xref: /linux/kernel/bpf/syscall.c (revision 246880958ac93989c97c73ae1e60b78b4c4c88c5)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  */
4 #include <linux/bpf.h>
5 #include <linux/bpf_trace.h>
6 #include <linux/bpf_lirc.h>
7 #include <linux/btf.h>
8 #include <linux/syscalls.h>
9 #include <linux/slab.h>
10 #include <linux/sched/signal.h>
11 #include <linux/vmalloc.h>
12 #include <linux/mmzone.h>
13 #include <linux/anon_inodes.h>
14 #include <linux/fdtable.h>
15 #include <linux/file.h>
16 #include <linux/fs.h>
17 #include <linux/license.h>
18 #include <linux/filter.h>
19 #include <linux/version.h>
20 #include <linux/kernel.h>
21 #include <linux/idr.h>
22 #include <linux/cred.h>
23 #include <linux/timekeeping.h>
24 #include <linux/ctype.h>
25 #include <linux/nospec.h>
26 #include <uapi/linux/btf.h>
27 
28 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
29 			   (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
30 			   (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
31 			   (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
32 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
33 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
34 
35 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
36 
37 DEFINE_PER_CPU(int, bpf_prog_active);
38 static DEFINE_IDR(prog_idr);
39 static DEFINE_SPINLOCK(prog_idr_lock);
40 static DEFINE_IDR(map_idr);
41 static DEFINE_SPINLOCK(map_idr_lock);
42 
43 int sysctl_unprivileged_bpf_disabled __read_mostly;
44 
45 static const struct bpf_map_ops * const bpf_map_types[] = {
46 #define BPF_PROG_TYPE(_id, _ops)
47 #define BPF_MAP_TYPE(_id, _ops) \
48 	[_id] = &_ops,
49 #include <linux/bpf_types.h>
50 #undef BPF_PROG_TYPE
51 #undef BPF_MAP_TYPE
52 };
53 
54 /*
55  * If we're handed a bigger struct than we know of, ensure all the unknown bits
56  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
57  * we don't know about yet.
58  *
59  * There is a ToCToU between this function call and the following
60  * copy_from_user() call. However, this is not a concern since this function is
61  * meant to be a future-proofing of bits.
62  */
63 int bpf_check_uarg_tail_zero(void __user *uaddr,
64 			     size_t expected_size,
65 			     size_t actual_size)
66 {
67 	unsigned char __user *addr;
68 	unsigned char __user *end;
69 	unsigned char val;
70 	int err;
71 
72 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
73 		return -E2BIG;
74 
75 	if (unlikely(!access_ok(uaddr, actual_size)))
76 		return -EFAULT;
77 
78 	if (actual_size <= expected_size)
79 		return 0;
80 
81 	addr = uaddr + expected_size;
82 	end  = uaddr + actual_size;
83 
84 	for (; addr < end; addr++) {
85 		err = get_user(val, addr);
86 		if (err)
87 			return err;
88 		if (val)
89 			return -E2BIG;
90 	}
91 
92 	return 0;
93 }
94 
95 const struct bpf_map_ops bpf_map_offload_ops = {
96 	.map_alloc = bpf_map_offload_map_alloc,
97 	.map_free = bpf_map_offload_map_free,
98 	.map_check_btf = map_check_no_btf,
99 };
100 
101 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
102 {
103 	const struct bpf_map_ops *ops;
104 	u32 type = attr->map_type;
105 	struct bpf_map *map;
106 	int err;
107 
108 	if (type >= ARRAY_SIZE(bpf_map_types))
109 		return ERR_PTR(-EINVAL);
110 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
111 	ops = bpf_map_types[type];
112 	if (!ops)
113 		return ERR_PTR(-EINVAL);
114 
115 	if (ops->map_alloc_check) {
116 		err = ops->map_alloc_check(attr);
117 		if (err)
118 			return ERR_PTR(err);
119 	}
120 	if (attr->map_ifindex)
121 		ops = &bpf_map_offload_ops;
122 	map = ops->map_alloc(attr);
123 	if (IS_ERR(map))
124 		return map;
125 	map->ops = ops;
126 	map->map_type = type;
127 	return map;
128 }
129 
130 void *bpf_map_area_alloc(size_t size, int numa_node)
131 {
132 	/* We really just want to fail instead of triggering OOM killer
133 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
134 	 * which is used for lower order allocation requests.
135 	 *
136 	 * It has been observed that higher order allocation requests done by
137 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
138 	 * to reclaim memory from the page cache, thus we set
139 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
140 	 */
141 
142 	const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
143 	void *area;
144 
145 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
146 		area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
147 				    numa_node);
148 		if (area != NULL)
149 			return area;
150 	}
151 
152 	return __vmalloc_node_flags_caller(size, numa_node,
153 					   GFP_KERNEL | __GFP_RETRY_MAYFAIL |
154 					   flags, __builtin_return_address(0));
155 }
156 
157 void bpf_map_area_free(void *area)
158 {
159 	kvfree(area);
160 }
161 
162 static u32 bpf_map_flags_retain_permanent(u32 flags)
163 {
164 	/* Some map creation flags are not tied to the map object but
165 	 * rather to the map fd instead, so they have no meaning upon
166 	 * map object inspection since multiple file descriptors with
167 	 * different (access) properties can exist here. Thus, given
168 	 * this has zero meaning for the map itself, lets clear these
169 	 * from here.
170 	 */
171 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
172 }
173 
174 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
175 {
176 	map->map_type = attr->map_type;
177 	map->key_size = attr->key_size;
178 	map->value_size = attr->value_size;
179 	map->max_entries = attr->max_entries;
180 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
181 	map->numa_node = bpf_map_attr_numa_node(attr);
182 }
183 
184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
185 {
186 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
187 
188 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
189 		atomic_long_sub(pages, &user->locked_vm);
190 		return -EPERM;
191 	}
192 	return 0;
193 }
194 
195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
196 {
197 	if (user)
198 		atomic_long_sub(pages, &user->locked_vm);
199 }
200 
201 int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size)
202 {
203 	u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
204 	struct user_struct *user;
205 	int ret;
206 
207 	if (size >= U32_MAX - PAGE_SIZE)
208 		return -E2BIG;
209 
210 	user = get_current_user();
211 	ret = bpf_charge_memlock(user, pages);
212 	if (ret) {
213 		free_uid(user);
214 		return ret;
215 	}
216 
217 	mem->pages = pages;
218 	mem->user = user;
219 
220 	return 0;
221 }
222 
223 void bpf_map_charge_finish(struct bpf_map_memory *mem)
224 {
225 	bpf_uncharge_memlock(mem->user, mem->pages);
226 	free_uid(mem->user);
227 }
228 
229 void bpf_map_charge_move(struct bpf_map_memory *dst,
230 			 struct bpf_map_memory *src)
231 {
232 	*dst = *src;
233 
234 	/* Make sure src will not be used for the redundant uncharging. */
235 	memset(src, 0, sizeof(struct bpf_map_memory));
236 }
237 
238 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
239 {
240 	int ret;
241 
242 	ret = bpf_charge_memlock(map->memory.user, pages);
243 	if (ret)
244 		return ret;
245 	map->memory.pages += pages;
246 	return ret;
247 }
248 
249 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
250 {
251 	bpf_uncharge_memlock(map->memory.user, pages);
252 	map->memory.pages -= pages;
253 }
254 
255 static int bpf_map_alloc_id(struct bpf_map *map)
256 {
257 	int id;
258 
259 	idr_preload(GFP_KERNEL);
260 	spin_lock_bh(&map_idr_lock);
261 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
262 	if (id > 0)
263 		map->id = id;
264 	spin_unlock_bh(&map_idr_lock);
265 	idr_preload_end();
266 
267 	if (WARN_ON_ONCE(!id))
268 		return -ENOSPC;
269 
270 	return id > 0 ? 0 : id;
271 }
272 
273 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
274 {
275 	unsigned long flags;
276 
277 	/* Offloaded maps are removed from the IDR store when their device
278 	 * disappears - even if someone holds an fd to them they are unusable,
279 	 * the memory is gone, all ops will fail; they are simply waiting for
280 	 * refcnt to drop to be freed.
281 	 */
282 	if (!map->id)
283 		return;
284 
285 	if (do_idr_lock)
286 		spin_lock_irqsave(&map_idr_lock, flags);
287 	else
288 		__acquire(&map_idr_lock);
289 
290 	idr_remove(&map_idr, map->id);
291 	map->id = 0;
292 
293 	if (do_idr_lock)
294 		spin_unlock_irqrestore(&map_idr_lock, flags);
295 	else
296 		__release(&map_idr_lock);
297 }
298 
299 /* called from workqueue */
300 static void bpf_map_free_deferred(struct work_struct *work)
301 {
302 	struct bpf_map *map = container_of(work, struct bpf_map, work);
303 	struct bpf_map_memory mem;
304 
305 	bpf_map_charge_move(&mem, &map->memory);
306 	security_bpf_map_free(map);
307 	/* implementation dependent freeing */
308 	map->ops->map_free(map);
309 	bpf_map_charge_finish(&mem);
310 }
311 
312 static void bpf_map_put_uref(struct bpf_map *map)
313 {
314 	if (atomic_dec_and_test(&map->usercnt)) {
315 		if (map->ops->map_release_uref)
316 			map->ops->map_release_uref(map);
317 	}
318 }
319 
320 /* decrement map refcnt and schedule it for freeing via workqueue
321  * (unrelying map implementation ops->map_free() might sleep)
322  */
323 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
324 {
325 	if (atomic_dec_and_test(&map->refcnt)) {
326 		/* bpf_map_free_id() must be called first */
327 		bpf_map_free_id(map, do_idr_lock);
328 		btf_put(map->btf);
329 		INIT_WORK(&map->work, bpf_map_free_deferred);
330 		schedule_work(&map->work);
331 	}
332 }
333 
334 void bpf_map_put(struct bpf_map *map)
335 {
336 	__bpf_map_put(map, true);
337 }
338 EXPORT_SYMBOL_GPL(bpf_map_put);
339 
340 void bpf_map_put_with_uref(struct bpf_map *map)
341 {
342 	bpf_map_put_uref(map);
343 	bpf_map_put(map);
344 }
345 
346 static int bpf_map_release(struct inode *inode, struct file *filp)
347 {
348 	struct bpf_map *map = filp->private_data;
349 
350 	if (map->ops->map_release)
351 		map->ops->map_release(map, filp);
352 
353 	bpf_map_put_with_uref(map);
354 	return 0;
355 }
356 
357 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
358 {
359 	fmode_t mode = f.file->f_mode;
360 
361 	/* Our file permissions may have been overridden by global
362 	 * map permissions facing syscall side.
363 	 */
364 	if (READ_ONCE(map->frozen))
365 		mode &= ~FMODE_CAN_WRITE;
366 	return mode;
367 }
368 
369 #ifdef CONFIG_PROC_FS
370 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
371 {
372 	const struct bpf_map *map = filp->private_data;
373 	const struct bpf_array *array;
374 	u32 owner_prog_type = 0;
375 	u32 owner_jited = 0;
376 
377 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
378 		array = container_of(map, struct bpf_array, map);
379 		owner_prog_type = array->owner_prog_type;
380 		owner_jited = array->owner_jited;
381 	}
382 
383 	seq_printf(m,
384 		   "map_type:\t%u\n"
385 		   "key_size:\t%u\n"
386 		   "value_size:\t%u\n"
387 		   "max_entries:\t%u\n"
388 		   "map_flags:\t%#x\n"
389 		   "memlock:\t%llu\n"
390 		   "map_id:\t%u\n"
391 		   "frozen:\t%u\n",
392 		   map->map_type,
393 		   map->key_size,
394 		   map->value_size,
395 		   map->max_entries,
396 		   map->map_flags,
397 		   map->memory.pages * 1ULL << PAGE_SHIFT,
398 		   map->id,
399 		   READ_ONCE(map->frozen));
400 
401 	if (owner_prog_type) {
402 		seq_printf(m, "owner_prog_type:\t%u\n",
403 			   owner_prog_type);
404 		seq_printf(m, "owner_jited:\t%u\n",
405 			   owner_jited);
406 	}
407 }
408 #endif
409 
410 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
411 			      loff_t *ppos)
412 {
413 	/* We need this handler such that alloc_file() enables
414 	 * f_mode with FMODE_CAN_READ.
415 	 */
416 	return -EINVAL;
417 }
418 
419 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
420 			       size_t siz, loff_t *ppos)
421 {
422 	/* We need this handler such that alloc_file() enables
423 	 * f_mode with FMODE_CAN_WRITE.
424 	 */
425 	return -EINVAL;
426 }
427 
428 const struct file_operations bpf_map_fops = {
429 #ifdef CONFIG_PROC_FS
430 	.show_fdinfo	= bpf_map_show_fdinfo,
431 #endif
432 	.release	= bpf_map_release,
433 	.read		= bpf_dummy_read,
434 	.write		= bpf_dummy_write,
435 };
436 
437 int bpf_map_new_fd(struct bpf_map *map, int flags)
438 {
439 	int ret;
440 
441 	ret = security_bpf_map(map, OPEN_FMODE(flags));
442 	if (ret < 0)
443 		return ret;
444 
445 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
446 				flags | O_CLOEXEC);
447 }
448 
449 int bpf_get_file_flag(int flags)
450 {
451 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
452 		return -EINVAL;
453 	if (flags & BPF_F_RDONLY)
454 		return O_RDONLY;
455 	if (flags & BPF_F_WRONLY)
456 		return O_WRONLY;
457 	return O_RDWR;
458 }
459 
460 /* helper macro to check that unused fields 'union bpf_attr' are zero */
461 #define CHECK_ATTR(CMD) \
462 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
463 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
464 		   sizeof(*attr) - \
465 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
466 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
467 
468 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
469  * Return 0 on success and < 0 on error.
470  */
471 static int bpf_obj_name_cpy(char *dst, const char *src)
472 {
473 	const char *end = src + BPF_OBJ_NAME_LEN;
474 
475 	memset(dst, 0, BPF_OBJ_NAME_LEN);
476 	/* Copy all isalnum(), '_' and '.' chars. */
477 	while (src < end && *src) {
478 		if (!isalnum(*src) &&
479 		    *src != '_' && *src != '.')
480 			return -EINVAL;
481 		*dst++ = *src++;
482 	}
483 
484 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
485 	if (src == end)
486 		return -EINVAL;
487 
488 	return 0;
489 }
490 
491 int map_check_no_btf(const struct bpf_map *map,
492 		     const struct btf *btf,
493 		     const struct btf_type *key_type,
494 		     const struct btf_type *value_type)
495 {
496 	return -ENOTSUPP;
497 }
498 
499 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
500 			 u32 btf_key_id, u32 btf_value_id)
501 {
502 	const struct btf_type *key_type, *value_type;
503 	u32 key_size, value_size;
504 	int ret = 0;
505 
506 	/* Some maps allow key to be unspecified. */
507 	if (btf_key_id) {
508 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
509 		if (!key_type || key_size != map->key_size)
510 			return -EINVAL;
511 	} else {
512 		key_type = btf_type_by_id(btf, 0);
513 		if (!map->ops->map_check_btf)
514 			return -EINVAL;
515 	}
516 
517 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
518 	if (!value_type || value_size != map->value_size)
519 		return -EINVAL;
520 
521 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
522 
523 	if (map_value_has_spin_lock(map)) {
524 		if (map->map_flags & BPF_F_RDONLY_PROG)
525 			return -EACCES;
526 		if (map->map_type != BPF_MAP_TYPE_HASH &&
527 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
528 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
529 		    map->map_type != BPF_MAP_TYPE_SK_STORAGE)
530 			return -ENOTSUPP;
531 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
532 		    map->value_size) {
533 			WARN_ONCE(1,
534 				  "verifier bug spin_lock_off %d value_size %d\n",
535 				  map->spin_lock_off, map->value_size);
536 			return -EFAULT;
537 		}
538 	}
539 
540 	if (map->ops->map_check_btf)
541 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
542 
543 	return ret;
544 }
545 
546 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
547 /* called via syscall */
548 static int map_create(union bpf_attr *attr)
549 {
550 	int numa_node = bpf_map_attr_numa_node(attr);
551 	struct bpf_map_memory mem;
552 	struct bpf_map *map;
553 	int f_flags;
554 	int err;
555 
556 	err = CHECK_ATTR(BPF_MAP_CREATE);
557 	if (err)
558 		return -EINVAL;
559 
560 	f_flags = bpf_get_file_flag(attr->map_flags);
561 	if (f_flags < 0)
562 		return f_flags;
563 
564 	if (numa_node != NUMA_NO_NODE &&
565 	    ((unsigned int)numa_node >= nr_node_ids ||
566 	     !node_online(numa_node)))
567 		return -EINVAL;
568 
569 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
570 	map = find_and_alloc_map(attr);
571 	if (IS_ERR(map))
572 		return PTR_ERR(map);
573 
574 	err = bpf_obj_name_cpy(map->name, attr->map_name);
575 	if (err)
576 		goto free_map;
577 
578 	atomic_set(&map->refcnt, 1);
579 	atomic_set(&map->usercnt, 1);
580 
581 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
582 		struct btf *btf;
583 
584 		if (!attr->btf_value_type_id) {
585 			err = -EINVAL;
586 			goto free_map;
587 		}
588 
589 		btf = btf_get_by_fd(attr->btf_fd);
590 		if (IS_ERR(btf)) {
591 			err = PTR_ERR(btf);
592 			goto free_map;
593 		}
594 
595 		err = map_check_btf(map, btf, attr->btf_key_type_id,
596 				    attr->btf_value_type_id);
597 		if (err) {
598 			btf_put(btf);
599 			goto free_map;
600 		}
601 
602 		map->btf = btf;
603 		map->btf_key_type_id = attr->btf_key_type_id;
604 		map->btf_value_type_id = attr->btf_value_type_id;
605 	} else {
606 		map->spin_lock_off = -EINVAL;
607 	}
608 
609 	err = security_bpf_map_alloc(map);
610 	if (err)
611 		goto free_map;
612 
613 	err = bpf_map_alloc_id(map);
614 	if (err)
615 		goto free_map_sec;
616 
617 	err = bpf_map_new_fd(map, f_flags);
618 	if (err < 0) {
619 		/* failed to allocate fd.
620 		 * bpf_map_put_with_uref() is needed because the above
621 		 * bpf_map_alloc_id() has published the map
622 		 * to the userspace and the userspace may
623 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
624 		 */
625 		bpf_map_put_with_uref(map);
626 		return err;
627 	}
628 
629 	return err;
630 
631 free_map_sec:
632 	security_bpf_map_free(map);
633 free_map:
634 	btf_put(map->btf);
635 	bpf_map_charge_move(&mem, &map->memory);
636 	map->ops->map_free(map);
637 	bpf_map_charge_finish(&mem);
638 	return err;
639 }
640 
641 /* if error is returned, fd is released.
642  * On success caller should complete fd access with matching fdput()
643  */
644 struct bpf_map *__bpf_map_get(struct fd f)
645 {
646 	if (!f.file)
647 		return ERR_PTR(-EBADF);
648 	if (f.file->f_op != &bpf_map_fops) {
649 		fdput(f);
650 		return ERR_PTR(-EINVAL);
651 	}
652 
653 	return f.file->private_data;
654 }
655 
656 /* prog's and map's refcnt limit */
657 #define BPF_MAX_REFCNT 32768
658 
659 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
660 {
661 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
662 		atomic_dec(&map->refcnt);
663 		return ERR_PTR(-EBUSY);
664 	}
665 	if (uref)
666 		atomic_inc(&map->usercnt);
667 	return map;
668 }
669 EXPORT_SYMBOL_GPL(bpf_map_inc);
670 
671 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
672 {
673 	struct fd f = fdget(ufd);
674 	struct bpf_map *map;
675 
676 	map = __bpf_map_get(f);
677 	if (IS_ERR(map))
678 		return map;
679 
680 	map = bpf_map_inc(map, true);
681 	fdput(f);
682 
683 	return map;
684 }
685 
686 /* map_idr_lock should have been held */
687 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map,
688 					      bool uref)
689 {
690 	int refold;
691 
692 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
693 
694 	if (refold >= BPF_MAX_REFCNT) {
695 		__bpf_map_put(map, false);
696 		return ERR_PTR(-EBUSY);
697 	}
698 
699 	if (!refold)
700 		return ERR_PTR(-ENOENT);
701 
702 	if (uref)
703 		atomic_inc(&map->usercnt);
704 
705 	return map;
706 }
707 
708 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
709 {
710 	spin_lock_bh(&map_idr_lock);
711 	map = __bpf_map_inc_not_zero(map, uref);
712 	spin_unlock_bh(&map_idr_lock);
713 
714 	return map;
715 }
716 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
717 
718 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
719 {
720 	return -ENOTSUPP;
721 }
722 
723 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
724 {
725 	if (key_size)
726 		return memdup_user(ukey, key_size);
727 
728 	if (ukey)
729 		return ERR_PTR(-EINVAL);
730 
731 	return NULL;
732 }
733 
734 /* last field in 'union bpf_attr' used by this command */
735 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
736 
737 static int map_lookup_elem(union bpf_attr *attr)
738 {
739 	void __user *ukey = u64_to_user_ptr(attr->key);
740 	void __user *uvalue = u64_to_user_ptr(attr->value);
741 	int ufd = attr->map_fd;
742 	struct bpf_map *map;
743 	void *key, *value, *ptr;
744 	u32 value_size;
745 	struct fd f;
746 	int err;
747 
748 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
749 		return -EINVAL;
750 
751 	if (attr->flags & ~BPF_F_LOCK)
752 		return -EINVAL;
753 
754 	f = fdget(ufd);
755 	map = __bpf_map_get(f);
756 	if (IS_ERR(map))
757 		return PTR_ERR(map);
758 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
759 		err = -EPERM;
760 		goto err_put;
761 	}
762 
763 	if ((attr->flags & BPF_F_LOCK) &&
764 	    !map_value_has_spin_lock(map)) {
765 		err = -EINVAL;
766 		goto err_put;
767 	}
768 
769 	key = __bpf_copy_key(ukey, map->key_size);
770 	if (IS_ERR(key)) {
771 		err = PTR_ERR(key);
772 		goto err_put;
773 	}
774 
775 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
776 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
777 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
778 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
779 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
780 	else if (IS_FD_MAP(map))
781 		value_size = sizeof(u32);
782 	else
783 		value_size = map->value_size;
784 
785 	err = -ENOMEM;
786 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
787 	if (!value)
788 		goto free_key;
789 
790 	if (bpf_map_is_dev_bound(map)) {
791 		err = bpf_map_offload_lookup_elem(map, key, value);
792 		goto done;
793 	}
794 
795 	preempt_disable();
796 	this_cpu_inc(bpf_prog_active);
797 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
798 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
799 		err = bpf_percpu_hash_copy(map, key, value);
800 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
801 		err = bpf_percpu_array_copy(map, key, value);
802 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
803 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
804 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
805 		err = bpf_stackmap_copy(map, key, value);
806 	} else if (IS_FD_ARRAY(map)) {
807 		err = bpf_fd_array_map_lookup_elem(map, key, value);
808 	} else if (IS_FD_HASH(map)) {
809 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
810 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
811 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
812 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
813 		   map->map_type == BPF_MAP_TYPE_STACK) {
814 		err = map->ops->map_peek_elem(map, value);
815 	} else {
816 		rcu_read_lock();
817 		if (map->ops->map_lookup_elem_sys_only)
818 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
819 		else
820 			ptr = map->ops->map_lookup_elem(map, key);
821 		if (IS_ERR(ptr)) {
822 			err = PTR_ERR(ptr);
823 		} else if (!ptr) {
824 			err = -ENOENT;
825 		} else {
826 			err = 0;
827 			if (attr->flags & BPF_F_LOCK)
828 				/* lock 'ptr' and copy everything but lock */
829 				copy_map_value_locked(map, value, ptr, true);
830 			else
831 				copy_map_value(map, value, ptr);
832 			/* mask lock, since value wasn't zero inited */
833 			check_and_init_map_lock(map, value);
834 		}
835 		rcu_read_unlock();
836 	}
837 	this_cpu_dec(bpf_prog_active);
838 	preempt_enable();
839 
840 done:
841 	if (err)
842 		goto free_value;
843 
844 	err = -EFAULT;
845 	if (copy_to_user(uvalue, value, value_size) != 0)
846 		goto free_value;
847 
848 	err = 0;
849 
850 free_value:
851 	kfree(value);
852 free_key:
853 	kfree(key);
854 err_put:
855 	fdput(f);
856 	return err;
857 }
858 
859 static void maybe_wait_bpf_programs(struct bpf_map *map)
860 {
861 	/* Wait for any running BPF programs to complete so that
862 	 * userspace, when we return to it, knows that all programs
863 	 * that could be running use the new map value.
864 	 */
865 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
866 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
867 		synchronize_rcu();
868 }
869 
870 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
871 
872 static int map_update_elem(union bpf_attr *attr)
873 {
874 	void __user *ukey = u64_to_user_ptr(attr->key);
875 	void __user *uvalue = u64_to_user_ptr(attr->value);
876 	int ufd = attr->map_fd;
877 	struct bpf_map *map;
878 	void *key, *value;
879 	u32 value_size;
880 	struct fd f;
881 	int err;
882 
883 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
884 		return -EINVAL;
885 
886 	f = fdget(ufd);
887 	map = __bpf_map_get(f);
888 	if (IS_ERR(map))
889 		return PTR_ERR(map);
890 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
891 		err = -EPERM;
892 		goto err_put;
893 	}
894 
895 	if ((attr->flags & BPF_F_LOCK) &&
896 	    !map_value_has_spin_lock(map)) {
897 		err = -EINVAL;
898 		goto err_put;
899 	}
900 
901 	key = __bpf_copy_key(ukey, map->key_size);
902 	if (IS_ERR(key)) {
903 		err = PTR_ERR(key);
904 		goto err_put;
905 	}
906 
907 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
908 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
909 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
910 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
911 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
912 	else
913 		value_size = map->value_size;
914 
915 	err = -ENOMEM;
916 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
917 	if (!value)
918 		goto free_key;
919 
920 	err = -EFAULT;
921 	if (copy_from_user(value, uvalue, value_size) != 0)
922 		goto free_value;
923 
924 	/* Need to create a kthread, thus must support schedule */
925 	if (bpf_map_is_dev_bound(map)) {
926 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
927 		goto out;
928 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
929 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
930 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
931 		err = map->ops->map_update_elem(map, key, value, attr->flags);
932 		goto out;
933 	}
934 
935 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
936 	 * inside bpf map update or delete otherwise deadlocks are possible
937 	 */
938 	preempt_disable();
939 	__this_cpu_inc(bpf_prog_active);
940 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
941 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
942 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
943 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
944 		err = bpf_percpu_array_update(map, key, value, attr->flags);
945 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
946 		err = bpf_percpu_cgroup_storage_update(map, key, value,
947 						       attr->flags);
948 	} else if (IS_FD_ARRAY(map)) {
949 		rcu_read_lock();
950 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
951 						   attr->flags);
952 		rcu_read_unlock();
953 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
954 		rcu_read_lock();
955 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
956 						  attr->flags);
957 		rcu_read_unlock();
958 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
959 		/* rcu_read_lock() is not needed */
960 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
961 							 attr->flags);
962 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
963 		   map->map_type == BPF_MAP_TYPE_STACK) {
964 		err = map->ops->map_push_elem(map, value, attr->flags);
965 	} else {
966 		rcu_read_lock();
967 		err = map->ops->map_update_elem(map, key, value, attr->flags);
968 		rcu_read_unlock();
969 	}
970 	__this_cpu_dec(bpf_prog_active);
971 	preempt_enable();
972 	maybe_wait_bpf_programs(map);
973 out:
974 free_value:
975 	kfree(value);
976 free_key:
977 	kfree(key);
978 err_put:
979 	fdput(f);
980 	return err;
981 }
982 
983 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
984 
985 static int map_delete_elem(union bpf_attr *attr)
986 {
987 	void __user *ukey = u64_to_user_ptr(attr->key);
988 	int ufd = attr->map_fd;
989 	struct bpf_map *map;
990 	struct fd f;
991 	void *key;
992 	int err;
993 
994 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
995 		return -EINVAL;
996 
997 	f = fdget(ufd);
998 	map = __bpf_map_get(f);
999 	if (IS_ERR(map))
1000 		return PTR_ERR(map);
1001 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1002 		err = -EPERM;
1003 		goto err_put;
1004 	}
1005 
1006 	key = __bpf_copy_key(ukey, map->key_size);
1007 	if (IS_ERR(key)) {
1008 		err = PTR_ERR(key);
1009 		goto err_put;
1010 	}
1011 
1012 	if (bpf_map_is_dev_bound(map)) {
1013 		err = bpf_map_offload_delete_elem(map, key);
1014 		goto out;
1015 	}
1016 
1017 	preempt_disable();
1018 	__this_cpu_inc(bpf_prog_active);
1019 	rcu_read_lock();
1020 	err = map->ops->map_delete_elem(map, key);
1021 	rcu_read_unlock();
1022 	__this_cpu_dec(bpf_prog_active);
1023 	preempt_enable();
1024 	maybe_wait_bpf_programs(map);
1025 out:
1026 	kfree(key);
1027 err_put:
1028 	fdput(f);
1029 	return err;
1030 }
1031 
1032 /* last field in 'union bpf_attr' used by this command */
1033 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1034 
1035 static int map_get_next_key(union bpf_attr *attr)
1036 {
1037 	void __user *ukey = u64_to_user_ptr(attr->key);
1038 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1039 	int ufd = attr->map_fd;
1040 	struct bpf_map *map;
1041 	void *key, *next_key;
1042 	struct fd f;
1043 	int err;
1044 
1045 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1046 		return -EINVAL;
1047 
1048 	f = fdget(ufd);
1049 	map = __bpf_map_get(f);
1050 	if (IS_ERR(map))
1051 		return PTR_ERR(map);
1052 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1053 		err = -EPERM;
1054 		goto err_put;
1055 	}
1056 
1057 	if (ukey) {
1058 		key = __bpf_copy_key(ukey, map->key_size);
1059 		if (IS_ERR(key)) {
1060 			err = PTR_ERR(key);
1061 			goto err_put;
1062 		}
1063 	} else {
1064 		key = NULL;
1065 	}
1066 
1067 	err = -ENOMEM;
1068 	next_key = kmalloc(map->key_size, GFP_USER);
1069 	if (!next_key)
1070 		goto free_key;
1071 
1072 	if (bpf_map_is_dev_bound(map)) {
1073 		err = bpf_map_offload_get_next_key(map, key, next_key);
1074 		goto out;
1075 	}
1076 
1077 	rcu_read_lock();
1078 	err = map->ops->map_get_next_key(map, key, next_key);
1079 	rcu_read_unlock();
1080 out:
1081 	if (err)
1082 		goto free_next_key;
1083 
1084 	err = -EFAULT;
1085 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1086 		goto free_next_key;
1087 
1088 	err = 0;
1089 
1090 free_next_key:
1091 	kfree(next_key);
1092 free_key:
1093 	kfree(key);
1094 err_put:
1095 	fdput(f);
1096 	return err;
1097 }
1098 
1099 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1100 
1101 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1102 {
1103 	void __user *ukey = u64_to_user_ptr(attr->key);
1104 	void __user *uvalue = u64_to_user_ptr(attr->value);
1105 	int ufd = attr->map_fd;
1106 	struct bpf_map *map;
1107 	void *key, *value;
1108 	u32 value_size;
1109 	struct fd f;
1110 	int err;
1111 
1112 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1113 		return -EINVAL;
1114 
1115 	f = fdget(ufd);
1116 	map = __bpf_map_get(f);
1117 	if (IS_ERR(map))
1118 		return PTR_ERR(map);
1119 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1120 		err = -EPERM;
1121 		goto err_put;
1122 	}
1123 
1124 	key = __bpf_copy_key(ukey, map->key_size);
1125 	if (IS_ERR(key)) {
1126 		err = PTR_ERR(key);
1127 		goto err_put;
1128 	}
1129 
1130 	value_size = map->value_size;
1131 
1132 	err = -ENOMEM;
1133 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1134 	if (!value)
1135 		goto free_key;
1136 
1137 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1138 	    map->map_type == BPF_MAP_TYPE_STACK) {
1139 		err = map->ops->map_pop_elem(map, value);
1140 	} else {
1141 		err = -ENOTSUPP;
1142 	}
1143 
1144 	if (err)
1145 		goto free_value;
1146 
1147 	if (copy_to_user(uvalue, value, value_size) != 0)
1148 		goto free_value;
1149 
1150 	err = 0;
1151 
1152 free_value:
1153 	kfree(value);
1154 free_key:
1155 	kfree(key);
1156 err_put:
1157 	fdput(f);
1158 	return err;
1159 }
1160 
1161 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1162 
1163 static int map_freeze(const union bpf_attr *attr)
1164 {
1165 	int err = 0, ufd = attr->map_fd;
1166 	struct bpf_map *map;
1167 	struct fd f;
1168 
1169 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1170 		return -EINVAL;
1171 
1172 	f = fdget(ufd);
1173 	map = __bpf_map_get(f);
1174 	if (IS_ERR(map))
1175 		return PTR_ERR(map);
1176 	if (READ_ONCE(map->frozen)) {
1177 		err = -EBUSY;
1178 		goto err_put;
1179 	}
1180 	if (!capable(CAP_SYS_ADMIN)) {
1181 		err = -EPERM;
1182 		goto err_put;
1183 	}
1184 
1185 	WRITE_ONCE(map->frozen, true);
1186 err_put:
1187 	fdput(f);
1188 	return err;
1189 }
1190 
1191 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1192 #define BPF_PROG_TYPE(_id, _name) \
1193 	[_id] = & _name ## _prog_ops,
1194 #define BPF_MAP_TYPE(_id, _ops)
1195 #include <linux/bpf_types.h>
1196 #undef BPF_PROG_TYPE
1197 #undef BPF_MAP_TYPE
1198 };
1199 
1200 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1201 {
1202 	const struct bpf_prog_ops *ops;
1203 
1204 	if (type >= ARRAY_SIZE(bpf_prog_types))
1205 		return -EINVAL;
1206 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1207 	ops = bpf_prog_types[type];
1208 	if (!ops)
1209 		return -EINVAL;
1210 
1211 	if (!bpf_prog_is_dev_bound(prog->aux))
1212 		prog->aux->ops = ops;
1213 	else
1214 		prog->aux->ops = &bpf_offload_prog_ops;
1215 	prog->type = type;
1216 	return 0;
1217 }
1218 
1219 /* drop refcnt on maps used by eBPF program and free auxilary data */
1220 static void free_used_maps(struct bpf_prog_aux *aux)
1221 {
1222 	enum bpf_cgroup_storage_type stype;
1223 	int i;
1224 
1225 	for_each_cgroup_storage_type(stype) {
1226 		if (!aux->cgroup_storage[stype])
1227 			continue;
1228 		bpf_cgroup_storage_release(aux->prog,
1229 					   aux->cgroup_storage[stype]);
1230 	}
1231 
1232 	for (i = 0; i < aux->used_map_cnt; i++)
1233 		bpf_map_put(aux->used_maps[i]);
1234 
1235 	kfree(aux->used_maps);
1236 }
1237 
1238 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1239 {
1240 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1241 	unsigned long user_bufs;
1242 
1243 	if (user) {
1244 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1245 		if (user_bufs > memlock_limit) {
1246 			atomic_long_sub(pages, &user->locked_vm);
1247 			return -EPERM;
1248 		}
1249 	}
1250 
1251 	return 0;
1252 }
1253 
1254 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1255 {
1256 	if (user)
1257 		atomic_long_sub(pages, &user->locked_vm);
1258 }
1259 
1260 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1261 {
1262 	struct user_struct *user = get_current_user();
1263 	int ret;
1264 
1265 	ret = __bpf_prog_charge(user, prog->pages);
1266 	if (ret) {
1267 		free_uid(user);
1268 		return ret;
1269 	}
1270 
1271 	prog->aux->user = user;
1272 	return 0;
1273 }
1274 
1275 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1276 {
1277 	struct user_struct *user = prog->aux->user;
1278 
1279 	__bpf_prog_uncharge(user, prog->pages);
1280 	free_uid(user);
1281 }
1282 
1283 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1284 {
1285 	int id;
1286 
1287 	idr_preload(GFP_KERNEL);
1288 	spin_lock_bh(&prog_idr_lock);
1289 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1290 	if (id > 0)
1291 		prog->aux->id = id;
1292 	spin_unlock_bh(&prog_idr_lock);
1293 	idr_preload_end();
1294 
1295 	/* id is in [1, INT_MAX) */
1296 	if (WARN_ON_ONCE(!id))
1297 		return -ENOSPC;
1298 
1299 	return id > 0 ? 0 : id;
1300 }
1301 
1302 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1303 {
1304 	/* cBPF to eBPF migrations are currently not in the idr store.
1305 	 * Offloaded programs are removed from the store when their device
1306 	 * disappears - even if someone grabs an fd to them they are unusable,
1307 	 * simply waiting for refcnt to drop to be freed.
1308 	 */
1309 	if (!prog->aux->id)
1310 		return;
1311 
1312 	if (do_idr_lock)
1313 		spin_lock_bh(&prog_idr_lock);
1314 	else
1315 		__acquire(&prog_idr_lock);
1316 
1317 	idr_remove(&prog_idr, prog->aux->id);
1318 	prog->aux->id = 0;
1319 
1320 	if (do_idr_lock)
1321 		spin_unlock_bh(&prog_idr_lock);
1322 	else
1323 		__release(&prog_idr_lock);
1324 }
1325 
1326 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1327 {
1328 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1329 
1330 	free_used_maps(aux);
1331 	bpf_prog_uncharge_memlock(aux->prog);
1332 	security_bpf_prog_free(aux);
1333 	bpf_prog_free(aux->prog);
1334 }
1335 
1336 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1337 {
1338 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1339 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1340 		/* bpf_prog_free_id() must be called first */
1341 		bpf_prog_free_id(prog, do_idr_lock);
1342 		bpf_prog_kallsyms_del_all(prog);
1343 		btf_put(prog->aux->btf);
1344 		kvfree(prog->aux->func_info);
1345 		bpf_prog_free_linfo(prog);
1346 
1347 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1348 	}
1349 }
1350 
1351 void bpf_prog_put(struct bpf_prog *prog)
1352 {
1353 	__bpf_prog_put(prog, true);
1354 }
1355 EXPORT_SYMBOL_GPL(bpf_prog_put);
1356 
1357 static int bpf_prog_release(struct inode *inode, struct file *filp)
1358 {
1359 	struct bpf_prog *prog = filp->private_data;
1360 
1361 	bpf_prog_put(prog);
1362 	return 0;
1363 }
1364 
1365 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1366 			       struct bpf_prog_stats *stats)
1367 {
1368 	u64 nsecs = 0, cnt = 0;
1369 	int cpu;
1370 
1371 	for_each_possible_cpu(cpu) {
1372 		const struct bpf_prog_stats *st;
1373 		unsigned int start;
1374 		u64 tnsecs, tcnt;
1375 
1376 		st = per_cpu_ptr(prog->aux->stats, cpu);
1377 		do {
1378 			start = u64_stats_fetch_begin_irq(&st->syncp);
1379 			tnsecs = st->nsecs;
1380 			tcnt = st->cnt;
1381 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1382 		nsecs += tnsecs;
1383 		cnt += tcnt;
1384 	}
1385 	stats->nsecs = nsecs;
1386 	stats->cnt = cnt;
1387 }
1388 
1389 #ifdef CONFIG_PROC_FS
1390 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1391 {
1392 	const struct bpf_prog *prog = filp->private_data;
1393 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1394 	struct bpf_prog_stats stats;
1395 
1396 	bpf_prog_get_stats(prog, &stats);
1397 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1398 	seq_printf(m,
1399 		   "prog_type:\t%u\n"
1400 		   "prog_jited:\t%u\n"
1401 		   "prog_tag:\t%s\n"
1402 		   "memlock:\t%llu\n"
1403 		   "prog_id:\t%u\n"
1404 		   "run_time_ns:\t%llu\n"
1405 		   "run_cnt:\t%llu\n",
1406 		   prog->type,
1407 		   prog->jited,
1408 		   prog_tag,
1409 		   prog->pages * 1ULL << PAGE_SHIFT,
1410 		   prog->aux->id,
1411 		   stats.nsecs,
1412 		   stats.cnt);
1413 }
1414 #endif
1415 
1416 const struct file_operations bpf_prog_fops = {
1417 #ifdef CONFIG_PROC_FS
1418 	.show_fdinfo	= bpf_prog_show_fdinfo,
1419 #endif
1420 	.release	= bpf_prog_release,
1421 	.read		= bpf_dummy_read,
1422 	.write		= bpf_dummy_write,
1423 };
1424 
1425 int bpf_prog_new_fd(struct bpf_prog *prog)
1426 {
1427 	int ret;
1428 
1429 	ret = security_bpf_prog(prog);
1430 	if (ret < 0)
1431 		return ret;
1432 
1433 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1434 				O_RDWR | O_CLOEXEC);
1435 }
1436 
1437 static struct bpf_prog *____bpf_prog_get(struct fd f)
1438 {
1439 	if (!f.file)
1440 		return ERR_PTR(-EBADF);
1441 	if (f.file->f_op != &bpf_prog_fops) {
1442 		fdput(f);
1443 		return ERR_PTR(-EINVAL);
1444 	}
1445 
1446 	return f.file->private_data;
1447 }
1448 
1449 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1450 {
1451 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1452 		atomic_sub(i, &prog->aux->refcnt);
1453 		return ERR_PTR(-EBUSY);
1454 	}
1455 	return prog;
1456 }
1457 EXPORT_SYMBOL_GPL(bpf_prog_add);
1458 
1459 void bpf_prog_sub(struct bpf_prog *prog, int i)
1460 {
1461 	/* Only to be used for undoing previous bpf_prog_add() in some
1462 	 * error path. We still know that another entity in our call
1463 	 * path holds a reference to the program, thus atomic_sub() can
1464 	 * be safely used in such cases!
1465 	 */
1466 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1467 }
1468 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1469 
1470 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1471 {
1472 	return bpf_prog_add(prog, 1);
1473 }
1474 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1475 
1476 /* prog_idr_lock should have been held */
1477 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1478 {
1479 	int refold;
1480 
1481 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1482 
1483 	if (refold >= BPF_MAX_REFCNT) {
1484 		__bpf_prog_put(prog, false);
1485 		return ERR_PTR(-EBUSY);
1486 	}
1487 
1488 	if (!refold)
1489 		return ERR_PTR(-ENOENT);
1490 
1491 	return prog;
1492 }
1493 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1494 
1495 bool bpf_prog_get_ok(struct bpf_prog *prog,
1496 			    enum bpf_prog_type *attach_type, bool attach_drv)
1497 {
1498 	/* not an attachment, just a refcount inc, always allow */
1499 	if (!attach_type)
1500 		return true;
1501 
1502 	if (prog->type != *attach_type)
1503 		return false;
1504 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1505 		return false;
1506 
1507 	return true;
1508 }
1509 
1510 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1511 				       bool attach_drv)
1512 {
1513 	struct fd f = fdget(ufd);
1514 	struct bpf_prog *prog;
1515 
1516 	prog = ____bpf_prog_get(f);
1517 	if (IS_ERR(prog))
1518 		return prog;
1519 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1520 		prog = ERR_PTR(-EINVAL);
1521 		goto out;
1522 	}
1523 
1524 	prog = bpf_prog_inc(prog);
1525 out:
1526 	fdput(f);
1527 	return prog;
1528 }
1529 
1530 struct bpf_prog *bpf_prog_get(u32 ufd)
1531 {
1532 	return __bpf_prog_get(ufd, NULL, false);
1533 }
1534 
1535 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1536 				       bool attach_drv)
1537 {
1538 	return __bpf_prog_get(ufd, &type, attach_drv);
1539 }
1540 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1541 
1542 /* Initially all BPF programs could be loaded w/o specifying
1543  * expected_attach_type. Later for some of them specifying expected_attach_type
1544  * at load time became required so that program could be validated properly.
1545  * Programs of types that are allowed to be loaded both w/ and w/o (for
1546  * backward compatibility) expected_attach_type, should have the default attach
1547  * type assigned to expected_attach_type for the latter case, so that it can be
1548  * validated later at attach time.
1549  *
1550  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1551  * prog type requires it but has some attach types that have to be backward
1552  * compatible.
1553  */
1554 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1555 {
1556 	switch (attr->prog_type) {
1557 	case BPF_PROG_TYPE_CGROUP_SOCK:
1558 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1559 		 * exist so checking for non-zero is the way to go here.
1560 		 */
1561 		if (!attr->expected_attach_type)
1562 			attr->expected_attach_type =
1563 				BPF_CGROUP_INET_SOCK_CREATE;
1564 		break;
1565 	}
1566 }
1567 
1568 static int
1569 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1570 			   enum bpf_attach_type expected_attach_type,
1571 			   u32 btf_id)
1572 {
1573 	switch (prog_type) {
1574 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
1575 		if (btf_id > BTF_MAX_TYPE)
1576 			return -EINVAL;
1577 		break;
1578 	default:
1579 		if (btf_id)
1580 			return -EINVAL;
1581 		break;
1582 	}
1583 
1584 	switch (prog_type) {
1585 	case BPF_PROG_TYPE_CGROUP_SOCK:
1586 		switch (expected_attach_type) {
1587 		case BPF_CGROUP_INET_SOCK_CREATE:
1588 		case BPF_CGROUP_INET4_POST_BIND:
1589 		case BPF_CGROUP_INET6_POST_BIND:
1590 			return 0;
1591 		default:
1592 			return -EINVAL;
1593 		}
1594 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1595 		switch (expected_attach_type) {
1596 		case BPF_CGROUP_INET4_BIND:
1597 		case BPF_CGROUP_INET6_BIND:
1598 		case BPF_CGROUP_INET4_CONNECT:
1599 		case BPF_CGROUP_INET6_CONNECT:
1600 		case BPF_CGROUP_UDP4_SENDMSG:
1601 		case BPF_CGROUP_UDP6_SENDMSG:
1602 		case BPF_CGROUP_UDP4_RECVMSG:
1603 		case BPF_CGROUP_UDP6_RECVMSG:
1604 			return 0;
1605 		default:
1606 			return -EINVAL;
1607 		}
1608 	case BPF_PROG_TYPE_CGROUP_SKB:
1609 		switch (expected_attach_type) {
1610 		case BPF_CGROUP_INET_INGRESS:
1611 		case BPF_CGROUP_INET_EGRESS:
1612 			return 0;
1613 		default:
1614 			return -EINVAL;
1615 		}
1616 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1617 		switch (expected_attach_type) {
1618 		case BPF_CGROUP_SETSOCKOPT:
1619 		case BPF_CGROUP_GETSOCKOPT:
1620 			return 0;
1621 		default:
1622 			return -EINVAL;
1623 		}
1624 	default:
1625 		return 0;
1626 	}
1627 }
1628 
1629 /* last field in 'union bpf_attr' used by this command */
1630 #define	BPF_PROG_LOAD_LAST_FIELD attach_btf_id
1631 
1632 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1633 {
1634 	enum bpf_prog_type type = attr->prog_type;
1635 	struct bpf_prog *prog;
1636 	int err;
1637 	char license[128];
1638 	bool is_gpl;
1639 
1640 	if (CHECK_ATTR(BPF_PROG_LOAD))
1641 		return -EINVAL;
1642 
1643 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1644 				 BPF_F_ANY_ALIGNMENT |
1645 				 BPF_F_TEST_STATE_FREQ |
1646 				 BPF_F_TEST_RND_HI32))
1647 		return -EINVAL;
1648 
1649 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1650 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1651 	    !capable(CAP_SYS_ADMIN))
1652 		return -EPERM;
1653 
1654 	/* copy eBPF program license from user space */
1655 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1656 			      sizeof(license) - 1) < 0)
1657 		return -EFAULT;
1658 	license[sizeof(license) - 1] = 0;
1659 
1660 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1661 	is_gpl = license_is_gpl_compatible(license);
1662 
1663 	if (attr->insn_cnt == 0 ||
1664 	    attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1665 		return -E2BIG;
1666 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1667 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1668 	    !capable(CAP_SYS_ADMIN))
1669 		return -EPERM;
1670 
1671 	bpf_prog_load_fixup_attach_type(attr);
1672 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1673 				       attr->attach_btf_id))
1674 		return -EINVAL;
1675 
1676 	/* plain bpf_prog allocation */
1677 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1678 	if (!prog)
1679 		return -ENOMEM;
1680 
1681 	prog->expected_attach_type = attr->expected_attach_type;
1682 	prog->aux->attach_btf_id = attr->attach_btf_id;
1683 
1684 	prog->aux->offload_requested = !!attr->prog_ifindex;
1685 
1686 	err = security_bpf_prog_alloc(prog->aux);
1687 	if (err)
1688 		goto free_prog_nouncharge;
1689 
1690 	err = bpf_prog_charge_memlock(prog);
1691 	if (err)
1692 		goto free_prog_sec;
1693 
1694 	prog->len = attr->insn_cnt;
1695 
1696 	err = -EFAULT;
1697 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1698 			   bpf_prog_insn_size(prog)) != 0)
1699 		goto free_prog;
1700 
1701 	prog->orig_prog = NULL;
1702 	prog->jited = 0;
1703 
1704 	atomic_set(&prog->aux->refcnt, 1);
1705 	prog->gpl_compatible = is_gpl ? 1 : 0;
1706 
1707 	if (bpf_prog_is_dev_bound(prog->aux)) {
1708 		err = bpf_prog_offload_init(prog, attr);
1709 		if (err)
1710 			goto free_prog;
1711 	}
1712 
1713 	/* find program type: socket_filter vs tracing_filter */
1714 	err = find_prog_type(type, prog);
1715 	if (err < 0)
1716 		goto free_prog;
1717 
1718 	prog->aux->load_time = ktime_get_boottime_ns();
1719 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1720 	if (err)
1721 		goto free_prog;
1722 
1723 	/* run eBPF verifier */
1724 	err = bpf_check(&prog, attr, uattr);
1725 	if (err < 0)
1726 		goto free_used_maps;
1727 
1728 	prog = bpf_prog_select_runtime(prog, &err);
1729 	if (err < 0)
1730 		goto free_used_maps;
1731 
1732 	err = bpf_prog_alloc_id(prog);
1733 	if (err)
1734 		goto free_used_maps;
1735 
1736 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
1737 	 * effectively publicly exposed. However, retrieving via
1738 	 * bpf_prog_get_fd_by_id() will take another reference,
1739 	 * therefore it cannot be gone underneath us.
1740 	 *
1741 	 * Only for the time /after/ successful bpf_prog_new_fd()
1742 	 * and before returning to userspace, we might just hold
1743 	 * one reference and any parallel close on that fd could
1744 	 * rip everything out. Hence, below notifications must
1745 	 * happen before bpf_prog_new_fd().
1746 	 *
1747 	 * Also, any failure handling from this point onwards must
1748 	 * be using bpf_prog_put() given the program is exposed.
1749 	 */
1750 	bpf_prog_kallsyms_add(prog);
1751 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1752 
1753 	err = bpf_prog_new_fd(prog);
1754 	if (err < 0)
1755 		bpf_prog_put(prog);
1756 	return err;
1757 
1758 free_used_maps:
1759 	bpf_prog_free_linfo(prog);
1760 	kvfree(prog->aux->func_info);
1761 	btf_put(prog->aux->btf);
1762 	bpf_prog_kallsyms_del_subprogs(prog);
1763 	free_used_maps(prog->aux);
1764 free_prog:
1765 	bpf_prog_uncharge_memlock(prog);
1766 free_prog_sec:
1767 	security_bpf_prog_free(prog->aux);
1768 free_prog_nouncharge:
1769 	bpf_prog_free(prog);
1770 	return err;
1771 }
1772 
1773 #define BPF_OBJ_LAST_FIELD file_flags
1774 
1775 static int bpf_obj_pin(const union bpf_attr *attr)
1776 {
1777 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1778 		return -EINVAL;
1779 
1780 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1781 }
1782 
1783 static int bpf_obj_get(const union bpf_attr *attr)
1784 {
1785 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1786 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1787 		return -EINVAL;
1788 
1789 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1790 				attr->file_flags);
1791 }
1792 
1793 struct bpf_raw_tracepoint {
1794 	struct bpf_raw_event_map *btp;
1795 	struct bpf_prog *prog;
1796 };
1797 
1798 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1799 {
1800 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1801 
1802 	if (raw_tp->prog) {
1803 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1804 		bpf_prog_put(raw_tp->prog);
1805 	}
1806 	bpf_put_raw_tracepoint(raw_tp->btp);
1807 	kfree(raw_tp);
1808 	return 0;
1809 }
1810 
1811 static const struct file_operations bpf_raw_tp_fops = {
1812 	.release	= bpf_raw_tracepoint_release,
1813 	.read		= bpf_dummy_read,
1814 	.write		= bpf_dummy_write,
1815 };
1816 
1817 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1818 
1819 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1820 {
1821 	struct bpf_raw_tracepoint *raw_tp;
1822 	struct bpf_raw_event_map *btp;
1823 	struct bpf_prog *prog;
1824 	const char *tp_name;
1825 	char buf[128];
1826 	int tp_fd, err;
1827 
1828 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1829 		return -EINVAL;
1830 
1831 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1832 	if (IS_ERR(prog))
1833 		return PTR_ERR(prog);
1834 
1835 	if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1836 	    prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1837 		err = -EINVAL;
1838 		goto out_put_prog;
1839 	}
1840 
1841 	if (prog->type == BPF_PROG_TYPE_RAW_TRACEPOINT &&
1842 	    prog->aux->attach_btf_id) {
1843 		if (attr->raw_tracepoint.name) {
1844 			/* raw_tp name should not be specified in raw_tp
1845 			 * programs that were verified via in-kernel BTF info
1846 			 */
1847 			err = -EINVAL;
1848 			goto out_put_prog;
1849 		}
1850 		/* raw_tp name is taken from type name instead */
1851 		tp_name = prog->aux->attach_func_name;
1852 	} else {
1853 		if (strncpy_from_user(buf,
1854 				      u64_to_user_ptr(attr->raw_tracepoint.name),
1855 				      sizeof(buf) - 1) < 0) {
1856 			err = -EFAULT;
1857 			goto out_put_prog;
1858 		}
1859 		buf[sizeof(buf) - 1] = 0;
1860 		tp_name = buf;
1861 	}
1862 
1863 	btp = bpf_get_raw_tracepoint(tp_name);
1864 	if (!btp) {
1865 		err = -ENOENT;
1866 		goto out_put_prog;
1867 	}
1868 
1869 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1870 	if (!raw_tp) {
1871 		err = -ENOMEM;
1872 		goto out_put_btp;
1873 	}
1874 	raw_tp->btp = btp;
1875 	raw_tp->prog = prog;
1876 
1877 	err = bpf_probe_register(raw_tp->btp, prog);
1878 	if (err)
1879 		goto out_free_tp;
1880 
1881 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1882 				 O_CLOEXEC);
1883 	if (tp_fd < 0) {
1884 		bpf_probe_unregister(raw_tp->btp, prog);
1885 		err = tp_fd;
1886 		goto out_free_tp;
1887 	}
1888 	return tp_fd;
1889 
1890 out_free_tp:
1891 	kfree(raw_tp);
1892 out_put_btp:
1893 	bpf_put_raw_tracepoint(btp);
1894 out_put_prog:
1895 	bpf_prog_put(prog);
1896 	return err;
1897 }
1898 
1899 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1900 					     enum bpf_attach_type attach_type)
1901 {
1902 	switch (prog->type) {
1903 	case BPF_PROG_TYPE_CGROUP_SOCK:
1904 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1905 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1906 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1907 	case BPF_PROG_TYPE_CGROUP_SKB:
1908 		return prog->enforce_expected_attach_type &&
1909 			prog->expected_attach_type != attach_type ?
1910 			-EINVAL : 0;
1911 	default:
1912 		return 0;
1913 	}
1914 }
1915 
1916 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1917 
1918 #define BPF_F_ATTACH_MASK \
1919 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1920 
1921 static int bpf_prog_attach(const union bpf_attr *attr)
1922 {
1923 	enum bpf_prog_type ptype;
1924 	struct bpf_prog *prog;
1925 	int ret;
1926 
1927 	if (!capable(CAP_NET_ADMIN))
1928 		return -EPERM;
1929 
1930 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1931 		return -EINVAL;
1932 
1933 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1934 		return -EINVAL;
1935 
1936 	switch (attr->attach_type) {
1937 	case BPF_CGROUP_INET_INGRESS:
1938 	case BPF_CGROUP_INET_EGRESS:
1939 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1940 		break;
1941 	case BPF_CGROUP_INET_SOCK_CREATE:
1942 	case BPF_CGROUP_INET4_POST_BIND:
1943 	case BPF_CGROUP_INET6_POST_BIND:
1944 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1945 		break;
1946 	case BPF_CGROUP_INET4_BIND:
1947 	case BPF_CGROUP_INET6_BIND:
1948 	case BPF_CGROUP_INET4_CONNECT:
1949 	case BPF_CGROUP_INET6_CONNECT:
1950 	case BPF_CGROUP_UDP4_SENDMSG:
1951 	case BPF_CGROUP_UDP6_SENDMSG:
1952 	case BPF_CGROUP_UDP4_RECVMSG:
1953 	case BPF_CGROUP_UDP6_RECVMSG:
1954 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1955 		break;
1956 	case BPF_CGROUP_SOCK_OPS:
1957 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1958 		break;
1959 	case BPF_CGROUP_DEVICE:
1960 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1961 		break;
1962 	case BPF_SK_MSG_VERDICT:
1963 		ptype = BPF_PROG_TYPE_SK_MSG;
1964 		break;
1965 	case BPF_SK_SKB_STREAM_PARSER:
1966 	case BPF_SK_SKB_STREAM_VERDICT:
1967 		ptype = BPF_PROG_TYPE_SK_SKB;
1968 		break;
1969 	case BPF_LIRC_MODE2:
1970 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1971 		break;
1972 	case BPF_FLOW_DISSECTOR:
1973 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1974 		break;
1975 	case BPF_CGROUP_SYSCTL:
1976 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1977 		break;
1978 	case BPF_CGROUP_GETSOCKOPT:
1979 	case BPF_CGROUP_SETSOCKOPT:
1980 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1981 		break;
1982 	default:
1983 		return -EINVAL;
1984 	}
1985 
1986 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1987 	if (IS_ERR(prog))
1988 		return PTR_ERR(prog);
1989 
1990 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1991 		bpf_prog_put(prog);
1992 		return -EINVAL;
1993 	}
1994 
1995 	switch (ptype) {
1996 	case BPF_PROG_TYPE_SK_SKB:
1997 	case BPF_PROG_TYPE_SK_MSG:
1998 		ret = sock_map_get_from_fd(attr, prog);
1999 		break;
2000 	case BPF_PROG_TYPE_LIRC_MODE2:
2001 		ret = lirc_prog_attach(attr, prog);
2002 		break;
2003 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2004 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2005 		break;
2006 	default:
2007 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2008 	}
2009 
2010 	if (ret)
2011 		bpf_prog_put(prog);
2012 	return ret;
2013 }
2014 
2015 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2016 
2017 static int bpf_prog_detach(const union bpf_attr *attr)
2018 {
2019 	enum bpf_prog_type ptype;
2020 
2021 	if (!capable(CAP_NET_ADMIN))
2022 		return -EPERM;
2023 
2024 	if (CHECK_ATTR(BPF_PROG_DETACH))
2025 		return -EINVAL;
2026 
2027 	switch (attr->attach_type) {
2028 	case BPF_CGROUP_INET_INGRESS:
2029 	case BPF_CGROUP_INET_EGRESS:
2030 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2031 		break;
2032 	case BPF_CGROUP_INET_SOCK_CREATE:
2033 	case BPF_CGROUP_INET4_POST_BIND:
2034 	case BPF_CGROUP_INET6_POST_BIND:
2035 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2036 		break;
2037 	case BPF_CGROUP_INET4_BIND:
2038 	case BPF_CGROUP_INET6_BIND:
2039 	case BPF_CGROUP_INET4_CONNECT:
2040 	case BPF_CGROUP_INET6_CONNECT:
2041 	case BPF_CGROUP_UDP4_SENDMSG:
2042 	case BPF_CGROUP_UDP6_SENDMSG:
2043 	case BPF_CGROUP_UDP4_RECVMSG:
2044 	case BPF_CGROUP_UDP6_RECVMSG:
2045 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2046 		break;
2047 	case BPF_CGROUP_SOCK_OPS:
2048 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2049 		break;
2050 	case BPF_CGROUP_DEVICE:
2051 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2052 		break;
2053 	case BPF_SK_MSG_VERDICT:
2054 		return sock_map_get_from_fd(attr, NULL);
2055 	case BPF_SK_SKB_STREAM_PARSER:
2056 	case BPF_SK_SKB_STREAM_VERDICT:
2057 		return sock_map_get_from_fd(attr, NULL);
2058 	case BPF_LIRC_MODE2:
2059 		return lirc_prog_detach(attr);
2060 	case BPF_FLOW_DISSECTOR:
2061 		return skb_flow_dissector_bpf_prog_detach(attr);
2062 	case BPF_CGROUP_SYSCTL:
2063 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2064 		break;
2065 	case BPF_CGROUP_GETSOCKOPT:
2066 	case BPF_CGROUP_SETSOCKOPT:
2067 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2068 		break;
2069 	default:
2070 		return -EINVAL;
2071 	}
2072 
2073 	return cgroup_bpf_prog_detach(attr, ptype);
2074 }
2075 
2076 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2077 
2078 static int bpf_prog_query(const union bpf_attr *attr,
2079 			  union bpf_attr __user *uattr)
2080 {
2081 	if (!capable(CAP_NET_ADMIN))
2082 		return -EPERM;
2083 	if (CHECK_ATTR(BPF_PROG_QUERY))
2084 		return -EINVAL;
2085 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2086 		return -EINVAL;
2087 
2088 	switch (attr->query.attach_type) {
2089 	case BPF_CGROUP_INET_INGRESS:
2090 	case BPF_CGROUP_INET_EGRESS:
2091 	case BPF_CGROUP_INET_SOCK_CREATE:
2092 	case BPF_CGROUP_INET4_BIND:
2093 	case BPF_CGROUP_INET6_BIND:
2094 	case BPF_CGROUP_INET4_POST_BIND:
2095 	case BPF_CGROUP_INET6_POST_BIND:
2096 	case BPF_CGROUP_INET4_CONNECT:
2097 	case BPF_CGROUP_INET6_CONNECT:
2098 	case BPF_CGROUP_UDP4_SENDMSG:
2099 	case BPF_CGROUP_UDP6_SENDMSG:
2100 	case BPF_CGROUP_UDP4_RECVMSG:
2101 	case BPF_CGROUP_UDP6_RECVMSG:
2102 	case BPF_CGROUP_SOCK_OPS:
2103 	case BPF_CGROUP_DEVICE:
2104 	case BPF_CGROUP_SYSCTL:
2105 	case BPF_CGROUP_GETSOCKOPT:
2106 	case BPF_CGROUP_SETSOCKOPT:
2107 		break;
2108 	case BPF_LIRC_MODE2:
2109 		return lirc_prog_query(attr, uattr);
2110 	case BPF_FLOW_DISSECTOR:
2111 		return skb_flow_dissector_prog_query(attr, uattr);
2112 	default:
2113 		return -EINVAL;
2114 	}
2115 
2116 	return cgroup_bpf_prog_query(attr, uattr);
2117 }
2118 
2119 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2120 
2121 static int bpf_prog_test_run(const union bpf_attr *attr,
2122 			     union bpf_attr __user *uattr)
2123 {
2124 	struct bpf_prog *prog;
2125 	int ret = -ENOTSUPP;
2126 
2127 	if (!capable(CAP_SYS_ADMIN))
2128 		return -EPERM;
2129 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2130 		return -EINVAL;
2131 
2132 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2133 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
2134 		return -EINVAL;
2135 
2136 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2137 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
2138 		return -EINVAL;
2139 
2140 	prog = bpf_prog_get(attr->test.prog_fd);
2141 	if (IS_ERR(prog))
2142 		return PTR_ERR(prog);
2143 
2144 	if (prog->aux->ops->test_run)
2145 		ret = prog->aux->ops->test_run(prog, attr, uattr);
2146 
2147 	bpf_prog_put(prog);
2148 	return ret;
2149 }
2150 
2151 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2152 
2153 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2154 			       union bpf_attr __user *uattr,
2155 			       struct idr *idr,
2156 			       spinlock_t *lock)
2157 {
2158 	u32 next_id = attr->start_id;
2159 	int err = 0;
2160 
2161 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2162 		return -EINVAL;
2163 
2164 	if (!capable(CAP_SYS_ADMIN))
2165 		return -EPERM;
2166 
2167 	next_id++;
2168 	spin_lock_bh(lock);
2169 	if (!idr_get_next(idr, &next_id))
2170 		err = -ENOENT;
2171 	spin_unlock_bh(lock);
2172 
2173 	if (!err)
2174 		err = put_user(next_id, &uattr->next_id);
2175 
2176 	return err;
2177 }
2178 
2179 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2180 
2181 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2182 {
2183 	struct bpf_prog *prog;
2184 	u32 id = attr->prog_id;
2185 	int fd;
2186 
2187 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2188 		return -EINVAL;
2189 
2190 	if (!capable(CAP_SYS_ADMIN))
2191 		return -EPERM;
2192 
2193 	spin_lock_bh(&prog_idr_lock);
2194 	prog = idr_find(&prog_idr, id);
2195 	if (prog)
2196 		prog = bpf_prog_inc_not_zero(prog);
2197 	else
2198 		prog = ERR_PTR(-ENOENT);
2199 	spin_unlock_bh(&prog_idr_lock);
2200 
2201 	if (IS_ERR(prog))
2202 		return PTR_ERR(prog);
2203 
2204 	fd = bpf_prog_new_fd(prog);
2205 	if (fd < 0)
2206 		bpf_prog_put(prog);
2207 
2208 	return fd;
2209 }
2210 
2211 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2212 
2213 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2214 {
2215 	struct bpf_map *map;
2216 	u32 id = attr->map_id;
2217 	int f_flags;
2218 	int fd;
2219 
2220 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2221 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2222 		return -EINVAL;
2223 
2224 	if (!capable(CAP_SYS_ADMIN))
2225 		return -EPERM;
2226 
2227 	f_flags = bpf_get_file_flag(attr->open_flags);
2228 	if (f_flags < 0)
2229 		return f_flags;
2230 
2231 	spin_lock_bh(&map_idr_lock);
2232 	map = idr_find(&map_idr, id);
2233 	if (map)
2234 		map = __bpf_map_inc_not_zero(map, true);
2235 	else
2236 		map = ERR_PTR(-ENOENT);
2237 	spin_unlock_bh(&map_idr_lock);
2238 
2239 	if (IS_ERR(map))
2240 		return PTR_ERR(map);
2241 
2242 	fd = bpf_map_new_fd(map, f_flags);
2243 	if (fd < 0)
2244 		bpf_map_put_with_uref(map);
2245 
2246 	return fd;
2247 }
2248 
2249 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2250 					      unsigned long addr, u32 *off,
2251 					      u32 *type)
2252 {
2253 	const struct bpf_map *map;
2254 	int i;
2255 
2256 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2257 		map = prog->aux->used_maps[i];
2258 		if (map == (void *)addr) {
2259 			*type = BPF_PSEUDO_MAP_FD;
2260 			return map;
2261 		}
2262 		if (!map->ops->map_direct_value_meta)
2263 			continue;
2264 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
2265 			*type = BPF_PSEUDO_MAP_VALUE;
2266 			return map;
2267 		}
2268 	}
2269 
2270 	return NULL;
2271 }
2272 
2273 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2274 {
2275 	const struct bpf_map *map;
2276 	struct bpf_insn *insns;
2277 	u32 off, type;
2278 	u64 imm;
2279 	int i;
2280 
2281 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2282 			GFP_USER);
2283 	if (!insns)
2284 		return insns;
2285 
2286 	for (i = 0; i < prog->len; i++) {
2287 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2288 			insns[i].code = BPF_JMP | BPF_CALL;
2289 			insns[i].imm = BPF_FUNC_tail_call;
2290 			/* fall-through */
2291 		}
2292 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2293 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2294 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2295 				insns[i].code = BPF_JMP | BPF_CALL;
2296 			if (!bpf_dump_raw_ok())
2297 				insns[i].imm = 0;
2298 			continue;
2299 		}
2300 
2301 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2302 			continue;
2303 
2304 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2305 		map = bpf_map_from_imm(prog, imm, &off, &type);
2306 		if (map) {
2307 			insns[i].src_reg = type;
2308 			insns[i].imm = map->id;
2309 			insns[i + 1].imm = off;
2310 			continue;
2311 		}
2312 	}
2313 
2314 	return insns;
2315 }
2316 
2317 static int set_info_rec_size(struct bpf_prog_info *info)
2318 {
2319 	/*
2320 	 * Ensure info.*_rec_size is the same as kernel expected size
2321 	 *
2322 	 * or
2323 	 *
2324 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2325 	 * zero.  In this case, the kernel will set the expected
2326 	 * _rec_size back to the info.
2327 	 */
2328 
2329 	if ((info->nr_func_info || info->func_info_rec_size) &&
2330 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2331 		return -EINVAL;
2332 
2333 	if ((info->nr_line_info || info->line_info_rec_size) &&
2334 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2335 		return -EINVAL;
2336 
2337 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2338 	    info->jited_line_info_rec_size != sizeof(__u64))
2339 		return -EINVAL;
2340 
2341 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2342 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2343 	info->jited_line_info_rec_size = sizeof(__u64);
2344 
2345 	return 0;
2346 }
2347 
2348 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2349 				   const union bpf_attr *attr,
2350 				   union bpf_attr __user *uattr)
2351 {
2352 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2353 	struct bpf_prog_info info = {};
2354 	u32 info_len = attr->info.info_len;
2355 	struct bpf_prog_stats stats;
2356 	char __user *uinsns;
2357 	u32 ulen;
2358 	int err;
2359 
2360 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2361 	if (err)
2362 		return err;
2363 	info_len = min_t(u32, sizeof(info), info_len);
2364 
2365 	if (copy_from_user(&info, uinfo, info_len))
2366 		return -EFAULT;
2367 
2368 	info.type = prog->type;
2369 	info.id = prog->aux->id;
2370 	info.load_time = prog->aux->load_time;
2371 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2372 					       prog->aux->user->uid);
2373 	info.gpl_compatible = prog->gpl_compatible;
2374 
2375 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2376 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2377 
2378 	ulen = info.nr_map_ids;
2379 	info.nr_map_ids = prog->aux->used_map_cnt;
2380 	ulen = min_t(u32, info.nr_map_ids, ulen);
2381 	if (ulen) {
2382 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2383 		u32 i;
2384 
2385 		for (i = 0; i < ulen; i++)
2386 			if (put_user(prog->aux->used_maps[i]->id,
2387 				     &user_map_ids[i]))
2388 				return -EFAULT;
2389 	}
2390 
2391 	err = set_info_rec_size(&info);
2392 	if (err)
2393 		return err;
2394 
2395 	bpf_prog_get_stats(prog, &stats);
2396 	info.run_time_ns = stats.nsecs;
2397 	info.run_cnt = stats.cnt;
2398 
2399 	if (!capable(CAP_SYS_ADMIN)) {
2400 		info.jited_prog_len = 0;
2401 		info.xlated_prog_len = 0;
2402 		info.nr_jited_ksyms = 0;
2403 		info.nr_jited_func_lens = 0;
2404 		info.nr_func_info = 0;
2405 		info.nr_line_info = 0;
2406 		info.nr_jited_line_info = 0;
2407 		goto done;
2408 	}
2409 
2410 	ulen = info.xlated_prog_len;
2411 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2412 	if (info.xlated_prog_len && ulen) {
2413 		struct bpf_insn *insns_sanitized;
2414 		bool fault;
2415 
2416 		if (prog->blinded && !bpf_dump_raw_ok()) {
2417 			info.xlated_prog_insns = 0;
2418 			goto done;
2419 		}
2420 		insns_sanitized = bpf_insn_prepare_dump(prog);
2421 		if (!insns_sanitized)
2422 			return -ENOMEM;
2423 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2424 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2425 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2426 		kfree(insns_sanitized);
2427 		if (fault)
2428 			return -EFAULT;
2429 	}
2430 
2431 	if (bpf_prog_is_dev_bound(prog->aux)) {
2432 		err = bpf_prog_offload_info_fill(&info, prog);
2433 		if (err)
2434 			return err;
2435 		goto done;
2436 	}
2437 
2438 	/* NOTE: the following code is supposed to be skipped for offload.
2439 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2440 	 * for offload.
2441 	 */
2442 	ulen = info.jited_prog_len;
2443 	if (prog->aux->func_cnt) {
2444 		u32 i;
2445 
2446 		info.jited_prog_len = 0;
2447 		for (i = 0; i < prog->aux->func_cnt; i++)
2448 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2449 	} else {
2450 		info.jited_prog_len = prog->jited_len;
2451 	}
2452 
2453 	if (info.jited_prog_len && ulen) {
2454 		if (bpf_dump_raw_ok()) {
2455 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2456 			ulen = min_t(u32, info.jited_prog_len, ulen);
2457 
2458 			/* for multi-function programs, copy the JITed
2459 			 * instructions for all the functions
2460 			 */
2461 			if (prog->aux->func_cnt) {
2462 				u32 len, free, i;
2463 				u8 *img;
2464 
2465 				free = ulen;
2466 				for (i = 0; i < prog->aux->func_cnt; i++) {
2467 					len = prog->aux->func[i]->jited_len;
2468 					len = min_t(u32, len, free);
2469 					img = (u8 *) prog->aux->func[i]->bpf_func;
2470 					if (copy_to_user(uinsns, img, len))
2471 						return -EFAULT;
2472 					uinsns += len;
2473 					free -= len;
2474 					if (!free)
2475 						break;
2476 				}
2477 			} else {
2478 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2479 					return -EFAULT;
2480 			}
2481 		} else {
2482 			info.jited_prog_insns = 0;
2483 		}
2484 	}
2485 
2486 	ulen = info.nr_jited_ksyms;
2487 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2488 	if (ulen) {
2489 		if (bpf_dump_raw_ok()) {
2490 			unsigned long ksym_addr;
2491 			u64 __user *user_ksyms;
2492 			u32 i;
2493 
2494 			/* copy the address of the kernel symbol
2495 			 * corresponding to each function
2496 			 */
2497 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2498 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2499 			if (prog->aux->func_cnt) {
2500 				for (i = 0; i < ulen; i++) {
2501 					ksym_addr = (unsigned long)
2502 						prog->aux->func[i]->bpf_func;
2503 					if (put_user((u64) ksym_addr,
2504 						     &user_ksyms[i]))
2505 						return -EFAULT;
2506 				}
2507 			} else {
2508 				ksym_addr = (unsigned long) prog->bpf_func;
2509 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2510 					return -EFAULT;
2511 			}
2512 		} else {
2513 			info.jited_ksyms = 0;
2514 		}
2515 	}
2516 
2517 	ulen = info.nr_jited_func_lens;
2518 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2519 	if (ulen) {
2520 		if (bpf_dump_raw_ok()) {
2521 			u32 __user *user_lens;
2522 			u32 func_len, i;
2523 
2524 			/* copy the JITed image lengths for each function */
2525 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2526 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2527 			if (prog->aux->func_cnt) {
2528 				for (i = 0; i < ulen; i++) {
2529 					func_len =
2530 						prog->aux->func[i]->jited_len;
2531 					if (put_user(func_len, &user_lens[i]))
2532 						return -EFAULT;
2533 				}
2534 			} else {
2535 				func_len = prog->jited_len;
2536 				if (put_user(func_len, &user_lens[0]))
2537 					return -EFAULT;
2538 			}
2539 		} else {
2540 			info.jited_func_lens = 0;
2541 		}
2542 	}
2543 
2544 	if (prog->aux->btf)
2545 		info.btf_id = btf_id(prog->aux->btf);
2546 
2547 	ulen = info.nr_func_info;
2548 	info.nr_func_info = prog->aux->func_info_cnt;
2549 	if (info.nr_func_info && ulen) {
2550 		char __user *user_finfo;
2551 
2552 		user_finfo = u64_to_user_ptr(info.func_info);
2553 		ulen = min_t(u32, info.nr_func_info, ulen);
2554 		if (copy_to_user(user_finfo, prog->aux->func_info,
2555 				 info.func_info_rec_size * ulen))
2556 			return -EFAULT;
2557 	}
2558 
2559 	ulen = info.nr_line_info;
2560 	info.nr_line_info = prog->aux->nr_linfo;
2561 	if (info.nr_line_info && ulen) {
2562 		__u8 __user *user_linfo;
2563 
2564 		user_linfo = u64_to_user_ptr(info.line_info);
2565 		ulen = min_t(u32, info.nr_line_info, ulen);
2566 		if (copy_to_user(user_linfo, prog->aux->linfo,
2567 				 info.line_info_rec_size * ulen))
2568 			return -EFAULT;
2569 	}
2570 
2571 	ulen = info.nr_jited_line_info;
2572 	if (prog->aux->jited_linfo)
2573 		info.nr_jited_line_info = prog->aux->nr_linfo;
2574 	else
2575 		info.nr_jited_line_info = 0;
2576 	if (info.nr_jited_line_info && ulen) {
2577 		if (bpf_dump_raw_ok()) {
2578 			__u64 __user *user_linfo;
2579 			u32 i;
2580 
2581 			user_linfo = u64_to_user_ptr(info.jited_line_info);
2582 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
2583 			for (i = 0; i < ulen; i++) {
2584 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2585 					     &user_linfo[i]))
2586 					return -EFAULT;
2587 			}
2588 		} else {
2589 			info.jited_line_info = 0;
2590 		}
2591 	}
2592 
2593 	ulen = info.nr_prog_tags;
2594 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2595 	if (ulen) {
2596 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2597 		u32 i;
2598 
2599 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
2600 		ulen = min_t(u32, info.nr_prog_tags, ulen);
2601 		if (prog->aux->func_cnt) {
2602 			for (i = 0; i < ulen; i++) {
2603 				if (copy_to_user(user_prog_tags[i],
2604 						 prog->aux->func[i]->tag,
2605 						 BPF_TAG_SIZE))
2606 					return -EFAULT;
2607 			}
2608 		} else {
2609 			if (copy_to_user(user_prog_tags[0],
2610 					 prog->tag, BPF_TAG_SIZE))
2611 				return -EFAULT;
2612 		}
2613 	}
2614 
2615 done:
2616 	if (copy_to_user(uinfo, &info, info_len) ||
2617 	    put_user(info_len, &uattr->info.info_len))
2618 		return -EFAULT;
2619 
2620 	return 0;
2621 }
2622 
2623 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2624 				  const union bpf_attr *attr,
2625 				  union bpf_attr __user *uattr)
2626 {
2627 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2628 	struct bpf_map_info info = {};
2629 	u32 info_len = attr->info.info_len;
2630 	int err;
2631 
2632 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2633 	if (err)
2634 		return err;
2635 	info_len = min_t(u32, sizeof(info), info_len);
2636 
2637 	info.type = map->map_type;
2638 	info.id = map->id;
2639 	info.key_size = map->key_size;
2640 	info.value_size = map->value_size;
2641 	info.max_entries = map->max_entries;
2642 	info.map_flags = map->map_flags;
2643 	memcpy(info.name, map->name, sizeof(map->name));
2644 
2645 	if (map->btf) {
2646 		info.btf_id = btf_id(map->btf);
2647 		info.btf_key_type_id = map->btf_key_type_id;
2648 		info.btf_value_type_id = map->btf_value_type_id;
2649 	}
2650 
2651 	if (bpf_map_is_dev_bound(map)) {
2652 		err = bpf_map_offload_info_fill(&info, map);
2653 		if (err)
2654 			return err;
2655 	}
2656 
2657 	if (copy_to_user(uinfo, &info, info_len) ||
2658 	    put_user(info_len, &uattr->info.info_len))
2659 		return -EFAULT;
2660 
2661 	return 0;
2662 }
2663 
2664 static int bpf_btf_get_info_by_fd(struct btf *btf,
2665 				  const union bpf_attr *attr,
2666 				  union bpf_attr __user *uattr)
2667 {
2668 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2669 	u32 info_len = attr->info.info_len;
2670 	int err;
2671 
2672 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2673 	if (err)
2674 		return err;
2675 
2676 	return btf_get_info_by_fd(btf, attr, uattr);
2677 }
2678 
2679 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2680 
2681 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2682 				  union bpf_attr __user *uattr)
2683 {
2684 	int ufd = attr->info.bpf_fd;
2685 	struct fd f;
2686 	int err;
2687 
2688 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2689 		return -EINVAL;
2690 
2691 	f = fdget(ufd);
2692 	if (!f.file)
2693 		return -EBADFD;
2694 
2695 	if (f.file->f_op == &bpf_prog_fops)
2696 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2697 					      uattr);
2698 	else if (f.file->f_op == &bpf_map_fops)
2699 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2700 					     uattr);
2701 	else if (f.file->f_op == &btf_fops)
2702 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2703 	else
2704 		err = -EINVAL;
2705 
2706 	fdput(f);
2707 	return err;
2708 }
2709 
2710 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2711 
2712 static int bpf_btf_load(const union bpf_attr *attr)
2713 {
2714 	if (CHECK_ATTR(BPF_BTF_LOAD))
2715 		return -EINVAL;
2716 
2717 	if (!capable(CAP_SYS_ADMIN))
2718 		return -EPERM;
2719 
2720 	return btf_new_fd(attr);
2721 }
2722 
2723 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2724 
2725 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2726 {
2727 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2728 		return -EINVAL;
2729 
2730 	if (!capable(CAP_SYS_ADMIN))
2731 		return -EPERM;
2732 
2733 	return btf_get_fd_by_id(attr->btf_id);
2734 }
2735 
2736 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2737 				    union bpf_attr __user *uattr,
2738 				    u32 prog_id, u32 fd_type,
2739 				    const char *buf, u64 probe_offset,
2740 				    u64 probe_addr)
2741 {
2742 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2743 	u32 len = buf ? strlen(buf) : 0, input_len;
2744 	int err = 0;
2745 
2746 	if (put_user(len, &uattr->task_fd_query.buf_len))
2747 		return -EFAULT;
2748 	input_len = attr->task_fd_query.buf_len;
2749 	if (input_len && ubuf) {
2750 		if (!len) {
2751 			/* nothing to copy, just make ubuf NULL terminated */
2752 			char zero = '\0';
2753 
2754 			if (put_user(zero, ubuf))
2755 				return -EFAULT;
2756 		} else if (input_len >= len + 1) {
2757 			/* ubuf can hold the string with NULL terminator */
2758 			if (copy_to_user(ubuf, buf, len + 1))
2759 				return -EFAULT;
2760 		} else {
2761 			/* ubuf cannot hold the string with NULL terminator,
2762 			 * do a partial copy with NULL terminator.
2763 			 */
2764 			char zero = '\0';
2765 
2766 			err = -ENOSPC;
2767 			if (copy_to_user(ubuf, buf, input_len - 1))
2768 				return -EFAULT;
2769 			if (put_user(zero, ubuf + input_len - 1))
2770 				return -EFAULT;
2771 		}
2772 	}
2773 
2774 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2775 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2776 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2777 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2778 		return -EFAULT;
2779 
2780 	return err;
2781 }
2782 
2783 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2784 
2785 static int bpf_task_fd_query(const union bpf_attr *attr,
2786 			     union bpf_attr __user *uattr)
2787 {
2788 	pid_t pid = attr->task_fd_query.pid;
2789 	u32 fd = attr->task_fd_query.fd;
2790 	const struct perf_event *event;
2791 	struct files_struct *files;
2792 	struct task_struct *task;
2793 	struct file *file;
2794 	int err;
2795 
2796 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2797 		return -EINVAL;
2798 
2799 	if (!capable(CAP_SYS_ADMIN))
2800 		return -EPERM;
2801 
2802 	if (attr->task_fd_query.flags != 0)
2803 		return -EINVAL;
2804 
2805 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2806 	if (!task)
2807 		return -ENOENT;
2808 
2809 	files = get_files_struct(task);
2810 	put_task_struct(task);
2811 	if (!files)
2812 		return -ENOENT;
2813 
2814 	err = 0;
2815 	spin_lock(&files->file_lock);
2816 	file = fcheck_files(files, fd);
2817 	if (!file)
2818 		err = -EBADF;
2819 	else
2820 		get_file(file);
2821 	spin_unlock(&files->file_lock);
2822 	put_files_struct(files);
2823 
2824 	if (err)
2825 		goto out;
2826 
2827 	if (file->f_op == &bpf_raw_tp_fops) {
2828 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2829 		struct bpf_raw_event_map *btp = raw_tp->btp;
2830 
2831 		err = bpf_task_fd_query_copy(attr, uattr,
2832 					     raw_tp->prog->aux->id,
2833 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2834 					     btp->tp->name, 0, 0);
2835 		goto put_file;
2836 	}
2837 
2838 	event = perf_get_event(file);
2839 	if (!IS_ERR(event)) {
2840 		u64 probe_offset, probe_addr;
2841 		u32 prog_id, fd_type;
2842 		const char *buf;
2843 
2844 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2845 					      &buf, &probe_offset,
2846 					      &probe_addr);
2847 		if (!err)
2848 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2849 						     fd_type, buf,
2850 						     probe_offset,
2851 						     probe_addr);
2852 		goto put_file;
2853 	}
2854 
2855 	err = -ENOTSUPP;
2856 put_file:
2857 	fput(file);
2858 out:
2859 	return err;
2860 }
2861 
2862 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2863 {
2864 	union bpf_attr attr = {};
2865 	int err;
2866 
2867 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2868 		return -EPERM;
2869 
2870 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2871 	if (err)
2872 		return err;
2873 	size = min_t(u32, size, sizeof(attr));
2874 
2875 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2876 	if (copy_from_user(&attr, uattr, size) != 0)
2877 		return -EFAULT;
2878 
2879 	err = security_bpf(cmd, &attr, size);
2880 	if (err < 0)
2881 		return err;
2882 
2883 	switch (cmd) {
2884 	case BPF_MAP_CREATE:
2885 		err = map_create(&attr);
2886 		break;
2887 	case BPF_MAP_LOOKUP_ELEM:
2888 		err = map_lookup_elem(&attr);
2889 		break;
2890 	case BPF_MAP_UPDATE_ELEM:
2891 		err = map_update_elem(&attr);
2892 		break;
2893 	case BPF_MAP_DELETE_ELEM:
2894 		err = map_delete_elem(&attr);
2895 		break;
2896 	case BPF_MAP_GET_NEXT_KEY:
2897 		err = map_get_next_key(&attr);
2898 		break;
2899 	case BPF_MAP_FREEZE:
2900 		err = map_freeze(&attr);
2901 		break;
2902 	case BPF_PROG_LOAD:
2903 		err = bpf_prog_load(&attr, uattr);
2904 		break;
2905 	case BPF_OBJ_PIN:
2906 		err = bpf_obj_pin(&attr);
2907 		break;
2908 	case BPF_OBJ_GET:
2909 		err = bpf_obj_get(&attr);
2910 		break;
2911 	case BPF_PROG_ATTACH:
2912 		err = bpf_prog_attach(&attr);
2913 		break;
2914 	case BPF_PROG_DETACH:
2915 		err = bpf_prog_detach(&attr);
2916 		break;
2917 	case BPF_PROG_QUERY:
2918 		err = bpf_prog_query(&attr, uattr);
2919 		break;
2920 	case BPF_PROG_TEST_RUN:
2921 		err = bpf_prog_test_run(&attr, uattr);
2922 		break;
2923 	case BPF_PROG_GET_NEXT_ID:
2924 		err = bpf_obj_get_next_id(&attr, uattr,
2925 					  &prog_idr, &prog_idr_lock);
2926 		break;
2927 	case BPF_MAP_GET_NEXT_ID:
2928 		err = bpf_obj_get_next_id(&attr, uattr,
2929 					  &map_idr, &map_idr_lock);
2930 		break;
2931 	case BPF_BTF_GET_NEXT_ID:
2932 		err = bpf_obj_get_next_id(&attr, uattr,
2933 					  &btf_idr, &btf_idr_lock);
2934 		break;
2935 	case BPF_PROG_GET_FD_BY_ID:
2936 		err = bpf_prog_get_fd_by_id(&attr);
2937 		break;
2938 	case BPF_MAP_GET_FD_BY_ID:
2939 		err = bpf_map_get_fd_by_id(&attr);
2940 		break;
2941 	case BPF_OBJ_GET_INFO_BY_FD:
2942 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2943 		break;
2944 	case BPF_RAW_TRACEPOINT_OPEN:
2945 		err = bpf_raw_tracepoint_open(&attr);
2946 		break;
2947 	case BPF_BTF_LOAD:
2948 		err = bpf_btf_load(&attr);
2949 		break;
2950 	case BPF_BTF_GET_FD_BY_ID:
2951 		err = bpf_btf_get_fd_by_id(&attr);
2952 		break;
2953 	case BPF_TASK_FD_QUERY:
2954 		err = bpf_task_fd_query(&attr, uattr);
2955 		break;
2956 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2957 		err = map_lookup_and_delete_elem(&attr);
2958 		break;
2959 	default:
2960 		err = -EINVAL;
2961 		break;
2962 	}
2963 
2964 	return err;
2965 }
2966