xref: /linux/kernel/bpf/syscall.c (revision 9d106c6dd81bb26ad7fc3ee89cb1d62557c8e2c9)
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 <linux/audit.h>
27 #include <uapi/linux/btf.h>
28 
29 #define IS_FD_ARRAY(map) ((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_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
33 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
34 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
35 			IS_FD_HASH(map))
36 
37 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
38 
39 DEFINE_PER_CPU(int, bpf_prog_active);
40 static DEFINE_IDR(prog_idr);
41 static DEFINE_SPINLOCK(prog_idr_lock);
42 static DEFINE_IDR(map_idr);
43 static DEFINE_SPINLOCK(map_idr_lock);
44 
45 int sysctl_unprivileged_bpf_disabled __read_mostly;
46 
47 static const struct bpf_map_ops * const bpf_map_types[] = {
48 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
49 #define BPF_MAP_TYPE(_id, _ops) \
50 	[_id] = &_ops,
51 #include <linux/bpf_types.h>
52 #undef BPF_PROG_TYPE
53 #undef BPF_MAP_TYPE
54 };
55 
56 /*
57  * If we're handed a bigger struct than we know of, ensure all the unknown bits
58  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
59  * we don't know about yet.
60  *
61  * There is a ToCToU between this function call and the following
62  * copy_from_user() call. However, this is not a concern since this function is
63  * meant to be a future-proofing of bits.
64  */
65 int bpf_check_uarg_tail_zero(void __user *uaddr,
66 			     size_t expected_size,
67 			     size_t actual_size)
68 {
69 	unsigned char __user *addr;
70 	unsigned char __user *end;
71 	unsigned char val;
72 	int err;
73 
74 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
75 		return -E2BIG;
76 
77 	if (unlikely(!access_ok(uaddr, actual_size)))
78 		return -EFAULT;
79 
80 	if (actual_size <= expected_size)
81 		return 0;
82 
83 	addr = uaddr + expected_size;
84 	end  = uaddr + actual_size;
85 
86 	for (; addr < end; addr++) {
87 		err = get_user(val, addr);
88 		if (err)
89 			return err;
90 		if (val)
91 			return -E2BIG;
92 	}
93 
94 	return 0;
95 }
96 
97 const struct bpf_map_ops bpf_map_offload_ops = {
98 	.map_alloc = bpf_map_offload_map_alloc,
99 	.map_free = bpf_map_offload_map_free,
100 	.map_check_btf = map_check_no_btf,
101 };
102 
103 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
104 {
105 	const struct bpf_map_ops *ops;
106 	u32 type = attr->map_type;
107 	struct bpf_map *map;
108 	int err;
109 
110 	if (type >= ARRAY_SIZE(bpf_map_types))
111 		return ERR_PTR(-EINVAL);
112 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
113 	ops = bpf_map_types[type];
114 	if (!ops)
115 		return ERR_PTR(-EINVAL);
116 
117 	if (ops->map_alloc_check) {
118 		err = ops->map_alloc_check(attr);
119 		if (err)
120 			return ERR_PTR(err);
121 	}
122 	if (attr->map_ifindex)
123 		ops = &bpf_map_offload_ops;
124 	map = ops->map_alloc(attr);
125 	if (IS_ERR(map))
126 		return map;
127 	map->ops = ops;
128 	map->map_type = type;
129 	return map;
130 }
131 
132 static u32 bpf_map_value_size(struct bpf_map *map)
133 {
134 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
135 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
136 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
137 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
138 		return round_up(map->value_size, 8) * num_possible_cpus();
139 	else if (IS_FD_MAP(map))
140 		return sizeof(u32);
141 	else
142 		return  map->value_size;
143 }
144 
145 static void maybe_wait_bpf_programs(struct bpf_map *map)
146 {
147 	/* Wait for any running BPF programs to complete so that
148 	 * userspace, when we return to it, knows that all programs
149 	 * that could be running use the new map value.
150 	 */
151 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
152 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
153 		synchronize_rcu();
154 }
155 
156 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
157 				void *value, __u64 flags)
158 {
159 	int err;
160 
161 	/* Need to create a kthread, thus must support schedule */
162 	if (bpf_map_is_dev_bound(map)) {
163 		return bpf_map_offload_update_elem(map, key, value, flags);
164 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
165 		   map->map_type == BPF_MAP_TYPE_SOCKHASH ||
166 		   map->map_type == BPF_MAP_TYPE_SOCKMAP ||
167 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
168 		return map->ops->map_update_elem(map, key, value, flags);
169 	} else if (IS_FD_PROG_ARRAY(map)) {
170 		return bpf_fd_array_map_update_elem(map, f.file, key, value,
171 						    flags);
172 	}
173 
174 	bpf_disable_instrumentation();
175 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
176 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
177 		err = bpf_percpu_hash_update(map, key, value, flags);
178 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
179 		err = bpf_percpu_array_update(map, key, value, flags);
180 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
181 		err = bpf_percpu_cgroup_storage_update(map, key, value,
182 						       flags);
183 	} else if (IS_FD_ARRAY(map)) {
184 		rcu_read_lock();
185 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
186 						   flags);
187 		rcu_read_unlock();
188 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
189 		rcu_read_lock();
190 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
191 						  flags);
192 		rcu_read_unlock();
193 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
194 		/* rcu_read_lock() is not needed */
195 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
196 							 flags);
197 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
198 		   map->map_type == BPF_MAP_TYPE_STACK) {
199 		err = map->ops->map_push_elem(map, value, flags);
200 	} else {
201 		rcu_read_lock();
202 		err = map->ops->map_update_elem(map, key, value, flags);
203 		rcu_read_unlock();
204 	}
205 	bpf_enable_instrumentation();
206 	maybe_wait_bpf_programs(map);
207 
208 	return err;
209 }
210 
211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
212 			      __u64 flags)
213 {
214 	void *ptr;
215 	int err;
216 
217 	if (bpf_map_is_dev_bound(map))
218 		return bpf_map_offload_lookup_elem(map, key, value);
219 
220 	bpf_disable_instrumentation();
221 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
222 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
223 		err = bpf_percpu_hash_copy(map, key, value);
224 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
225 		err = bpf_percpu_array_copy(map, key, value);
226 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
227 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
228 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
229 		err = bpf_stackmap_copy(map, key, value);
230 	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
231 		err = bpf_fd_array_map_lookup_elem(map, key, value);
232 	} else if (IS_FD_HASH(map)) {
233 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
234 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
235 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
236 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
237 		   map->map_type == BPF_MAP_TYPE_STACK) {
238 		err = map->ops->map_peek_elem(map, value);
239 	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
240 		/* struct_ops map requires directly updating "value" */
241 		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
242 	} else {
243 		rcu_read_lock();
244 		if (map->ops->map_lookup_elem_sys_only)
245 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
246 		else
247 			ptr = map->ops->map_lookup_elem(map, key);
248 		if (IS_ERR(ptr)) {
249 			err = PTR_ERR(ptr);
250 		} else if (!ptr) {
251 			err = -ENOENT;
252 		} else {
253 			err = 0;
254 			if (flags & BPF_F_LOCK)
255 				/* lock 'ptr' and copy everything but lock */
256 				copy_map_value_locked(map, value, ptr, true);
257 			else
258 				copy_map_value(map, value, ptr);
259 			/* mask lock, since value wasn't zero inited */
260 			check_and_init_map_lock(map, value);
261 		}
262 		rcu_read_unlock();
263 	}
264 
265 	bpf_enable_instrumentation();
266 	maybe_wait_bpf_programs(map);
267 
268 	return err;
269 }
270 
271 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
272 {
273 	/* We really just want to fail instead of triggering OOM killer
274 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
275 	 * which is used for lower order allocation requests.
276 	 *
277 	 * It has been observed that higher order allocation requests done by
278 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
279 	 * to reclaim memory from the page cache, thus we set
280 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
281 	 */
282 
283 	const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
284 	void *area;
285 
286 	if (size >= SIZE_MAX)
287 		return NULL;
288 
289 	/* kmalloc()'ed memory can't be mmap()'ed */
290 	if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
291 		area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
292 				    numa_node);
293 		if (area != NULL)
294 			return area;
295 	}
296 	if (mmapable) {
297 		BUG_ON(!PAGE_ALIGNED(size));
298 		return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
299 					       __GFP_RETRY_MAYFAIL | flags);
300 	}
301 	return __vmalloc_node_flags_caller(size, numa_node,
302 					   GFP_KERNEL | __GFP_RETRY_MAYFAIL |
303 					   flags, __builtin_return_address(0));
304 }
305 
306 void *bpf_map_area_alloc(u64 size, int numa_node)
307 {
308 	return __bpf_map_area_alloc(size, numa_node, false);
309 }
310 
311 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
312 {
313 	return __bpf_map_area_alloc(size, numa_node, true);
314 }
315 
316 void bpf_map_area_free(void *area)
317 {
318 	kvfree(area);
319 }
320 
321 static u32 bpf_map_flags_retain_permanent(u32 flags)
322 {
323 	/* Some map creation flags are not tied to the map object but
324 	 * rather to the map fd instead, so they have no meaning upon
325 	 * map object inspection since multiple file descriptors with
326 	 * different (access) properties can exist here. Thus, given
327 	 * this has zero meaning for the map itself, lets clear these
328 	 * from here.
329 	 */
330 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
331 }
332 
333 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
334 {
335 	map->map_type = attr->map_type;
336 	map->key_size = attr->key_size;
337 	map->value_size = attr->value_size;
338 	map->max_entries = attr->max_entries;
339 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
340 	map->numa_node = bpf_map_attr_numa_node(attr);
341 }
342 
343 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
344 {
345 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
346 
347 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
348 		atomic_long_sub(pages, &user->locked_vm);
349 		return -EPERM;
350 	}
351 	return 0;
352 }
353 
354 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
355 {
356 	if (user)
357 		atomic_long_sub(pages, &user->locked_vm);
358 }
359 
360 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
361 {
362 	u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
363 	struct user_struct *user;
364 	int ret;
365 
366 	if (size >= U32_MAX - PAGE_SIZE)
367 		return -E2BIG;
368 
369 	user = get_current_user();
370 	ret = bpf_charge_memlock(user, pages);
371 	if (ret) {
372 		free_uid(user);
373 		return ret;
374 	}
375 
376 	mem->pages = pages;
377 	mem->user = user;
378 
379 	return 0;
380 }
381 
382 void bpf_map_charge_finish(struct bpf_map_memory *mem)
383 {
384 	bpf_uncharge_memlock(mem->user, mem->pages);
385 	free_uid(mem->user);
386 }
387 
388 void bpf_map_charge_move(struct bpf_map_memory *dst,
389 			 struct bpf_map_memory *src)
390 {
391 	*dst = *src;
392 
393 	/* Make sure src will not be used for the redundant uncharging. */
394 	memset(src, 0, sizeof(struct bpf_map_memory));
395 }
396 
397 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
398 {
399 	int ret;
400 
401 	ret = bpf_charge_memlock(map->memory.user, pages);
402 	if (ret)
403 		return ret;
404 	map->memory.pages += pages;
405 	return ret;
406 }
407 
408 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
409 {
410 	bpf_uncharge_memlock(map->memory.user, pages);
411 	map->memory.pages -= pages;
412 }
413 
414 static int bpf_map_alloc_id(struct bpf_map *map)
415 {
416 	int id;
417 
418 	idr_preload(GFP_KERNEL);
419 	spin_lock_bh(&map_idr_lock);
420 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
421 	if (id > 0)
422 		map->id = id;
423 	spin_unlock_bh(&map_idr_lock);
424 	idr_preload_end();
425 
426 	if (WARN_ON_ONCE(!id))
427 		return -ENOSPC;
428 
429 	return id > 0 ? 0 : id;
430 }
431 
432 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
433 {
434 	unsigned long flags;
435 
436 	/* Offloaded maps are removed from the IDR store when their device
437 	 * disappears - even if someone holds an fd to them they are unusable,
438 	 * the memory is gone, all ops will fail; they are simply waiting for
439 	 * refcnt to drop to be freed.
440 	 */
441 	if (!map->id)
442 		return;
443 
444 	if (do_idr_lock)
445 		spin_lock_irqsave(&map_idr_lock, flags);
446 	else
447 		__acquire(&map_idr_lock);
448 
449 	idr_remove(&map_idr, map->id);
450 	map->id = 0;
451 
452 	if (do_idr_lock)
453 		spin_unlock_irqrestore(&map_idr_lock, flags);
454 	else
455 		__release(&map_idr_lock);
456 }
457 
458 /* called from workqueue */
459 static void bpf_map_free_deferred(struct work_struct *work)
460 {
461 	struct bpf_map *map = container_of(work, struct bpf_map, work);
462 	struct bpf_map_memory mem;
463 
464 	bpf_map_charge_move(&mem, &map->memory);
465 	security_bpf_map_free(map);
466 	/* implementation dependent freeing */
467 	map->ops->map_free(map);
468 	bpf_map_charge_finish(&mem);
469 }
470 
471 static void bpf_map_put_uref(struct bpf_map *map)
472 {
473 	if (atomic64_dec_and_test(&map->usercnt)) {
474 		if (map->ops->map_release_uref)
475 			map->ops->map_release_uref(map);
476 	}
477 }
478 
479 /* decrement map refcnt and schedule it for freeing via workqueue
480  * (unrelying map implementation ops->map_free() might sleep)
481  */
482 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
483 {
484 	if (atomic64_dec_and_test(&map->refcnt)) {
485 		/* bpf_map_free_id() must be called first */
486 		bpf_map_free_id(map, do_idr_lock);
487 		btf_put(map->btf);
488 		INIT_WORK(&map->work, bpf_map_free_deferred);
489 		schedule_work(&map->work);
490 	}
491 }
492 
493 void bpf_map_put(struct bpf_map *map)
494 {
495 	__bpf_map_put(map, true);
496 }
497 EXPORT_SYMBOL_GPL(bpf_map_put);
498 
499 void bpf_map_put_with_uref(struct bpf_map *map)
500 {
501 	bpf_map_put_uref(map);
502 	bpf_map_put(map);
503 }
504 
505 static int bpf_map_release(struct inode *inode, struct file *filp)
506 {
507 	struct bpf_map *map = filp->private_data;
508 
509 	if (map->ops->map_release)
510 		map->ops->map_release(map, filp);
511 
512 	bpf_map_put_with_uref(map);
513 	return 0;
514 }
515 
516 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
517 {
518 	fmode_t mode = f.file->f_mode;
519 
520 	/* Our file permissions may have been overridden by global
521 	 * map permissions facing syscall side.
522 	 */
523 	if (READ_ONCE(map->frozen))
524 		mode &= ~FMODE_CAN_WRITE;
525 	return mode;
526 }
527 
528 #ifdef CONFIG_PROC_FS
529 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
530 {
531 	const struct bpf_map *map = filp->private_data;
532 	const struct bpf_array *array;
533 	u32 type = 0, jited = 0;
534 
535 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
536 		array = container_of(map, struct bpf_array, map);
537 		type  = array->aux->type;
538 		jited = array->aux->jited;
539 	}
540 
541 	seq_printf(m,
542 		   "map_type:\t%u\n"
543 		   "key_size:\t%u\n"
544 		   "value_size:\t%u\n"
545 		   "max_entries:\t%u\n"
546 		   "map_flags:\t%#x\n"
547 		   "memlock:\t%llu\n"
548 		   "map_id:\t%u\n"
549 		   "frozen:\t%u\n",
550 		   map->map_type,
551 		   map->key_size,
552 		   map->value_size,
553 		   map->max_entries,
554 		   map->map_flags,
555 		   map->memory.pages * 1ULL << PAGE_SHIFT,
556 		   map->id,
557 		   READ_ONCE(map->frozen));
558 	if (type) {
559 		seq_printf(m, "owner_prog_type:\t%u\n", type);
560 		seq_printf(m, "owner_jited:\t%u\n", jited);
561 	}
562 }
563 #endif
564 
565 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
566 			      loff_t *ppos)
567 {
568 	/* We need this handler such that alloc_file() enables
569 	 * f_mode with FMODE_CAN_READ.
570 	 */
571 	return -EINVAL;
572 }
573 
574 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
575 			       size_t siz, loff_t *ppos)
576 {
577 	/* We need this handler such that alloc_file() enables
578 	 * f_mode with FMODE_CAN_WRITE.
579 	 */
580 	return -EINVAL;
581 }
582 
583 /* called for any extra memory-mapped regions (except initial) */
584 static void bpf_map_mmap_open(struct vm_area_struct *vma)
585 {
586 	struct bpf_map *map = vma->vm_file->private_data;
587 
588 	bpf_map_inc_with_uref(map);
589 
590 	if (vma->vm_flags & VM_WRITE) {
591 		mutex_lock(&map->freeze_mutex);
592 		map->writecnt++;
593 		mutex_unlock(&map->freeze_mutex);
594 	}
595 }
596 
597 /* called for all unmapped memory region (including initial) */
598 static void bpf_map_mmap_close(struct vm_area_struct *vma)
599 {
600 	struct bpf_map *map = vma->vm_file->private_data;
601 
602 	if (vma->vm_flags & VM_WRITE) {
603 		mutex_lock(&map->freeze_mutex);
604 		map->writecnt--;
605 		mutex_unlock(&map->freeze_mutex);
606 	}
607 
608 	bpf_map_put_with_uref(map);
609 }
610 
611 static const struct vm_operations_struct bpf_map_default_vmops = {
612 	.open		= bpf_map_mmap_open,
613 	.close		= bpf_map_mmap_close,
614 };
615 
616 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
617 {
618 	struct bpf_map *map = filp->private_data;
619 	int err;
620 
621 	if (!map->ops->map_mmap || map_value_has_spin_lock(map))
622 		return -ENOTSUPP;
623 
624 	if (!(vma->vm_flags & VM_SHARED))
625 		return -EINVAL;
626 
627 	mutex_lock(&map->freeze_mutex);
628 
629 	if ((vma->vm_flags & VM_WRITE) && map->frozen) {
630 		err = -EPERM;
631 		goto out;
632 	}
633 
634 	/* set default open/close callbacks */
635 	vma->vm_ops = &bpf_map_default_vmops;
636 	vma->vm_private_data = map;
637 
638 	err = map->ops->map_mmap(map, vma);
639 	if (err)
640 		goto out;
641 
642 	bpf_map_inc_with_uref(map);
643 
644 	if (vma->vm_flags & VM_WRITE)
645 		map->writecnt++;
646 out:
647 	mutex_unlock(&map->freeze_mutex);
648 	return err;
649 }
650 
651 const struct file_operations bpf_map_fops = {
652 #ifdef CONFIG_PROC_FS
653 	.show_fdinfo	= bpf_map_show_fdinfo,
654 #endif
655 	.release	= bpf_map_release,
656 	.read		= bpf_dummy_read,
657 	.write		= bpf_dummy_write,
658 	.mmap		= bpf_map_mmap,
659 };
660 
661 int bpf_map_new_fd(struct bpf_map *map, int flags)
662 {
663 	int ret;
664 
665 	ret = security_bpf_map(map, OPEN_FMODE(flags));
666 	if (ret < 0)
667 		return ret;
668 
669 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
670 				flags | O_CLOEXEC);
671 }
672 
673 int bpf_get_file_flag(int flags)
674 {
675 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
676 		return -EINVAL;
677 	if (flags & BPF_F_RDONLY)
678 		return O_RDONLY;
679 	if (flags & BPF_F_WRONLY)
680 		return O_WRONLY;
681 	return O_RDWR;
682 }
683 
684 /* helper macro to check that unused fields 'union bpf_attr' are zero */
685 #define CHECK_ATTR(CMD) \
686 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
687 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
688 		   sizeof(*attr) - \
689 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
690 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
691 
692 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
693  * Return 0 on success and < 0 on error.
694  */
695 static int bpf_obj_name_cpy(char *dst, const char *src)
696 {
697 	const char *end = src + BPF_OBJ_NAME_LEN;
698 
699 	memset(dst, 0, BPF_OBJ_NAME_LEN);
700 	/* Copy all isalnum(), '_' and '.' chars. */
701 	while (src < end && *src) {
702 		if (!isalnum(*src) &&
703 		    *src != '_' && *src != '.')
704 			return -EINVAL;
705 		*dst++ = *src++;
706 	}
707 
708 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
709 	if (src == end)
710 		return -EINVAL;
711 
712 	return 0;
713 }
714 
715 int map_check_no_btf(const struct bpf_map *map,
716 		     const struct btf *btf,
717 		     const struct btf_type *key_type,
718 		     const struct btf_type *value_type)
719 {
720 	return -ENOTSUPP;
721 }
722 
723 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
724 			 u32 btf_key_id, u32 btf_value_id)
725 {
726 	const struct btf_type *key_type, *value_type;
727 	u32 key_size, value_size;
728 	int ret = 0;
729 
730 	/* Some maps allow key to be unspecified. */
731 	if (btf_key_id) {
732 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
733 		if (!key_type || key_size != map->key_size)
734 			return -EINVAL;
735 	} else {
736 		key_type = btf_type_by_id(btf, 0);
737 		if (!map->ops->map_check_btf)
738 			return -EINVAL;
739 	}
740 
741 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
742 	if (!value_type || value_size != map->value_size)
743 		return -EINVAL;
744 
745 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
746 
747 	if (map_value_has_spin_lock(map)) {
748 		if (map->map_flags & BPF_F_RDONLY_PROG)
749 			return -EACCES;
750 		if (map->map_type != BPF_MAP_TYPE_HASH &&
751 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
752 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
753 		    map->map_type != BPF_MAP_TYPE_SK_STORAGE)
754 			return -ENOTSUPP;
755 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
756 		    map->value_size) {
757 			WARN_ONCE(1,
758 				  "verifier bug spin_lock_off %d value_size %d\n",
759 				  map->spin_lock_off, map->value_size);
760 			return -EFAULT;
761 		}
762 	}
763 
764 	if (map->ops->map_check_btf)
765 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
766 
767 	return ret;
768 }
769 
770 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
771 /* called via syscall */
772 static int map_create(union bpf_attr *attr)
773 {
774 	int numa_node = bpf_map_attr_numa_node(attr);
775 	struct bpf_map_memory mem;
776 	struct bpf_map *map;
777 	int f_flags;
778 	int err;
779 
780 	err = CHECK_ATTR(BPF_MAP_CREATE);
781 	if (err)
782 		return -EINVAL;
783 
784 	if (attr->btf_vmlinux_value_type_id) {
785 		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
786 		    attr->btf_key_type_id || attr->btf_value_type_id)
787 			return -EINVAL;
788 	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
789 		return -EINVAL;
790 	}
791 
792 	f_flags = bpf_get_file_flag(attr->map_flags);
793 	if (f_flags < 0)
794 		return f_flags;
795 
796 	if (numa_node != NUMA_NO_NODE &&
797 	    ((unsigned int)numa_node >= nr_node_ids ||
798 	     !node_online(numa_node)))
799 		return -EINVAL;
800 
801 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
802 	map = find_and_alloc_map(attr);
803 	if (IS_ERR(map))
804 		return PTR_ERR(map);
805 
806 	err = bpf_obj_name_cpy(map->name, attr->map_name);
807 	if (err)
808 		goto free_map;
809 
810 	atomic64_set(&map->refcnt, 1);
811 	atomic64_set(&map->usercnt, 1);
812 	mutex_init(&map->freeze_mutex);
813 
814 	map->spin_lock_off = -EINVAL;
815 	if (attr->btf_key_type_id || attr->btf_value_type_id ||
816 	    /* Even the map's value is a kernel's struct,
817 	     * the bpf_prog.o must have BTF to begin with
818 	     * to figure out the corresponding kernel's
819 	     * counter part.  Thus, attr->btf_fd has
820 	     * to be valid also.
821 	     */
822 	    attr->btf_vmlinux_value_type_id) {
823 		struct btf *btf;
824 
825 		btf = btf_get_by_fd(attr->btf_fd);
826 		if (IS_ERR(btf)) {
827 			err = PTR_ERR(btf);
828 			goto free_map;
829 		}
830 		map->btf = btf;
831 
832 		if (attr->btf_value_type_id) {
833 			err = map_check_btf(map, btf, attr->btf_key_type_id,
834 					    attr->btf_value_type_id);
835 			if (err)
836 				goto free_map;
837 		}
838 
839 		map->btf_key_type_id = attr->btf_key_type_id;
840 		map->btf_value_type_id = attr->btf_value_type_id;
841 		map->btf_vmlinux_value_type_id =
842 			attr->btf_vmlinux_value_type_id;
843 	}
844 
845 	err = security_bpf_map_alloc(map);
846 	if (err)
847 		goto free_map;
848 
849 	err = bpf_map_alloc_id(map);
850 	if (err)
851 		goto free_map_sec;
852 
853 	err = bpf_map_new_fd(map, f_flags);
854 	if (err < 0) {
855 		/* failed to allocate fd.
856 		 * bpf_map_put_with_uref() is needed because the above
857 		 * bpf_map_alloc_id() has published the map
858 		 * to the userspace and the userspace may
859 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
860 		 */
861 		bpf_map_put_with_uref(map);
862 		return err;
863 	}
864 
865 	return err;
866 
867 free_map_sec:
868 	security_bpf_map_free(map);
869 free_map:
870 	btf_put(map->btf);
871 	bpf_map_charge_move(&mem, &map->memory);
872 	map->ops->map_free(map);
873 	bpf_map_charge_finish(&mem);
874 	return err;
875 }
876 
877 /* if error is returned, fd is released.
878  * On success caller should complete fd access with matching fdput()
879  */
880 struct bpf_map *__bpf_map_get(struct fd f)
881 {
882 	if (!f.file)
883 		return ERR_PTR(-EBADF);
884 	if (f.file->f_op != &bpf_map_fops) {
885 		fdput(f);
886 		return ERR_PTR(-EINVAL);
887 	}
888 
889 	return f.file->private_data;
890 }
891 
892 void bpf_map_inc(struct bpf_map *map)
893 {
894 	atomic64_inc(&map->refcnt);
895 }
896 EXPORT_SYMBOL_GPL(bpf_map_inc);
897 
898 void bpf_map_inc_with_uref(struct bpf_map *map)
899 {
900 	atomic64_inc(&map->refcnt);
901 	atomic64_inc(&map->usercnt);
902 }
903 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
904 
905 struct bpf_map *bpf_map_get(u32 ufd)
906 {
907 	struct fd f = fdget(ufd);
908 	struct bpf_map *map;
909 
910 	map = __bpf_map_get(f);
911 	if (IS_ERR(map))
912 		return map;
913 
914 	bpf_map_inc(map);
915 	fdput(f);
916 
917 	return map;
918 }
919 
920 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
921 {
922 	struct fd f = fdget(ufd);
923 	struct bpf_map *map;
924 
925 	map = __bpf_map_get(f);
926 	if (IS_ERR(map))
927 		return map;
928 
929 	bpf_map_inc_with_uref(map);
930 	fdput(f);
931 
932 	return map;
933 }
934 
935 /* map_idr_lock should have been held */
936 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
937 {
938 	int refold;
939 
940 	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
941 	if (!refold)
942 		return ERR_PTR(-ENOENT);
943 	if (uref)
944 		atomic64_inc(&map->usercnt);
945 
946 	return map;
947 }
948 
949 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
950 {
951 	spin_lock_bh(&map_idr_lock);
952 	map = __bpf_map_inc_not_zero(map, false);
953 	spin_unlock_bh(&map_idr_lock);
954 
955 	return map;
956 }
957 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
958 
959 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
960 {
961 	return -ENOTSUPP;
962 }
963 
964 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
965 {
966 	if (key_size)
967 		return memdup_user(ukey, key_size);
968 
969 	if (ukey)
970 		return ERR_PTR(-EINVAL);
971 
972 	return NULL;
973 }
974 
975 /* last field in 'union bpf_attr' used by this command */
976 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
977 
978 static int map_lookup_elem(union bpf_attr *attr)
979 {
980 	void __user *ukey = u64_to_user_ptr(attr->key);
981 	void __user *uvalue = u64_to_user_ptr(attr->value);
982 	int ufd = attr->map_fd;
983 	struct bpf_map *map;
984 	void *key, *value;
985 	u32 value_size;
986 	struct fd f;
987 	int err;
988 
989 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
990 		return -EINVAL;
991 
992 	if (attr->flags & ~BPF_F_LOCK)
993 		return -EINVAL;
994 
995 	f = fdget(ufd);
996 	map = __bpf_map_get(f);
997 	if (IS_ERR(map))
998 		return PTR_ERR(map);
999 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1000 		err = -EPERM;
1001 		goto err_put;
1002 	}
1003 
1004 	if ((attr->flags & BPF_F_LOCK) &&
1005 	    !map_value_has_spin_lock(map)) {
1006 		err = -EINVAL;
1007 		goto err_put;
1008 	}
1009 
1010 	key = __bpf_copy_key(ukey, map->key_size);
1011 	if (IS_ERR(key)) {
1012 		err = PTR_ERR(key);
1013 		goto err_put;
1014 	}
1015 
1016 	value_size = bpf_map_value_size(map);
1017 
1018 	err = -ENOMEM;
1019 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1020 	if (!value)
1021 		goto free_key;
1022 
1023 	err = bpf_map_copy_value(map, key, value, attr->flags);
1024 	if (err)
1025 		goto free_value;
1026 
1027 	err = -EFAULT;
1028 	if (copy_to_user(uvalue, value, value_size) != 0)
1029 		goto free_value;
1030 
1031 	err = 0;
1032 
1033 free_value:
1034 	kfree(value);
1035 free_key:
1036 	kfree(key);
1037 err_put:
1038 	fdput(f);
1039 	return err;
1040 }
1041 
1042 
1043 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1044 
1045 static int map_update_elem(union bpf_attr *attr)
1046 {
1047 	void __user *ukey = u64_to_user_ptr(attr->key);
1048 	void __user *uvalue = u64_to_user_ptr(attr->value);
1049 	int ufd = attr->map_fd;
1050 	struct bpf_map *map;
1051 	void *key, *value;
1052 	u32 value_size;
1053 	struct fd f;
1054 	int err;
1055 
1056 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1057 		return -EINVAL;
1058 
1059 	f = fdget(ufd);
1060 	map = __bpf_map_get(f);
1061 	if (IS_ERR(map))
1062 		return PTR_ERR(map);
1063 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1064 		err = -EPERM;
1065 		goto err_put;
1066 	}
1067 
1068 	if ((attr->flags & BPF_F_LOCK) &&
1069 	    !map_value_has_spin_lock(map)) {
1070 		err = -EINVAL;
1071 		goto err_put;
1072 	}
1073 
1074 	key = __bpf_copy_key(ukey, map->key_size);
1075 	if (IS_ERR(key)) {
1076 		err = PTR_ERR(key);
1077 		goto err_put;
1078 	}
1079 
1080 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1081 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1082 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1083 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1084 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
1085 	else
1086 		value_size = map->value_size;
1087 
1088 	err = -ENOMEM;
1089 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1090 	if (!value)
1091 		goto free_key;
1092 
1093 	err = -EFAULT;
1094 	if (copy_from_user(value, uvalue, value_size) != 0)
1095 		goto free_value;
1096 
1097 	err = bpf_map_update_value(map, f, key, value, attr->flags);
1098 
1099 free_value:
1100 	kfree(value);
1101 free_key:
1102 	kfree(key);
1103 err_put:
1104 	fdput(f);
1105 	return err;
1106 }
1107 
1108 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1109 
1110 static int map_delete_elem(union bpf_attr *attr)
1111 {
1112 	void __user *ukey = u64_to_user_ptr(attr->key);
1113 	int ufd = attr->map_fd;
1114 	struct bpf_map *map;
1115 	struct fd f;
1116 	void *key;
1117 	int err;
1118 
1119 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1120 		return -EINVAL;
1121 
1122 	f = fdget(ufd);
1123 	map = __bpf_map_get(f);
1124 	if (IS_ERR(map))
1125 		return PTR_ERR(map);
1126 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1127 		err = -EPERM;
1128 		goto err_put;
1129 	}
1130 
1131 	key = __bpf_copy_key(ukey, map->key_size);
1132 	if (IS_ERR(key)) {
1133 		err = PTR_ERR(key);
1134 		goto err_put;
1135 	}
1136 
1137 	if (bpf_map_is_dev_bound(map)) {
1138 		err = bpf_map_offload_delete_elem(map, key);
1139 		goto out;
1140 	} else if (IS_FD_PROG_ARRAY(map) ||
1141 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1142 		/* These maps require sleepable context */
1143 		err = map->ops->map_delete_elem(map, key);
1144 		goto out;
1145 	}
1146 
1147 	bpf_disable_instrumentation();
1148 	rcu_read_lock();
1149 	err = map->ops->map_delete_elem(map, key);
1150 	rcu_read_unlock();
1151 	bpf_enable_instrumentation();
1152 	maybe_wait_bpf_programs(map);
1153 out:
1154 	kfree(key);
1155 err_put:
1156 	fdput(f);
1157 	return err;
1158 }
1159 
1160 /* last field in 'union bpf_attr' used by this command */
1161 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1162 
1163 static int map_get_next_key(union bpf_attr *attr)
1164 {
1165 	void __user *ukey = u64_to_user_ptr(attr->key);
1166 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1167 	int ufd = attr->map_fd;
1168 	struct bpf_map *map;
1169 	void *key, *next_key;
1170 	struct fd f;
1171 	int err;
1172 
1173 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1174 		return -EINVAL;
1175 
1176 	f = fdget(ufd);
1177 	map = __bpf_map_get(f);
1178 	if (IS_ERR(map))
1179 		return PTR_ERR(map);
1180 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1181 		err = -EPERM;
1182 		goto err_put;
1183 	}
1184 
1185 	if (ukey) {
1186 		key = __bpf_copy_key(ukey, map->key_size);
1187 		if (IS_ERR(key)) {
1188 			err = PTR_ERR(key);
1189 			goto err_put;
1190 		}
1191 	} else {
1192 		key = NULL;
1193 	}
1194 
1195 	err = -ENOMEM;
1196 	next_key = kmalloc(map->key_size, GFP_USER);
1197 	if (!next_key)
1198 		goto free_key;
1199 
1200 	if (bpf_map_is_dev_bound(map)) {
1201 		err = bpf_map_offload_get_next_key(map, key, next_key);
1202 		goto out;
1203 	}
1204 
1205 	rcu_read_lock();
1206 	err = map->ops->map_get_next_key(map, key, next_key);
1207 	rcu_read_unlock();
1208 out:
1209 	if (err)
1210 		goto free_next_key;
1211 
1212 	err = -EFAULT;
1213 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1214 		goto free_next_key;
1215 
1216 	err = 0;
1217 
1218 free_next_key:
1219 	kfree(next_key);
1220 free_key:
1221 	kfree(key);
1222 err_put:
1223 	fdput(f);
1224 	return err;
1225 }
1226 
1227 int generic_map_delete_batch(struct bpf_map *map,
1228 			     const union bpf_attr *attr,
1229 			     union bpf_attr __user *uattr)
1230 {
1231 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1232 	u32 cp, max_count;
1233 	int err = 0;
1234 	void *key;
1235 
1236 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1237 		return -EINVAL;
1238 
1239 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1240 	    !map_value_has_spin_lock(map)) {
1241 		return -EINVAL;
1242 	}
1243 
1244 	max_count = attr->batch.count;
1245 	if (!max_count)
1246 		return 0;
1247 
1248 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1249 	if (!key)
1250 		return -ENOMEM;
1251 
1252 	for (cp = 0; cp < max_count; cp++) {
1253 		err = -EFAULT;
1254 		if (copy_from_user(key, keys + cp * map->key_size,
1255 				   map->key_size))
1256 			break;
1257 
1258 		if (bpf_map_is_dev_bound(map)) {
1259 			err = bpf_map_offload_delete_elem(map, key);
1260 			break;
1261 		}
1262 
1263 		bpf_disable_instrumentation();
1264 		rcu_read_lock();
1265 		err = map->ops->map_delete_elem(map, key);
1266 		rcu_read_unlock();
1267 		bpf_enable_instrumentation();
1268 		maybe_wait_bpf_programs(map);
1269 		if (err)
1270 			break;
1271 	}
1272 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1273 		err = -EFAULT;
1274 
1275 	kfree(key);
1276 	return err;
1277 }
1278 
1279 int generic_map_update_batch(struct bpf_map *map,
1280 			     const union bpf_attr *attr,
1281 			     union bpf_attr __user *uattr)
1282 {
1283 	void __user *values = u64_to_user_ptr(attr->batch.values);
1284 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1285 	u32 value_size, cp, max_count;
1286 	int ufd = attr->map_fd;
1287 	void *key, *value;
1288 	struct fd f;
1289 	int err = 0;
1290 
1291 	f = fdget(ufd);
1292 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1293 		return -EINVAL;
1294 
1295 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1296 	    !map_value_has_spin_lock(map)) {
1297 		return -EINVAL;
1298 	}
1299 
1300 	value_size = bpf_map_value_size(map);
1301 
1302 	max_count = attr->batch.count;
1303 	if (!max_count)
1304 		return 0;
1305 
1306 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1307 	if (!key)
1308 		return -ENOMEM;
1309 
1310 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1311 	if (!value) {
1312 		kfree(key);
1313 		return -ENOMEM;
1314 	}
1315 
1316 	for (cp = 0; cp < max_count; cp++) {
1317 		err = -EFAULT;
1318 		if (copy_from_user(key, keys + cp * map->key_size,
1319 		    map->key_size) ||
1320 		    copy_from_user(value, values + cp * value_size, value_size))
1321 			break;
1322 
1323 		err = bpf_map_update_value(map, f, key, value,
1324 					   attr->batch.elem_flags);
1325 
1326 		if (err)
1327 			break;
1328 	}
1329 
1330 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1331 		err = -EFAULT;
1332 
1333 	kfree(value);
1334 	kfree(key);
1335 	return err;
1336 }
1337 
1338 #define MAP_LOOKUP_RETRIES 3
1339 
1340 int generic_map_lookup_batch(struct bpf_map *map,
1341 				    const union bpf_attr *attr,
1342 				    union bpf_attr __user *uattr)
1343 {
1344 	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1345 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1346 	void __user *values = u64_to_user_ptr(attr->batch.values);
1347 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1348 	void *buf, *buf_prevkey, *prev_key, *key, *value;
1349 	int err, retry = MAP_LOOKUP_RETRIES;
1350 	u32 value_size, cp, max_count;
1351 
1352 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1353 		return -EINVAL;
1354 
1355 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1356 	    !map_value_has_spin_lock(map))
1357 		return -EINVAL;
1358 
1359 	value_size = bpf_map_value_size(map);
1360 
1361 	max_count = attr->batch.count;
1362 	if (!max_count)
1363 		return 0;
1364 
1365 	if (put_user(0, &uattr->batch.count))
1366 		return -EFAULT;
1367 
1368 	buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1369 	if (!buf_prevkey)
1370 		return -ENOMEM;
1371 
1372 	buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1373 	if (!buf) {
1374 		kvfree(buf_prevkey);
1375 		return -ENOMEM;
1376 	}
1377 
1378 	err = -EFAULT;
1379 	prev_key = NULL;
1380 	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1381 		goto free_buf;
1382 	key = buf;
1383 	value = key + map->key_size;
1384 	if (ubatch)
1385 		prev_key = buf_prevkey;
1386 
1387 	for (cp = 0; cp < max_count;) {
1388 		rcu_read_lock();
1389 		err = map->ops->map_get_next_key(map, prev_key, key);
1390 		rcu_read_unlock();
1391 		if (err)
1392 			break;
1393 		err = bpf_map_copy_value(map, key, value,
1394 					 attr->batch.elem_flags);
1395 
1396 		if (err == -ENOENT) {
1397 			if (retry) {
1398 				retry--;
1399 				continue;
1400 			}
1401 			err = -EINTR;
1402 			break;
1403 		}
1404 
1405 		if (err)
1406 			goto free_buf;
1407 
1408 		if (copy_to_user(keys + cp * map->key_size, key,
1409 				 map->key_size)) {
1410 			err = -EFAULT;
1411 			goto free_buf;
1412 		}
1413 		if (copy_to_user(values + cp * value_size, value, value_size)) {
1414 			err = -EFAULT;
1415 			goto free_buf;
1416 		}
1417 
1418 		if (!prev_key)
1419 			prev_key = buf_prevkey;
1420 
1421 		swap(prev_key, key);
1422 		retry = MAP_LOOKUP_RETRIES;
1423 		cp++;
1424 	}
1425 
1426 	if (err == -EFAULT)
1427 		goto free_buf;
1428 
1429 	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1430 		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1431 		err = -EFAULT;
1432 
1433 free_buf:
1434 	kfree(buf_prevkey);
1435 	kfree(buf);
1436 	return err;
1437 }
1438 
1439 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1440 
1441 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1442 {
1443 	void __user *ukey = u64_to_user_ptr(attr->key);
1444 	void __user *uvalue = u64_to_user_ptr(attr->value);
1445 	int ufd = attr->map_fd;
1446 	struct bpf_map *map;
1447 	void *key, *value;
1448 	u32 value_size;
1449 	struct fd f;
1450 	int err;
1451 
1452 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1453 		return -EINVAL;
1454 
1455 	f = fdget(ufd);
1456 	map = __bpf_map_get(f);
1457 	if (IS_ERR(map))
1458 		return PTR_ERR(map);
1459 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1460 		err = -EPERM;
1461 		goto err_put;
1462 	}
1463 
1464 	key = __bpf_copy_key(ukey, map->key_size);
1465 	if (IS_ERR(key)) {
1466 		err = PTR_ERR(key);
1467 		goto err_put;
1468 	}
1469 
1470 	value_size = map->value_size;
1471 
1472 	err = -ENOMEM;
1473 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1474 	if (!value)
1475 		goto free_key;
1476 
1477 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1478 	    map->map_type == BPF_MAP_TYPE_STACK) {
1479 		err = map->ops->map_pop_elem(map, value);
1480 	} else {
1481 		err = -ENOTSUPP;
1482 	}
1483 
1484 	if (err)
1485 		goto free_value;
1486 
1487 	if (copy_to_user(uvalue, value, value_size) != 0)
1488 		goto free_value;
1489 
1490 	err = 0;
1491 
1492 free_value:
1493 	kfree(value);
1494 free_key:
1495 	kfree(key);
1496 err_put:
1497 	fdput(f);
1498 	return err;
1499 }
1500 
1501 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1502 
1503 static int map_freeze(const union bpf_attr *attr)
1504 {
1505 	int err = 0, ufd = attr->map_fd;
1506 	struct bpf_map *map;
1507 	struct fd f;
1508 
1509 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1510 		return -EINVAL;
1511 
1512 	f = fdget(ufd);
1513 	map = __bpf_map_get(f);
1514 	if (IS_ERR(map))
1515 		return PTR_ERR(map);
1516 
1517 	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1518 		fdput(f);
1519 		return -ENOTSUPP;
1520 	}
1521 
1522 	mutex_lock(&map->freeze_mutex);
1523 
1524 	if (map->writecnt) {
1525 		err = -EBUSY;
1526 		goto err_put;
1527 	}
1528 	if (READ_ONCE(map->frozen)) {
1529 		err = -EBUSY;
1530 		goto err_put;
1531 	}
1532 	if (!capable(CAP_SYS_ADMIN)) {
1533 		err = -EPERM;
1534 		goto err_put;
1535 	}
1536 
1537 	WRITE_ONCE(map->frozen, true);
1538 err_put:
1539 	mutex_unlock(&map->freeze_mutex);
1540 	fdput(f);
1541 	return err;
1542 }
1543 
1544 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1545 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1546 	[_id] = & _name ## _prog_ops,
1547 #define BPF_MAP_TYPE(_id, _ops)
1548 #include <linux/bpf_types.h>
1549 #undef BPF_PROG_TYPE
1550 #undef BPF_MAP_TYPE
1551 };
1552 
1553 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1554 {
1555 	const struct bpf_prog_ops *ops;
1556 
1557 	if (type >= ARRAY_SIZE(bpf_prog_types))
1558 		return -EINVAL;
1559 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1560 	ops = bpf_prog_types[type];
1561 	if (!ops)
1562 		return -EINVAL;
1563 
1564 	if (!bpf_prog_is_dev_bound(prog->aux))
1565 		prog->aux->ops = ops;
1566 	else
1567 		prog->aux->ops = &bpf_offload_prog_ops;
1568 	prog->type = type;
1569 	return 0;
1570 }
1571 
1572 enum bpf_audit {
1573 	BPF_AUDIT_LOAD,
1574 	BPF_AUDIT_UNLOAD,
1575 	BPF_AUDIT_MAX,
1576 };
1577 
1578 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1579 	[BPF_AUDIT_LOAD]   = "LOAD",
1580 	[BPF_AUDIT_UNLOAD] = "UNLOAD",
1581 };
1582 
1583 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1584 {
1585 	struct audit_context *ctx = NULL;
1586 	struct audit_buffer *ab;
1587 
1588 	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1589 		return;
1590 	if (audit_enabled == AUDIT_OFF)
1591 		return;
1592 	if (op == BPF_AUDIT_LOAD)
1593 		ctx = audit_context();
1594 	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1595 	if (unlikely(!ab))
1596 		return;
1597 	audit_log_format(ab, "prog-id=%u op=%s",
1598 			 prog->aux->id, bpf_audit_str[op]);
1599 	audit_log_end(ab);
1600 }
1601 
1602 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1603 {
1604 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1605 	unsigned long user_bufs;
1606 
1607 	if (user) {
1608 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1609 		if (user_bufs > memlock_limit) {
1610 			atomic_long_sub(pages, &user->locked_vm);
1611 			return -EPERM;
1612 		}
1613 	}
1614 
1615 	return 0;
1616 }
1617 
1618 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1619 {
1620 	if (user)
1621 		atomic_long_sub(pages, &user->locked_vm);
1622 }
1623 
1624 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1625 {
1626 	struct user_struct *user = get_current_user();
1627 	int ret;
1628 
1629 	ret = __bpf_prog_charge(user, prog->pages);
1630 	if (ret) {
1631 		free_uid(user);
1632 		return ret;
1633 	}
1634 
1635 	prog->aux->user = user;
1636 	return 0;
1637 }
1638 
1639 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1640 {
1641 	struct user_struct *user = prog->aux->user;
1642 
1643 	__bpf_prog_uncharge(user, prog->pages);
1644 	free_uid(user);
1645 }
1646 
1647 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1648 {
1649 	int id;
1650 
1651 	idr_preload(GFP_KERNEL);
1652 	spin_lock_bh(&prog_idr_lock);
1653 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1654 	if (id > 0)
1655 		prog->aux->id = id;
1656 	spin_unlock_bh(&prog_idr_lock);
1657 	idr_preload_end();
1658 
1659 	/* id is in [1, INT_MAX) */
1660 	if (WARN_ON_ONCE(!id))
1661 		return -ENOSPC;
1662 
1663 	return id > 0 ? 0 : id;
1664 }
1665 
1666 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1667 {
1668 	/* cBPF to eBPF migrations are currently not in the idr store.
1669 	 * Offloaded programs are removed from the store when their device
1670 	 * disappears - even if someone grabs an fd to them they are unusable,
1671 	 * simply waiting for refcnt to drop to be freed.
1672 	 */
1673 	if (!prog->aux->id)
1674 		return;
1675 
1676 	if (do_idr_lock)
1677 		spin_lock_bh(&prog_idr_lock);
1678 	else
1679 		__acquire(&prog_idr_lock);
1680 
1681 	idr_remove(&prog_idr, prog->aux->id);
1682 	prog->aux->id = 0;
1683 
1684 	if (do_idr_lock)
1685 		spin_unlock_bh(&prog_idr_lock);
1686 	else
1687 		__release(&prog_idr_lock);
1688 }
1689 
1690 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1691 {
1692 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1693 
1694 	kvfree(aux->func_info);
1695 	kfree(aux->func_info_aux);
1696 	bpf_prog_uncharge_memlock(aux->prog);
1697 	security_bpf_prog_free(aux);
1698 	bpf_prog_free(aux->prog);
1699 }
1700 
1701 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1702 {
1703 	bpf_prog_kallsyms_del_all(prog);
1704 	btf_put(prog->aux->btf);
1705 	bpf_prog_free_linfo(prog);
1706 
1707 	if (deferred)
1708 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1709 	else
1710 		__bpf_prog_put_rcu(&prog->aux->rcu);
1711 }
1712 
1713 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1714 {
1715 	if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1716 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1717 		bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1718 		/* bpf_prog_free_id() must be called first */
1719 		bpf_prog_free_id(prog, do_idr_lock);
1720 		__bpf_prog_put_noref(prog, true);
1721 	}
1722 }
1723 
1724 void bpf_prog_put(struct bpf_prog *prog)
1725 {
1726 	__bpf_prog_put(prog, true);
1727 }
1728 EXPORT_SYMBOL_GPL(bpf_prog_put);
1729 
1730 static int bpf_prog_release(struct inode *inode, struct file *filp)
1731 {
1732 	struct bpf_prog *prog = filp->private_data;
1733 
1734 	bpf_prog_put(prog);
1735 	return 0;
1736 }
1737 
1738 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1739 			       struct bpf_prog_stats *stats)
1740 {
1741 	u64 nsecs = 0, cnt = 0;
1742 	int cpu;
1743 
1744 	for_each_possible_cpu(cpu) {
1745 		const struct bpf_prog_stats *st;
1746 		unsigned int start;
1747 		u64 tnsecs, tcnt;
1748 
1749 		st = per_cpu_ptr(prog->aux->stats, cpu);
1750 		do {
1751 			start = u64_stats_fetch_begin_irq(&st->syncp);
1752 			tnsecs = st->nsecs;
1753 			tcnt = st->cnt;
1754 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1755 		nsecs += tnsecs;
1756 		cnt += tcnt;
1757 	}
1758 	stats->nsecs = nsecs;
1759 	stats->cnt = cnt;
1760 }
1761 
1762 #ifdef CONFIG_PROC_FS
1763 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1764 {
1765 	const struct bpf_prog *prog = filp->private_data;
1766 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1767 	struct bpf_prog_stats stats;
1768 
1769 	bpf_prog_get_stats(prog, &stats);
1770 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1771 	seq_printf(m,
1772 		   "prog_type:\t%u\n"
1773 		   "prog_jited:\t%u\n"
1774 		   "prog_tag:\t%s\n"
1775 		   "memlock:\t%llu\n"
1776 		   "prog_id:\t%u\n"
1777 		   "run_time_ns:\t%llu\n"
1778 		   "run_cnt:\t%llu\n",
1779 		   prog->type,
1780 		   prog->jited,
1781 		   prog_tag,
1782 		   prog->pages * 1ULL << PAGE_SHIFT,
1783 		   prog->aux->id,
1784 		   stats.nsecs,
1785 		   stats.cnt);
1786 }
1787 #endif
1788 
1789 const struct file_operations bpf_prog_fops = {
1790 #ifdef CONFIG_PROC_FS
1791 	.show_fdinfo	= bpf_prog_show_fdinfo,
1792 #endif
1793 	.release	= bpf_prog_release,
1794 	.read		= bpf_dummy_read,
1795 	.write		= bpf_dummy_write,
1796 };
1797 
1798 int bpf_prog_new_fd(struct bpf_prog *prog)
1799 {
1800 	int ret;
1801 
1802 	ret = security_bpf_prog(prog);
1803 	if (ret < 0)
1804 		return ret;
1805 
1806 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1807 				O_RDWR | O_CLOEXEC);
1808 }
1809 
1810 static struct bpf_prog *____bpf_prog_get(struct fd f)
1811 {
1812 	if (!f.file)
1813 		return ERR_PTR(-EBADF);
1814 	if (f.file->f_op != &bpf_prog_fops) {
1815 		fdput(f);
1816 		return ERR_PTR(-EINVAL);
1817 	}
1818 
1819 	return f.file->private_data;
1820 }
1821 
1822 void bpf_prog_add(struct bpf_prog *prog, int i)
1823 {
1824 	atomic64_add(i, &prog->aux->refcnt);
1825 }
1826 EXPORT_SYMBOL_GPL(bpf_prog_add);
1827 
1828 void bpf_prog_sub(struct bpf_prog *prog, int i)
1829 {
1830 	/* Only to be used for undoing previous bpf_prog_add() in some
1831 	 * error path. We still know that another entity in our call
1832 	 * path holds a reference to the program, thus atomic_sub() can
1833 	 * be safely used in such cases!
1834 	 */
1835 	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1836 }
1837 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1838 
1839 void bpf_prog_inc(struct bpf_prog *prog)
1840 {
1841 	atomic64_inc(&prog->aux->refcnt);
1842 }
1843 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1844 
1845 /* prog_idr_lock should have been held */
1846 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1847 {
1848 	int refold;
1849 
1850 	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1851 
1852 	if (!refold)
1853 		return ERR_PTR(-ENOENT);
1854 
1855 	return prog;
1856 }
1857 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1858 
1859 bool bpf_prog_get_ok(struct bpf_prog *prog,
1860 			    enum bpf_prog_type *attach_type, bool attach_drv)
1861 {
1862 	/* not an attachment, just a refcount inc, always allow */
1863 	if (!attach_type)
1864 		return true;
1865 
1866 	if (prog->type != *attach_type)
1867 		return false;
1868 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1869 		return false;
1870 
1871 	return true;
1872 }
1873 
1874 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1875 				       bool attach_drv)
1876 {
1877 	struct fd f = fdget(ufd);
1878 	struct bpf_prog *prog;
1879 
1880 	prog = ____bpf_prog_get(f);
1881 	if (IS_ERR(prog))
1882 		return prog;
1883 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1884 		prog = ERR_PTR(-EINVAL);
1885 		goto out;
1886 	}
1887 
1888 	bpf_prog_inc(prog);
1889 out:
1890 	fdput(f);
1891 	return prog;
1892 }
1893 
1894 struct bpf_prog *bpf_prog_get(u32 ufd)
1895 {
1896 	return __bpf_prog_get(ufd, NULL, false);
1897 }
1898 
1899 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1900 				       bool attach_drv)
1901 {
1902 	return __bpf_prog_get(ufd, &type, attach_drv);
1903 }
1904 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1905 
1906 /* Initially all BPF programs could be loaded w/o specifying
1907  * expected_attach_type. Later for some of them specifying expected_attach_type
1908  * at load time became required so that program could be validated properly.
1909  * Programs of types that are allowed to be loaded both w/ and w/o (for
1910  * backward compatibility) expected_attach_type, should have the default attach
1911  * type assigned to expected_attach_type for the latter case, so that it can be
1912  * validated later at attach time.
1913  *
1914  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1915  * prog type requires it but has some attach types that have to be backward
1916  * compatible.
1917  */
1918 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1919 {
1920 	switch (attr->prog_type) {
1921 	case BPF_PROG_TYPE_CGROUP_SOCK:
1922 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1923 		 * exist so checking for non-zero is the way to go here.
1924 		 */
1925 		if (!attr->expected_attach_type)
1926 			attr->expected_attach_type =
1927 				BPF_CGROUP_INET_SOCK_CREATE;
1928 		break;
1929 	}
1930 }
1931 
1932 static int
1933 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1934 			   enum bpf_attach_type expected_attach_type,
1935 			   u32 btf_id, u32 prog_fd)
1936 {
1937 	if (btf_id) {
1938 		if (btf_id > BTF_MAX_TYPE)
1939 			return -EINVAL;
1940 
1941 		switch (prog_type) {
1942 		case BPF_PROG_TYPE_TRACING:
1943 		case BPF_PROG_TYPE_STRUCT_OPS:
1944 		case BPF_PROG_TYPE_EXT:
1945 			break;
1946 		default:
1947 			return -EINVAL;
1948 		}
1949 	}
1950 
1951 	if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
1952 	    prog_type != BPF_PROG_TYPE_EXT)
1953 		return -EINVAL;
1954 
1955 	switch (prog_type) {
1956 	case BPF_PROG_TYPE_CGROUP_SOCK:
1957 		switch (expected_attach_type) {
1958 		case BPF_CGROUP_INET_SOCK_CREATE:
1959 		case BPF_CGROUP_INET4_POST_BIND:
1960 		case BPF_CGROUP_INET6_POST_BIND:
1961 			return 0;
1962 		default:
1963 			return -EINVAL;
1964 		}
1965 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1966 		switch (expected_attach_type) {
1967 		case BPF_CGROUP_INET4_BIND:
1968 		case BPF_CGROUP_INET6_BIND:
1969 		case BPF_CGROUP_INET4_CONNECT:
1970 		case BPF_CGROUP_INET6_CONNECT:
1971 		case BPF_CGROUP_UDP4_SENDMSG:
1972 		case BPF_CGROUP_UDP6_SENDMSG:
1973 		case BPF_CGROUP_UDP4_RECVMSG:
1974 		case BPF_CGROUP_UDP6_RECVMSG:
1975 			return 0;
1976 		default:
1977 			return -EINVAL;
1978 		}
1979 	case BPF_PROG_TYPE_CGROUP_SKB:
1980 		switch (expected_attach_type) {
1981 		case BPF_CGROUP_INET_INGRESS:
1982 		case BPF_CGROUP_INET_EGRESS:
1983 			return 0;
1984 		default:
1985 			return -EINVAL;
1986 		}
1987 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1988 		switch (expected_attach_type) {
1989 		case BPF_CGROUP_SETSOCKOPT:
1990 		case BPF_CGROUP_GETSOCKOPT:
1991 			return 0;
1992 		default:
1993 			return -EINVAL;
1994 		}
1995 	case BPF_PROG_TYPE_EXT:
1996 		if (expected_attach_type)
1997 			return -EINVAL;
1998 		/* fallthrough */
1999 	default:
2000 		return 0;
2001 	}
2002 }
2003 
2004 /* last field in 'union bpf_attr' used by this command */
2005 #define	BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2006 
2007 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2008 {
2009 	enum bpf_prog_type type = attr->prog_type;
2010 	struct bpf_prog *prog;
2011 	int err;
2012 	char license[128];
2013 	bool is_gpl;
2014 
2015 	if (CHECK_ATTR(BPF_PROG_LOAD))
2016 		return -EINVAL;
2017 
2018 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2019 				 BPF_F_ANY_ALIGNMENT |
2020 				 BPF_F_TEST_STATE_FREQ |
2021 				 BPF_F_TEST_RND_HI32))
2022 		return -EINVAL;
2023 
2024 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2025 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2026 	    !capable(CAP_SYS_ADMIN))
2027 		return -EPERM;
2028 
2029 	/* copy eBPF program license from user space */
2030 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2031 			      sizeof(license) - 1) < 0)
2032 		return -EFAULT;
2033 	license[sizeof(license) - 1] = 0;
2034 
2035 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2036 	is_gpl = license_is_gpl_compatible(license);
2037 
2038 	if (attr->insn_cnt == 0 ||
2039 	    attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2040 		return -E2BIG;
2041 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2042 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2043 	    !capable(CAP_SYS_ADMIN))
2044 		return -EPERM;
2045 
2046 	bpf_prog_load_fixup_attach_type(attr);
2047 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2048 				       attr->attach_btf_id,
2049 				       attr->attach_prog_fd))
2050 		return -EINVAL;
2051 
2052 	/* plain bpf_prog allocation */
2053 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2054 	if (!prog)
2055 		return -ENOMEM;
2056 
2057 	prog->expected_attach_type = attr->expected_attach_type;
2058 	prog->aux->attach_btf_id = attr->attach_btf_id;
2059 	if (attr->attach_prog_fd) {
2060 		struct bpf_prog *tgt_prog;
2061 
2062 		tgt_prog = bpf_prog_get(attr->attach_prog_fd);
2063 		if (IS_ERR(tgt_prog)) {
2064 			err = PTR_ERR(tgt_prog);
2065 			goto free_prog_nouncharge;
2066 		}
2067 		prog->aux->linked_prog = tgt_prog;
2068 	}
2069 
2070 	prog->aux->offload_requested = !!attr->prog_ifindex;
2071 
2072 	err = security_bpf_prog_alloc(prog->aux);
2073 	if (err)
2074 		goto free_prog_nouncharge;
2075 
2076 	err = bpf_prog_charge_memlock(prog);
2077 	if (err)
2078 		goto free_prog_sec;
2079 
2080 	prog->len = attr->insn_cnt;
2081 
2082 	err = -EFAULT;
2083 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2084 			   bpf_prog_insn_size(prog)) != 0)
2085 		goto free_prog;
2086 
2087 	prog->orig_prog = NULL;
2088 	prog->jited = 0;
2089 
2090 	atomic64_set(&prog->aux->refcnt, 1);
2091 	prog->gpl_compatible = is_gpl ? 1 : 0;
2092 
2093 	if (bpf_prog_is_dev_bound(prog->aux)) {
2094 		err = bpf_prog_offload_init(prog, attr);
2095 		if (err)
2096 			goto free_prog;
2097 	}
2098 
2099 	/* find program type: socket_filter vs tracing_filter */
2100 	err = find_prog_type(type, prog);
2101 	if (err < 0)
2102 		goto free_prog;
2103 
2104 	prog->aux->load_time = ktime_get_boottime_ns();
2105 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
2106 	if (err)
2107 		goto free_prog;
2108 
2109 	/* run eBPF verifier */
2110 	err = bpf_check(&prog, attr, uattr);
2111 	if (err < 0)
2112 		goto free_used_maps;
2113 
2114 	prog = bpf_prog_select_runtime(prog, &err);
2115 	if (err < 0)
2116 		goto free_used_maps;
2117 
2118 	err = bpf_prog_alloc_id(prog);
2119 	if (err)
2120 		goto free_used_maps;
2121 
2122 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2123 	 * effectively publicly exposed. However, retrieving via
2124 	 * bpf_prog_get_fd_by_id() will take another reference,
2125 	 * therefore it cannot be gone underneath us.
2126 	 *
2127 	 * Only for the time /after/ successful bpf_prog_new_fd()
2128 	 * and before returning to userspace, we might just hold
2129 	 * one reference and any parallel close on that fd could
2130 	 * rip everything out. Hence, below notifications must
2131 	 * happen before bpf_prog_new_fd().
2132 	 *
2133 	 * Also, any failure handling from this point onwards must
2134 	 * be using bpf_prog_put() given the program is exposed.
2135 	 */
2136 	bpf_prog_kallsyms_add(prog);
2137 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2138 	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2139 
2140 	err = bpf_prog_new_fd(prog);
2141 	if (err < 0)
2142 		bpf_prog_put(prog);
2143 	return err;
2144 
2145 free_used_maps:
2146 	/* In case we have subprogs, we need to wait for a grace
2147 	 * period before we can tear down JIT memory since symbols
2148 	 * are already exposed under kallsyms.
2149 	 */
2150 	__bpf_prog_put_noref(prog, prog->aux->func_cnt);
2151 	return err;
2152 free_prog:
2153 	bpf_prog_uncharge_memlock(prog);
2154 free_prog_sec:
2155 	security_bpf_prog_free(prog->aux);
2156 free_prog_nouncharge:
2157 	bpf_prog_free(prog);
2158 	return err;
2159 }
2160 
2161 #define BPF_OBJ_LAST_FIELD file_flags
2162 
2163 static int bpf_obj_pin(const union bpf_attr *attr)
2164 {
2165 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2166 		return -EINVAL;
2167 
2168 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2169 }
2170 
2171 static int bpf_obj_get(const union bpf_attr *attr)
2172 {
2173 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2174 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2175 		return -EINVAL;
2176 
2177 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2178 				attr->file_flags);
2179 }
2180 
2181 struct bpf_link {
2182 	atomic64_t refcnt;
2183 	const struct bpf_link_ops *ops;
2184 	struct bpf_prog *prog;
2185 	struct work_struct work;
2186 };
2187 
2188 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops,
2189 		   struct bpf_prog *prog)
2190 {
2191 	atomic64_set(&link->refcnt, 1);
2192 	link->ops = ops;
2193 	link->prog = prog;
2194 }
2195 
2196 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2197  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2198  * anon_inode's release() call. This helper manages marking bpf_link as
2199  * defunct, releases anon_inode file and puts reserved FD.
2200  */
2201 static void bpf_link_cleanup(struct bpf_link *link, struct file *link_file,
2202 			     int link_fd)
2203 {
2204 	link->prog = NULL;
2205 	fput(link_file);
2206 	put_unused_fd(link_fd);
2207 }
2208 
2209 void bpf_link_inc(struct bpf_link *link)
2210 {
2211 	atomic64_inc(&link->refcnt);
2212 }
2213 
2214 /* bpf_link_free is guaranteed to be called from process context */
2215 static void bpf_link_free(struct bpf_link *link)
2216 {
2217 	if (link->prog) {
2218 		/* detach BPF program, clean up used resources */
2219 		link->ops->release(link);
2220 		bpf_prog_put(link->prog);
2221 	}
2222 	/* free bpf_link and its containing memory */
2223 	link->ops->dealloc(link);
2224 }
2225 
2226 static void bpf_link_put_deferred(struct work_struct *work)
2227 {
2228 	struct bpf_link *link = container_of(work, struct bpf_link, work);
2229 
2230 	bpf_link_free(link);
2231 }
2232 
2233 /* bpf_link_put can be called from atomic context, but ensures that resources
2234  * are freed from process context
2235  */
2236 void bpf_link_put(struct bpf_link *link)
2237 {
2238 	if (!atomic64_dec_and_test(&link->refcnt))
2239 		return;
2240 
2241 	if (in_atomic()) {
2242 		INIT_WORK(&link->work, bpf_link_put_deferred);
2243 		schedule_work(&link->work);
2244 	} else {
2245 		bpf_link_free(link);
2246 	}
2247 }
2248 
2249 static int bpf_link_release(struct inode *inode, struct file *filp)
2250 {
2251 	struct bpf_link *link = filp->private_data;
2252 
2253 	bpf_link_put(link);
2254 	return 0;
2255 }
2256 
2257 #ifdef CONFIG_PROC_FS
2258 static const struct bpf_link_ops bpf_raw_tp_lops;
2259 static const struct bpf_link_ops bpf_tracing_link_lops;
2260 static const struct bpf_link_ops bpf_xdp_link_lops;
2261 
2262 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2263 {
2264 	const struct bpf_link *link = filp->private_data;
2265 	const struct bpf_prog *prog = link->prog;
2266 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2267 	const char *link_type;
2268 
2269 	if (link->ops == &bpf_raw_tp_lops)
2270 		link_type = "raw_tracepoint";
2271 	else if (link->ops == &bpf_tracing_link_lops)
2272 		link_type = "tracing";
2273 	else
2274 		link_type = "unknown";
2275 
2276 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2277 	seq_printf(m,
2278 		   "link_type:\t%s\n"
2279 		   "prog_tag:\t%s\n"
2280 		   "prog_id:\t%u\n",
2281 		   link_type,
2282 		   prog_tag,
2283 		   prog->aux->id);
2284 }
2285 #endif
2286 
2287 const struct file_operations bpf_link_fops = {
2288 #ifdef CONFIG_PROC_FS
2289 	.show_fdinfo	= bpf_link_show_fdinfo,
2290 #endif
2291 	.release	= bpf_link_release,
2292 	.read		= bpf_dummy_read,
2293 	.write		= bpf_dummy_write,
2294 };
2295 
2296 int bpf_link_new_fd(struct bpf_link *link)
2297 {
2298 	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2299 }
2300 
2301 /* Similar to bpf_link_new_fd, create anon_inode for given bpf_link, but
2302  * instead of immediately installing fd in fdtable, just reserve it and
2303  * return. Caller then need to either install it with fd_install(fd, file) or
2304  * release with put_unused_fd(fd).
2305  * This is useful for cases when bpf_link attachment/detachment are
2306  * complicated and expensive operations and should be delayed until all the fd
2307  * reservation and anon_inode creation succeeds.
2308  */
2309 struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd)
2310 {
2311 	struct file *file;
2312 	int fd;
2313 
2314 	fd = get_unused_fd_flags(O_CLOEXEC);
2315 	if (fd < 0)
2316 		return ERR_PTR(fd);
2317 
2318 	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2319 	if (IS_ERR(file)) {
2320 		put_unused_fd(fd);
2321 		return file;
2322 	}
2323 
2324 	*reserved_fd = fd;
2325 	return file;
2326 }
2327 
2328 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2329 {
2330 	struct fd f = fdget(ufd);
2331 	struct bpf_link *link;
2332 
2333 	if (!f.file)
2334 		return ERR_PTR(-EBADF);
2335 	if (f.file->f_op != &bpf_link_fops) {
2336 		fdput(f);
2337 		return ERR_PTR(-EINVAL);
2338 	}
2339 
2340 	link = f.file->private_data;
2341 	bpf_link_inc(link);
2342 	fdput(f);
2343 
2344 	return link;
2345 }
2346 
2347 struct bpf_tracing_link {
2348 	struct bpf_link link;
2349 };
2350 
2351 static void bpf_tracing_link_release(struct bpf_link *link)
2352 {
2353 	WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog));
2354 }
2355 
2356 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2357 {
2358 	struct bpf_tracing_link *tr_link =
2359 		container_of(link, struct bpf_tracing_link, link);
2360 
2361 	kfree(tr_link);
2362 }
2363 
2364 static const struct bpf_link_ops bpf_tracing_link_lops = {
2365 	.release = bpf_tracing_link_release,
2366 	.dealloc = bpf_tracing_link_dealloc,
2367 };
2368 
2369 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
2370 {
2371 	struct bpf_tracing_link *link;
2372 	struct file *link_file;
2373 	int link_fd, err;
2374 
2375 	if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2376 	    prog->expected_attach_type != BPF_TRACE_FEXIT &&
2377 	    prog->expected_attach_type != BPF_MODIFY_RETURN &&
2378 	    prog->type != BPF_PROG_TYPE_EXT) {
2379 		err = -EINVAL;
2380 		goto out_put_prog;
2381 	}
2382 
2383 	link = kzalloc(sizeof(*link), GFP_USER);
2384 	if (!link) {
2385 		err = -ENOMEM;
2386 		goto out_put_prog;
2387 	}
2388 	bpf_link_init(&link->link, &bpf_tracing_link_lops, prog);
2389 
2390 	link_file = bpf_link_new_file(&link->link, &link_fd);
2391 	if (IS_ERR(link_file)) {
2392 		kfree(link);
2393 		err = PTR_ERR(link_file);
2394 		goto out_put_prog;
2395 	}
2396 
2397 	err = bpf_trampoline_link_prog(prog);
2398 	if (err) {
2399 		bpf_link_cleanup(&link->link, link_file, link_fd);
2400 		goto out_put_prog;
2401 	}
2402 
2403 	fd_install(link_fd, link_file);
2404 	return link_fd;
2405 
2406 out_put_prog:
2407 	bpf_prog_put(prog);
2408 	return err;
2409 }
2410 
2411 struct bpf_raw_tp_link {
2412 	struct bpf_link link;
2413 	struct bpf_raw_event_map *btp;
2414 };
2415 
2416 static void bpf_raw_tp_link_release(struct bpf_link *link)
2417 {
2418 	struct bpf_raw_tp_link *raw_tp =
2419 		container_of(link, struct bpf_raw_tp_link, link);
2420 
2421 	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2422 	bpf_put_raw_tracepoint(raw_tp->btp);
2423 }
2424 
2425 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2426 {
2427 	struct bpf_raw_tp_link *raw_tp =
2428 		container_of(link, struct bpf_raw_tp_link, link);
2429 
2430 	kfree(raw_tp);
2431 }
2432 
2433 static const struct bpf_link_ops bpf_raw_tp_lops = {
2434 	.release = bpf_raw_tp_link_release,
2435 	.dealloc = bpf_raw_tp_link_dealloc,
2436 };
2437 
2438 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2439 
2440 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2441 {
2442 	struct bpf_raw_tp_link *link;
2443 	struct bpf_raw_event_map *btp;
2444 	struct file *link_file;
2445 	struct bpf_prog *prog;
2446 	const char *tp_name;
2447 	char buf[128];
2448 	int link_fd, err;
2449 
2450 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2451 		return -EINVAL;
2452 
2453 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2454 	if (IS_ERR(prog))
2455 		return PTR_ERR(prog);
2456 
2457 	if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
2458 	    prog->type != BPF_PROG_TYPE_TRACING &&
2459 	    prog->type != BPF_PROG_TYPE_EXT &&
2460 	    prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
2461 		err = -EINVAL;
2462 		goto out_put_prog;
2463 	}
2464 
2465 	if (prog->type == BPF_PROG_TYPE_TRACING ||
2466 	    prog->type == BPF_PROG_TYPE_EXT) {
2467 		if (attr->raw_tracepoint.name) {
2468 			/* The attach point for this category of programs
2469 			 * should be specified via btf_id during program load.
2470 			 */
2471 			err = -EINVAL;
2472 			goto out_put_prog;
2473 		}
2474 		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
2475 			tp_name = prog->aux->attach_func_name;
2476 		else
2477 			return bpf_tracing_prog_attach(prog);
2478 	} else {
2479 		if (strncpy_from_user(buf,
2480 				      u64_to_user_ptr(attr->raw_tracepoint.name),
2481 				      sizeof(buf) - 1) < 0) {
2482 			err = -EFAULT;
2483 			goto out_put_prog;
2484 		}
2485 		buf[sizeof(buf) - 1] = 0;
2486 		tp_name = buf;
2487 	}
2488 
2489 	btp = bpf_get_raw_tracepoint(tp_name);
2490 	if (!btp) {
2491 		err = -ENOENT;
2492 		goto out_put_prog;
2493 	}
2494 
2495 	link = kzalloc(sizeof(*link), GFP_USER);
2496 	if (!link) {
2497 		err = -ENOMEM;
2498 		goto out_put_btp;
2499 	}
2500 	bpf_link_init(&link->link, &bpf_raw_tp_lops, prog);
2501 	link->btp = btp;
2502 
2503 	link_file = bpf_link_new_file(&link->link, &link_fd);
2504 	if (IS_ERR(link_file)) {
2505 		kfree(link);
2506 		err = PTR_ERR(link_file);
2507 		goto out_put_btp;
2508 	}
2509 
2510 	err = bpf_probe_register(link->btp, prog);
2511 	if (err) {
2512 		bpf_link_cleanup(&link->link, link_file, link_fd);
2513 		goto out_put_btp;
2514 	}
2515 
2516 	fd_install(link_fd, link_file);
2517 	return link_fd;
2518 
2519 out_put_btp:
2520 	bpf_put_raw_tracepoint(btp);
2521 out_put_prog:
2522 	bpf_prog_put(prog);
2523 	return err;
2524 }
2525 
2526 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2527 					     enum bpf_attach_type attach_type)
2528 {
2529 	switch (prog->type) {
2530 	case BPF_PROG_TYPE_CGROUP_SOCK:
2531 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2532 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2533 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2534 	case BPF_PROG_TYPE_CGROUP_SKB:
2535 		return prog->enforce_expected_attach_type &&
2536 			prog->expected_attach_type != attach_type ?
2537 			-EINVAL : 0;
2538 	default:
2539 		return 0;
2540 	}
2541 }
2542 
2543 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2544 
2545 #define BPF_F_ATTACH_MASK \
2546 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2547 
2548 static int bpf_prog_attach(const union bpf_attr *attr)
2549 {
2550 	enum bpf_prog_type ptype;
2551 	struct bpf_prog *prog;
2552 	int ret;
2553 
2554 	if (!capable(CAP_NET_ADMIN))
2555 		return -EPERM;
2556 
2557 	if (CHECK_ATTR(BPF_PROG_ATTACH))
2558 		return -EINVAL;
2559 
2560 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2561 		return -EINVAL;
2562 
2563 	switch (attr->attach_type) {
2564 	case BPF_CGROUP_INET_INGRESS:
2565 	case BPF_CGROUP_INET_EGRESS:
2566 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2567 		break;
2568 	case BPF_CGROUP_INET_SOCK_CREATE:
2569 	case BPF_CGROUP_INET4_POST_BIND:
2570 	case BPF_CGROUP_INET6_POST_BIND:
2571 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2572 		break;
2573 	case BPF_CGROUP_INET4_BIND:
2574 	case BPF_CGROUP_INET6_BIND:
2575 	case BPF_CGROUP_INET4_CONNECT:
2576 	case BPF_CGROUP_INET6_CONNECT:
2577 	case BPF_CGROUP_UDP4_SENDMSG:
2578 	case BPF_CGROUP_UDP6_SENDMSG:
2579 	case BPF_CGROUP_UDP4_RECVMSG:
2580 	case BPF_CGROUP_UDP6_RECVMSG:
2581 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2582 		break;
2583 	case BPF_CGROUP_SOCK_OPS:
2584 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2585 		break;
2586 	case BPF_CGROUP_DEVICE:
2587 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2588 		break;
2589 	case BPF_SK_MSG_VERDICT:
2590 		ptype = BPF_PROG_TYPE_SK_MSG;
2591 		break;
2592 	case BPF_SK_SKB_STREAM_PARSER:
2593 	case BPF_SK_SKB_STREAM_VERDICT:
2594 		ptype = BPF_PROG_TYPE_SK_SKB;
2595 		break;
2596 	case BPF_LIRC_MODE2:
2597 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
2598 		break;
2599 	case BPF_FLOW_DISSECTOR:
2600 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2601 		break;
2602 	case BPF_CGROUP_SYSCTL:
2603 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2604 		break;
2605 	case BPF_CGROUP_GETSOCKOPT:
2606 	case BPF_CGROUP_SETSOCKOPT:
2607 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2608 		break;
2609 	default:
2610 		return -EINVAL;
2611 	}
2612 
2613 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2614 	if (IS_ERR(prog))
2615 		return PTR_ERR(prog);
2616 
2617 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2618 		bpf_prog_put(prog);
2619 		return -EINVAL;
2620 	}
2621 
2622 	switch (ptype) {
2623 	case BPF_PROG_TYPE_SK_SKB:
2624 	case BPF_PROG_TYPE_SK_MSG:
2625 		ret = sock_map_get_from_fd(attr, prog);
2626 		break;
2627 	case BPF_PROG_TYPE_LIRC_MODE2:
2628 		ret = lirc_prog_attach(attr, prog);
2629 		break;
2630 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2631 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2632 		break;
2633 	default:
2634 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2635 	}
2636 
2637 	if (ret)
2638 		bpf_prog_put(prog);
2639 	return ret;
2640 }
2641 
2642 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2643 
2644 static int bpf_prog_detach(const union bpf_attr *attr)
2645 {
2646 	enum bpf_prog_type ptype;
2647 
2648 	if (!capable(CAP_NET_ADMIN))
2649 		return -EPERM;
2650 
2651 	if (CHECK_ATTR(BPF_PROG_DETACH))
2652 		return -EINVAL;
2653 
2654 	switch (attr->attach_type) {
2655 	case BPF_CGROUP_INET_INGRESS:
2656 	case BPF_CGROUP_INET_EGRESS:
2657 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2658 		break;
2659 	case BPF_CGROUP_INET_SOCK_CREATE:
2660 	case BPF_CGROUP_INET4_POST_BIND:
2661 	case BPF_CGROUP_INET6_POST_BIND:
2662 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2663 		break;
2664 	case BPF_CGROUP_INET4_BIND:
2665 	case BPF_CGROUP_INET6_BIND:
2666 	case BPF_CGROUP_INET4_CONNECT:
2667 	case BPF_CGROUP_INET6_CONNECT:
2668 	case BPF_CGROUP_UDP4_SENDMSG:
2669 	case BPF_CGROUP_UDP6_SENDMSG:
2670 	case BPF_CGROUP_UDP4_RECVMSG:
2671 	case BPF_CGROUP_UDP6_RECVMSG:
2672 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2673 		break;
2674 	case BPF_CGROUP_SOCK_OPS:
2675 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2676 		break;
2677 	case BPF_CGROUP_DEVICE:
2678 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2679 		break;
2680 	case BPF_SK_MSG_VERDICT:
2681 		return sock_map_get_from_fd(attr, NULL);
2682 	case BPF_SK_SKB_STREAM_PARSER:
2683 	case BPF_SK_SKB_STREAM_VERDICT:
2684 		return sock_map_get_from_fd(attr, NULL);
2685 	case BPF_LIRC_MODE2:
2686 		return lirc_prog_detach(attr);
2687 	case BPF_FLOW_DISSECTOR:
2688 		return skb_flow_dissector_bpf_prog_detach(attr);
2689 	case BPF_CGROUP_SYSCTL:
2690 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2691 		break;
2692 	case BPF_CGROUP_GETSOCKOPT:
2693 	case BPF_CGROUP_SETSOCKOPT:
2694 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2695 		break;
2696 	default:
2697 		return -EINVAL;
2698 	}
2699 
2700 	return cgroup_bpf_prog_detach(attr, ptype);
2701 }
2702 
2703 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2704 
2705 static int bpf_prog_query(const union bpf_attr *attr,
2706 			  union bpf_attr __user *uattr)
2707 {
2708 	if (!capable(CAP_NET_ADMIN))
2709 		return -EPERM;
2710 	if (CHECK_ATTR(BPF_PROG_QUERY))
2711 		return -EINVAL;
2712 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2713 		return -EINVAL;
2714 
2715 	switch (attr->query.attach_type) {
2716 	case BPF_CGROUP_INET_INGRESS:
2717 	case BPF_CGROUP_INET_EGRESS:
2718 	case BPF_CGROUP_INET_SOCK_CREATE:
2719 	case BPF_CGROUP_INET4_BIND:
2720 	case BPF_CGROUP_INET6_BIND:
2721 	case BPF_CGROUP_INET4_POST_BIND:
2722 	case BPF_CGROUP_INET6_POST_BIND:
2723 	case BPF_CGROUP_INET4_CONNECT:
2724 	case BPF_CGROUP_INET6_CONNECT:
2725 	case BPF_CGROUP_UDP4_SENDMSG:
2726 	case BPF_CGROUP_UDP6_SENDMSG:
2727 	case BPF_CGROUP_UDP4_RECVMSG:
2728 	case BPF_CGROUP_UDP6_RECVMSG:
2729 	case BPF_CGROUP_SOCK_OPS:
2730 	case BPF_CGROUP_DEVICE:
2731 	case BPF_CGROUP_SYSCTL:
2732 	case BPF_CGROUP_GETSOCKOPT:
2733 	case BPF_CGROUP_SETSOCKOPT:
2734 		break;
2735 	case BPF_LIRC_MODE2:
2736 		return lirc_prog_query(attr, uattr);
2737 	case BPF_FLOW_DISSECTOR:
2738 		return skb_flow_dissector_prog_query(attr, uattr);
2739 	default:
2740 		return -EINVAL;
2741 	}
2742 
2743 	return cgroup_bpf_prog_query(attr, uattr);
2744 }
2745 
2746 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2747 
2748 static int bpf_prog_test_run(const union bpf_attr *attr,
2749 			     union bpf_attr __user *uattr)
2750 {
2751 	struct bpf_prog *prog;
2752 	int ret = -ENOTSUPP;
2753 
2754 	if (!capable(CAP_SYS_ADMIN))
2755 		return -EPERM;
2756 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2757 		return -EINVAL;
2758 
2759 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2760 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
2761 		return -EINVAL;
2762 
2763 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2764 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
2765 		return -EINVAL;
2766 
2767 	prog = bpf_prog_get(attr->test.prog_fd);
2768 	if (IS_ERR(prog))
2769 		return PTR_ERR(prog);
2770 
2771 	if (prog->aux->ops->test_run)
2772 		ret = prog->aux->ops->test_run(prog, attr, uattr);
2773 
2774 	bpf_prog_put(prog);
2775 	return ret;
2776 }
2777 
2778 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2779 
2780 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2781 			       union bpf_attr __user *uattr,
2782 			       struct idr *idr,
2783 			       spinlock_t *lock)
2784 {
2785 	u32 next_id = attr->start_id;
2786 	int err = 0;
2787 
2788 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2789 		return -EINVAL;
2790 
2791 	if (!capable(CAP_SYS_ADMIN))
2792 		return -EPERM;
2793 
2794 	next_id++;
2795 	spin_lock_bh(lock);
2796 	if (!idr_get_next(idr, &next_id))
2797 		err = -ENOENT;
2798 	spin_unlock_bh(lock);
2799 
2800 	if (!err)
2801 		err = put_user(next_id, &uattr->next_id);
2802 
2803 	return err;
2804 }
2805 
2806 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2807 
2808 struct bpf_prog *bpf_prog_by_id(u32 id)
2809 {
2810 	struct bpf_prog *prog;
2811 
2812 	if (!id)
2813 		return ERR_PTR(-ENOENT);
2814 
2815 	spin_lock_bh(&prog_idr_lock);
2816 	prog = idr_find(&prog_idr, id);
2817 	if (prog)
2818 		prog = bpf_prog_inc_not_zero(prog);
2819 	else
2820 		prog = ERR_PTR(-ENOENT);
2821 	spin_unlock_bh(&prog_idr_lock);
2822 	return prog;
2823 }
2824 
2825 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2826 {
2827 	struct bpf_prog *prog;
2828 	u32 id = attr->prog_id;
2829 	int fd;
2830 
2831 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2832 		return -EINVAL;
2833 
2834 	if (!capable(CAP_SYS_ADMIN))
2835 		return -EPERM;
2836 
2837 	prog = bpf_prog_by_id(id);
2838 	if (IS_ERR(prog))
2839 		return PTR_ERR(prog);
2840 
2841 	fd = bpf_prog_new_fd(prog);
2842 	if (fd < 0)
2843 		bpf_prog_put(prog);
2844 
2845 	return fd;
2846 }
2847 
2848 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2849 
2850 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2851 {
2852 	struct bpf_map *map;
2853 	u32 id = attr->map_id;
2854 	int f_flags;
2855 	int fd;
2856 
2857 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2858 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2859 		return -EINVAL;
2860 
2861 	if (!capable(CAP_SYS_ADMIN))
2862 		return -EPERM;
2863 
2864 	f_flags = bpf_get_file_flag(attr->open_flags);
2865 	if (f_flags < 0)
2866 		return f_flags;
2867 
2868 	spin_lock_bh(&map_idr_lock);
2869 	map = idr_find(&map_idr, id);
2870 	if (map)
2871 		map = __bpf_map_inc_not_zero(map, true);
2872 	else
2873 		map = ERR_PTR(-ENOENT);
2874 	spin_unlock_bh(&map_idr_lock);
2875 
2876 	if (IS_ERR(map))
2877 		return PTR_ERR(map);
2878 
2879 	fd = bpf_map_new_fd(map, f_flags);
2880 	if (fd < 0)
2881 		bpf_map_put_with_uref(map);
2882 
2883 	return fd;
2884 }
2885 
2886 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2887 					      unsigned long addr, u32 *off,
2888 					      u32 *type)
2889 {
2890 	const struct bpf_map *map;
2891 	int i;
2892 
2893 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2894 		map = prog->aux->used_maps[i];
2895 		if (map == (void *)addr) {
2896 			*type = BPF_PSEUDO_MAP_FD;
2897 			return map;
2898 		}
2899 		if (!map->ops->map_direct_value_meta)
2900 			continue;
2901 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
2902 			*type = BPF_PSEUDO_MAP_VALUE;
2903 			return map;
2904 		}
2905 	}
2906 
2907 	return NULL;
2908 }
2909 
2910 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2911 {
2912 	const struct bpf_map *map;
2913 	struct bpf_insn *insns;
2914 	u32 off, type;
2915 	u64 imm;
2916 	int i;
2917 
2918 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2919 			GFP_USER);
2920 	if (!insns)
2921 		return insns;
2922 
2923 	for (i = 0; i < prog->len; i++) {
2924 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2925 			insns[i].code = BPF_JMP | BPF_CALL;
2926 			insns[i].imm = BPF_FUNC_tail_call;
2927 			/* fall-through */
2928 		}
2929 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2930 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2931 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2932 				insns[i].code = BPF_JMP | BPF_CALL;
2933 			if (!bpf_dump_raw_ok())
2934 				insns[i].imm = 0;
2935 			continue;
2936 		}
2937 
2938 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2939 			continue;
2940 
2941 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2942 		map = bpf_map_from_imm(prog, imm, &off, &type);
2943 		if (map) {
2944 			insns[i].src_reg = type;
2945 			insns[i].imm = map->id;
2946 			insns[i + 1].imm = off;
2947 			continue;
2948 		}
2949 	}
2950 
2951 	return insns;
2952 }
2953 
2954 static int set_info_rec_size(struct bpf_prog_info *info)
2955 {
2956 	/*
2957 	 * Ensure info.*_rec_size is the same as kernel expected size
2958 	 *
2959 	 * or
2960 	 *
2961 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2962 	 * zero.  In this case, the kernel will set the expected
2963 	 * _rec_size back to the info.
2964 	 */
2965 
2966 	if ((info->nr_func_info || info->func_info_rec_size) &&
2967 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2968 		return -EINVAL;
2969 
2970 	if ((info->nr_line_info || info->line_info_rec_size) &&
2971 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2972 		return -EINVAL;
2973 
2974 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2975 	    info->jited_line_info_rec_size != sizeof(__u64))
2976 		return -EINVAL;
2977 
2978 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2979 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2980 	info->jited_line_info_rec_size = sizeof(__u64);
2981 
2982 	return 0;
2983 }
2984 
2985 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2986 				   const union bpf_attr *attr,
2987 				   union bpf_attr __user *uattr)
2988 {
2989 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2990 	struct bpf_prog_info info = {};
2991 	u32 info_len = attr->info.info_len;
2992 	struct bpf_prog_stats stats;
2993 	char __user *uinsns;
2994 	u32 ulen;
2995 	int err;
2996 
2997 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2998 	if (err)
2999 		return err;
3000 	info_len = min_t(u32, sizeof(info), info_len);
3001 
3002 	if (copy_from_user(&info, uinfo, info_len))
3003 		return -EFAULT;
3004 
3005 	info.type = prog->type;
3006 	info.id = prog->aux->id;
3007 	info.load_time = prog->aux->load_time;
3008 	info.created_by_uid = from_kuid_munged(current_user_ns(),
3009 					       prog->aux->user->uid);
3010 	info.gpl_compatible = prog->gpl_compatible;
3011 
3012 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
3013 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3014 
3015 	ulen = info.nr_map_ids;
3016 	info.nr_map_ids = prog->aux->used_map_cnt;
3017 	ulen = min_t(u32, info.nr_map_ids, ulen);
3018 	if (ulen) {
3019 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3020 		u32 i;
3021 
3022 		for (i = 0; i < ulen; i++)
3023 			if (put_user(prog->aux->used_maps[i]->id,
3024 				     &user_map_ids[i]))
3025 				return -EFAULT;
3026 	}
3027 
3028 	err = set_info_rec_size(&info);
3029 	if (err)
3030 		return err;
3031 
3032 	bpf_prog_get_stats(prog, &stats);
3033 	info.run_time_ns = stats.nsecs;
3034 	info.run_cnt = stats.cnt;
3035 
3036 	if (!capable(CAP_SYS_ADMIN)) {
3037 		info.jited_prog_len = 0;
3038 		info.xlated_prog_len = 0;
3039 		info.nr_jited_ksyms = 0;
3040 		info.nr_jited_func_lens = 0;
3041 		info.nr_func_info = 0;
3042 		info.nr_line_info = 0;
3043 		info.nr_jited_line_info = 0;
3044 		goto done;
3045 	}
3046 
3047 	ulen = info.xlated_prog_len;
3048 	info.xlated_prog_len = bpf_prog_insn_size(prog);
3049 	if (info.xlated_prog_len && ulen) {
3050 		struct bpf_insn *insns_sanitized;
3051 		bool fault;
3052 
3053 		if (prog->blinded && !bpf_dump_raw_ok()) {
3054 			info.xlated_prog_insns = 0;
3055 			goto done;
3056 		}
3057 		insns_sanitized = bpf_insn_prepare_dump(prog);
3058 		if (!insns_sanitized)
3059 			return -ENOMEM;
3060 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3061 		ulen = min_t(u32, info.xlated_prog_len, ulen);
3062 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
3063 		kfree(insns_sanitized);
3064 		if (fault)
3065 			return -EFAULT;
3066 	}
3067 
3068 	if (bpf_prog_is_dev_bound(prog->aux)) {
3069 		err = bpf_prog_offload_info_fill(&info, prog);
3070 		if (err)
3071 			return err;
3072 		goto done;
3073 	}
3074 
3075 	/* NOTE: the following code is supposed to be skipped for offload.
3076 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
3077 	 * for offload.
3078 	 */
3079 	ulen = info.jited_prog_len;
3080 	if (prog->aux->func_cnt) {
3081 		u32 i;
3082 
3083 		info.jited_prog_len = 0;
3084 		for (i = 0; i < prog->aux->func_cnt; i++)
3085 			info.jited_prog_len += prog->aux->func[i]->jited_len;
3086 	} else {
3087 		info.jited_prog_len = prog->jited_len;
3088 	}
3089 
3090 	if (info.jited_prog_len && ulen) {
3091 		if (bpf_dump_raw_ok()) {
3092 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
3093 			ulen = min_t(u32, info.jited_prog_len, ulen);
3094 
3095 			/* for multi-function programs, copy the JITed
3096 			 * instructions for all the functions
3097 			 */
3098 			if (prog->aux->func_cnt) {
3099 				u32 len, free, i;
3100 				u8 *img;
3101 
3102 				free = ulen;
3103 				for (i = 0; i < prog->aux->func_cnt; i++) {
3104 					len = prog->aux->func[i]->jited_len;
3105 					len = min_t(u32, len, free);
3106 					img = (u8 *) prog->aux->func[i]->bpf_func;
3107 					if (copy_to_user(uinsns, img, len))
3108 						return -EFAULT;
3109 					uinsns += len;
3110 					free -= len;
3111 					if (!free)
3112 						break;
3113 				}
3114 			} else {
3115 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
3116 					return -EFAULT;
3117 			}
3118 		} else {
3119 			info.jited_prog_insns = 0;
3120 		}
3121 	}
3122 
3123 	ulen = info.nr_jited_ksyms;
3124 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3125 	if (ulen) {
3126 		if (bpf_dump_raw_ok()) {
3127 			unsigned long ksym_addr;
3128 			u64 __user *user_ksyms;
3129 			u32 i;
3130 
3131 			/* copy the address of the kernel symbol
3132 			 * corresponding to each function
3133 			 */
3134 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3135 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3136 			if (prog->aux->func_cnt) {
3137 				for (i = 0; i < ulen; i++) {
3138 					ksym_addr = (unsigned long)
3139 						prog->aux->func[i]->bpf_func;
3140 					if (put_user((u64) ksym_addr,
3141 						     &user_ksyms[i]))
3142 						return -EFAULT;
3143 				}
3144 			} else {
3145 				ksym_addr = (unsigned long) prog->bpf_func;
3146 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
3147 					return -EFAULT;
3148 			}
3149 		} else {
3150 			info.jited_ksyms = 0;
3151 		}
3152 	}
3153 
3154 	ulen = info.nr_jited_func_lens;
3155 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3156 	if (ulen) {
3157 		if (bpf_dump_raw_ok()) {
3158 			u32 __user *user_lens;
3159 			u32 func_len, i;
3160 
3161 			/* copy the JITed image lengths for each function */
3162 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3163 			user_lens = u64_to_user_ptr(info.jited_func_lens);
3164 			if (prog->aux->func_cnt) {
3165 				for (i = 0; i < ulen; i++) {
3166 					func_len =
3167 						prog->aux->func[i]->jited_len;
3168 					if (put_user(func_len, &user_lens[i]))
3169 						return -EFAULT;
3170 				}
3171 			} else {
3172 				func_len = prog->jited_len;
3173 				if (put_user(func_len, &user_lens[0]))
3174 					return -EFAULT;
3175 			}
3176 		} else {
3177 			info.jited_func_lens = 0;
3178 		}
3179 	}
3180 
3181 	if (prog->aux->btf)
3182 		info.btf_id = btf_id(prog->aux->btf);
3183 
3184 	ulen = info.nr_func_info;
3185 	info.nr_func_info = prog->aux->func_info_cnt;
3186 	if (info.nr_func_info && ulen) {
3187 		char __user *user_finfo;
3188 
3189 		user_finfo = u64_to_user_ptr(info.func_info);
3190 		ulen = min_t(u32, info.nr_func_info, ulen);
3191 		if (copy_to_user(user_finfo, prog->aux->func_info,
3192 				 info.func_info_rec_size * ulen))
3193 			return -EFAULT;
3194 	}
3195 
3196 	ulen = info.nr_line_info;
3197 	info.nr_line_info = prog->aux->nr_linfo;
3198 	if (info.nr_line_info && ulen) {
3199 		__u8 __user *user_linfo;
3200 
3201 		user_linfo = u64_to_user_ptr(info.line_info);
3202 		ulen = min_t(u32, info.nr_line_info, ulen);
3203 		if (copy_to_user(user_linfo, prog->aux->linfo,
3204 				 info.line_info_rec_size * ulen))
3205 			return -EFAULT;
3206 	}
3207 
3208 	ulen = info.nr_jited_line_info;
3209 	if (prog->aux->jited_linfo)
3210 		info.nr_jited_line_info = prog->aux->nr_linfo;
3211 	else
3212 		info.nr_jited_line_info = 0;
3213 	if (info.nr_jited_line_info && ulen) {
3214 		if (bpf_dump_raw_ok()) {
3215 			__u64 __user *user_linfo;
3216 			u32 i;
3217 
3218 			user_linfo = u64_to_user_ptr(info.jited_line_info);
3219 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
3220 			for (i = 0; i < ulen; i++) {
3221 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3222 					     &user_linfo[i]))
3223 					return -EFAULT;
3224 			}
3225 		} else {
3226 			info.jited_line_info = 0;
3227 		}
3228 	}
3229 
3230 	ulen = info.nr_prog_tags;
3231 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3232 	if (ulen) {
3233 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3234 		u32 i;
3235 
3236 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
3237 		ulen = min_t(u32, info.nr_prog_tags, ulen);
3238 		if (prog->aux->func_cnt) {
3239 			for (i = 0; i < ulen; i++) {
3240 				if (copy_to_user(user_prog_tags[i],
3241 						 prog->aux->func[i]->tag,
3242 						 BPF_TAG_SIZE))
3243 					return -EFAULT;
3244 			}
3245 		} else {
3246 			if (copy_to_user(user_prog_tags[0],
3247 					 prog->tag, BPF_TAG_SIZE))
3248 				return -EFAULT;
3249 		}
3250 	}
3251 
3252 done:
3253 	if (copy_to_user(uinfo, &info, info_len) ||
3254 	    put_user(info_len, &uattr->info.info_len))
3255 		return -EFAULT;
3256 
3257 	return 0;
3258 }
3259 
3260 static int bpf_map_get_info_by_fd(struct bpf_map *map,
3261 				  const union bpf_attr *attr,
3262 				  union bpf_attr __user *uattr)
3263 {
3264 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3265 	struct bpf_map_info info = {};
3266 	u32 info_len = attr->info.info_len;
3267 	int err;
3268 
3269 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3270 	if (err)
3271 		return err;
3272 	info_len = min_t(u32, sizeof(info), info_len);
3273 
3274 	info.type = map->map_type;
3275 	info.id = map->id;
3276 	info.key_size = map->key_size;
3277 	info.value_size = map->value_size;
3278 	info.max_entries = map->max_entries;
3279 	info.map_flags = map->map_flags;
3280 	memcpy(info.name, map->name, sizeof(map->name));
3281 
3282 	if (map->btf) {
3283 		info.btf_id = btf_id(map->btf);
3284 		info.btf_key_type_id = map->btf_key_type_id;
3285 		info.btf_value_type_id = map->btf_value_type_id;
3286 	}
3287 	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3288 
3289 	if (bpf_map_is_dev_bound(map)) {
3290 		err = bpf_map_offload_info_fill(&info, map);
3291 		if (err)
3292 			return err;
3293 	}
3294 
3295 	if (copy_to_user(uinfo, &info, info_len) ||
3296 	    put_user(info_len, &uattr->info.info_len))
3297 		return -EFAULT;
3298 
3299 	return 0;
3300 }
3301 
3302 static int bpf_btf_get_info_by_fd(struct btf *btf,
3303 				  const union bpf_attr *attr,
3304 				  union bpf_attr __user *uattr)
3305 {
3306 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3307 	u32 info_len = attr->info.info_len;
3308 	int err;
3309 
3310 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3311 	if (err)
3312 		return err;
3313 
3314 	return btf_get_info_by_fd(btf, attr, uattr);
3315 }
3316 
3317 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3318 
3319 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3320 				  union bpf_attr __user *uattr)
3321 {
3322 	int ufd = attr->info.bpf_fd;
3323 	struct fd f;
3324 	int err;
3325 
3326 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3327 		return -EINVAL;
3328 
3329 	f = fdget(ufd);
3330 	if (!f.file)
3331 		return -EBADFD;
3332 
3333 	if (f.file->f_op == &bpf_prog_fops)
3334 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
3335 					      uattr);
3336 	else if (f.file->f_op == &bpf_map_fops)
3337 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
3338 					     uattr);
3339 	else if (f.file->f_op == &btf_fops)
3340 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
3341 	else
3342 		err = -EINVAL;
3343 
3344 	fdput(f);
3345 	return err;
3346 }
3347 
3348 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3349 
3350 static int bpf_btf_load(const union bpf_attr *attr)
3351 {
3352 	if (CHECK_ATTR(BPF_BTF_LOAD))
3353 		return -EINVAL;
3354 
3355 	if (!capable(CAP_SYS_ADMIN))
3356 		return -EPERM;
3357 
3358 	return btf_new_fd(attr);
3359 }
3360 
3361 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3362 
3363 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3364 {
3365 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3366 		return -EINVAL;
3367 
3368 	if (!capable(CAP_SYS_ADMIN))
3369 		return -EPERM;
3370 
3371 	return btf_get_fd_by_id(attr->btf_id);
3372 }
3373 
3374 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3375 				    union bpf_attr __user *uattr,
3376 				    u32 prog_id, u32 fd_type,
3377 				    const char *buf, u64 probe_offset,
3378 				    u64 probe_addr)
3379 {
3380 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3381 	u32 len = buf ? strlen(buf) : 0, input_len;
3382 	int err = 0;
3383 
3384 	if (put_user(len, &uattr->task_fd_query.buf_len))
3385 		return -EFAULT;
3386 	input_len = attr->task_fd_query.buf_len;
3387 	if (input_len && ubuf) {
3388 		if (!len) {
3389 			/* nothing to copy, just make ubuf NULL terminated */
3390 			char zero = '\0';
3391 
3392 			if (put_user(zero, ubuf))
3393 				return -EFAULT;
3394 		} else if (input_len >= len + 1) {
3395 			/* ubuf can hold the string with NULL terminator */
3396 			if (copy_to_user(ubuf, buf, len + 1))
3397 				return -EFAULT;
3398 		} else {
3399 			/* ubuf cannot hold the string with NULL terminator,
3400 			 * do a partial copy with NULL terminator.
3401 			 */
3402 			char zero = '\0';
3403 
3404 			err = -ENOSPC;
3405 			if (copy_to_user(ubuf, buf, input_len - 1))
3406 				return -EFAULT;
3407 			if (put_user(zero, ubuf + input_len - 1))
3408 				return -EFAULT;
3409 		}
3410 	}
3411 
3412 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3413 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3414 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3415 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3416 		return -EFAULT;
3417 
3418 	return err;
3419 }
3420 
3421 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3422 
3423 static int bpf_task_fd_query(const union bpf_attr *attr,
3424 			     union bpf_attr __user *uattr)
3425 {
3426 	pid_t pid = attr->task_fd_query.pid;
3427 	u32 fd = attr->task_fd_query.fd;
3428 	const struct perf_event *event;
3429 	struct files_struct *files;
3430 	struct task_struct *task;
3431 	struct file *file;
3432 	int err;
3433 
3434 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3435 		return -EINVAL;
3436 
3437 	if (!capable(CAP_SYS_ADMIN))
3438 		return -EPERM;
3439 
3440 	if (attr->task_fd_query.flags != 0)
3441 		return -EINVAL;
3442 
3443 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3444 	if (!task)
3445 		return -ENOENT;
3446 
3447 	files = get_files_struct(task);
3448 	put_task_struct(task);
3449 	if (!files)
3450 		return -ENOENT;
3451 
3452 	err = 0;
3453 	spin_lock(&files->file_lock);
3454 	file = fcheck_files(files, fd);
3455 	if (!file)
3456 		err = -EBADF;
3457 	else
3458 		get_file(file);
3459 	spin_unlock(&files->file_lock);
3460 	put_files_struct(files);
3461 
3462 	if (err)
3463 		goto out;
3464 
3465 	if (file->f_op == &bpf_link_fops) {
3466 		struct bpf_link *link = file->private_data;
3467 
3468 		if (link->ops == &bpf_raw_tp_lops) {
3469 			struct bpf_raw_tp_link *raw_tp =
3470 				container_of(link, struct bpf_raw_tp_link, link);
3471 			struct bpf_raw_event_map *btp = raw_tp->btp;
3472 
3473 			err = bpf_task_fd_query_copy(attr, uattr,
3474 						     raw_tp->link.prog->aux->id,
3475 						     BPF_FD_TYPE_RAW_TRACEPOINT,
3476 						     btp->tp->name, 0, 0);
3477 			goto put_file;
3478 		}
3479 		goto out_not_supp;
3480 	}
3481 
3482 	event = perf_get_event(file);
3483 	if (!IS_ERR(event)) {
3484 		u64 probe_offset, probe_addr;
3485 		u32 prog_id, fd_type;
3486 		const char *buf;
3487 
3488 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3489 					      &buf, &probe_offset,
3490 					      &probe_addr);
3491 		if (!err)
3492 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3493 						     fd_type, buf,
3494 						     probe_offset,
3495 						     probe_addr);
3496 		goto put_file;
3497 	}
3498 
3499 out_not_supp:
3500 	err = -ENOTSUPP;
3501 put_file:
3502 	fput(file);
3503 out:
3504 	return err;
3505 }
3506 
3507 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3508 
3509 #define BPF_DO_BATCH(fn)			\
3510 	do {					\
3511 		if (!fn) {			\
3512 			err = -ENOTSUPP;	\
3513 			goto err_put;		\
3514 		}				\
3515 		err = fn(map, attr, uattr);	\
3516 	} while (0)
3517 
3518 static int bpf_map_do_batch(const union bpf_attr *attr,
3519 			    union bpf_attr __user *uattr,
3520 			    int cmd)
3521 {
3522 	struct bpf_map *map;
3523 	int err, ufd;
3524 	struct fd f;
3525 
3526 	if (CHECK_ATTR(BPF_MAP_BATCH))
3527 		return -EINVAL;
3528 
3529 	ufd = attr->batch.map_fd;
3530 	f = fdget(ufd);
3531 	map = __bpf_map_get(f);
3532 	if (IS_ERR(map))
3533 		return PTR_ERR(map);
3534 
3535 	if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3536 	     cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3537 	    !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3538 		err = -EPERM;
3539 		goto err_put;
3540 	}
3541 
3542 	if (cmd != BPF_MAP_LOOKUP_BATCH &&
3543 	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3544 		err = -EPERM;
3545 		goto err_put;
3546 	}
3547 
3548 	if (cmd == BPF_MAP_LOOKUP_BATCH)
3549 		BPF_DO_BATCH(map->ops->map_lookup_batch);
3550 	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3551 		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3552 	else if (cmd == BPF_MAP_UPDATE_BATCH)
3553 		BPF_DO_BATCH(map->ops->map_update_batch);
3554 	else
3555 		BPF_DO_BATCH(map->ops->map_delete_batch);
3556 
3557 err_put:
3558 	fdput(f);
3559 	return err;
3560 }
3561 
3562 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3563 {
3564 	union bpf_attr attr = {};
3565 	int err;
3566 
3567 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3568 		return -EPERM;
3569 
3570 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3571 	if (err)
3572 		return err;
3573 	size = min_t(u32, size, sizeof(attr));
3574 
3575 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
3576 	if (copy_from_user(&attr, uattr, size) != 0)
3577 		return -EFAULT;
3578 
3579 	err = security_bpf(cmd, &attr, size);
3580 	if (err < 0)
3581 		return err;
3582 
3583 	switch (cmd) {
3584 	case BPF_MAP_CREATE:
3585 		err = map_create(&attr);
3586 		break;
3587 	case BPF_MAP_LOOKUP_ELEM:
3588 		err = map_lookup_elem(&attr);
3589 		break;
3590 	case BPF_MAP_UPDATE_ELEM:
3591 		err = map_update_elem(&attr);
3592 		break;
3593 	case BPF_MAP_DELETE_ELEM:
3594 		err = map_delete_elem(&attr);
3595 		break;
3596 	case BPF_MAP_GET_NEXT_KEY:
3597 		err = map_get_next_key(&attr);
3598 		break;
3599 	case BPF_MAP_FREEZE:
3600 		err = map_freeze(&attr);
3601 		break;
3602 	case BPF_PROG_LOAD:
3603 		err = bpf_prog_load(&attr, uattr);
3604 		break;
3605 	case BPF_OBJ_PIN:
3606 		err = bpf_obj_pin(&attr);
3607 		break;
3608 	case BPF_OBJ_GET:
3609 		err = bpf_obj_get(&attr);
3610 		break;
3611 	case BPF_PROG_ATTACH:
3612 		err = bpf_prog_attach(&attr);
3613 		break;
3614 	case BPF_PROG_DETACH:
3615 		err = bpf_prog_detach(&attr);
3616 		break;
3617 	case BPF_PROG_QUERY:
3618 		err = bpf_prog_query(&attr, uattr);
3619 		break;
3620 	case BPF_PROG_TEST_RUN:
3621 		err = bpf_prog_test_run(&attr, uattr);
3622 		break;
3623 	case BPF_PROG_GET_NEXT_ID:
3624 		err = bpf_obj_get_next_id(&attr, uattr,
3625 					  &prog_idr, &prog_idr_lock);
3626 		break;
3627 	case BPF_MAP_GET_NEXT_ID:
3628 		err = bpf_obj_get_next_id(&attr, uattr,
3629 					  &map_idr, &map_idr_lock);
3630 		break;
3631 	case BPF_BTF_GET_NEXT_ID:
3632 		err = bpf_obj_get_next_id(&attr, uattr,
3633 					  &btf_idr, &btf_idr_lock);
3634 		break;
3635 	case BPF_PROG_GET_FD_BY_ID:
3636 		err = bpf_prog_get_fd_by_id(&attr);
3637 		break;
3638 	case BPF_MAP_GET_FD_BY_ID:
3639 		err = bpf_map_get_fd_by_id(&attr);
3640 		break;
3641 	case BPF_OBJ_GET_INFO_BY_FD:
3642 		err = bpf_obj_get_info_by_fd(&attr, uattr);
3643 		break;
3644 	case BPF_RAW_TRACEPOINT_OPEN:
3645 		err = bpf_raw_tracepoint_open(&attr);
3646 		break;
3647 	case BPF_BTF_LOAD:
3648 		err = bpf_btf_load(&attr);
3649 		break;
3650 	case BPF_BTF_GET_FD_BY_ID:
3651 		err = bpf_btf_get_fd_by_id(&attr);
3652 		break;
3653 	case BPF_TASK_FD_QUERY:
3654 		err = bpf_task_fd_query(&attr, uattr);
3655 		break;
3656 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3657 		err = map_lookup_and_delete_elem(&attr);
3658 		break;
3659 	case BPF_MAP_LOOKUP_BATCH:
3660 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
3661 		break;
3662 	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
3663 		err = bpf_map_do_batch(&attr, uattr,
3664 				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
3665 		break;
3666 	case BPF_MAP_UPDATE_BATCH:
3667 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
3668 		break;
3669 	case BPF_MAP_DELETE_BATCH:
3670 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
3671 		break;
3672 	default:
3673 		err = -EINVAL;
3674 		break;
3675 	}
3676 
3677 	return err;
3678 }
3679