xref: /linux/kernel/bpf/syscall.c (revision 4d66c56f7efe122d09d06cd3ebfa52a43d51a9cb)
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(u64 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 >= SIZE_MAX)
146 		return NULL;
147 
148 	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
149 		area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
150 				    numa_node);
151 		if (area != NULL)
152 			return area;
153 	}
154 
155 	return __vmalloc_node_flags_caller(size, numa_node,
156 					   GFP_KERNEL | __GFP_RETRY_MAYFAIL |
157 					   flags, __builtin_return_address(0));
158 }
159 
160 void bpf_map_area_free(void *area)
161 {
162 	kvfree(area);
163 }
164 
165 static u32 bpf_map_flags_retain_permanent(u32 flags)
166 {
167 	/* Some map creation flags are not tied to the map object but
168 	 * rather to the map fd instead, so they have no meaning upon
169 	 * map object inspection since multiple file descriptors with
170 	 * different (access) properties can exist here. Thus, given
171 	 * this has zero meaning for the map itself, lets clear these
172 	 * from here.
173 	 */
174 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
175 }
176 
177 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
178 {
179 	map->map_type = attr->map_type;
180 	map->key_size = attr->key_size;
181 	map->value_size = attr->value_size;
182 	map->max_entries = attr->max_entries;
183 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
184 	map->numa_node = bpf_map_attr_numa_node(attr);
185 }
186 
187 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
188 {
189 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
190 
191 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
192 		atomic_long_sub(pages, &user->locked_vm);
193 		return -EPERM;
194 	}
195 	return 0;
196 }
197 
198 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
199 {
200 	if (user)
201 		atomic_long_sub(pages, &user->locked_vm);
202 }
203 
204 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
205 {
206 	u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
207 	struct user_struct *user;
208 	int ret;
209 
210 	if (size >= U32_MAX - PAGE_SIZE)
211 		return -E2BIG;
212 
213 	user = get_current_user();
214 	ret = bpf_charge_memlock(user, pages);
215 	if (ret) {
216 		free_uid(user);
217 		return ret;
218 	}
219 
220 	mem->pages = pages;
221 	mem->user = user;
222 
223 	return 0;
224 }
225 
226 void bpf_map_charge_finish(struct bpf_map_memory *mem)
227 {
228 	bpf_uncharge_memlock(mem->user, mem->pages);
229 	free_uid(mem->user);
230 }
231 
232 void bpf_map_charge_move(struct bpf_map_memory *dst,
233 			 struct bpf_map_memory *src)
234 {
235 	*dst = *src;
236 
237 	/* Make sure src will not be used for the redundant uncharging. */
238 	memset(src, 0, sizeof(struct bpf_map_memory));
239 }
240 
241 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
242 {
243 	int ret;
244 
245 	ret = bpf_charge_memlock(map->memory.user, pages);
246 	if (ret)
247 		return ret;
248 	map->memory.pages += pages;
249 	return ret;
250 }
251 
252 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
253 {
254 	bpf_uncharge_memlock(map->memory.user, pages);
255 	map->memory.pages -= pages;
256 }
257 
258 static int bpf_map_alloc_id(struct bpf_map *map)
259 {
260 	int id;
261 
262 	idr_preload(GFP_KERNEL);
263 	spin_lock_bh(&map_idr_lock);
264 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
265 	if (id > 0)
266 		map->id = id;
267 	spin_unlock_bh(&map_idr_lock);
268 	idr_preload_end();
269 
270 	if (WARN_ON_ONCE(!id))
271 		return -ENOSPC;
272 
273 	return id > 0 ? 0 : id;
274 }
275 
276 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
277 {
278 	unsigned long flags;
279 
280 	/* Offloaded maps are removed from the IDR store when their device
281 	 * disappears - even if someone holds an fd to them they are unusable,
282 	 * the memory is gone, all ops will fail; they are simply waiting for
283 	 * refcnt to drop to be freed.
284 	 */
285 	if (!map->id)
286 		return;
287 
288 	if (do_idr_lock)
289 		spin_lock_irqsave(&map_idr_lock, flags);
290 	else
291 		__acquire(&map_idr_lock);
292 
293 	idr_remove(&map_idr, map->id);
294 	map->id = 0;
295 
296 	if (do_idr_lock)
297 		spin_unlock_irqrestore(&map_idr_lock, flags);
298 	else
299 		__release(&map_idr_lock);
300 }
301 
302 /* called from workqueue */
303 static void bpf_map_free_deferred(struct work_struct *work)
304 {
305 	struct bpf_map *map = container_of(work, struct bpf_map, work);
306 	struct bpf_map_memory mem;
307 
308 	bpf_map_charge_move(&mem, &map->memory);
309 	security_bpf_map_free(map);
310 	/* implementation dependent freeing */
311 	map->ops->map_free(map);
312 	bpf_map_charge_finish(&mem);
313 }
314 
315 static void bpf_map_put_uref(struct bpf_map *map)
316 {
317 	if (atomic_dec_and_test(&map->usercnt)) {
318 		if (map->ops->map_release_uref)
319 			map->ops->map_release_uref(map);
320 	}
321 }
322 
323 /* decrement map refcnt and schedule it for freeing via workqueue
324  * (unrelying map implementation ops->map_free() might sleep)
325  */
326 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
327 {
328 	if (atomic_dec_and_test(&map->refcnt)) {
329 		/* bpf_map_free_id() must be called first */
330 		bpf_map_free_id(map, do_idr_lock);
331 		btf_put(map->btf);
332 		INIT_WORK(&map->work, bpf_map_free_deferred);
333 		schedule_work(&map->work);
334 	}
335 }
336 
337 void bpf_map_put(struct bpf_map *map)
338 {
339 	__bpf_map_put(map, true);
340 }
341 EXPORT_SYMBOL_GPL(bpf_map_put);
342 
343 void bpf_map_put_with_uref(struct bpf_map *map)
344 {
345 	bpf_map_put_uref(map);
346 	bpf_map_put(map);
347 }
348 
349 static int bpf_map_release(struct inode *inode, struct file *filp)
350 {
351 	struct bpf_map *map = filp->private_data;
352 
353 	if (map->ops->map_release)
354 		map->ops->map_release(map, filp);
355 
356 	bpf_map_put_with_uref(map);
357 	return 0;
358 }
359 
360 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
361 {
362 	fmode_t mode = f.file->f_mode;
363 
364 	/* Our file permissions may have been overridden by global
365 	 * map permissions facing syscall side.
366 	 */
367 	if (READ_ONCE(map->frozen))
368 		mode &= ~FMODE_CAN_WRITE;
369 	return mode;
370 }
371 
372 #ifdef CONFIG_PROC_FS
373 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
374 {
375 	const struct bpf_map *map = filp->private_data;
376 	const struct bpf_array *array;
377 	u32 owner_prog_type = 0;
378 	u32 owner_jited = 0;
379 
380 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
381 		array = container_of(map, struct bpf_array, map);
382 		owner_prog_type = array->owner_prog_type;
383 		owner_jited = array->owner_jited;
384 	}
385 
386 	seq_printf(m,
387 		   "map_type:\t%u\n"
388 		   "key_size:\t%u\n"
389 		   "value_size:\t%u\n"
390 		   "max_entries:\t%u\n"
391 		   "map_flags:\t%#x\n"
392 		   "memlock:\t%llu\n"
393 		   "map_id:\t%u\n"
394 		   "frozen:\t%u\n",
395 		   map->map_type,
396 		   map->key_size,
397 		   map->value_size,
398 		   map->max_entries,
399 		   map->map_flags,
400 		   map->memory.pages * 1ULL << PAGE_SHIFT,
401 		   map->id,
402 		   READ_ONCE(map->frozen));
403 
404 	if (owner_prog_type) {
405 		seq_printf(m, "owner_prog_type:\t%u\n",
406 			   owner_prog_type);
407 		seq_printf(m, "owner_jited:\t%u\n",
408 			   owner_jited);
409 	}
410 }
411 #endif
412 
413 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
414 			      loff_t *ppos)
415 {
416 	/* We need this handler such that alloc_file() enables
417 	 * f_mode with FMODE_CAN_READ.
418 	 */
419 	return -EINVAL;
420 }
421 
422 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
423 			       size_t siz, loff_t *ppos)
424 {
425 	/* We need this handler such that alloc_file() enables
426 	 * f_mode with FMODE_CAN_WRITE.
427 	 */
428 	return -EINVAL;
429 }
430 
431 const struct file_operations bpf_map_fops = {
432 #ifdef CONFIG_PROC_FS
433 	.show_fdinfo	= bpf_map_show_fdinfo,
434 #endif
435 	.release	= bpf_map_release,
436 	.read		= bpf_dummy_read,
437 	.write		= bpf_dummy_write,
438 };
439 
440 int bpf_map_new_fd(struct bpf_map *map, int flags)
441 {
442 	int ret;
443 
444 	ret = security_bpf_map(map, OPEN_FMODE(flags));
445 	if (ret < 0)
446 		return ret;
447 
448 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
449 				flags | O_CLOEXEC);
450 }
451 
452 int bpf_get_file_flag(int flags)
453 {
454 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
455 		return -EINVAL;
456 	if (flags & BPF_F_RDONLY)
457 		return O_RDONLY;
458 	if (flags & BPF_F_WRONLY)
459 		return O_WRONLY;
460 	return O_RDWR;
461 }
462 
463 /* helper macro to check that unused fields 'union bpf_attr' are zero */
464 #define CHECK_ATTR(CMD) \
465 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
466 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
467 		   sizeof(*attr) - \
468 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
469 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
470 
471 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
472  * Return 0 on success and < 0 on error.
473  */
474 static int bpf_obj_name_cpy(char *dst, const char *src)
475 {
476 	const char *end = src + BPF_OBJ_NAME_LEN;
477 
478 	memset(dst, 0, BPF_OBJ_NAME_LEN);
479 	/* Copy all isalnum(), '_' and '.' chars. */
480 	while (src < end && *src) {
481 		if (!isalnum(*src) &&
482 		    *src != '_' && *src != '.')
483 			return -EINVAL;
484 		*dst++ = *src++;
485 	}
486 
487 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
488 	if (src == end)
489 		return -EINVAL;
490 
491 	return 0;
492 }
493 
494 int map_check_no_btf(const struct bpf_map *map,
495 		     const struct btf *btf,
496 		     const struct btf_type *key_type,
497 		     const struct btf_type *value_type)
498 {
499 	return -ENOTSUPP;
500 }
501 
502 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
503 			 u32 btf_key_id, u32 btf_value_id)
504 {
505 	const struct btf_type *key_type, *value_type;
506 	u32 key_size, value_size;
507 	int ret = 0;
508 
509 	/* Some maps allow key to be unspecified. */
510 	if (btf_key_id) {
511 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
512 		if (!key_type || key_size != map->key_size)
513 			return -EINVAL;
514 	} else {
515 		key_type = btf_type_by_id(btf, 0);
516 		if (!map->ops->map_check_btf)
517 			return -EINVAL;
518 	}
519 
520 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
521 	if (!value_type || value_size != map->value_size)
522 		return -EINVAL;
523 
524 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
525 
526 	if (map_value_has_spin_lock(map)) {
527 		if (map->map_flags & BPF_F_RDONLY_PROG)
528 			return -EACCES;
529 		if (map->map_type != BPF_MAP_TYPE_HASH &&
530 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
531 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
532 		    map->map_type != BPF_MAP_TYPE_SK_STORAGE)
533 			return -ENOTSUPP;
534 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
535 		    map->value_size) {
536 			WARN_ONCE(1,
537 				  "verifier bug spin_lock_off %d value_size %d\n",
538 				  map->spin_lock_off, map->value_size);
539 			return -EFAULT;
540 		}
541 	}
542 
543 	if (map->ops->map_check_btf)
544 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
545 
546 	return ret;
547 }
548 
549 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
550 /* called via syscall */
551 static int map_create(union bpf_attr *attr)
552 {
553 	int numa_node = bpf_map_attr_numa_node(attr);
554 	struct bpf_map_memory mem;
555 	struct bpf_map *map;
556 	int f_flags;
557 	int err;
558 
559 	err = CHECK_ATTR(BPF_MAP_CREATE);
560 	if (err)
561 		return -EINVAL;
562 
563 	f_flags = bpf_get_file_flag(attr->map_flags);
564 	if (f_flags < 0)
565 		return f_flags;
566 
567 	if (numa_node != NUMA_NO_NODE &&
568 	    ((unsigned int)numa_node >= nr_node_ids ||
569 	     !node_online(numa_node)))
570 		return -EINVAL;
571 
572 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
573 	map = find_and_alloc_map(attr);
574 	if (IS_ERR(map))
575 		return PTR_ERR(map);
576 
577 	err = bpf_obj_name_cpy(map->name, attr->map_name);
578 	if (err)
579 		goto free_map;
580 
581 	atomic_set(&map->refcnt, 1);
582 	atomic_set(&map->usercnt, 1);
583 
584 	if (attr->btf_key_type_id || attr->btf_value_type_id) {
585 		struct btf *btf;
586 
587 		if (!attr->btf_value_type_id) {
588 			err = -EINVAL;
589 			goto free_map;
590 		}
591 
592 		btf = btf_get_by_fd(attr->btf_fd);
593 		if (IS_ERR(btf)) {
594 			err = PTR_ERR(btf);
595 			goto free_map;
596 		}
597 
598 		err = map_check_btf(map, btf, attr->btf_key_type_id,
599 				    attr->btf_value_type_id);
600 		if (err) {
601 			btf_put(btf);
602 			goto free_map;
603 		}
604 
605 		map->btf = btf;
606 		map->btf_key_type_id = attr->btf_key_type_id;
607 		map->btf_value_type_id = attr->btf_value_type_id;
608 	} else {
609 		map->spin_lock_off = -EINVAL;
610 	}
611 
612 	err = security_bpf_map_alloc(map);
613 	if (err)
614 		goto free_map;
615 
616 	err = bpf_map_alloc_id(map);
617 	if (err)
618 		goto free_map_sec;
619 
620 	err = bpf_map_new_fd(map, f_flags);
621 	if (err < 0) {
622 		/* failed to allocate fd.
623 		 * bpf_map_put_with_uref() is needed because the above
624 		 * bpf_map_alloc_id() has published the map
625 		 * to the userspace and the userspace may
626 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
627 		 */
628 		bpf_map_put_with_uref(map);
629 		return err;
630 	}
631 
632 	return err;
633 
634 free_map_sec:
635 	security_bpf_map_free(map);
636 free_map:
637 	btf_put(map->btf);
638 	bpf_map_charge_move(&mem, &map->memory);
639 	map->ops->map_free(map);
640 	bpf_map_charge_finish(&mem);
641 	return err;
642 }
643 
644 /* if error is returned, fd is released.
645  * On success caller should complete fd access with matching fdput()
646  */
647 struct bpf_map *__bpf_map_get(struct fd f)
648 {
649 	if (!f.file)
650 		return ERR_PTR(-EBADF);
651 	if (f.file->f_op != &bpf_map_fops) {
652 		fdput(f);
653 		return ERR_PTR(-EINVAL);
654 	}
655 
656 	return f.file->private_data;
657 }
658 
659 /* prog's and map's refcnt limit */
660 #define BPF_MAX_REFCNT 32768
661 
662 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
663 {
664 	if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
665 		atomic_dec(&map->refcnt);
666 		return ERR_PTR(-EBUSY);
667 	}
668 	if (uref)
669 		atomic_inc(&map->usercnt);
670 	return map;
671 }
672 EXPORT_SYMBOL_GPL(bpf_map_inc);
673 
674 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
675 {
676 	struct fd f = fdget(ufd);
677 	struct bpf_map *map;
678 
679 	map = __bpf_map_get(f);
680 	if (IS_ERR(map))
681 		return map;
682 
683 	map = bpf_map_inc(map, true);
684 	fdput(f);
685 
686 	return map;
687 }
688 
689 /* map_idr_lock should have been held */
690 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map,
691 					      bool uref)
692 {
693 	int refold;
694 
695 	refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
696 
697 	if (refold >= BPF_MAX_REFCNT) {
698 		__bpf_map_put(map, false);
699 		return ERR_PTR(-EBUSY);
700 	}
701 
702 	if (!refold)
703 		return ERR_PTR(-ENOENT);
704 
705 	if (uref)
706 		atomic_inc(&map->usercnt);
707 
708 	return map;
709 }
710 
711 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
712 {
713 	spin_lock_bh(&map_idr_lock);
714 	map = __bpf_map_inc_not_zero(map, uref);
715 	spin_unlock_bh(&map_idr_lock);
716 
717 	return map;
718 }
719 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
720 
721 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
722 {
723 	return -ENOTSUPP;
724 }
725 
726 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
727 {
728 	if (key_size)
729 		return memdup_user(ukey, key_size);
730 
731 	if (ukey)
732 		return ERR_PTR(-EINVAL);
733 
734 	return NULL;
735 }
736 
737 /* last field in 'union bpf_attr' used by this command */
738 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
739 
740 static int map_lookup_elem(union bpf_attr *attr)
741 {
742 	void __user *ukey = u64_to_user_ptr(attr->key);
743 	void __user *uvalue = u64_to_user_ptr(attr->value);
744 	int ufd = attr->map_fd;
745 	struct bpf_map *map;
746 	void *key, *value, *ptr;
747 	u32 value_size;
748 	struct fd f;
749 	int err;
750 
751 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
752 		return -EINVAL;
753 
754 	if (attr->flags & ~BPF_F_LOCK)
755 		return -EINVAL;
756 
757 	f = fdget(ufd);
758 	map = __bpf_map_get(f);
759 	if (IS_ERR(map))
760 		return PTR_ERR(map);
761 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
762 		err = -EPERM;
763 		goto err_put;
764 	}
765 
766 	if ((attr->flags & BPF_F_LOCK) &&
767 	    !map_value_has_spin_lock(map)) {
768 		err = -EINVAL;
769 		goto err_put;
770 	}
771 
772 	key = __bpf_copy_key(ukey, map->key_size);
773 	if (IS_ERR(key)) {
774 		err = PTR_ERR(key);
775 		goto err_put;
776 	}
777 
778 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
779 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
780 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
781 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
782 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
783 	else if (IS_FD_MAP(map))
784 		value_size = sizeof(u32);
785 	else
786 		value_size = map->value_size;
787 
788 	err = -ENOMEM;
789 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
790 	if (!value)
791 		goto free_key;
792 
793 	if (bpf_map_is_dev_bound(map)) {
794 		err = bpf_map_offload_lookup_elem(map, key, value);
795 		goto done;
796 	}
797 
798 	preempt_disable();
799 	this_cpu_inc(bpf_prog_active);
800 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
801 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
802 		err = bpf_percpu_hash_copy(map, key, value);
803 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
804 		err = bpf_percpu_array_copy(map, key, value);
805 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
806 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
807 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
808 		err = bpf_stackmap_copy(map, key, value);
809 	} else if (IS_FD_ARRAY(map)) {
810 		err = bpf_fd_array_map_lookup_elem(map, key, value);
811 	} else if (IS_FD_HASH(map)) {
812 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
813 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
814 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
815 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
816 		   map->map_type == BPF_MAP_TYPE_STACK) {
817 		err = map->ops->map_peek_elem(map, value);
818 	} else {
819 		rcu_read_lock();
820 		if (map->ops->map_lookup_elem_sys_only)
821 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
822 		else
823 			ptr = map->ops->map_lookup_elem(map, key);
824 		if (IS_ERR(ptr)) {
825 			err = PTR_ERR(ptr);
826 		} else if (!ptr) {
827 			err = -ENOENT;
828 		} else {
829 			err = 0;
830 			if (attr->flags & BPF_F_LOCK)
831 				/* lock 'ptr' and copy everything but lock */
832 				copy_map_value_locked(map, value, ptr, true);
833 			else
834 				copy_map_value(map, value, ptr);
835 			/* mask lock, since value wasn't zero inited */
836 			check_and_init_map_lock(map, value);
837 		}
838 		rcu_read_unlock();
839 	}
840 	this_cpu_dec(bpf_prog_active);
841 	preempt_enable();
842 
843 done:
844 	if (err)
845 		goto free_value;
846 
847 	err = -EFAULT;
848 	if (copy_to_user(uvalue, value, value_size) != 0)
849 		goto free_value;
850 
851 	err = 0;
852 
853 free_value:
854 	kfree(value);
855 free_key:
856 	kfree(key);
857 err_put:
858 	fdput(f);
859 	return err;
860 }
861 
862 static void maybe_wait_bpf_programs(struct bpf_map *map)
863 {
864 	/* Wait for any running BPF programs to complete so that
865 	 * userspace, when we return to it, knows that all programs
866 	 * that could be running use the new map value.
867 	 */
868 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
869 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
870 		synchronize_rcu();
871 }
872 
873 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
874 
875 static int map_update_elem(union bpf_attr *attr)
876 {
877 	void __user *ukey = u64_to_user_ptr(attr->key);
878 	void __user *uvalue = u64_to_user_ptr(attr->value);
879 	int ufd = attr->map_fd;
880 	struct bpf_map *map;
881 	void *key, *value;
882 	u32 value_size;
883 	struct fd f;
884 	int err;
885 
886 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
887 		return -EINVAL;
888 
889 	f = fdget(ufd);
890 	map = __bpf_map_get(f);
891 	if (IS_ERR(map))
892 		return PTR_ERR(map);
893 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
894 		err = -EPERM;
895 		goto err_put;
896 	}
897 
898 	if ((attr->flags & BPF_F_LOCK) &&
899 	    !map_value_has_spin_lock(map)) {
900 		err = -EINVAL;
901 		goto err_put;
902 	}
903 
904 	key = __bpf_copy_key(ukey, map->key_size);
905 	if (IS_ERR(key)) {
906 		err = PTR_ERR(key);
907 		goto err_put;
908 	}
909 
910 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
911 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
912 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
913 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
914 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
915 	else
916 		value_size = map->value_size;
917 
918 	err = -ENOMEM;
919 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
920 	if (!value)
921 		goto free_key;
922 
923 	err = -EFAULT;
924 	if (copy_from_user(value, uvalue, value_size) != 0)
925 		goto free_value;
926 
927 	/* Need to create a kthread, thus must support schedule */
928 	if (bpf_map_is_dev_bound(map)) {
929 		err = bpf_map_offload_update_elem(map, key, value, attr->flags);
930 		goto out;
931 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
932 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
933 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
934 		err = map->ops->map_update_elem(map, key, value, attr->flags);
935 		goto out;
936 	}
937 
938 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
939 	 * inside bpf map update or delete otherwise deadlocks are possible
940 	 */
941 	preempt_disable();
942 	__this_cpu_inc(bpf_prog_active);
943 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
944 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
945 		err = bpf_percpu_hash_update(map, key, value, attr->flags);
946 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
947 		err = bpf_percpu_array_update(map, key, value, attr->flags);
948 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
949 		err = bpf_percpu_cgroup_storage_update(map, key, value,
950 						       attr->flags);
951 	} else if (IS_FD_ARRAY(map)) {
952 		rcu_read_lock();
953 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
954 						   attr->flags);
955 		rcu_read_unlock();
956 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
957 		rcu_read_lock();
958 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
959 						  attr->flags);
960 		rcu_read_unlock();
961 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
962 		/* rcu_read_lock() is not needed */
963 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
964 							 attr->flags);
965 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
966 		   map->map_type == BPF_MAP_TYPE_STACK) {
967 		err = map->ops->map_push_elem(map, value, attr->flags);
968 	} else {
969 		rcu_read_lock();
970 		err = map->ops->map_update_elem(map, key, value, attr->flags);
971 		rcu_read_unlock();
972 	}
973 	__this_cpu_dec(bpf_prog_active);
974 	preempt_enable();
975 	maybe_wait_bpf_programs(map);
976 out:
977 free_value:
978 	kfree(value);
979 free_key:
980 	kfree(key);
981 err_put:
982 	fdput(f);
983 	return err;
984 }
985 
986 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
987 
988 static int map_delete_elem(union bpf_attr *attr)
989 {
990 	void __user *ukey = u64_to_user_ptr(attr->key);
991 	int ufd = attr->map_fd;
992 	struct bpf_map *map;
993 	struct fd f;
994 	void *key;
995 	int err;
996 
997 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
998 		return -EINVAL;
999 
1000 	f = fdget(ufd);
1001 	map = __bpf_map_get(f);
1002 	if (IS_ERR(map))
1003 		return PTR_ERR(map);
1004 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1005 		err = -EPERM;
1006 		goto err_put;
1007 	}
1008 
1009 	key = __bpf_copy_key(ukey, map->key_size);
1010 	if (IS_ERR(key)) {
1011 		err = PTR_ERR(key);
1012 		goto err_put;
1013 	}
1014 
1015 	if (bpf_map_is_dev_bound(map)) {
1016 		err = bpf_map_offload_delete_elem(map, key);
1017 		goto out;
1018 	}
1019 
1020 	preempt_disable();
1021 	__this_cpu_inc(bpf_prog_active);
1022 	rcu_read_lock();
1023 	err = map->ops->map_delete_elem(map, key);
1024 	rcu_read_unlock();
1025 	__this_cpu_dec(bpf_prog_active);
1026 	preempt_enable();
1027 	maybe_wait_bpf_programs(map);
1028 out:
1029 	kfree(key);
1030 err_put:
1031 	fdput(f);
1032 	return err;
1033 }
1034 
1035 /* last field in 'union bpf_attr' used by this command */
1036 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1037 
1038 static int map_get_next_key(union bpf_attr *attr)
1039 {
1040 	void __user *ukey = u64_to_user_ptr(attr->key);
1041 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1042 	int ufd = attr->map_fd;
1043 	struct bpf_map *map;
1044 	void *key, *next_key;
1045 	struct fd f;
1046 	int err;
1047 
1048 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1049 		return -EINVAL;
1050 
1051 	f = fdget(ufd);
1052 	map = __bpf_map_get(f);
1053 	if (IS_ERR(map))
1054 		return PTR_ERR(map);
1055 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1056 		err = -EPERM;
1057 		goto err_put;
1058 	}
1059 
1060 	if (ukey) {
1061 		key = __bpf_copy_key(ukey, map->key_size);
1062 		if (IS_ERR(key)) {
1063 			err = PTR_ERR(key);
1064 			goto err_put;
1065 		}
1066 	} else {
1067 		key = NULL;
1068 	}
1069 
1070 	err = -ENOMEM;
1071 	next_key = kmalloc(map->key_size, GFP_USER);
1072 	if (!next_key)
1073 		goto free_key;
1074 
1075 	if (bpf_map_is_dev_bound(map)) {
1076 		err = bpf_map_offload_get_next_key(map, key, next_key);
1077 		goto out;
1078 	}
1079 
1080 	rcu_read_lock();
1081 	err = map->ops->map_get_next_key(map, key, next_key);
1082 	rcu_read_unlock();
1083 out:
1084 	if (err)
1085 		goto free_next_key;
1086 
1087 	err = -EFAULT;
1088 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1089 		goto free_next_key;
1090 
1091 	err = 0;
1092 
1093 free_next_key:
1094 	kfree(next_key);
1095 free_key:
1096 	kfree(key);
1097 err_put:
1098 	fdput(f);
1099 	return err;
1100 }
1101 
1102 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1103 
1104 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1105 {
1106 	void __user *ukey = u64_to_user_ptr(attr->key);
1107 	void __user *uvalue = u64_to_user_ptr(attr->value);
1108 	int ufd = attr->map_fd;
1109 	struct bpf_map *map;
1110 	void *key, *value;
1111 	u32 value_size;
1112 	struct fd f;
1113 	int err;
1114 
1115 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1116 		return -EINVAL;
1117 
1118 	f = fdget(ufd);
1119 	map = __bpf_map_get(f);
1120 	if (IS_ERR(map))
1121 		return PTR_ERR(map);
1122 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1123 		err = -EPERM;
1124 		goto err_put;
1125 	}
1126 
1127 	key = __bpf_copy_key(ukey, map->key_size);
1128 	if (IS_ERR(key)) {
1129 		err = PTR_ERR(key);
1130 		goto err_put;
1131 	}
1132 
1133 	value_size = map->value_size;
1134 
1135 	err = -ENOMEM;
1136 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1137 	if (!value)
1138 		goto free_key;
1139 
1140 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1141 	    map->map_type == BPF_MAP_TYPE_STACK) {
1142 		err = map->ops->map_pop_elem(map, value);
1143 	} else {
1144 		err = -ENOTSUPP;
1145 	}
1146 
1147 	if (err)
1148 		goto free_value;
1149 
1150 	if (copy_to_user(uvalue, value, value_size) != 0)
1151 		goto free_value;
1152 
1153 	err = 0;
1154 
1155 free_value:
1156 	kfree(value);
1157 free_key:
1158 	kfree(key);
1159 err_put:
1160 	fdput(f);
1161 	return err;
1162 }
1163 
1164 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1165 
1166 static int map_freeze(const union bpf_attr *attr)
1167 {
1168 	int err = 0, ufd = attr->map_fd;
1169 	struct bpf_map *map;
1170 	struct fd f;
1171 
1172 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1173 		return -EINVAL;
1174 
1175 	f = fdget(ufd);
1176 	map = __bpf_map_get(f);
1177 	if (IS_ERR(map))
1178 		return PTR_ERR(map);
1179 	if (READ_ONCE(map->frozen)) {
1180 		err = -EBUSY;
1181 		goto err_put;
1182 	}
1183 	if (!capable(CAP_SYS_ADMIN)) {
1184 		err = -EPERM;
1185 		goto err_put;
1186 	}
1187 
1188 	WRITE_ONCE(map->frozen, true);
1189 err_put:
1190 	fdput(f);
1191 	return err;
1192 }
1193 
1194 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1195 #define BPF_PROG_TYPE(_id, _name) \
1196 	[_id] = & _name ## _prog_ops,
1197 #define BPF_MAP_TYPE(_id, _ops)
1198 #include <linux/bpf_types.h>
1199 #undef BPF_PROG_TYPE
1200 #undef BPF_MAP_TYPE
1201 };
1202 
1203 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1204 {
1205 	const struct bpf_prog_ops *ops;
1206 
1207 	if (type >= ARRAY_SIZE(bpf_prog_types))
1208 		return -EINVAL;
1209 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1210 	ops = bpf_prog_types[type];
1211 	if (!ops)
1212 		return -EINVAL;
1213 
1214 	if (!bpf_prog_is_dev_bound(prog->aux))
1215 		prog->aux->ops = ops;
1216 	else
1217 		prog->aux->ops = &bpf_offload_prog_ops;
1218 	prog->type = type;
1219 	return 0;
1220 }
1221 
1222 /* drop refcnt on maps used by eBPF program and free auxilary data */
1223 static void free_used_maps(struct bpf_prog_aux *aux)
1224 {
1225 	enum bpf_cgroup_storage_type stype;
1226 	int i;
1227 
1228 	for_each_cgroup_storage_type(stype) {
1229 		if (!aux->cgroup_storage[stype])
1230 			continue;
1231 		bpf_cgroup_storage_release(aux->prog,
1232 					   aux->cgroup_storage[stype]);
1233 	}
1234 
1235 	for (i = 0; i < aux->used_map_cnt; i++)
1236 		bpf_map_put(aux->used_maps[i]);
1237 
1238 	kfree(aux->used_maps);
1239 }
1240 
1241 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1242 {
1243 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1244 	unsigned long user_bufs;
1245 
1246 	if (user) {
1247 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1248 		if (user_bufs > memlock_limit) {
1249 			atomic_long_sub(pages, &user->locked_vm);
1250 			return -EPERM;
1251 		}
1252 	}
1253 
1254 	return 0;
1255 }
1256 
1257 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1258 {
1259 	if (user)
1260 		atomic_long_sub(pages, &user->locked_vm);
1261 }
1262 
1263 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1264 {
1265 	struct user_struct *user = get_current_user();
1266 	int ret;
1267 
1268 	ret = __bpf_prog_charge(user, prog->pages);
1269 	if (ret) {
1270 		free_uid(user);
1271 		return ret;
1272 	}
1273 
1274 	prog->aux->user = user;
1275 	return 0;
1276 }
1277 
1278 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1279 {
1280 	struct user_struct *user = prog->aux->user;
1281 
1282 	__bpf_prog_uncharge(user, prog->pages);
1283 	free_uid(user);
1284 }
1285 
1286 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1287 {
1288 	int id;
1289 
1290 	idr_preload(GFP_KERNEL);
1291 	spin_lock_bh(&prog_idr_lock);
1292 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1293 	if (id > 0)
1294 		prog->aux->id = id;
1295 	spin_unlock_bh(&prog_idr_lock);
1296 	idr_preload_end();
1297 
1298 	/* id is in [1, INT_MAX) */
1299 	if (WARN_ON_ONCE(!id))
1300 		return -ENOSPC;
1301 
1302 	return id > 0 ? 0 : id;
1303 }
1304 
1305 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1306 {
1307 	/* cBPF to eBPF migrations are currently not in the idr store.
1308 	 * Offloaded programs are removed from the store when their device
1309 	 * disappears - even if someone grabs an fd to them they are unusable,
1310 	 * simply waiting for refcnt to drop to be freed.
1311 	 */
1312 	if (!prog->aux->id)
1313 		return;
1314 
1315 	if (do_idr_lock)
1316 		spin_lock_bh(&prog_idr_lock);
1317 	else
1318 		__acquire(&prog_idr_lock);
1319 
1320 	idr_remove(&prog_idr, prog->aux->id);
1321 	prog->aux->id = 0;
1322 
1323 	if (do_idr_lock)
1324 		spin_unlock_bh(&prog_idr_lock);
1325 	else
1326 		__release(&prog_idr_lock);
1327 }
1328 
1329 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1330 {
1331 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1332 
1333 	kvfree(aux->func_info);
1334 	free_used_maps(aux);
1335 	bpf_prog_uncharge_memlock(aux->prog);
1336 	security_bpf_prog_free(aux);
1337 	bpf_prog_free(aux->prog);
1338 }
1339 
1340 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1341 {
1342 	bpf_prog_kallsyms_del_all(prog);
1343 	btf_put(prog->aux->btf);
1344 	bpf_prog_free_linfo(prog);
1345 
1346 	if (deferred)
1347 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1348 	else
1349 		__bpf_prog_put_rcu(&prog->aux->rcu);
1350 }
1351 
1352 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1353 {
1354 	if (atomic_dec_and_test(&prog->aux->refcnt)) {
1355 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1356 		/* bpf_prog_free_id() must be called first */
1357 		bpf_prog_free_id(prog, do_idr_lock);
1358 		__bpf_prog_put_noref(prog, true);
1359 	}
1360 }
1361 
1362 void bpf_prog_put(struct bpf_prog *prog)
1363 {
1364 	__bpf_prog_put(prog, true);
1365 }
1366 EXPORT_SYMBOL_GPL(bpf_prog_put);
1367 
1368 static int bpf_prog_release(struct inode *inode, struct file *filp)
1369 {
1370 	struct bpf_prog *prog = filp->private_data;
1371 
1372 	bpf_prog_put(prog);
1373 	return 0;
1374 }
1375 
1376 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1377 			       struct bpf_prog_stats *stats)
1378 {
1379 	u64 nsecs = 0, cnt = 0;
1380 	int cpu;
1381 
1382 	for_each_possible_cpu(cpu) {
1383 		const struct bpf_prog_stats *st;
1384 		unsigned int start;
1385 		u64 tnsecs, tcnt;
1386 
1387 		st = per_cpu_ptr(prog->aux->stats, cpu);
1388 		do {
1389 			start = u64_stats_fetch_begin_irq(&st->syncp);
1390 			tnsecs = st->nsecs;
1391 			tcnt = st->cnt;
1392 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1393 		nsecs += tnsecs;
1394 		cnt += tcnt;
1395 	}
1396 	stats->nsecs = nsecs;
1397 	stats->cnt = cnt;
1398 }
1399 
1400 #ifdef CONFIG_PROC_FS
1401 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1402 {
1403 	const struct bpf_prog *prog = filp->private_data;
1404 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1405 	struct bpf_prog_stats stats;
1406 
1407 	bpf_prog_get_stats(prog, &stats);
1408 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1409 	seq_printf(m,
1410 		   "prog_type:\t%u\n"
1411 		   "prog_jited:\t%u\n"
1412 		   "prog_tag:\t%s\n"
1413 		   "memlock:\t%llu\n"
1414 		   "prog_id:\t%u\n"
1415 		   "run_time_ns:\t%llu\n"
1416 		   "run_cnt:\t%llu\n",
1417 		   prog->type,
1418 		   prog->jited,
1419 		   prog_tag,
1420 		   prog->pages * 1ULL << PAGE_SHIFT,
1421 		   prog->aux->id,
1422 		   stats.nsecs,
1423 		   stats.cnt);
1424 }
1425 #endif
1426 
1427 const struct file_operations bpf_prog_fops = {
1428 #ifdef CONFIG_PROC_FS
1429 	.show_fdinfo	= bpf_prog_show_fdinfo,
1430 #endif
1431 	.release	= bpf_prog_release,
1432 	.read		= bpf_dummy_read,
1433 	.write		= bpf_dummy_write,
1434 };
1435 
1436 int bpf_prog_new_fd(struct bpf_prog *prog)
1437 {
1438 	int ret;
1439 
1440 	ret = security_bpf_prog(prog);
1441 	if (ret < 0)
1442 		return ret;
1443 
1444 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1445 				O_RDWR | O_CLOEXEC);
1446 }
1447 
1448 static struct bpf_prog *____bpf_prog_get(struct fd f)
1449 {
1450 	if (!f.file)
1451 		return ERR_PTR(-EBADF);
1452 	if (f.file->f_op != &bpf_prog_fops) {
1453 		fdput(f);
1454 		return ERR_PTR(-EINVAL);
1455 	}
1456 
1457 	return f.file->private_data;
1458 }
1459 
1460 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1461 {
1462 	if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1463 		atomic_sub(i, &prog->aux->refcnt);
1464 		return ERR_PTR(-EBUSY);
1465 	}
1466 	return prog;
1467 }
1468 EXPORT_SYMBOL_GPL(bpf_prog_add);
1469 
1470 void bpf_prog_sub(struct bpf_prog *prog, int i)
1471 {
1472 	/* Only to be used for undoing previous bpf_prog_add() in some
1473 	 * error path. We still know that another entity in our call
1474 	 * path holds a reference to the program, thus atomic_sub() can
1475 	 * be safely used in such cases!
1476 	 */
1477 	WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1478 }
1479 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1480 
1481 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1482 {
1483 	return bpf_prog_add(prog, 1);
1484 }
1485 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1486 
1487 /* prog_idr_lock should have been held */
1488 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1489 {
1490 	int refold;
1491 
1492 	refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1493 
1494 	if (refold >= BPF_MAX_REFCNT) {
1495 		__bpf_prog_put(prog, false);
1496 		return ERR_PTR(-EBUSY);
1497 	}
1498 
1499 	if (!refold)
1500 		return ERR_PTR(-ENOENT);
1501 
1502 	return prog;
1503 }
1504 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1505 
1506 bool bpf_prog_get_ok(struct bpf_prog *prog,
1507 			    enum bpf_prog_type *attach_type, bool attach_drv)
1508 {
1509 	/* not an attachment, just a refcount inc, always allow */
1510 	if (!attach_type)
1511 		return true;
1512 
1513 	if (prog->type != *attach_type)
1514 		return false;
1515 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1516 		return false;
1517 
1518 	return true;
1519 }
1520 
1521 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1522 				       bool attach_drv)
1523 {
1524 	struct fd f = fdget(ufd);
1525 	struct bpf_prog *prog;
1526 
1527 	prog = ____bpf_prog_get(f);
1528 	if (IS_ERR(prog))
1529 		return prog;
1530 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1531 		prog = ERR_PTR(-EINVAL);
1532 		goto out;
1533 	}
1534 
1535 	prog = bpf_prog_inc(prog);
1536 out:
1537 	fdput(f);
1538 	return prog;
1539 }
1540 
1541 struct bpf_prog *bpf_prog_get(u32 ufd)
1542 {
1543 	return __bpf_prog_get(ufd, NULL, false);
1544 }
1545 
1546 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1547 				       bool attach_drv)
1548 {
1549 	return __bpf_prog_get(ufd, &type, attach_drv);
1550 }
1551 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1552 
1553 /* Initially all BPF programs could be loaded w/o specifying
1554  * expected_attach_type. Later for some of them specifying expected_attach_type
1555  * at load time became required so that program could be validated properly.
1556  * Programs of types that are allowed to be loaded both w/ and w/o (for
1557  * backward compatibility) expected_attach_type, should have the default attach
1558  * type assigned to expected_attach_type for the latter case, so that it can be
1559  * validated later at attach time.
1560  *
1561  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1562  * prog type requires it but has some attach types that have to be backward
1563  * compatible.
1564  */
1565 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1566 {
1567 	switch (attr->prog_type) {
1568 	case BPF_PROG_TYPE_CGROUP_SOCK:
1569 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1570 		 * exist so checking for non-zero is the way to go here.
1571 		 */
1572 		if (!attr->expected_attach_type)
1573 			attr->expected_attach_type =
1574 				BPF_CGROUP_INET_SOCK_CREATE;
1575 		break;
1576 	}
1577 }
1578 
1579 static int
1580 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1581 			   enum bpf_attach_type expected_attach_type,
1582 			   u32 btf_id)
1583 {
1584 	switch (prog_type) {
1585 	case BPF_PROG_TYPE_TRACING:
1586 		if (btf_id > BTF_MAX_TYPE)
1587 			return -EINVAL;
1588 		break;
1589 	default:
1590 		if (btf_id)
1591 			return -EINVAL;
1592 		break;
1593 	}
1594 
1595 	switch (prog_type) {
1596 	case BPF_PROG_TYPE_CGROUP_SOCK:
1597 		switch (expected_attach_type) {
1598 		case BPF_CGROUP_INET_SOCK_CREATE:
1599 		case BPF_CGROUP_INET4_POST_BIND:
1600 		case BPF_CGROUP_INET6_POST_BIND:
1601 			return 0;
1602 		default:
1603 			return -EINVAL;
1604 		}
1605 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1606 		switch (expected_attach_type) {
1607 		case BPF_CGROUP_INET4_BIND:
1608 		case BPF_CGROUP_INET6_BIND:
1609 		case BPF_CGROUP_INET4_CONNECT:
1610 		case BPF_CGROUP_INET6_CONNECT:
1611 		case BPF_CGROUP_UDP4_SENDMSG:
1612 		case BPF_CGROUP_UDP6_SENDMSG:
1613 		case BPF_CGROUP_UDP4_RECVMSG:
1614 		case BPF_CGROUP_UDP6_RECVMSG:
1615 			return 0;
1616 		default:
1617 			return -EINVAL;
1618 		}
1619 	case BPF_PROG_TYPE_CGROUP_SKB:
1620 		switch (expected_attach_type) {
1621 		case BPF_CGROUP_INET_INGRESS:
1622 		case BPF_CGROUP_INET_EGRESS:
1623 			return 0;
1624 		default:
1625 			return -EINVAL;
1626 		}
1627 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1628 		switch (expected_attach_type) {
1629 		case BPF_CGROUP_SETSOCKOPT:
1630 		case BPF_CGROUP_GETSOCKOPT:
1631 			return 0;
1632 		default:
1633 			return -EINVAL;
1634 		}
1635 	default:
1636 		return 0;
1637 	}
1638 }
1639 
1640 /* last field in 'union bpf_attr' used by this command */
1641 #define	BPF_PROG_LOAD_LAST_FIELD attach_btf_id
1642 
1643 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1644 {
1645 	enum bpf_prog_type type = attr->prog_type;
1646 	struct bpf_prog *prog;
1647 	int err;
1648 	char license[128];
1649 	bool is_gpl;
1650 
1651 	if (CHECK_ATTR(BPF_PROG_LOAD))
1652 		return -EINVAL;
1653 
1654 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1655 				 BPF_F_ANY_ALIGNMENT |
1656 				 BPF_F_TEST_STATE_FREQ |
1657 				 BPF_F_TEST_RND_HI32))
1658 		return -EINVAL;
1659 
1660 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1661 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1662 	    !capable(CAP_SYS_ADMIN))
1663 		return -EPERM;
1664 
1665 	/* copy eBPF program license from user space */
1666 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1667 			      sizeof(license) - 1) < 0)
1668 		return -EFAULT;
1669 	license[sizeof(license) - 1] = 0;
1670 
1671 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
1672 	is_gpl = license_is_gpl_compatible(license);
1673 
1674 	if (attr->insn_cnt == 0 ||
1675 	    attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1676 		return -E2BIG;
1677 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1678 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
1679 	    !capable(CAP_SYS_ADMIN))
1680 		return -EPERM;
1681 
1682 	bpf_prog_load_fixup_attach_type(attr);
1683 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1684 				       attr->attach_btf_id))
1685 		return -EINVAL;
1686 
1687 	/* plain bpf_prog allocation */
1688 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1689 	if (!prog)
1690 		return -ENOMEM;
1691 
1692 	prog->expected_attach_type = attr->expected_attach_type;
1693 	prog->aux->attach_btf_id = attr->attach_btf_id;
1694 
1695 	prog->aux->offload_requested = !!attr->prog_ifindex;
1696 
1697 	err = security_bpf_prog_alloc(prog->aux);
1698 	if (err)
1699 		goto free_prog_nouncharge;
1700 
1701 	err = bpf_prog_charge_memlock(prog);
1702 	if (err)
1703 		goto free_prog_sec;
1704 
1705 	prog->len = attr->insn_cnt;
1706 
1707 	err = -EFAULT;
1708 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1709 			   bpf_prog_insn_size(prog)) != 0)
1710 		goto free_prog;
1711 
1712 	prog->orig_prog = NULL;
1713 	prog->jited = 0;
1714 
1715 	atomic_set(&prog->aux->refcnt, 1);
1716 	prog->gpl_compatible = is_gpl ? 1 : 0;
1717 
1718 	if (bpf_prog_is_dev_bound(prog->aux)) {
1719 		err = bpf_prog_offload_init(prog, attr);
1720 		if (err)
1721 			goto free_prog;
1722 	}
1723 
1724 	/* find program type: socket_filter vs tracing_filter */
1725 	err = find_prog_type(type, prog);
1726 	if (err < 0)
1727 		goto free_prog;
1728 
1729 	prog->aux->load_time = ktime_get_boottime_ns();
1730 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1731 	if (err)
1732 		goto free_prog;
1733 
1734 	/* run eBPF verifier */
1735 	err = bpf_check(&prog, attr, uattr);
1736 	if (err < 0)
1737 		goto free_used_maps;
1738 
1739 	prog = bpf_prog_select_runtime(prog, &err);
1740 	if (err < 0)
1741 		goto free_used_maps;
1742 
1743 	err = bpf_prog_alloc_id(prog);
1744 	if (err)
1745 		goto free_used_maps;
1746 
1747 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
1748 	 * effectively publicly exposed. However, retrieving via
1749 	 * bpf_prog_get_fd_by_id() will take another reference,
1750 	 * therefore it cannot be gone underneath us.
1751 	 *
1752 	 * Only for the time /after/ successful bpf_prog_new_fd()
1753 	 * and before returning to userspace, we might just hold
1754 	 * one reference and any parallel close on that fd could
1755 	 * rip everything out. Hence, below notifications must
1756 	 * happen before bpf_prog_new_fd().
1757 	 *
1758 	 * Also, any failure handling from this point onwards must
1759 	 * be using bpf_prog_put() given the program is exposed.
1760 	 */
1761 	bpf_prog_kallsyms_add(prog);
1762 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1763 
1764 	err = bpf_prog_new_fd(prog);
1765 	if (err < 0)
1766 		bpf_prog_put(prog);
1767 	return err;
1768 
1769 free_used_maps:
1770 	/* In case we have subprogs, we need to wait for a grace
1771 	 * period before we can tear down JIT memory since symbols
1772 	 * are already exposed under kallsyms.
1773 	 */
1774 	__bpf_prog_put_noref(prog, prog->aux->func_cnt);
1775 	return err;
1776 free_prog:
1777 	bpf_prog_uncharge_memlock(prog);
1778 free_prog_sec:
1779 	security_bpf_prog_free(prog->aux);
1780 free_prog_nouncharge:
1781 	bpf_prog_free(prog);
1782 	return err;
1783 }
1784 
1785 #define BPF_OBJ_LAST_FIELD file_flags
1786 
1787 static int bpf_obj_pin(const union bpf_attr *attr)
1788 {
1789 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1790 		return -EINVAL;
1791 
1792 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1793 }
1794 
1795 static int bpf_obj_get(const union bpf_attr *attr)
1796 {
1797 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1798 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1799 		return -EINVAL;
1800 
1801 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1802 				attr->file_flags);
1803 }
1804 
1805 struct bpf_raw_tracepoint {
1806 	struct bpf_raw_event_map *btp;
1807 	struct bpf_prog *prog;
1808 };
1809 
1810 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1811 {
1812 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1813 
1814 	if (raw_tp->prog) {
1815 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1816 		bpf_prog_put(raw_tp->prog);
1817 	}
1818 	bpf_put_raw_tracepoint(raw_tp->btp);
1819 	kfree(raw_tp);
1820 	return 0;
1821 }
1822 
1823 static const struct file_operations bpf_raw_tp_fops = {
1824 	.release	= bpf_raw_tracepoint_release,
1825 	.read		= bpf_dummy_read,
1826 	.write		= bpf_dummy_write,
1827 };
1828 
1829 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1830 
1831 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1832 {
1833 	struct bpf_raw_tracepoint *raw_tp;
1834 	struct bpf_raw_event_map *btp;
1835 	struct bpf_prog *prog;
1836 	const char *tp_name;
1837 	char buf[128];
1838 	int tp_fd, err;
1839 
1840 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1841 		return -EINVAL;
1842 
1843 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1844 	if (IS_ERR(prog))
1845 		return PTR_ERR(prog);
1846 
1847 	if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1848 	    prog->type != BPF_PROG_TYPE_TRACING &&
1849 	    prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1850 		err = -EINVAL;
1851 		goto out_put_prog;
1852 	}
1853 
1854 	if (prog->type == BPF_PROG_TYPE_TRACING) {
1855 		if (attr->raw_tracepoint.name) {
1856 			/* raw_tp name should not be specified in raw_tp
1857 			 * programs that were verified via in-kernel BTF info
1858 			 */
1859 			err = -EINVAL;
1860 			goto out_put_prog;
1861 		}
1862 		/* raw_tp name is taken from type name instead */
1863 		tp_name = prog->aux->attach_func_name;
1864 	} else {
1865 		if (strncpy_from_user(buf,
1866 				      u64_to_user_ptr(attr->raw_tracepoint.name),
1867 				      sizeof(buf) - 1) < 0) {
1868 			err = -EFAULT;
1869 			goto out_put_prog;
1870 		}
1871 		buf[sizeof(buf) - 1] = 0;
1872 		tp_name = buf;
1873 	}
1874 
1875 	btp = bpf_get_raw_tracepoint(tp_name);
1876 	if (!btp) {
1877 		err = -ENOENT;
1878 		goto out_put_prog;
1879 	}
1880 
1881 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1882 	if (!raw_tp) {
1883 		err = -ENOMEM;
1884 		goto out_put_btp;
1885 	}
1886 	raw_tp->btp = btp;
1887 	raw_tp->prog = prog;
1888 
1889 	err = bpf_probe_register(raw_tp->btp, prog);
1890 	if (err)
1891 		goto out_free_tp;
1892 
1893 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1894 				 O_CLOEXEC);
1895 	if (tp_fd < 0) {
1896 		bpf_probe_unregister(raw_tp->btp, prog);
1897 		err = tp_fd;
1898 		goto out_free_tp;
1899 	}
1900 	return tp_fd;
1901 
1902 out_free_tp:
1903 	kfree(raw_tp);
1904 out_put_btp:
1905 	bpf_put_raw_tracepoint(btp);
1906 out_put_prog:
1907 	bpf_prog_put(prog);
1908 	return err;
1909 }
1910 
1911 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1912 					     enum bpf_attach_type attach_type)
1913 {
1914 	switch (prog->type) {
1915 	case BPF_PROG_TYPE_CGROUP_SOCK:
1916 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1917 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1918 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1919 	case BPF_PROG_TYPE_CGROUP_SKB:
1920 		return prog->enforce_expected_attach_type &&
1921 			prog->expected_attach_type != attach_type ?
1922 			-EINVAL : 0;
1923 	default:
1924 		return 0;
1925 	}
1926 }
1927 
1928 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1929 
1930 #define BPF_F_ATTACH_MASK \
1931 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1932 
1933 static int bpf_prog_attach(const union bpf_attr *attr)
1934 {
1935 	enum bpf_prog_type ptype;
1936 	struct bpf_prog *prog;
1937 	int ret;
1938 
1939 	if (!capable(CAP_NET_ADMIN))
1940 		return -EPERM;
1941 
1942 	if (CHECK_ATTR(BPF_PROG_ATTACH))
1943 		return -EINVAL;
1944 
1945 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1946 		return -EINVAL;
1947 
1948 	switch (attr->attach_type) {
1949 	case BPF_CGROUP_INET_INGRESS:
1950 	case BPF_CGROUP_INET_EGRESS:
1951 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
1952 		break;
1953 	case BPF_CGROUP_INET_SOCK_CREATE:
1954 	case BPF_CGROUP_INET4_POST_BIND:
1955 	case BPF_CGROUP_INET6_POST_BIND:
1956 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1957 		break;
1958 	case BPF_CGROUP_INET4_BIND:
1959 	case BPF_CGROUP_INET6_BIND:
1960 	case BPF_CGROUP_INET4_CONNECT:
1961 	case BPF_CGROUP_INET6_CONNECT:
1962 	case BPF_CGROUP_UDP4_SENDMSG:
1963 	case BPF_CGROUP_UDP6_SENDMSG:
1964 	case BPF_CGROUP_UDP4_RECVMSG:
1965 	case BPF_CGROUP_UDP6_RECVMSG:
1966 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1967 		break;
1968 	case BPF_CGROUP_SOCK_OPS:
1969 		ptype = BPF_PROG_TYPE_SOCK_OPS;
1970 		break;
1971 	case BPF_CGROUP_DEVICE:
1972 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1973 		break;
1974 	case BPF_SK_MSG_VERDICT:
1975 		ptype = BPF_PROG_TYPE_SK_MSG;
1976 		break;
1977 	case BPF_SK_SKB_STREAM_PARSER:
1978 	case BPF_SK_SKB_STREAM_VERDICT:
1979 		ptype = BPF_PROG_TYPE_SK_SKB;
1980 		break;
1981 	case BPF_LIRC_MODE2:
1982 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
1983 		break;
1984 	case BPF_FLOW_DISSECTOR:
1985 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1986 		break;
1987 	case BPF_CGROUP_SYSCTL:
1988 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1989 		break;
1990 	case BPF_CGROUP_GETSOCKOPT:
1991 	case BPF_CGROUP_SETSOCKOPT:
1992 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1993 		break;
1994 	default:
1995 		return -EINVAL;
1996 	}
1997 
1998 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1999 	if (IS_ERR(prog))
2000 		return PTR_ERR(prog);
2001 
2002 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2003 		bpf_prog_put(prog);
2004 		return -EINVAL;
2005 	}
2006 
2007 	switch (ptype) {
2008 	case BPF_PROG_TYPE_SK_SKB:
2009 	case BPF_PROG_TYPE_SK_MSG:
2010 		ret = sock_map_get_from_fd(attr, prog);
2011 		break;
2012 	case BPF_PROG_TYPE_LIRC_MODE2:
2013 		ret = lirc_prog_attach(attr, prog);
2014 		break;
2015 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2016 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2017 		break;
2018 	default:
2019 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2020 	}
2021 
2022 	if (ret)
2023 		bpf_prog_put(prog);
2024 	return ret;
2025 }
2026 
2027 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2028 
2029 static int bpf_prog_detach(const union bpf_attr *attr)
2030 {
2031 	enum bpf_prog_type ptype;
2032 
2033 	if (!capable(CAP_NET_ADMIN))
2034 		return -EPERM;
2035 
2036 	if (CHECK_ATTR(BPF_PROG_DETACH))
2037 		return -EINVAL;
2038 
2039 	switch (attr->attach_type) {
2040 	case BPF_CGROUP_INET_INGRESS:
2041 	case BPF_CGROUP_INET_EGRESS:
2042 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2043 		break;
2044 	case BPF_CGROUP_INET_SOCK_CREATE:
2045 	case BPF_CGROUP_INET4_POST_BIND:
2046 	case BPF_CGROUP_INET6_POST_BIND:
2047 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2048 		break;
2049 	case BPF_CGROUP_INET4_BIND:
2050 	case BPF_CGROUP_INET6_BIND:
2051 	case BPF_CGROUP_INET4_CONNECT:
2052 	case BPF_CGROUP_INET6_CONNECT:
2053 	case BPF_CGROUP_UDP4_SENDMSG:
2054 	case BPF_CGROUP_UDP6_SENDMSG:
2055 	case BPF_CGROUP_UDP4_RECVMSG:
2056 	case BPF_CGROUP_UDP6_RECVMSG:
2057 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2058 		break;
2059 	case BPF_CGROUP_SOCK_OPS:
2060 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2061 		break;
2062 	case BPF_CGROUP_DEVICE:
2063 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2064 		break;
2065 	case BPF_SK_MSG_VERDICT:
2066 		return sock_map_get_from_fd(attr, NULL);
2067 	case BPF_SK_SKB_STREAM_PARSER:
2068 	case BPF_SK_SKB_STREAM_VERDICT:
2069 		return sock_map_get_from_fd(attr, NULL);
2070 	case BPF_LIRC_MODE2:
2071 		return lirc_prog_detach(attr);
2072 	case BPF_FLOW_DISSECTOR:
2073 		return skb_flow_dissector_bpf_prog_detach(attr);
2074 	case BPF_CGROUP_SYSCTL:
2075 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2076 		break;
2077 	case BPF_CGROUP_GETSOCKOPT:
2078 	case BPF_CGROUP_SETSOCKOPT:
2079 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2080 		break;
2081 	default:
2082 		return -EINVAL;
2083 	}
2084 
2085 	return cgroup_bpf_prog_detach(attr, ptype);
2086 }
2087 
2088 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2089 
2090 static int bpf_prog_query(const union bpf_attr *attr,
2091 			  union bpf_attr __user *uattr)
2092 {
2093 	if (!capable(CAP_NET_ADMIN))
2094 		return -EPERM;
2095 	if (CHECK_ATTR(BPF_PROG_QUERY))
2096 		return -EINVAL;
2097 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2098 		return -EINVAL;
2099 
2100 	switch (attr->query.attach_type) {
2101 	case BPF_CGROUP_INET_INGRESS:
2102 	case BPF_CGROUP_INET_EGRESS:
2103 	case BPF_CGROUP_INET_SOCK_CREATE:
2104 	case BPF_CGROUP_INET4_BIND:
2105 	case BPF_CGROUP_INET6_BIND:
2106 	case BPF_CGROUP_INET4_POST_BIND:
2107 	case BPF_CGROUP_INET6_POST_BIND:
2108 	case BPF_CGROUP_INET4_CONNECT:
2109 	case BPF_CGROUP_INET6_CONNECT:
2110 	case BPF_CGROUP_UDP4_SENDMSG:
2111 	case BPF_CGROUP_UDP6_SENDMSG:
2112 	case BPF_CGROUP_UDP4_RECVMSG:
2113 	case BPF_CGROUP_UDP6_RECVMSG:
2114 	case BPF_CGROUP_SOCK_OPS:
2115 	case BPF_CGROUP_DEVICE:
2116 	case BPF_CGROUP_SYSCTL:
2117 	case BPF_CGROUP_GETSOCKOPT:
2118 	case BPF_CGROUP_SETSOCKOPT:
2119 		break;
2120 	case BPF_LIRC_MODE2:
2121 		return lirc_prog_query(attr, uattr);
2122 	case BPF_FLOW_DISSECTOR:
2123 		return skb_flow_dissector_prog_query(attr, uattr);
2124 	default:
2125 		return -EINVAL;
2126 	}
2127 
2128 	return cgroup_bpf_prog_query(attr, uattr);
2129 }
2130 
2131 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2132 
2133 static int bpf_prog_test_run(const union bpf_attr *attr,
2134 			     union bpf_attr __user *uattr)
2135 {
2136 	struct bpf_prog *prog;
2137 	int ret = -ENOTSUPP;
2138 
2139 	if (!capable(CAP_SYS_ADMIN))
2140 		return -EPERM;
2141 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2142 		return -EINVAL;
2143 
2144 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2145 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
2146 		return -EINVAL;
2147 
2148 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2149 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
2150 		return -EINVAL;
2151 
2152 	prog = bpf_prog_get(attr->test.prog_fd);
2153 	if (IS_ERR(prog))
2154 		return PTR_ERR(prog);
2155 
2156 	if (prog->aux->ops->test_run)
2157 		ret = prog->aux->ops->test_run(prog, attr, uattr);
2158 
2159 	bpf_prog_put(prog);
2160 	return ret;
2161 }
2162 
2163 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2164 
2165 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2166 			       union bpf_attr __user *uattr,
2167 			       struct idr *idr,
2168 			       spinlock_t *lock)
2169 {
2170 	u32 next_id = attr->start_id;
2171 	int err = 0;
2172 
2173 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2174 		return -EINVAL;
2175 
2176 	if (!capable(CAP_SYS_ADMIN))
2177 		return -EPERM;
2178 
2179 	next_id++;
2180 	spin_lock_bh(lock);
2181 	if (!idr_get_next(idr, &next_id))
2182 		err = -ENOENT;
2183 	spin_unlock_bh(lock);
2184 
2185 	if (!err)
2186 		err = put_user(next_id, &uattr->next_id);
2187 
2188 	return err;
2189 }
2190 
2191 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2192 
2193 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2194 {
2195 	struct bpf_prog *prog;
2196 	u32 id = attr->prog_id;
2197 	int fd;
2198 
2199 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2200 		return -EINVAL;
2201 
2202 	if (!capable(CAP_SYS_ADMIN))
2203 		return -EPERM;
2204 
2205 	spin_lock_bh(&prog_idr_lock);
2206 	prog = idr_find(&prog_idr, id);
2207 	if (prog)
2208 		prog = bpf_prog_inc_not_zero(prog);
2209 	else
2210 		prog = ERR_PTR(-ENOENT);
2211 	spin_unlock_bh(&prog_idr_lock);
2212 
2213 	if (IS_ERR(prog))
2214 		return PTR_ERR(prog);
2215 
2216 	fd = bpf_prog_new_fd(prog);
2217 	if (fd < 0)
2218 		bpf_prog_put(prog);
2219 
2220 	return fd;
2221 }
2222 
2223 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2224 
2225 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2226 {
2227 	struct bpf_map *map;
2228 	u32 id = attr->map_id;
2229 	int f_flags;
2230 	int fd;
2231 
2232 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2233 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2234 		return -EINVAL;
2235 
2236 	if (!capable(CAP_SYS_ADMIN))
2237 		return -EPERM;
2238 
2239 	f_flags = bpf_get_file_flag(attr->open_flags);
2240 	if (f_flags < 0)
2241 		return f_flags;
2242 
2243 	spin_lock_bh(&map_idr_lock);
2244 	map = idr_find(&map_idr, id);
2245 	if (map)
2246 		map = __bpf_map_inc_not_zero(map, true);
2247 	else
2248 		map = ERR_PTR(-ENOENT);
2249 	spin_unlock_bh(&map_idr_lock);
2250 
2251 	if (IS_ERR(map))
2252 		return PTR_ERR(map);
2253 
2254 	fd = bpf_map_new_fd(map, f_flags);
2255 	if (fd < 0)
2256 		bpf_map_put_with_uref(map);
2257 
2258 	return fd;
2259 }
2260 
2261 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2262 					      unsigned long addr, u32 *off,
2263 					      u32 *type)
2264 {
2265 	const struct bpf_map *map;
2266 	int i;
2267 
2268 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2269 		map = prog->aux->used_maps[i];
2270 		if (map == (void *)addr) {
2271 			*type = BPF_PSEUDO_MAP_FD;
2272 			return map;
2273 		}
2274 		if (!map->ops->map_direct_value_meta)
2275 			continue;
2276 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
2277 			*type = BPF_PSEUDO_MAP_VALUE;
2278 			return map;
2279 		}
2280 	}
2281 
2282 	return NULL;
2283 }
2284 
2285 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2286 {
2287 	const struct bpf_map *map;
2288 	struct bpf_insn *insns;
2289 	u32 off, type;
2290 	u64 imm;
2291 	int i;
2292 
2293 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2294 			GFP_USER);
2295 	if (!insns)
2296 		return insns;
2297 
2298 	for (i = 0; i < prog->len; i++) {
2299 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2300 			insns[i].code = BPF_JMP | BPF_CALL;
2301 			insns[i].imm = BPF_FUNC_tail_call;
2302 			/* fall-through */
2303 		}
2304 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2305 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2306 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2307 				insns[i].code = BPF_JMP | BPF_CALL;
2308 			if (!bpf_dump_raw_ok())
2309 				insns[i].imm = 0;
2310 			continue;
2311 		}
2312 
2313 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2314 			continue;
2315 
2316 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2317 		map = bpf_map_from_imm(prog, imm, &off, &type);
2318 		if (map) {
2319 			insns[i].src_reg = type;
2320 			insns[i].imm = map->id;
2321 			insns[i + 1].imm = off;
2322 			continue;
2323 		}
2324 	}
2325 
2326 	return insns;
2327 }
2328 
2329 static int set_info_rec_size(struct bpf_prog_info *info)
2330 {
2331 	/*
2332 	 * Ensure info.*_rec_size is the same as kernel expected size
2333 	 *
2334 	 * or
2335 	 *
2336 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2337 	 * zero.  In this case, the kernel will set the expected
2338 	 * _rec_size back to the info.
2339 	 */
2340 
2341 	if ((info->nr_func_info || info->func_info_rec_size) &&
2342 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2343 		return -EINVAL;
2344 
2345 	if ((info->nr_line_info || info->line_info_rec_size) &&
2346 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2347 		return -EINVAL;
2348 
2349 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2350 	    info->jited_line_info_rec_size != sizeof(__u64))
2351 		return -EINVAL;
2352 
2353 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2354 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2355 	info->jited_line_info_rec_size = sizeof(__u64);
2356 
2357 	return 0;
2358 }
2359 
2360 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2361 				   const union bpf_attr *attr,
2362 				   union bpf_attr __user *uattr)
2363 {
2364 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2365 	struct bpf_prog_info info = {};
2366 	u32 info_len = attr->info.info_len;
2367 	struct bpf_prog_stats stats;
2368 	char __user *uinsns;
2369 	u32 ulen;
2370 	int err;
2371 
2372 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2373 	if (err)
2374 		return err;
2375 	info_len = min_t(u32, sizeof(info), info_len);
2376 
2377 	if (copy_from_user(&info, uinfo, info_len))
2378 		return -EFAULT;
2379 
2380 	info.type = prog->type;
2381 	info.id = prog->aux->id;
2382 	info.load_time = prog->aux->load_time;
2383 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2384 					       prog->aux->user->uid);
2385 	info.gpl_compatible = prog->gpl_compatible;
2386 
2387 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2388 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2389 
2390 	ulen = info.nr_map_ids;
2391 	info.nr_map_ids = prog->aux->used_map_cnt;
2392 	ulen = min_t(u32, info.nr_map_ids, ulen);
2393 	if (ulen) {
2394 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2395 		u32 i;
2396 
2397 		for (i = 0; i < ulen; i++)
2398 			if (put_user(prog->aux->used_maps[i]->id,
2399 				     &user_map_ids[i]))
2400 				return -EFAULT;
2401 	}
2402 
2403 	err = set_info_rec_size(&info);
2404 	if (err)
2405 		return err;
2406 
2407 	bpf_prog_get_stats(prog, &stats);
2408 	info.run_time_ns = stats.nsecs;
2409 	info.run_cnt = stats.cnt;
2410 
2411 	if (!capable(CAP_SYS_ADMIN)) {
2412 		info.jited_prog_len = 0;
2413 		info.xlated_prog_len = 0;
2414 		info.nr_jited_ksyms = 0;
2415 		info.nr_jited_func_lens = 0;
2416 		info.nr_func_info = 0;
2417 		info.nr_line_info = 0;
2418 		info.nr_jited_line_info = 0;
2419 		goto done;
2420 	}
2421 
2422 	ulen = info.xlated_prog_len;
2423 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2424 	if (info.xlated_prog_len && ulen) {
2425 		struct bpf_insn *insns_sanitized;
2426 		bool fault;
2427 
2428 		if (prog->blinded && !bpf_dump_raw_ok()) {
2429 			info.xlated_prog_insns = 0;
2430 			goto done;
2431 		}
2432 		insns_sanitized = bpf_insn_prepare_dump(prog);
2433 		if (!insns_sanitized)
2434 			return -ENOMEM;
2435 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2436 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2437 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2438 		kfree(insns_sanitized);
2439 		if (fault)
2440 			return -EFAULT;
2441 	}
2442 
2443 	if (bpf_prog_is_dev_bound(prog->aux)) {
2444 		err = bpf_prog_offload_info_fill(&info, prog);
2445 		if (err)
2446 			return err;
2447 		goto done;
2448 	}
2449 
2450 	/* NOTE: the following code is supposed to be skipped for offload.
2451 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2452 	 * for offload.
2453 	 */
2454 	ulen = info.jited_prog_len;
2455 	if (prog->aux->func_cnt) {
2456 		u32 i;
2457 
2458 		info.jited_prog_len = 0;
2459 		for (i = 0; i < prog->aux->func_cnt; i++)
2460 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2461 	} else {
2462 		info.jited_prog_len = prog->jited_len;
2463 	}
2464 
2465 	if (info.jited_prog_len && ulen) {
2466 		if (bpf_dump_raw_ok()) {
2467 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2468 			ulen = min_t(u32, info.jited_prog_len, ulen);
2469 
2470 			/* for multi-function programs, copy the JITed
2471 			 * instructions for all the functions
2472 			 */
2473 			if (prog->aux->func_cnt) {
2474 				u32 len, free, i;
2475 				u8 *img;
2476 
2477 				free = ulen;
2478 				for (i = 0; i < prog->aux->func_cnt; i++) {
2479 					len = prog->aux->func[i]->jited_len;
2480 					len = min_t(u32, len, free);
2481 					img = (u8 *) prog->aux->func[i]->bpf_func;
2482 					if (copy_to_user(uinsns, img, len))
2483 						return -EFAULT;
2484 					uinsns += len;
2485 					free -= len;
2486 					if (!free)
2487 						break;
2488 				}
2489 			} else {
2490 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2491 					return -EFAULT;
2492 			}
2493 		} else {
2494 			info.jited_prog_insns = 0;
2495 		}
2496 	}
2497 
2498 	ulen = info.nr_jited_ksyms;
2499 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2500 	if (ulen) {
2501 		if (bpf_dump_raw_ok()) {
2502 			unsigned long ksym_addr;
2503 			u64 __user *user_ksyms;
2504 			u32 i;
2505 
2506 			/* copy the address of the kernel symbol
2507 			 * corresponding to each function
2508 			 */
2509 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2510 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2511 			if (prog->aux->func_cnt) {
2512 				for (i = 0; i < ulen; i++) {
2513 					ksym_addr = (unsigned long)
2514 						prog->aux->func[i]->bpf_func;
2515 					if (put_user((u64) ksym_addr,
2516 						     &user_ksyms[i]))
2517 						return -EFAULT;
2518 				}
2519 			} else {
2520 				ksym_addr = (unsigned long) prog->bpf_func;
2521 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2522 					return -EFAULT;
2523 			}
2524 		} else {
2525 			info.jited_ksyms = 0;
2526 		}
2527 	}
2528 
2529 	ulen = info.nr_jited_func_lens;
2530 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2531 	if (ulen) {
2532 		if (bpf_dump_raw_ok()) {
2533 			u32 __user *user_lens;
2534 			u32 func_len, i;
2535 
2536 			/* copy the JITed image lengths for each function */
2537 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2538 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2539 			if (prog->aux->func_cnt) {
2540 				for (i = 0; i < ulen; i++) {
2541 					func_len =
2542 						prog->aux->func[i]->jited_len;
2543 					if (put_user(func_len, &user_lens[i]))
2544 						return -EFAULT;
2545 				}
2546 			} else {
2547 				func_len = prog->jited_len;
2548 				if (put_user(func_len, &user_lens[0]))
2549 					return -EFAULT;
2550 			}
2551 		} else {
2552 			info.jited_func_lens = 0;
2553 		}
2554 	}
2555 
2556 	if (prog->aux->btf)
2557 		info.btf_id = btf_id(prog->aux->btf);
2558 
2559 	ulen = info.nr_func_info;
2560 	info.nr_func_info = prog->aux->func_info_cnt;
2561 	if (info.nr_func_info && ulen) {
2562 		char __user *user_finfo;
2563 
2564 		user_finfo = u64_to_user_ptr(info.func_info);
2565 		ulen = min_t(u32, info.nr_func_info, ulen);
2566 		if (copy_to_user(user_finfo, prog->aux->func_info,
2567 				 info.func_info_rec_size * ulen))
2568 			return -EFAULT;
2569 	}
2570 
2571 	ulen = info.nr_line_info;
2572 	info.nr_line_info = prog->aux->nr_linfo;
2573 	if (info.nr_line_info && ulen) {
2574 		__u8 __user *user_linfo;
2575 
2576 		user_linfo = u64_to_user_ptr(info.line_info);
2577 		ulen = min_t(u32, info.nr_line_info, ulen);
2578 		if (copy_to_user(user_linfo, prog->aux->linfo,
2579 				 info.line_info_rec_size * ulen))
2580 			return -EFAULT;
2581 	}
2582 
2583 	ulen = info.nr_jited_line_info;
2584 	if (prog->aux->jited_linfo)
2585 		info.nr_jited_line_info = prog->aux->nr_linfo;
2586 	else
2587 		info.nr_jited_line_info = 0;
2588 	if (info.nr_jited_line_info && ulen) {
2589 		if (bpf_dump_raw_ok()) {
2590 			__u64 __user *user_linfo;
2591 			u32 i;
2592 
2593 			user_linfo = u64_to_user_ptr(info.jited_line_info);
2594 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
2595 			for (i = 0; i < ulen; i++) {
2596 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2597 					     &user_linfo[i]))
2598 					return -EFAULT;
2599 			}
2600 		} else {
2601 			info.jited_line_info = 0;
2602 		}
2603 	}
2604 
2605 	ulen = info.nr_prog_tags;
2606 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2607 	if (ulen) {
2608 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2609 		u32 i;
2610 
2611 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
2612 		ulen = min_t(u32, info.nr_prog_tags, ulen);
2613 		if (prog->aux->func_cnt) {
2614 			for (i = 0; i < ulen; i++) {
2615 				if (copy_to_user(user_prog_tags[i],
2616 						 prog->aux->func[i]->tag,
2617 						 BPF_TAG_SIZE))
2618 					return -EFAULT;
2619 			}
2620 		} else {
2621 			if (copy_to_user(user_prog_tags[0],
2622 					 prog->tag, BPF_TAG_SIZE))
2623 				return -EFAULT;
2624 		}
2625 	}
2626 
2627 done:
2628 	if (copy_to_user(uinfo, &info, info_len) ||
2629 	    put_user(info_len, &uattr->info.info_len))
2630 		return -EFAULT;
2631 
2632 	return 0;
2633 }
2634 
2635 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2636 				  const union bpf_attr *attr,
2637 				  union bpf_attr __user *uattr)
2638 {
2639 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2640 	struct bpf_map_info info = {};
2641 	u32 info_len = attr->info.info_len;
2642 	int err;
2643 
2644 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2645 	if (err)
2646 		return err;
2647 	info_len = min_t(u32, sizeof(info), info_len);
2648 
2649 	info.type = map->map_type;
2650 	info.id = map->id;
2651 	info.key_size = map->key_size;
2652 	info.value_size = map->value_size;
2653 	info.max_entries = map->max_entries;
2654 	info.map_flags = map->map_flags;
2655 	memcpy(info.name, map->name, sizeof(map->name));
2656 
2657 	if (map->btf) {
2658 		info.btf_id = btf_id(map->btf);
2659 		info.btf_key_type_id = map->btf_key_type_id;
2660 		info.btf_value_type_id = map->btf_value_type_id;
2661 	}
2662 
2663 	if (bpf_map_is_dev_bound(map)) {
2664 		err = bpf_map_offload_info_fill(&info, map);
2665 		if (err)
2666 			return err;
2667 	}
2668 
2669 	if (copy_to_user(uinfo, &info, info_len) ||
2670 	    put_user(info_len, &uattr->info.info_len))
2671 		return -EFAULT;
2672 
2673 	return 0;
2674 }
2675 
2676 static int bpf_btf_get_info_by_fd(struct btf *btf,
2677 				  const union bpf_attr *attr,
2678 				  union bpf_attr __user *uattr)
2679 {
2680 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2681 	u32 info_len = attr->info.info_len;
2682 	int err;
2683 
2684 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2685 	if (err)
2686 		return err;
2687 
2688 	return btf_get_info_by_fd(btf, attr, uattr);
2689 }
2690 
2691 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2692 
2693 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2694 				  union bpf_attr __user *uattr)
2695 {
2696 	int ufd = attr->info.bpf_fd;
2697 	struct fd f;
2698 	int err;
2699 
2700 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2701 		return -EINVAL;
2702 
2703 	f = fdget(ufd);
2704 	if (!f.file)
2705 		return -EBADFD;
2706 
2707 	if (f.file->f_op == &bpf_prog_fops)
2708 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2709 					      uattr);
2710 	else if (f.file->f_op == &bpf_map_fops)
2711 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2712 					     uattr);
2713 	else if (f.file->f_op == &btf_fops)
2714 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2715 	else
2716 		err = -EINVAL;
2717 
2718 	fdput(f);
2719 	return err;
2720 }
2721 
2722 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2723 
2724 static int bpf_btf_load(const union bpf_attr *attr)
2725 {
2726 	if (CHECK_ATTR(BPF_BTF_LOAD))
2727 		return -EINVAL;
2728 
2729 	if (!capable(CAP_SYS_ADMIN))
2730 		return -EPERM;
2731 
2732 	return btf_new_fd(attr);
2733 }
2734 
2735 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2736 
2737 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2738 {
2739 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2740 		return -EINVAL;
2741 
2742 	if (!capable(CAP_SYS_ADMIN))
2743 		return -EPERM;
2744 
2745 	return btf_get_fd_by_id(attr->btf_id);
2746 }
2747 
2748 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2749 				    union bpf_attr __user *uattr,
2750 				    u32 prog_id, u32 fd_type,
2751 				    const char *buf, u64 probe_offset,
2752 				    u64 probe_addr)
2753 {
2754 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2755 	u32 len = buf ? strlen(buf) : 0, input_len;
2756 	int err = 0;
2757 
2758 	if (put_user(len, &uattr->task_fd_query.buf_len))
2759 		return -EFAULT;
2760 	input_len = attr->task_fd_query.buf_len;
2761 	if (input_len && ubuf) {
2762 		if (!len) {
2763 			/* nothing to copy, just make ubuf NULL terminated */
2764 			char zero = '\0';
2765 
2766 			if (put_user(zero, ubuf))
2767 				return -EFAULT;
2768 		} else if (input_len >= len + 1) {
2769 			/* ubuf can hold the string with NULL terminator */
2770 			if (copy_to_user(ubuf, buf, len + 1))
2771 				return -EFAULT;
2772 		} else {
2773 			/* ubuf cannot hold the string with NULL terminator,
2774 			 * do a partial copy with NULL terminator.
2775 			 */
2776 			char zero = '\0';
2777 
2778 			err = -ENOSPC;
2779 			if (copy_to_user(ubuf, buf, input_len - 1))
2780 				return -EFAULT;
2781 			if (put_user(zero, ubuf + input_len - 1))
2782 				return -EFAULT;
2783 		}
2784 	}
2785 
2786 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2787 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2788 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2789 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2790 		return -EFAULT;
2791 
2792 	return err;
2793 }
2794 
2795 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2796 
2797 static int bpf_task_fd_query(const union bpf_attr *attr,
2798 			     union bpf_attr __user *uattr)
2799 {
2800 	pid_t pid = attr->task_fd_query.pid;
2801 	u32 fd = attr->task_fd_query.fd;
2802 	const struct perf_event *event;
2803 	struct files_struct *files;
2804 	struct task_struct *task;
2805 	struct file *file;
2806 	int err;
2807 
2808 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2809 		return -EINVAL;
2810 
2811 	if (!capable(CAP_SYS_ADMIN))
2812 		return -EPERM;
2813 
2814 	if (attr->task_fd_query.flags != 0)
2815 		return -EINVAL;
2816 
2817 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2818 	if (!task)
2819 		return -ENOENT;
2820 
2821 	files = get_files_struct(task);
2822 	put_task_struct(task);
2823 	if (!files)
2824 		return -ENOENT;
2825 
2826 	err = 0;
2827 	spin_lock(&files->file_lock);
2828 	file = fcheck_files(files, fd);
2829 	if (!file)
2830 		err = -EBADF;
2831 	else
2832 		get_file(file);
2833 	spin_unlock(&files->file_lock);
2834 	put_files_struct(files);
2835 
2836 	if (err)
2837 		goto out;
2838 
2839 	if (file->f_op == &bpf_raw_tp_fops) {
2840 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
2841 		struct bpf_raw_event_map *btp = raw_tp->btp;
2842 
2843 		err = bpf_task_fd_query_copy(attr, uattr,
2844 					     raw_tp->prog->aux->id,
2845 					     BPF_FD_TYPE_RAW_TRACEPOINT,
2846 					     btp->tp->name, 0, 0);
2847 		goto put_file;
2848 	}
2849 
2850 	event = perf_get_event(file);
2851 	if (!IS_ERR(event)) {
2852 		u64 probe_offset, probe_addr;
2853 		u32 prog_id, fd_type;
2854 		const char *buf;
2855 
2856 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2857 					      &buf, &probe_offset,
2858 					      &probe_addr);
2859 		if (!err)
2860 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2861 						     fd_type, buf,
2862 						     probe_offset,
2863 						     probe_addr);
2864 		goto put_file;
2865 	}
2866 
2867 	err = -ENOTSUPP;
2868 put_file:
2869 	fput(file);
2870 out:
2871 	return err;
2872 }
2873 
2874 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2875 {
2876 	union bpf_attr attr = {};
2877 	int err;
2878 
2879 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2880 		return -EPERM;
2881 
2882 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2883 	if (err)
2884 		return err;
2885 	size = min_t(u32, size, sizeof(attr));
2886 
2887 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
2888 	if (copy_from_user(&attr, uattr, size) != 0)
2889 		return -EFAULT;
2890 
2891 	err = security_bpf(cmd, &attr, size);
2892 	if (err < 0)
2893 		return err;
2894 
2895 	switch (cmd) {
2896 	case BPF_MAP_CREATE:
2897 		err = map_create(&attr);
2898 		break;
2899 	case BPF_MAP_LOOKUP_ELEM:
2900 		err = map_lookup_elem(&attr);
2901 		break;
2902 	case BPF_MAP_UPDATE_ELEM:
2903 		err = map_update_elem(&attr);
2904 		break;
2905 	case BPF_MAP_DELETE_ELEM:
2906 		err = map_delete_elem(&attr);
2907 		break;
2908 	case BPF_MAP_GET_NEXT_KEY:
2909 		err = map_get_next_key(&attr);
2910 		break;
2911 	case BPF_MAP_FREEZE:
2912 		err = map_freeze(&attr);
2913 		break;
2914 	case BPF_PROG_LOAD:
2915 		err = bpf_prog_load(&attr, uattr);
2916 		break;
2917 	case BPF_OBJ_PIN:
2918 		err = bpf_obj_pin(&attr);
2919 		break;
2920 	case BPF_OBJ_GET:
2921 		err = bpf_obj_get(&attr);
2922 		break;
2923 	case BPF_PROG_ATTACH:
2924 		err = bpf_prog_attach(&attr);
2925 		break;
2926 	case BPF_PROG_DETACH:
2927 		err = bpf_prog_detach(&attr);
2928 		break;
2929 	case BPF_PROG_QUERY:
2930 		err = bpf_prog_query(&attr, uattr);
2931 		break;
2932 	case BPF_PROG_TEST_RUN:
2933 		err = bpf_prog_test_run(&attr, uattr);
2934 		break;
2935 	case BPF_PROG_GET_NEXT_ID:
2936 		err = bpf_obj_get_next_id(&attr, uattr,
2937 					  &prog_idr, &prog_idr_lock);
2938 		break;
2939 	case BPF_MAP_GET_NEXT_ID:
2940 		err = bpf_obj_get_next_id(&attr, uattr,
2941 					  &map_idr, &map_idr_lock);
2942 		break;
2943 	case BPF_BTF_GET_NEXT_ID:
2944 		err = bpf_obj_get_next_id(&attr, uattr,
2945 					  &btf_idr, &btf_idr_lock);
2946 		break;
2947 	case BPF_PROG_GET_FD_BY_ID:
2948 		err = bpf_prog_get_fd_by_id(&attr);
2949 		break;
2950 	case BPF_MAP_GET_FD_BY_ID:
2951 		err = bpf_map_get_fd_by_id(&attr);
2952 		break;
2953 	case BPF_OBJ_GET_INFO_BY_FD:
2954 		err = bpf_obj_get_info_by_fd(&attr, uattr);
2955 		break;
2956 	case BPF_RAW_TRACEPOINT_OPEN:
2957 		err = bpf_raw_tracepoint_open(&attr);
2958 		break;
2959 	case BPF_BTF_LOAD:
2960 		err = bpf_btf_load(&attr);
2961 		break;
2962 	case BPF_BTF_GET_FD_BY_ID:
2963 		err = bpf_btf_get_fd_by_id(&attr);
2964 		break;
2965 	case BPF_TASK_FD_QUERY:
2966 		err = bpf_task_fd_query(&attr, uattr);
2967 		break;
2968 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2969 		err = map_lookup_and_delete_elem(&attr);
2970 		break;
2971 	default:
2972 		err = -EINVAL;
2973 		break;
2974 	}
2975 
2976 	return err;
2977 }
2978