
 680			bpf_timer_cancel_and_free(field_ptr);
 681			break;
 682		case BPF_KPTR_UNREF:
 683			WRITE_ONCE(*(u64 *)field_ptr, 0);
 684			break;
 685		case BPF_KPTR_REF:
 686		case BPF_KPTR_PERCPU:
 687			xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 688			if (!xchgd_field)
 689				break;
 690
 691			if (!btf_is_kernel(field->kptr.btf)) {
 692				pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 693									   field->kptr.btf_id);
 694				migrate_disable();
 695				__bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 696								 pointee_struct_meta->record : NULL,
 697								 fields[i].type == BPF_KPTR_PERCPU);
 698				migrate_enable();
 699			} else {
 700				field->kptr.dtor(xchgd_field);
 701			}
 702			break;
 703		case BPF_LIST_HEAD:
 704			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 705				continue;
 706			bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 707			break;
 708		case BPF_RB_ROOT:
 709			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 710				continue;
 711			bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 712			break;
 713		case BPF_LIST_NODE:
 714		case BPF_RB_NODE:
 715		case BPF_REFCOUNT:
 716			break;
 717		default:
 718			WARN_ON_ONCE(1);
 719			continue;
 720		}
 721	}
 722}
 723
 724/* called from workqueue */
 725static void bpf_map_free_deferred(struct work_struct *work)
 726{
 727	struct bpf_map *map = container_of(work, struct bpf_map, work);
 
 728	struct btf_record *rec = map->record;
 729	struct btf *btf = map->btf;
 730
 731	security_bpf_map_free(map);
 732	bpf_map_release_memcg(map);
 733	/* implementation dependent freeing */
 734	map->ops->map_free(map);
 735	/* Delay freeing of btf_record for maps, as map_free
 736	 * callback usually needs access to them. It is better to do it here
 737	 * than require each callback to do the free itself manually.
 738	 *
 739	 * Note that the btf_record stashed in map->inner_map_meta->record was
 740	 * already freed using the map_free callback for map in map case which
 741	 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 742	 * template bpf_map struct used during verification.
 743	 */
 
 744	btf_record_free(rec);
 745	/* Delay freeing of btf for maps, as map_free callback may need
 746	 * struct_meta info which will be freed with btf_put().
 747	 */
 748	btf_put(btf);
 749}
 750
 751static void bpf_map_put_uref(struct bpf_map *map)
 752{
 753	if (atomic64_dec_and_test(&map->usercnt)) {
 754		if (map->ops->map_release_uref)
 755			map->ops->map_release_uref(map);
 756	}
 757}
 758
 759static void bpf_map_free_in_work(struct bpf_map *map)
 760{
 761	INIT_WORK(&map->work, bpf_map_free_deferred);
 762	/* Avoid spawning kworkers, since they all might contend
 763	 * for the same mutex like slab_mutex.
 764	 */
 765	queue_work(system_unbound_wq, &map->work);
 766}
 767
 768static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
 769{
 770	bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
 771}
 772
 773static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
 774{
 775	if (rcu_trace_implies_rcu_gp())
 776		bpf_map_free_rcu_gp(rcu);
 777	else
 778		call_rcu(rcu, bpf_map_free_rcu_gp);
 779}
 780
 781/* decrement map refcnt and schedule it for freeing via workqueue
 782 * (underlying map implementation ops->map_free() might sleep)
 783 */
 784void bpf_map_put(struct bpf_map *map)
 785{
 786	if (atomic64_dec_and_test(&map->refcnt)) {
 787		/* bpf_map_free_id() must be called first */
 788		bpf_map_free_id(map);
 
 
 
 
 
 
 
 
 789
 790		WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
 791		if (READ_ONCE(map->free_after_mult_rcu_gp))
 792			call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
 793		else if (READ_ONCE(map->free_after_rcu_gp))
 794			call_rcu(&map->rcu, bpf_map_free_rcu_gp);
 795		else
 796			bpf_map_free_in_work(map);
 797	}
 798}
 799EXPORT_SYMBOL_GPL(bpf_map_put);
 800
 801void bpf_map_put_with_uref(struct bpf_map *map)
 802{
 803	bpf_map_put_uref(map);
 804	bpf_map_put(map);
 805}
 806
 807static int bpf_map_release(struct inode *inode, struct file *filp)
 808{
 809	struct bpf_map *map = filp->private_data;
 810
 811	if (map->ops->map_release)
 812		map->ops->map_release(map, filp);
 813
 814	bpf_map_put_with_uref(map);
 815	return 0;
 816}
 817
 818static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 819{
 820	fmode_t mode = f.file->f_mode;
 821
 822	/* Our file permissions may have been overridden by global
 823	 * map permissions facing syscall side.
 824	 */
 825	if (READ_ONCE(map->frozen))
 826		mode &= ~FMODE_CAN_WRITE;
 827	return mode;
 828}
 829
 830#ifdef CONFIG_PROC_FS
 831/* Show the memory usage of a bpf map */
 832static u64 bpf_map_memory_usage(const struct bpf_map *map)
 
 
 
 833{
 834	return map->ops->map_mem_usage(map);
 
 
 
 
 835}
 836
 837static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 838{
 839	struct bpf_map *map = filp->private_data;
 840	u32 type = 0, jited = 0;
 841
 842	if (map_type_contains_progs(map)) {
 843		spin_lock(&map->owner.lock);
 844		type  = map->owner.type;
 845		jited = map->owner.jited;
 846		spin_unlock(&map->owner.lock);
 847	}
 848
 849	seq_printf(m,
 850		   "map_type:\t%u\n"
 851		   "key_size:\t%u\n"
 852		   "value_size:\t%u\n"
 853		   "max_entries:\t%u\n"
 854		   "map_flags:\t%#x\n"
 855		   "map_extra:\t%#llx\n"
 856		   "memlock:\t%llu\n"
 857		   "map_id:\t%u\n"
 858		   "frozen:\t%u\n",
 859		   map->map_type,
 860		   map->key_size,
 861		   map->value_size,
 862		   map->max_entries,
 863		   map->map_flags,
 864		   (unsigned long long)map->map_extra,
 865		   bpf_map_memory_usage(map),
 866		   map->id,
 867		   READ_ONCE(map->frozen));
 868	if (type) {
 869		seq_printf(m, "owner_prog_type:\t%u\n", type);
 870		seq_printf(m, "owner_jited:\t%u\n", jited);
 871	}
 872}
 873#endif
 874
 875static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 876			      loff_t *ppos)
 877{
 878	/* We need this handler such that alloc_file() enables
 879	 * f_mode with FMODE_CAN_READ.
 880	 */
 881	return -EINVAL;
 882}
 883
 884static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 885			       size_t siz, loff_t *ppos)
 886{
 887	/* We need this handler such that alloc_file() enables
 888	 * f_mode with FMODE_CAN_WRITE.
 889	 */
 890	return -EINVAL;
 891}
 892
 893/* called for any extra memory-mapped regions (except initial) */
 894static void bpf_map_mmap_open(struct vm_area_struct *vma)
 895{
 896	struct bpf_map *map = vma->vm_file->private_data;
 897
 898	if (vma->vm_flags & VM_MAYWRITE)
 899		bpf_map_write_active_inc(map);
 900}
 901
 902/* called for all unmapped memory region (including initial) */
 903static void bpf_map_mmap_close(struct vm_area_struct *vma)
 904{
 905	struct bpf_map *map = vma->vm_file->private_data;
 906
 907	if (vma->vm_flags & VM_MAYWRITE)
 908		bpf_map_write_active_dec(map);
 909}
 910
 911static const struct vm_operations_struct bpf_map_default_vmops = {
 912	.open		= bpf_map_mmap_open,
 913	.close		= bpf_map_mmap_close,
 914};
 915
 916static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 917{
 918	struct bpf_map *map = filp->private_data;
 919	int err;
 920
 921	if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 922		return -ENOTSUPP;
 923
 924	if (!(vma->vm_flags & VM_SHARED))
 925		return -EINVAL;
 926
 927	mutex_lock(&map->freeze_mutex);
 928
 929	if (vma->vm_flags & VM_WRITE) {
 930		if (map->frozen) {
 931			err = -EPERM;
 932			goto out;
 933		}
 934		/* map is meant to be read-only, so do not allow mapping as
 935		 * writable, because it's possible to leak a writable page
 936		 * reference and allows user-space to still modify it after
 937		 * freezing, while verifier will assume contents do not change
 938		 */
 939		if (map->map_flags & BPF_F_RDONLY_PROG) {
 940			err = -EACCES;
 941			goto out;
 942		}
 943	}
 944
 945	/* set default open/close callbacks */
 946	vma->vm_ops = &bpf_map_default_vmops;
 947	vma->vm_private_data = map;
 948	vm_flags_clear(vma, VM_MAYEXEC);
 949	if (!(vma->vm_flags & VM_WRITE))
 950		/* disallow re-mapping with PROT_WRITE */
 951		vm_flags_clear(vma, VM_MAYWRITE);
 952
 953	err = map->ops->map_mmap(map, vma);
 954	if (err)
 955		goto out;
 956
 957	if (vma->vm_flags & VM_MAYWRITE)
 958		bpf_map_write_active_inc(map);
 959out:
 960	mutex_unlock(&map->freeze_mutex);
 961	return err;
 962}
 963
 964static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 965{
 966	struct bpf_map *map = filp->private_data;
 967
 968	if (map->ops->map_poll)
 969		return map->ops->map_poll(map, filp, pts);
 970
 971	return EPOLLERR;
 972}
 973
 974static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
 975					   unsigned long len, unsigned long pgoff,
 976					   unsigned long flags)
 977{
 978	struct bpf_map *map = filp->private_data;
 979
 980	if (map->ops->map_get_unmapped_area)
 981		return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
 982#ifdef CONFIG_MMU
 983	return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
 984#else
 985	return addr;
 986#endif
 987}
 988
 989const struct file_operations bpf_map_fops = {
 990#ifdef CONFIG_PROC_FS
 991	.show_fdinfo	= bpf_map_show_fdinfo,
 992#endif
 993	.release	= bpf_map_release,
 994	.read		= bpf_dummy_read,
 995	.write		= bpf_dummy_write,
 996	.mmap		= bpf_map_mmap,
 997	.poll		= bpf_map_poll,
 998	.get_unmapped_area = bpf_get_unmapped_area,
 999};
1000
1001int bpf_map_new_fd(struct bpf_map *map, int flags)
1002{
1003	int ret;
1004
1005	ret = security_bpf_map(map, OPEN_FMODE(flags));
1006	if (ret < 0)
1007		return ret;
1008
1009	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1010				flags | O_CLOEXEC);
1011}
1012
1013int bpf_get_file_flag(int flags)
1014{
1015	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1016		return -EINVAL;
1017	if (flags & BPF_F_RDONLY)
1018		return O_RDONLY;
1019	if (flags & BPF_F_WRONLY)
1020		return O_WRONLY;
1021	return O_RDWR;
1022}
1023
1024/* helper macro to check that unused fields 'union bpf_attr' are zero */
1025#define CHECK_ATTR(CMD) \
1026	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1027		   sizeof(attr->CMD##_LAST_FIELD), 0, \
1028		   sizeof(*attr) - \
1029		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1030		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
1031
1032/* dst and src must have at least "size" number of bytes.
1033 * Return strlen on success and < 0 on error.
1034 */
1035int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1036{
1037	const char *end = src + size;
1038	const char *orig_src = src;
1039
1040	memset(dst, 0, size);
1041	/* Copy all isalnum(), '_' and '.' chars. */
1042	while (src < end && *src) {
1043		if (!isalnum(*src) &&
1044		    *src != '_' && *src != '.')
1045			return -EINVAL;
1046		*dst++ = *src++;
1047	}
1048
1049	/* No '\0' found in "size" number of bytes */
1050	if (src == end)
1051		return -EINVAL;
1052
1053	return src - orig_src;
1054}
1055
1056int map_check_no_btf(const struct bpf_map *map,
1057		     const struct btf *btf,
1058		     const struct btf_type *key_type,
1059		     const struct btf_type *value_type)
1060{
1061	return -ENOTSUPP;
1062}
1063
1064static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1065			 const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1066{
1067	const struct btf_type *key_type, *value_type;
1068	u32 key_size, value_size;
1069	int ret = 0;
1070
1071	/* Some maps allow key to be unspecified. */
1072	if (btf_key_id) {
1073		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1074		if (!key_type || key_size != map->key_size)
1075			return -EINVAL;
1076	} else {
1077		key_type = btf_type_by_id(btf, 0);
1078		if (!map->ops->map_check_btf)
1079			return -EINVAL;
1080	}
1081
1082	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1083	if (!value_type || value_size != map->value_size)
1084		return -EINVAL;
1085
1086	map->record = btf_parse_fields(btf, value_type,
1087				       BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1088				       BPF_RB_ROOT | BPF_REFCOUNT,
1089				       map->value_size);
1090	if (!IS_ERR_OR_NULL(map->record)) {
1091		int i;
1092
1093		if (!bpf_token_capable(token, CAP_BPF)) {
1094			ret = -EPERM;
1095			goto free_map_tab;
1096		}
1097		if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1098			ret = -EACCES;
1099			goto free_map_tab;
1100		}
1101		for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1102			switch (map->record->field_mask & (1 << i)) {
1103			case 0:
1104				continue;
1105			case BPF_SPIN_LOCK:
1106				if (map->map_type != BPF_MAP_TYPE_HASH &&
1107				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1108				    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1109				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1110				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1111				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1112				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1113					ret = -EOPNOTSUPP;
1114					goto free_map_tab;
1115				}
1116				break;
1117			case BPF_TIMER:
1118				if (map->map_type != BPF_MAP_TYPE_HASH &&
1119				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1120				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1121					ret = -EOPNOTSUPP;
1122					goto free_map_tab;
1123				}
1124				break;
1125			case BPF_KPTR_UNREF:
1126			case BPF_KPTR_REF:
1127			case BPF_KPTR_PERCPU:
1128			case BPF_REFCOUNT:
1129				if (map->map_type != BPF_MAP_TYPE_HASH &&
1130				    map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1131				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1132				    map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1133				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1134				    map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1135				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1136				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1137				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1138				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1139					ret = -EOPNOTSUPP;
1140					goto free_map_tab;
1141				}
1142				break;
1143			case BPF_LIST_HEAD:
1144			case BPF_RB_ROOT:
1145				if (map->map_type != BPF_MAP_TYPE_HASH &&
1146				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1147				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1148					ret = -EOPNOTSUPP;
1149					goto free_map_tab;
1150				}
1151				break;
1152			default:
1153				/* Fail if map_type checks are missing for a field type */
1154				ret = -EOPNOTSUPP;
1155				goto free_map_tab;
1156			}
1157		}
1158	}
1159
1160	ret = btf_check_and_fixup_fields(btf, map->record);
1161	if (ret < 0)
1162		goto free_map_tab;
1163
1164	if (map->ops->map_check_btf) {
1165		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1166		if (ret < 0)
1167			goto free_map_tab;
1168	}
1169
1170	return ret;
1171free_map_tab:
1172	bpf_map_free_record(map);
1173	return ret;
1174}
1175
1176static bool bpf_net_capable(void)
1177{
1178	return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1179}
1180
1181#define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1182/* called via syscall */
1183static int map_create(union bpf_attr *attr)
1184{
1185	const struct bpf_map_ops *ops;
1186	struct bpf_token *token = NULL;
1187	int numa_node = bpf_map_attr_numa_node(attr);
1188	u32 map_type = attr->map_type;
1189	struct bpf_map *map;
1190	bool token_flag;
1191	int f_flags;
1192	int err;
1193
1194	err = CHECK_ATTR(BPF_MAP_CREATE);
1195	if (err)
1196		return -EINVAL;
1197
1198	/* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1199	 * to avoid per-map type checks tripping on unknown flag
1200	 */
1201	token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1202	attr->map_flags &= ~BPF_F_TOKEN_FD;
1203
1204	if (attr->btf_vmlinux_value_type_id) {
1205		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1206		    attr->btf_key_type_id || attr->btf_value_type_id)
1207			return -EINVAL;
1208	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1209		return -EINVAL;
1210	}
1211
1212	if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1213	    attr->map_type != BPF_MAP_TYPE_ARENA &&
1214	    attr->map_extra != 0)
1215		return -EINVAL;
1216
1217	f_flags = bpf_get_file_flag(attr->map_flags);
1218	if (f_flags < 0)
1219		return f_flags;
1220
1221	if (numa_node != NUMA_NO_NODE &&
1222	    ((unsigned int)numa_node >= nr_node_ids ||
1223	     !node_online(numa_node)))
1224		return -EINVAL;
1225
1226	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1227	map_type = attr->map_type;
1228	if (map_type >= ARRAY_SIZE(bpf_map_types))
1229		return -EINVAL;
1230	map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1231	ops = bpf_map_types[map_type];
1232	if (!ops)
1233		return -EINVAL;
1234
1235	if (ops->map_alloc_check) {
1236		err = ops->map_alloc_check(attr);
1237		if (err)
1238			return err;
1239	}
1240	if (attr->map_ifindex)
1241		ops = &bpf_map_offload_ops;
1242	if (!ops->map_mem_usage)
1243		return -EINVAL;
1244
1245	if (token_flag) {
1246		token = bpf_token_get_from_fd(attr->map_token_fd);
1247		if (IS_ERR(token))
1248			return PTR_ERR(token);
1249
1250		/* if current token doesn't grant map creation permissions,
1251		 * then we can't use this token, so ignore it and rely on
1252		 * system-wide capabilities checks
1253		 */
1254		if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1255		    !bpf_token_allow_map_type(token, attr->map_type)) {
1256			bpf_token_put(token);
1257			token = NULL;
1258		}
1259	}
1260
1261	err = -EPERM;
1262
1263	/* Intent here is for unprivileged_bpf_disabled to block BPF map
1264	 * creation for unprivileged users; other actions depend
1265	 * on fd availability and access to bpffs, so are dependent on
1266	 * object creation success. Even with unprivileged BPF disabled,
1267	 * capability checks are still carried out.
1268	 */
1269	if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1270		goto put_token;
1271
1272	/* check privileged map type permissions */
1273	switch (map_type) {
1274	case BPF_MAP_TYPE_ARRAY:
1275	case BPF_MAP_TYPE_PERCPU_ARRAY:
1276	case BPF_MAP_TYPE_PROG_ARRAY:
1277	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1278	case BPF_MAP_TYPE_CGROUP_ARRAY:
1279	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1280	case BPF_MAP_TYPE_HASH:
1281	case BPF_MAP_TYPE_PERCPU_HASH:
1282	case BPF_MAP_TYPE_HASH_OF_MAPS:
1283	case BPF_MAP_TYPE_RINGBUF:
1284	case BPF_MAP_TYPE_USER_RINGBUF:
1285	case BPF_MAP_TYPE_CGROUP_STORAGE:
1286	case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1287		/* unprivileged */
1288		break;
1289	case BPF_MAP_TYPE_SK_STORAGE:
1290	case BPF_MAP_TYPE_INODE_STORAGE:
1291	case BPF_MAP_TYPE_TASK_STORAGE:
1292	case BPF_MAP_TYPE_CGRP_STORAGE:
1293	case BPF_MAP_TYPE_BLOOM_FILTER:
1294	case BPF_MAP_TYPE_LPM_TRIE:
1295	case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1296	case BPF_MAP_TYPE_STACK_TRACE:
1297	case BPF_MAP_TYPE_QUEUE:
1298	case BPF_MAP_TYPE_STACK:
1299	case BPF_MAP_TYPE_LRU_HASH:
1300	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1301	case BPF_MAP_TYPE_STRUCT_OPS:
1302	case BPF_MAP_TYPE_CPUMAP:
1303	case BPF_MAP_TYPE_ARENA:
1304		if (!bpf_token_capable(token, CAP_BPF))
1305			goto put_token;
1306		break;
1307	case BPF_MAP_TYPE_SOCKMAP:
1308	case BPF_MAP_TYPE_SOCKHASH:
1309	case BPF_MAP_TYPE_DEVMAP:
1310	case BPF_MAP_TYPE_DEVMAP_HASH:
1311	case BPF_MAP_TYPE_XSKMAP:
1312		if (!bpf_token_capable(token, CAP_NET_ADMIN))
1313			goto put_token;
1314		break;
1315	default:
1316		WARN(1, "unsupported map type %d", map_type);
1317		goto put_token;
1318	}
1319
1320	map = ops->map_alloc(attr);
1321	if (IS_ERR(map)) {
1322		err = PTR_ERR(map);
1323		goto put_token;
1324	}
1325	map->ops = ops;
1326	map->map_type = map_type;
1327
1328	err = bpf_obj_name_cpy(map->name, attr->map_name,
1329			       sizeof(attr->map_name));
1330	if (err < 0)
1331		goto free_map;
1332
1333	atomic64_set(&map->refcnt, 1);
1334	atomic64_set(&map->usercnt, 1);
1335	mutex_init(&map->freeze_mutex);
1336	spin_lock_init(&map->owner.lock);
1337
1338	if (attr->btf_key_type_id || attr->btf_value_type_id ||
1339	    /* Even the map's value is a kernel's struct,
1340	     * the bpf_prog.o must have BTF to begin with
1341	     * to figure out the corresponding kernel's
1342	     * counter part.  Thus, attr->btf_fd has
1343	     * to be valid also.
1344	     */
1345	    attr->btf_vmlinux_value_type_id) {
1346		struct btf *btf;
1347
1348		btf = btf_get_by_fd(attr->btf_fd);
1349		if (IS_ERR(btf)) {
1350			err = PTR_ERR(btf);
1351			goto free_map;
1352		}
1353		if (btf_is_kernel(btf)) {
1354			btf_put(btf);
1355			err = -EACCES;
1356			goto free_map;
1357		}
1358		map->btf = btf;
1359
1360		if (attr->btf_value_type_id) {
1361			err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1362					    attr->btf_value_type_id);
1363			if (err)
1364				goto free_map;
1365		}
1366
1367		map->btf_key_type_id = attr->btf_key_type_id;
1368		map->btf_value_type_id = attr->btf_value_type_id;
1369		map->btf_vmlinux_value_type_id =
1370			attr->btf_vmlinux_value_type_id;
1371	}
1372
1373	err = security_bpf_map_create(map, attr, token);
 
 
 
 
 
 
 
 
1374	if (err)
1375		goto free_map_sec;
1376
1377	err = bpf_map_alloc_id(map);
1378	if (err)
1379		goto free_map_sec;
1380
1381	bpf_map_save_memcg(map);
1382	bpf_token_put(token);
1383
1384	err = bpf_map_new_fd(map, f_flags);
1385	if (err < 0) {
1386		/* failed to allocate fd.
1387		 * bpf_map_put_with_uref() is needed because the above
1388		 * bpf_map_alloc_id() has published the map
1389		 * to the userspace and the userspace may
1390		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1391		 */
1392		bpf_map_put_with_uref(map);
1393		return err;
1394	}
1395
1396	return err;
1397
1398free_map_sec:
1399	security_bpf_map_free(map);
 
 
1400free_map:
1401	btf_put(map->btf);
1402	map->ops->map_free(map);
1403put_token:
1404	bpf_token_put(token);
1405	return err;
1406}
1407
1408/* if error is returned, fd is released.
1409 * On success caller should complete fd access with matching fdput()
1410 */
1411struct bpf_map *__bpf_map_get(struct fd f)
1412{
1413	if (!f.file)
1414		return ERR_PTR(-EBADF);
1415	if (f.file->f_op != &bpf_map_fops) {
1416		fdput(f);
1417		return ERR_PTR(-EINVAL);
1418	}
1419
1420	return f.file->private_data;
1421}
1422
1423void bpf_map_inc(struct bpf_map *map)
1424{
1425	atomic64_inc(&map->refcnt);
1426}
1427EXPORT_SYMBOL_GPL(bpf_map_inc);
1428
1429void bpf_map_inc_with_uref(struct bpf_map *map)
1430{
1431	atomic64_inc(&map->refcnt);
1432	atomic64_inc(&map->usercnt);
1433}
1434EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1435
1436struct bpf_map *bpf_map_get(u32 ufd)
1437{
1438	struct fd f = fdget(ufd);
1439	struct bpf_map *map;
1440
1441	map = __bpf_map_get(f);
1442	if (IS_ERR(map))
1443		return map;
1444
1445	bpf_map_inc(map);
1446	fdput(f);
1447
1448	return map;
1449}
1450EXPORT_SYMBOL(bpf_map_get);
1451
1452struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1453{
1454	struct fd f = fdget(ufd);
1455	struct bpf_map *map;
1456
1457	map = __bpf_map_get(f);
1458	if (IS_ERR(map))
1459		return map;
1460
1461	bpf_map_inc_with_uref(map);
1462	fdput(f);
1463
1464	return map;
1465}
1466
1467/* map_idr_lock should have been held or the map should have been
1468 * protected by rcu read lock.
1469 */
1470struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1471{
1472	int refold;
1473
1474	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1475	if (!refold)
1476		return ERR_PTR(-ENOENT);
1477	if (uref)
1478		atomic64_inc(&map->usercnt);
1479
1480	return map;
1481}
1482
1483struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1484{
1485	spin_lock_bh(&map_idr_lock);
1486	map = __bpf_map_inc_not_zero(map, false);
1487	spin_unlock_bh(&map_idr_lock);
1488
1489	return map;
1490}
1491EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1492
1493int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1494{
1495	return -ENOTSUPP;
1496}
1497
1498static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1499{
1500	if (key_size)
1501		return vmemdup_user(ukey, key_size);
1502
1503	if (ukey)
1504		return ERR_PTR(-EINVAL);
1505
1506	return NULL;
1507}
1508
1509static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1510{
1511	if (key_size)
1512		return kvmemdup_bpfptr(ukey, key_size);
1513
1514	if (!bpfptr_is_null(ukey))
1515		return ERR_PTR(-EINVAL);
1516
1517	return NULL;
1518}
1519
1520/* last field in 'union bpf_attr' used by this command */
1521#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1522
1523static int map_lookup_elem(union bpf_attr *attr)
1524{
1525	void __user *ukey = u64_to_user_ptr(attr->key);
1526	void __user *uvalue = u64_to_user_ptr(attr->value);
1527	int ufd = attr->map_fd;
1528	struct bpf_map *map;
1529	void *key, *value;
1530	u32 value_size;
1531	struct fd f;
1532	int err;
1533
1534	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1535		return -EINVAL;
1536
1537	if (attr->flags & ~BPF_F_LOCK)
1538		return -EINVAL;
1539
1540	f = fdget(ufd);
1541	map = __bpf_map_get(f);
1542	if (IS_ERR(map))
1543		return PTR_ERR(map);
1544	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1545		err = -EPERM;
1546		goto err_put;
1547	}
1548
1549	if ((attr->flags & BPF_F_LOCK) &&
1550	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1551		err = -EINVAL;
1552		goto err_put;
1553	}
1554
1555	key = __bpf_copy_key(ukey, map->key_size);
1556	if (IS_ERR(key)) {
1557		err = PTR_ERR(key);
1558		goto err_put;
1559	}
1560
1561	value_size = bpf_map_value_size(map);
1562
1563	err = -ENOMEM;
1564	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1565	if (!value)
1566		goto free_key;
1567
1568	if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1569		if (copy_from_user(value, uvalue, value_size))
1570			err = -EFAULT;
1571		else
1572			err = bpf_map_copy_value(map, key, value, attr->flags);
1573		goto free_value;
1574	}
1575
1576	err = bpf_map_copy_value(map, key, value, attr->flags);
1577	if (err)
1578		goto free_value;
1579
1580	err = -EFAULT;
1581	if (copy_to_user(uvalue, value, value_size) != 0)
1582		goto free_value;
1583
1584	err = 0;
1585
1586free_value:
1587	kvfree(value);
1588free_key:
1589	kvfree(key);
1590err_put:
1591	fdput(f);
1592	return err;
1593}
1594
1595
1596#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1597
1598static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1599{
1600	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1601	bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1602	int ufd = attr->map_fd;
1603	struct bpf_map *map;
1604	void *key, *value;
1605	u32 value_size;
1606	struct fd f;
1607	int err;
1608
1609	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1610		return -EINVAL;
1611
1612	f = fdget(ufd);
1613	map = __bpf_map_get(f);
1614	if (IS_ERR(map))
1615		return PTR_ERR(map);
1616	bpf_map_write_active_inc(map);
1617	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1618		err = -EPERM;
1619		goto err_put;
1620	}
1621
1622	if ((attr->flags & BPF_F_LOCK) &&
1623	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1624		err = -EINVAL;
1625		goto err_put;
1626	}
1627
1628	key = ___bpf_copy_key(ukey, map->key_size);
1629	if (IS_ERR(key)) {
1630		err = PTR_ERR(key);
1631		goto err_put;
1632	}
1633
1634	value_size = bpf_map_value_size(map);
1635	value = kvmemdup_bpfptr(uvalue, value_size);
1636	if (IS_ERR(value)) {
1637		err = PTR_ERR(value);
1638		goto free_key;
1639	}
1640
1641	err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1642	if (!err)
1643		maybe_wait_bpf_programs(map);
1644
1645	kvfree(value);
1646free_key:
1647	kvfree(key);
1648err_put:
1649	bpf_map_write_active_dec(map);
1650	fdput(f);
1651	return err;
1652}
1653
1654#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1655
1656static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1657{
1658	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1659	int ufd = attr->map_fd;
1660	struct bpf_map *map;
1661	struct fd f;
1662	void *key;
1663	int err;
1664
1665	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1666		return -EINVAL;
1667
1668	f = fdget(ufd);
1669	map = __bpf_map_get(f);
1670	if (IS_ERR(map))
1671		return PTR_ERR(map);
1672	bpf_map_write_active_inc(map);
1673	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1674		err = -EPERM;
1675		goto err_put;
1676	}
1677
1678	key = ___bpf_copy_key(ukey, map->key_size);
1679	if (IS_ERR(key)) {
1680		err = PTR_ERR(key);
1681		goto err_put;
1682	}
1683
1684	if (bpf_map_is_offloaded(map)) {
1685		err = bpf_map_offload_delete_elem(map, key);
1686		goto out;
1687	} else if (IS_FD_PROG_ARRAY(map) ||
1688		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1689		/* These maps require sleepable context */
1690		err = map->ops->map_delete_elem(map, key);
1691		goto out;
1692	}
1693
1694	bpf_disable_instrumentation();
1695	rcu_read_lock();
1696	err = map->ops->map_delete_elem(map, key);
1697	rcu_read_unlock();
1698	bpf_enable_instrumentation();
1699	if (!err)
1700		maybe_wait_bpf_programs(map);
1701out:
1702	kvfree(key);
1703err_put:
1704	bpf_map_write_active_dec(map);
1705	fdput(f);
1706	return err;
1707}
1708
1709/* last field in 'union bpf_attr' used by this command */
1710#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1711
1712static int map_get_next_key(union bpf_attr *attr)
1713{
1714	void __user *ukey = u64_to_user_ptr(attr->key);
1715	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1716	int ufd = attr->map_fd;
1717	struct bpf_map *map;
1718	void *key, *next_key;
1719	struct fd f;
1720	int err;
1721
1722	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1723		return -EINVAL;
1724
1725	f = fdget(ufd);
1726	map = __bpf_map_get(f);
1727	if (IS_ERR(map))
1728		return PTR_ERR(map);
1729	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1730		err = -EPERM;
1731		goto err_put;
1732	}
1733
1734	if (ukey) {
1735		key = __bpf_copy_key(ukey, map->key_size);
1736		if (IS_ERR(key)) {
1737			err = PTR_ERR(key);
1738			goto err_put;
1739		}
1740	} else {
1741		key = NULL;
1742	}
1743
1744	err = -ENOMEM;
1745	next_key = kvmalloc(map->key_size, GFP_USER);
1746	if (!next_key)
1747		goto free_key;
1748
1749	if (bpf_map_is_offloaded(map)) {
1750		err = bpf_map_offload_get_next_key(map, key, next_key);
1751		goto out;
1752	}
1753
1754	rcu_read_lock();
1755	err = map->ops->map_get_next_key(map, key, next_key);
1756	rcu_read_unlock();
1757out:
1758	if (err)
1759		goto free_next_key;
1760
1761	err = -EFAULT;
1762	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1763		goto free_next_key;
1764
1765	err = 0;
1766
1767free_next_key:
1768	kvfree(next_key);
1769free_key:
1770	kvfree(key);
1771err_put:
1772	fdput(f);
1773	return err;
1774}
1775
1776int generic_map_delete_batch(struct bpf_map *map,
1777			     const union bpf_attr *attr,
1778			     union bpf_attr __user *uattr)
1779{
1780	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1781	u32 cp, max_count;
1782	int err = 0;
1783	void *key;
1784
1785	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1786		return -EINVAL;
1787
1788	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1789	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1790		return -EINVAL;
1791	}
1792
1793	max_count = attr->batch.count;
1794	if (!max_count)
1795		return 0;
1796
1797	if (put_user(0, &uattr->batch.count))
1798		return -EFAULT;
1799
1800	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1801	if (!key)
1802		return -ENOMEM;
1803
1804	for (cp = 0; cp < max_count; cp++) {
1805		err = -EFAULT;
1806		if (copy_from_user(key, keys + cp * map->key_size,
1807				   map->key_size))
1808			break;
1809
1810		if (bpf_map_is_offloaded(map)) {
1811			err = bpf_map_offload_delete_elem(map, key);
1812			break;
1813		}
1814
1815		bpf_disable_instrumentation();
1816		rcu_read_lock();
1817		err = map->ops->map_delete_elem(map, key);
1818		rcu_read_unlock();
1819		bpf_enable_instrumentation();
1820		if (err)
1821			break;
1822		cond_resched();
1823	}
1824	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1825		err = -EFAULT;
1826
1827	kvfree(key);
1828
 
1829	return err;
1830}
1831
1832int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1833			     const union bpf_attr *attr,
1834			     union bpf_attr __user *uattr)
1835{
1836	void __user *values = u64_to_user_ptr(attr->batch.values);
1837	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1838	u32 value_size, cp, max_count;
1839	void *key, *value;
1840	int err = 0;
1841
1842	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1843		return -EINVAL;
1844
1845	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1846	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1847		return -EINVAL;
1848	}
1849
1850	value_size = bpf_map_value_size(map);
1851
1852	max_count = attr->batch.count;
1853	if (!max_count)
1854		return 0;
1855
1856	if (put_user(0, &uattr->batch.count))
1857		return -EFAULT;
1858
1859	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1860	if (!key)
1861		return -ENOMEM;
1862
1863	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1864	if (!value) {
1865		kvfree(key);
1866		return -ENOMEM;
1867	}
1868
1869	for (cp = 0; cp < max_count; cp++) {
1870		err = -EFAULT;
1871		if (copy_from_user(key, keys + cp * map->key_size,
1872		    map->key_size) ||
1873		    copy_from_user(value, values + cp * value_size, value_size))
1874			break;
1875
1876		err = bpf_map_update_value(map, map_file, key, value,
1877					   attr->batch.elem_flags);
1878
1879		if (err)
1880			break;
1881		cond_resched();
1882	}
1883
1884	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1885		err = -EFAULT;
1886
1887	kvfree(value);
1888	kvfree(key);
1889
1890	return err;
1891}
1892
1893#define MAP_LOOKUP_RETRIES 3
1894
1895int generic_map_lookup_batch(struct bpf_map *map,
1896				    const union bpf_attr *attr,
1897				    union bpf_attr __user *uattr)
1898{
1899	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1900	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1901	void __user *values = u64_to_user_ptr(attr->batch.values);
1902	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1903	void *buf, *buf_prevkey, *prev_key, *key, *value;
1904	int err, retry = MAP_LOOKUP_RETRIES;
1905	u32 value_size, cp, max_count;
1906
1907	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1908		return -EINVAL;
1909
1910	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1911	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1912		return -EINVAL;
1913
1914	value_size = bpf_map_value_size(map);
1915
1916	max_count = attr->batch.count;
1917	if (!max_count)
1918		return 0;
1919
1920	if (put_user(0, &uattr->batch.count))
1921		return -EFAULT;
1922
1923	buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1924	if (!buf_prevkey)
1925		return -ENOMEM;
1926
1927	buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1928	if (!buf) {
1929		kvfree(buf_prevkey);
1930		return -ENOMEM;
1931	}
1932
1933	err = -EFAULT;
1934	prev_key = NULL;
1935	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1936		goto free_buf;
1937	key = buf;
1938	value = key + map->key_size;
1939	if (ubatch)
1940		prev_key = buf_prevkey;
1941
1942	for (cp = 0; cp < max_count;) {
1943		rcu_read_lock();
1944		err = map->ops->map_get_next_key(map, prev_key, key);
1945		rcu_read_unlock();
1946		if (err)
1947			break;
1948		err = bpf_map_copy_value(map, key, value,
1949					 attr->batch.elem_flags);
1950
1951		if (err == -ENOENT) {
1952			if (retry) {
1953				retry--;
1954				continue;
1955			}
1956			err = -EINTR;
1957			break;
1958		}
1959
1960		if (err)
1961			goto free_buf;
1962
1963		if (copy_to_user(keys + cp * map->key_size, key,
1964				 map->key_size)) {
1965			err = -EFAULT;
1966			goto free_buf;
1967		}
1968		if (copy_to_user(values + cp * value_size, value, value_size)) {
1969			err = -EFAULT;
1970			goto free_buf;
1971		}
1972
1973		if (!prev_key)
1974			prev_key = buf_prevkey;
1975
1976		swap(prev_key, key);
1977		retry = MAP_LOOKUP_RETRIES;
1978		cp++;
1979		cond_resched();
1980	}
1981
1982	if (err == -EFAULT)
1983		goto free_buf;
1984
1985	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1986		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1987		err = -EFAULT;
1988
1989free_buf:
1990	kvfree(buf_prevkey);
1991	kvfree(buf);
1992	return err;
1993}
1994
1995#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1996
1997static int map_lookup_and_delete_elem(union bpf_attr *attr)
1998{
1999	void __user *ukey = u64_to_user_ptr(attr->key);
2000	void __user *uvalue = u64_to_user_ptr(attr->value);
2001	int ufd = attr->map_fd;
2002	struct bpf_map *map;
2003	void *key, *value;
2004	u32 value_size;
2005	struct fd f;
2006	int err;
2007
2008	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2009		return -EINVAL;
2010
2011	if (attr->flags & ~BPF_F_LOCK)
2012		return -EINVAL;
2013
2014	f = fdget(ufd);
2015	map = __bpf_map_get(f);
2016	if (IS_ERR(map))
2017		return PTR_ERR(map);
2018	bpf_map_write_active_inc(map);
2019	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2020	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2021		err = -EPERM;
2022		goto err_put;
2023	}
2024
2025	if (attr->flags &&
2026	    (map->map_type == BPF_MAP_TYPE_QUEUE ||
2027	     map->map_type == BPF_MAP_TYPE_STACK)) {
2028		err = -EINVAL;
2029		goto err_put;
2030	}
2031
2032	if ((attr->flags & BPF_F_LOCK) &&
2033	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2034		err = -EINVAL;
2035		goto err_put;
2036	}
2037
2038	key = __bpf_copy_key(ukey, map->key_size);
2039	if (IS_ERR(key)) {
2040		err = PTR_ERR(key);
2041		goto err_put;
2042	}
2043
2044	value_size = bpf_map_value_size(map);
2045
2046	err = -ENOMEM;
2047	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2048	if (!value)
2049		goto free_key;
2050
2051	err = -ENOTSUPP;
2052	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2053	    map->map_type == BPF_MAP_TYPE_STACK) {
2054		err = map->ops->map_pop_elem(map, value);
2055	} else if (map->map_type == BPF_MAP_TYPE_HASH ||
2056		   map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2057		   map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2058		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2059		if (!bpf_map_is_offloaded(map)) {
2060			bpf_disable_instrumentation();
2061			rcu_read_lock();
2062			err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2063			rcu_read_unlock();
2064			bpf_enable_instrumentation();
2065		}
2066	}
2067
2068	if (err)
2069		goto free_value;
2070
2071	if (copy_to_user(uvalue, value, value_size) != 0) {
2072		err = -EFAULT;
2073		goto free_value;
2074	}
2075
2076	err = 0;
2077
2078free_value:
2079	kvfree(value);
2080free_key:
2081	kvfree(key);
2082err_put:
2083	bpf_map_write_active_dec(map);
2084	fdput(f);
2085	return err;
2086}
2087
2088#define BPF_MAP_FREEZE_LAST_FIELD map_fd
2089
2090static int map_freeze(const union bpf_attr *attr)
2091{
2092	int err = 0, ufd = attr->map_fd;
2093	struct bpf_map *map;
2094	struct fd f;
2095
2096	if (CHECK_ATTR(BPF_MAP_FREEZE))
2097		return -EINVAL;
2098
2099	f = fdget(ufd);
2100	map = __bpf_map_get(f);
2101	if (IS_ERR(map))
2102		return PTR_ERR(map);
2103
2104	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
2105		fdput(f);
2106		return -ENOTSUPP;
2107	}
2108
2109	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2110		fdput(f);
2111		return -EPERM;
2112	}
2113
2114	mutex_lock(&map->freeze_mutex);
2115	if (bpf_map_write_active(map)) {
2116		err = -EBUSY;
2117		goto err_put;
2118	}
2119	if (READ_ONCE(map->frozen)) {
2120		err = -EBUSY;
2121		goto err_put;
2122	}
 
 
 
 
2123
2124	WRITE_ONCE(map->frozen, true);
2125err_put:
2126	mutex_unlock(&map->freeze_mutex);
2127	fdput(f);
2128	return err;
2129}
2130
2131static const struct bpf_prog_ops * const bpf_prog_types[] = {
2132#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2133	[_id] = & _name ## _prog_ops,
2134#define BPF_MAP_TYPE(_id, _ops)
2135#define BPF_LINK_TYPE(_id, _name)
2136#include <linux/bpf_types.h>
2137#undef BPF_PROG_TYPE
2138#undef BPF_MAP_TYPE
2139#undef BPF_LINK_TYPE
2140};
2141
2142static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2143{
2144	const struct bpf_prog_ops *ops;
2145
2146	if (type >= ARRAY_SIZE(bpf_prog_types))
2147		return -EINVAL;
2148	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2149	ops = bpf_prog_types[type];
2150	if (!ops)
2151		return -EINVAL;
2152
2153	if (!bpf_prog_is_offloaded(prog->aux))
2154		prog->aux->ops = ops;
2155	else
2156		prog->aux->ops = &bpf_offload_prog_ops;
2157	prog->type = type;
2158	return 0;
2159}
2160
2161enum bpf_audit {
2162	BPF_AUDIT_LOAD,
2163	BPF_AUDIT_UNLOAD,
2164	BPF_AUDIT_MAX,
2165};
2166
2167static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2168	[BPF_AUDIT_LOAD]   = "LOAD",
2169	[BPF_AUDIT_UNLOAD] = "UNLOAD",
2170};
2171
2172static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2173{
2174	struct audit_context *ctx = NULL;
2175	struct audit_buffer *ab;
2176
2177	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2178		return;
2179	if (audit_enabled == AUDIT_OFF)
2180		return;
2181	if (!in_irq() && !irqs_disabled())
2182		ctx = audit_context();
2183	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2184	if (unlikely(!ab))
2185		return;
2186	audit_log_format(ab, "prog-id=%u op=%s",
2187			 prog->aux->id, bpf_audit_str[op]);
2188	audit_log_end(ab);
2189}
2190
2191static int bpf_prog_alloc_id(struct bpf_prog *prog)
2192{
2193	int id;
2194
2195	idr_preload(GFP_KERNEL);
2196	spin_lock_bh(&prog_idr_lock);
2197	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2198	if (id > 0)
2199		prog->aux->id = id;
2200	spin_unlock_bh(&prog_idr_lock);
2201	idr_preload_end();
2202
2203	/* id is in [1, INT_MAX) */
2204	if (WARN_ON_ONCE(!id))
2205		return -ENOSPC;
2206
2207	return id > 0 ? 0 : id;
2208}
2209
2210void bpf_prog_free_id(struct bpf_prog *prog)
2211{
2212	unsigned long flags;
2213
2214	/* cBPF to eBPF migrations are currently not in the idr store.
2215	 * Offloaded programs are removed from the store when their device
2216	 * disappears - even if someone grabs an fd to them they are unusable,
2217	 * simply waiting for refcnt to drop to be freed.
2218	 */
2219	if (!prog->aux->id)
2220		return;
2221
2222	spin_lock_irqsave(&prog_idr_lock, flags);
2223	idr_remove(&prog_idr, prog->aux->id);
2224	prog->aux->id = 0;
2225	spin_unlock_irqrestore(&prog_idr_lock, flags);
2226}
2227
2228static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2229{
2230	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2231
2232	kvfree(aux->func_info);
2233	kfree(aux->func_info_aux);
2234	free_uid(aux->user);
2235	security_bpf_prog_free(aux->prog);
2236	bpf_prog_free(aux->prog);
2237}
2238
2239static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2240{
2241	bpf_prog_kallsyms_del_all(prog);
2242	btf_put(prog->aux->btf);
2243	module_put(prog->aux->mod);
2244	kvfree(prog->aux->jited_linfo);
2245	kvfree(prog->aux->linfo);
2246	kfree(prog->aux->kfunc_tab);
2247	if (prog->aux->attach_btf)
2248		btf_put(prog->aux->attach_btf);
2249
2250	if (deferred) {
2251		if (prog->sleepable)
2252			call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2253		else
2254			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2255	} else {
2256		__bpf_prog_put_rcu(&prog->aux->rcu);
2257	}
2258}
2259
2260static void bpf_prog_put_deferred(struct work_struct *work)
2261{
2262	struct bpf_prog_aux *aux;
2263	struct bpf_prog *prog;
2264
2265	aux = container_of(work, struct bpf_prog_aux, work);
2266	prog = aux->prog;
2267	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2268	bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2269	bpf_prog_free_id(prog);
2270	__bpf_prog_put_noref(prog, true);
2271}
2272
2273static void __bpf_prog_put(struct bpf_prog *prog)
2274{
2275	struct bpf_prog_aux *aux = prog->aux;
2276
2277	if (atomic64_dec_and_test(&aux->refcnt)) {
2278		if (in_irq() || irqs_disabled()) {
2279			INIT_WORK(&aux->work, bpf_prog_put_deferred);
2280			schedule_work(&aux->work);
2281		} else {
2282			bpf_prog_put_deferred(&aux->work);
2283		}
2284	}
2285}
2286
2287void bpf_prog_put(struct bpf_prog *prog)
2288{
2289	__bpf_prog_put(prog);
2290}
2291EXPORT_SYMBOL_GPL(bpf_prog_put);
2292
2293static int bpf_prog_release(struct inode *inode, struct file *filp)
2294{
2295	struct bpf_prog *prog = filp->private_data;
2296
2297	bpf_prog_put(prog);
2298	return 0;
2299}
2300
2301struct bpf_prog_kstats {
2302	u64 nsecs;
2303	u64 cnt;
2304	u64 misses;
2305};
2306
2307void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2308{
2309	struct bpf_prog_stats *stats;
2310	unsigned int flags;
2311
2312	stats = this_cpu_ptr(prog->stats);
2313	flags = u64_stats_update_begin_irqsave(&stats->syncp);
2314	u64_stats_inc(&stats->misses);
2315	u64_stats_update_end_irqrestore(&stats->syncp, flags);
2316}
2317
2318static void bpf_prog_get_stats(const struct bpf_prog *prog,
2319			       struct bpf_prog_kstats *stats)
2320{
2321	u64 nsecs = 0, cnt = 0, misses = 0;
2322	int cpu;
2323
2324	for_each_possible_cpu(cpu) {
2325		const struct bpf_prog_stats *st;
2326		unsigned int start;
2327		u64 tnsecs, tcnt, tmisses;
2328
2329		st = per_cpu_ptr(prog->stats, cpu);
2330		do {
2331			start = u64_stats_fetch_begin(&st->syncp);
2332			tnsecs = u64_stats_read(&st->nsecs);
2333			tcnt = u64_stats_read(&st->cnt);
2334			tmisses = u64_stats_read(&st->misses);
2335		} while (u64_stats_fetch_retry(&st->syncp, start));
2336		nsecs += tnsecs;
2337		cnt += tcnt;
2338		misses += tmisses;
2339	}
2340	stats->nsecs = nsecs;
2341	stats->cnt = cnt;
2342	stats->misses = misses;
2343}
2344
2345#ifdef CONFIG_PROC_FS
2346static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2347{
2348	const struct bpf_prog *prog = filp->private_data;
2349	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2350	struct bpf_prog_kstats stats;
2351
2352	bpf_prog_get_stats(prog, &stats);
2353	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2354	seq_printf(m,
2355		   "prog_type:\t%u\n"
2356		   "prog_jited:\t%u\n"
2357		   "prog_tag:\t%s\n"
2358		   "memlock:\t%llu\n"
2359		   "prog_id:\t%u\n"
2360		   "run_time_ns:\t%llu\n"
2361		   "run_cnt:\t%llu\n"
2362		   "recursion_misses:\t%llu\n"
2363		   "verified_insns:\t%u\n",
2364		   prog->type,
2365		   prog->jited,
2366		   prog_tag,
2367		   prog->pages * 1ULL << PAGE_SHIFT,
2368		   prog->aux->id,
2369		   stats.nsecs,
2370		   stats.cnt,
2371		   stats.misses,
2372		   prog->aux->verified_insns);
2373}
2374#endif
2375
2376const struct file_operations bpf_prog_fops = {
2377#ifdef CONFIG_PROC_FS
2378	.show_fdinfo	= bpf_prog_show_fdinfo,
2379#endif
2380	.release	= bpf_prog_release,
2381	.read		= bpf_dummy_read,
2382	.write		= bpf_dummy_write,
2383};
2384
2385int bpf_prog_new_fd(struct bpf_prog *prog)
2386{
2387	int ret;
2388
2389	ret = security_bpf_prog(prog);
2390	if (ret < 0)
2391		return ret;
2392
2393	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2394				O_RDWR | O_CLOEXEC);
2395}
2396
2397static struct bpf_prog *____bpf_prog_get(struct fd f)
2398{
2399	if (!f.file)
2400		return ERR_PTR(-EBADF);
2401	if (f.file->f_op != &bpf_prog_fops) {
2402		fdput(f);
2403		return ERR_PTR(-EINVAL);
2404	}
2405
2406	return f.file->private_data;
2407}
2408
2409void bpf_prog_add(struct bpf_prog *prog, int i)
2410{
2411	atomic64_add(i, &prog->aux->refcnt);
2412}
2413EXPORT_SYMBOL_GPL(bpf_prog_add);
2414
2415void bpf_prog_sub(struct bpf_prog *prog, int i)
2416{
2417	/* Only to be used for undoing previous bpf_prog_add() in some
2418	 * error path. We still know that another entity in our call
2419	 * path holds a reference to the program, thus atomic_sub() can
2420	 * be safely used in such cases!
2421	 */
2422	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2423}
2424EXPORT_SYMBOL_GPL(bpf_prog_sub);
2425
2426void bpf_prog_inc(struct bpf_prog *prog)
2427{
2428	atomic64_inc(&prog->aux->refcnt);
2429}
2430EXPORT_SYMBOL_GPL(bpf_prog_inc);
2431
2432/* prog_idr_lock should have been held */
2433struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2434{
2435	int refold;
2436
2437	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2438
2439	if (!refold)
2440		return ERR_PTR(-ENOENT);
2441
2442	return prog;
2443}
2444EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2445
2446bool bpf_prog_get_ok(struct bpf_prog *prog,
2447			    enum bpf_prog_type *attach_type, bool attach_drv)
2448{
2449	/* not an attachment, just a refcount inc, always allow */
2450	if (!attach_type)
2451		return true;
2452
2453	if (prog->type != *attach_type)
2454		return false;
2455	if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2456		return false;
2457
2458	return true;
2459}
2460
2461static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2462				       bool attach_drv)
2463{
2464	struct fd f = fdget(ufd);
2465	struct bpf_prog *prog;
2466
2467	prog = ____bpf_prog_get(f);
2468	if (IS_ERR(prog))
2469		return prog;
2470	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2471		prog = ERR_PTR(-EINVAL);
2472		goto out;
2473	}
2474
2475	bpf_prog_inc(prog);
2476out:
2477	fdput(f);
2478	return prog;
2479}
2480
2481struct bpf_prog *bpf_prog_get(u32 ufd)
2482{
2483	return __bpf_prog_get(ufd, NULL, false);
2484}
2485
2486struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2487				       bool attach_drv)
2488{
2489	return __bpf_prog_get(ufd, &type, attach_drv);
2490}
2491EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2492
2493/* Initially all BPF programs could be loaded w/o specifying
2494 * expected_attach_type. Later for some of them specifying expected_attach_type
2495 * at load time became required so that program could be validated properly.
2496 * Programs of types that are allowed to be loaded both w/ and w/o (for
2497 * backward compatibility) expected_attach_type, should have the default attach
2498 * type assigned to expected_attach_type for the latter case, so that it can be
2499 * validated later at attach time.
2500 *
2501 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2502 * prog type requires it but has some attach types that have to be backward
2503 * compatible.
2504 */
2505static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2506{
2507	switch (attr->prog_type) {
2508	case BPF_PROG_TYPE_CGROUP_SOCK:
2509		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2510		 * exist so checking for non-zero is the way to go here.
2511		 */
2512		if (!attr->expected_attach_type)
2513			attr->expected_attach_type =
2514				BPF_CGROUP_INET_SOCK_CREATE;
2515		break;
2516	case BPF_PROG_TYPE_SK_REUSEPORT:
2517		if (!attr->expected_attach_type)
2518			attr->expected_attach_type =
2519				BPF_SK_REUSEPORT_SELECT;
2520		break;
2521	}
2522}
2523
2524static int
2525bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2526			   enum bpf_attach_type expected_attach_type,
2527			   struct btf *attach_btf, u32 btf_id,
2528			   struct bpf_prog *dst_prog)
2529{
2530	if (btf_id) {
2531		if (btf_id > BTF_MAX_TYPE)
2532			return -EINVAL;
2533
2534		if (!attach_btf && !dst_prog)
2535			return -EINVAL;
2536
2537		switch (prog_type) {
2538		case BPF_PROG_TYPE_TRACING:
2539		case BPF_PROG_TYPE_LSM:
2540		case BPF_PROG_TYPE_STRUCT_OPS:
2541		case BPF_PROG_TYPE_EXT:
2542			break;
2543		default:
2544			return -EINVAL;
2545		}
2546	}
2547
2548	if (attach_btf && (!btf_id || dst_prog))
2549		return -EINVAL;
2550
2551	if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2552	    prog_type != BPF_PROG_TYPE_EXT)
2553		return -EINVAL;
2554
2555	switch (prog_type) {
2556	case BPF_PROG_TYPE_CGROUP_SOCK:
2557		switch (expected_attach_type) {
2558		case BPF_CGROUP_INET_SOCK_CREATE:
2559		case BPF_CGROUP_INET_SOCK_RELEASE:
2560		case BPF_CGROUP_INET4_POST_BIND:
2561		case BPF_CGROUP_INET6_POST_BIND:
2562			return 0;
2563		default:
2564			return -EINVAL;
2565		}
2566	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2567		switch (expected_attach_type) {
2568		case BPF_CGROUP_INET4_BIND:
2569		case BPF_CGROUP_INET6_BIND:
2570		case BPF_CGROUP_INET4_CONNECT:
2571		case BPF_CGROUP_INET6_CONNECT:
2572		case BPF_CGROUP_UNIX_CONNECT:
2573		case BPF_CGROUP_INET4_GETPEERNAME:
2574		case BPF_CGROUP_INET6_GETPEERNAME:
2575		case BPF_CGROUP_UNIX_GETPEERNAME:
2576		case BPF_CGROUP_INET4_GETSOCKNAME:
2577		case BPF_CGROUP_INET6_GETSOCKNAME:
2578		case BPF_CGROUP_UNIX_GETSOCKNAME:
2579		case BPF_CGROUP_UDP4_SENDMSG:
2580		case BPF_CGROUP_UDP6_SENDMSG:
2581		case BPF_CGROUP_UNIX_SENDMSG:
2582		case BPF_CGROUP_UDP4_RECVMSG:
2583		case BPF_CGROUP_UDP6_RECVMSG:
2584		case BPF_CGROUP_UNIX_RECVMSG:
2585			return 0;
2586		default:
2587			return -EINVAL;
2588		}
2589	case BPF_PROG_TYPE_CGROUP_SKB:
2590		switch (expected_attach_type) {
2591		case BPF_CGROUP_INET_INGRESS:
2592		case BPF_CGROUP_INET_EGRESS:
2593			return 0;
2594		default:
2595			return -EINVAL;
2596		}
2597	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2598		switch (expected_attach_type) {
2599		case BPF_CGROUP_SETSOCKOPT:
2600		case BPF_CGROUP_GETSOCKOPT:
2601			return 0;
2602		default:
2603			return -EINVAL;
2604		}
2605	case BPF_PROG_TYPE_SK_LOOKUP:
2606		if (expected_attach_type == BPF_SK_LOOKUP)
2607			return 0;
2608		return -EINVAL;
2609	case BPF_PROG_TYPE_SK_REUSEPORT:
2610		switch (expected_attach_type) {
2611		case BPF_SK_REUSEPORT_SELECT:
2612		case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2613			return 0;
2614		default:
2615			return -EINVAL;
2616		}
2617	case BPF_PROG_TYPE_NETFILTER:
2618		if (expected_attach_type == BPF_NETFILTER)
2619			return 0;
2620		return -EINVAL;
2621	case BPF_PROG_TYPE_SYSCALL:
2622	case BPF_PROG_TYPE_EXT:
2623		if (expected_attach_type)
2624			return -EINVAL;
2625		fallthrough;
2626	default:
2627		return 0;
2628	}
2629}
2630
2631static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2632{
2633	switch (prog_type) {
2634	case BPF_PROG_TYPE_SCHED_CLS:
2635	case BPF_PROG_TYPE_SCHED_ACT:
2636	case BPF_PROG_TYPE_XDP:
2637	case BPF_PROG_TYPE_LWT_IN:
2638	case BPF_PROG_TYPE_LWT_OUT:
2639	case BPF_PROG_TYPE_LWT_XMIT:
2640	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2641	case BPF_PROG_TYPE_SK_SKB:
2642	case BPF_PROG_TYPE_SK_MSG:
 
2643	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2644	case BPF_PROG_TYPE_CGROUP_DEVICE:
2645	case BPF_PROG_TYPE_CGROUP_SOCK:
2646	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2647	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2648	case BPF_PROG_TYPE_CGROUP_SYSCTL:
2649	case BPF_PROG_TYPE_SOCK_OPS:
2650	case BPF_PROG_TYPE_EXT: /* extends any prog */
2651	case BPF_PROG_TYPE_NETFILTER:
2652		return true;
2653	case BPF_PROG_TYPE_CGROUP_SKB:
2654		/* always unpriv */
2655	case BPF_PROG_TYPE_SK_REUSEPORT:
2656		/* equivalent to SOCKET_FILTER. need CAP_BPF only */
2657	default:
2658		return false;
2659	}
2660}
2661
2662static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2663{
2664	switch (prog_type) {
2665	case BPF_PROG_TYPE_KPROBE:
2666	case BPF_PROG_TYPE_TRACEPOINT:
2667	case BPF_PROG_TYPE_PERF_EVENT:
2668	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2669	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2670	case BPF_PROG_TYPE_TRACING:
2671	case BPF_PROG_TYPE_LSM:
2672	case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2673	case BPF_PROG_TYPE_EXT: /* extends any prog */
2674		return true;
2675	default:
2676		return false;
2677	}
2678}
2679
2680/* last field in 'union bpf_attr' used by this command */
2681#define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2682
2683static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2684{
2685	enum bpf_prog_type type = attr->prog_type;
2686	struct bpf_prog *prog, *dst_prog = NULL;
2687	struct btf *attach_btf = NULL;
2688	struct bpf_token *token = NULL;
2689	bool bpf_cap;
2690	int err;
2691	char license[128];
 
2692
2693	if (CHECK_ATTR(BPF_PROG_LOAD))
2694		return -EINVAL;
2695
2696	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2697				 BPF_F_ANY_ALIGNMENT |
2698				 BPF_F_TEST_STATE_FREQ |
2699				 BPF_F_SLEEPABLE |
2700				 BPF_F_TEST_RND_HI32 |
2701				 BPF_F_XDP_HAS_FRAGS |
2702				 BPF_F_XDP_DEV_BOUND_ONLY |
2703				 BPF_F_TEST_REG_INVARIANTS |
2704				 BPF_F_TOKEN_FD))
2705		return -EINVAL;
2706
2707	bpf_prog_load_fixup_attach_type(attr);
2708
2709	if (attr->prog_flags & BPF_F_TOKEN_FD) {
2710		token = bpf_token_get_from_fd(attr->prog_token_fd);
2711		if (IS_ERR(token))
2712			return PTR_ERR(token);
2713		/* if current token doesn't grant prog loading permissions,
2714		 * then we can't use this token, so ignore it and rely on
2715		 * system-wide capabilities checks
2716		 */
2717		if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2718		    !bpf_token_allow_prog_type(token, attr->prog_type,
2719					       attr->expected_attach_type)) {
2720			bpf_token_put(token);
2721			token = NULL;
2722		}
2723	}
2724
2725	bpf_cap = bpf_token_capable(token, CAP_BPF);
2726	err = -EPERM;
2727
2728	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2729	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2730	    !bpf_cap)
2731		goto put_token;
 
 
 
 
 
 
 
2732
2733	/* Intent here is for unprivileged_bpf_disabled to block BPF program
2734	 * creation for unprivileged users; other actions depend
2735	 * on fd availability and access to bpffs, so are dependent on
2736	 * object creation success. Even with unprivileged BPF disabled,
2737	 * capability checks are still carried out for these
2738	 * and other operations.
2739	 */
2740	if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2741		goto put_token;
2742
2743	if (attr->insn_cnt == 0 ||
2744	    attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2745		err = -E2BIG;
2746		goto put_token;
2747	}
2748	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2749	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2750	    !bpf_cap)
2751		goto put_token;
2752
2753	if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2754		goto put_token;
2755	if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2756		goto put_token;
2757
2758	/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2759	 * or btf, we need to check which one it is
2760	 */
2761	if (attr->attach_prog_fd) {
2762		dst_prog = bpf_prog_get(attr->attach_prog_fd);
2763		if (IS_ERR(dst_prog)) {
2764			dst_prog = NULL;
2765			attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2766			if (IS_ERR(attach_btf)) {
2767				err = -EINVAL;
2768				goto put_token;
2769			}
2770			if (!btf_is_kernel(attach_btf)) {
2771				/* attaching through specifying bpf_prog's BTF
2772				 * objects directly might be supported eventually
2773				 */
2774				btf_put(attach_btf);
2775				err = -ENOTSUPP;
2776				goto put_token;
2777			}
2778		}
2779	} else if (attr->attach_btf_id) {
2780		/* fall back to vmlinux BTF, if BTF type ID is specified */
2781		attach_btf = bpf_get_btf_vmlinux();
2782		if (IS_ERR(attach_btf)) {
2783			err = PTR_ERR(attach_btf);
2784			goto put_token;
2785		}
2786		if (!attach_btf) {
2787			err = -EINVAL;
2788			goto put_token;
2789		}
2790		btf_get(attach_btf);
2791	}
2792
 
2793	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2794				       attach_btf, attr->attach_btf_id,
2795				       dst_prog)) {
2796		if (dst_prog)
2797			bpf_prog_put(dst_prog);
2798		if (attach_btf)
2799			btf_put(attach_btf);
2800		err = -EINVAL;
2801		goto put_token;
2802	}
2803
2804	/* plain bpf_prog allocation */
2805	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2806	if (!prog) {
2807		if (dst_prog)
2808			bpf_prog_put(dst_prog);
2809		if (attach_btf)
2810			btf_put(attach_btf);
2811		err = -EINVAL;
2812		goto put_token;
2813	}
2814
2815	prog->expected_attach_type = attr->expected_attach_type;
2816	prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2817	prog->aux->attach_btf = attach_btf;
2818	prog->aux->attach_btf_id = attr->attach_btf_id;
2819	prog->aux->dst_prog = dst_prog;
2820	prog->aux->dev_bound = !!attr->prog_ifindex;
 
2821	prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2822
2823	/* move token into prog->aux, reuse taken refcnt */
2824	prog->aux->token = token;
2825	token = NULL;
2826
2827	prog->aux->user = get_current_user();
2828	prog->len = attr->insn_cnt;
2829
2830	err = -EFAULT;
2831	if (copy_from_bpfptr(prog->insns,
2832			     make_bpfptr(attr->insns, uattr.is_kernel),
2833			     bpf_prog_insn_size(prog)) != 0)
2834		goto free_prog;
2835	/* copy eBPF program license from user space */
2836	if (strncpy_from_bpfptr(license,
2837				make_bpfptr(attr->license, uattr.is_kernel),
2838				sizeof(license) - 1) < 0)
2839		goto free_prog;
2840	license[sizeof(license) - 1] = 0;
2841
2842	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2843	prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2844
2845	prog->orig_prog = NULL;
2846	prog->jited = 0;
2847
2848	atomic64_set(&prog->aux->refcnt, 1);
 
2849
2850	if (bpf_prog_is_dev_bound(prog->aux)) {
2851		err = bpf_prog_dev_bound_init(prog, attr);
2852		if (err)
2853			goto free_prog;
2854	}
2855
2856	if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2857	    bpf_prog_is_dev_bound(dst_prog->aux)) {
2858		err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2859		if (err)
2860			goto free_prog;
2861	}
2862
2863	/*
2864	 * Bookkeeping for managing the program attachment chain.
2865	 *
2866	 * It might be tempting to set attach_tracing_prog flag at the attachment
2867	 * time, but this will not prevent from loading bunch of tracing prog
2868	 * first, then attach them one to another.
2869	 *
2870	 * The flag attach_tracing_prog is set for the whole program lifecycle, and
2871	 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2872	 * programs cannot change attachment target.
2873	 */
2874	if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2875	    dst_prog->type == BPF_PROG_TYPE_TRACING) {
2876		prog->aux->attach_tracing_prog = true;
2877	}
2878
2879	/* find program type: socket_filter vs tracing_filter */
2880	err = find_prog_type(type, prog);
2881	if (err < 0)
2882		goto free_prog;
2883
2884	prog->aux->load_time = ktime_get_boottime_ns();
2885	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2886			       sizeof(attr->prog_name));
2887	if (err < 0)
2888		goto free_prog;
2889
2890	err = security_bpf_prog_load(prog, attr, token);
2891	if (err)
2892		goto free_prog_sec;
2893
2894	/* run eBPF verifier */
2895	err = bpf_check(&prog, attr, uattr, uattr_size);
2896	if (err < 0)
2897		goto free_used_maps;
2898
2899	prog = bpf_prog_select_runtime(prog, &err);
2900	if (err < 0)
2901		goto free_used_maps;
2902
2903	err = bpf_prog_alloc_id(prog);
2904	if (err)
2905		goto free_used_maps;
2906
2907	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2908	 * effectively publicly exposed. However, retrieving via
2909	 * bpf_prog_get_fd_by_id() will take another reference,
2910	 * therefore it cannot be gone underneath us.
2911	 *
2912	 * Only for the time /after/ successful bpf_prog_new_fd()
2913	 * and before returning to userspace, we might just hold
2914	 * one reference and any parallel close on that fd could
2915	 * rip everything out. Hence, below notifications must
2916	 * happen before bpf_prog_new_fd().
2917	 *
2918	 * Also, any failure handling from this point onwards must
2919	 * be using bpf_prog_put() given the program is exposed.
2920	 */
2921	bpf_prog_kallsyms_add(prog);
2922	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2923	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2924
2925	err = bpf_prog_new_fd(prog);
2926	if (err < 0)
2927		bpf_prog_put(prog);
2928	return err;
2929
2930free_used_maps:
2931	/* In case we have subprogs, we need to wait for a grace
2932	 * period before we can tear down JIT memory since symbols
2933	 * are already exposed under kallsyms.
2934	 */
2935	__bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2936	return err;
2937
2938free_prog_sec:
2939	security_bpf_prog_free(prog);
 
2940free_prog:
2941	free_uid(prog->aux->user);
2942	if (prog->aux->attach_btf)
2943		btf_put(prog->aux->attach_btf);
2944	bpf_prog_free(prog);
2945put_token:
2946	bpf_token_put(token);
2947	return err;
2948}
2949
2950#define BPF_OBJ_LAST_FIELD path_fd
2951
2952static int bpf_obj_pin(const union bpf_attr *attr)
2953{
2954	int path_fd;
2955
2956	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2957		return -EINVAL;
2958
2959	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2960	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2961		return -EINVAL;
2962
2963	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2964	return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2965				u64_to_user_ptr(attr->pathname));
2966}
2967
2968static int bpf_obj_get(const union bpf_attr *attr)
2969{
2970	int path_fd;
2971
2972	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2973	    attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2974		return -EINVAL;
2975
2976	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2977	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2978		return -EINVAL;
2979
2980	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2981	return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2982				attr->file_flags);
2983}
2984
2985void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2986		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
2987{
2988	atomic64_set(&link->refcnt, 1);
2989	link->type = type;
2990	link->id = 0;
2991	link->ops = ops;
2992	link->prog = prog;
2993}
2994
2995static void bpf_link_free_id(int id)
2996{
2997	if (!id)
2998		return;
2999
3000	spin_lock_bh(&link_idr_lock);
3001	idr_remove(&link_idr, id);
3002	spin_unlock_bh(&link_idr_lock);
3003}
3004
3005/* Clean up bpf_link and corresponding anon_inode file and FD. After
3006 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3007 * anon_inode's release() call. This helper marks bpf_link as
3008 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3009 * is not decremented, it's the responsibility of a calling code that failed
3010 * to complete bpf_link initialization.
3011 * This helper eventually calls link's dealloc callback, but does not call
3012 * link's release callback.
3013 */
3014void bpf_link_cleanup(struct bpf_link_primer *primer)
3015{
3016	primer->link->prog = NULL;
3017	bpf_link_free_id(primer->id);
3018	fput(primer->file);
3019	put_unused_fd(primer->fd);
3020}
3021
3022void bpf_link_inc(struct bpf_link *link)
3023{
3024	atomic64_inc(&link->refcnt);
3025}
3026
3027static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3028{
3029	struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3030
3031	/* free bpf_link and its containing memory */
3032	link->ops->dealloc_deferred(link);
3033}
3034
3035static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
3036{
3037	if (rcu_trace_implies_rcu_gp())
3038		bpf_link_defer_dealloc_rcu_gp(rcu);
3039	else
3040		call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
3041}
3042
3043/* bpf_link_free is guaranteed to be called from process context */
3044static void bpf_link_free(struct bpf_link *link)
3045{
3046	bool sleepable = false;
3047
3048	bpf_link_free_id(link->id);
3049	if (link->prog) {
3050		sleepable = link->prog->sleepable;
3051		/* detach BPF program, clean up used resources */
3052		link->ops->release(link);
3053		bpf_prog_put(link->prog);
3054	}
3055	if (link->ops->dealloc_deferred) {
3056		/* schedule BPF link deallocation; if underlying BPF program
3057		 * is sleepable, we need to first wait for RCU tasks trace
3058		 * sync, then go through "classic" RCU grace period
3059		 */
3060		if (sleepable)
3061			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3062		else
3063			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3064	}
3065	if (link->ops->dealloc)
3066		link->ops->dealloc(link);
3067}
3068
3069static void bpf_link_put_deferred(struct work_struct *work)
3070{
3071	struct bpf_link *link = container_of(work, struct bpf_link, work);
3072
3073	bpf_link_free(link);
3074}
3075
3076/* bpf_link_put might be called from atomic context. It needs to be called
3077 * from sleepable context in order to acquire sleeping locks during the process.
3078 */
3079void bpf_link_put(struct bpf_link *link)
3080{
3081	if (!atomic64_dec_and_test(&link->refcnt))
3082		return;
3083
3084	INIT_WORK(&link->work, bpf_link_put_deferred);
3085	schedule_work(&link->work);
 
 
 
 
3086}
3087EXPORT_SYMBOL(bpf_link_put);
3088
3089static void bpf_link_put_direct(struct bpf_link *link)
3090{
3091	if (!atomic64_dec_and_test(&link->refcnt))
3092		return;
3093	bpf_link_free(link);
3094}
3095
3096static int bpf_link_release(struct inode *inode, struct file *filp)
3097{
3098	struct bpf_link *link = filp->private_data;
3099
3100	bpf_link_put_direct(link);
3101	return 0;
3102}
3103
3104#ifdef CONFIG_PROC_FS
3105#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3106#define BPF_MAP_TYPE(_id, _ops)
3107#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3108static const char *bpf_link_type_strs[] = {
3109	[BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3110#include <linux/bpf_types.h>
3111};
3112#undef BPF_PROG_TYPE
3113#undef BPF_MAP_TYPE
3114#undef BPF_LINK_TYPE
3115
3116static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3117{
3118	const struct bpf_link *link = filp->private_data;
3119	const struct bpf_prog *prog = link->prog;
3120	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3121
 
3122	seq_printf(m,
3123		   "link_type:\t%s\n"
3124		   "link_id:\t%u\n",
 
 
3125		   bpf_link_type_strs[link->type],
3126		   link->id);
3127	if (prog) {
3128		bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3129		seq_printf(m,
3130			   "prog_tag:\t%s\n"
3131			   "prog_id:\t%u\n",
3132			   prog_tag,
3133			   prog->aux->id);
3134	}
3135	if (link->ops->show_fdinfo)
3136		link->ops->show_fdinfo(link, m);
3137}
3138#endif
3139
3140static const struct file_operations bpf_link_fops = {
3141#ifdef CONFIG_PROC_FS
3142	.show_fdinfo	= bpf_link_show_fdinfo,
3143#endif
3144	.release	= bpf_link_release,
3145	.read		= bpf_dummy_read,
3146	.write		= bpf_dummy_write,
3147};
3148
3149static int bpf_link_alloc_id(struct bpf_link *link)
3150{
3151	int id;
3152
3153	idr_preload(GFP_KERNEL);
3154	spin_lock_bh(&link_idr_lock);
3155	id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3156	spin_unlock_bh(&link_idr_lock);
3157	idr_preload_end();
3158
3159	return id;
3160}
3161
3162/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3163 * reserving unused FD and allocating ID from link_idr. This is to be paired
3164 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3165 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3166 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3167 * transient state is passed around in struct bpf_link_primer.
3168 * This is preferred way to create and initialize bpf_link, especially when
3169 * there are complicated and expensive operations in between creating bpf_link
3170 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3171 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3172 * expensive (and potentially failing) roll back operations in a rare case
3173 * that file, FD, or ID can't be allocated.
3174 */
3175int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3176{
3177	struct file *file;
3178	int fd, id;
3179
3180	fd = get_unused_fd_flags(O_CLOEXEC);
3181	if (fd < 0)
3182		return fd;
3183
3184
3185	id = bpf_link_alloc_id(link);
3186	if (id < 0) {
3187		put_unused_fd(fd);
3188		return id;
3189	}
3190
3191	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
3192	if (IS_ERR(file)) {
3193		bpf_link_free_id(id);
3194		put_unused_fd(fd);
3195		return PTR_ERR(file);
3196	}
3197
3198	primer->link = link;
3199	primer->file = file;
3200	primer->fd = fd;
3201	primer->id = id;
3202	return 0;
3203}
3204
3205int bpf_link_settle(struct bpf_link_primer *primer)
3206{
3207	/* make bpf_link fetchable by ID */
3208	spin_lock_bh(&link_idr_lock);
3209	primer->link->id = primer->id;
3210	spin_unlock_bh(&link_idr_lock);
3211	/* make bpf_link fetchable by FD */
3212	fd_install(primer->fd, primer->file);
3213	/* pass through installed FD */
3214	return primer->fd;
3215}
3216
3217int bpf_link_new_fd(struct bpf_link *link)
3218{
3219	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3220}
3221
3222struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3223{
3224	struct fd f = fdget(ufd);
3225	struct bpf_link *link;
3226
3227	if (!f.file)
3228		return ERR_PTR(-EBADF);
3229	if (f.file->f_op != &bpf_link_fops) {
3230		fdput(f);
3231		return ERR_PTR(-EINVAL);
3232	}
3233
3234	link = f.file->private_data;
3235	bpf_link_inc(link);
3236	fdput(f);
3237
3238	return link;
3239}
3240EXPORT_SYMBOL(bpf_link_get_from_fd);
3241
3242static void bpf_tracing_link_release(struct bpf_link *link)
3243{
3244	struct bpf_tracing_link *tr_link =
3245		container_of(link, struct bpf_tracing_link, link.link);
3246
3247	WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3248						tr_link->trampoline));
3249
3250	bpf_trampoline_put(tr_link->trampoline);
3251
3252	/* tgt_prog is NULL if target is a kernel function */
3253	if (tr_link->tgt_prog)
3254		bpf_prog_put(tr_link->tgt_prog);
3255}
3256
3257static void bpf_tracing_link_dealloc(struct bpf_link *link)
3258{
3259	struct bpf_tracing_link *tr_link =
3260		container_of(link, struct bpf_tracing_link, link.link);
3261
3262	kfree(tr_link);
3263}
3264
3265static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3266					 struct seq_file *seq)
3267{
3268	struct bpf_tracing_link *tr_link =
3269		container_of(link, struct bpf_tracing_link, link.link);
3270	u32 target_btf_id, target_obj_id;
3271
3272	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3273				  &target_obj_id, &target_btf_id);
3274	seq_printf(seq,
3275		   "attach_type:\t%d\n"
3276		   "target_obj_id:\t%u\n"
3277		   "target_btf_id:\t%u\n",
3278		   tr_link->attach_type,
3279		   target_obj_id,
3280		   target_btf_id);
3281}
3282
3283static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3284					   struct bpf_link_info *info)
3285{
3286	struct bpf_tracing_link *tr_link =
3287		container_of(link, struct bpf_tracing_link, link.link);
3288
3289	info->tracing.attach_type = tr_link->attach_type;
3290	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3291				  &info->tracing.target_obj_id,
3292				  &info->tracing.target_btf_id);
3293
3294	return 0;
3295}
3296
3297static const struct bpf_link_ops bpf_tracing_link_lops = {
3298	.release = bpf_tracing_link_release,
3299	.dealloc = bpf_tracing_link_dealloc,
3300	.show_fdinfo = bpf_tracing_link_show_fdinfo,
3301	.fill_link_info = bpf_tracing_link_fill_link_info,
3302};
3303
3304static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3305				   int tgt_prog_fd,
3306				   u32 btf_id,
3307				   u64 bpf_cookie)
3308{
3309	struct bpf_link_primer link_primer;
3310	struct bpf_prog *tgt_prog = NULL;
3311	struct bpf_trampoline *tr = NULL;
3312	struct bpf_tracing_link *link;
3313	u64 key = 0;
3314	int err;
3315
3316	switch (prog->type) {
3317	case BPF_PROG_TYPE_TRACING:
3318		if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3319		    prog->expected_attach_type != BPF_TRACE_FEXIT &&
3320		    prog->expected_attach_type != BPF_MODIFY_RETURN) {
3321			err = -EINVAL;
3322			goto out_put_prog;
3323		}
3324		break;
3325	case BPF_PROG_TYPE_EXT:
3326		if (prog->expected_attach_type != 0) {
3327			err = -EINVAL;
3328			goto out_put_prog;
3329		}
3330		break;
3331	case BPF_PROG_TYPE_LSM:
3332		if (prog->expected_attach_type != BPF_LSM_MAC) {
3333			err = -EINVAL;
3334			goto out_put_prog;
3335		}
3336		break;
3337	default:
3338		err = -EINVAL;
3339		goto out_put_prog;
3340	}
3341
3342	if (!!tgt_prog_fd != !!btf_id) {
3343		err = -EINVAL;
3344		goto out_put_prog;
3345	}
3346
3347	if (tgt_prog_fd) {
3348		/*
3349		 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3350		 * part would be changed to implement the same for
3351		 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3352		 * attach_tracing_prog flag is set.
3353		 */
3354		if (prog->type != BPF_PROG_TYPE_EXT) {
3355			err = -EINVAL;
3356			goto out_put_prog;
3357		}
3358
3359		tgt_prog = bpf_prog_get(tgt_prog_fd);
3360		if (IS_ERR(tgt_prog)) {
3361			err = PTR_ERR(tgt_prog);
3362			tgt_prog = NULL;
3363			goto out_put_prog;
3364		}
3365
3366		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3367	}
3368
3369	link = kzalloc(sizeof(*link), GFP_USER);
3370	if (!link) {
3371		err = -ENOMEM;
3372		goto out_put_prog;
3373	}
3374	bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3375		      &bpf_tracing_link_lops, prog);
3376	link->attach_type = prog->expected_attach_type;
3377	link->link.cookie = bpf_cookie;
3378
3379	mutex_lock(&prog->aux->dst_mutex);
3380
3381	/* There are a few possible cases here:
3382	 *
3383	 * - if prog->aux->dst_trampoline is set, the program was just loaded
3384	 *   and not yet attached to anything, so we can use the values stored
3385	 *   in prog->aux
3386	 *
3387	 * - if prog->aux->dst_trampoline is NULL, the program has already been
3388         *   attached to a target and its initial target was cleared (below)
3389	 *
3390	 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3391	 *   target_btf_id using the link_create API.
3392	 *
3393	 * - if tgt_prog == NULL when this function was called using the old
3394	 *   raw_tracepoint_open API, and we need a target from prog->aux
3395	 *
3396	 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3397	 *   was detached and is going for re-attachment.
3398	 *
3399	 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3400	 *   are NULL, then program was already attached and user did not provide
3401	 *   tgt_prog_fd so we have no way to find out or create trampoline
3402	 */
3403	if (!prog->aux->dst_trampoline && !tgt_prog) {
3404		/*
3405		 * Allow re-attach for TRACING and LSM programs. If it's
3406		 * currently linked, bpf_trampoline_link_prog will fail.
3407		 * EXT programs need to specify tgt_prog_fd, so they
3408		 * re-attach in separate code path.
3409		 */
3410		if (prog->type != BPF_PROG_TYPE_TRACING &&
3411		    prog->type != BPF_PROG_TYPE_LSM) {
3412			err = -EINVAL;
3413			goto out_unlock;
3414		}
3415		/* We can allow re-attach only if we have valid attach_btf. */
3416		if (!prog->aux->attach_btf) {
3417			err = -EINVAL;
3418			goto out_unlock;
3419		}
3420		btf_id = prog->aux->attach_btf_id;
3421		key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3422	}
3423
3424	if (!prog->aux->dst_trampoline ||
3425	    (key && key != prog->aux->dst_trampoline->key)) {
3426		/* If there is no saved target, or the specified target is
3427		 * different from the destination specified at load time, we
3428		 * need a new trampoline and a check for compatibility
3429		 */
3430		struct bpf_attach_target_info tgt_info = {};
3431
3432		err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3433					      &tgt_info);
3434		if (err)
3435			goto out_unlock;
3436
3437		if (tgt_info.tgt_mod) {
3438			module_put(prog->aux->mod);
3439			prog->aux->mod = tgt_info.tgt_mod;
3440		}
3441
3442		tr = bpf_trampoline_get(key, &tgt_info);
3443		if (!tr) {
3444			err = -ENOMEM;
3445			goto out_unlock;
3446		}
3447	} else {
3448		/* The caller didn't specify a target, or the target was the
3449		 * same as the destination supplied during program load. This
3450		 * means we can reuse the trampoline and reference from program
3451		 * load time, and there is no need to allocate a new one. This
3452		 * can only happen once for any program, as the saved values in
3453		 * prog->aux are cleared below.
3454		 */
3455		tr = prog->aux->dst_trampoline;
3456		tgt_prog = prog->aux->dst_prog;
3457	}
3458
3459	err = bpf_link_prime(&link->link.link, &link_primer);
3460	if (err)
3461		goto out_unlock;
3462
3463	err = bpf_trampoline_link_prog(&link->link, tr);
3464	if (err) {
3465		bpf_link_cleanup(&link_primer);
3466		link = NULL;
3467		goto out_unlock;
3468	}
3469
3470	link->tgt_prog = tgt_prog;
3471	link->trampoline = tr;
3472
3473	/* Always clear the trampoline and target prog from prog->aux to make
3474	 * sure the original attach destination is not kept alive after a
3475	 * program is (re-)attached to another target.
3476	 */
3477	if (prog->aux->dst_prog &&
3478	    (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3479		/* got extra prog ref from syscall, or attaching to different prog */
3480		bpf_prog_put(prog->aux->dst_prog);
3481	if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3482		/* we allocated a new trampoline, so free the old one */
3483		bpf_trampoline_put(prog->aux->dst_trampoline);
3484
3485	prog->aux->dst_prog = NULL;
3486	prog->aux->dst_trampoline = NULL;
3487	mutex_unlock(&prog->aux->dst_mutex);
3488
3489	return bpf_link_settle(&link_primer);
3490out_unlock:
3491	if (tr && tr != prog->aux->dst_trampoline)
3492		bpf_trampoline_put(tr);
3493	mutex_unlock(&prog->aux->dst_mutex);
3494	kfree(link);
3495out_put_prog:
3496	if (tgt_prog_fd && tgt_prog)
3497		bpf_prog_put(tgt_prog);
3498	return err;
3499}
3500
3501struct bpf_raw_tp_link {
3502	struct bpf_link link;
3503	struct bpf_raw_event_map *btp;
3504};
3505
3506static void bpf_raw_tp_link_release(struct bpf_link *link)
3507{
3508	struct bpf_raw_tp_link *raw_tp =
3509		container_of(link, struct bpf_raw_tp_link, link);
3510
3511	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3512	bpf_put_raw_tracepoint(raw_tp->btp);
3513}
3514
3515static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3516{
3517	struct bpf_raw_tp_link *raw_tp =
3518		container_of(link, struct bpf_raw_tp_link, link);
3519
3520	kfree(raw_tp);
3521}
3522
3523static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3524					struct seq_file *seq)
3525{
3526	struct bpf_raw_tp_link *raw_tp_link =
3527		container_of(link, struct bpf_raw_tp_link, link);
3528
3529	seq_printf(seq,
3530		   "tp_name:\t%s\n",
3531		   raw_tp_link->btp->tp->name);
3532}
3533
3534static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3535			    u32 len)
3536{
3537	if (ulen >= len + 1) {
3538		if (copy_to_user(ubuf, buf, len + 1))
3539			return -EFAULT;
3540	} else {
3541		char zero = '\0';
3542
3543		if (copy_to_user(ubuf, buf, ulen - 1))
3544			return -EFAULT;
3545		if (put_user(zero, ubuf + ulen - 1))
3546			return -EFAULT;
3547		return -ENOSPC;
3548	}
3549
3550	return 0;
3551}
3552
3553static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3554					  struct bpf_link_info *info)
3555{
3556	struct bpf_raw_tp_link *raw_tp_link =
3557		container_of(link, struct bpf_raw_tp_link, link);
3558	char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3559	const char *tp_name = raw_tp_link->btp->tp->name;
3560	u32 ulen = info->raw_tracepoint.tp_name_len;
3561	size_t tp_len = strlen(tp_name);
3562
3563	if (!ulen ^ !ubuf)
3564		return -EINVAL;
3565
3566	info->raw_tracepoint.tp_name_len = tp_len + 1;
3567
3568	if (!ubuf)
3569		return 0;
3570
3571	return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
 
 
 
 
 
 
 
 
 
 
 
 
 
3572}
3573
3574static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3575	.release = bpf_raw_tp_link_release,
3576	.dealloc_deferred = bpf_raw_tp_link_dealloc,
3577	.show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3578	.fill_link_info = bpf_raw_tp_link_fill_link_info,
3579};
3580
3581#ifdef CONFIG_PERF_EVENTS
3582struct bpf_perf_link {
3583	struct bpf_link link;
3584	struct file *perf_file;
3585};
3586
3587static void bpf_perf_link_release(struct bpf_link *link)
3588{
3589	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3590	struct perf_event *event = perf_link->perf_file->private_data;
3591
3592	perf_event_free_bpf_prog(event);
3593	fput(perf_link->perf_file);
3594}
3595
3596static void bpf_perf_link_dealloc(struct bpf_link *link)
3597{
3598	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3599
3600	kfree(perf_link);
3601}
3602
3603static int bpf_perf_link_fill_common(const struct perf_event *event,
3604				     char __user *uname, u32 ulen,
3605				     u64 *probe_offset, u64 *probe_addr,
3606				     u32 *fd_type, unsigned long *missed)
3607{
3608	const char *buf;
3609	u32 prog_id;
3610	size_t len;
3611	int err;
3612
3613	if (!ulen ^ !uname)
3614		return -EINVAL;
3615
3616	err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3617				      probe_offset, probe_addr, missed);
3618	if (err)
3619		return err;
3620	if (!uname)
3621		return 0;
3622	if (buf) {
3623		len = strlen(buf);
3624		err = bpf_copy_to_user(uname, buf, ulen, len);
3625		if (err)
3626			return err;
3627	} else {
3628		char zero = '\0';
3629
3630		if (put_user(zero, uname))
3631			return -EFAULT;
3632	}
3633	return 0;
3634}
3635
3636#ifdef CONFIG_KPROBE_EVENTS
3637static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3638				     struct bpf_link_info *info)
3639{
3640	unsigned long missed;
3641	char __user *uname;
3642	u64 addr, offset;
3643	u32 ulen, type;
3644	int err;
3645
3646	uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3647	ulen = info->perf_event.kprobe.name_len;
3648	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3649					&type, &missed);
3650	if (err)
3651		return err;
3652	if (type == BPF_FD_TYPE_KRETPROBE)
3653		info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3654	else
3655		info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3656
3657	info->perf_event.kprobe.offset = offset;
3658	info->perf_event.kprobe.missed = missed;
3659	if (!kallsyms_show_value(current_cred()))
3660		addr = 0;
3661	info->perf_event.kprobe.addr = addr;
3662	info->perf_event.kprobe.cookie = event->bpf_cookie;
3663	return 0;
3664}
3665#endif
3666
3667#ifdef CONFIG_UPROBE_EVENTS
3668static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3669				     struct bpf_link_info *info)
3670{
3671	char __user *uname;
3672	u64 addr, offset;
3673	u32 ulen, type;
3674	int err;
3675
3676	uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3677	ulen = info->perf_event.uprobe.name_len;
3678	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3679					&type, NULL);
3680	if (err)
3681		return err;
3682
3683	if (type == BPF_FD_TYPE_URETPROBE)
3684		info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3685	else
3686		info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3687	info->perf_event.uprobe.offset = offset;
3688	info->perf_event.uprobe.cookie = event->bpf_cookie;
3689	return 0;
3690}
3691#endif
3692
3693static int bpf_perf_link_fill_probe(const struct perf_event *event,
3694				    struct bpf_link_info *info)
3695{
3696#ifdef CONFIG_KPROBE_EVENTS
3697	if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3698		return bpf_perf_link_fill_kprobe(event, info);
3699#endif
3700#ifdef CONFIG_UPROBE_EVENTS
3701	if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3702		return bpf_perf_link_fill_uprobe(event, info);
3703#endif
3704	return -EOPNOTSUPP;
3705}
3706
3707static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3708					 struct bpf_link_info *info)
3709{
3710	char __user *uname;
3711	u32 ulen;
3712
3713	uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3714	ulen = info->perf_event.tracepoint.name_len;
3715	info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3716	info->perf_event.tracepoint.cookie = event->bpf_cookie;
3717	return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3718}
3719
3720static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3721					 struct bpf_link_info *info)
3722{
3723	info->perf_event.event.type = event->attr.type;
3724	info->perf_event.event.config = event->attr.config;
3725	info->perf_event.event.cookie = event->bpf_cookie;
3726	info->perf_event.type = BPF_PERF_EVENT_EVENT;
3727	return 0;
3728}
3729
3730static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3731					struct bpf_link_info *info)
3732{
3733	struct bpf_perf_link *perf_link;
3734	const struct perf_event *event;
3735
3736	perf_link = container_of(link, struct bpf_perf_link, link);
3737	event = perf_get_event(perf_link->perf_file);
3738	if (IS_ERR(event))
3739		return PTR_ERR(event);
3740
3741	switch (event->prog->type) {
3742	case BPF_PROG_TYPE_PERF_EVENT:
3743		return bpf_perf_link_fill_perf_event(event, info);
3744	case BPF_PROG_TYPE_TRACEPOINT:
3745		return bpf_perf_link_fill_tracepoint(event, info);
3746	case BPF_PROG_TYPE_KPROBE:
3747		return bpf_perf_link_fill_probe(event, info);
3748	default:
3749		return -EOPNOTSUPP;
3750	}
3751}
3752
3753static const struct bpf_link_ops bpf_perf_link_lops = {
3754	.release = bpf_perf_link_release,
3755	.dealloc = bpf_perf_link_dealloc,
3756	.fill_link_info = bpf_perf_link_fill_link_info,
3757};
3758
3759static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3760{
3761	struct bpf_link_primer link_primer;
3762	struct bpf_perf_link *link;
3763	struct perf_event *event;
3764	struct file *perf_file;
3765	int err;
3766
3767	if (attr->link_create.flags)
3768		return -EINVAL;
3769
3770	perf_file = perf_event_get(attr->link_create.target_fd);
3771	if (IS_ERR(perf_file))
3772		return PTR_ERR(perf_file);
3773
3774	link = kzalloc(sizeof(*link), GFP_USER);
3775	if (!link) {
3776		err = -ENOMEM;
3777		goto out_put_file;
3778	}
3779	bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3780	link->perf_file = perf_file;
3781
3782	err = bpf_link_prime(&link->link, &link_primer);
3783	if (err) {
3784		kfree(link);
3785		goto out_put_file;
3786	}
3787
3788	event = perf_file->private_data;
3789	err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3790	if (err) {
3791		bpf_link_cleanup(&link_primer);
3792		goto out_put_file;
3793	}
3794	/* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3795	bpf_prog_inc(prog);
3796
3797	return bpf_link_settle(&link_primer);
3798
3799out_put_file:
3800	fput(perf_file);
3801	return err;
3802}
3803#else
3804static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3805{
3806	return -EOPNOTSUPP;
3807}
3808#endif /* CONFIG_PERF_EVENTS */
3809
3810static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3811				  const char __user *user_tp_name)
3812{
3813	struct bpf_link_primer link_primer;
3814	struct bpf_raw_tp_link *link;
3815	struct bpf_raw_event_map *btp;
3816	const char *tp_name;
3817	char buf[128];
3818	int err;
3819
3820	switch (prog->type) {
3821	case BPF_PROG_TYPE_TRACING:
3822	case BPF_PROG_TYPE_EXT:
3823	case BPF_PROG_TYPE_LSM:
3824		if (user_tp_name)
3825			/* The attach point for this category of programs
3826			 * should be specified via btf_id during program load.
3827			 */
3828			return -EINVAL;
3829		if (prog->type == BPF_PROG_TYPE_TRACING &&
3830		    prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3831			tp_name = prog->aux->attach_func_name;
3832			break;
3833		}
3834		return bpf_tracing_prog_attach(prog, 0, 0, 0);
3835	case BPF_PROG_TYPE_RAW_TRACEPOINT:
3836	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3837		if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3838			return -EFAULT;
3839		buf[sizeof(buf) - 1] = 0;
3840		tp_name = buf;
3841		break;
3842	default:
3843		return -EINVAL;
3844	}
3845
3846	btp = bpf_get_raw_tracepoint(tp_name);
3847	if (!btp)
3848		return -ENOENT;
3849
3850	link = kzalloc(sizeof(*link), GFP_USER);
3851	if (!link) {
3852		err = -ENOMEM;
3853		goto out_put_btp;
3854	}
3855	bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3856		      &bpf_raw_tp_link_lops, prog);
3857	link->btp = btp;
3858
3859	err = bpf_link_prime(&link->link, &link_primer);
3860	if (err) {
3861		kfree(link);
3862		goto out_put_btp;
3863	}
3864
3865	err = bpf_probe_register(link->btp, prog);
3866	if (err) {
3867		bpf_link_cleanup(&link_primer);
3868		goto out_put_btp;
3869	}
3870
3871	return bpf_link_settle(&link_primer);
3872
3873out_put_btp:
3874	bpf_put_raw_tracepoint(btp);
3875	return err;
3876}
3877
3878#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3879
3880static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3881{
3882	struct bpf_prog *prog;
3883	int fd;
3884
3885	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3886		return -EINVAL;
3887
3888	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3889	if (IS_ERR(prog))
3890		return PTR_ERR(prog);
3891
3892	fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3893	if (fd < 0)
3894		bpf_prog_put(prog);
3895	return fd;
3896}
3897
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3898static enum bpf_prog_type
3899attach_type_to_prog_type(enum bpf_attach_type attach_type)
3900{
3901	switch (attach_type) {
3902	case BPF_CGROUP_INET_INGRESS:
3903	case BPF_CGROUP_INET_EGRESS:
3904		return BPF_PROG_TYPE_CGROUP_SKB;
3905	case BPF_CGROUP_INET_SOCK_CREATE:
3906	case BPF_CGROUP_INET_SOCK_RELEASE:
3907	case BPF_CGROUP_INET4_POST_BIND:
3908	case BPF_CGROUP_INET6_POST_BIND:
3909		return BPF_PROG_TYPE_CGROUP_SOCK;
3910	case BPF_CGROUP_INET4_BIND:
3911	case BPF_CGROUP_INET6_BIND:
3912	case BPF_CGROUP_INET4_CONNECT:
3913	case BPF_CGROUP_INET6_CONNECT:
3914	case BPF_CGROUP_UNIX_CONNECT:
3915	case BPF_CGROUP_INET4_GETPEERNAME:
3916	case BPF_CGROUP_INET6_GETPEERNAME:
3917	case BPF_CGROUP_UNIX_GETPEERNAME:
3918	case BPF_CGROUP_INET4_GETSOCKNAME:
3919	case BPF_CGROUP_INET6_GETSOCKNAME:
3920	case BPF_CGROUP_UNIX_GETSOCKNAME:
3921	case BPF_CGROUP_UDP4_SENDMSG:
3922	case BPF_CGROUP_UDP6_SENDMSG:
3923	case BPF_CGROUP_UNIX_SENDMSG:
3924	case BPF_CGROUP_UDP4_RECVMSG:
3925	case BPF_CGROUP_UDP6_RECVMSG:
3926	case BPF_CGROUP_UNIX_RECVMSG:
3927		return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3928	case BPF_CGROUP_SOCK_OPS:
3929		return BPF_PROG_TYPE_SOCK_OPS;
3930	case BPF_CGROUP_DEVICE:
3931		return BPF_PROG_TYPE_CGROUP_DEVICE;
3932	case BPF_SK_MSG_VERDICT:
3933		return BPF_PROG_TYPE_SK_MSG;
3934	case BPF_SK_SKB_STREAM_PARSER:
3935	case BPF_SK_SKB_STREAM_VERDICT:
3936	case BPF_SK_SKB_VERDICT:
3937		return BPF_PROG_TYPE_SK_SKB;
3938	case BPF_LIRC_MODE2:
3939		return BPF_PROG_TYPE_LIRC_MODE2;
3940	case BPF_FLOW_DISSECTOR:
3941		return BPF_PROG_TYPE_FLOW_DISSECTOR;
3942	case BPF_CGROUP_SYSCTL:
3943		return BPF_PROG_TYPE_CGROUP_SYSCTL;
3944	case BPF_CGROUP_GETSOCKOPT:
3945	case BPF_CGROUP_SETSOCKOPT:
3946		return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3947	case BPF_TRACE_ITER:
3948	case BPF_TRACE_RAW_TP:
3949	case BPF_TRACE_FENTRY:
3950	case BPF_TRACE_FEXIT:
3951	case BPF_MODIFY_RETURN:
3952		return BPF_PROG_TYPE_TRACING;
3953	case BPF_LSM_MAC:
3954		return BPF_PROG_TYPE_LSM;
3955	case BPF_SK_LOOKUP:
3956		return BPF_PROG_TYPE_SK_LOOKUP;
3957	case BPF_XDP:
3958		return BPF_PROG_TYPE_XDP;
3959	case BPF_LSM_CGROUP:
3960		return BPF_PROG_TYPE_LSM;
3961	case BPF_TCX_INGRESS:
3962	case BPF_TCX_EGRESS:
3963	case BPF_NETKIT_PRIMARY:
3964	case BPF_NETKIT_PEER:
3965		return BPF_PROG_TYPE_SCHED_CLS;
3966	default:
3967		return BPF_PROG_TYPE_UNSPEC;
3968	}
3969}
3970
3971static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3972					     enum bpf_attach_type attach_type)
3973{
3974	enum bpf_prog_type ptype;
3975
3976	switch (prog->type) {
3977	case BPF_PROG_TYPE_CGROUP_SOCK:
3978	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3979	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3980	case BPF_PROG_TYPE_SK_LOOKUP:
3981		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3982	case BPF_PROG_TYPE_CGROUP_SKB:
3983		if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
3984			/* cg-skb progs can be loaded by unpriv user.
3985			 * check permissions at attach time.
3986			 */
3987			return -EPERM;
3988
3989		ptype = attach_type_to_prog_type(attach_type);
3990		if (prog->type != ptype)
3991			return -EINVAL;
3992
3993		return prog->enforce_expected_attach_type &&
3994			prog->expected_attach_type != attach_type ?
3995			-EINVAL : 0;
3996	case BPF_PROG_TYPE_EXT:
3997		return 0;
3998	case BPF_PROG_TYPE_NETFILTER:
3999		if (attach_type != BPF_NETFILTER)
4000			return -EINVAL;
4001		return 0;
4002	case BPF_PROG_TYPE_PERF_EVENT:
4003	case BPF_PROG_TYPE_TRACEPOINT:
4004		if (attach_type != BPF_PERF_EVENT)
4005			return -EINVAL;
4006		return 0;
4007	case BPF_PROG_TYPE_KPROBE:
4008		if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4009		    attach_type != BPF_TRACE_KPROBE_MULTI)
4010			return -EINVAL;
4011		if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4012		    attach_type != BPF_TRACE_UPROBE_MULTI)
4013			return -EINVAL;
4014		if (attach_type != BPF_PERF_EVENT &&
4015		    attach_type != BPF_TRACE_KPROBE_MULTI &&
4016		    attach_type != BPF_TRACE_UPROBE_MULTI)
4017			return -EINVAL;
4018		return 0;
4019	case BPF_PROG_TYPE_SCHED_CLS:
4020		if (attach_type != BPF_TCX_INGRESS &&
4021		    attach_type != BPF_TCX_EGRESS &&
4022		    attach_type != BPF_NETKIT_PRIMARY &&
4023		    attach_type != BPF_NETKIT_PEER)
4024			return -EINVAL;
4025		return 0;
4026	default:
4027		ptype = attach_type_to_prog_type(attach_type);
4028		if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4029			return -EINVAL;
4030		return 0;
4031	}
4032}
4033
4034#define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4035
4036#define BPF_F_ATTACH_MASK_BASE	\
4037	(BPF_F_ALLOW_OVERRIDE |	\
4038	 BPF_F_ALLOW_MULTI |	\
4039	 BPF_F_REPLACE)
4040
4041#define BPF_F_ATTACH_MASK_MPROG	\
4042	(BPF_F_REPLACE |	\
4043	 BPF_F_BEFORE |		\
4044	 BPF_F_AFTER |		\
4045	 BPF_F_ID |		\
4046	 BPF_F_LINK)
4047
4048static int bpf_prog_attach(const union bpf_attr *attr)
4049{
4050	enum bpf_prog_type ptype;
4051	struct bpf_prog *prog;
4052	int ret;
4053
4054	if (CHECK_ATTR(BPF_PROG_ATTACH))
4055		return -EINVAL;
4056
 
 
 
4057	ptype = attach_type_to_prog_type(attr->attach_type);
4058	if (ptype == BPF_PROG_TYPE_UNSPEC)
4059		return -EINVAL;
4060	if (bpf_mprog_supported(ptype)) {
4061		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4062			return -EINVAL;
4063	} else {
4064		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4065			return -EINVAL;
4066		if (attr->relative_fd ||
4067		    attr->expected_revision)
4068			return -EINVAL;
4069	}
4070
4071	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4072	if (IS_ERR(prog))
4073		return PTR_ERR(prog);
4074
4075	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4076		bpf_prog_put(prog);
4077		return -EINVAL;
4078	}
4079
4080	switch (ptype) {
4081	case BPF_PROG_TYPE_SK_SKB:
4082	case BPF_PROG_TYPE_SK_MSG:
4083		ret = sock_map_get_from_fd(attr, prog);
4084		break;
4085	case BPF_PROG_TYPE_LIRC_MODE2:
4086		ret = lirc_prog_attach(attr, prog);
4087		break;
4088	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4089		ret = netns_bpf_prog_attach(attr, prog);
4090		break;
4091	case BPF_PROG_TYPE_CGROUP_DEVICE:
4092	case BPF_PROG_TYPE_CGROUP_SKB:
4093	case BPF_PROG_TYPE_CGROUP_SOCK:
4094	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4095	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4096	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4097	case BPF_PROG_TYPE_SOCK_OPS:
4098	case BPF_PROG_TYPE_LSM:
4099		if (ptype == BPF_PROG_TYPE_LSM &&
4100		    prog->expected_attach_type != BPF_LSM_CGROUP)
4101			ret = -EINVAL;
4102		else
4103			ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4104		break;
4105	case BPF_PROG_TYPE_SCHED_CLS:
4106		if (attr->attach_type == BPF_TCX_INGRESS ||
4107		    attr->attach_type == BPF_TCX_EGRESS)
4108			ret = tcx_prog_attach(attr, prog);
4109		else
4110			ret = netkit_prog_attach(attr, prog);
4111		break;
4112	default:
4113		ret = -EINVAL;
4114	}
4115
4116	if (ret)
4117		bpf_prog_put(prog);
4118	return ret;
4119}
4120
4121#define BPF_PROG_DETACH_LAST_FIELD expected_revision
4122
4123static int bpf_prog_detach(const union bpf_attr *attr)
4124{
4125	struct bpf_prog *prog = NULL;
4126	enum bpf_prog_type ptype;
4127	int ret;
4128
4129	if (CHECK_ATTR(BPF_PROG_DETACH))
4130		return -EINVAL;
4131
4132	ptype = attach_type_to_prog_type(attr->attach_type);
4133	if (bpf_mprog_supported(ptype)) {
4134		if (ptype == BPF_PROG_TYPE_UNSPEC)
4135			return -EINVAL;
4136		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4137			return -EINVAL;
4138		if (attr->attach_bpf_fd) {
4139			prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4140			if (IS_ERR(prog))
4141				return PTR_ERR(prog);
4142		}
4143	} else if (attr->attach_flags ||
4144		   attr->relative_fd ||
4145		   attr->expected_revision) {
4146		return -EINVAL;
4147	}
4148
4149	switch (ptype) {
4150	case BPF_PROG_TYPE_SK_MSG:
4151	case BPF_PROG_TYPE_SK_SKB:
4152		ret = sock_map_prog_detach(attr, ptype);
4153		break;
4154	case BPF_PROG_TYPE_LIRC_MODE2:
4155		ret = lirc_prog_detach(attr);
4156		break;
4157	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4158		ret = netns_bpf_prog_detach(attr, ptype);
4159		break;
4160	case BPF_PROG_TYPE_CGROUP_DEVICE:
4161	case BPF_PROG_TYPE_CGROUP_SKB:
4162	case BPF_PROG_TYPE_CGROUP_SOCK:
4163	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4164	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4165	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4166	case BPF_PROG_TYPE_SOCK_OPS:
4167	case BPF_PROG_TYPE_LSM:
4168		ret = cgroup_bpf_prog_detach(attr, ptype);
4169		break;
4170	case BPF_PROG_TYPE_SCHED_CLS:
4171		if (attr->attach_type == BPF_TCX_INGRESS ||
4172		    attr->attach_type == BPF_TCX_EGRESS)
4173			ret = tcx_prog_detach(attr, prog);
4174		else
4175			ret = netkit_prog_detach(attr, prog);
4176		break;
4177	default:
4178		ret = -EINVAL;
4179	}
4180
4181	if (prog)
4182		bpf_prog_put(prog);
4183	return ret;
4184}
4185
4186#define BPF_PROG_QUERY_LAST_FIELD query.revision
4187
4188static int bpf_prog_query(const union bpf_attr *attr,
4189			  union bpf_attr __user *uattr)
4190{
4191	if (!bpf_net_capable())
4192		return -EPERM;
4193	if (CHECK_ATTR(BPF_PROG_QUERY))
4194		return -EINVAL;
4195	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4196		return -EINVAL;
4197
4198	switch (attr->query.attach_type) {
4199	case BPF_CGROUP_INET_INGRESS:
4200	case BPF_CGROUP_INET_EGRESS:
4201	case BPF_CGROUP_INET_SOCK_CREATE:
4202	case BPF_CGROUP_INET_SOCK_RELEASE:
4203	case BPF_CGROUP_INET4_BIND:
4204	case BPF_CGROUP_INET6_BIND:
4205	case BPF_CGROUP_INET4_POST_BIND:
4206	case BPF_CGROUP_INET6_POST_BIND:
4207	case BPF_CGROUP_INET4_CONNECT:
4208	case BPF_CGROUP_INET6_CONNECT:
4209	case BPF_CGROUP_UNIX_CONNECT:
4210	case BPF_CGROUP_INET4_GETPEERNAME:
4211	case BPF_CGROUP_INET6_GETPEERNAME:
4212	case BPF_CGROUP_UNIX_GETPEERNAME:
4213	case BPF_CGROUP_INET4_GETSOCKNAME:
4214	case BPF_CGROUP_INET6_GETSOCKNAME:
4215	case BPF_CGROUP_UNIX_GETSOCKNAME:
4216	case BPF_CGROUP_UDP4_SENDMSG:
4217	case BPF_CGROUP_UDP6_SENDMSG:
4218	case BPF_CGROUP_UNIX_SENDMSG:
4219	case BPF_CGROUP_UDP4_RECVMSG:
4220	case BPF_CGROUP_UDP6_RECVMSG:
4221	case BPF_CGROUP_UNIX_RECVMSG:
4222	case BPF_CGROUP_SOCK_OPS:
4223	case BPF_CGROUP_DEVICE:
4224	case BPF_CGROUP_SYSCTL:
4225	case BPF_CGROUP_GETSOCKOPT:
4226	case BPF_CGROUP_SETSOCKOPT:
4227	case BPF_LSM_CGROUP:
4228		return cgroup_bpf_prog_query(attr, uattr);
4229	case BPF_LIRC_MODE2:
4230		return lirc_prog_query(attr, uattr);
4231	case BPF_FLOW_DISSECTOR:
4232	case BPF_SK_LOOKUP:
4233		return netns_bpf_prog_query(attr, uattr);
4234	case BPF_SK_SKB_STREAM_PARSER:
4235	case BPF_SK_SKB_STREAM_VERDICT:
4236	case BPF_SK_MSG_VERDICT:
4237	case BPF_SK_SKB_VERDICT:
4238		return sock_map_bpf_prog_query(attr, uattr);
4239	case BPF_TCX_INGRESS:
4240	case BPF_TCX_EGRESS:
4241		return tcx_prog_query(attr, uattr);
4242	case BPF_NETKIT_PRIMARY:
4243	case BPF_NETKIT_PEER:
4244		return netkit_prog_query(attr, uattr);
4245	default:
4246		return -EINVAL;
4247	}
4248}
4249
4250#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4251
4252static int bpf_prog_test_run(const union bpf_attr *attr,
4253			     union bpf_attr __user *uattr)
4254{
4255	struct bpf_prog *prog;
4256	int ret = -ENOTSUPP;
4257
4258	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4259		return -EINVAL;
4260
4261	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4262	    (!attr->test.ctx_size_in && attr->test.ctx_in))
4263		return -EINVAL;
4264
4265	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4266	    (!attr->test.ctx_size_out && attr->test.ctx_out))
4267		return -EINVAL;
4268
4269	prog = bpf_prog_get(attr->test.prog_fd);
4270	if (IS_ERR(prog))
4271		return PTR_ERR(prog);
4272
4273	if (prog->aux->ops->test_run)
4274		ret = prog->aux->ops->test_run(prog, attr, uattr);
4275
4276	bpf_prog_put(prog);
4277	return ret;
4278}
4279
4280#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4281
4282static int bpf_obj_get_next_id(const union bpf_attr *attr,
4283			       union bpf_attr __user *uattr,
4284			       struct idr *idr,
4285			       spinlock_t *lock)
4286{
4287	u32 next_id = attr->start_id;
4288	int err = 0;
4289
4290	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4291		return -EINVAL;
4292
4293	if (!capable(CAP_SYS_ADMIN))
4294		return -EPERM;
4295
4296	next_id++;
4297	spin_lock_bh(lock);
4298	if (!idr_get_next(idr, &next_id))
4299		err = -ENOENT;
4300	spin_unlock_bh(lock);
4301
4302	if (!err)
4303		err = put_user(next_id, &uattr->next_id);
4304
4305	return err;
4306}
4307
4308struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4309{
4310	struct bpf_map *map;
4311
4312	spin_lock_bh(&map_idr_lock);
4313again:
4314	map = idr_get_next(&map_idr, id);
4315	if (map) {
4316		map = __bpf_map_inc_not_zero(map, false);
4317		if (IS_ERR(map)) {
4318			(*id)++;
4319			goto again;
4320		}
4321	}
4322	spin_unlock_bh(&map_idr_lock);
4323
4324	return map;
4325}
4326
4327struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4328{
4329	struct bpf_prog *prog;
4330
4331	spin_lock_bh(&prog_idr_lock);
4332again:
4333	prog = idr_get_next(&prog_idr, id);
4334	if (prog) {
4335		prog = bpf_prog_inc_not_zero(prog);
4336		if (IS_ERR(prog)) {
4337			(*id)++;
4338			goto again;
4339		}
4340	}
4341	spin_unlock_bh(&prog_idr_lock);
4342
4343	return prog;
4344}
4345
4346#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4347
4348struct bpf_prog *bpf_prog_by_id(u32 id)
4349{
4350	struct bpf_prog *prog;
4351
4352	if (!id)
4353		return ERR_PTR(-ENOENT);
4354
4355	spin_lock_bh(&prog_idr_lock);
4356	prog = idr_find(&prog_idr, id);
4357	if (prog)
4358		prog = bpf_prog_inc_not_zero(prog);
4359	else
4360		prog = ERR_PTR(-ENOENT);
4361	spin_unlock_bh(&prog_idr_lock);
4362	return prog;
4363}
4364
4365static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4366{
4367	struct bpf_prog *prog;
4368	u32 id = attr->prog_id;
4369	int fd;
4370
4371	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4372		return -EINVAL;
4373
4374	if (!capable(CAP_SYS_ADMIN))
4375		return -EPERM;
4376
4377	prog = bpf_prog_by_id(id);
4378	if (IS_ERR(prog))
4379		return PTR_ERR(prog);
4380
4381	fd = bpf_prog_new_fd(prog);
4382	if (fd < 0)
4383		bpf_prog_put(prog);
4384
4385	return fd;
4386}
4387
4388#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4389
4390static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4391{
4392	struct bpf_map *map;
4393	u32 id = attr->map_id;
4394	int f_flags;
4395	int fd;
4396
4397	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4398	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4399		return -EINVAL;
4400
4401	if (!capable(CAP_SYS_ADMIN))
4402		return -EPERM;
4403
4404	f_flags = bpf_get_file_flag(attr->open_flags);
4405	if (f_flags < 0)
4406		return f_flags;
4407
4408	spin_lock_bh(&map_idr_lock);
4409	map = idr_find(&map_idr, id);
4410	if (map)
4411		map = __bpf_map_inc_not_zero(map, true);
4412	else
4413		map = ERR_PTR(-ENOENT);
4414	spin_unlock_bh(&map_idr_lock);
4415
4416	if (IS_ERR(map))
4417		return PTR_ERR(map);
4418
4419	fd = bpf_map_new_fd(map, f_flags);
4420	if (fd < 0)
4421		bpf_map_put_with_uref(map);
4422
4423	return fd;
4424}
4425
4426static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4427					      unsigned long addr, u32 *off,
4428					      u32 *type)
4429{
4430	const struct bpf_map *map;
4431	int i;
4432
4433	mutex_lock(&prog->aux->used_maps_mutex);
4434	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4435		map = prog->aux->used_maps[i];
4436		if (map == (void *)addr) {
4437			*type = BPF_PSEUDO_MAP_FD;
4438			goto out;
4439		}
4440		if (!map->ops->map_direct_value_meta)
4441			continue;
4442		if (!map->ops->map_direct_value_meta(map, addr, off)) {
4443			*type = BPF_PSEUDO_MAP_VALUE;
4444			goto out;
4445		}
4446	}
4447	map = NULL;
4448
4449out:
4450	mutex_unlock(&prog->aux->used_maps_mutex);
4451	return map;
4452}
4453
4454static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4455					      const struct cred *f_cred)
4456{
4457	const struct bpf_map *map;
4458	struct bpf_insn *insns;
4459	u32 off, type;
4460	u64 imm;
4461	u8 code;
4462	int i;
4463
4464	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4465			GFP_USER);
4466	if (!insns)
4467		return insns;
4468
4469	for (i = 0; i < prog->len; i++) {
4470		code = insns[i].code;
4471
4472		if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4473			insns[i].code = BPF_JMP | BPF_CALL;
4474			insns[i].imm = BPF_FUNC_tail_call;
4475			/* fall-through */
4476		}
4477		if (code == (BPF_JMP | BPF_CALL) ||
4478		    code == (BPF_JMP | BPF_CALL_ARGS)) {
4479			if (code == (BPF_JMP | BPF_CALL_ARGS))
4480				insns[i].code = BPF_JMP | BPF_CALL;
4481			if (!bpf_dump_raw_ok(f_cred))
4482				insns[i].imm = 0;
4483			continue;
4484		}
4485		if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4486			insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4487			continue;
4488		}
4489
4490		if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4491		     BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4492			insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4493			continue;
4494		}
4495
4496		if (code != (BPF_LD | BPF_IMM | BPF_DW))
4497			continue;
4498
4499		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4500		map = bpf_map_from_imm(prog, imm, &off, &type);
4501		if (map) {
4502			insns[i].src_reg = type;
4503			insns[i].imm = map->id;
4504			insns[i + 1].imm = off;
4505			continue;
4506		}
4507	}
4508
4509	return insns;
4510}
4511
4512static int set_info_rec_size(struct bpf_prog_info *info)
4513{
4514	/*
4515	 * Ensure info.*_rec_size is the same as kernel expected size
4516	 *
4517	 * or
4518	 *
4519	 * Only allow zero *_rec_size if both _rec_size and _cnt are
4520	 * zero.  In this case, the kernel will set the expected
4521	 * _rec_size back to the info.
4522	 */
4523
4524	if ((info->nr_func_info || info->func_info_rec_size) &&
4525	    info->func_info_rec_size != sizeof(struct bpf_func_info))
4526		return -EINVAL;
4527
4528	if ((info->nr_line_info || info->line_info_rec_size) &&
4529	    info->line_info_rec_size != sizeof(struct bpf_line_info))
4530		return -EINVAL;
4531
4532	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4533	    info->jited_line_info_rec_size != sizeof(__u64))
4534		return -EINVAL;
4535
4536	info->func_info_rec_size = sizeof(struct bpf_func_info);
4537	info->line_info_rec_size = sizeof(struct bpf_line_info);
4538	info->jited_line_info_rec_size = sizeof(__u64);
4539
4540	return 0;
4541}
4542
4543static int bpf_prog_get_info_by_fd(struct file *file,
4544				   struct bpf_prog *prog,
4545				   const union bpf_attr *attr,
4546				   union bpf_attr __user *uattr)
4547{
4548	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4549	struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4550	struct bpf_prog_info info;
4551	u32 info_len = attr->info.info_len;
4552	struct bpf_prog_kstats stats;
4553	char __user *uinsns;
4554	u32 ulen;
4555	int err;
4556
4557	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4558	if (err)
4559		return err;
4560	info_len = min_t(u32, sizeof(info), info_len);
4561
4562	memset(&info, 0, sizeof(info));
4563	if (copy_from_user(&info, uinfo, info_len))
4564		return -EFAULT;
4565
4566	info.type = prog->type;
4567	info.id = prog->aux->id;
4568	info.load_time = prog->aux->load_time;
4569	info.created_by_uid = from_kuid_munged(current_user_ns(),
4570					       prog->aux->user->uid);
4571	info.gpl_compatible = prog->gpl_compatible;
4572
4573	memcpy(info.tag, prog->tag, sizeof(prog->tag));
4574	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4575
4576	mutex_lock(&prog->aux->used_maps_mutex);
4577	ulen = info.nr_map_ids;
4578	info.nr_map_ids = prog->aux->used_map_cnt;
4579	ulen = min_t(u32, info.nr_map_ids, ulen);
4580	if (ulen) {
4581		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4582		u32 i;
4583
4584		for (i = 0; i < ulen; i++)
4585			if (put_user(prog->aux->used_maps[i]->id,
4586				     &user_map_ids[i])) {
4587				mutex_unlock(&prog->aux->used_maps_mutex);
4588				return -EFAULT;
4589			}
4590	}
4591	mutex_unlock(&prog->aux->used_maps_mutex);
4592
4593	err = set_info_rec_size(&info);
4594	if (err)
4595		return err;
4596
4597	bpf_prog_get_stats(prog, &stats);
4598	info.run_time_ns = stats.nsecs;
4599	info.run_cnt = stats.cnt;
4600	info.recursion_misses = stats.misses;
4601
4602	info.verified_insns = prog->aux->verified_insns;
4603
4604	if (!bpf_capable()) {
4605		info.jited_prog_len = 0;
4606		info.xlated_prog_len = 0;
4607		info.nr_jited_ksyms = 0;
4608		info.nr_jited_func_lens = 0;
4609		info.nr_func_info = 0;
4610		info.nr_line_info = 0;
4611		info.nr_jited_line_info = 0;
4612		goto done;
4613	}
4614
4615	ulen = info.xlated_prog_len;
4616	info.xlated_prog_len = bpf_prog_insn_size(prog);
4617	if (info.xlated_prog_len && ulen) {
4618		struct bpf_insn *insns_sanitized;
4619		bool fault;
4620
4621		if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4622			info.xlated_prog_insns = 0;
4623			goto done;
4624		}
4625		insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4626		if (!insns_sanitized)
4627			return -ENOMEM;
4628		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4629		ulen = min_t(u32, info.xlated_prog_len, ulen);
4630		fault = copy_to_user(uinsns, insns_sanitized, ulen);
4631		kfree(insns_sanitized);
4632		if (fault)
4633			return -EFAULT;
4634	}
4635
4636	if (bpf_prog_is_offloaded(prog->aux)) {
4637		err = bpf_prog_offload_info_fill(&info, prog);
4638		if (err)
4639			return err;
4640		goto done;
4641	}
4642
4643	/* NOTE: the following code is supposed to be skipped for offload.
4644	 * bpf_prog_offload_info_fill() is the place to fill similar fields
4645	 * for offload.
4646	 */
4647	ulen = info.jited_prog_len;
4648	if (prog->aux->func_cnt) {
4649		u32 i;
4650
4651		info.jited_prog_len = 0;
4652		for (i = 0; i < prog->aux->func_cnt; i++)
4653			info.jited_prog_len += prog->aux->func[i]->jited_len;
4654	} else {
4655		info.jited_prog_len = prog->jited_len;
4656	}
4657
4658	if (info.jited_prog_len && ulen) {
4659		if (bpf_dump_raw_ok(file->f_cred)) {
4660			uinsns = u64_to_user_ptr(info.jited_prog_insns);
4661			ulen = min_t(u32, info.jited_prog_len, ulen);
4662
4663			/* for multi-function programs, copy the JITed
4664			 * instructions for all the functions
4665			 */
4666			if (prog->aux->func_cnt) {
4667				u32 len, free, i;
4668				u8 *img;
4669
4670				free = ulen;
4671				for (i = 0; i < prog->aux->func_cnt; i++) {
4672					len = prog->aux->func[i]->jited_len;
4673					len = min_t(u32, len, free);
4674					img = (u8 *) prog->aux->func[i]->bpf_func;
4675					if (copy_to_user(uinsns, img, len))
4676						return -EFAULT;
4677					uinsns += len;
4678					free -= len;
4679					if (!free)
4680						break;
4681				}
4682			} else {
4683				if (copy_to_user(uinsns, prog->bpf_func, ulen))
4684					return -EFAULT;
4685			}
4686		} else {
4687			info.jited_prog_insns = 0;
4688		}
4689	}
4690
4691	ulen = info.nr_jited_ksyms;
4692	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4693	if (ulen) {
4694		if (bpf_dump_raw_ok(file->f_cred)) {
4695			unsigned long ksym_addr;
4696			u64 __user *user_ksyms;
4697			u32 i;
4698
4699			/* copy the address of the kernel symbol
4700			 * corresponding to each function
4701			 */
4702			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4703			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4704			if (prog->aux->func_cnt) {
4705				for (i = 0; i < ulen; i++) {
4706					ksym_addr = (unsigned long)
4707						prog->aux->func[i]->bpf_func;
4708					if (put_user((u64) ksym_addr,
4709						     &user_ksyms[i]))
4710						return -EFAULT;
4711				}
4712			} else {
4713				ksym_addr = (unsigned long) prog->bpf_func;
4714				if (put_user((u64) ksym_addr, &user_ksyms[0]))
4715					return -EFAULT;
4716			}
4717		} else {
4718			info.jited_ksyms = 0;
4719		}
4720	}
4721
4722	ulen = info.nr_jited_func_lens;
4723	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4724	if (ulen) {
4725		if (bpf_dump_raw_ok(file->f_cred)) {
4726			u32 __user *user_lens;
4727			u32 func_len, i;
4728
4729			/* copy the JITed image lengths for each function */
4730			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4731			user_lens = u64_to_user_ptr(info.jited_func_lens);
4732			if (prog->aux->func_cnt) {
4733				for (i = 0; i < ulen; i++) {
4734					func_len =
4735						prog->aux->func[i]->jited_len;
4736					if (put_user(func_len, &user_lens[i]))
4737						return -EFAULT;
4738				}
4739			} else {
4740				func_len = prog->jited_len;
4741				if (put_user(func_len, &user_lens[0]))
4742					return -EFAULT;
4743			}
4744		} else {
4745			info.jited_func_lens = 0;
4746		}
4747	}
4748
4749	if (prog->aux->btf)
4750		info.btf_id = btf_obj_id(prog->aux->btf);
4751	info.attach_btf_id = prog->aux->attach_btf_id;
4752	if (attach_btf)
4753		info.attach_btf_obj_id = btf_obj_id(attach_btf);
4754
4755	ulen = info.nr_func_info;
4756	info.nr_func_info = prog->aux->func_info_cnt;
4757	if (info.nr_func_info && ulen) {
4758		char __user *user_finfo;
4759
4760		user_finfo = u64_to_user_ptr(info.func_info);
4761		ulen = min_t(u32, info.nr_func_info, ulen);
4762		if (copy_to_user(user_finfo, prog->aux->func_info,
4763				 info.func_info_rec_size * ulen))
4764			return -EFAULT;
4765	}
4766
4767	ulen = info.nr_line_info;
4768	info.nr_line_info = prog->aux->nr_linfo;
4769	if (info.nr_line_info && ulen) {
4770		__u8 __user *user_linfo;
4771
4772		user_linfo = u64_to_user_ptr(info.line_info);
4773		ulen = min_t(u32, info.nr_line_info, ulen);
4774		if (copy_to_user(user_linfo, prog->aux->linfo,
4775				 info.line_info_rec_size * ulen))
4776			return -EFAULT;
4777	}
4778
4779	ulen = info.nr_jited_line_info;
4780	if (prog->aux->jited_linfo)
4781		info.nr_jited_line_info = prog->aux->nr_linfo;
4782	else
4783		info.nr_jited_line_info = 0;
4784	if (info.nr_jited_line_info && ulen) {
4785		if (bpf_dump_raw_ok(file->f_cred)) {
4786			unsigned long line_addr;
4787			__u64 __user *user_linfo;
4788			u32 i;
4789
4790			user_linfo = u64_to_user_ptr(info.jited_line_info);
4791			ulen = min_t(u32, info.nr_jited_line_info, ulen);
4792			for (i = 0; i < ulen; i++) {
4793				line_addr = (unsigned long)prog->aux->jited_linfo[i];
4794				if (put_user((__u64)line_addr, &user_linfo[i]))
4795					return -EFAULT;
4796			}
4797		} else {
4798			info.jited_line_info = 0;
4799		}
4800	}
4801
4802	ulen = info.nr_prog_tags;
4803	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4804	if (ulen) {
4805		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4806		u32 i;
4807
4808		user_prog_tags = u64_to_user_ptr(info.prog_tags);
4809		ulen = min_t(u32, info.nr_prog_tags, ulen);
4810		if (prog->aux->func_cnt) {
4811			for (i = 0; i < ulen; i++) {
4812				if (copy_to_user(user_prog_tags[i],
4813						 prog->aux->func[i]->tag,
4814						 BPF_TAG_SIZE))
4815					return -EFAULT;
4816			}
4817		} else {
4818			if (copy_to_user(user_prog_tags[0],
4819					 prog->tag, BPF_TAG_SIZE))
4820				return -EFAULT;
4821		}
4822	}
4823
4824done:
4825	if (copy_to_user(uinfo, &info, info_len) ||
4826	    put_user(info_len, &uattr->info.info_len))
4827		return -EFAULT;
4828
4829	return 0;
4830}
4831
4832static int bpf_map_get_info_by_fd(struct file *file,
4833				  struct bpf_map *map,
4834				  const union bpf_attr *attr,
4835				  union bpf_attr __user *uattr)
4836{
4837	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4838	struct bpf_map_info info;
4839	u32 info_len = attr->info.info_len;
4840	int err;
4841
4842	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4843	if (err)
4844		return err;
4845	info_len = min_t(u32, sizeof(info), info_len);
4846
4847	memset(&info, 0, sizeof(info));
4848	info.type = map->map_type;
4849	info.id = map->id;
4850	info.key_size = map->key_size;
4851	info.value_size = map->value_size;
4852	info.max_entries = map->max_entries;
4853	info.map_flags = map->map_flags;
4854	info.map_extra = map->map_extra;
4855	memcpy(info.name, map->name, sizeof(map->name));
4856
4857	if (map->btf) {
4858		info.btf_id = btf_obj_id(map->btf);
4859		info.btf_key_type_id = map->btf_key_type_id;
4860		info.btf_value_type_id = map->btf_value_type_id;
4861	}
4862	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4863	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4864		bpf_map_struct_ops_info_fill(&info, map);
4865
4866	if (bpf_map_is_offloaded(map)) {
4867		err = bpf_map_offload_info_fill(&info, map);
4868		if (err)
4869			return err;
4870	}
4871
4872	if (copy_to_user(uinfo, &info, info_len) ||
4873	    put_user(info_len, &uattr->info.info_len))
4874		return -EFAULT;
4875
4876	return 0;
4877}
4878
4879static int bpf_btf_get_info_by_fd(struct file *file,
4880				  struct btf *btf,
4881				  const union bpf_attr *attr,
4882				  union bpf_attr __user *uattr)
4883{
4884	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4885	u32 info_len = attr->info.info_len;
4886	int err;
4887
4888	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4889	if (err)
4890		return err;
4891
4892	return btf_get_info_by_fd(btf, attr, uattr);
4893}
4894
4895static int bpf_link_get_info_by_fd(struct file *file,
4896				  struct bpf_link *link,
4897				  const union bpf_attr *attr,
4898				  union bpf_attr __user *uattr)
4899{
4900	struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4901	struct bpf_link_info info;
4902	u32 info_len = attr->info.info_len;
4903	int err;
4904
4905	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4906	if (err)
4907		return err;
4908	info_len = min_t(u32, sizeof(info), info_len);
4909
4910	memset(&info, 0, sizeof(info));
4911	if (copy_from_user(&info, uinfo, info_len))
4912		return -EFAULT;
4913
4914	info.type = link->type;
4915	info.id = link->id;
4916	if (link->prog)
4917		info.prog_id = link->prog->aux->id;
4918
4919	if (link->ops->fill_link_info) {
4920		err = link->ops->fill_link_info(link, &info);
4921		if (err)
4922			return err;
4923	}
4924
4925	if (copy_to_user(uinfo, &info, info_len) ||
4926	    put_user(info_len, &uattr->info.info_len))
4927		return -EFAULT;
4928
4929	return 0;
4930}
4931
4932
4933#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4934
4935static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4936				  union bpf_attr __user *uattr)
4937{
4938	int ufd = attr->info.bpf_fd;
4939	struct fd f;
4940	int err;
4941
4942	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4943		return -EINVAL;
4944
4945	f = fdget(ufd);
4946	if (!f.file)
4947		return -EBADFD;
4948
4949	if (f.file->f_op == &bpf_prog_fops)
4950		err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4951					      uattr);
4952	else if (f.file->f_op == &bpf_map_fops)
4953		err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4954					     uattr);
4955	else if (f.file->f_op == &btf_fops)
4956		err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4957	else if (f.file->f_op == &bpf_link_fops)
4958		err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4959					      attr, uattr);
4960	else
4961		err = -EINVAL;
4962
4963	fdput(f);
4964	return err;
4965}
4966
4967#define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
4968
4969static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4970{
4971	struct bpf_token *token = NULL;
4972
4973	if (CHECK_ATTR(BPF_BTF_LOAD))
4974		return -EINVAL;
4975
4976	if (attr->btf_flags & ~BPF_F_TOKEN_FD)
4977		return -EINVAL;
4978
4979	if (attr->btf_flags & BPF_F_TOKEN_FD) {
4980		token = bpf_token_get_from_fd(attr->btf_token_fd);
4981		if (IS_ERR(token))
4982			return PTR_ERR(token);
4983		if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
4984			bpf_token_put(token);
4985			token = NULL;
4986		}
4987	}
4988
4989	if (!bpf_token_capable(token, CAP_BPF)) {
4990		bpf_token_put(token);
4991		return -EPERM;
4992	}
4993
4994	bpf_token_put(token);
4995
4996	return btf_new_fd(attr, uattr, uattr_size);
4997}
4998
4999#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
5000
5001static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
5002{
5003	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
5004		return -EINVAL;
5005
5006	if (!capable(CAP_SYS_ADMIN))
5007		return -EPERM;
5008
5009	return btf_get_fd_by_id(attr->btf_id);
5010}
5011
5012static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5013				    union bpf_attr __user *uattr,
5014				    u32 prog_id, u32 fd_type,
5015				    const char *buf, u64 probe_offset,
5016				    u64 probe_addr)
5017{
5018	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5019	u32 len = buf ? strlen(buf) : 0, input_len;
5020	int err = 0;
5021
5022	if (put_user(len, &uattr->task_fd_query.buf_len))
5023		return -EFAULT;
5024	input_len = attr->task_fd_query.buf_len;
5025	if (input_len && ubuf) {
5026		if (!len) {
5027			/* nothing to copy, just make ubuf NULL terminated */
5028			char zero = '\0';
5029
5030			if (put_user(zero, ubuf))
5031				return -EFAULT;
5032		} else if (input_len >= len + 1) {
5033			/* ubuf can hold the string with NULL terminator */
5034			if (copy_to_user(ubuf, buf, len + 1))
5035				return -EFAULT;
5036		} else {
5037			/* ubuf cannot hold the string with NULL terminator,
5038			 * do a partial copy with NULL terminator.
5039			 */
5040			char zero = '\0';
5041
5042			err = -ENOSPC;
5043			if (copy_to_user(ubuf, buf, input_len - 1))
5044				return -EFAULT;
5045			if (put_user(zero, ubuf + input_len - 1))
5046				return -EFAULT;
5047		}
5048	}
5049
5050	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5051	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5052	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5053	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5054		return -EFAULT;
5055
5056	return err;
5057}
5058
5059#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5060
5061static int bpf_task_fd_query(const union bpf_attr *attr,
5062			     union bpf_attr __user *uattr)
5063{
5064	pid_t pid = attr->task_fd_query.pid;
5065	u32 fd = attr->task_fd_query.fd;
5066	const struct perf_event *event;
5067	struct task_struct *task;
5068	struct file *file;
5069	int err;
5070
5071	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5072		return -EINVAL;
5073
5074	if (!capable(CAP_SYS_ADMIN))
5075		return -EPERM;
5076
5077	if (attr->task_fd_query.flags != 0)
5078		return -EINVAL;
5079
5080	rcu_read_lock();
5081	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5082	rcu_read_unlock();
5083	if (!task)
5084		return -ENOENT;
5085
5086	err = 0;
5087	file = fget_task(task, fd);
5088	put_task_struct(task);
5089	if (!file)
5090		return -EBADF;
5091
5092	if (file->f_op == &bpf_link_fops) {
5093		struct bpf_link *link = file->private_data;
5094
5095		if (link->ops == &bpf_raw_tp_link_lops) {
5096			struct bpf_raw_tp_link *raw_tp =
5097				container_of(link, struct bpf_raw_tp_link, link);
5098			struct bpf_raw_event_map *btp = raw_tp->btp;
5099
5100			err = bpf_task_fd_query_copy(attr, uattr,
5101						     raw_tp->link.prog->aux->id,
5102						     BPF_FD_TYPE_RAW_TRACEPOINT,
5103						     btp->tp->name, 0, 0);
5104			goto put_file;
5105		}
5106		goto out_not_supp;
5107	}
5108
5109	event = perf_get_event(file);
5110	if (!IS_ERR(event)) {
5111		u64 probe_offset, probe_addr;
5112		u32 prog_id, fd_type;
5113		const char *buf;
5114
5115		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5116					      &buf, &probe_offset,
5117					      &probe_addr, NULL);
5118		if (!err)
5119			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5120						     fd_type, buf,
5121						     probe_offset,
5122						     probe_addr);
5123		goto put_file;
5124	}
5125
5126out_not_supp:
5127	err = -ENOTSUPP;
5128put_file:
5129	fput(file);
5130	return err;
5131}
5132
5133#define BPF_MAP_BATCH_LAST_FIELD batch.flags
5134
5135#define BPF_DO_BATCH(fn, ...)			\
5136	do {					\
5137		if (!fn) {			\
5138			err = -ENOTSUPP;	\
5139			goto err_put;		\
5140		}				\
5141		err = fn(__VA_ARGS__);		\
5142	} while (0)
5143
5144static int bpf_map_do_batch(const union bpf_attr *attr,
5145			    union bpf_attr __user *uattr,
5146			    int cmd)
5147{
5148	bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
5149			 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5150	bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5151	struct bpf_map *map;
5152	int err, ufd;
5153	struct fd f;
5154
5155	if (CHECK_ATTR(BPF_MAP_BATCH))
5156		return -EINVAL;
5157
5158	ufd = attr->batch.map_fd;
5159	f = fdget(ufd);
5160	map = __bpf_map_get(f);
5161	if (IS_ERR(map))
5162		return PTR_ERR(map);
5163	if (has_write)
5164		bpf_map_write_active_inc(map);
5165	if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5166		err = -EPERM;
5167		goto err_put;
5168	}
5169	if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5170		err = -EPERM;
5171		goto err_put;
5172	}
5173
5174	if (cmd == BPF_MAP_LOOKUP_BATCH)
5175		BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5176	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5177		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5178	else if (cmd == BPF_MAP_UPDATE_BATCH)
5179		BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
5180	else
5181		BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5182err_put:
5183	if (has_write) {
5184		maybe_wait_bpf_programs(map);
5185		bpf_map_write_active_dec(map);
5186	}
5187	fdput(f);
5188	return err;
5189}
5190
5191#define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5192static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5193{
 
5194	struct bpf_prog *prog;
5195	int ret;
5196
5197	if (CHECK_ATTR(BPF_LINK_CREATE))
5198		return -EINVAL;
5199
5200	if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5201		return bpf_struct_ops_link_create(attr);
5202
5203	prog = bpf_prog_get(attr->link_create.prog_fd);
5204	if (IS_ERR(prog))
5205		return PTR_ERR(prog);
5206
5207	ret = bpf_prog_attach_check_attach_type(prog,
5208						attr->link_create.attach_type);
5209	if (ret)
5210		goto out;
5211
5212	switch (prog->type) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5213	case BPF_PROG_TYPE_CGROUP_SKB:
5214	case BPF_PROG_TYPE_CGROUP_SOCK:
5215	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5216	case BPF_PROG_TYPE_SOCK_OPS:
5217	case BPF_PROG_TYPE_CGROUP_DEVICE:
5218	case BPF_PROG_TYPE_CGROUP_SYSCTL:
5219	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5220		ret = cgroup_bpf_link_attach(attr, prog);
5221		break;
5222	case BPF_PROG_TYPE_EXT:
5223		ret = bpf_tracing_prog_attach(prog,
5224					      attr->link_create.target_fd,
5225					      attr->link_create.target_btf_id,
5226					      attr->link_create.tracing.cookie);
5227		break;
5228	case BPF_PROG_TYPE_LSM:
5229	case BPF_PROG_TYPE_TRACING:
5230		if (attr->link_create.attach_type != prog->expected_attach_type) {
5231			ret = -EINVAL;
5232			goto out;
5233		}
5234		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5235			ret = bpf_raw_tp_link_attach(prog, NULL);
5236		else if (prog->expected_attach_type == BPF_TRACE_ITER)
5237			ret = bpf_iter_link_attach(attr, uattr, prog);
5238		else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5239			ret = cgroup_bpf_link_attach(attr, prog);
5240		else
5241			ret = bpf_tracing_prog_attach(prog,
5242						      attr->link_create.target_fd,
5243						      attr->link_create.target_btf_id,
5244						      attr->link_create.tracing.cookie);
5245		break;
5246	case BPF_PROG_TYPE_FLOW_DISSECTOR:
5247	case BPF_PROG_TYPE_SK_LOOKUP:
5248		ret = netns_bpf_link_create(attr, prog);
5249		break;
5250#ifdef CONFIG_NET
5251	case BPF_PROG_TYPE_XDP:
5252		ret = bpf_xdp_link_attach(attr, prog);
5253		break;
5254	case BPF_PROG_TYPE_SCHED_CLS:
5255		if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5256		    attr->link_create.attach_type == BPF_TCX_EGRESS)
5257			ret = tcx_link_attach(attr, prog);
5258		else
5259			ret = netkit_link_attach(attr, prog);
5260		break;
5261	case BPF_PROG_TYPE_NETFILTER:
5262		ret = bpf_nf_link_attach(attr, prog);
5263		break;
5264#endif
5265	case BPF_PROG_TYPE_PERF_EVENT:
5266	case BPF_PROG_TYPE_TRACEPOINT:
5267		ret = bpf_perf_link_attach(attr, prog);
5268		break;
5269	case BPF_PROG_TYPE_KPROBE:
5270		if (attr->link_create.attach_type == BPF_PERF_EVENT)
5271			ret = bpf_perf_link_attach(attr, prog);
5272		else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5273			ret = bpf_kprobe_multi_link_attach(attr, prog);
5274		else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5275			ret = bpf_uprobe_multi_link_attach(attr, prog);
5276		break;
5277	default:
5278		ret = -EINVAL;
5279	}
5280
5281out:
5282	if (ret < 0)
5283		bpf_prog_put(prog);
5284	return ret;
5285}
5286
5287static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5288{
5289	struct bpf_map *new_map, *old_map = NULL;
5290	int ret;
5291
5292	new_map = bpf_map_get(attr->link_update.new_map_fd);
5293	if (IS_ERR(new_map))
5294		return PTR_ERR(new_map);
5295
5296	if (attr->link_update.flags & BPF_F_REPLACE) {
5297		old_map = bpf_map_get(attr->link_update.old_map_fd);
5298		if (IS_ERR(old_map)) {
5299			ret = PTR_ERR(old_map);
5300			goto out_put;
5301		}
5302	} else if (attr->link_update.old_map_fd) {
5303		ret = -EINVAL;
5304		goto out_put;
5305	}
5306
5307	ret = link->ops->update_map(link, new_map, old_map);
5308
5309	if (old_map)
5310		bpf_map_put(old_map);
5311out_put:
5312	bpf_map_put(new_map);
5313	return ret;
5314}
5315
5316#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5317
5318static int link_update(union bpf_attr *attr)
5319{
5320	struct bpf_prog *old_prog = NULL, *new_prog;
5321	struct bpf_link *link;
5322	u32 flags;
5323	int ret;
5324
5325	if (CHECK_ATTR(BPF_LINK_UPDATE))
5326		return -EINVAL;
5327
5328	flags = attr->link_update.flags;
5329	if (flags & ~BPF_F_REPLACE)
5330		return -EINVAL;
5331
5332	link = bpf_link_get_from_fd(attr->link_update.link_fd);
5333	if (IS_ERR(link))
5334		return PTR_ERR(link);
5335
5336	if (link->ops->update_map) {
5337		ret = link_update_map(link, attr);
5338		goto out_put_link;
5339	}
5340
5341	new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5342	if (IS_ERR(new_prog)) {
5343		ret = PTR_ERR(new_prog);
5344		goto out_put_link;
5345	}
5346
5347	if (flags & BPF_F_REPLACE) {
5348		old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5349		if (IS_ERR(old_prog)) {
5350			ret = PTR_ERR(old_prog);
5351			old_prog = NULL;
5352			goto out_put_progs;
5353		}
5354	} else if (attr->link_update.old_prog_fd) {
5355		ret = -EINVAL;
5356		goto out_put_progs;
5357	}
5358
5359	if (link->ops->update_prog)
5360		ret = link->ops->update_prog(link, new_prog, old_prog);
5361	else
5362		ret = -EINVAL;
5363
5364out_put_progs:
5365	if (old_prog)
5366		bpf_prog_put(old_prog);
5367	if (ret)
5368		bpf_prog_put(new_prog);
5369out_put_link:
5370	bpf_link_put_direct(link);
5371	return ret;
5372}
5373
5374#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5375
5376static int link_detach(union bpf_attr *attr)
5377{
5378	struct bpf_link *link;
5379	int ret;
5380
5381	if (CHECK_ATTR(BPF_LINK_DETACH))
5382		return -EINVAL;
5383
5384	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5385	if (IS_ERR(link))
5386		return PTR_ERR(link);
5387
5388	if (link->ops->detach)
5389		ret = link->ops->detach(link);
5390	else
5391		ret = -EOPNOTSUPP;
5392
5393	bpf_link_put_direct(link);
5394	return ret;
5395}
5396
5397static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5398{
5399	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5400}
5401
5402struct bpf_link *bpf_link_by_id(u32 id)
5403{
5404	struct bpf_link *link;
5405
5406	if (!id)
5407		return ERR_PTR(-ENOENT);
5408
5409	spin_lock_bh(&link_idr_lock);
5410	/* before link is "settled", ID is 0, pretend it doesn't exist yet */
5411	link = idr_find(&link_idr, id);
5412	if (link) {
5413		if (link->id)
5414			link = bpf_link_inc_not_zero(link);
5415		else
5416			link = ERR_PTR(-EAGAIN);
5417	} else {
5418		link = ERR_PTR(-ENOENT);
5419	}
5420	spin_unlock_bh(&link_idr_lock);
5421	return link;
5422}
5423
5424struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5425{
5426	struct bpf_link *link;
5427
5428	spin_lock_bh(&link_idr_lock);
5429again:
5430	link = idr_get_next(&link_idr, id);
5431	if (link) {
5432		link = bpf_link_inc_not_zero(link);
5433		if (IS_ERR(link)) {
5434			(*id)++;
5435			goto again;
5436		}
5437	}
5438	spin_unlock_bh(&link_idr_lock);
5439
5440	return link;
5441}
5442
5443#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5444
5445static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5446{
5447	struct bpf_link *link;
5448	u32 id = attr->link_id;
5449	int fd;
5450
5451	if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5452		return -EINVAL;
5453
5454	if (!capable(CAP_SYS_ADMIN))
5455		return -EPERM;
5456
5457	link = bpf_link_by_id(id);
5458	if (IS_ERR(link))
5459		return PTR_ERR(link);
5460
5461	fd = bpf_link_new_fd(link);
5462	if (fd < 0)
5463		bpf_link_put_direct(link);
5464
5465	return fd;
5466}
5467
5468DEFINE_MUTEX(bpf_stats_enabled_mutex);
5469
5470static int bpf_stats_release(struct inode *inode, struct file *file)
5471{
5472	mutex_lock(&bpf_stats_enabled_mutex);
5473	static_key_slow_dec(&bpf_stats_enabled_key.key);
5474	mutex_unlock(&bpf_stats_enabled_mutex);
5475	return 0;
5476}
5477
5478static const struct file_operations bpf_stats_fops = {
5479	.release = bpf_stats_release,
5480};
5481
5482static int bpf_enable_runtime_stats(void)
5483{
5484	int fd;
5485
5486	mutex_lock(&bpf_stats_enabled_mutex);
5487
5488	/* Set a very high limit to avoid overflow */
5489	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5490		mutex_unlock(&bpf_stats_enabled_mutex);
5491		return -EBUSY;
5492	}
5493
5494	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5495	if (fd >= 0)
5496		static_key_slow_inc(&bpf_stats_enabled_key.key);
5497
5498	mutex_unlock(&bpf_stats_enabled_mutex);
5499	return fd;
5500}
5501
5502#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5503
5504static int bpf_enable_stats(union bpf_attr *attr)
5505{
5506
5507	if (CHECK_ATTR(BPF_ENABLE_STATS))
5508		return -EINVAL;
5509
5510	if (!capable(CAP_SYS_ADMIN))
5511		return -EPERM;
5512
5513	switch (attr->enable_stats.type) {
5514	case BPF_STATS_RUN_TIME:
5515		return bpf_enable_runtime_stats();
5516	default:
5517		break;
5518	}
5519	return -EINVAL;
5520}
5521
5522#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5523
5524static int bpf_iter_create(union bpf_attr *attr)
5525{
5526	struct bpf_link *link;
5527	int err;
5528
5529	if (CHECK_ATTR(BPF_ITER_CREATE))
5530		return -EINVAL;
5531
5532	if (attr->iter_create.flags)
5533		return -EINVAL;
5534
5535	link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5536	if (IS_ERR(link))
5537		return PTR_ERR(link);
5538
5539	err = bpf_iter_new_fd(link);
5540	bpf_link_put_direct(link);
5541
5542	return err;
5543}
5544
5545#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5546
5547static int bpf_prog_bind_map(union bpf_attr *attr)
5548{
5549	struct bpf_prog *prog;
5550	struct bpf_map *map;
5551	struct bpf_map **used_maps_old, **used_maps_new;
5552	int i, ret = 0;
5553
5554	if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5555		return -EINVAL;
5556
5557	if (attr->prog_bind_map.flags)
5558		return -EINVAL;
5559
5560	prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5561	if (IS_ERR(prog))
5562		return PTR_ERR(prog);
5563
5564	map = bpf_map_get(attr->prog_bind_map.map_fd);
5565	if (IS_ERR(map)) {
5566		ret = PTR_ERR(map);
5567		goto out_prog_put;
5568	}
5569
5570	mutex_lock(&prog->aux->used_maps_mutex);
5571
5572	used_maps_old = prog->aux->used_maps;
5573
5574	for (i = 0; i < prog->aux->used_map_cnt; i++)
5575		if (used_maps_old[i] == map) {
5576			bpf_map_put(map);
5577			goto out_unlock;
5578		}
5579
5580	used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5581				      sizeof(used_maps_new[0]),
5582				      GFP_KERNEL);
5583	if (!used_maps_new) {
5584		ret = -ENOMEM;
5585		goto out_unlock;
5586	}
5587
5588	/* The bpf program will not access the bpf map, but for the sake of
5589	 * simplicity, increase sleepable_refcnt for sleepable program as well.
5590	 */
5591	if (prog->sleepable)
5592		atomic64_inc(&map->sleepable_refcnt);
5593	memcpy(used_maps_new, used_maps_old,
5594	       sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5595	used_maps_new[prog->aux->used_map_cnt] = map;
5596
5597	prog->aux->used_map_cnt++;
5598	prog->aux->used_maps = used_maps_new;
5599
5600	kfree(used_maps_old);
5601
5602out_unlock:
5603	mutex_unlock(&prog->aux->used_maps_mutex);
5604
5605	if (ret)
5606		bpf_map_put(map);
5607out_prog_put:
5608	bpf_prog_put(prog);
5609	return ret;
5610}
5611
5612#define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5613
5614static int token_create(union bpf_attr *attr)
5615{
5616	if (CHECK_ATTR(BPF_TOKEN_CREATE))
5617		return -EINVAL;
5618
5619	/* no flags are supported yet */
5620	if (attr->token_create.flags)
5621		return -EINVAL;
5622
5623	return bpf_token_create(attr);
5624}
5625
5626static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5627{
5628	union bpf_attr attr;
 
5629	int err;
5630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5631	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5632	if (err)
5633		return err;
5634	size = min_t(u32, size, sizeof(attr));
5635
5636	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
5637	memset(&attr, 0, sizeof(attr));
5638	if (copy_from_bpfptr(&attr, uattr, size) != 0)
5639		return -EFAULT;
5640
5641	err = security_bpf(cmd, &attr, size);
5642	if (err < 0)
5643		return err;
5644
5645	switch (cmd) {
5646	case BPF_MAP_CREATE:
5647		err = map_create(&attr);
5648		break;
5649	case BPF_MAP_LOOKUP_ELEM:
5650		err = map_lookup_elem(&attr);
5651		break;
5652	case BPF_MAP_UPDATE_ELEM:
5653		err = map_update_elem(&attr, uattr);
5654		break;
5655	case BPF_MAP_DELETE_ELEM:
5656		err = map_delete_elem(&attr, uattr);
5657		break;
5658	case BPF_MAP_GET_NEXT_KEY:
5659		err = map_get_next_key(&attr);
5660		break;
5661	case BPF_MAP_FREEZE:
5662		err = map_freeze(&attr);
5663		break;
5664	case BPF_PROG_LOAD:
5665		err = bpf_prog_load(&attr, uattr, size);
5666		break;
5667	case BPF_OBJ_PIN:
5668		err = bpf_obj_pin(&attr);
5669		break;
5670	case BPF_OBJ_GET:
5671		err = bpf_obj_get(&attr);
5672		break;
5673	case BPF_PROG_ATTACH:
5674		err = bpf_prog_attach(&attr);
5675		break;
5676	case BPF_PROG_DETACH:
5677		err = bpf_prog_detach(&attr);
5678		break;
5679	case BPF_PROG_QUERY:
5680		err = bpf_prog_query(&attr, uattr.user);
5681		break;
5682	case BPF_PROG_TEST_RUN:
5683		err = bpf_prog_test_run(&attr, uattr.user);
5684		break;
5685	case BPF_PROG_GET_NEXT_ID:
5686		err = bpf_obj_get_next_id(&attr, uattr.user,
5687					  &prog_idr, &prog_idr_lock);
5688		break;
5689	case BPF_MAP_GET_NEXT_ID:
5690		err = bpf_obj_get_next_id(&attr, uattr.user,
5691					  &map_idr, &map_idr_lock);
5692		break;
5693	case BPF_BTF_GET_NEXT_ID:
5694		err = bpf_obj_get_next_id(&attr, uattr.user,
5695					  &btf_idr, &btf_idr_lock);
5696		break;
5697	case BPF_PROG_GET_FD_BY_ID:
5698		err = bpf_prog_get_fd_by_id(&attr);
5699		break;
5700	case BPF_MAP_GET_FD_BY_ID:
5701		err = bpf_map_get_fd_by_id(&attr);
5702		break;
5703	case BPF_OBJ_GET_INFO_BY_FD:
5704		err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5705		break;
5706	case BPF_RAW_TRACEPOINT_OPEN:
5707		err = bpf_raw_tracepoint_open(&attr);
5708		break;
5709	case BPF_BTF_LOAD:
5710		err = bpf_btf_load(&attr, uattr, size);
5711		break;
5712	case BPF_BTF_GET_FD_BY_ID:
5713		err = bpf_btf_get_fd_by_id(&attr);
5714		break;
5715	case BPF_TASK_FD_QUERY:
5716		err = bpf_task_fd_query(&attr, uattr.user);
5717		break;
5718	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5719		err = map_lookup_and_delete_elem(&attr);
5720		break;
5721	case BPF_MAP_LOOKUP_BATCH:
5722		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5723		break;
5724	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5725		err = bpf_map_do_batch(&attr, uattr.user,
5726				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5727		break;
5728	case BPF_MAP_UPDATE_BATCH:
5729		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5730		break;
5731	case BPF_MAP_DELETE_BATCH:
5732		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5733		break;
5734	case BPF_LINK_CREATE:
5735		err = link_create(&attr, uattr);
5736		break;
5737	case BPF_LINK_UPDATE:
5738		err = link_update(&attr);
5739		break;
5740	case BPF_LINK_GET_FD_BY_ID:
5741		err = bpf_link_get_fd_by_id(&attr);
5742		break;
5743	case BPF_LINK_GET_NEXT_ID:
5744		err = bpf_obj_get_next_id(&attr, uattr.user,
5745					  &link_idr, &link_idr_lock);
5746		break;
5747	case BPF_ENABLE_STATS:
5748		err = bpf_enable_stats(&attr);
5749		break;
5750	case BPF_ITER_CREATE:
5751		err = bpf_iter_create(&attr);
5752		break;
5753	case BPF_LINK_DETACH:
5754		err = link_detach(&attr);
5755		break;
5756	case BPF_PROG_BIND_MAP:
5757		err = bpf_prog_bind_map(&attr);
5758		break;
5759	case BPF_TOKEN_CREATE:
5760		err = token_create(&attr);
5761		break;
5762	default:
5763		err = -EINVAL;
5764		break;
5765	}
5766
5767	return err;
5768}
5769
5770SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5771{
5772	return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5773}
5774
5775static bool syscall_prog_is_valid_access(int off, int size,
5776					 enum bpf_access_type type,
5777					 const struct bpf_prog *prog,
5778					 struct bpf_insn_access_aux *info)
5779{
5780	if (off < 0 || off >= U16_MAX)
5781		return false;
5782	if (off % size != 0)
5783		return false;
5784	return true;
5785}
5786
5787BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5788{
5789	switch (cmd) {
5790	case BPF_MAP_CREATE:
5791	case BPF_MAP_DELETE_ELEM:
5792	case BPF_MAP_UPDATE_ELEM:
5793	case BPF_MAP_FREEZE:
5794	case BPF_MAP_GET_FD_BY_ID:
5795	case BPF_PROG_LOAD:
5796	case BPF_BTF_LOAD:
5797	case BPF_LINK_CREATE:
5798	case BPF_RAW_TRACEPOINT_OPEN:
5799		break;
5800	default:
5801		return -EINVAL;
5802	}
5803	return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5804}
5805
5806
5807/* To shut up -Wmissing-prototypes.
5808 * This function is used by the kernel light skeleton
5809 * to load bpf programs when modules are loaded or during kernel boot.
5810 * See tools/lib/bpf/skel_internal.h
5811 */
5812int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5813
5814int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5815{
5816	struct bpf_prog * __maybe_unused prog;
5817	struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5818
5819	switch (cmd) {
5820#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5821	case BPF_PROG_TEST_RUN:
5822		if (attr->test.data_in || attr->test.data_out ||
5823		    attr->test.ctx_out || attr->test.duration ||
5824		    attr->test.repeat || attr->test.flags)
5825			return -EINVAL;
5826
5827		prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5828		if (IS_ERR(prog))
5829			return PTR_ERR(prog);
5830
5831		if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5832		    attr->test.ctx_size_in > U16_MAX) {
5833			bpf_prog_put(prog);
5834			return -EINVAL;
5835		}
5836
5837		run_ctx.bpf_cookie = 0;
 
5838		if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5839			/* recursion detected */
5840			__bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5841			bpf_prog_put(prog);
5842			return -EBUSY;
5843		}
5844		attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5845		__bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5846						&run_ctx);
5847		bpf_prog_put(prog);
5848		return 0;
5849#endif
5850	default:
5851		return ____bpf_sys_bpf(cmd, attr, size);
5852	}
5853}
5854EXPORT_SYMBOL(kern_sys_bpf);
5855
5856static const struct bpf_func_proto bpf_sys_bpf_proto = {
5857	.func		= bpf_sys_bpf,
5858	.gpl_only	= false,
5859	.ret_type	= RET_INTEGER,
5860	.arg1_type	= ARG_ANYTHING,
5861	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
5862	.arg3_type	= ARG_CONST_SIZE,
5863};
5864
5865const struct bpf_func_proto * __weak
5866tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5867{
5868	return bpf_base_func_proto(func_id, prog);
5869}
5870
5871BPF_CALL_1(bpf_sys_close, u32, fd)
5872{
5873	/* When bpf program calls this helper there should not be
5874	 * an fdget() without matching completed fdput().
5875	 * This helper is allowed in the following callchain only:
5876	 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5877	 */
5878	return close_fd(fd);
5879}
5880
5881static const struct bpf_func_proto bpf_sys_close_proto = {
5882	.func		= bpf_sys_close,
5883	.gpl_only	= false,
5884	.ret_type	= RET_INTEGER,
5885	.arg1_type	= ARG_ANYTHING,
5886};
5887
5888BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5889{
5890	if (flags)
5891		return -EINVAL;
5892
5893	if (name_sz <= 1 || name[name_sz - 1])
5894		return -EINVAL;
5895
5896	if (!bpf_dump_raw_ok(current_cred()))
5897		return -EPERM;
5898
5899	*res = kallsyms_lookup_name(name);
5900	return *res ? 0 : -ENOENT;
5901}
5902
5903static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5904	.func		= bpf_kallsyms_lookup_name,
5905	.gpl_only	= false,
5906	.ret_type	= RET_INTEGER,
5907	.arg1_type	= ARG_PTR_TO_MEM,
5908	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
5909	.arg3_type	= ARG_ANYTHING,
5910	.arg4_type	= ARG_PTR_TO_LONG,
5911};
5912
5913static const struct bpf_func_proto *
5914syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5915{
5916	switch (func_id) {
5917	case BPF_FUNC_sys_bpf:
5918		return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
5919		       ? NULL : &bpf_sys_bpf_proto;
5920	case BPF_FUNC_btf_find_by_name_kind:
5921		return &bpf_btf_find_by_name_kind_proto;
5922	case BPF_FUNC_sys_close:
5923		return &bpf_sys_close_proto;
5924	case BPF_FUNC_kallsyms_lookup_name:
5925		return &bpf_kallsyms_lookup_name_proto;
5926	default:
5927		return tracing_prog_func_proto(func_id, prog);
5928	}
5929}
5930
5931const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5932	.get_func_proto  = syscall_prog_func_proto,
5933	.is_valid_access = syscall_prog_is_valid_access,
5934};
5935
5936const struct bpf_prog_ops bpf_syscall_prog_ops = {
5937	.test_run = bpf_prog_test_run_syscall,
5938};
5939
5940#ifdef CONFIG_SYSCTL
5941static int bpf_stats_handler(struct ctl_table *table, int write,
5942			     void *buffer, size_t *lenp, loff_t *ppos)
5943{
5944	struct static_key *key = (struct static_key *)table->data;
5945	static int saved_val;
5946	int val, ret;
5947	struct ctl_table tmp = {
5948		.data   = &val,
5949		.maxlen = sizeof(val),
5950		.mode   = table->mode,
5951		.extra1 = SYSCTL_ZERO,
5952		.extra2 = SYSCTL_ONE,
5953	};
5954
5955	if (write && !capable(CAP_SYS_ADMIN))
5956		return -EPERM;
5957
5958	mutex_lock(&bpf_stats_enabled_mutex);
5959	val = saved_val;
5960	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5961	if (write && !ret && val != saved_val) {
5962		if (val)
5963			static_key_slow_inc(key);
5964		else
5965			static_key_slow_dec(key);
5966		saved_val = val;
5967	}
5968	mutex_unlock(&bpf_stats_enabled_mutex);
5969	return ret;
5970}
5971
5972void __weak unpriv_ebpf_notify(int new_state)
5973{
5974}
5975
5976static int bpf_unpriv_handler(struct ctl_table *table, int write,
5977			      void *buffer, size_t *lenp, loff_t *ppos)
5978{
5979	int ret, unpriv_enable = *(int *)table->data;
5980	bool locked_state = unpriv_enable == 1;
5981	struct ctl_table tmp = *table;
5982
5983	if (write && !capable(CAP_SYS_ADMIN))
5984		return -EPERM;
5985
5986	tmp.data = &unpriv_enable;
5987	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5988	if (write && !ret) {
5989		if (locked_state && unpriv_enable != 1)
5990			return -EPERM;
5991		*(int *)table->data = unpriv_enable;
5992	}
5993
5994	if (write)
5995		unpriv_ebpf_notify(unpriv_enable);
5996
5997	return ret;
5998}
5999
6000static struct ctl_table bpf_syscall_table[] = {
6001	{
6002		.procname	= "unprivileged_bpf_disabled",
6003		.data		= &sysctl_unprivileged_bpf_disabled,
6004		.maxlen		= sizeof(sysctl_unprivileged_bpf_disabled),
6005		.mode		= 0644,
6006		.proc_handler	= bpf_unpriv_handler,
6007		.extra1		= SYSCTL_ZERO,
6008		.extra2		= SYSCTL_TWO,
6009	},
6010	{
6011		.procname	= "bpf_stats_enabled",
6012		.data		= &bpf_stats_enabled_key.key,
 
6013		.mode		= 0644,
6014		.proc_handler	= bpf_stats_handler,
6015	},
6016	{ }
6017};
6018
6019static int __init bpf_syscall_sysctl_init(void)
6020{
6021	register_sysctl_init("kernel", bpf_syscall_table);
6022	return 0;
6023}
6024late_initcall(bpf_syscall_sysctl_init);
6025#endif /* CONFIG_SYSCTL */