Linux Audio

Check our new training course

Loading...
v6.8
   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_STRUCT_OPS) {
 168		return map->ops->map_update_elem(map, key, value, flags);
 169	} else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 170		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 171		return sock_map_update_elem_sys(map, key, value, flags);
 172	} else if (IS_FD_PROG_ARRAY(map)) {
 173		return bpf_fd_array_map_update_elem(map, map_file, key, value,
 174						    flags);
 175	}
 176
 177	bpf_disable_instrumentation();
 178	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 179	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 180		err = bpf_percpu_hash_update(map, key, value, flags);
 181	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 182		err = bpf_percpu_array_update(map, key, value, flags);
 183	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 184		err = bpf_percpu_cgroup_storage_update(map, key, value,
 185						       flags);
 186	} else if (IS_FD_ARRAY(map)) {
 
 187		err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 188						   flags);
 
 189	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 
 190		err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 191						  flags);
 
 192	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 193		/* rcu_read_lock() is not needed */
 194		err = bpf_fd_reuseport_array_update_elem(map, key, value,
 195							 flags);
 196	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 197		   map->map_type == BPF_MAP_TYPE_STACK ||
 198		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 199		err = map->ops->map_push_elem(map, value, flags);
 200	} else {
 201		rcu_read_lock();
 202		err = map->ops->map_update_elem(map, key, value, flags);
 203		rcu_read_unlock();
 204	}
 205	bpf_enable_instrumentation();
 
 206
 207	return err;
 208}
 209
 210static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 211			      __u64 flags)
 212{
 213	void *ptr;
 214	int err;
 215
 216	if (bpf_map_is_offloaded(map))
 217		return bpf_map_offload_lookup_elem(map, key, value);
 218
 219	bpf_disable_instrumentation();
 220	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 221	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 222		err = bpf_percpu_hash_copy(map, key, value);
 223	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 224		err = bpf_percpu_array_copy(map, key, value);
 225	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 226		err = bpf_percpu_cgroup_storage_copy(map, key, value);
 227	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 228		err = bpf_stackmap_copy(map, key, value);
 229	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 230		err = bpf_fd_array_map_lookup_elem(map, key, value);
 231	} else if (IS_FD_HASH(map)) {
 232		err = bpf_fd_htab_map_lookup_elem(map, key, value);
 233	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 234		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 235	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 236		   map->map_type == BPF_MAP_TYPE_STACK ||
 237		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 238		err = map->ops->map_peek_elem(map, value);
 239	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 240		/* struct_ops map requires directly updating "value" */
 241		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 242	} else {
 243		rcu_read_lock();
 244		if (map->ops->map_lookup_elem_sys_only)
 245			ptr = map->ops->map_lookup_elem_sys_only(map, key);
 246		else
 247			ptr = map->ops->map_lookup_elem(map, key);
 248		if (IS_ERR(ptr)) {
 249			err = PTR_ERR(ptr);
 250		} else if (!ptr) {
 251			err = -ENOENT;
 252		} else {
 253			err = 0;
 254			if (flags & BPF_F_LOCK)
 255				/* lock 'ptr' and copy everything but lock */
 256				copy_map_value_locked(map, value, ptr, true);
 257			else
 258				copy_map_value(map, value, ptr);
 259			/* mask lock and timer, since value wasn't zero inited */
 260			check_and_init_map_value(map, value);
 261		}
 262		rcu_read_unlock();
 263	}
 264
 265	bpf_enable_instrumentation();
 
 266
 267	return err;
 268}
 269
 270/* Please, do not use this function outside from the map creation path
 271 * (e.g. in map update path) without taking care of setting the active
 272 * memory cgroup (see at bpf_map_kmalloc_node() for example).
 273 */
 274static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 275{
 276	/* We really just want to fail instead of triggering OOM killer
 277	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 278	 * which is used for lower order allocation requests.
 279	 *
 280	 * It has been observed that higher order allocation requests done by
 281	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
 282	 * to reclaim memory from the page cache, thus we set
 283	 * __GFP_RETRY_MAYFAIL to avoid such situations.
 284	 */
 285
 286	gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 287	unsigned int flags = 0;
 288	unsigned long align = 1;
 289	void *area;
 290
 291	if (size >= SIZE_MAX)
 292		return NULL;
 293
 294	/* kmalloc()'ed memory can't be mmap()'ed */
 295	if (mmapable) {
 296		BUG_ON(!PAGE_ALIGNED(size));
 297		align = SHMLBA;
 298		flags = VM_USERMAP;
 299	} else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 300		area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 301				    numa_node);
 302		if (area != NULL)
 303			return area;
 304	}
 305
 306	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 307			gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 308			flags, numa_node, __builtin_return_address(0));
 309}
 310
 311void *bpf_map_area_alloc(u64 size, int numa_node)
 312{
 313	return __bpf_map_area_alloc(size, numa_node, false);
 314}
 315
 316void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 317{
 318	return __bpf_map_area_alloc(size, numa_node, true);
 319}
 320
 321void bpf_map_area_free(void *area)
 322{
 323	kvfree(area);
 324}
 325
 326static u32 bpf_map_flags_retain_permanent(u32 flags)
 327{
 328	/* Some map creation flags are not tied to the map object but
 329	 * rather to the map fd instead, so they have no meaning upon
 330	 * map object inspection since multiple file descriptors with
 331	 * different (access) properties can exist here. Thus, given
 332	 * this has zero meaning for the map itself, lets clear these
 333	 * from here.
 334	 */
 335	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 336}
 337
 338void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 339{
 340	map->map_type = attr->map_type;
 341	map->key_size = attr->key_size;
 342	map->value_size = attr->value_size;
 343	map->max_entries = attr->max_entries;
 344	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 345	map->numa_node = bpf_map_attr_numa_node(attr);
 346	map->map_extra = attr->map_extra;
 347}
 348
 349static int bpf_map_alloc_id(struct bpf_map *map)
 350{
 351	int id;
 352
 353	idr_preload(GFP_KERNEL);
 354	spin_lock_bh(&map_idr_lock);
 355	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 356	if (id > 0)
 357		map->id = id;
 358	spin_unlock_bh(&map_idr_lock);
 359	idr_preload_end();
 360
 361	if (WARN_ON_ONCE(!id))
 362		return -ENOSPC;
 363
 364	return id > 0 ? 0 : id;
 365}
 366
 367void bpf_map_free_id(struct bpf_map *map)
 368{
 369	unsigned long flags;
 370
 371	/* Offloaded maps are removed from the IDR store when their device
 372	 * disappears - even if someone holds an fd to them they are unusable,
 373	 * the memory is gone, all ops will fail; they are simply waiting for
 374	 * refcnt to drop to be freed.
 375	 */
 376	if (!map->id)
 377		return;
 378
 379	spin_lock_irqsave(&map_idr_lock, flags);
 
 
 
 380
 381	idr_remove(&map_idr, map->id);
 382	map->id = 0;
 383
 384	spin_unlock_irqrestore(&map_idr_lock, flags);
 
 
 
 385}
 386
 387#ifdef CONFIG_MEMCG_KMEM
 388static void bpf_map_save_memcg(struct bpf_map *map)
 389{
 390	/* Currently if a map is created by a process belonging to the root
 391	 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 392	 * So we have to check map->objcg for being NULL each time it's
 393	 * being used.
 394	 */
 395	if (memcg_bpf_enabled())
 396		map->objcg = get_obj_cgroup_from_current();
 397}
 398
 399static void bpf_map_release_memcg(struct bpf_map *map)
 400{
 401	if (map->objcg)
 402		obj_cgroup_put(map->objcg);
 403}
 404
 405static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 406{
 407	if (map->objcg)
 408		return get_mem_cgroup_from_objcg(map->objcg);
 409
 410	return root_mem_cgroup;
 411}
 412
 413void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 414			   int node)
 415{
 416	struct mem_cgroup *memcg, *old_memcg;
 417	void *ptr;
 418
 419	memcg = bpf_map_get_memcg(map);
 420	old_memcg = set_active_memcg(memcg);
 421	ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 422	set_active_memcg(old_memcg);
 423	mem_cgroup_put(memcg);
 424
 425	return ptr;
 426}
 427
 428void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 429{
 430	struct mem_cgroup *memcg, *old_memcg;
 431	void *ptr;
 432
 433	memcg = bpf_map_get_memcg(map);
 434	old_memcg = set_active_memcg(memcg);
 435	ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 436	set_active_memcg(old_memcg);
 437	mem_cgroup_put(memcg);
 438
 439	return ptr;
 440}
 441
 442void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 443		       gfp_t flags)
 444{
 445	struct mem_cgroup *memcg, *old_memcg;
 446	void *ptr;
 447
 448	memcg = bpf_map_get_memcg(map);
 449	old_memcg = set_active_memcg(memcg);
 450	ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 451	set_active_memcg(old_memcg);
 452	mem_cgroup_put(memcg);
 453
 454	return ptr;
 455}
 456
 457void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 458				    size_t align, gfp_t flags)
 459{
 460	struct mem_cgroup *memcg, *old_memcg;
 461	void __percpu *ptr;
 462
 463	memcg = bpf_map_get_memcg(map);
 464	old_memcg = set_active_memcg(memcg);
 465	ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 466	set_active_memcg(old_memcg);
 467	mem_cgroup_put(memcg);
 468
 469	return ptr;
 470}
 471
 472#else
 473static void bpf_map_save_memcg(struct bpf_map *map)
 474{
 475}
 476
 477static void bpf_map_release_memcg(struct bpf_map *map)
 478{
 479}
 480#endif
 481
 482static int btf_field_cmp(const void *a, const void *b)
 483{
 484	const struct btf_field *f1 = a, *f2 = b;
 485
 486	if (f1->offset < f2->offset)
 487		return -1;
 488	else if (f1->offset > f2->offset)
 489		return 1;
 490	return 0;
 491}
 492
 493struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 494				  u32 field_mask)
 495{
 496	struct btf_field *field;
 497
 498	if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 499		return NULL;
 500	field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 501	if (!field || !(field->type & field_mask))
 502		return NULL;
 503	return field;
 504}
 505
 506void btf_record_free(struct btf_record *rec)
 507{
 508	int i;
 509
 510	if (IS_ERR_OR_NULL(rec))
 511		return;
 512	for (i = 0; i < rec->cnt; i++) {
 513		switch (rec->fields[i].type) {
 
 
 
 514		case BPF_KPTR_UNREF:
 515		case BPF_KPTR_REF:
 516		case BPF_KPTR_PERCPU:
 517			if (rec->fields[i].kptr.module)
 518				module_put(rec->fields[i].kptr.module);
 519			btf_put(rec->fields[i].kptr.btf);
 520			break;
 521		case BPF_LIST_HEAD:
 522		case BPF_LIST_NODE:
 523		case BPF_RB_ROOT:
 524		case BPF_RB_NODE:
 525		case BPF_SPIN_LOCK:
 526		case BPF_TIMER:
 527		case BPF_REFCOUNT:
 528			/* Nothing to release */
 529			break;
 530		default:
 531			WARN_ON_ONCE(1);
 532			continue;
 533		}
 534	}
 535	kfree(rec);
 536}
 537
 538void bpf_map_free_record(struct bpf_map *map)
 539{
 540	btf_record_free(map->record);
 541	map->record = NULL;
 542}
 543
 544struct btf_record *btf_record_dup(const struct btf_record *rec)
 545{
 546	const struct btf_field *fields;
 547	struct btf_record *new_rec;
 548	int ret, size, i;
 549
 550	if (IS_ERR_OR_NULL(rec))
 551		return NULL;
 552	size = offsetof(struct btf_record, fields[rec->cnt]);
 553	new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 554	if (!new_rec)
 555		return ERR_PTR(-ENOMEM);
 556	/* Do a deep copy of the btf_record */
 557	fields = rec->fields;
 558	new_rec->cnt = 0;
 559	for (i = 0; i < rec->cnt; i++) {
 560		switch (fields[i].type) {
 
 
 
 561		case BPF_KPTR_UNREF:
 562		case BPF_KPTR_REF:
 563		case BPF_KPTR_PERCPU:
 564			btf_get(fields[i].kptr.btf);
 565			if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 566				ret = -ENXIO;
 567				goto free;
 568			}
 569			break;
 570		case BPF_LIST_HEAD:
 571		case BPF_LIST_NODE:
 572		case BPF_RB_ROOT:
 573		case BPF_RB_NODE:
 574		case BPF_SPIN_LOCK:
 575		case BPF_TIMER:
 576		case BPF_REFCOUNT:
 577			/* Nothing to acquire */
 578			break;
 579		default:
 580			ret = -EFAULT;
 581			WARN_ON_ONCE(1);
 582			goto free;
 583		}
 584		new_rec->cnt++;
 585	}
 586	return new_rec;
 587free:
 588	btf_record_free(new_rec);
 589	return ERR_PTR(ret);
 590}
 591
 592bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 593{
 594	bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 595	int size;
 596
 597	if (!a_has_fields && !b_has_fields)
 598		return true;
 599	if (a_has_fields != b_has_fields)
 600		return false;
 601	if (rec_a->cnt != rec_b->cnt)
 602		return false;
 603	size = offsetof(struct btf_record, fields[rec_a->cnt]);
 604	/* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 605	 * members are zeroed out. So memcmp is safe to do without worrying
 606	 * about padding/unused fields.
 607	 *
 608	 * While spin_lock, timer, and kptr have no relation to map BTF,
 609	 * list_head metadata is specific to map BTF, the btf and value_rec
 610	 * members in particular. btf is the map BTF, while value_rec points to
 611	 * btf_record in that map BTF.
 612	 *
 613	 * So while by default, we don't rely on the map BTF (which the records
 614	 * were parsed from) matching for both records, which is not backwards
 615	 * compatible, in case list_head is part of it, we implicitly rely on
 616	 * that by way of depending on memcmp succeeding for it.
 617	 */
 618	return !memcmp(rec_a, rec_b, size);
 619}
 620
 621void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 622{
 623	if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 624		return;
 625	bpf_timer_cancel_and_free(obj + rec->timer_off);
 626}
 627
 628void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 629{
 630	const struct btf_field *fields;
 631	int i;
 632
 633	if (IS_ERR_OR_NULL(rec))
 634		return;
 635	fields = rec->fields;
 636	for (i = 0; i < rec->cnt; i++) {
 637		struct btf_struct_meta *pointee_struct_meta;
 638		const struct btf_field *field = &fields[i];
 639		void *field_ptr = obj + field->offset;
 640		void *xchgd_field;
 641
 642		switch (fields[i].type) {
 643		case BPF_SPIN_LOCK:
 644			break;
 645		case BPF_TIMER:
 646			bpf_timer_cancel_and_free(field_ptr);
 647			break;
 648		case BPF_KPTR_UNREF:
 649			WRITE_ONCE(*(u64 *)field_ptr, 0);
 650			break;
 651		case BPF_KPTR_REF:
 652		case BPF_KPTR_PERCPU:
 653			xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 654			if (!xchgd_field)
 655				break;
 656
 657			if (!btf_is_kernel(field->kptr.btf)) {
 658				pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 659									   field->kptr.btf_id);
 660				migrate_disable();
 661				__bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 662								 pointee_struct_meta->record : NULL,
 663								 fields[i].type == BPF_KPTR_PERCPU);
 664				migrate_enable();
 665			} else {
 666				field->kptr.dtor(xchgd_field);
 667			}
 668			break;
 669		case BPF_LIST_HEAD:
 670			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 671				continue;
 672			bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 673			break;
 674		case BPF_RB_ROOT:
 675			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 676				continue;
 677			bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 678			break;
 679		case BPF_LIST_NODE:
 680		case BPF_RB_NODE:
 681		case BPF_REFCOUNT:
 682			break;
 683		default:
 684			WARN_ON_ONCE(1);
 685			continue;
 686		}
 687	}
 688}
 689
 690/* called from workqueue */
 691static void bpf_map_free_deferred(struct work_struct *work)
 692{
 693	struct bpf_map *map = container_of(work, struct bpf_map, work);
 
 694	struct btf_record *rec = map->record;
 695	struct btf *btf = map->btf;
 696
 697	security_bpf_map_free(map);
 698	bpf_map_release_memcg(map);
 699	/* implementation dependent freeing */
 700	map->ops->map_free(map);
 701	/* Delay freeing of btf_record for maps, as map_free
 702	 * callback usually needs access to them. It is better to do it here
 703	 * than require each callback to do the free itself manually.
 704	 *
 705	 * Note that the btf_record stashed in map->inner_map_meta->record was
 706	 * already freed using the map_free callback for map in map case which
 707	 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 708	 * template bpf_map struct used during verification.
 709	 */
 
 710	btf_record_free(rec);
 711	/* Delay freeing of btf for maps, as map_free callback may need
 712	 * struct_meta info which will be freed with btf_put().
 713	 */
 714	btf_put(btf);
 715}
 716
 717static void bpf_map_put_uref(struct bpf_map *map)
 718{
 719	if (atomic64_dec_and_test(&map->usercnt)) {
 720		if (map->ops->map_release_uref)
 721			map->ops->map_release_uref(map);
 722	}
 723}
 724
 725static void bpf_map_free_in_work(struct bpf_map *map)
 726{
 727	INIT_WORK(&map->work, bpf_map_free_deferred);
 728	/* Avoid spawning kworkers, since they all might contend
 729	 * for the same mutex like slab_mutex.
 730	 */
 731	queue_work(system_unbound_wq, &map->work);
 732}
 733
 734static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
 735{
 736	bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
 737}
 738
 739static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
 740{
 741	if (rcu_trace_implies_rcu_gp())
 742		bpf_map_free_rcu_gp(rcu);
 743	else
 744		call_rcu(rcu, bpf_map_free_rcu_gp);
 745}
 746
 747/* decrement map refcnt and schedule it for freeing via workqueue
 748 * (underlying map implementation ops->map_free() might sleep)
 749 */
 750void bpf_map_put(struct bpf_map *map)
 751{
 752	if (atomic64_dec_and_test(&map->refcnt)) {
 753		/* bpf_map_free_id() must be called first */
 754		bpf_map_free_id(map);
 755
 756		WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
 757		if (READ_ONCE(map->free_after_mult_rcu_gp))
 758			call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
 759		else if (READ_ONCE(map->free_after_rcu_gp))
 760			call_rcu(&map->rcu, bpf_map_free_rcu_gp);
 761		else
 762			bpf_map_free_in_work(map);
 763	}
 764}
 
 
 
 
 
 765EXPORT_SYMBOL_GPL(bpf_map_put);
 766
 767void bpf_map_put_with_uref(struct bpf_map *map)
 768{
 769	bpf_map_put_uref(map);
 770	bpf_map_put(map);
 771}
 772
 773static int bpf_map_release(struct inode *inode, struct file *filp)
 774{
 775	struct bpf_map *map = filp->private_data;
 776
 777	if (map->ops->map_release)
 778		map->ops->map_release(map, filp);
 779
 780	bpf_map_put_with_uref(map);
 781	return 0;
 782}
 783
 784static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 785{
 786	fmode_t mode = f.file->f_mode;
 787
 788	/* Our file permissions may have been overridden by global
 789	 * map permissions facing syscall side.
 790	 */
 791	if (READ_ONCE(map->frozen))
 792		mode &= ~FMODE_CAN_WRITE;
 793	return mode;
 794}
 795
 796#ifdef CONFIG_PROC_FS
 797/* Show the memory usage of a bpf map */
 798static u64 bpf_map_memory_usage(const struct bpf_map *map)
 
 
 
 799{
 800	return map->ops->map_mem_usage(map);
 
 
 
 
 801}
 802
 803static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 804{
 805	struct bpf_map *map = filp->private_data;
 806	u32 type = 0, jited = 0;
 807
 808	if (map_type_contains_progs(map)) {
 809		spin_lock(&map->owner.lock);
 810		type  = map->owner.type;
 811		jited = map->owner.jited;
 812		spin_unlock(&map->owner.lock);
 813	}
 814
 815	seq_printf(m,
 816		   "map_type:\t%u\n"
 817		   "key_size:\t%u\n"
 818		   "value_size:\t%u\n"
 819		   "max_entries:\t%u\n"
 820		   "map_flags:\t%#x\n"
 821		   "map_extra:\t%#llx\n"
 822		   "memlock:\t%llu\n"
 823		   "map_id:\t%u\n"
 824		   "frozen:\t%u\n",
 825		   map->map_type,
 826		   map->key_size,
 827		   map->value_size,
 828		   map->max_entries,
 829		   map->map_flags,
 830		   (unsigned long long)map->map_extra,
 831		   bpf_map_memory_usage(map),
 832		   map->id,
 833		   READ_ONCE(map->frozen));
 834	if (type) {
 835		seq_printf(m, "owner_prog_type:\t%u\n", type);
 836		seq_printf(m, "owner_jited:\t%u\n", jited);
 837	}
 838}
 839#endif
 840
 841static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 842			      loff_t *ppos)
 843{
 844	/* We need this handler such that alloc_file() enables
 845	 * f_mode with FMODE_CAN_READ.
 846	 */
 847	return -EINVAL;
 848}
 849
 850static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 851			       size_t siz, loff_t *ppos)
 852{
 853	/* We need this handler such that alloc_file() enables
 854	 * f_mode with FMODE_CAN_WRITE.
 855	 */
 856	return -EINVAL;
 857}
 858
 859/* called for any extra memory-mapped regions (except initial) */
 860static void bpf_map_mmap_open(struct vm_area_struct *vma)
 861{
 862	struct bpf_map *map = vma->vm_file->private_data;
 863
 864	if (vma->vm_flags & VM_MAYWRITE)
 865		bpf_map_write_active_inc(map);
 866}
 867
 868/* called for all unmapped memory region (including initial) */
 869static void bpf_map_mmap_close(struct vm_area_struct *vma)
 870{
 871	struct bpf_map *map = vma->vm_file->private_data;
 872
 873	if (vma->vm_flags & VM_MAYWRITE)
 874		bpf_map_write_active_dec(map);
 875}
 876
 877static const struct vm_operations_struct bpf_map_default_vmops = {
 878	.open		= bpf_map_mmap_open,
 879	.close		= bpf_map_mmap_close,
 880};
 881
 882static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 883{
 884	struct bpf_map *map = filp->private_data;
 885	int err;
 886
 887	if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 888		return -ENOTSUPP;
 889
 890	if (!(vma->vm_flags & VM_SHARED))
 891		return -EINVAL;
 892
 893	mutex_lock(&map->freeze_mutex);
 894
 895	if (vma->vm_flags & VM_WRITE) {
 896		if (map->frozen) {
 897			err = -EPERM;
 898			goto out;
 899		}
 900		/* map is meant to be read-only, so do not allow mapping as
 901		 * writable, because it's possible to leak a writable page
 902		 * reference and allows user-space to still modify it after
 903		 * freezing, while verifier will assume contents do not change
 904		 */
 905		if (map->map_flags & BPF_F_RDONLY_PROG) {
 906			err = -EACCES;
 907			goto out;
 908		}
 909	}
 910
 911	/* set default open/close callbacks */
 912	vma->vm_ops = &bpf_map_default_vmops;
 913	vma->vm_private_data = map;
 914	vm_flags_clear(vma, VM_MAYEXEC);
 915	if (!(vma->vm_flags & VM_WRITE))
 916		/* disallow re-mapping with PROT_WRITE */
 917		vm_flags_clear(vma, VM_MAYWRITE);
 918
 919	err = map->ops->map_mmap(map, vma);
 920	if (err)
 921		goto out;
 922
 923	if (vma->vm_flags & VM_MAYWRITE)
 924		bpf_map_write_active_inc(map);
 925out:
 926	mutex_unlock(&map->freeze_mutex);
 927	return err;
 928}
 929
 930static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 931{
 932	struct bpf_map *map = filp->private_data;
 933
 934	if (map->ops->map_poll)
 935		return map->ops->map_poll(map, filp, pts);
 936
 937	return EPOLLERR;
 938}
 939
 940const struct file_operations bpf_map_fops = {
 941#ifdef CONFIG_PROC_FS
 942	.show_fdinfo	= bpf_map_show_fdinfo,
 943#endif
 944	.release	= bpf_map_release,
 945	.read		= bpf_dummy_read,
 946	.write		= bpf_dummy_write,
 947	.mmap		= bpf_map_mmap,
 948	.poll		= bpf_map_poll,
 949};
 950
 951int bpf_map_new_fd(struct bpf_map *map, int flags)
 952{
 953	int ret;
 954
 955	ret = security_bpf_map(map, OPEN_FMODE(flags));
 956	if (ret < 0)
 957		return ret;
 958
 959	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 960				flags | O_CLOEXEC);
 961}
 962
 963int bpf_get_file_flag(int flags)
 964{
 965	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 966		return -EINVAL;
 967	if (flags & BPF_F_RDONLY)
 968		return O_RDONLY;
 969	if (flags & BPF_F_WRONLY)
 970		return O_WRONLY;
 971	return O_RDWR;
 972}
 973
 974/* helper macro to check that unused fields 'union bpf_attr' are zero */
 975#define CHECK_ATTR(CMD) \
 976	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 977		   sizeof(attr->CMD##_LAST_FIELD), 0, \
 978		   sizeof(*attr) - \
 979		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 980		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
 981
 982/* dst and src must have at least "size" number of bytes.
 983 * Return strlen on success and < 0 on error.
 984 */
 985int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 986{
 987	const char *end = src + size;
 988	const char *orig_src = src;
 989
 990	memset(dst, 0, size);
 991	/* Copy all isalnum(), '_' and '.' chars. */
 992	while (src < end && *src) {
 993		if (!isalnum(*src) &&
 994		    *src != '_' && *src != '.')
 995			return -EINVAL;
 996		*dst++ = *src++;
 997	}
 998
 999	/* No '\0' found in "size" number of bytes */
1000	if (src == end)
1001		return -EINVAL;
1002
1003	return src - orig_src;
1004}
1005
1006int map_check_no_btf(const struct bpf_map *map,
1007		     const struct btf *btf,
1008		     const struct btf_type *key_type,
1009		     const struct btf_type *value_type)
1010{
1011	return -ENOTSUPP;
1012}
1013
1014static int map_check_btf(struct bpf_map *map, const struct btf *btf,
1015			 u32 btf_key_id, u32 btf_value_id)
1016{
1017	const struct btf_type *key_type, *value_type;
1018	u32 key_size, value_size;
1019	int ret = 0;
1020
1021	/* Some maps allow key to be unspecified. */
1022	if (btf_key_id) {
1023		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1024		if (!key_type || key_size != map->key_size)
1025			return -EINVAL;
1026	} else {
1027		key_type = btf_type_by_id(btf, 0);
1028		if (!map->ops->map_check_btf)
1029			return -EINVAL;
1030	}
1031
1032	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1033	if (!value_type || value_size != map->value_size)
1034		return -EINVAL;
1035
1036	map->record = btf_parse_fields(btf, value_type,
1037				       BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1038				       BPF_RB_ROOT | BPF_REFCOUNT,
1039				       map->value_size);
1040	if (!IS_ERR_OR_NULL(map->record)) {
1041		int i;
1042
1043		if (!bpf_capable()) {
1044			ret = -EPERM;
1045			goto free_map_tab;
1046		}
1047		if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1048			ret = -EACCES;
1049			goto free_map_tab;
1050		}
1051		for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1052			switch (map->record->field_mask & (1 << i)) {
1053			case 0:
1054				continue;
1055			case BPF_SPIN_LOCK:
1056				if (map->map_type != BPF_MAP_TYPE_HASH &&
1057				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1058				    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1059				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1060				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1061				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1062				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1063					ret = -EOPNOTSUPP;
1064					goto free_map_tab;
1065				}
1066				break;
1067			case BPF_TIMER:
1068				if (map->map_type != BPF_MAP_TYPE_HASH &&
1069				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1070				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1071					ret = -EOPNOTSUPP;
1072					goto free_map_tab;
1073				}
1074				break;
1075			case BPF_KPTR_UNREF:
1076			case BPF_KPTR_REF:
1077			case BPF_KPTR_PERCPU:
1078			case BPF_REFCOUNT:
1079				if (map->map_type != BPF_MAP_TYPE_HASH &&
1080				    map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1081				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1082				    map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1083				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1084				    map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1085				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1086				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1087				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1088				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1089					ret = -EOPNOTSUPP;
1090					goto free_map_tab;
1091				}
1092				break;
1093			case BPF_LIST_HEAD:
1094			case BPF_RB_ROOT:
1095				if (map->map_type != BPF_MAP_TYPE_HASH &&
1096				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1097				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1098					ret = -EOPNOTSUPP;
1099					goto free_map_tab;
1100				}
1101				break;
1102			default:
1103				/* Fail if map_type checks are missing for a field type */
1104				ret = -EOPNOTSUPP;
1105				goto free_map_tab;
1106			}
1107		}
1108	}
1109
1110	ret = btf_check_and_fixup_fields(btf, map->record);
1111	if (ret < 0)
1112		goto free_map_tab;
1113
1114	if (map->ops->map_check_btf) {
1115		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1116		if (ret < 0)
1117			goto free_map_tab;
1118	}
1119
1120	return ret;
1121free_map_tab:
1122	bpf_map_free_record(map);
1123	return ret;
1124}
1125
1126#define BPF_MAP_CREATE_LAST_FIELD map_extra
1127/* called via syscall */
1128static int map_create(union bpf_attr *attr)
1129{
1130	const struct bpf_map_ops *ops;
1131	int numa_node = bpf_map_attr_numa_node(attr);
1132	u32 map_type = attr->map_type;
1133	struct bpf_map *map;
1134	int f_flags;
1135	int err;
1136
1137	err = CHECK_ATTR(BPF_MAP_CREATE);
1138	if (err)
1139		return -EINVAL;
1140
1141	if (attr->btf_vmlinux_value_type_id) {
1142		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1143		    attr->btf_key_type_id || attr->btf_value_type_id)
1144			return -EINVAL;
1145	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1146		return -EINVAL;
1147	}
1148
1149	if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1150	    attr->map_extra != 0)
1151		return -EINVAL;
1152
1153	f_flags = bpf_get_file_flag(attr->map_flags);
1154	if (f_flags < 0)
1155		return f_flags;
1156
1157	if (numa_node != NUMA_NO_NODE &&
1158	    ((unsigned int)numa_node >= nr_node_ids ||
1159	     !node_online(numa_node)))
1160		return -EINVAL;
1161
1162	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1163	map_type = attr->map_type;
1164	if (map_type >= ARRAY_SIZE(bpf_map_types))
1165		return -EINVAL;
1166	map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1167	ops = bpf_map_types[map_type];
1168	if (!ops)
1169		return -EINVAL;
1170
1171	if (ops->map_alloc_check) {
1172		err = ops->map_alloc_check(attr);
1173		if (err)
1174			return err;
1175	}
1176	if (attr->map_ifindex)
1177		ops = &bpf_map_offload_ops;
1178	if (!ops->map_mem_usage)
1179		return -EINVAL;
1180
1181	/* Intent here is for unprivileged_bpf_disabled to block BPF map
1182	 * creation for unprivileged users; other actions depend
1183	 * on fd availability and access to bpffs, so are dependent on
1184	 * object creation success. Even with unprivileged BPF disabled,
1185	 * capability checks are still carried out.
1186	 */
1187	if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
1188		return -EPERM;
1189
1190	/* check privileged map type permissions */
1191	switch (map_type) {
1192	case BPF_MAP_TYPE_ARRAY:
1193	case BPF_MAP_TYPE_PERCPU_ARRAY:
1194	case BPF_MAP_TYPE_PROG_ARRAY:
1195	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1196	case BPF_MAP_TYPE_CGROUP_ARRAY:
1197	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1198	case BPF_MAP_TYPE_HASH:
1199	case BPF_MAP_TYPE_PERCPU_HASH:
1200	case BPF_MAP_TYPE_HASH_OF_MAPS:
1201	case BPF_MAP_TYPE_RINGBUF:
1202	case BPF_MAP_TYPE_USER_RINGBUF:
1203	case BPF_MAP_TYPE_CGROUP_STORAGE:
1204	case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1205		/* unprivileged */
1206		break;
1207	case BPF_MAP_TYPE_SK_STORAGE:
1208	case BPF_MAP_TYPE_INODE_STORAGE:
1209	case BPF_MAP_TYPE_TASK_STORAGE:
1210	case BPF_MAP_TYPE_CGRP_STORAGE:
1211	case BPF_MAP_TYPE_BLOOM_FILTER:
1212	case BPF_MAP_TYPE_LPM_TRIE:
1213	case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1214	case BPF_MAP_TYPE_STACK_TRACE:
1215	case BPF_MAP_TYPE_QUEUE:
1216	case BPF_MAP_TYPE_STACK:
1217	case BPF_MAP_TYPE_LRU_HASH:
1218	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1219	case BPF_MAP_TYPE_STRUCT_OPS:
1220	case BPF_MAP_TYPE_CPUMAP:
1221		if (!bpf_capable())
1222			return -EPERM;
1223		break;
1224	case BPF_MAP_TYPE_SOCKMAP:
1225	case BPF_MAP_TYPE_SOCKHASH:
1226	case BPF_MAP_TYPE_DEVMAP:
1227	case BPF_MAP_TYPE_DEVMAP_HASH:
1228	case BPF_MAP_TYPE_XSKMAP:
1229		if (!capable(CAP_NET_ADMIN))
1230			return -EPERM;
1231		break;
1232	default:
1233		WARN(1, "unsupported map type %d", map_type);
1234		return -EPERM;
1235	}
1236
1237	map = ops->map_alloc(attr);
1238	if (IS_ERR(map))
1239		return PTR_ERR(map);
1240	map->ops = ops;
1241	map->map_type = map_type;
1242
1243	err = bpf_obj_name_cpy(map->name, attr->map_name,
1244			       sizeof(attr->map_name));
1245	if (err < 0)
1246		goto free_map;
1247
1248	atomic64_set(&map->refcnt, 1);
1249	atomic64_set(&map->usercnt, 1);
1250	mutex_init(&map->freeze_mutex);
1251	spin_lock_init(&map->owner.lock);
1252
1253	if (attr->btf_key_type_id || attr->btf_value_type_id ||
1254	    /* Even the map's value is a kernel's struct,
1255	     * the bpf_prog.o must have BTF to begin with
1256	     * to figure out the corresponding kernel's
1257	     * counter part.  Thus, attr->btf_fd has
1258	     * to be valid also.
1259	     */
1260	    attr->btf_vmlinux_value_type_id) {
1261		struct btf *btf;
1262
1263		btf = btf_get_by_fd(attr->btf_fd);
1264		if (IS_ERR(btf)) {
1265			err = PTR_ERR(btf);
1266			goto free_map;
1267		}
1268		if (btf_is_kernel(btf)) {
1269			btf_put(btf);
1270			err = -EACCES;
1271			goto free_map;
1272		}
1273		map->btf = btf;
1274
1275		if (attr->btf_value_type_id) {
1276			err = map_check_btf(map, btf, attr->btf_key_type_id,
1277					    attr->btf_value_type_id);
1278			if (err)
1279				goto free_map;
1280		}
1281
1282		map->btf_key_type_id = attr->btf_key_type_id;
1283		map->btf_value_type_id = attr->btf_value_type_id;
1284		map->btf_vmlinux_value_type_id =
1285			attr->btf_vmlinux_value_type_id;
1286	}
1287
 
 
 
 
 
 
 
 
1288	err = security_bpf_map_alloc(map);
1289	if (err)
1290		goto free_map;
1291
1292	err = bpf_map_alloc_id(map);
1293	if (err)
1294		goto free_map_sec;
1295
1296	bpf_map_save_memcg(map);
1297
1298	err = bpf_map_new_fd(map, f_flags);
1299	if (err < 0) {
1300		/* failed to allocate fd.
1301		 * bpf_map_put_with_uref() is needed because the above
1302		 * bpf_map_alloc_id() has published the map
1303		 * to the userspace and the userspace may
1304		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1305		 */
1306		bpf_map_put_with_uref(map);
1307		return err;
1308	}
1309
1310	return err;
1311
1312free_map_sec:
1313	security_bpf_map_free(map);
 
 
1314free_map:
1315	btf_put(map->btf);
1316	map->ops->map_free(map);
1317	return err;
1318}
1319
1320/* if error is returned, fd is released.
1321 * On success caller should complete fd access with matching fdput()
1322 */
1323struct bpf_map *__bpf_map_get(struct fd f)
1324{
1325	if (!f.file)
1326		return ERR_PTR(-EBADF);
1327	if (f.file->f_op != &bpf_map_fops) {
1328		fdput(f);
1329		return ERR_PTR(-EINVAL);
1330	}
1331
1332	return f.file->private_data;
1333}
1334
1335void bpf_map_inc(struct bpf_map *map)
1336{
1337	atomic64_inc(&map->refcnt);
1338}
1339EXPORT_SYMBOL_GPL(bpf_map_inc);
1340
1341void bpf_map_inc_with_uref(struct bpf_map *map)
1342{
1343	atomic64_inc(&map->refcnt);
1344	atomic64_inc(&map->usercnt);
1345}
1346EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1347
1348struct bpf_map *bpf_map_get(u32 ufd)
1349{
1350	struct fd f = fdget(ufd);
1351	struct bpf_map *map;
1352
1353	map = __bpf_map_get(f);
1354	if (IS_ERR(map))
1355		return map;
1356
1357	bpf_map_inc(map);
1358	fdput(f);
1359
1360	return map;
1361}
1362EXPORT_SYMBOL(bpf_map_get);
1363
1364struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1365{
1366	struct fd f = fdget(ufd);
1367	struct bpf_map *map;
1368
1369	map = __bpf_map_get(f);
1370	if (IS_ERR(map))
1371		return map;
1372
1373	bpf_map_inc_with_uref(map);
1374	fdput(f);
1375
1376	return map;
1377}
1378
1379/* map_idr_lock should have been held or the map should have been
1380 * protected by rcu read lock.
1381 */
1382struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1383{
1384	int refold;
1385
1386	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1387	if (!refold)
1388		return ERR_PTR(-ENOENT);
1389	if (uref)
1390		atomic64_inc(&map->usercnt);
1391
1392	return map;
1393}
1394
1395struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1396{
1397	spin_lock_bh(&map_idr_lock);
1398	map = __bpf_map_inc_not_zero(map, false);
1399	spin_unlock_bh(&map_idr_lock);
1400
1401	return map;
1402}
1403EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1404
1405int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1406{
1407	return -ENOTSUPP;
1408}
1409
1410static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1411{
1412	if (key_size)
1413		return vmemdup_user(ukey, key_size);
1414
1415	if (ukey)
1416		return ERR_PTR(-EINVAL);
1417
1418	return NULL;
1419}
1420
1421static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1422{
1423	if (key_size)
1424		return kvmemdup_bpfptr(ukey, key_size);
1425
1426	if (!bpfptr_is_null(ukey))
1427		return ERR_PTR(-EINVAL);
1428
1429	return NULL;
1430}
1431
1432/* last field in 'union bpf_attr' used by this command */
1433#define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1434
1435static int map_lookup_elem(union bpf_attr *attr)
1436{
1437	void __user *ukey = u64_to_user_ptr(attr->key);
1438	void __user *uvalue = u64_to_user_ptr(attr->value);
1439	int ufd = attr->map_fd;
1440	struct bpf_map *map;
1441	void *key, *value;
1442	u32 value_size;
1443	struct fd f;
1444	int err;
1445
1446	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1447		return -EINVAL;
1448
1449	if (attr->flags & ~BPF_F_LOCK)
1450		return -EINVAL;
1451
1452	f = fdget(ufd);
1453	map = __bpf_map_get(f);
1454	if (IS_ERR(map))
1455		return PTR_ERR(map);
1456	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1457		err = -EPERM;
1458		goto err_put;
1459	}
1460
1461	if ((attr->flags & BPF_F_LOCK) &&
1462	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1463		err = -EINVAL;
1464		goto err_put;
1465	}
1466
1467	key = __bpf_copy_key(ukey, map->key_size);
1468	if (IS_ERR(key)) {
1469		err = PTR_ERR(key);
1470		goto err_put;
1471	}
1472
1473	value_size = bpf_map_value_size(map);
1474
1475	err = -ENOMEM;
1476	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1477	if (!value)
1478		goto free_key;
1479
1480	if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1481		if (copy_from_user(value, uvalue, value_size))
1482			err = -EFAULT;
1483		else
1484			err = bpf_map_copy_value(map, key, value, attr->flags);
1485		goto free_value;
1486	}
1487
1488	err = bpf_map_copy_value(map, key, value, attr->flags);
1489	if (err)
1490		goto free_value;
1491
1492	err = -EFAULT;
1493	if (copy_to_user(uvalue, value, value_size) != 0)
1494		goto free_value;
1495
1496	err = 0;
1497
1498free_value:
1499	kvfree(value);
1500free_key:
1501	kvfree(key);
1502err_put:
1503	fdput(f);
1504	return err;
1505}
1506
1507
1508#define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1509
1510static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1511{
1512	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1513	bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1514	int ufd = attr->map_fd;
1515	struct bpf_map *map;
1516	void *key, *value;
1517	u32 value_size;
1518	struct fd f;
1519	int err;
1520
1521	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1522		return -EINVAL;
1523
1524	f = fdget(ufd);
1525	map = __bpf_map_get(f);
1526	if (IS_ERR(map))
1527		return PTR_ERR(map);
1528	bpf_map_write_active_inc(map);
1529	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1530		err = -EPERM;
1531		goto err_put;
1532	}
1533
1534	if ((attr->flags & BPF_F_LOCK) &&
1535	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1536		err = -EINVAL;
1537		goto err_put;
1538	}
1539
1540	key = ___bpf_copy_key(ukey, map->key_size);
1541	if (IS_ERR(key)) {
1542		err = PTR_ERR(key);
1543		goto err_put;
1544	}
1545
1546	value_size = bpf_map_value_size(map);
1547	value = kvmemdup_bpfptr(uvalue, value_size);
1548	if (IS_ERR(value)) {
1549		err = PTR_ERR(value);
1550		goto free_key;
1551	}
1552
1553	err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1554	if (!err)
1555		maybe_wait_bpf_programs(map);
1556
1557	kvfree(value);
1558free_key:
1559	kvfree(key);
1560err_put:
1561	bpf_map_write_active_dec(map);
1562	fdput(f);
1563	return err;
1564}
1565
1566#define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1567
1568static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1569{
1570	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1571	int ufd = attr->map_fd;
1572	struct bpf_map *map;
1573	struct fd f;
1574	void *key;
1575	int err;
1576
1577	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1578		return -EINVAL;
1579
1580	f = fdget(ufd);
1581	map = __bpf_map_get(f);
1582	if (IS_ERR(map))
1583		return PTR_ERR(map);
1584	bpf_map_write_active_inc(map);
1585	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1586		err = -EPERM;
1587		goto err_put;
1588	}
1589
1590	key = ___bpf_copy_key(ukey, map->key_size);
1591	if (IS_ERR(key)) {
1592		err = PTR_ERR(key);
1593		goto err_put;
1594	}
1595
1596	if (bpf_map_is_offloaded(map)) {
1597		err = bpf_map_offload_delete_elem(map, key);
1598		goto out;
1599	} else if (IS_FD_PROG_ARRAY(map) ||
1600		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1601		/* These maps require sleepable context */
1602		err = map->ops->map_delete_elem(map, key);
1603		goto out;
1604	}
1605
1606	bpf_disable_instrumentation();
1607	rcu_read_lock();
1608	err = map->ops->map_delete_elem(map, key);
1609	rcu_read_unlock();
1610	bpf_enable_instrumentation();
1611	if (!err)
1612		maybe_wait_bpf_programs(map);
1613out:
1614	kvfree(key);
1615err_put:
1616	bpf_map_write_active_dec(map);
1617	fdput(f);
1618	return err;
1619}
1620
1621/* last field in 'union bpf_attr' used by this command */
1622#define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1623
1624static int map_get_next_key(union bpf_attr *attr)
1625{
1626	void __user *ukey = u64_to_user_ptr(attr->key);
1627	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1628	int ufd = attr->map_fd;
1629	struct bpf_map *map;
1630	void *key, *next_key;
1631	struct fd f;
1632	int err;
1633
1634	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1635		return -EINVAL;
1636
1637	f = fdget(ufd);
1638	map = __bpf_map_get(f);
1639	if (IS_ERR(map))
1640		return PTR_ERR(map);
1641	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1642		err = -EPERM;
1643		goto err_put;
1644	}
1645
1646	if (ukey) {
1647		key = __bpf_copy_key(ukey, map->key_size);
1648		if (IS_ERR(key)) {
1649			err = PTR_ERR(key);
1650			goto err_put;
1651		}
1652	} else {
1653		key = NULL;
1654	}
1655
1656	err = -ENOMEM;
1657	next_key = kvmalloc(map->key_size, GFP_USER);
1658	if (!next_key)
1659		goto free_key;
1660
1661	if (bpf_map_is_offloaded(map)) {
1662		err = bpf_map_offload_get_next_key(map, key, next_key);
1663		goto out;
1664	}
1665
1666	rcu_read_lock();
1667	err = map->ops->map_get_next_key(map, key, next_key);
1668	rcu_read_unlock();
1669out:
1670	if (err)
1671		goto free_next_key;
1672
1673	err = -EFAULT;
1674	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1675		goto free_next_key;
1676
1677	err = 0;
1678
1679free_next_key:
1680	kvfree(next_key);
1681free_key:
1682	kvfree(key);
1683err_put:
1684	fdput(f);
1685	return err;
1686}
1687
1688int generic_map_delete_batch(struct bpf_map *map,
1689			     const union bpf_attr *attr,
1690			     union bpf_attr __user *uattr)
1691{
1692	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1693	u32 cp, max_count;
1694	int err = 0;
1695	void *key;
1696
1697	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1698		return -EINVAL;
1699
1700	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1701	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1702		return -EINVAL;
1703	}
1704
1705	max_count = attr->batch.count;
1706	if (!max_count)
1707		return 0;
1708
1709	if (put_user(0, &uattr->batch.count))
1710		return -EFAULT;
1711
1712	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1713	if (!key)
1714		return -ENOMEM;
1715
1716	for (cp = 0; cp < max_count; cp++) {
1717		err = -EFAULT;
1718		if (copy_from_user(key, keys + cp * map->key_size,
1719				   map->key_size))
1720			break;
1721
1722		if (bpf_map_is_offloaded(map)) {
1723			err = bpf_map_offload_delete_elem(map, key);
1724			break;
1725		}
1726
1727		bpf_disable_instrumentation();
1728		rcu_read_lock();
1729		err = map->ops->map_delete_elem(map, key);
1730		rcu_read_unlock();
1731		bpf_enable_instrumentation();
1732		if (err)
1733			break;
1734		cond_resched();
1735	}
1736	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1737		err = -EFAULT;
1738
1739	kvfree(key);
1740
 
1741	return err;
1742}
1743
1744int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1745			     const union bpf_attr *attr,
1746			     union bpf_attr __user *uattr)
1747{
1748	void __user *values = u64_to_user_ptr(attr->batch.values);
1749	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1750	u32 value_size, cp, max_count;
1751	void *key, *value;
1752	int err = 0;
1753
1754	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1755		return -EINVAL;
1756
1757	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1758	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1759		return -EINVAL;
1760	}
1761
1762	value_size = bpf_map_value_size(map);
1763
1764	max_count = attr->batch.count;
1765	if (!max_count)
1766		return 0;
1767
1768	if (put_user(0, &uattr->batch.count))
1769		return -EFAULT;
1770
1771	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1772	if (!key)
1773		return -ENOMEM;
1774
1775	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1776	if (!value) {
1777		kvfree(key);
1778		return -ENOMEM;
1779	}
1780
1781	for (cp = 0; cp < max_count; cp++) {
1782		err = -EFAULT;
1783		if (copy_from_user(key, keys + cp * map->key_size,
1784		    map->key_size) ||
1785		    copy_from_user(value, values + cp * value_size, value_size))
1786			break;
1787
1788		err = bpf_map_update_value(map, map_file, key, value,
1789					   attr->batch.elem_flags);
1790
1791		if (err)
1792			break;
1793		cond_resched();
1794	}
1795
1796	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1797		err = -EFAULT;
1798
1799	kvfree(value);
1800	kvfree(key);
1801
1802	return err;
1803}
1804
1805#define MAP_LOOKUP_RETRIES 3
1806
1807int generic_map_lookup_batch(struct bpf_map *map,
1808				    const union bpf_attr *attr,
1809				    union bpf_attr __user *uattr)
1810{
1811	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1812	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1813	void __user *values = u64_to_user_ptr(attr->batch.values);
1814	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1815	void *buf, *buf_prevkey, *prev_key, *key, *value;
1816	int err, retry = MAP_LOOKUP_RETRIES;
1817	u32 value_size, cp, max_count;
1818
1819	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1820		return -EINVAL;
1821
1822	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1823	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1824		return -EINVAL;
1825
1826	value_size = bpf_map_value_size(map);
1827
1828	max_count = attr->batch.count;
1829	if (!max_count)
1830		return 0;
1831
1832	if (put_user(0, &uattr->batch.count))
1833		return -EFAULT;
1834
1835	buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1836	if (!buf_prevkey)
1837		return -ENOMEM;
1838
1839	buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1840	if (!buf) {
1841		kvfree(buf_prevkey);
1842		return -ENOMEM;
1843	}
1844
1845	err = -EFAULT;
1846	prev_key = NULL;
1847	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1848		goto free_buf;
1849	key = buf;
1850	value = key + map->key_size;
1851	if (ubatch)
1852		prev_key = buf_prevkey;
1853
1854	for (cp = 0; cp < max_count;) {
1855		rcu_read_lock();
1856		err = map->ops->map_get_next_key(map, prev_key, key);
1857		rcu_read_unlock();
1858		if (err)
1859			break;
1860		err = bpf_map_copy_value(map, key, value,
1861					 attr->batch.elem_flags);
1862
1863		if (err == -ENOENT) {
1864			if (retry) {
1865				retry--;
1866				continue;
1867			}
1868			err = -EINTR;
1869			break;
1870		}
1871
1872		if (err)
1873			goto free_buf;
1874
1875		if (copy_to_user(keys + cp * map->key_size, key,
1876				 map->key_size)) {
1877			err = -EFAULT;
1878			goto free_buf;
1879		}
1880		if (copy_to_user(values + cp * value_size, value, value_size)) {
1881			err = -EFAULT;
1882			goto free_buf;
1883		}
1884
1885		if (!prev_key)
1886			prev_key = buf_prevkey;
1887
1888		swap(prev_key, key);
1889		retry = MAP_LOOKUP_RETRIES;
1890		cp++;
1891		cond_resched();
1892	}
1893
1894	if (err == -EFAULT)
1895		goto free_buf;
1896
1897	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1898		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1899		err = -EFAULT;
1900
1901free_buf:
1902	kvfree(buf_prevkey);
1903	kvfree(buf);
1904	return err;
1905}
1906
1907#define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1908
1909static int map_lookup_and_delete_elem(union bpf_attr *attr)
1910{
1911	void __user *ukey = u64_to_user_ptr(attr->key);
1912	void __user *uvalue = u64_to_user_ptr(attr->value);
1913	int ufd = attr->map_fd;
1914	struct bpf_map *map;
1915	void *key, *value;
1916	u32 value_size;
1917	struct fd f;
1918	int err;
1919
1920	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1921		return -EINVAL;
1922
1923	if (attr->flags & ~BPF_F_LOCK)
1924		return -EINVAL;
1925
1926	f = fdget(ufd);
1927	map = __bpf_map_get(f);
1928	if (IS_ERR(map))
1929		return PTR_ERR(map);
1930	bpf_map_write_active_inc(map);
1931	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1932	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1933		err = -EPERM;
1934		goto err_put;
1935	}
1936
1937	if (attr->flags &&
1938	    (map->map_type == BPF_MAP_TYPE_QUEUE ||
1939	     map->map_type == BPF_MAP_TYPE_STACK)) {
1940		err = -EINVAL;
1941		goto err_put;
1942	}
1943
1944	if ((attr->flags & BPF_F_LOCK) &&
1945	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1946		err = -EINVAL;
1947		goto err_put;
1948	}
1949
1950	key = __bpf_copy_key(ukey, map->key_size);
1951	if (IS_ERR(key)) {
1952		err = PTR_ERR(key);
1953		goto err_put;
1954	}
1955
1956	value_size = bpf_map_value_size(map);
1957
1958	err = -ENOMEM;
1959	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1960	if (!value)
1961		goto free_key;
1962
1963	err = -ENOTSUPP;
1964	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1965	    map->map_type == BPF_MAP_TYPE_STACK) {
1966		err = map->ops->map_pop_elem(map, value);
1967	} else if (map->map_type == BPF_MAP_TYPE_HASH ||
1968		   map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1969		   map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1970		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1971		if (!bpf_map_is_offloaded(map)) {
1972			bpf_disable_instrumentation();
1973			rcu_read_lock();
1974			err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1975			rcu_read_unlock();
1976			bpf_enable_instrumentation();
1977		}
1978	}
1979
1980	if (err)
1981		goto free_value;
1982
1983	if (copy_to_user(uvalue, value, value_size) != 0) {
1984		err = -EFAULT;
1985		goto free_value;
1986	}
1987
1988	err = 0;
1989
1990free_value:
1991	kvfree(value);
1992free_key:
1993	kvfree(key);
1994err_put:
1995	bpf_map_write_active_dec(map);
1996	fdput(f);
1997	return err;
1998}
1999
2000#define BPF_MAP_FREEZE_LAST_FIELD map_fd
2001
2002static int map_freeze(const union bpf_attr *attr)
2003{
2004	int err = 0, ufd = attr->map_fd;
2005	struct bpf_map *map;
2006	struct fd f;
2007
2008	if (CHECK_ATTR(BPF_MAP_FREEZE))
2009		return -EINVAL;
2010
2011	f = fdget(ufd);
2012	map = __bpf_map_get(f);
2013	if (IS_ERR(map))
2014		return PTR_ERR(map);
2015
2016	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
2017		fdput(f);
2018		return -ENOTSUPP;
2019	}
2020
2021	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2022		fdput(f);
2023		return -EPERM;
2024	}
2025
2026	mutex_lock(&map->freeze_mutex);
2027	if (bpf_map_write_active(map)) {
2028		err = -EBUSY;
2029		goto err_put;
2030	}
2031	if (READ_ONCE(map->frozen)) {
2032		err = -EBUSY;
2033		goto err_put;
2034	}
 
 
 
 
2035
2036	WRITE_ONCE(map->frozen, true);
2037err_put:
2038	mutex_unlock(&map->freeze_mutex);
2039	fdput(f);
2040	return err;
2041}
2042
2043static const struct bpf_prog_ops * const bpf_prog_types[] = {
2044#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2045	[_id] = & _name ## _prog_ops,
2046#define BPF_MAP_TYPE(_id, _ops)
2047#define BPF_LINK_TYPE(_id, _name)
2048#include <linux/bpf_types.h>
2049#undef BPF_PROG_TYPE
2050#undef BPF_MAP_TYPE
2051#undef BPF_LINK_TYPE
2052};
2053
2054static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2055{
2056	const struct bpf_prog_ops *ops;
2057
2058	if (type >= ARRAY_SIZE(bpf_prog_types))
2059		return -EINVAL;
2060	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2061	ops = bpf_prog_types[type];
2062	if (!ops)
2063		return -EINVAL;
2064
2065	if (!bpf_prog_is_offloaded(prog->aux))
2066		prog->aux->ops = ops;
2067	else
2068		prog->aux->ops = &bpf_offload_prog_ops;
2069	prog->type = type;
2070	return 0;
2071}
2072
2073enum bpf_audit {
2074	BPF_AUDIT_LOAD,
2075	BPF_AUDIT_UNLOAD,
2076	BPF_AUDIT_MAX,
2077};
2078
2079static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2080	[BPF_AUDIT_LOAD]   = "LOAD",
2081	[BPF_AUDIT_UNLOAD] = "UNLOAD",
2082};
2083
2084static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2085{
2086	struct audit_context *ctx = NULL;
2087	struct audit_buffer *ab;
2088
2089	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2090		return;
2091	if (audit_enabled == AUDIT_OFF)
2092		return;
2093	if (!in_irq() && !irqs_disabled())
2094		ctx = audit_context();
2095	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2096	if (unlikely(!ab))
2097		return;
2098	audit_log_format(ab, "prog-id=%u op=%s",
2099			 prog->aux->id, bpf_audit_str[op]);
2100	audit_log_end(ab);
2101}
2102
2103static int bpf_prog_alloc_id(struct bpf_prog *prog)
2104{
2105	int id;
2106
2107	idr_preload(GFP_KERNEL);
2108	spin_lock_bh(&prog_idr_lock);
2109	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2110	if (id > 0)
2111		prog->aux->id = id;
2112	spin_unlock_bh(&prog_idr_lock);
2113	idr_preload_end();
2114
2115	/* id is in [1, INT_MAX) */
2116	if (WARN_ON_ONCE(!id))
2117		return -ENOSPC;
2118
2119	return id > 0 ? 0 : id;
2120}
2121
2122void bpf_prog_free_id(struct bpf_prog *prog)
2123{
2124	unsigned long flags;
2125
2126	/* cBPF to eBPF migrations are currently not in the idr store.
2127	 * Offloaded programs are removed from the store when their device
2128	 * disappears - even if someone grabs an fd to them they are unusable,
2129	 * simply waiting for refcnt to drop to be freed.
2130	 */
2131	if (!prog->aux->id)
2132		return;
2133
2134	spin_lock_irqsave(&prog_idr_lock, flags);
2135	idr_remove(&prog_idr, prog->aux->id);
2136	prog->aux->id = 0;
2137	spin_unlock_irqrestore(&prog_idr_lock, flags);
2138}
2139
2140static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2141{
2142	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2143
2144	kvfree(aux->func_info);
2145	kfree(aux->func_info_aux);
2146	free_uid(aux->user);
2147	security_bpf_prog_free(aux);
2148	bpf_prog_free(aux->prog);
2149}
2150
2151static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2152{
2153	bpf_prog_kallsyms_del_all(prog);
2154	btf_put(prog->aux->btf);
2155	module_put(prog->aux->mod);
2156	kvfree(prog->aux->jited_linfo);
2157	kvfree(prog->aux->linfo);
2158	kfree(prog->aux->kfunc_tab);
2159	if (prog->aux->attach_btf)
2160		btf_put(prog->aux->attach_btf);
2161
2162	if (deferred) {
2163		if (prog->aux->sleepable)
2164			call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2165		else
2166			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2167	} else {
2168		__bpf_prog_put_rcu(&prog->aux->rcu);
2169	}
2170}
2171
2172static void bpf_prog_put_deferred(struct work_struct *work)
2173{
2174	struct bpf_prog_aux *aux;
2175	struct bpf_prog *prog;
2176
2177	aux = container_of(work, struct bpf_prog_aux, work);
2178	prog = aux->prog;
2179	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2180	bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2181	bpf_prog_free_id(prog);
2182	__bpf_prog_put_noref(prog, true);
2183}
2184
2185static void __bpf_prog_put(struct bpf_prog *prog)
2186{
2187	struct bpf_prog_aux *aux = prog->aux;
2188
2189	if (atomic64_dec_and_test(&aux->refcnt)) {
2190		if (in_irq() || irqs_disabled()) {
2191			INIT_WORK(&aux->work, bpf_prog_put_deferred);
2192			schedule_work(&aux->work);
2193		} else {
2194			bpf_prog_put_deferred(&aux->work);
2195		}
2196	}
2197}
2198
2199void bpf_prog_put(struct bpf_prog *prog)
2200{
2201	__bpf_prog_put(prog);
2202}
2203EXPORT_SYMBOL_GPL(bpf_prog_put);
2204
2205static int bpf_prog_release(struct inode *inode, struct file *filp)
2206{
2207	struct bpf_prog *prog = filp->private_data;
2208
2209	bpf_prog_put(prog);
2210	return 0;
2211}
2212
2213struct bpf_prog_kstats {
2214	u64 nsecs;
2215	u64 cnt;
2216	u64 misses;
2217};
2218
2219void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2220{
2221	struct bpf_prog_stats *stats;
2222	unsigned int flags;
2223
2224	stats = this_cpu_ptr(prog->stats);
2225	flags = u64_stats_update_begin_irqsave(&stats->syncp);
2226	u64_stats_inc(&stats->misses);
2227	u64_stats_update_end_irqrestore(&stats->syncp, flags);
2228}
2229
2230static void bpf_prog_get_stats(const struct bpf_prog *prog,
2231			       struct bpf_prog_kstats *stats)
2232{
2233	u64 nsecs = 0, cnt = 0, misses = 0;
2234	int cpu;
2235
2236	for_each_possible_cpu(cpu) {
2237		const struct bpf_prog_stats *st;
2238		unsigned int start;
2239		u64 tnsecs, tcnt, tmisses;
2240
2241		st = per_cpu_ptr(prog->stats, cpu);
2242		do {
2243			start = u64_stats_fetch_begin(&st->syncp);
2244			tnsecs = u64_stats_read(&st->nsecs);
2245			tcnt = u64_stats_read(&st->cnt);
2246			tmisses = u64_stats_read(&st->misses);
2247		} while (u64_stats_fetch_retry(&st->syncp, start));
2248		nsecs += tnsecs;
2249		cnt += tcnt;
2250		misses += tmisses;
2251	}
2252	stats->nsecs = nsecs;
2253	stats->cnt = cnt;
2254	stats->misses = misses;
2255}
2256
2257#ifdef CONFIG_PROC_FS
2258static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2259{
2260	const struct bpf_prog *prog = filp->private_data;
2261	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2262	struct bpf_prog_kstats stats;
2263
2264	bpf_prog_get_stats(prog, &stats);
2265	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2266	seq_printf(m,
2267		   "prog_type:\t%u\n"
2268		   "prog_jited:\t%u\n"
2269		   "prog_tag:\t%s\n"
2270		   "memlock:\t%llu\n"
2271		   "prog_id:\t%u\n"
2272		   "run_time_ns:\t%llu\n"
2273		   "run_cnt:\t%llu\n"
2274		   "recursion_misses:\t%llu\n"
2275		   "verified_insns:\t%u\n",
2276		   prog->type,
2277		   prog->jited,
2278		   prog_tag,
2279		   prog->pages * 1ULL << PAGE_SHIFT,
2280		   prog->aux->id,
2281		   stats.nsecs,
2282		   stats.cnt,
2283		   stats.misses,
2284		   prog->aux->verified_insns);
2285}
2286#endif
2287
2288const struct file_operations bpf_prog_fops = {
2289#ifdef CONFIG_PROC_FS
2290	.show_fdinfo	= bpf_prog_show_fdinfo,
2291#endif
2292	.release	= bpf_prog_release,
2293	.read		= bpf_dummy_read,
2294	.write		= bpf_dummy_write,
2295};
2296
2297int bpf_prog_new_fd(struct bpf_prog *prog)
2298{
2299	int ret;
2300
2301	ret = security_bpf_prog(prog);
2302	if (ret < 0)
2303		return ret;
2304
2305	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2306				O_RDWR | O_CLOEXEC);
2307}
2308
2309static struct bpf_prog *____bpf_prog_get(struct fd f)
2310{
2311	if (!f.file)
2312		return ERR_PTR(-EBADF);
2313	if (f.file->f_op != &bpf_prog_fops) {
2314		fdput(f);
2315		return ERR_PTR(-EINVAL);
2316	}
2317
2318	return f.file->private_data;
2319}
2320
2321void bpf_prog_add(struct bpf_prog *prog, int i)
2322{
2323	atomic64_add(i, &prog->aux->refcnt);
2324}
2325EXPORT_SYMBOL_GPL(bpf_prog_add);
2326
2327void bpf_prog_sub(struct bpf_prog *prog, int i)
2328{
2329	/* Only to be used for undoing previous bpf_prog_add() in some
2330	 * error path. We still know that another entity in our call
2331	 * path holds a reference to the program, thus atomic_sub() can
2332	 * be safely used in such cases!
2333	 */
2334	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2335}
2336EXPORT_SYMBOL_GPL(bpf_prog_sub);
2337
2338void bpf_prog_inc(struct bpf_prog *prog)
2339{
2340	atomic64_inc(&prog->aux->refcnt);
2341}
2342EXPORT_SYMBOL_GPL(bpf_prog_inc);
2343
2344/* prog_idr_lock should have been held */
2345struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2346{
2347	int refold;
2348
2349	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2350
2351	if (!refold)
2352		return ERR_PTR(-ENOENT);
2353
2354	return prog;
2355}
2356EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2357
2358bool bpf_prog_get_ok(struct bpf_prog *prog,
2359			    enum bpf_prog_type *attach_type, bool attach_drv)
2360{
2361	/* not an attachment, just a refcount inc, always allow */
2362	if (!attach_type)
2363		return true;
2364
2365	if (prog->type != *attach_type)
2366		return false;
2367	if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2368		return false;
2369
2370	return true;
2371}
2372
2373static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2374				       bool attach_drv)
2375{
2376	struct fd f = fdget(ufd);
2377	struct bpf_prog *prog;
2378
2379	prog = ____bpf_prog_get(f);
2380	if (IS_ERR(prog))
2381		return prog;
2382	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2383		prog = ERR_PTR(-EINVAL);
2384		goto out;
2385	}
2386
2387	bpf_prog_inc(prog);
2388out:
2389	fdput(f);
2390	return prog;
2391}
2392
2393struct bpf_prog *bpf_prog_get(u32 ufd)
2394{
2395	return __bpf_prog_get(ufd, NULL, false);
2396}
2397
2398struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2399				       bool attach_drv)
2400{
2401	return __bpf_prog_get(ufd, &type, attach_drv);
2402}
2403EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2404
2405/* Initially all BPF programs could be loaded w/o specifying
2406 * expected_attach_type. Later for some of them specifying expected_attach_type
2407 * at load time became required so that program could be validated properly.
2408 * Programs of types that are allowed to be loaded both w/ and w/o (for
2409 * backward compatibility) expected_attach_type, should have the default attach
2410 * type assigned to expected_attach_type for the latter case, so that it can be
2411 * validated later at attach time.
2412 *
2413 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2414 * prog type requires it but has some attach types that have to be backward
2415 * compatible.
2416 */
2417static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2418{
2419	switch (attr->prog_type) {
2420	case BPF_PROG_TYPE_CGROUP_SOCK:
2421		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2422		 * exist so checking for non-zero is the way to go here.
2423		 */
2424		if (!attr->expected_attach_type)
2425			attr->expected_attach_type =
2426				BPF_CGROUP_INET_SOCK_CREATE;
2427		break;
2428	case BPF_PROG_TYPE_SK_REUSEPORT:
2429		if (!attr->expected_attach_type)
2430			attr->expected_attach_type =
2431				BPF_SK_REUSEPORT_SELECT;
2432		break;
2433	}
2434}
2435
2436static int
2437bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2438			   enum bpf_attach_type expected_attach_type,
2439			   struct btf *attach_btf, u32 btf_id,
2440			   struct bpf_prog *dst_prog)
2441{
2442	if (btf_id) {
2443		if (btf_id > BTF_MAX_TYPE)
2444			return -EINVAL;
2445
2446		if (!attach_btf && !dst_prog)
2447			return -EINVAL;
2448
2449		switch (prog_type) {
2450		case BPF_PROG_TYPE_TRACING:
2451		case BPF_PROG_TYPE_LSM:
2452		case BPF_PROG_TYPE_STRUCT_OPS:
2453		case BPF_PROG_TYPE_EXT:
2454			break;
2455		default:
2456			return -EINVAL;
2457		}
2458	}
2459
2460	if (attach_btf && (!btf_id || dst_prog))
2461		return -EINVAL;
2462
2463	if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2464	    prog_type != BPF_PROG_TYPE_EXT)
2465		return -EINVAL;
2466
2467	switch (prog_type) {
2468	case BPF_PROG_TYPE_CGROUP_SOCK:
2469		switch (expected_attach_type) {
2470		case BPF_CGROUP_INET_SOCK_CREATE:
2471		case BPF_CGROUP_INET_SOCK_RELEASE:
2472		case BPF_CGROUP_INET4_POST_BIND:
2473		case BPF_CGROUP_INET6_POST_BIND:
2474			return 0;
2475		default:
2476			return -EINVAL;
2477		}
2478	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2479		switch (expected_attach_type) {
2480		case BPF_CGROUP_INET4_BIND:
2481		case BPF_CGROUP_INET6_BIND:
2482		case BPF_CGROUP_INET4_CONNECT:
2483		case BPF_CGROUP_INET6_CONNECT:
2484		case BPF_CGROUP_UNIX_CONNECT:
2485		case BPF_CGROUP_INET4_GETPEERNAME:
2486		case BPF_CGROUP_INET6_GETPEERNAME:
2487		case BPF_CGROUP_UNIX_GETPEERNAME:
2488		case BPF_CGROUP_INET4_GETSOCKNAME:
2489		case BPF_CGROUP_INET6_GETSOCKNAME:
2490		case BPF_CGROUP_UNIX_GETSOCKNAME:
2491		case BPF_CGROUP_UDP4_SENDMSG:
2492		case BPF_CGROUP_UDP6_SENDMSG:
2493		case BPF_CGROUP_UNIX_SENDMSG:
2494		case BPF_CGROUP_UDP4_RECVMSG:
2495		case BPF_CGROUP_UDP6_RECVMSG:
2496		case BPF_CGROUP_UNIX_RECVMSG:
2497			return 0;
2498		default:
2499			return -EINVAL;
2500		}
2501	case BPF_PROG_TYPE_CGROUP_SKB:
2502		switch (expected_attach_type) {
2503		case BPF_CGROUP_INET_INGRESS:
2504		case BPF_CGROUP_INET_EGRESS:
2505			return 0;
2506		default:
2507			return -EINVAL;
2508		}
2509	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2510		switch (expected_attach_type) {
2511		case BPF_CGROUP_SETSOCKOPT:
2512		case BPF_CGROUP_GETSOCKOPT:
2513			return 0;
2514		default:
2515			return -EINVAL;
2516		}
2517	case BPF_PROG_TYPE_SK_LOOKUP:
2518		if (expected_attach_type == BPF_SK_LOOKUP)
2519			return 0;
2520		return -EINVAL;
2521	case BPF_PROG_TYPE_SK_REUSEPORT:
2522		switch (expected_attach_type) {
2523		case BPF_SK_REUSEPORT_SELECT:
2524		case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2525			return 0;
2526		default:
2527			return -EINVAL;
2528		}
2529	case BPF_PROG_TYPE_NETFILTER:
2530		if (expected_attach_type == BPF_NETFILTER)
2531			return 0;
2532		return -EINVAL;
2533	case BPF_PROG_TYPE_SYSCALL:
2534	case BPF_PROG_TYPE_EXT:
2535		if (expected_attach_type)
2536			return -EINVAL;
2537		fallthrough;
2538	default:
2539		return 0;
2540	}
2541}
2542
2543static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2544{
2545	switch (prog_type) {
2546	case BPF_PROG_TYPE_SCHED_CLS:
2547	case BPF_PROG_TYPE_SCHED_ACT:
2548	case BPF_PROG_TYPE_XDP:
2549	case BPF_PROG_TYPE_LWT_IN:
2550	case BPF_PROG_TYPE_LWT_OUT:
2551	case BPF_PROG_TYPE_LWT_XMIT:
2552	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2553	case BPF_PROG_TYPE_SK_SKB:
2554	case BPF_PROG_TYPE_SK_MSG:
 
2555	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2556	case BPF_PROG_TYPE_CGROUP_DEVICE:
2557	case BPF_PROG_TYPE_CGROUP_SOCK:
2558	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2559	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2560	case BPF_PROG_TYPE_CGROUP_SYSCTL:
2561	case BPF_PROG_TYPE_SOCK_OPS:
2562	case BPF_PROG_TYPE_EXT: /* extends any prog */
2563	case BPF_PROG_TYPE_NETFILTER:
2564		return true;
2565	case BPF_PROG_TYPE_CGROUP_SKB:
2566		/* always unpriv */
2567	case BPF_PROG_TYPE_SK_REUSEPORT:
2568		/* equivalent to SOCKET_FILTER. need CAP_BPF only */
2569	default:
2570		return false;
2571	}
2572}
2573
2574static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2575{
2576	switch (prog_type) {
2577	case BPF_PROG_TYPE_KPROBE:
2578	case BPF_PROG_TYPE_TRACEPOINT:
2579	case BPF_PROG_TYPE_PERF_EVENT:
2580	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2581	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2582	case BPF_PROG_TYPE_TRACING:
2583	case BPF_PROG_TYPE_LSM:
2584	case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2585	case BPF_PROG_TYPE_EXT: /* extends any prog */
2586		return true;
2587	default:
2588		return false;
2589	}
2590}
2591
2592/* last field in 'union bpf_attr' used by this command */
2593#define	BPF_PROG_LOAD_LAST_FIELD log_true_size
2594
2595static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2596{
2597	enum bpf_prog_type type = attr->prog_type;
2598	struct bpf_prog *prog, *dst_prog = NULL;
2599	struct btf *attach_btf = NULL;
2600	int err;
2601	char license[128];
 
2602
2603	if (CHECK_ATTR(BPF_PROG_LOAD))
2604		return -EINVAL;
2605
2606	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2607				 BPF_F_ANY_ALIGNMENT |
2608				 BPF_F_TEST_STATE_FREQ |
2609				 BPF_F_SLEEPABLE |
2610				 BPF_F_TEST_RND_HI32 |
2611				 BPF_F_XDP_HAS_FRAGS |
2612				 BPF_F_XDP_DEV_BOUND_ONLY |
2613				 BPF_F_TEST_REG_INVARIANTS))
2614		return -EINVAL;
2615
2616	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2617	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2618	    !bpf_capable())
2619		return -EPERM;
2620
2621	/* Intent here is for unprivileged_bpf_disabled to block BPF program
2622	 * creation for unprivileged users; other actions depend
2623	 * on fd availability and access to bpffs, so are dependent on
2624	 * object creation success. Even with unprivileged BPF disabled,
2625	 * capability checks are still carried out for these
2626	 * and other operations.
2627	 */
2628	if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
2629		return -EPERM;
2630
2631	if (attr->insn_cnt == 0 ||
2632	    attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2633		return -E2BIG;
2634	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2635	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2636	    !bpf_capable())
2637		return -EPERM;
2638
2639	if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2640		return -EPERM;
2641	if (is_perfmon_prog_type(type) && !perfmon_capable())
2642		return -EPERM;
2643
2644	/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2645	 * or btf, we need to check which one it is
2646	 */
2647	if (attr->attach_prog_fd) {
2648		dst_prog = bpf_prog_get(attr->attach_prog_fd);
2649		if (IS_ERR(dst_prog)) {
2650			dst_prog = NULL;
2651			attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2652			if (IS_ERR(attach_btf))
2653				return -EINVAL;
2654			if (!btf_is_kernel(attach_btf)) {
2655				/* attaching through specifying bpf_prog's BTF
2656				 * objects directly might be supported eventually
2657				 */
2658				btf_put(attach_btf);
2659				return -ENOTSUPP;
2660			}
2661		}
2662	} else if (attr->attach_btf_id) {
2663		/* fall back to vmlinux BTF, if BTF type ID is specified */
2664		attach_btf = bpf_get_btf_vmlinux();
2665		if (IS_ERR(attach_btf))
2666			return PTR_ERR(attach_btf);
2667		if (!attach_btf)
2668			return -EINVAL;
2669		btf_get(attach_btf);
2670	}
2671
2672	bpf_prog_load_fixup_attach_type(attr);
2673	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2674				       attach_btf, attr->attach_btf_id,
2675				       dst_prog)) {
2676		if (dst_prog)
2677			bpf_prog_put(dst_prog);
2678		if (attach_btf)
2679			btf_put(attach_btf);
2680		return -EINVAL;
2681	}
2682
2683	/* plain bpf_prog allocation */
2684	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2685	if (!prog) {
2686		if (dst_prog)
2687			bpf_prog_put(dst_prog);
2688		if (attach_btf)
2689			btf_put(attach_btf);
2690		return -ENOMEM;
2691	}
2692
2693	prog->expected_attach_type = attr->expected_attach_type;
2694	prog->aux->attach_btf = attach_btf;
2695	prog->aux->attach_btf_id = attr->attach_btf_id;
2696	prog->aux->dst_prog = dst_prog;
2697	prog->aux->dev_bound = !!attr->prog_ifindex;
2698	prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2699	prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2700
2701	err = security_bpf_prog_alloc(prog->aux);
2702	if (err)
2703		goto free_prog;
2704
2705	prog->aux->user = get_current_user();
2706	prog->len = attr->insn_cnt;
2707
2708	err = -EFAULT;
2709	if (copy_from_bpfptr(prog->insns,
2710			     make_bpfptr(attr->insns, uattr.is_kernel),
2711			     bpf_prog_insn_size(prog)) != 0)
2712		goto free_prog_sec;
2713	/* copy eBPF program license from user space */
2714	if (strncpy_from_bpfptr(license,
2715				make_bpfptr(attr->license, uattr.is_kernel),
2716				sizeof(license) - 1) < 0)
2717		goto free_prog_sec;
2718	license[sizeof(license) - 1] = 0;
2719
2720	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2721	prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2722
2723	prog->orig_prog = NULL;
2724	prog->jited = 0;
2725
2726	atomic64_set(&prog->aux->refcnt, 1);
 
2727
2728	if (bpf_prog_is_dev_bound(prog->aux)) {
2729		err = bpf_prog_dev_bound_init(prog, attr);
2730		if (err)
2731			goto free_prog_sec;
2732	}
2733
2734	if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2735	    bpf_prog_is_dev_bound(dst_prog->aux)) {
2736		err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2737		if (err)
2738			goto free_prog_sec;
2739	}
2740
2741	/*
2742	 * Bookkeeping for managing the program attachment chain.
2743	 *
2744	 * It might be tempting to set attach_tracing_prog flag at the attachment
2745	 * time, but this will not prevent from loading bunch of tracing prog
2746	 * first, then attach them one to another.
2747	 *
2748	 * The flag attach_tracing_prog is set for the whole program lifecycle, and
2749	 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2750	 * programs cannot change attachment target.
2751	 */
2752	if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2753	    dst_prog->type == BPF_PROG_TYPE_TRACING) {
2754		prog->aux->attach_tracing_prog = true;
2755	}
2756
2757	/* find program type: socket_filter vs tracing_filter */
2758	err = find_prog_type(type, prog);
2759	if (err < 0)
2760		goto free_prog_sec;
2761
2762	prog->aux->load_time = ktime_get_boottime_ns();
2763	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2764			       sizeof(attr->prog_name));
2765	if (err < 0)
2766		goto free_prog_sec;
2767
2768	/* run eBPF verifier */
2769	err = bpf_check(&prog, attr, uattr, uattr_size);
2770	if (err < 0)
2771		goto free_used_maps;
2772
2773	prog = bpf_prog_select_runtime(prog, &err);
2774	if (err < 0)
2775		goto free_used_maps;
2776
2777	err = bpf_prog_alloc_id(prog);
2778	if (err)
2779		goto free_used_maps;
2780
2781	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2782	 * effectively publicly exposed. However, retrieving via
2783	 * bpf_prog_get_fd_by_id() will take another reference,
2784	 * therefore it cannot be gone underneath us.
2785	 *
2786	 * Only for the time /after/ successful bpf_prog_new_fd()
2787	 * and before returning to userspace, we might just hold
2788	 * one reference and any parallel close on that fd could
2789	 * rip everything out. Hence, below notifications must
2790	 * happen before bpf_prog_new_fd().
2791	 *
2792	 * Also, any failure handling from this point onwards must
2793	 * be using bpf_prog_put() given the program is exposed.
2794	 */
2795	bpf_prog_kallsyms_add(prog);
2796	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2797	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2798
2799	err = bpf_prog_new_fd(prog);
2800	if (err < 0)
2801		bpf_prog_put(prog);
2802	return err;
2803
2804free_used_maps:
2805	/* In case we have subprogs, we need to wait for a grace
2806	 * period before we can tear down JIT memory since symbols
2807	 * are already exposed under kallsyms.
2808	 */
2809	__bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2810	return err;
2811free_prog_sec:
2812	free_uid(prog->aux->user);
2813	security_bpf_prog_free(prog->aux);
2814free_prog:
2815	if (prog->aux->attach_btf)
2816		btf_put(prog->aux->attach_btf);
2817	bpf_prog_free(prog);
2818	return err;
2819}
2820
2821#define BPF_OBJ_LAST_FIELD path_fd
2822
2823static int bpf_obj_pin(const union bpf_attr *attr)
2824{
2825	int path_fd;
2826
2827	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2828		return -EINVAL;
2829
2830	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2831	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2832		return -EINVAL;
2833
2834	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2835	return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2836				u64_to_user_ptr(attr->pathname));
2837}
2838
2839static int bpf_obj_get(const union bpf_attr *attr)
2840{
2841	int path_fd;
2842
2843	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2844	    attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2845		return -EINVAL;
2846
2847	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2848	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2849		return -EINVAL;
2850
2851	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2852	return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2853				attr->file_flags);
2854}
2855
2856void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2857		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
2858{
2859	atomic64_set(&link->refcnt, 1);
2860	link->type = type;
2861	link->id = 0;
2862	link->ops = ops;
2863	link->prog = prog;
2864}
2865
2866static void bpf_link_free_id(int id)
2867{
2868	if (!id)
2869		return;
2870
2871	spin_lock_bh(&link_idr_lock);
2872	idr_remove(&link_idr, id);
2873	spin_unlock_bh(&link_idr_lock);
2874}
2875
2876/* Clean up bpf_link and corresponding anon_inode file and FD. After
2877 * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2878 * anon_inode's release() call. This helper marks bpf_link as
2879 * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2880 * is not decremented, it's the responsibility of a calling code that failed
2881 * to complete bpf_link initialization.
2882 * This helper eventually calls link's dealloc callback, but does not call
2883 * link's release callback.
2884 */
2885void bpf_link_cleanup(struct bpf_link_primer *primer)
2886{
2887	primer->link->prog = NULL;
2888	bpf_link_free_id(primer->id);
2889	fput(primer->file);
2890	put_unused_fd(primer->fd);
2891}
2892
2893void bpf_link_inc(struct bpf_link *link)
2894{
2895	atomic64_inc(&link->refcnt);
2896}
2897
2898/* bpf_link_free is guaranteed to be called from process context */
2899static void bpf_link_free(struct bpf_link *link)
2900{
2901	bpf_link_free_id(link->id);
2902	if (link->prog) {
2903		/* detach BPF program, clean up used resources */
2904		link->ops->release(link);
2905		bpf_prog_put(link->prog);
2906	}
2907	/* free bpf_link and its containing memory */
2908	link->ops->dealloc(link);
2909}
2910
2911static void bpf_link_put_deferred(struct work_struct *work)
2912{
2913	struct bpf_link *link = container_of(work, struct bpf_link, work);
2914
2915	bpf_link_free(link);
2916}
2917
2918/* bpf_link_put might be called from atomic context. It needs to be called
2919 * from sleepable context in order to acquire sleeping locks during the process.
2920 */
2921void bpf_link_put(struct bpf_link *link)
2922{
2923	if (!atomic64_dec_and_test(&link->refcnt))
2924		return;
2925
2926	INIT_WORK(&link->work, bpf_link_put_deferred);
2927	schedule_work(&link->work);
 
 
 
 
2928}
2929EXPORT_SYMBOL(bpf_link_put);
2930
2931static void bpf_link_put_direct(struct bpf_link *link)
2932{
2933	if (!atomic64_dec_and_test(&link->refcnt))
2934		return;
2935	bpf_link_free(link);
2936}
2937
2938static int bpf_link_release(struct inode *inode, struct file *filp)
2939{
2940	struct bpf_link *link = filp->private_data;
2941
2942	bpf_link_put_direct(link);
2943	return 0;
2944}
2945
2946#ifdef CONFIG_PROC_FS
2947#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2948#define BPF_MAP_TYPE(_id, _ops)
2949#define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2950static const char *bpf_link_type_strs[] = {
2951	[BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2952#include <linux/bpf_types.h>
2953};
2954#undef BPF_PROG_TYPE
2955#undef BPF_MAP_TYPE
2956#undef BPF_LINK_TYPE
2957
2958static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2959{
2960	const struct bpf_link *link = filp->private_data;
2961	const struct bpf_prog *prog = link->prog;
2962	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2963
 
2964	seq_printf(m,
2965		   "link_type:\t%s\n"
2966		   "link_id:\t%u\n",
 
 
2967		   bpf_link_type_strs[link->type],
2968		   link->id);
2969	if (prog) {
2970		bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2971		seq_printf(m,
2972			   "prog_tag:\t%s\n"
2973			   "prog_id:\t%u\n",
2974			   prog_tag,
2975			   prog->aux->id);
2976	}
2977	if (link->ops->show_fdinfo)
2978		link->ops->show_fdinfo(link, m);
2979}
2980#endif
2981
2982static const struct file_operations bpf_link_fops = {
2983#ifdef CONFIG_PROC_FS
2984	.show_fdinfo	= bpf_link_show_fdinfo,
2985#endif
2986	.release	= bpf_link_release,
2987	.read		= bpf_dummy_read,
2988	.write		= bpf_dummy_write,
2989};
2990
2991static int bpf_link_alloc_id(struct bpf_link *link)
2992{
2993	int id;
2994
2995	idr_preload(GFP_KERNEL);
2996	spin_lock_bh(&link_idr_lock);
2997	id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2998	spin_unlock_bh(&link_idr_lock);
2999	idr_preload_end();
3000
3001	return id;
3002}
3003
3004/* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3005 * reserving unused FD and allocating ID from link_idr. This is to be paired
3006 * with bpf_link_settle() to install FD and ID and expose bpf_link to
3007 * user-space, if bpf_link is successfully attached. If not, bpf_link and
3008 * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3009 * transient state is passed around in struct bpf_link_primer.
3010 * This is preferred way to create and initialize bpf_link, especially when
3011 * there are complicated and expensive operations in between creating bpf_link
3012 * itself and attaching it to BPF hook. By using bpf_link_prime() and
3013 * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3014 * expensive (and potentially failing) roll back operations in a rare case
3015 * that file, FD, or ID can't be allocated.
3016 */
3017int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3018{
3019	struct file *file;
3020	int fd, id;
3021
3022	fd = get_unused_fd_flags(O_CLOEXEC);
3023	if (fd < 0)
3024		return fd;
3025
3026
3027	id = bpf_link_alloc_id(link);
3028	if (id < 0) {
3029		put_unused_fd(fd);
3030		return id;
3031	}
3032
3033	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
3034	if (IS_ERR(file)) {
3035		bpf_link_free_id(id);
3036		put_unused_fd(fd);
3037		return PTR_ERR(file);
3038	}
3039
3040	primer->link = link;
3041	primer->file = file;
3042	primer->fd = fd;
3043	primer->id = id;
3044	return 0;
3045}
3046
3047int bpf_link_settle(struct bpf_link_primer *primer)
3048{
3049	/* make bpf_link fetchable by ID */
3050	spin_lock_bh(&link_idr_lock);
3051	primer->link->id = primer->id;
3052	spin_unlock_bh(&link_idr_lock);
3053	/* make bpf_link fetchable by FD */
3054	fd_install(primer->fd, primer->file);
3055	/* pass through installed FD */
3056	return primer->fd;
3057}
3058
3059int bpf_link_new_fd(struct bpf_link *link)
3060{
3061	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3062}
3063
3064struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3065{
3066	struct fd f = fdget(ufd);
3067	struct bpf_link *link;
3068
3069	if (!f.file)
3070		return ERR_PTR(-EBADF);
3071	if (f.file->f_op != &bpf_link_fops) {
3072		fdput(f);
3073		return ERR_PTR(-EINVAL);
3074	}
3075
3076	link = f.file->private_data;
3077	bpf_link_inc(link);
3078	fdput(f);
3079
3080	return link;
3081}
3082EXPORT_SYMBOL(bpf_link_get_from_fd);
3083
3084static void bpf_tracing_link_release(struct bpf_link *link)
3085{
3086	struct bpf_tracing_link *tr_link =
3087		container_of(link, struct bpf_tracing_link, link.link);
3088
3089	WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3090						tr_link->trampoline));
3091
3092	bpf_trampoline_put(tr_link->trampoline);
3093
3094	/* tgt_prog is NULL if target is a kernel function */
3095	if (tr_link->tgt_prog)
3096		bpf_prog_put(tr_link->tgt_prog);
3097}
3098
3099static void bpf_tracing_link_dealloc(struct bpf_link *link)
3100{
3101	struct bpf_tracing_link *tr_link =
3102		container_of(link, struct bpf_tracing_link, link.link);
3103
3104	kfree(tr_link);
3105}
3106
3107static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3108					 struct seq_file *seq)
3109{
3110	struct bpf_tracing_link *tr_link =
3111		container_of(link, struct bpf_tracing_link, link.link);
3112	u32 target_btf_id, target_obj_id;
3113
3114	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3115				  &target_obj_id, &target_btf_id);
3116	seq_printf(seq,
3117		   "attach_type:\t%d\n"
3118		   "target_obj_id:\t%u\n"
3119		   "target_btf_id:\t%u\n",
3120		   tr_link->attach_type,
3121		   target_obj_id,
3122		   target_btf_id);
3123}
3124
3125static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3126					   struct bpf_link_info *info)
3127{
3128	struct bpf_tracing_link *tr_link =
3129		container_of(link, struct bpf_tracing_link, link.link);
3130
3131	info->tracing.attach_type = tr_link->attach_type;
3132	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3133				  &info->tracing.target_obj_id,
3134				  &info->tracing.target_btf_id);
3135
3136	return 0;
3137}
3138
3139static const struct bpf_link_ops bpf_tracing_link_lops = {
3140	.release = bpf_tracing_link_release,
3141	.dealloc = bpf_tracing_link_dealloc,
3142	.show_fdinfo = bpf_tracing_link_show_fdinfo,
3143	.fill_link_info = bpf_tracing_link_fill_link_info,
3144};
3145
3146static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3147				   int tgt_prog_fd,
3148				   u32 btf_id,
3149				   u64 bpf_cookie)
3150{
3151	struct bpf_link_primer link_primer;
3152	struct bpf_prog *tgt_prog = NULL;
3153	struct bpf_trampoline *tr = NULL;
3154	struct bpf_tracing_link *link;
3155	u64 key = 0;
3156	int err;
3157
3158	switch (prog->type) {
3159	case BPF_PROG_TYPE_TRACING:
3160		if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3161		    prog->expected_attach_type != BPF_TRACE_FEXIT &&
3162		    prog->expected_attach_type != BPF_MODIFY_RETURN) {
3163			err = -EINVAL;
3164			goto out_put_prog;
3165		}
3166		break;
3167	case BPF_PROG_TYPE_EXT:
3168		if (prog->expected_attach_type != 0) {
3169			err = -EINVAL;
3170			goto out_put_prog;
3171		}
3172		break;
3173	case BPF_PROG_TYPE_LSM:
3174		if (prog->expected_attach_type != BPF_LSM_MAC) {
3175			err = -EINVAL;
3176			goto out_put_prog;
3177		}
3178		break;
3179	default:
3180		err = -EINVAL;
3181		goto out_put_prog;
3182	}
3183
3184	if (!!tgt_prog_fd != !!btf_id) {
3185		err = -EINVAL;
3186		goto out_put_prog;
3187	}
3188
3189	if (tgt_prog_fd) {
3190		/*
3191		 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3192		 * part would be changed to implement the same for
3193		 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3194		 * attach_tracing_prog flag is set.
3195		 */
3196		if (prog->type != BPF_PROG_TYPE_EXT) {
3197			err = -EINVAL;
3198			goto out_put_prog;
3199		}
3200
3201		tgt_prog = bpf_prog_get(tgt_prog_fd);
3202		if (IS_ERR(tgt_prog)) {
3203			err = PTR_ERR(tgt_prog);
3204			tgt_prog = NULL;
3205			goto out_put_prog;
3206		}
3207
3208		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3209	}
3210
3211	link = kzalloc(sizeof(*link), GFP_USER);
3212	if (!link) {
3213		err = -ENOMEM;
3214		goto out_put_prog;
3215	}
3216	bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3217		      &bpf_tracing_link_lops, prog);
3218	link->attach_type = prog->expected_attach_type;
3219	link->link.cookie = bpf_cookie;
3220
3221	mutex_lock(&prog->aux->dst_mutex);
3222
3223	/* There are a few possible cases here:
3224	 *
3225	 * - if prog->aux->dst_trampoline is set, the program was just loaded
3226	 *   and not yet attached to anything, so we can use the values stored
3227	 *   in prog->aux
3228	 *
3229	 * - if prog->aux->dst_trampoline is NULL, the program has already been
3230         *   attached to a target and its initial target was cleared (below)
3231	 *
3232	 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3233	 *   target_btf_id using the link_create API.
3234	 *
3235	 * - if tgt_prog == NULL when this function was called using the old
3236	 *   raw_tracepoint_open API, and we need a target from prog->aux
3237	 *
3238	 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3239	 *   was detached and is going for re-attachment.
3240	 *
3241	 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3242	 *   are NULL, then program was already attached and user did not provide
3243	 *   tgt_prog_fd so we have no way to find out or create trampoline
3244	 */
3245	if (!prog->aux->dst_trampoline && !tgt_prog) {
3246		/*
3247		 * Allow re-attach for TRACING and LSM programs. If it's
3248		 * currently linked, bpf_trampoline_link_prog will fail.
3249		 * EXT programs need to specify tgt_prog_fd, so they
3250		 * re-attach in separate code path.
3251		 */
3252		if (prog->type != BPF_PROG_TYPE_TRACING &&
3253		    prog->type != BPF_PROG_TYPE_LSM) {
3254			err = -EINVAL;
3255			goto out_unlock;
3256		}
3257		/* We can allow re-attach only if we have valid attach_btf. */
3258		if (!prog->aux->attach_btf) {
3259			err = -EINVAL;
3260			goto out_unlock;
3261		}
3262		btf_id = prog->aux->attach_btf_id;
3263		key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3264	}
3265
3266	if (!prog->aux->dst_trampoline ||
3267	    (key && key != prog->aux->dst_trampoline->key)) {
3268		/* If there is no saved target, or the specified target is
3269		 * different from the destination specified at load time, we
3270		 * need a new trampoline and a check for compatibility
3271		 */
3272		struct bpf_attach_target_info tgt_info = {};
3273
3274		err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3275					      &tgt_info);
3276		if (err)
3277			goto out_unlock;
3278
3279		if (tgt_info.tgt_mod) {
3280			module_put(prog->aux->mod);
3281			prog->aux->mod = tgt_info.tgt_mod;
3282		}
3283
3284		tr = bpf_trampoline_get(key, &tgt_info);
3285		if (!tr) {
3286			err = -ENOMEM;
3287			goto out_unlock;
3288		}
3289	} else {
3290		/* The caller didn't specify a target, or the target was the
3291		 * same as the destination supplied during program load. This
3292		 * means we can reuse the trampoline and reference from program
3293		 * load time, and there is no need to allocate a new one. This
3294		 * can only happen once for any program, as the saved values in
3295		 * prog->aux are cleared below.
3296		 */
3297		tr = prog->aux->dst_trampoline;
3298		tgt_prog = prog->aux->dst_prog;
3299	}
3300
3301	err = bpf_link_prime(&link->link.link, &link_primer);
3302	if (err)
3303		goto out_unlock;
3304
3305	err = bpf_trampoline_link_prog(&link->link, tr);
3306	if (err) {
3307		bpf_link_cleanup(&link_primer);
3308		link = NULL;
3309		goto out_unlock;
3310	}
3311
3312	link->tgt_prog = tgt_prog;
3313	link->trampoline = tr;
3314
3315	/* Always clear the trampoline and target prog from prog->aux to make
3316	 * sure the original attach destination is not kept alive after a
3317	 * program is (re-)attached to another target.
3318	 */
3319	if (prog->aux->dst_prog &&
3320	    (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3321		/* got extra prog ref from syscall, or attaching to different prog */
3322		bpf_prog_put(prog->aux->dst_prog);
3323	if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3324		/* we allocated a new trampoline, so free the old one */
3325		bpf_trampoline_put(prog->aux->dst_trampoline);
3326
3327	prog->aux->dst_prog = NULL;
3328	prog->aux->dst_trampoline = NULL;
3329	mutex_unlock(&prog->aux->dst_mutex);
3330
3331	return bpf_link_settle(&link_primer);
3332out_unlock:
3333	if (tr && tr != prog->aux->dst_trampoline)
3334		bpf_trampoline_put(tr);
3335	mutex_unlock(&prog->aux->dst_mutex);
3336	kfree(link);
3337out_put_prog:
3338	if (tgt_prog_fd && tgt_prog)
3339		bpf_prog_put(tgt_prog);
3340	return err;
3341}
3342
3343struct bpf_raw_tp_link {
3344	struct bpf_link link;
3345	struct bpf_raw_event_map *btp;
3346};
3347
3348static void bpf_raw_tp_link_release(struct bpf_link *link)
3349{
3350	struct bpf_raw_tp_link *raw_tp =
3351		container_of(link, struct bpf_raw_tp_link, link);
3352
3353	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3354	bpf_put_raw_tracepoint(raw_tp->btp);
3355}
3356
3357static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3358{
3359	struct bpf_raw_tp_link *raw_tp =
3360		container_of(link, struct bpf_raw_tp_link, link);
3361
3362	kfree(raw_tp);
3363}
3364
3365static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3366					struct seq_file *seq)
3367{
3368	struct bpf_raw_tp_link *raw_tp_link =
3369		container_of(link, struct bpf_raw_tp_link, link);
3370
3371	seq_printf(seq,
3372		   "tp_name:\t%s\n",
3373		   raw_tp_link->btp->tp->name);
3374}
3375
3376static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3377			    u32 len)
3378{
3379	if (ulen >= len + 1) {
3380		if (copy_to_user(ubuf, buf, len + 1))
3381			return -EFAULT;
3382	} else {
3383		char zero = '\0';
3384
3385		if (copy_to_user(ubuf, buf, ulen - 1))
3386			return -EFAULT;
3387		if (put_user(zero, ubuf + ulen - 1))
3388			return -EFAULT;
3389		return -ENOSPC;
3390	}
3391
3392	return 0;
3393}
3394
3395static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3396					  struct bpf_link_info *info)
3397{
3398	struct bpf_raw_tp_link *raw_tp_link =
3399		container_of(link, struct bpf_raw_tp_link, link);
3400	char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3401	const char *tp_name = raw_tp_link->btp->tp->name;
3402	u32 ulen = info->raw_tracepoint.tp_name_len;
3403	size_t tp_len = strlen(tp_name);
3404
3405	if (!ulen ^ !ubuf)
3406		return -EINVAL;
3407
3408	info->raw_tracepoint.tp_name_len = tp_len + 1;
3409
3410	if (!ubuf)
3411		return 0;
3412
3413	return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
 
 
 
 
 
 
 
 
 
 
 
 
 
3414}
3415
3416static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3417	.release = bpf_raw_tp_link_release,
3418	.dealloc = bpf_raw_tp_link_dealloc,
3419	.show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3420	.fill_link_info = bpf_raw_tp_link_fill_link_info,
3421};
3422
3423#ifdef CONFIG_PERF_EVENTS
3424struct bpf_perf_link {
3425	struct bpf_link link;
3426	struct file *perf_file;
3427};
3428
3429static void bpf_perf_link_release(struct bpf_link *link)
3430{
3431	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3432	struct perf_event *event = perf_link->perf_file->private_data;
3433
3434	perf_event_free_bpf_prog(event);
3435	fput(perf_link->perf_file);
3436}
3437
3438static void bpf_perf_link_dealloc(struct bpf_link *link)
3439{
3440	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3441
3442	kfree(perf_link);
3443}
3444
3445static int bpf_perf_link_fill_common(const struct perf_event *event,
3446				     char __user *uname, u32 ulen,
3447				     u64 *probe_offset, u64 *probe_addr,
3448				     u32 *fd_type, unsigned long *missed)
3449{
3450	const char *buf;
3451	u32 prog_id;
3452	size_t len;
3453	int err;
3454
3455	if (!ulen ^ !uname)
3456		return -EINVAL;
3457
3458	err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3459				      probe_offset, probe_addr, missed);
3460	if (err)
3461		return err;
3462	if (!uname)
3463		return 0;
3464	if (buf) {
3465		len = strlen(buf);
3466		err = bpf_copy_to_user(uname, buf, ulen, len);
3467		if (err)
3468			return err;
3469	} else {
3470		char zero = '\0';
3471
3472		if (put_user(zero, uname))
3473			return -EFAULT;
3474	}
3475	return 0;
3476}
3477
3478#ifdef CONFIG_KPROBE_EVENTS
3479static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3480				     struct bpf_link_info *info)
3481{
3482	unsigned long missed;
3483	char __user *uname;
3484	u64 addr, offset;
3485	u32 ulen, type;
3486	int err;
3487
3488	uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3489	ulen = info->perf_event.kprobe.name_len;
3490	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3491					&type, &missed);
3492	if (err)
3493		return err;
3494	if (type == BPF_FD_TYPE_KRETPROBE)
3495		info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3496	else
3497		info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3498
3499	info->perf_event.kprobe.offset = offset;
3500	info->perf_event.kprobe.missed = missed;
3501	if (!kallsyms_show_value(current_cred()))
3502		addr = 0;
3503	info->perf_event.kprobe.addr = addr;
3504	return 0;
3505}
3506#endif
3507
3508#ifdef CONFIG_UPROBE_EVENTS
3509static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3510				     struct bpf_link_info *info)
3511{
3512	char __user *uname;
3513	u64 addr, offset;
3514	u32 ulen, type;
3515	int err;
3516
3517	uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3518	ulen = info->perf_event.uprobe.name_len;
3519	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3520					&type, NULL);
3521	if (err)
3522		return err;
3523
3524	if (type == BPF_FD_TYPE_URETPROBE)
3525		info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3526	else
3527		info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3528	info->perf_event.uprobe.offset = offset;
3529	return 0;
3530}
3531#endif
3532
3533static int bpf_perf_link_fill_probe(const struct perf_event *event,
3534				    struct bpf_link_info *info)
3535{
3536#ifdef CONFIG_KPROBE_EVENTS
3537	if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3538		return bpf_perf_link_fill_kprobe(event, info);
3539#endif
3540#ifdef CONFIG_UPROBE_EVENTS
3541	if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3542		return bpf_perf_link_fill_uprobe(event, info);
3543#endif
3544	return -EOPNOTSUPP;
3545}
3546
3547static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3548					 struct bpf_link_info *info)
3549{
3550	char __user *uname;
3551	u32 ulen;
3552
3553	uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3554	ulen = info->perf_event.tracepoint.name_len;
3555	info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3556	return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3557}
3558
3559static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3560					 struct bpf_link_info *info)
3561{
3562	info->perf_event.event.type = event->attr.type;
3563	info->perf_event.event.config = event->attr.config;
3564	info->perf_event.type = BPF_PERF_EVENT_EVENT;
3565	return 0;
3566}
3567
3568static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3569					struct bpf_link_info *info)
3570{
3571	struct bpf_perf_link *perf_link;
3572	const struct perf_event *event;
3573
3574	perf_link = container_of(link, struct bpf_perf_link, link);
3575	event = perf_get_event(perf_link->perf_file);
3576	if (IS_ERR(event))
3577		return PTR_ERR(event);
3578
3579	switch (event->prog->type) {
3580	case BPF_PROG_TYPE_PERF_EVENT:
3581		return bpf_perf_link_fill_perf_event(event, info);
3582	case BPF_PROG_TYPE_TRACEPOINT:
3583		return bpf_perf_link_fill_tracepoint(event, info);
3584	case BPF_PROG_TYPE_KPROBE:
3585		return bpf_perf_link_fill_probe(event, info);
3586	default:
3587		return -EOPNOTSUPP;
3588	}
3589}
3590
3591static const struct bpf_link_ops bpf_perf_link_lops = {
3592	.release = bpf_perf_link_release,
3593	.dealloc = bpf_perf_link_dealloc,
3594	.fill_link_info = bpf_perf_link_fill_link_info,
3595};
3596
3597static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3598{
3599	struct bpf_link_primer link_primer;
3600	struct bpf_perf_link *link;
3601	struct perf_event *event;
3602	struct file *perf_file;
3603	int err;
3604
3605	if (attr->link_create.flags)
3606		return -EINVAL;
3607
3608	perf_file = perf_event_get(attr->link_create.target_fd);
3609	if (IS_ERR(perf_file))
3610		return PTR_ERR(perf_file);
3611
3612	link = kzalloc(sizeof(*link), GFP_USER);
3613	if (!link) {
3614		err = -ENOMEM;
3615		goto out_put_file;
3616	}
3617	bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3618	link->perf_file = perf_file;
3619
3620	err = bpf_link_prime(&link->link, &link_primer);
3621	if (err) {
3622		kfree(link);
3623		goto out_put_file;
3624	}
3625
3626	event = perf_file->private_data;
3627	err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3628	if (err) {
3629		bpf_link_cleanup(&link_primer);
3630		goto out_put_file;
3631	}
3632	/* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3633	bpf_prog_inc(prog);
3634
3635	return bpf_link_settle(&link_primer);
3636
3637out_put_file:
3638	fput(perf_file);
3639	return err;
3640}
3641#else
3642static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3643{
3644	return -EOPNOTSUPP;
3645}
3646#endif /* CONFIG_PERF_EVENTS */
3647
3648static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3649				  const char __user *user_tp_name)
3650{
3651	struct bpf_link_primer link_primer;
3652	struct bpf_raw_tp_link *link;
3653	struct bpf_raw_event_map *btp;
3654	const char *tp_name;
3655	char buf[128];
3656	int err;
3657
3658	switch (prog->type) {
3659	case BPF_PROG_TYPE_TRACING:
3660	case BPF_PROG_TYPE_EXT:
3661	case BPF_PROG_TYPE_LSM:
3662		if (user_tp_name)
3663			/* The attach point for this category of programs
3664			 * should be specified via btf_id during program load.
3665			 */
3666			return -EINVAL;
3667		if (prog->type == BPF_PROG_TYPE_TRACING &&
3668		    prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3669			tp_name = prog->aux->attach_func_name;
3670			break;
3671		}
3672		return bpf_tracing_prog_attach(prog, 0, 0, 0);
3673	case BPF_PROG_TYPE_RAW_TRACEPOINT:
3674	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3675		if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3676			return -EFAULT;
3677		buf[sizeof(buf) - 1] = 0;
3678		tp_name = buf;
3679		break;
3680	default:
3681		return -EINVAL;
3682	}
3683
3684	btp = bpf_get_raw_tracepoint(tp_name);
3685	if (!btp)
3686		return -ENOENT;
3687
3688	link = kzalloc(sizeof(*link), GFP_USER);
3689	if (!link) {
3690		err = -ENOMEM;
3691		goto out_put_btp;
3692	}
3693	bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3694		      &bpf_raw_tp_link_lops, prog);
3695	link->btp = btp;
3696
3697	err = bpf_link_prime(&link->link, &link_primer);
3698	if (err) {
3699		kfree(link);
3700		goto out_put_btp;
3701	}
3702
3703	err = bpf_probe_register(link->btp, prog);
3704	if (err) {
3705		bpf_link_cleanup(&link_primer);
3706		goto out_put_btp;
3707	}
3708
3709	return bpf_link_settle(&link_primer);
3710
3711out_put_btp:
3712	bpf_put_raw_tracepoint(btp);
3713	return err;
3714}
3715
3716#define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3717
3718static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3719{
3720	struct bpf_prog *prog;
3721	int fd;
3722
3723	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3724		return -EINVAL;
3725
3726	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3727	if (IS_ERR(prog))
3728		return PTR_ERR(prog);
3729
3730	fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3731	if (fd < 0)
3732		bpf_prog_put(prog);
3733	return fd;
3734}
3735
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3736static enum bpf_prog_type
3737attach_type_to_prog_type(enum bpf_attach_type attach_type)
3738{
3739	switch (attach_type) {
3740	case BPF_CGROUP_INET_INGRESS:
3741	case BPF_CGROUP_INET_EGRESS:
3742		return BPF_PROG_TYPE_CGROUP_SKB;
3743	case BPF_CGROUP_INET_SOCK_CREATE:
3744	case BPF_CGROUP_INET_SOCK_RELEASE:
3745	case BPF_CGROUP_INET4_POST_BIND:
3746	case BPF_CGROUP_INET6_POST_BIND:
3747		return BPF_PROG_TYPE_CGROUP_SOCK;
3748	case BPF_CGROUP_INET4_BIND:
3749	case BPF_CGROUP_INET6_BIND:
3750	case BPF_CGROUP_INET4_CONNECT:
3751	case BPF_CGROUP_INET6_CONNECT:
3752	case BPF_CGROUP_UNIX_CONNECT:
3753	case BPF_CGROUP_INET4_GETPEERNAME:
3754	case BPF_CGROUP_INET6_GETPEERNAME:
3755	case BPF_CGROUP_UNIX_GETPEERNAME:
3756	case BPF_CGROUP_INET4_GETSOCKNAME:
3757	case BPF_CGROUP_INET6_GETSOCKNAME:
3758	case BPF_CGROUP_UNIX_GETSOCKNAME:
3759	case BPF_CGROUP_UDP4_SENDMSG:
3760	case BPF_CGROUP_UDP6_SENDMSG:
3761	case BPF_CGROUP_UNIX_SENDMSG:
3762	case BPF_CGROUP_UDP4_RECVMSG:
3763	case BPF_CGROUP_UDP6_RECVMSG:
3764	case BPF_CGROUP_UNIX_RECVMSG:
3765		return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3766	case BPF_CGROUP_SOCK_OPS:
3767		return BPF_PROG_TYPE_SOCK_OPS;
3768	case BPF_CGROUP_DEVICE:
3769		return BPF_PROG_TYPE_CGROUP_DEVICE;
3770	case BPF_SK_MSG_VERDICT:
3771		return BPF_PROG_TYPE_SK_MSG;
3772	case BPF_SK_SKB_STREAM_PARSER:
3773	case BPF_SK_SKB_STREAM_VERDICT:
3774	case BPF_SK_SKB_VERDICT:
3775		return BPF_PROG_TYPE_SK_SKB;
3776	case BPF_LIRC_MODE2:
3777		return BPF_PROG_TYPE_LIRC_MODE2;
3778	case BPF_FLOW_DISSECTOR:
3779		return BPF_PROG_TYPE_FLOW_DISSECTOR;
3780	case BPF_CGROUP_SYSCTL:
3781		return BPF_PROG_TYPE_CGROUP_SYSCTL;
3782	case BPF_CGROUP_GETSOCKOPT:
3783	case BPF_CGROUP_SETSOCKOPT:
3784		return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3785	case BPF_TRACE_ITER:
3786	case BPF_TRACE_RAW_TP:
3787	case BPF_TRACE_FENTRY:
3788	case BPF_TRACE_FEXIT:
3789	case BPF_MODIFY_RETURN:
3790		return BPF_PROG_TYPE_TRACING;
3791	case BPF_LSM_MAC:
3792		return BPF_PROG_TYPE_LSM;
3793	case BPF_SK_LOOKUP:
3794		return BPF_PROG_TYPE_SK_LOOKUP;
3795	case BPF_XDP:
3796		return BPF_PROG_TYPE_XDP;
3797	case BPF_LSM_CGROUP:
3798		return BPF_PROG_TYPE_LSM;
3799	case BPF_TCX_INGRESS:
3800	case BPF_TCX_EGRESS:
3801	case BPF_NETKIT_PRIMARY:
3802	case BPF_NETKIT_PEER:
3803		return BPF_PROG_TYPE_SCHED_CLS;
3804	default:
3805		return BPF_PROG_TYPE_UNSPEC;
3806	}
3807}
3808
3809static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3810					     enum bpf_attach_type attach_type)
3811{
3812	enum bpf_prog_type ptype;
3813
3814	switch (prog->type) {
3815	case BPF_PROG_TYPE_CGROUP_SOCK:
3816	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3817	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3818	case BPF_PROG_TYPE_SK_LOOKUP:
3819		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3820	case BPF_PROG_TYPE_CGROUP_SKB:
3821		if (!capable(CAP_NET_ADMIN))
3822			/* cg-skb progs can be loaded by unpriv user.
3823			 * check permissions at attach time.
3824			 */
3825			return -EPERM;
3826		return prog->enforce_expected_attach_type &&
3827			prog->expected_attach_type != attach_type ?
3828			-EINVAL : 0;
3829	case BPF_PROG_TYPE_EXT:
3830		return 0;
3831	case BPF_PROG_TYPE_NETFILTER:
3832		if (attach_type != BPF_NETFILTER)
3833			return -EINVAL;
3834		return 0;
3835	case BPF_PROG_TYPE_PERF_EVENT:
3836	case BPF_PROG_TYPE_TRACEPOINT:
3837		if (attach_type != BPF_PERF_EVENT)
3838			return -EINVAL;
3839		return 0;
3840	case BPF_PROG_TYPE_KPROBE:
3841		if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3842		    attach_type != BPF_TRACE_KPROBE_MULTI)
3843			return -EINVAL;
3844		if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
3845		    attach_type != BPF_TRACE_UPROBE_MULTI)
3846			return -EINVAL;
3847		if (attach_type != BPF_PERF_EVENT &&
3848		    attach_type != BPF_TRACE_KPROBE_MULTI &&
3849		    attach_type != BPF_TRACE_UPROBE_MULTI)
3850			return -EINVAL;
3851		return 0;
3852	case BPF_PROG_TYPE_SCHED_CLS:
3853		if (attach_type != BPF_TCX_INGRESS &&
3854		    attach_type != BPF_TCX_EGRESS &&
3855		    attach_type != BPF_NETKIT_PRIMARY &&
3856		    attach_type != BPF_NETKIT_PEER)
3857			return -EINVAL;
3858		return 0;
3859	default:
3860		ptype = attach_type_to_prog_type(attach_type);
3861		if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
3862			return -EINVAL;
3863		return 0;
3864	}
3865}
3866
3867#define BPF_PROG_ATTACH_LAST_FIELD expected_revision
3868
3869#define BPF_F_ATTACH_MASK_BASE	\
3870	(BPF_F_ALLOW_OVERRIDE |	\
3871	 BPF_F_ALLOW_MULTI |	\
3872	 BPF_F_REPLACE)
3873
3874#define BPF_F_ATTACH_MASK_MPROG	\
3875	(BPF_F_REPLACE |	\
3876	 BPF_F_BEFORE |		\
3877	 BPF_F_AFTER |		\
3878	 BPF_F_ID |		\
3879	 BPF_F_LINK)
3880
3881static int bpf_prog_attach(const union bpf_attr *attr)
3882{
3883	enum bpf_prog_type ptype;
3884	struct bpf_prog *prog;
3885	int ret;
3886
3887	if (CHECK_ATTR(BPF_PROG_ATTACH))
3888		return -EINVAL;
3889
 
 
 
3890	ptype = attach_type_to_prog_type(attr->attach_type);
3891	if (ptype == BPF_PROG_TYPE_UNSPEC)
3892		return -EINVAL;
3893	if (bpf_mprog_supported(ptype)) {
3894		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3895			return -EINVAL;
3896	} else {
3897		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
3898			return -EINVAL;
3899		if (attr->relative_fd ||
3900		    attr->expected_revision)
3901			return -EINVAL;
3902	}
3903
3904	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3905	if (IS_ERR(prog))
3906		return PTR_ERR(prog);
3907
3908	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3909		bpf_prog_put(prog);
3910		return -EINVAL;
3911	}
3912
3913	switch (ptype) {
3914	case BPF_PROG_TYPE_SK_SKB:
3915	case BPF_PROG_TYPE_SK_MSG:
3916		ret = sock_map_get_from_fd(attr, prog);
3917		break;
3918	case BPF_PROG_TYPE_LIRC_MODE2:
3919		ret = lirc_prog_attach(attr, prog);
3920		break;
3921	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3922		ret = netns_bpf_prog_attach(attr, prog);
3923		break;
3924	case BPF_PROG_TYPE_CGROUP_DEVICE:
3925	case BPF_PROG_TYPE_CGROUP_SKB:
3926	case BPF_PROG_TYPE_CGROUP_SOCK:
3927	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3928	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3929	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3930	case BPF_PROG_TYPE_SOCK_OPS:
3931	case BPF_PROG_TYPE_LSM:
3932		if (ptype == BPF_PROG_TYPE_LSM &&
3933		    prog->expected_attach_type != BPF_LSM_CGROUP)
3934			ret = -EINVAL;
3935		else
3936			ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3937		break;
3938	case BPF_PROG_TYPE_SCHED_CLS:
3939		if (attr->attach_type == BPF_TCX_INGRESS ||
3940		    attr->attach_type == BPF_TCX_EGRESS)
3941			ret = tcx_prog_attach(attr, prog);
3942		else
3943			ret = netkit_prog_attach(attr, prog);
3944		break;
3945	default:
3946		ret = -EINVAL;
3947	}
3948
3949	if (ret)
3950		bpf_prog_put(prog);
3951	return ret;
3952}
3953
3954#define BPF_PROG_DETACH_LAST_FIELD expected_revision
3955
3956static int bpf_prog_detach(const union bpf_attr *attr)
3957{
3958	struct bpf_prog *prog = NULL;
3959	enum bpf_prog_type ptype;
3960	int ret;
3961
3962	if (CHECK_ATTR(BPF_PROG_DETACH))
3963		return -EINVAL;
3964
3965	ptype = attach_type_to_prog_type(attr->attach_type);
3966	if (bpf_mprog_supported(ptype)) {
3967		if (ptype == BPF_PROG_TYPE_UNSPEC)
3968			return -EINVAL;
3969		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3970			return -EINVAL;
3971		if (attr->attach_bpf_fd) {
3972			prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3973			if (IS_ERR(prog))
3974				return PTR_ERR(prog);
3975		}
3976	} else if (attr->attach_flags ||
3977		   attr->relative_fd ||
3978		   attr->expected_revision) {
3979		return -EINVAL;
3980	}
3981
3982	switch (ptype) {
3983	case BPF_PROG_TYPE_SK_MSG:
3984	case BPF_PROG_TYPE_SK_SKB:
3985		ret = sock_map_prog_detach(attr, ptype);
3986		break;
3987	case BPF_PROG_TYPE_LIRC_MODE2:
3988		ret = lirc_prog_detach(attr);
3989		break;
3990	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3991		ret = netns_bpf_prog_detach(attr, ptype);
3992		break;
3993	case BPF_PROG_TYPE_CGROUP_DEVICE:
3994	case BPF_PROG_TYPE_CGROUP_SKB:
3995	case BPF_PROG_TYPE_CGROUP_SOCK:
3996	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3997	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3998	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3999	case BPF_PROG_TYPE_SOCK_OPS:
4000	case BPF_PROG_TYPE_LSM:
4001		ret = cgroup_bpf_prog_detach(attr, ptype);
4002		break;
4003	case BPF_PROG_TYPE_SCHED_CLS:
4004		if (attr->attach_type == BPF_TCX_INGRESS ||
4005		    attr->attach_type == BPF_TCX_EGRESS)
4006			ret = tcx_prog_detach(attr, prog);
4007		else
4008			ret = netkit_prog_detach(attr, prog);
4009		break;
4010	default:
4011		ret = -EINVAL;
4012	}
4013
4014	if (prog)
4015		bpf_prog_put(prog);
4016	return ret;
4017}
4018
4019#define BPF_PROG_QUERY_LAST_FIELD query.revision
4020
4021static int bpf_prog_query(const union bpf_attr *attr,
4022			  union bpf_attr __user *uattr)
4023{
4024	if (!capable(CAP_NET_ADMIN))
4025		return -EPERM;
4026	if (CHECK_ATTR(BPF_PROG_QUERY))
4027		return -EINVAL;
4028	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4029		return -EINVAL;
4030
4031	switch (attr->query.attach_type) {
4032	case BPF_CGROUP_INET_INGRESS:
4033	case BPF_CGROUP_INET_EGRESS:
4034	case BPF_CGROUP_INET_SOCK_CREATE:
4035	case BPF_CGROUP_INET_SOCK_RELEASE:
4036	case BPF_CGROUP_INET4_BIND:
4037	case BPF_CGROUP_INET6_BIND:
4038	case BPF_CGROUP_INET4_POST_BIND:
4039	case BPF_CGROUP_INET6_POST_BIND:
4040	case BPF_CGROUP_INET4_CONNECT:
4041	case BPF_CGROUP_INET6_CONNECT:
4042	case BPF_CGROUP_UNIX_CONNECT:
4043	case BPF_CGROUP_INET4_GETPEERNAME:
4044	case BPF_CGROUP_INET6_GETPEERNAME:
4045	case BPF_CGROUP_UNIX_GETPEERNAME:
4046	case BPF_CGROUP_INET4_GETSOCKNAME:
4047	case BPF_CGROUP_INET6_GETSOCKNAME:
4048	case BPF_CGROUP_UNIX_GETSOCKNAME:
4049	case BPF_CGROUP_UDP4_SENDMSG:
4050	case BPF_CGROUP_UDP6_SENDMSG:
4051	case BPF_CGROUP_UNIX_SENDMSG:
4052	case BPF_CGROUP_UDP4_RECVMSG:
4053	case BPF_CGROUP_UDP6_RECVMSG:
4054	case BPF_CGROUP_UNIX_RECVMSG:
4055	case BPF_CGROUP_SOCK_OPS:
4056	case BPF_CGROUP_DEVICE:
4057	case BPF_CGROUP_SYSCTL:
4058	case BPF_CGROUP_GETSOCKOPT:
4059	case BPF_CGROUP_SETSOCKOPT:
4060	case BPF_LSM_CGROUP:
4061		return cgroup_bpf_prog_query(attr, uattr);
4062	case BPF_LIRC_MODE2:
4063		return lirc_prog_query(attr, uattr);
4064	case BPF_FLOW_DISSECTOR:
4065	case BPF_SK_LOOKUP:
4066		return netns_bpf_prog_query(attr, uattr);
4067	case BPF_SK_SKB_STREAM_PARSER:
4068	case BPF_SK_SKB_STREAM_VERDICT:
4069	case BPF_SK_MSG_VERDICT:
4070	case BPF_SK_SKB_VERDICT:
4071		return sock_map_bpf_prog_query(attr, uattr);
4072	case BPF_TCX_INGRESS:
4073	case BPF_TCX_EGRESS:
4074		return tcx_prog_query(attr, uattr);
4075	case BPF_NETKIT_PRIMARY:
4076	case BPF_NETKIT_PEER:
4077		return netkit_prog_query(attr, uattr);
4078	default:
4079		return -EINVAL;
4080	}
4081}
4082
4083#define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4084
4085static int bpf_prog_test_run(const union bpf_attr *attr,
4086			     union bpf_attr __user *uattr)
4087{
4088	struct bpf_prog *prog;
4089	int ret = -ENOTSUPP;
4090
4091	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4092		return -EINVAL;
4093
4094	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4095	    (!attr->test.ctx_size_in && attr->test.ctx_in))
4096		return -EINVAL;
4097
4098	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4099	    (!attr->test.ctx_size_out && attr->test.ctx_out))
4100		return -EINVAL;
4101
4102	prog = bpf_prog_get(attr->test.prog_fd);
4103	if (IS_ERR(prog))
4104		return PTR_ERR(prog);
4105
4106	if (prog->aux->ops->test_run)
4107		ret = prog->aux->ops->test_run(prog, attr, uattr);
4108
4109	bpf_prog_put(prog);
4110	return ret;
4111}
4112
4113#define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4114
4115static int bpf_obj_get_next_id(const union bpf_attr *attr,
4116			       union bpf_attr __user *uattr,
4117			       struct idr *idr,
4118			       spinlock_t *lock)
4119{
4120	u32 next_id = attr->start_id;
4121	int err = 0;
4122
4123	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4124		return -EINVAL;
4125
4126	if (!capable(CAP_SYS_ADMIN))
4127		return -EPERM;
4128
4129	next_id++;
4130	spin_lock_bh(lock);
4131	if (!idr_get_next(idr, &next_id))
4132		err = -ENOENT;
4133	spin_unlock_bh(lock);
4134
4135	if (!err)
4136		err = put_user(next_id, &uattr->next_id);
4137
4138	return err;
4139}
4140
4141struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4142{
4143	struct bpf_map *map;
4144
4145	spin_lock_bh(&map_idr_lock);
4146again:
4147	map = idr_get_next(&map_idr, id);
4148	if (map) {
4149		map = __bpf_map_inc_not_zero(map, false);
4150		if (IS_ERR(map)) {
4151			(*id)++;
4152			goto again;
4153		}
4154	}
4155	spin_unlock_bh(&map_idr_lock);
4156
4157	return map;
4158}
4159
4160struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4161{
4162	struct bpf_prog *prog;
4163
4164	spin_lock_bh(&prog_idr_lock);
4165again:
4166	prog = idr_get_next(&prog_idr, id);
4167	if (prog) {
4168		prog = bpf_prog_inc_not_zero(prog);
4169		if (IS_ERR(prog)) {
4170			(*id)++;
4171			goto again;
4172		}
4173	}
4174	spin_unlock_bh(&prog_idr_lock);
4175
4176	return prog;
4177}
4178
4179#define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4180
4181struct bpf_prog *bpf_prog_by_id(u32 id)
4182{
4183	struct bpf_prog *prog;
4184
4185	if (!id)
4186		return ERR_PTR(-ENOENT);
4187
4188	spin_lock_bh(&prog_idr_lock);
4189	prog = idr_find(&prog_idr, id);
4190	if (prog)
4191		prog = bpf_prog_inc_not_zero(prog);
4192	else
4193		prog = ERR_PTR(-ENOENT);
4194	spin_unlock_bh(&prog_idr_lock);
4195	return prog;
4196}
4197
4198static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4199{
4200	struct bpf_prog *prog;
4201	u32 id = attr->prog_id;
4202	int fd;
4203
4204	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4205		return -EINVAL;
4206
4207	if (!capable(CAP_SYS_ADMIN))
4208		return -EPERM;
4209
4210	prog = bpf_prog_by_id(id);
4211	if (IS_ERR(prog))
4212		return PTR_ERR(prog);
4213
4214	fd = bpf_prog_new_fd(prog);
4215	if (fd < 0)
4216		bpf_prog_put(prog);
4217
4218	return fd;
4219}
4220
4221#define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4222
4223static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4224{
4225	struct bpf_map *map;
4226	u32 id = attr->map_id;
4227	int f_flags;
4228	int fd;
4229
4230	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4231	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4232		return -EINVAL;
4233
4234	if (!capable(CAP_SYS_ADMIN))
4235		return -EPERM;
4236
4237	f_flags = bpf_get_file_flag(attr->open_flags);
4238	if (f_flags < 0)
4239		return f_flags;
4240
4241	spin_lock_bh(&map_idr_lock);
4242	map = idr_find(&map_idr, id);
4243	if (map)
4244		map = __bpf_map_inc_not_zero(map, true);
4245	else
4246		map = ERR_PTR(-ENOENT);
4247	spin_unlock_bh(&map_idr_lock);
4248
4249	if (IS_ERR(map))
4250		return PTR_ERR(map);
4251
4252	fd = bpf_map_new_fd(map, f_flags);
4253	if (fd < 0)
4254		bpf_map_put_with_uref(map);
4255
4256	return fd;
4257}
4258
4259static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4260					      unsigned long addr, u32 *off,
4261					      u32 *type)
4262{
4263	const struct bpf_map *map;
4264	int i;
4265
4266	mutex_lock(&prog->aux->used_maps_mutex);
4267	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4268		map = prog->aux->used_maps[i];
4269		if (map == (void *)addr) {
4270			*type = BPF_PSEUDO_MAP_FD;
4271			goto out;
4272		}
4273		if (!map->ops->map_direct_value_meta)
4274			continue;
4275		if (!map->ops->map_direct_value_meta(map, addr, off)) {
4276			*type = BPF_PSEUDO_MAP_VALUE;
4277			goto out;
4278		}
4279	}
4280	map = NULL;
4281
4282out:
4283	mutex_unlock(&prog->aux->used_maps_mutex);
4284	return map;
4285}
4286
4287static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4288					      const struct cred *f_cred)
4289{
4290	const struct bpf_map *map;
4291	struct bpf_insn *insns;
4292	u32 off, type;
4293	u64 imm;
4294	u8 code;
4295	int i;
4296
4297	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4298			GFP_USER);
4299	if (!insns)
4300		return insns;
4301
4302	for (i = 0; i < prog->len; i++) {
4303		code = insns[i].code;
4304
4305		if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4306			insns[i].code = BPF_JMP | BPF_CALL;
4307			insns[i].imm = BPF_FUNC_tail_call;
4308			/* fall-through */
4309		}
4310		if (code == (BPF_JMP | BPF_CALL) ||
4311		    code == (BPF_JMP | BPF_CALL_ARGS)) {
4312			if (code == (BPF_JMP | BPF_CALL_ARGS))
4313				insns[i].code = BPF_JMP | BPF_CALL;
4314			if (!bpf_dump_raw_ok(f_cred))
4315				insns[i].imm = 0;
4316			continue;
4317		}
4318		if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4319			insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4320			continue;
4321		}
4322
4323		if (code != (BPF_LD | BPF_IMM | BPF_DW))
4324			continue;
4325
4326		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4327		map = bpf_map_from_imm(prog, imm, &off, &type);
4328		if (map) {
4329			insns[i].src_reg = type;
4330			insns[i].imm = map->id;
4331			insns[i + 1].imm = off;
4332			continue;
4333		}
4334	}
4335
4336	return insns;
4337}
4338
4339static int set_info_rec_size(struct bpf_prog_info *info)
4340{
4341	/*
4342	 * Ensure info.*_rec_size is the same as kernel expected size
4343	 *
4344	 * or
4345	 *
4346	 * Only allow zero *_rec_size if both _rec_size and _cnt are
4347	 * zero.  In this case, the kernel will set the expected
4348	 * _rec_size back to the info.
4349	 */
4350
4351	if ((info->nr_func_info || info->func_info_rec_size) &&
4352	    info->func_info_rec_size != sizeof(struct bpf_func_info))
4353		return -EINVAL;
4354
4355	if ((info->nr_line_info || info->line_info_rec_size) &&
4356	    info->line_info_rec_size != sizeof(struct bpf_line_info))
4357		return -EINVAL;
4358
4359	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4360	    info->jited_line_info_rec_size != sizeof(__u64))
4361		return -EINVAL;
4362
4363	info->func_info_rec_size = sizeof(struct bpf_func_info);
4364	info->line_info_rec_size = sizeof(struct bpf_line_info);
4365	info->jited_line_info_rec_size = sizeof(__u64);
4366
4367	return 0;
4368}
4369
4370static int bpf_prog_get_info_by_fd(struct file *file,
4371				   struct bpf_prog *prog,
4372				   const union bpf_attr *attr,
4373				   union bpf_attr __user *uattr)
4374{
4375	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4376	struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4377	struct bpf_prog_info info;
4378	u32 info_len = attr->info.info_len;
4379	struct bpf_prog_kstats stats;
4380	char __user *uinsns;
4381	u32 ulen;
4382	int err;
4383
4384	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4385	if (err)
4386		return err;
4387	info_len = min_t(u32, sizeof(info), info_len);
4388
4389	memset(&info, 0, sizeof(info));
4390	if (copy_from_user(&info, uinfo, info_len))
4391		return -EFAULT;
4392
4393	info.type = prog->type;
4394	info.id = prog->aux->id;
4395	info.load_time = prog->aux->load_time;
4396	info.created_by_uid = from_kuid_munged(current_user_ns(),
4397					       prog->aux->user->uid);
4398	info.gpl_compatible = prog->gpl_compatible;
4399
4400	memcpy(info.tag, prog->tag, sizeof(prog->tag));
4401	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4402
4403	mutex_lock(&prog->aux->used_maps_mutex);
4404	ulen = info.nr_map_ids;
4405	info.nr_map_ids = prog->aux->used_map_cnt;
4406	ulen = min_t(u32, info.nr_map_ids, ulen);
4407	if (ulen) {
4408		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4409		u32 i;
4410
4411		for (i = 0; i < ulen; i++)
4412			if (put_user(prog->aux->used_maps[i]->id,
4413				     &user_map_ids[i])) {
4414				mutex_unlock(&prog->aux->used_maps_mutex);
4415				return -EFAULT;
4416			}
4417	}
4418	mutex_unlock(&prog->aux->used_maps_mutex);
4419
4420	err = set_info_rec_size(&info);
4421	if (err)
4422		return err;
4423
4424	bpf_prog_get_stats(prog, &stats);
4425	info.run_time_ns = stats.nsecs;
4426	info.run_cnt = stats.cnt;
4427	info.recursion_misses = stats.misses;
4428
4429	info.verified_insns = prog->aux->verified_insns;
4430
4431	if (!bpf_capable()) {
4432		info.jited_prog_len = 0;
4433		info.xlated_prog_len = 0;
4434		info.nr_jited_ksyms = 0;
4435		info.nr_jited_func_lens = 0;
4436		info.nr_func_info = 0;
4437		info.nr_line_info = 0;
4438		info.nr_jited_line_info = 0;
4439		goto done;
4440	}
4441
4442	ulen = info.xlated_prog_len;
4443	info.xlated_prog_len = bpf_prog_insn_size(prog);
4444	if (info.xlated_prog_len && ulen) {
4445		struct bpf_insn *insns_sanitized;
4446		bool fault;
4447
4448		if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4449			info.xlated_prog_insns = 0;
4450			goto done;
4451		}
4452		insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4453		if (!insns_sanitized)
4454			return -ENOMEM;
4455		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4456		ulen = min_t(u32, info.xlated_prog_len, ulen);
4457		fault = copy_to_user(uinsns, insns_sanitized, ulen);
4458		kfree(insns_sanitized);
4459		if (fault)
4460			return -EFAULT;
4461	}
4462
4463	if (bpf_prog_is_offloaded(prog->aux)) {
4464		err = bpf_prog_offload_info_fill(&info, prog);
4465		if (err)
4466			return err;
4467		goto done;
4468	}
4469
4470	/* NOTE: the following code is supposed to be skipped for offload.
4471	 * bpf_prog_offload_info_fill() is the place to fill similar fields
4472	 * for offload.
4473	 */
4474	ulen = info.jited_prog_len;
4475	if (prog->aux->func_cnt) {
4476		u32 i;
4477
4478		info.jited_prog_len = 0;
4479		for (i = 0; i < prog->aux->func_cnt; i++)
4480			info.jited_prog_len += prog->aux->func[i]->jited_len;
4481	} else {
4482		info.jited_prog_len = prog->jited_len;
4483	}
4484
4485	if (info.jited_prog_len && ulen) {
4486		if (bpf_dump_raw_ok(file->f_cred)) {
4487			uinsns = u64_to_user_ptr(info.jited_prog_insns);
4488			ulen = min_t(u32, info.jited_prog_len, ulen);
4489
4490			/* for multi-function programs, copy the JITed
4491			 * instructions for all the functions
4492			 */
4493			if (prog->aux->func_cnt) {
4494				u32 len, free, i;
4495				u8 *img;
4496
4497				free = ulen;
4498				for (i = 0; i < prog->aux->func_cnt; i++) {
4499					len = prog->aux->func[i]->jited_len;
4500					len = min_t(u32, len, free);
4501					img = (u8 *) prog->aux->func[i]->bpf_func;
4502					if (copy_to_user(uinsns, img, len))
4503						return -EFAULT;
4504					uinsns += len;
4505					free -= len;
4506					if (!free)
4507						break;
4508				}
4509			} else {
4510				if (copy_to_user(uinsns, prog->bpf_func, ulen))
4511					return -EFAULT;
4512			}
4513		} else {
4514			info.jited_prog_insns = 0;
4515		}
4516	}
4517
4518	ulen = info.nr_jited_ksyms;
4519	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4520	if (ulen) {
4521		if (bpf_dump_raw_ok(file->f_cred)) {
4522			unsigned long ksym_addr;
4523			u64 __user *user_ksyms;
4524			u32 i;
4525
4526			/* copy the address of the kernel symbol
4527			 * corresponding to each function
4528			 */
4529			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4530			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4531			if (prog->aux->func_cnt) {
4532				for (i = 0; i < ulen; i++) {
4533					ksym_addr = (unsigned long)
4534						prog->aux->func[i]->bpf_func;
4535					if (put_user((u64) ksym_addr,
4536						     &user_ksyms[i]))
4537						return -EFAULT;
4538				}
4539			} else {
4540				ksym_addr = (unsigned long) prog->bpf_func;
4541				if (put_user((u64) ksym_addr, &user_ksyms[0]))
4542					return -EFAULT;
4543			}
4544		} else {
4545			info.jited_ksyms = 0;
4546		}
4547	}
4548
4549	ulen = info.nr_jited_func_lens;
4550	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4551	if (ulen) {
4552		if (bpf_dump_raw_ok(file->f_cred)) {
4553			u32 __user *user_lens;
4554			u32 func_len, i;
4555
4556			/* copy the JITed image lengths for each function */
4557			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4558			user_lens = u64_to_user_ptr(info.jited_func_lens);
4559			if (prog->aux->func_cnt) {
4560				for (i = 0; i < ulen; i++) {
4561					func_len =
4562						prog->aux->func[i]->jited_len;
4563					if (put_user(func_len, &user_lens[i]))
4564						return -EFAULT;
4565				}
4566			} else {
4567				func_len = prog->jited_len;
4568				if (put_user(func_len, &user_lens[0]))
4569					return -EFAULT;
4570			}
4571		} else {
4572			info.jited_func_lens = 0;
4573		}
4574	}
4575
4576	if (prog->aux->btf)
4577		info.btf_id = btf_obj_id(prog->aux->btf);
4578	info.attach_btf_id = prog->aux->attach_btf_id;
4579	if (attach_btf)
4580		info.attach_btf_obj_id = btf_obj_id(attach_btf);
4581
4582	ulen = info.nr_func_info;
4583	info.nr_func_info = prog->aux->func_info_cnt;
4584	if (info.nr_func_info && ulen) {
4585		char __user *user_finfo;
4586
4587		user_finfo = u64_to_user_ptr(info.func_info);
4588		ulen = min_t(u32, info.nr_func_info, ulen);
4589		if (copy_to_user(user_finfo, prog->aux->func_info,
4590				 info.func_info_rec_size * ulen))
4591			return -EFAULT;
4592	}
4593
4594	ulen = info.nr_line_info;
4595	info.nr_line_info = prog->aux->nr_linfo;
4596	if (info.nr_line_info && ulen) {
4597		__u8 __user *user_linfo;
4598
4599		user_linfo = u64_to_user_ptr(info.line_info);
4600		ulen = min_t(u32, info.nr_line_info, ulen);
4601		if (copy_to_user(user_linfo, prog->aux->linfo,
4602				 info.line_info_rec_size * ulen))
4603			return -EFAULT;
4604	}
4605
4606	ulen = info.nr_jited_line_info;
4607	if (prog->aux->jited_linfo)
4608		info.nr_jited_line_info = prog->aux->nr_linfo;
4609	else
4610		info.nr_jited_line_info = 0;
4611	if (info.nr_jited_line_info && ulen) {
4612		if (bpf_dump_raw_ok(file->f_cred)) {
4613			unsigned long line_addr;
4614			__u64 __user *user_linfo;
4615			u32 i;
4616
4617			user_linfo = u64_to_user_ptr(info.jited_line_info);
4618			ulen = min_t(u32, info.nr_jited_line_info, ulen);
4619			for (i = 0; i < ulen; i++) {
4620				line_addr = (unsigned long)prog->aux->jited_linfo[i];
4621				if (put_user((__u64)line_addr, &user_linfo[i]))
4622					return -EFAULT;
4623			}
4624		} else {
4625			info.jited_line_info = 0;
4626		}
4627	}
4628
4629	ulen = info.nr_prog_tags;
4630	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4631	if (ulen) {
4632		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4633		u32 i;
4634
4635		user_prog_tags = u64_to_user_ptr(info.prog_tags);
4636		ulen = min_t(u32, info.nr_prog_tags, ulen);
4637		if (prog->aux->func_cnt) {
4638			for (i = 0; i < ulen; i++) {
4639				if (copy_to_user(user_prog_tags[i],
4640						 prog->aux->func[i]->tag,
4641						 BPF_TAG_SIZE))
4642					return -EFAULT;
4643			}
4644		} else {
4645			if (copy_to_user(user_prog_tags[0],
4646					 prog->tag, BPF_TAG_SIZE))
4647				return -EFAULT;
4648		}
4649	}
4650
4651done:
4652	if (copy_to_user(uinfo, &info, info_len) ||
4653	    put_user(info_len, &uattr->info.info_len))
4654		return -EFAULT;
4655
4656	return 0;
4657}
4658
4659static int bpf_map_get_info_by_fd(struct file *file,
4660				  struct bpf_map *map,
4661				  const union bpf_attr *attr,
4662				  union bpf_attr __user *uattr)
4663{
4664	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4665	struct bpf_map_info info;
4666	u32 info_len = attr->info.info_len;
4667	int err;
4668
4669	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4670	if (err)
4671		return err;
4672	info_len = min_t(u32, sizeof(info), info_len);
4673
4674	memset(&info, 0, sizeof(info));
4675	info.type = map->map_type;
4676	info.id = map->id;
4677	info.key_size = map->key_size;
4678	info.value_size = map->value_size;
4679	info.max_entries = map->max_entries;
4680	info.map_flags = map->map_flags;
4681	info.map_extra = map->map_extra;
4682	memcpy(info.name, map->name, sizeof(map->name));
4683
4684	if (map->btf) {
4685		info.btf_id = btf_obj_id(map->btf);
4686		info.btf_key_type_id = map->btf_key_type_id;
4687		info.btf_value_type_id = map->btf_value_type_id;
4688	}
4689	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4690
4691	if (bpf_map_is_offloaded(map)) {
4692		err = bpf_map_offload_info_fill(&info, map);
4693		if (err)
4694			return err;
4695	}
4696
4697	if (copy_to_user(uinfo, &info, info_len) ||
4698	    put_user(info_len, &uattr->info.info_len))
4699		return -EFAULT;
4700
4701	return 0;
4702}
4703
4704static int bpf_btf_get_info_by_fd(struct file *file,
4705				  struct btf *btf,
4706				  const union bpf_attr *attr,
4707				  union bpf_attr __user *uattr)
4708{
4709	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4710	u32 info_len = attr->info.info_len;
4711	int err;
4712
4713	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4714	if (err)
4715		return err;
4716
4717	return btf_get_info_by_fd(btf, attr, uattr);
4718}
4719
4720static int bpf_link_get_info_by_fd(struct file *file,
4721				  struct bpf_link *link,
4722				  const union bpf_attr *attr,
4723				  union bpf_attr __user *uattr)
4724{
4725	struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4726	struct bpf_link_info info;
4727	u32 info_len = attr->info.info_len;
4728	int err;
4729
4730	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4731	if (err)
4732		return err;
4733	info_len = min_t(u32, sizeof(info), info_len);
4734
4735	memset(&info, 0, sizeof(info));
4736	if (copy_from_user(&info, uinfo, info_len))
4737		return -EFAULT;
4738
4739	info.type = link->type;
4740	info.id = link->id;
4741	if (link->prog)
4742		info.prog_id = link->prog->aux->id;
4743
4744	if (link->ops->fill_link_info) {
4745		err = link->ops->fill_link_info(link, &info);
4746		if (err)
4747			return err;
4748	}
4749
4750	if (copy_to_user(uinfo, &info, info_len) ||
4751	    put_user(info_len, &uattr->info.info_len))
4752		return -EFAULT;
4753
4754	return 0;
4755}
4756
4757
4758#define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4759
4760static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4761				  union bpf_attr __user *uattr)
4762{
4763	int ufd = attr->info.bpf_fd;
4764	struct fd f;
4765	int err;
4766
4767	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4768		return -EINVAL;
4769
4770	f = fdget(ufd);
4771	if (!f.file)
4772		return -EBADFD;
4773
4774	if (f.file->f_op == &bpf_prog_fops)
4775		err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4776					      uattr);
4777	else if (f.file->f_op == &bpf_map_fops)
4778		err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4779					     uattr);
4780	else if (f.file->f_op == &btf_fops)
4781		err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4782	else if (f.file->f_op == &bpf_link_fops)
4783		err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4784					      attr, uattr);
4785	else
4786		err = -EINVAL;
4787
4788	fdput(f);
4789	return err;
4790}
4791
4792#define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4793
4794static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4795{
4796	if (CHECK_ATTR(BPF_BTF_LOAD))
4797		return -EINVAL;
4798
4799	if (!bpf_capable())
4800		return -EPERM;
4801
4802	return btf_new_fd(attr, uattr, uattr_size);
4803}
4804
4805#define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4806
4807static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4808{
4809	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4810		return -EINVAL;
4811
4812	if (!capable(CAP_SYS_ADMIN))
4813		return -EPERM;
4814
4815	return btf_get_fd_by_id(attr->btf_id);
4816}
4817
4818static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4819				    union bpf_attr __user *uattr,
4820				    u32 prog_id, u32 fd_type,
4821				    const char *buf, u64 probe_offset,
4822				    u64 probe_addr)
4823{
4824	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4825	u32 len = buf ? strlen(buf) : 0, input_len;
4826	int err = 0;
4827
4828	if (put_user(len, &uattr->task_fd_query.buf_len))
4829		return -EFAULT;
4830	input_len = attr->task_fd_query.buf_len;
4831	if (input_len && ubuf) {
4832		if (!len) {
4833			/* nothing to copy, just make ubuf NULL terminated */
4834			char zero = '\0';
4835
4836			if (put_user(zero, ubuf))
4837				return -EFAULT;
4838		} else if (input_len >= len + 1) {
4839			/* ubuf can hold the string with NULL terminator */
4840			if (copy_to_user(ubuf, buf, len + 1))
4841				return -EFAULT;
4842		} else {
4843			/* ubuf cannot hold the string with NULL terminator,
4844			 * do a partial copy with NULL terminator.
4845			 */
4846			char zero = '\0';
4847
4848			err = -ENOSPC;
4849			if (copy_to_user(ubuf, buf, input_len - 1))
4850				return -EFAULT;
4851			if (put_user(zero, ubuf + input_len - 1))
4852				return -EFAULT;
4853		}
4854	}
4855
4856	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4857	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4858	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4859	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4860		return -EFAULT;
4861
4862	return err;
4863}
4864
4865#define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4866
4867static int bpf_task_fd_query(const union bpf_attr *attr,
4868			     union bpf_attr __user *uattr)
4869{
4870	pid_t pid = attr->task_fd_query.pid;
4871	u32 fd = attr->task_fd_query.fd;
4872	const struct perf_event *event;
4873	struct task_struct *task;
4874	struct file *file;
4875	int err;
4876
4877	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4878		return -EINVAL;
4879
4880	if (!capable(CAP_SYS_ADMIN))
4881		return -EPERM;
4882
4883	if (attr->task_fd_query.flags != 0)
4884		return -EINVAL;
4885
4886	rcu_read_lock();
4887	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4888	rcu_read_unlock();
4889	if (!task)
4890		return -ENOENT;
4891
4892	err = 0;
4893	file = fget_task(task, fd);
4894	put_task_struct(task);
4895	if (!file)
4896		return -EBADF;
4897
4898	if (file->f_op == &bpf_link_fops) {
4899		struct bpf_link *link = file->private_data;
4900
4901		if (link->ops == &bpf_raw_tp_link_lops) {
4902			struct bpf_raw_tp_link *raw_tp =
4903				container_of(link, struct bpf_raw_tp_link, link);
4904			struct bpf_raw_event_map *btp = raw_tp->btp;
4905
4906			err = bpf_task_fd_query_copy(attr, uattr,
4907						     raw_tp->link.prog->aux->id,
4908						     BPF_FD_TYPE_RAW_TRACEPOINT,
4909						     btp->tp->name, 0, 0);
4910			goto put_file;
4911		}
4912		goto out_not_supp;
4913	}
4914
4915	event = perf_get_event(file);
4916	if (!IS_ERR(event)) {
4917		u64 probe_offset, probe_addr;
4918		u32 prog_id, fd_type;
4919		const char *buf;
4920
4921		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4922					      &buf, &probe_offset,
4923					      &probe_addr, NULL);
4924		if (!err)
4925			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4926						     fd_type, buf,
4927						     probe_offset,
4928						     probe_addr);
4929		goto put_file;
4930	}
4931
4932out_not_supp:
4933	err = -ENOTSUPP;
4934put_file:
4935	fput(file);
4936	return err;
4937}
4938
4939#define BPF_MAP_BATCH_LAST_FIELD batch.flags
4940
4941#define BPF_DO_BATCH(fn, ...)			\
4942	do {					\
4943		if (!fn) {			\
4944			err = -ENOTSUPP;	\
4945			goto err_put;		\
4946		}				\
4947		err = fn(__VA_ARGS__);		\
4948	} while (0)
4949
4950static int bpf_map_do_batch(const union bpf_attr *attr,
4951			    union bpf_attr __user *uattr,
4952			    int cmd)
4953{
4954	bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4955			 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4956	bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4957	struct bpf_map *map;
4958	int err, ufd;
4959	struct fd f;
4960
4961	if (CHECK_ATTR(BPF_MAP_BATCH))
4962		return -EINVAL;
4963
4964	ufd = attr->batch.map_fd;
4965	f = fdget(ufd);
4966	map = __bpf_map_get(f);
4967	if (IS_ERR(map))
4968		return PTR_ERR(map);
4969	if (has_write)
4970		bpf_map_write_active_inc(map);
4971	if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4972		err = -EPERM;
4973		goto err_put;
4974	}
4975	if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4976		err = -EPERM;
4977		goto err_put;
4978	}
4979
4980	if (cmd == BPF_MAP_LOOKUP_BATCH)
4981		BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4982	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4983		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4984	else if (cmd == BPF_MAP_UPDATE_BATCH)
4985		BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4986	else
4987		BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4988err_put:
4989	if (has_write) {
4990		maybe_wait_bpf_programs(map);
4991		bpf_map_write_active_dec(map);
4992	}
4993	fdput(f);
4994	return err;
4995}
4996
4997#define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
4998static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4999{
 
5000	struct bpf_prog *prog;
5001	int ret;
5002
5003	if (CHECK_ATTR(BPF_LINK_CREATE))
5004		return -EINVAL;
5005
5006	if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5007		return bpf_struct_ops_link_create(attr);
5008
5009	prog = bpf_prog_get(attr->link_create.prog_fd);
5010	if (IS_ERR(prog))
5011		return PTR_ERR(prog);
5012
5013	ret = bpf_prog_attach_check_attach_type(prog,
5014						attr->link_create.attach_type);
5015	if (ret)
5016		goto out;
5017
5018	switch (prog->type) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5019	case BPF_PROG_TYPE_CGROUP_SKB:
5020	case BPF_PROG_TYPE_CGROUP_SOCK:
5021	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5022	case BPF_PROG_TYPE_SOCK_OPS:
5023	case BPF_PROG_TYPE_CGROUP_DEVICE:
5024	case BPF_PROG_TYPE_CGROUP_SYSCTL:
5025	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5026		ret = cgroup_bpf_link_attach(attr, prog);
5027		break;
5028	case BPF_PROG_TYPE_EXT:
5029		ret = bpf_tracing_prog_attach(prog,
5030					      attr->link_create.target_fd,
5031					      attr->link_create.target_btf_id,
5032					      attr->link_create.tracing.cookie);
5033		break;
5034	case BPF_PROG_TYPE_LSM:
5035	case BPF_PROG_TYPE_TRACING:
5036		if (attr->link_create.attach_type != prog->expected_attach_type) {
5037			ret = -EINVAL;
5038			goto out;
5039		}
5040		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5041			ret = bpf_raw_tp_link_attach(prog, NULL);
5042		else if (prog->expected_attach_type == BPF_TRACE_ITER)
5043			ret = bpf_iter_link_attach(attr, uattr, prog);
5044		else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5045			ret = cgroup_bpf_link_attach(attr, prog);
5046		else
5047			ret = bpf_tracing_prog_attach(prog,
5048						      attr->link_create.target_fd,
5049						      attr->link_create.target_btf_id,
5050						      attr->link_create.tracing.cookie);
5051		break;
5052	case BPF_PROG_TYPE_FLOW_DISSECTOR:
5053	case BPF_PROG_TYPE_SK_LOOKUP:
5054		ret = netns_bpf_link_create(attr, prog);
5055		break;
5056#ifdef CONFIG_NET
5057	case BPF_PROG_TYPE_XDP:
5058		ret = bpf_xdp_link_attach(attr, prog);
5059		break;
5060	case BPF_PROG_TYPE_SCHED_CLS:
5061		if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5062		    attr->link_create.attach_type == BPF_TCX_EGRESS)
5063			ret = tcx_link_attach(attr, prog);
5064		else
5065			ret = netkit_link_attach(attr, prog);
5066		break;
5067	case BPF_PROG_TYPE_NETFILTER:
5068		ret = bpf_nf_link_attach(attr, prog);
5069		break;
5070#endif
5071	case BPF_PROG_TYPE_PERF_EVENT:
5072	case BPF_PROG_TYPE_TRACEPOINT:
5073		ret = bpf_perf_link_attach(attr, prog);
5074		break;
5075	case BPF_PROG_TYPE_KPROBE:
5076		if (attr->link_create.attach_type == BPF_PERF_EVENT)
5077			ret = bpf_perf_link_attach(attr, prog);
5078		else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5079			ret = bpf_kprobe_multi_link_attach(attr, prog);
5080		else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5081			ret = bpf_uprobe_multi_link_attach(attr, prog);
5082		break;
5083	default:
5084		ret = -EINVAL;
5085	}
5086
5087out:
5088	if (ret < 0)
5089		bpf_prog_put(prog);
5090	return ret;
5091}
5092
5093static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5094{
5095	struct bpf_map *new_map, *old_map = NULL;
5096	int ret;
5097
5098	new_map = bpf_map_get(attr->link_update.new_map_fd);
5099	if (IS_ERR(new_map))
5100		return PTR_ERR(new_map);
5101
5102	if (attr->link_update.flags & BPF_F_REPLACE) {
5103		old_map = bpf_map_get(attr->link_update.old_map_fd);
5104		if (IS_ERR(old_map)) {
5105			ret = PTR_ERR(old_map);
5106			goto out_put;
5107		}
5108	} else if (attr->link_update.old_map_fd) {
5109		ret = -EINVAL;
5110		goto out_put;
5111	}
5112
5113	ret = link->ops->update_map(link, new_map, old_map);
5114
5115	if (old_map)
5116		bpf_map_put(old_map);
5117out_put:
5118	bpf_map_put(new_map);
5119	return ret;
5120}
5121
5122#define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5123
5124static int link_update(union bpf_attr *attr)
5125{
5126	struct bpf_prog *old_prog = NULL, *new_prog;
5127	struct bpf_link *link;
5128	u32 flags;
5129	int ret;
5130
5131	if (CHECK_ATTR(BPF_LINK_UPDATE))
5132		return -EINVAL;
5133
5134	flags = attr->link_update.flags;
5135	if (flags & ~BPF_F_REPLACE)
5136		return -EINVAL;
5137
5138	link = bpf_link_get_from_fd(attr->link_update.link_fd);
5139	if (IS_ERR(link))
5140		return PTR_ERR(link);
5141
5142	if (link->ops->update_map) {
5143		ret = link_update_map(link, attr);
5144		goto out_put_link;
5145	}
5146
5147	new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5148	if (IS_ERR(new_prog)) {
5149		ret = PTR_ERR(new_prog);
5150		goto out_put_link;
5151	}
5152
5153	if (flags & BPF_F_REPLACE) {
5154		old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5155		if (IS_ERR(old_prog)) {
5156			ret = PTR_ERR(old_prog);
5157			old_prog = NULL;
5158			goto out_put_progs;
5159		}
5160	} else if (attr->link_update.old_prog_fd) {
5161		ret = -EINVAL;
5162		goto out_put_progs;
5163	}
5164
5165	if (link->ops->update_prog)
5166		ret = link->ops->update_prog(link, new_prog, old_prog);
5167	else
5168		ret = -EINVAL;
5169
5170out_put_progs:
5171	if (old_prog)
5172		bpf_prog_put(old_prog);
5173	if (ret)
5174		bpf_prog_put(new_prog);
5175out_put_link:
5176	bpf_link_put_direct(link);
5177	return ret;
5178}
5179
5180#define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5181
5182static int link_detach(union bpf_attr *attr)
5183{
5184	struct bpf_link *link;
5185	int ret;
5186
5187	if (CHECK_ATTR(BPF_LINK_DETACH))
5188		return -EINVAL;
5189
5190	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5191	if (IS_ERR(link))
5192		return PTR_ERR(link);
5193
5194	if (link->ops->detach)
5195		ret = link->ops->detach(link);
5196	else
5197		ret = -EOPNOTSUPP;
5198
5199	bpf_link_put_direct(link);
5200	return ret;
5201}
5202
5203static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5204{
5205	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5206}
5207
5208struct bpf_link *bpf_link_by_id(u32 id)
5209{
5210	struct bpf_link *link;
5211
5212	if (!id)
5213		return ERR_PTR(-ENOENT);
5214
5215	spin_lock_bh(&link_idr_lock);
5216	/* before link is "settled", ID is 0, pretend it doesn't exist yet */
5217	link = idr_find(&link_idr, id);
5218	if (link) {
5219		if (link->id)
5220			link = bpf_link_inc_not_zero(link);
5221		else
5222			link = ERR_PTR(-EAGAIN);
5223	} else {
5224		link = ERR_PTR(-ENOENT);
5225	}
5226	spin_unlock_bh(&link_idr_lock);
5227	return link;
5228}
5229
5230struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5231{
5232	struct bpf_link *link;
5233
5234	spin_lock_bh(&link_idr_lock);
5235again:
5236	link = idr_get_next(&link_idr, id);
5237	if (link) {
5238		link = bpf_link_inc_not_zero(link);
5239		if (IS_ERR(link)) {
5240			(*id)++;
5241			goto again;
5242		}
5243	}
5244	spin_unlock_bh(&link_idr_lock);
5245
5246	return link;
5247}
5248
5249#define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5250
5251static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5252{
5253	struct bpf_link *link;
5254	u32 id = attr->link_id;
5255	int fd;
5256
5257	if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5258		return -EINVAL;
5259
5260	if (!capable(CAP_SYS_ADMIN))
5261		return -EPERM;
5262
5263	link = bpf_link_by_id(id);
5264	if (IS_ERR(link))
5265		return PTR_ERR(link);
5266
5267	fd = bpf_link_new_fd(link);
5268	if (fd < 0)
5269		bpf_link_put_direct(link);
5270
5271	return fd;
5272}
5273
5274DEFINE_MUTEX(bpf_stats_enabled_mutex);
5275
5276static int bpf_stats_release(struct inode *inode, struct file *file)
5277{
5278	mutex_lock(&bpf_stats_enabled_mutex);
5279	static_key_slow_dec(&bpf_stats_enabled_key.key);
5280	mutex_unlock(&bpf_stats_enabled_mutex);
5281	return 0;
5282}
5283
5284static const struct file_operations bpf_stats_fops = {
5285	.release = bpf_stats_release,
5286};
5287
5288static int bpf_enable_runtime_stats(void)
5289{
5290	int fd;
5291
5292	mutex_lock(&bpf_stats_enabled_mutex);
5293
5294	/* Set a very high limit to avoid overflow */
5295	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5296		mutex_unlock(&bpf_stats_enabled_mutex);
5297		return -EBUSY;
5298	}
5299
5300	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5301	if (fd >= 0)
5302		static_key_slow_inc(&bpf_stats_enabled_key.key);
5303
5304	mutex_unlock(&bpf_stats_enabled_mutex);
5305	return fd;
5306}
5307
5308#define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5309
5310static int bpf_enable_stats(union bpf_attr *attr)
5311{
5312
5313	if (CHECK_ATTR(BPF_ENABLE_STATS))
5314		return -EINVAL;
5315
5316	if (!capable(CAP_SYS_ADMIN))
5317		return -EPERM;
5318
5319	switch (attr->enable_stats.type) {
5320	case BPF_STATS_RUN_TIME:
5321		return bpf_enable_runtime_stats();
5322	default:
5323		break;
5324	}
5325	return -EINVAL;
5326}
5327
5328#define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5329
5330static int bpf_iter_create(union bpf_attr *attr)
5331{
5332	struct bpf_link *link;
5333	int err;
5334
5335	if (CHECK_ATTR(BPF_ITER_CREATE))
5336		return -EINVAL;
5337
5338	if (attr->iter_create.flags)
5339		return -EINVAL;
5340
5341	link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5342	if (IS_ERR(link))
5343		return PTR_ERR(link);
5344
5345	err = bpf_iter_new_fd(link);
5346	bpf_link_put_direct(link);
5347
5348	return err;
5349}
5350
5351#define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5352
5353static int bpf_prog_bind_map(union bpf_attr *attr)
5354{
5355	struct bpf_prog *prog;
5356	struct bpf_map *map;
5357	struct bpf_map **used_maps_old, **used_maps_new;
5358	int i, ret = 0;
5359
5360	if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5361		return -EINVAL;
5362
5363	if (attr->prog_bind_map.flags)
5364		return -EINVAL;
5365
5366	prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5367	if (IS_ERR(prog))
5368		return PTR_ERR(prog);
5369
5370	map = bpf_map_get(attr->prog_bind_map.map_fd);
5371	if (IS_ERR(map)) {
5372		ret = PTR_ERR(map);
5373		goto out_prog_put;
5374	}
5375
5376	mutex_lock(&prog->aux->used_maps_mutex);
5377
5378	used_maps_old = prog->aux->used_maps;
5379
5380	for (i = 0; i < prog->aux->used_map_cnt; i++)
5381		if (used_maps_old[i] == map) {
5382			bpf_map_put(map);
5383			goto out_unlock;
5384		}
5385
5386	used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5387				      sizeof(used_maps_new[0]),
5388				      GFP_KERNEL);
5389	if (!used_maps_new) {
5390		ret = -ENOMEM;
5391		goto out_unlock;
5392	}
5393
5394	/* The bpf program will not access the bpf map, but for the sake of
5395	 * simplicity, increase sleepable_refcnt for sleepable program as well.
5396	 */
5397	if (prog->aux->sleepable)
5398		atomic64_inc(&map->sleepable_refcnt);
5399	memcpy(used_maps_new, used_maps_old,
5400	       sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5401	used_maps_new[prog->aux->used_map_cnt] = map;
5402
5403	prog->aux->used_map_cnt++;
5404	prog->aux->used_maps = used_maps_new;
5405
5406	kfree(used_maps_old);
5407
5408out_unlock:
5409	mutex_unlock(&prog->aux->used_maps_mutex);
5410
5411	if (ret)
5412		bpf_map_put(map);
5413out_prog_put:
5414	bpf_prog_put(prog);
5415	return ret;
5416}
5417
5418static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5419{
5420	union bpf_attr attr;
 
5421	int err;
5422
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5423	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5424	if (err)
5425		return err;
5426	size = min_t(u32, size, sizeof(attr));
5427
5428	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
5429	memset(&attr, 0, sizeof(attr));
5430	if (copy_from_bpfptr(&attr, uattr, size) != 0)
5431		return -EFAULT;
5432
5433	err = security_bpf(cmd, &attr, size);
5434	if (err < 0)
5435		return err;
5436
5437	switch (cmd) {
5438	case BPF_MAP_CREATE:
5439		err = map_create(&attr);
5440		break;
5441	case BPF_MAP_LOOKUP_ELEM:
5442		err = map_lookup_elem(&attr);
5443		break;
5444	case BPF_MAP_UPDATE_ELEM:
5445		err = map_update_elem(&attr, uattr);
5446		break;
5447	case BPF_MAP_DELETE_ELEM:
5448		err = map_delete_elem(&attr, uattr);
5449		break;
5450	case BPF_MAP_GET_NEXT_KEY:
5451		err = map_get_next_key(&attr);
5452		break;
5453	case BPF_MAP_FREEZE:
5454		err = map_freeze(&attr);
5455		break;
5456	case BPF_PROG_LOAD:
5457		err = bpf_prog_load(&attr, uattr, size);
5458		break;
5459	case BPF_OBJ_PIN:
5460		err = bpf_obj_pin(&attr);
5461		break;
5462	case BPF_OBJ_GET:
5463		err = bpf_obj_get(&attr);
5464		break;
5465	case BPF_PROG_ATTACH:
5466		err = bpf_prog_attach(&attr);
5467		break;
5468	case BPF_PROG_DETACH:
5469		err = bpf_prog_detach(&attr);
5470		break;
5471	case BPF_PROG_QUERY:
5472		err = bpf_prog_query(&attr, uattr.user);
5473		break;
5474	case BPF_PROG_TEST_RUN:
5475		err = bpf_prog_test_run(&attr, uattr.user);
5476		break;
5477	case BPF_PROG_GET_NEXT_ID:
5478		err = bpf_obj_get_next_id(&attr, uattr.user,
5479					  &prog_idr, &prog_idr_lock);
5480		break;
5481	case BPF_MAP_GET_NEXT_ID:
5482		err = bpf_obj_get_next_id(&attr, uattr.user,
5483					  &map_idr, &map_idr_lock);
5484		break;
5485	case BPF_BTF_GET_NEXT_ID:
5486		err = bpf_obj_get_next_id(&attr, uattr.user,
5487					  &btf_idr, &btf_idr_lock);
5488		break;
5489	case BPF_PROG_GET_FD_BY_ID:
5490		err = bpf_prog_get_fd_by_id(&attr);
5491		break;
5492	case BPF_MAP_GET_FD_BY_ID:
5493		err = bpf_map_get_fd_by_id(&attr);
5494		break;
5495	case BPF_OBJ_GET_INFO_BY_FD:
5496		err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5497		break;
5498	case BPF_RAW_TRACEPOINT_OPEN:
5499		err = bpf_raw_tracepoint_open(&attr);
5500		break;
5501	case BPF_BTF_LOAD:
5502		err = bpf_btf_load(&attr, uattr, size);
5503		break;
5504	case BPF_BTF_GET_FD_BY_ID:
5505		err = bpf_btf_get_fd_by_id(&attr);
5506		break;
5507	case BPF_TASK_FD_QUERY:
5508		err = bpf_task_fd_query(&attr, uattr.user);
5509		break;
5510	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5511		err = map_lookup_and_delete_elem(&attr);
5512		break;
5513	case BPF_MAP_LOOKUP_BATCH:
5514		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5515		break;
5516	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5517		err = bpf_map_do_batch(&attr, uattr.user,
5518				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5519		break;
5520	case BPF_MAP_UPDATE_BATCH:
5521		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5522		break;
5523	case BPF_MAP_DELETE_BATCH:
5524		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5525		break;
5526	case BPF_LINK_CREATE:
5527		err = link_create(&attr, uattr);
5528		break;
5529	case BPF_LINK_UPDATE:
5530		err = link_update(&attr);
5531		break;
5532	case BPF_LINK_GET_FD_BY_ID:
5533		err = bpf_link_get_fd_by_id(&attr);
5534		break;
5535	case BPF_LINK_GET_NEXT_ID:
5536		err = bpf_obj_get_next_id(&attr, uattr.user,
5537					  &link_idr, &link_idr_lock);
5538		break;
5539	case BPF_ENABLE_STATS:
5540		err = bpf_enable_stats(&attr);
5541		break;
5542	case BPF_ITER_CREATE:
5543		err = bpf_iter_create(&attr);
5544		break;
5545	case BPF_LINK_DETACH:
5546		err = link_detach(&attr);
5547		break;
5548	case BPF_PROG_BIND_MAP:
5549		err = bpf_prog_bind_map(&attr);
5550		break;
5551	default:
5552		err = -EINVAL;
5553		break;
5554	}
5555
5556	return err;
5557}
5558
5559SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5560{
5561	return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5562}
5563
5564static bool syscall_prog_is_valid_access(int off, int size,
5565					 enum bpf_access_type type,
5566					 const struct bpf_prog *prog,
5567					 struct bpf_insn_access_aux *info)
5568{
5569	if (off < 0 || off >= U16_MAX)
5570		return false;
5571	if (off % size != 0)
5572		return false;
5573	return true;
5574}
5575
5576BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5577{
5578	switch (cmd) {
5579	case BPF_MAP_CREATE:
5580	case BPF_MAP_DELETE_ELEM:
5581	case BPF_MAP_UPDATE_ELEM:
5582	case BPF_MAP_FREEZE:
5583	case BPF_MAP_GET_FD_BY_ID:
5584	case BPF_PROG_LOAD:
5585	case BPF_BTF_LOAD:
5586	case BPF_LINK_CREATE:
5587	case BPF_RAW_TRACEPOINT_OPEN:
5588		break;
5589	default:
5590		return -EINVAL;
5591	}
5592	return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5593}
5594
5595
5596/* To shut up -Wmissing-prototypes.
5597 * This function is used by the kernel light skeleton
5598 * to load bpf programs when modules are loaded or during kernel boot.
5599 * See tools/lib/bpf/skel_internal.h
5600 */
5601int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5602
5603int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5604{
5605	struct bpf_prog * __maybe_unused prog;
5606	struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5607
5608	switch (cmd) {
5609#ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5610	case BPF_PROG_TEST_RUN:
5611		if (attr->test.data_in || attr->test.data_out ||
5612		    attr->test.ctx_out || attr->test.duration ||
5613		    attr->test.repeat || attr->test.flags)
5614			return -EINVAL;
5615
5616		prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5617		if (IS_ERR(prog))
5618			return PTR_ERR(prog);
5619
5620		if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5621		    attr->test.ctx_size_in > U16_MAX) {
5622			bpf_prog_put(prog);
5623			return -EINVAL;
5624		}
5625
5626		run_ctx.bpf_cookie = 0;
 
5627		if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5628			/* recursion detected */
5629			__bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5630			bpf_prog_put(prog);
5631			return -EBUSY;
5632		}
5633		attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5634		__bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5635						&run_ctx);
5636		bpf_prog_put(prog);
5637		return 0;
5638#endif
5639	default:
5640		return ____bpf_sys_bpf(cmd, attr, size);
5641	}
5642}
5643EXPORT_SYMBOL(kern_sys_bpf);
5644
5645static const struct bpf_func_proto bpf_sys_bpf_proto = {
5646	.func		= bpf_sys_bpf,
5647	.gpl_only	= false,
5648	.ret_type	= RET_INTEGER,
5649	.arg1_type	= ARG_ANYTHING,
5650	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
5651	.arg3_type	= ARG_CONST_SIZE,
5652};
5653
5654const struct bpf_func_proto * __weak
5655tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5656{
5657	return bpf_base_func_proto(func_id);
5658}
5659
5660BPF_CALL_1(bpf_sys_close, u32, fd)
5661{
5662	/* When bpf program calls this helper there should not be
5663	 * an fdget() without matching completed fdput().
5664	 * This helper is allowed in the following callchain only:
5665	 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5666	 */
5667	return close_fd(fd);
5668}
5669
5670static const struct bpf_func_proto bpf_sys_close_proto = {
5671	.func		= bpf_sys_close,
5672	.gpl_only	= false,
5673	.ret_type	= RET_INTEGER,
5674	.arg1_type	= ARG_ANYTHING,
5675};
5676
5677BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5678{
5679	if (flags)
5680		return -EINVAL;
5681
5682	if (name_sz <= 1 || name[name_sz - 1])
5683		return -EINVAL;
5684
5685	if (!bpf_dump_raw_ok(current_cred()))
5686		return -EPERM;
5687
5688	*res = kallsyms_lookup_name(name);
5689	return *res ? 0 : -ENOENT;
5690}
5691
5692static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5693	.func		= bpf_kallsyms_lookup_name,
5694	.gpl_only	= false,
5695	.ret_type	= RET_INTEGER,
5696	.arg1_type	= ARG_PTR_TO_MEM,
5697	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
5698	.arg3_type	= ARG_ANYTHING,
5699	.arg4_type	= ARG_PTR_TO_LONG,
5700};
5701
5702static const struct bpf_func_proto *
5703syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5704{
5705	switch (func_id) {
5706	case BPF_FUNC_sys_bpf:
5707		return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5708	case BPF_FUNC_btf_find_by_name_kind:
5709		return &bpf_btf_find_by_name_kind_proto;
5710	case BPF_FUNC_sys_close:
5711		return &bpf_sys_close_proto;
5712	case BPF_FUNC_kallsyms_lookup_name:
5713		return &bpf_kallsyms_lookup_name_proto;
5714	default:
5715		return tracing_prog_func_proto(func_id, prog);
5716	}
5717}
5718
5719const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5720	.get_func_proto  = syscall_prog_func_proto,
5721	.is_valid_access = syscall_prog_is_valid_access,
5722};
5723
5724const struct bpf_prog_ops bpf_syscall_prog_ops = {
5725	.test_run = bpf_prog_test_run_syscall,
5726};
5727
5728#ifdef CONFIG_SYSCTL
5729static int bpf_stats_handler(struct ctl_table *table, int write,
5730			     void *buffer, size_t *lenp, loff_t *ppos)
5731{
5732	struct static_key *key = (struct static_key *)table->data;
5733	static int saved_val;
5734	int val, ret;
5735	struct ctl_table tmp = {
5736		.data   = &val,
5737		.maxlen = sizeof(val),
5738		.mode   = table->mode,
5739		.extra1 = SYSCTL_ZERO,
5740		.extra2 = SYSCTL_ONE,
5741	};
5742
5743	if (write && !capable(CAP_SYS_ADMIN))
5744		return -EPERM;
5745
5746	mutex_lock(&bpf_stats_enabled_mutex);
5747	val = saved_val;
5748	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5749	if (write && !ret && val != saved_val) {
5750		if (val)
5751			static_key_slow_inc(key);
5752		else
5753			static_key_slow_dec(key);
5754		saved_val = val;
5755	}
5756	mutex_unlock(&bpf_stats_enabled_mutex);
5757	return ret;
5758}
5759
5760void __weak unpriv_ebpf_notify(int new_state)
5761{
5762}
5763
5764static int bpf_unpriv_handler(struct ctl_table *table, int write,
5765			      void *buffer, size_t *lenp, loff_t *ppos)
5766{
5767	int ret, unpriv_enable = *(int *)table->data;
5768	bool locked_state = unpriv_enable == 1;
5769	struct ctl_table tmp = *table;
5770
5771	if (write && !capable(CAP_SYS_ADMIN))
5772		return -EPERM;
5773
5774	tmp.data = &unpriv_enable;
5775	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5776	if (write && !ret) {
5777		if (locked_state && unpriv_enable != 1)
5778			return -EPERM;
5779		*(int *)table->data = unpriv_enable;
5780	}
5781
5782	if (write)
5783		unpriv_ebpf_notify(unpriv_enable);
5784
5785	return ret;
5786}
5787
5788static struct ctl_table bpf_syscall_table[] = {
5789	{
5790		.procname	= "unprivileged_bpf_disabled",
5791		.data		= &sysctl_unprivileged_bpf_disabled,
5792		.maxlen		= sizeof(sysctl_unprivileged_bpf_disabled),
5793		.mode		= 0644,
5794		.proc_handler	= bpf_unpriv_handler,
5795		.extra1		= SYSCTL_ZERO,
5796		.extra2		= SYSCTL_TWO,
5797	},
5798	{
5799		.procname	= "bpf_stats_enabled",
5800		.data		= &bpf_stats_enabled_key.key,
 
5801		.mode		= 0644,
5802		.proc_handler	= bpf_stats_handler,
5803	},
5804	{ }
5805};
5806
5807static int __init bpf_syscall_sysctl_init(void)
5808{
5809	register_sysctl_init("kernel", bpf_syscall_table);
5810	return 0;
5811}
5812late_initcall(bpf_syscall_sysctl_init);
5813#endif /* CONFIG_SYSCTL */
v6.2
   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 */