xref: /linux-6.15/kernel/bpf/syscall.c (revision d3ec10aa)
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 	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
175 	 * inside bpf map update or delete otherwise deadlocks are possible
176 	 */
177 	preempt_disable();
178 	__this_cpu_inc(bpf_prog_active);
179 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
180 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
181 		err = bpf_percpu_hash_update(map, key, value, flags);
182 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
183 		err = bpf_percpu_array_update(map, key, value, flags);
184 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
185 		err = bpf_percpu_cgroup_storage_update(map, key, value,
186 						       flags);
187 	} else if (IS_FD_ARRAY(map)) {
188 		rcu_read_lock();
189 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
190 						   flags);
191 		rcu_read_unlock();
192 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
193 		rcu_read_lock();
194 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
195 						  flags);
196 		rcu_read_unlock();
197 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
198 		/* rcu_read_lock() is not needed */
199 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
200 							 flags);
201 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
202 		   map->map_type == BPF_MAP_TYPE_STACK) {
203 		err = map->ops->map_push_elem(map, value, flags);
204 	} else {
205 		rcu_read_lock();
206 		err = map->ops->map_update_elem(map, key, value, flags);
207 		rcu_read_unlock();
208 	}
209 	__this_cpu_dec(bpf_prog_active);
210 	preempt_enable();
211 	maybe_wait_bpf_programs(map);
212 
213 	return err;
214 }
215 
216 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
217 			      __u64 flags)
218 {
219 	void *ptr;
220 	int err;
221 
222 	if (bpf_map_is_dev_bound(map))
223 		return bpf_map_offload_lookup_elem(map, key, value);
224 
225 	preempt_disable();
226 	this_cpu_inc(bpf_prog_active);
227 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
228 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
229 		err = bpf_percpu_hash_copy(map, key, value);
230 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
231 		err = bpf_percpu_array_copy(map, key, value);
232 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
233 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
234 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
235 		err = bpf_stackmap_copy(map, key, value);
236 	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
237 		err = bpf_fd_array_map_lookup_elem(map, key, value);
238 	} else if (IS_FD_HASH(map)) {
239 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
240 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
241 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
242 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
243 		   map->map_type == BPF_MAP_TYPE_STACK) {
244 		err = map->ops->map_peek_elem(map, value);
245 	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
246 		/* struct_ops map requires directly updating "value" */
247 		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
248 	} else {
249 		rcu_read_lock();
250 		if (map->ops->map_lookup_elem_sys_only)
251 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
252 		else
253 			ptr = map->ops->map_lookup_elem(map, key);
254 		if (IS_ERR(ptr)) {
255 			err = PTR_ERR(ptr);
256 		} else if (!ptr) {
257 			err = -ENOENT;
258 		} else {
259 			err = 0;
260 			if (flags & BPF_F_LOCK)
261 				/* lock 'ptr' and copy everything but lock */
262 				copy_map_value_locked(map, value, ptr, true);
263 			else
264 				copy_map_value(map, value, ptr);
265 			/* mask lock, since value wasn't zero inited */
266 			check_and_init_map_lock(map, value);
267 		}
268 		rcu_read_unlock();
269 	}
270 
271 	this_cpu_dec(bpf_prog_active);
272 	preempt_enable();
273 	maybe_wait_bpf_programs(map);
274 
275 	return err;
276 }
277 
278 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
279 {
280 	/* We really just want to fail instead of triggering OOM killer
281 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
282 	 * which is used for lower order allocation requests.
283 	 *
284 	 * It has been observed that higher order allocation requests done by
285 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
286 	 * to reclaim memory from the page cache, thus we set
287 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
288 	 */
289 
290 	const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
291 	void *area;
292 
293 	if (size >= SIZE_MAX)
294 		return NULL;
295 
296 	/* kmalloc()'ed memory can't be mmap()'ed */
297 	if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
298 		area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
299 				    numa_node);
300 		if (area != NULL)
301 			return area;
302 	}
303 	if (mmapable) {
304 		BUG_ON(!PAGE_ALIGNED(size));
305 		return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
306 					       __GFP_RETRY_MAYFAIL | flags);
307 	}
308 	return __vmalloc_node_flags_caller(size, numa_node,
309 					   GFP_KERNEL | __GFP_RETRY_MAYFAIL |
310 					   flags, __builtin_return_address(0));
311 }
312 
313 void *bpf_map_area_alloc(u64 size, int numa_node)
314 {
315 	return __bpf_map_area_alloc(size, numa_node, false);
316 }
317 
318 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
319 {
320 	return __bpf_map_area_alloc(size, numa_node, true);
321 }
322 
323 void bpf_map_area_free(void *area)
324 {
325 	kvfree(area);
326 }
327 
328 static u32 bpf_map_flags_retain_permanent(u32 flags)
329 {
330 	/* Some map creation flags are not tied to the map object but
331 	 * rather to the map fd instead, so they have no meaning upon
332 	 * map object inspection since multiple file descriptors with
333 	 * different (access) properties can exist here. Thus, given
334 	 * this has zero meaning for the map itself, lets clear these
335 	 * from here.
336 	 */
337 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
338 }
339 
340 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
341 {
342 	map->map_type = attr->map_type;
343 	map->key_size = attr->key_size;
344 	map->value_size = attr->value_size;
345 	map->max_entries = attr->max_entries;
346 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
347 	map->numa_node = bpf_map_attr_numa_node(attr);
348 }
349 
350 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
351 {
352 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
353 
354 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
355 		atomic_long_sub(pages, &user->locked_vm);
356 		return -EPERM;
357 	}
358 	return 0;
359 }
360 
361 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
362 {
363 	if (user)
364 		atomic_long_sub(pages, &user->locked_vm);
365 }
366 
367 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
368 {
369 	u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
370 	struct user_struct *user;
371 	int ret;
372 
373 	if (size >= U32_MAX - PAGE_SIZE)
374 		return -E2BIG;
375 
376 	user = get_current_user();
377 	ret = bpf_charge_memlock(user, pages);
378 	if (ret) {
379 		free_uid(user);
380 		return ret;
381 	}
382 
383 	mem->pages = pages;
384 	mem->user = user;
385 
386 	return 0;
387 }
388 
389 void bpf_map_charge_finish(struct bpf_map_memory *mem)
390 {
391 	bpf_uncharge_memlock(mem->user, mem->pages);
392 	free_uid(mem->user);
393 }
394 
395 void bpf_map_charge_move(struct bpf_map_memory *dst,
396 			 struct bpf_map_memory *src)
397 {
398 	*dst = *src;
399 
400 	/* Make sure src will not be used for the redundant uncharging. */
401 	memset(src, 0, sizeof(struct bpf_map_memory));
402 }
403 
404 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
405 {
406 	int ret;
407 
408 	ret = bpf_charge_memlock(map->memory.user, pages);
409 	if (ret)
410 		return ret;
411 	map->memory.pages += pages;
412 	return ret;
413 }
414 
415 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
416 {
417 	bpf_uncharge_memlock(map->memory.user, pages);
418 	map->memory.pages -= pages;
419 }
420 
421 static int bpf_map_alloc_id(struct bpf_map *map)
422 {
423 	int id;
424 
425 	idr_preload(GFP_KERNEL);
426 	spin_lock_bh(&map_idr_lock);
427 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
428 	if (id > 0)
429 		map->id = id;
430 	spin_unlock_bh(&map_idr_lock);
431 	idr_preload_end();
432 
433 	if (WARN_ON_ONCE(!id))
434 		return -ENOSPC;
435 
436 	return id > 0 ? 0 : id;
437 }
438 
439 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
440 {
441 	unsigned long flags;
442 
443 	/* Offloaded maps are removed from the IDR store when their device
444 	 * disappears - even if someone holds an fd to them they are unusable,
445 	 * the memory is gone, all ops will fail; they are simply waiting for
446 	 * refcnt to drop to be freed.
447 	 */
448 	if (!map->id)
449 		return;
450 
451 	if (do_idr_lock)
452 		spin_lock_irqsave(&map_idr_lock, flags);
453 	else
454 		__acquire(&map_idr_lock);
455 
456 	idr_remove(&map_idr, map->id);
457 	map->id = 0;
458 
459 	if (do_idr_lock)
460 		spin_unlock_irqrestore(&map_idr_lock, flags);
461 	else
462 		__release(&map_idr_lock);
463 }
464 
465 /* called from workqueue */
466 static void bpf_map_free_deferred(struct work_struct *work)
467 {
468 	struct bpf_map *map = container_of(work, struct bpf_map, work);
469 	struct bpf_map_memory mem;
470 
471 	bpf_map_charge_move(&mem, &map->memory);
472 	security_bpf_map_free(map);
473 	/* implementation dependent freeing */
474 	map->ops->map_free(map);
475 	bpf_map_charge_finish(&mem);
476 }
477 
478 static void bpf_map_put_uref(struct bpf_map *map)
479 {
480 	if (atomic64_dec_and_test(&map->usercnt)) {
481 		if (map->ops->map_release_uref)
482 			map->ops->map_release_uref(map);
483 	}
484 }
485 
486 /* decrement map refcnt and schedule it for freeing via workqueue
487  * (unrelying map implementation ops->map_free() might sleep)
488  */
489 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
490 {
491 	if (atomic64_dec_and_test(&map->refcnt)) {
492 		/* bpf_map_free_id() must be called first */
493 		bpf_map_free_id(map, do_idr_lock);
494 		btf_put(map->btf);
495 		INIT_WORK(&map->work, bpf_map_free_deferred);
496 		schedule_work(&map->work);
497 	}
498 }
499 
500 void bpf_map_put(struct bpf_map *map)
501 {
502 	__bpf_map_put(map, true);
503 }
504 EXPORT_SYMBOL_GPL(bpf_map_put);
505 
506 void bpf_map_put_with_uref(struct bpf_map *map)
507 {
508 	bpf_map_put_uref(map);
509 	bpf_map_put(map);
510 }
511 
512 static int bpf_map_release(struct inode *inode, struct file *filp)
513 {
514 	struct bpf_map *map = filp->private_data;
515 
516 	if (map->ops->map_release)
517 		map->ops->map_release(map, filp);
518 
519 	bpf_map_put_with_uref(map);
520 	return 0;
521 }
522 
523 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
524 {
525 	fmode_t mode = f.file->f_mode;
526 
527 	/* Our file permissions may have been overridden by global
528 	 * map permissions facing syscall side.
529 	 */
530 	if (READ_ONCE(map->frozen))
531 		mode &= ~FMODE_CAN_WRITE;
532 	return mode;
533 }
534 
535 #ifdef CONFIG_PROC_FS
536 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
537 {
538 	const struct bpf_map *map = filp->private_data;
539 	const struct bpf_array *array;
540 	u32 type = 0, jited = 0;
541 
542 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
543 		array = container_of(map, struct bpf_array, map);
544 		type  = array->aux->type;
545 		jited = array->aux->jited;
546 	}
547 
548 	seq_printf(m,
549 		   "map_type:\t%u\n"
550 		   "key_size:\t%u\n"
551 		   "value_size:\t%u\n"
552 		   "max_entries:\t%u\n"
553 		   "map_flags:\t%#x\n"
554 		   "memlock:\t%llu\n"
555 		   "map_id:\t%u\n"
556 		   "frozen:\t%u\n",
557 		   map->map_type,
558 		   map->key_size,
559 		   map->value_size,
560 		   map->max_entries,
561 		   map->map_flags,
562 		   map->memory.pages * 1ULL << PAGE_SHIFT,
563 		   map->id,
564 		   READ_ONCE(map->frozen));
565 	if (type) {
566 		seq_printf(m, "owner_prog_type:\t%u\n", type);
567 		seq_printf(m, "owner_jited:\t%u\n", jited);
568 	}
569 }
570 #endif
571 
572 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
573 			      loff_t *ppos)
574 {
575 	/* We need this handler such that alloc_file() enables
576 	 * f_mode with FMODE_CAN_READ.
577 	 */
578 	return -EINVAL;
579 }
580 
581 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
582 			       size_t siz, loff_t *ppos)
583 {
584 	/* We need this handler such that alloc_file() enables
585 	 * f_mode with FMODE_CAN_WRITE.
586 	 */
587 	return -EINVAL;
588 }
589 
590 /* called for any extra memory-mapped regions (except initial) */
591 static void bpf_map_mmap_open(struct vm_area_struct *vma)
592 {
593 	struct bpf_map *map = vma->vm_file->private_data;
594 
595 	bpf_map_inc_with_uref(map);
596 
597 	if (vma->vm_flags & VM_WRITE) {
598 		mutex_lock(&map->freeze_mutex);
599 		map->writecnt++;
600 		mutex_unlock(&map->freeze_mutex);
601 	}
602 }
603 
604 /* called for all unmapped memory region (including initial) */
605 static void bpf_map_mmap_close(struct vm_area_struct *vma)
606 {
607 	struct bpf_map *map = vma->vm_file->private_data;
608 
609 	if (vma->vm_flags & VM_WRITE) {
610 		mutex_lock(&map->freeze_mutex);
611 		map->writecnt--;
612 		mutex_unlock(&map->freeze_mutex);
613 	}
614 
615 	bpf_map_put_with_uref(map);
616 }
617 
618 static const struct vm_operations_struct bpf_map_default_vmops = {
619 	.open		= bpf_map_mmap_open,
620 	.close		= bpf_map_mmap_close,
621 };
622 
623 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
624 {
625 	struct bpf_map *map = filp->private_data;
626 	int err;
627 
628 	if (!map->ops->map_mmap || map_value_has_spin_lock(map))
629 		return -ENOTSUPP;
630 
631 	if (!(vma->vm_flags & VM_SHARED))
632 		return -EINVAL;
633 
634 	mutex_lock(&map->freeze_mutex);
635 
636 	if ((vma->vm_flags & VM_WRITE) && map->frozen) {
637 		err = -EPERM;
638 		goto out;
639 	}
640 
641 	/* set default open/close callbacks */
642 	vma->vm_ops = &bpf_map_default_vmops;
643 	vma->vm_private_data = map;
644 
645 	err = map->ops->map_mmap(map, vma);
646 	if (err)
647 		goto out;
648 
649 	bpf_map_inc_with_uref(map);
650 
651 	if (vma->vm_flags & VM_WRITE)
652 		map->writecnt++;
653 out:
654 	mutex_unlock(&map->freeze_mutex);
655 	return err;
656 }
657 
658 const struct file_operations bpf_map_fops = {
659 #ifdef CONFIG_PROC_FS
660 	.show_fdinfo	= bpf_map_show_fdinfo,
661 #endif
662 	.release	= bpf_map_release,
663 	.read		= bpf_dummy_read,
664 	.write		= bpf_dummy_write,
665 	.mmap		= bpf_map_mmap,
666 };
667 
668 int bpf_map_new_fd(struct bpf_map *map, int flags)
669 {
670 	int ret;
671 
672 	ret = security_bpf_map(map, OPEN_FMODE(flags));
673 	if (ret < 0)
674 		return ret;
675 
676 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
677 				flags | O_CLOEXEC);
678 }
679 
680 int bpf_get_file_flag(int flags)
681 {
682 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
683 		return -EINVAL;
684 	if (flags & BPF_F_RDONLY)
685 		return O_RDONLY;
686 	if (flags & BPF_F_WRONLY)
687 		return O_WRONLY;
688 	return O_RDWR;
689 }
690 
691 /* helper macro to check that unused fields 'union bpf_attr' are zero */
692 #define CHECK_ATTR(CMD) \
693 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
694 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
695 		   sizeof(*attr) - \
696 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
697 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
698 
699 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
700  * Return 0 on success and < 0 on error.
701  */
702 static int bpf_obj_name_cpy(char *dst, const char *src)
703 {
704 	const char *end = src + BPF_OBJ_NAME_LEN;
705 
706 	memset(dst, 0, BPF_OBJ_NAME_LEN);
707 	/* Copy all isalnum(), '_' and '.' chars. */
708 	while (src < end && *src) {
709 		if (!isalnum(*src) &&
710 		    *src != '_' && *src != '.')
711 			return -EINVAL;
712 		*dst++ = *src++;
713 	}
714 
715 	/* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
716 	if (src == end)
717 		return -EINVAL;
718 
719 	return 0;
720 }
721 
722 int map_check_no_btf(const struct bpf_map *map,
723 		     const struct btf *btf,
724 		     const struct btf_type *key_type,
725 		     const struct btf_type *value_type)
726 {
727 	return -ENOTSUPP;
728 }
729 
730 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
731 			 u32 btf_key_id, u32 btf_value_id)
732 {
733 	const struct btf_type *key_type, *value_type;
734 	u32 key_size, value_size;
735 	int ret = 0;
736 
737 	/* Some maps allow key to be unspecified. */
738 	if (btf_key_id) {
739 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
740 		if (!key_type || key_size != map->key_size)
741 			return -EINVAL;
742 	} else {
743 		key_type = btf_type_by_id(btf, 0);
744 		if (!map->ops->map_check_btf)
745 			return -EINVAL;
746 	}
747 
748 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
749 	if (!value_type || value_size != map->value_size)
750 		return -EINVAL;
751 
752 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
753 
754 	if (map_value_has_spin_lock(map)) {
755 		if (map->map_flags & BPF_F_RDONLY_PROG)
756 			return -EACCES;
757 		if (map->map_type != BPF_MAP_TYPE_HASH &&
758 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
759 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
760 		    map->map_type != BPF_MAP_TYPE_SK_STORAGE)
761 			return -ENOTSUPP;
762 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
763 		    map->value_size) {
764 			WARN_ONCE(1,
765 				  "verifier bug spin_lock_off %d value_size %d\n",
766 				  map->spin_lock_off, map->value_size);
767 			return -EFAULT;
768 		}
769 	}
770 
771 	if (map->ops->map_check_btf)
772 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
773 
774 	return ret;
775 }
776 
777 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
778 /* called via syscall */
779 static int map_create(union bpf_attr *attr)
780 {
781 	int numa_node = bpf_map_attr_numa_node(attr);
782 	struct bpf_map_memory mem;
783 	struct bpf_map *map;
784 	int f_flags;
785 	int err;
786 
787 	err = CHECK_ATTR(BPF_MAP_CREATE);
788 	if (err)
789 		return -EINVAL;
790 
791 	if (attr->btf_vmlinux_value_type_id) {
792 		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
793 		    attr->btf_key_type_id || attr->btf_value_type_id)
794 			return -EINVAL;
795 	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
796 		return -EINVAL;
797 	}
798 
799 	f_flags = bpf_get_file_flag(attr->map_flags);
800 	if (f_flags < 0)
801 		return f_flags;
802 
803 	if (numa_node != NUMA_NO_NODE &&
804 	    ((unsigned int)numa_node >= nr_node_ids ||
805 	     !node_online(numa_node)))
806 		return -EINVAL;
807 
808 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
809 	map = find_and_alloc_map(attr);
810 	if (IS_ERR(map))
811 		return PTR_ERR(map);
812 
813 	err = bpf_obj_name_cpy(map->name, attr->map_name);
814 	if (err)
815 		goto free_map;
816 
817 	atomic64_set(&map->refcnt, 1);
818 	atomic64_set(&map->usercnt, 1);
819 	mutex_init(&map->freeze_mutex);
820 
821 	map->spin_lock_off = -EINVAL;
822 	if (attr->btf_key_type_id || attr->btf_value_type_id ||
823 	    /* Even the map's value is a kernel's struct,
824 	     * the bpf_prog.o must have BTF to begin with
825 	     * to figure out the corresponding kernel's
826 	     * counter part.  Thus, attr->btf_fd has
827 	     * to be valid also.
828 	     */
829 	    attr->btf_vmlinux_value_type_id) {
830 		struct btf *btf;
831 
832 		btf = btf_get_by_fd(attr->btf_fd);
833 		if (IS_ERR(btf)) {
834 			err = PTR_ERR(btf);
835 			goto free_map;
836 		}
837 		map->btf = btf;
838 
839 		if (attr->btf_value_type_id) {
840 			err = map_check_btf(map, btf, attr->btf_key_type_id,
841 					    attr->btf_value_type_id);
842 			if (err)
843 				goto free_map;
844 		}
845 
846 		map->btf_key_type_id = attr->btf_key_type_id;
847 		map->btf_value_type_id = attr->btf_value_type_id;
848 		map->btf_vmlinux_value_type_id =
849 			attr->btf_vmlinux_value_type_id;
850 	}
851 
852 	err = security_bpf_map_alloc(map);
853 	if (err)
854 		goto free_map;
855 
856 	err = bpf_map_alloc_id(map);
857 	if (err)
858 		goto free_map_sec;
859 
860 	err = bpf_map_new_fd(map, f_flags);
861 	if (err < 0) {
862 		/* failed to allocate fd.
863 		 * bpf_map_put_with_uref() is needed because the above
864 		 * bpf_map_alloc_id() has published the map
865 		 * to the userspace and the userspace may
866 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
867 		 */
868 		bpf_map_put_with_uref(map);
869 		return err;
870 	}
871 
872 	return err;
873 
874 free_map_sec:
875 	security_bpf_map_free(map);
876 free_map:
877 	btf_put(map->btf);
878 	bpf_map_charge_move(&mem, &map->memory);
879 	map->ops->map_free(map);
880 	bpf_map_charge_finish(&mem);
881 	return err;
882 }
883 
884 /* if error is returned, fd is released.
885  * On success caller should complete fd access with matching fdput()
886  */
887 struct bpf_map *__bpf_map_get(struct fd f)
888 {
889 	if (!f.file)
890 		return ERR_PTR(-EBADF);
891 	if (f.file->f_op != &bpf_map_fops) {
892 		fdput(f);
893 		return ERR_PTR(-EINVAL);
894 	}
895 
896 	return f.file->private_data;
897 }
898 
899 void bpf_map_inc(struct bpf_map *map)
900 {
901 	atomic64_inc(&map->refcnt);
902 }
903 EXPORT_SYMBOL_GPL(bpf_map_inc);
904 
905 void bpf_map_inc_with_uref(struct bpf_map *map)
906 {
907 	atomic64_inc(&map->refcnt);
908 	atomic64_inc(&map->usercnt);
909 }
910 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
911 
912 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
913 {
914 	struct fd f = fdget(ufd);
915 	struct bpf_map *map;
916 
917 	map = __bpf_map_get(f);
918 	if (IS_ERR(map))
919 		return map;
920 
921 	bpf_map_inc_with_uref(map);
922 	fdput(f);
923 
924 	return map;
925 }
926 
927 /* map_idr_lock should have been held */
928 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
929 {
930 	int refold;
931 
932 	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
933 	if (!refold)
934 		return ERR_PTR(-ENOENT);
935 	if (uref)
936 		atomic64_inc(&map->usercnt);
937 
938 	return map;
939 }
940 
941 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
942 {
943 	spin_lock_bh(&map_idr_lock);
944 	map = __bpf_map_inc_not_zero(map, false);
945 	spin_unlock_bh(&map_idr_lock);
946 
947 	return map;
948 }
949 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
950 
951 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
952 {
953 	return -ENOTSUPP;
954 }
955 
956 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
957 {
958 	if (key_size)
959 		return memdup_user(ukey, key_size);
960 
961 	if (ukey)
962 		return ERR_PTR(-EINVAL);
963 
964 	return NULL;
965 }
966 
967 /* last field in 'union bpf_attr' used by this command */
968 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
969 
970 static int map_lookup_elem(union bpf_attr *attr)
971 {
972 	void __user *ukey = u64_to_user_ptr(attr->key);
973 	void __user *uvalue = u64_to_user_ptr(attr->value);
974 	int ufd = attr->map_fd;
975 	struct bpf_map *map;
976 	void *key, *value;
977 	u32 value_size;
978 	struct fd f;
979 	int err;
980 
981 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
982 		return -EINVAL;
983 
984 	if (attr->flags & ~BPF_F_LOCK)
985 		return -EINVAL;
986 
987 	f = fdget(ufd);
988 	map = __bpf_map_get(f);
989 	if (IS_ERR(map))
990 		return PTR_ERR(map);
991 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
992 		err = -EPERM;
993 		goto err_put;
994 	}
995 
996 	if ((attr->flags & BPF_F_LOCK) &&
997 	    !map_value_has_spin_lock(map)) {
998 		err = -EINVAL;
999 		goto err_put;
1000 	}
1001 
1002 	key = __bpf_copy_key(ukey, map->key_size);
1003 	if (IS_ERR(key)) {
1004 		err = PTR_ERR(key);
1005 		goto err_put;
1006 	}
1007 
1008 	value_size = bpf_map_value_size(map);
1009 
1010 	err = -ENOMEM;
1011 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1012 	if (!value)
1013 		goto free_key;
1014 
1015 	err = bpf_map_copy_value(map, key, value, attr->flags);
1016 	if (err)
1017 		goto free_value;
1018 
1019 	err = -EFAULT;
1020 	if (copy_to_user(uvalue, value, value_size) != 0)
1021 		goto free_value;
1022 
1023 	err = 0;
1024 
1025 free_value:
1026 	kfree(value);
1027 free_key:
1028 	kfree(key);
1029 err_put:
1030 	fdput(f);
1031 	return err;
1032 }
1033 
1034 
1035 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1036 
1037 static int map_update_elem(union bpf_attr *attr)
1038 {
1039 	void __user *ukey = u64_to_user_ptr(attr->key);
1040 	void __user *uvalue = u64_to_user_ptr(attr->value);
1041 	int ufd = attr->map_fd;
1042 	struct bpf_map *map;
1043 	void *key, *value;
1044 	u32 value_size;
1045 	struct fd f;
1046 	int err;
1047 
1048 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1049 		return -EINVAL;
1050 
1051 	f = fdget(ufd);
1052 	map = __bpf_map_get(f);
1053 	if (IS_ERR(map))
1054 		return PTR_ERR(map);
1055 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1056 		err = -EPERM;
1057 		goto err_put;
1058 	}
1059 
1060 	if ((attr->flags & BPF_F_LOCK) &&
1061 	    !map_value_has_spin_lock(map)) {
1062 		err = -EINVAL;
1063 		goto err_put;
1064 	}
1065 
1066 	key = __bpf_copy_key(ukey, map->key_size);
1067 	if (IS_ERR(key)) {
1068 		err = PTR_ERR(key);
1069 		goto err_put;
1070 	}
1071 
1072 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1073 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1074 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1075 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1076 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
1077 	else
1078 		value_size = map->value_size;
1079 
1080 	err = -ENOMEM;
1081 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1082 	if (!value)
1083 		goto free_key;
1084 
1085 	err = -EFAULT;
1086 	if (copy_from_user(value, uvalue, value_size) != 0)
1087 		goto free_value;
1088 
1089 	err = bpf_map_update_value(map, f, key, value, attr->flags);
1090 
1091 free_value:
1092 	kfree(value);
1093 free_key:
1094 	kfree(key);
1095 err_put:
1096 	fdput(f);
1097 	return err;
1098 }
1099 
1100 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1101 
1102 static int map_delete_elem(union bpf_attr *attr)
1103 {
1104 	void __user *ukey = u64_to_user_ptr(attr->key);
1105 	int ufd = attr->map_fd;
1106 	struct bpf_map *map;
1107 	struct fd f;
1108 	void *key;
1109 	int err;
1110 
1111 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1112 		return -EINVAL;
1113 
1114 	f = fdget(ufd);
1115 	map = __bpf_map_get(f);
1116 	if (IS_ERR(map))
1117 		return PTR_ERR(map);
1118 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1119 		err = -EPERM;
1120 		goto err_put;
1121 	}
1122 
1123 	key = __bpf_copy_key(ukey, map->key_size);
1124 	if (IS_ERR(key)) {
1125 		err = PTR_ERR(key);
1126 		goto err_put;
1127 	}
1128 
1129 	if (bpf_map_is_dev_bound(map)) {
1130 		err = bpf_map_offload_delete_elem(map, key);
1131 		goto out;
1132 	} else if (IS_FD_PROG_ARRAY(map) ||
1133 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1134 		/* These maps require sleepable context */
1135 		err = map->ops->map_delete_elem(map, key);
1136 		goto out;
1137 	}
1138 
1139 	preempt_disable();
1140 	__this_cpu_inc(bpf_prog_active);
1141 	rcu_read_lock();
1142 	err = map->ops->map_delete_elem(map, key);
1143 	rcu_read_unlock();
1144 	__this_cpu_dec(bpf_prog_active);
1145 	preempt_enable();
1146 	maybe_wait_bpf_programs(map);
1147 out:
1148 	kfree(key);
1149 err_put:
1150 	fdput(f);
1151 	return err;
1152 }
1153 
1154 /* last field in 'union bpf_attr' used by this command */
1155 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1156 
1157 static int map_get_next_key(union bpf_attr *attr)
1158 {
1159 	void __user *ukey = u64_to_user_ptr(attr->key);
1160 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1161 	int ufd = attr->map_fd;
1162 	struct bpf_map *map;
1163 	void *key, *next_key;
1164 	struct fd f;
1165 	int err;
1166 
1167 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1168 		return -EINVAL;
1169 
1170 	f = fdget(ufd);
1171 	map = __bpf_map_get(f);
1172 	if (IS_ERR(map))
1173 		return PTR_ERR(map);
1174 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1175 		err = -EPERM;
1176 		goto err_put;
1177 	}
1178 
1179 	if (ukey) {
1180 		key = __bpf_copy_key(ukey, map->key_size);
1181 		if (IS_ERR(key)) {
1182 			err = PTR_ERR(key);
1183 			goto err_put;
1184 		}
1185 	} else {
1186 		key = NULL;
1187 	}
1188 
1189 	err = -ENOMEM;
1190 	next_key = kmalloc(map->key_size, GFP_USER);
1191 	if (!next_key)
1192 		goto free_key;
1193 
1194 	if (bpf_map_is_dev_bound(map)) {
1195 		err = bpf_map_offload_get_next_key(map, key, next_key);
1196 		goto out;
1197 	}
1198 
1199 	rcu_read_lock();
1200 	err = map->ops->map_get_next_key(map, key, next_key);
1201 	rcu_read_unlock();
1202 out:
1203 	if (err)
1204 		goto free_next_key;
1205 
1206 	err = -EFAULT;
1207 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1208 		goto free_next_key;
1209 
1210 	err = 0;
1211 
1212 free_next_key:
1213 	kfree(next_key);
1214 free_key:
1215 	kfree(key);
1216 err_put:
1217 	fdput(f);
1218 	return err;
1219 }
1220 
1221 int generic_map_delete_batch(struct bpf_map *map,
1222 			     const union bpf_attr *attr,
1223 			     union bpf_attr __user *uattr)
1224 {
1225 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1226 	u32 cp, max_count;
1227 	int err = 0;
1228 	void *key;
1229 
1230 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1231 		return -EINVAL;
1232 
1233 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1234 	    !map_value_has_spin_lock(map)) {
1235 		return -EINVAL;
1236 	}
1237 
1238 	max_count = attr->batch.count;
1239 	if (!max_count)
1240 		return 0;
1241 
1242 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1243 	if (!key)
1244 		return -ENOMEM;
1245 
1246 	for (cp = 0; cp < max_count; cp++) {
1247 		err = -EFAULT;
1248 		if (copy_from_user(key, keys + cp * map->key_size,
1249 				   map->key_size))
1250 			break;
1251 
1252 		if (bpf_map_is_dev_bound(map)) {
1253 			err = bpf_map_offload_delete_elem(map, key);
1254 			break;
1255 		}
1256 
1257 		preempt_disable();
1258 		__this_cpu_inc(bpf_prog_active);
1259 		rcu_read_lock();
1260 		err = map->ops->map_delete_elem(map, key);
1261 		rcu_read_unlock();
1262 		__this_cpu_dec(bpf_prog_active);
1263 		preempt_enable();
1264 		maybe_wait_bpf_programs(map);
1265 		if (err)
1266 			break;
1267 	}
1268 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1269 		err = -EFAULT;
1270 
1271 	kfree(key);
1272 	return err;
1273 }
1274 
1275 int generic_map_update_batch(struct bpf_map *map,
1276 			     const union bpf_attr *attr,
1277 			     union bpf_attr __user *uattr)
1278 {
1279 	void __user *values = u64_to_user_ptr(attr->batch.values);
1280 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1281 	u32 value_size, cp, max_count;
1282 	int ufd = attr->map_fd;
1283 	void *key, *value;
1284 	struct fd f;
1285 	int err = 0;
1286 
1287 	f = fdget(ufd);
1288 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1289 		return -EINVAL;
1290 
1291 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1292 	    !map_value_has_spin_lock(map)) {
1293 		return -EINVAL;
1294 	}
1295 
1296 	value_size = bpf_map_value_size(map);
1297 
1298 	max_count = attr->batch.count;
1299 	if (!max_count)
1300 		return 0;
1301 
1302 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1303 	if (!key)
1304 		return -ENOMEM;
1305 
1306 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1307 	if (!value) {
1308 		kfree(key);
1309 		return -ENOMEM;
1310 	}
1311 
1312 	for (cp = 0; cp < max_count; cp++) {
1313 		err = -EFAULT;
1314 		if (copy_from_user(key, keys + cp * map->key_size,
1315 		    map->key_size) ||
1316 		    copy_from_user(value, values + cp * value_size, value_size))
1317 			break;
1318 
1319 		err = bpf_map_update_value(map, f, key, value,
1320 					   attr->batch.elem_flags);
1321 
1322 		if (err)
1323 			break;
1324 	}
1325 
1326 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1327 		err = -EFAULT;
1328 
1329 	kfree(value);
1330 	kfree(key);
1331 	return err;
1332 }
1333 
1334 #define MAP_LOOKUP_RETRIES 3
1335 
1336 int generic_map_lookup_batch(struct bpf_map *map,
1337 				    const union bpf_attr *attr,
1338 				    union bpf_attr __user *uattr)
1339 {
1340 	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1341 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1342 	void __user *values = u64_to_user_ptr(attr->batch.values);
1343 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1344 	void *buf, *buf_prevkey, *prev_key, *key, *value;
1345 	int err, retry = MAP_LOOKUP_RETRIES;
1346 	u32 value_size, cp, max_count;
1347 
1348 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1349 		return -EINVAL;
1350 
1351 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1352 	    !map_value_has_spin_lock(map))
1353 		return -EINVAL;
1354 
1355 	value_size = bpf_map_value_size(map);
1356 
1357 	max_count = attr->batch.count;
1358 	if (!max_count)
1359 		return 0;
1360 
1361 	if (put_user(0, &uattr->batch.count))
1362 		return -EFAULT;
1363 
1364 	buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1365 	if (!buf_prevkey)
1366 		return -ENOMEM;
1367 
1368 	buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1369 	if (!buf) {
1370 		kvfree(buf_prevkey);
1371 		return -ENOMEM;
1372 	}
1373 
1374 	err = -EFAULT;
1375 	prev_key = NULL;
1376 	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1377 		goto free_buf;
1378 	key = buf;
1379 	value = key + map->key_size;
1380 	if (ubatch)
1381 		prev_key = buf_prevkey;
1382 
1383 	for (cp = 0; cp < max_count;) {
1384 		rcu_read_lock();
1385 		err = map->ops->map_get_next_key(map, prev_key, key);
1386 		rcu_read_unlock();
1387 		if (err)
1388 			break;
1389 		err = bpf_map_copy_value(map, key, value,
1390 					 attr->batch.elem_flags);
1391 
1392 		if (err == -ENOENT) {
1393 			if (retry) {
1394 				retry--;
1395 				continue;
1396 			}
1397 			err = -EINTR;
1398 			break;
1399 		}
1400 
1401 		if (err)
1402 			goto free_buf;
1403 
1404 		if (copy_to_user(keys + cp * map->key_size, key,
1405 				 map->key_size)) {
1406 			err = -EFAULT;
1407 			goto free_buf;
1408 		}
1409 		if (copy_to_user(values + cp * value_size, value, value_size)) {
1410 			err = -EFAULT;
1411 			goto free_buf;
1412 		}
1413 
1414 		if (!prev_key)
1415 			prev_key = buf_prevkey;
1416 
1417 		swap(prev_key, key);
1418 		retry = MAP_LOOKUP_RETRIES;
1419 		cp++;
1420 	}
1421 
1422 	if (err == -EFAULT)
1423 		goto free_buf;
1424 
1425 	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1426 		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1427 		err = -EFAULT;
1428 
1429 free_buf:
1430 	kfree(buf_prevkey);
1431 	kfree(buf);
1432 	return err;
1433 }
1434 
1435 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1436 
1437 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1438 {
1439 	void __user *ukey = u64_to_user_ptr(attr->key);
1440 	void __user *uvalue = u64_to_user_ptr(attr->value);
1441 	int ufd = attr->map_fd;
1442 	struct bpf_map *map;
1443 	void *key, *value;
1444 	u32 value_size;
1445 	struct fd f;
1446 	int err;
1447 
1448 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1449 		return -EINVAL;
1450 
1451 	f = fdget(ufd);
1452 	map = __bpf_map_get(f);
1453 	if (IS_ERR(map))
1454 		return PTR_ERR(map);
1455 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1456 		err = -EPERM;
1457 		goto err_put;
1458 	}
1459 
1460 	key = __bpf_copy_key(ukey, map->key_size);
1461 	if (IS_ERR(key)) {
1462 		err = PTR_ERR(key);
1463 		goto err_put;
1464 	}
1465 
1466 	value_size = map->value_size;
1467 
1468 	err = -ENOMEM;
1469 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1470 	if (!value)
1471 		goto free_key;
1472 
1473 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1474 	    map->map_type == BPF_MAP_TYPE_STACK) {
1475 		err = map->ops->map_pop_elem(map, value);
1476 	} else {
1477 		err = -ENOTSUPP;
1478 	}
1479 
1480 	if (err)
1481 		goto free_value;
1482 
1483 	if (copy_to_user(uvalue, value, value_size) != 0)
1484 		goto free_value;
1485 
1486 	err = 0;
1487 
1488 free_value:
1489 	kfree(value);
1490 free_key:
1491 	kfree(key);
1492 err_put:
1493 	fdput(f);
1494 	return err;
1495 }
1496 
1497 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1498 
1499 static int map_freeze(const union bpf_attr *attr)
1500 {
1501 	int err = 0, ufd = attr->map_fd;
1502 	struct bpf_map *map;
1503 	struct fd f;
1504 
1505 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1506 		return -EINVAL;
1507 
1508 	f = fdget(ufd);
1509 	map = __bpf_map_get(f);
1510 	if (IS_ERR(map))
1511 		return PTR_ERR(map);
1512 
1513 	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1514 		fdput(f);
1515 		return -ENOTSUPP;
1516 	}
1517 
1518 	mutex_lock(&map->freeze_mutex);
1519 
1520 	if (map->writecnt) {
1521 		err = -EBUSY;
1522 		goto err_put;
1523 	}
1524 	if (READ_ONCE(map->frozen)) {
1525 		err = -EBUSY;
1526 		goto err_put;
1527 	}
1528 	if (!capable(CAP_SYS_ADMIN)) {
1529 		err = -EPERM;
1530 		goto err_put;
1531 	}
1532 
1533 	WRITE_ONCE(map->frozen, true);
1534 err_put:
1535 	mutex_unlock(&map->freeze_mutex);
1536 	fdput(f);
1537 	return err;
1538 }
1539 
1540 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1541 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1542 	[_id] = & _name ## _prog_ops,
1543 #define BPF_MAP_TYPE(_id, _ops)
1544 #include <linux/bpf_types.h>
1545 #undef BPF_PROG_TYPE
1546 #undef BPF_MAP_TYPE
1547 };
1548 
1549 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1550 {
1551 	const struct bpf_prog_ops *ops;
1552 
1553 	if (type >= ARRAY_SIZE(bpf_prog_types))
1554 		return -EINVAL;
1555 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1556 	ops = bpf_prog_types[type];
1557 	if (!ops)
1558 		return -EINVAL;
1559 
1560 	if (!bpf_prog_is_dev_bound(prog->aux))
1561 		prog->aux->ops = ops;
1562 	else
1563 		prog->aux->ops = &bpf_offload_prog_ops;
1564 	prog->type = type;
1565 	return 0;
1566 }
1567 
1568 enum bpf_audit {
1569 	BPF_AUDIT_LOAD,
1570 	BPF_AUDIT_UNLOAD,
1571 	BPF_AUDIT_MAX,
1572 };
1573 
1574 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1575 	[BPF_AUDIT_LOAD]   = "LOAD",
1576 	[BPF_AUDIT_UNLOAD] = "UNLOAD",
1577 };
1578 
1579 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1580 {
1581 	struct audit_context *ctx = NULL;
1582 	struct audit_buffer *ab;
1583 
1584 	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1585 		return;
1586 	if (audit_enabled == AUDIT_OFF)
1587 		return;
1588 	if (op == BPF_AUDIT_LOAD)
1589 		ctx = audit_context();
1590 	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1591 	if (unlikely(!ab))
1592 		return;
1593 	audit_log_format(ab, "prog-id=%u op=%s",
1594 			 prog->aux->id, bpf_audit_str[op]);
1595 	audit_log_end(ab);
1596 }
1597 
1598 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1599 {
1600 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1601 	unsigned long user_bufs;
1602 
1603 	if (user) {
1604 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1605 		if (user_bufs > memlock_limit) {
1606 			atomic_long_sub(pages, &user->locked_vm);
1607 			return -EPERM;
1608 		}
1609 	}
1610 
1611 	return 0;
1612 }
1613 
1614 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1615 {
1616 	if (user)
1617 		atomic_long_sub(pages, &user->locked_vm);
1618 }
1619 
1620 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1621 {
1622 	struct user_struct *user = get_current_user();
1623 	int ret;
1624 
1625 	ret = __bpf_prog_charge(user, prog->pages);
1626 	if (ret) {
1627 		free_uid(user);
1628 		return ret;
1629 	}
1630 
1631 	prog->aux->user = user;
1632 	return 0;
1633 }
1634 
1635 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1636 {
1637 	struct user_struct *user = prog->aux->user;
1638 
1639 	__bpf_prog_uncharge(user, prog->pages);
1640 	free_uid(user);
1641 }
1642 
1643 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1644 {
1645 	int id;
1646 
1647 	idr_preload(GFP_KERNEL);
1648 	spin_lock_bh(&prog_idr_lock);
1649 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1650 	if (id > 0)
1651 		prog->aux->id = id;
1652 	spin_unlock_bh(&prog_idr_lock);
1653 	idr_preload_end();
1654 
1655 	/* id is in [1, INT_MAX) */
1656 	if (WARN_ON_ONCE(!id))
1657 		return -ENOSPC;
1658 
1659 	return id > 0 ? 0 : id;
1660 }
1661 
1662 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1663 {
1664 	/* cBPF to eBPF migrations are currently not in the idr store.
1665 	 * Offloaded programs are removed from the store when their device
1666 	 * disappears - even if someone grabs an fd to them they are unusable,
1667 	 * simply waiting for refcnt to drop to be freed.
1668 	 */
1669 	if (!prog->aux->id)
1670 		return;
1671 
1672 	if (do_idr_lock)
1673 		spin_lock_bh(&prog_idr_lock);
1674 	else
1675 		__acquire(&prog_idr_lock);
1676 
1677 	idr_remove(&prog_idr, prog->aux->id);
1678 	prog->aux->id = 0;
1679 
1680 	if (do_idr_lock)
1681 		spin_unlock_bh(&prog_idr_lock);
1682 	else
1683 		__release(&prog_idr_lock);
1684 }
1685 
1686 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1687 {
1688 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1689 
1690 	kvfree(aux->func_info);
1691 	kfree(aux->func_info_aux);
1692 	bpf_prog_uncharge_memlock(aux->prog);
1693 	security_bpf_prog_free(aux);
1694 	bpf_prog_free(aux->prog);
1695 }
1696 
1697 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1698 {
1699 	bpf_prog_kallsyms_del_all(prog);
1700 	btf_put(prog->aux->btf);
1701 	bpf_prog_free_linfo(prog);
1702 
1703 	if (deferred)
1704 		call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1705 	else
1706 		__bpf_prog_put_rcu(&prog->aux->rcu);
1707 }
1708 
1709 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1710 {
1711 	if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1712 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1713 		bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1714 		/* bpf_prog_free_id() must be called first */
1715 		bpf_prog_free_id(prog, do_idr_lock);
1716 		__bpf_prog_put_noref(prog, true);
1717 	}
1718 }
1719 
1720 void bpf_prog_put(struct bpf_prog *prog)
1721 {
1722 	__bpf_prog_put(prog, true);
1723 }
1724 EXPORT_SYMBOL_GPL(bpf_prog_put);
1725 
1726 static int bpf_prog_release(struct inode *inode, struct file *filp)
1727 {
1728 	struct bpf_prog *prog = filp->private_data;
1729 
1730 	bpf_prog_put(prog);
1731 	return 0;
1732 }
1733 
1734 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1735 			       struct bpf_prog_stats *stats)
1736 {
1737 	u64 nsecs = 0, cnt = 0;
1738 	int cpu;
1739 
1740 	for_each_possible_cpu(cpu) {
1741 		const struct bpf_prog_stats *st;
1742 		unsigned int start;
1743 		u64 tnsecs, tcnt;
1744 
1745 		st = per_cpu_ptr(prog->aux->stats, cpu);
1746 		do {
1747 			start = u64_stats_fetch_begin_irq(&st->syncp);
1748 			tnsecs = st->nsecs;
1749 			tcnt = st->cnt;
1750 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1751 		nsecs += tnsecs;
1752 		cnt += tcnt;
1753 	}
1754 	stats->nsecs = nsecs;
1755 	stats->cnt = cnt;
1756 }
1757 
1758 #ifdef CONFIG_PROC_FS
1759 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1760 {
1761 	const struct bpf_prog *prog = filp->private_data;
1762 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1763 	struct bpf_prog_stats stats;
1764 
1765 	bpf_prog_get_stats(prog, &stats);
1766 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1767 	seq_printf(m,
1768 		   "prog_type:\t%u\n"
1769 		   "prog_jited:\t%u\n"
1770 		   "prog_tag:\t%s\n"
1771 		   "memlock:\t%llu\n"
1772 		   "prog_id:\t%u\n"
1773 		   "run_time_ns:\t%llu\n"
1774 		   "run_cnt:\t%llu\n",
1775 		   prog->type,
1776 		   prog->jited,
1777 		   prog_tag,
1778 		   prog->pages * 1ULL << PAGE_SHIFT,
1779 		   prog->aux->id,
1780 		   stats.nsecs,
1781 		   stats.cnt);
1782 }
1783 #endif
1784 
1785 const struct file_operations bpf_prog_fops = {
1786 #ifdef CONFIG_PROC_FS
1787 	.show_fdinfo	= bpf_prog_show_fdinfo,
1788 #endif
1789 	.release	= bpf_prog_release,
1790 	.read		= bpf_dummy_read,
1791 	.write		= bpf_dummy_write,
1792 };
1793 
1794 int bpf_prog_new_fd(struct bpf_prog *prog)
1795 {
1796 	int ret;
1797 
1798 	ret = security_bpf_prog(prog);
1799 	if (ret < 0)
1800 		return ret;
1801 
1802 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1803 				O_RDWR | O_CLOEXEC);
1804 }
1805 
1806 static struct bpf_prog *____bpf_prog_get(struct fd f)
1807 {
1808 	if (!f.file)
1809 		return ERR_PTR(-EBADF);
1810 	if (f.file->f_op != &bpf_prog_fops) {
1811 		fdput(f);
1812 		return ERR_PTR(-EINVAL);
1813 	}
1814 
1815 	return f.file->private_data;
1816 }
1817 
1818 void bpf_prog_add(struct bpf_prog *prog, int i)
1819 {
1820 	atomic64_add(i, &prog->aux->refcnt);
1821 }
1822 EXPORT_SYMBOL_GPL(bpf_prog_add);
1823 
1824 void bpf_prog_sub(struct bpf_prog *prog, int i)
1825 {
1826 	/* Only to be used for undoing previous bpf_prog_add() in some
1827 	 * error path. We still know that another entity in our call
1828 	 * path holds a reference to the program, thus atomic_sub() can
1829 	 * be safely used in such cases!
1830 	 */
1831 	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1832 }
1833 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1834 
1835 void bpf_prog_inc(struct bpf_prog *prog)
1836 {
1837 	atomic64_inc(&prog->aux->refcnt);
1838 }
1839 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1840 
1841 /* prog_idr_lock should have been held */
1842 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1843 {
1844 	int refold;
1845 
1846 	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1847 
1848 	if (!refold)
1849 		return ERR_PTR(-ENOENT);
1850 
1851 	return prog;
1852 }
1853 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1854 
1855 bool bpf_prog_get_ok(struct bpf_prog *prog,
1856 			    enum bpf_prog_type *attach_type, bool attach_drv)
1857 {
1858 	/* not an attachment, just a refcount inc, always allow */
1859 	if (!attach_type)
1860 		return true;
1861 
1862 	if (prog->type != *attach_type)
1863 		return false;
1864 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1865 		return false;
1866 
1867 	return true;
1868 }
1869 
1870 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1871 				       bool attach_drv)
1872 {
1873 	struct fd f = fdget(ufd);
1874 	struct bpf_prog *prog;
1875 
1876 	prog = ____bpf_prog_get(f);
1877 	if (IS_ERR(prog))
1878 		return prog;
1879 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1880 		prog = ERR_PTR(-EINVAL);
1881 		goto out;
1882 	}
1883 
1884 	bpf_prog_inc(prog);
1885 out:
1886 	fdput(f);
1887 	return prog;
1888 }
1889 
1890 struct bpf_prog *bpf_prog_get(u32 ufd)
1891 {
1892 	return __bpf_prog_get(ufd, NULL, false);
1893 }
1894 
1895 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1896 				       bool attach_drv)
1897 {
1898 	return __bpf_prog_get(ufd, &type, attach_drv);
1899 }
1900 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1901 
1902 /* Initially all BPF programs could be loaded w/o specifying
1903  * expected_attach_type. Later for some of them specifying expected_attach_type
1904  * at load time became required so that program could be validated properly.
1905  * Programs of types that are allowed to be loaded both w/ and w/o (for
1906  * backward compatibility) expected_attach_type, should have the default attach
1907  * type assigned to expected_attach_type for the latter case, so that it can be
1908  * validated later at attach time.
1909  *
1910  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1911  * prog type requires it but has some attach types that have to be backward
1912  * compatible.
1913  */
1914 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1915 {
1916 	switch (attr->prog_type) {
1917 	case BPF_PROG_TYPE_CGROUP_SOCK:
1918 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1919 		 * exist so checking for non-zero is the way to go here.
1920 		 */
1921 		if (!attr->expected_attach_type)
1922 			attr->expected_attach_type =
1923 				BPF_CGROUP_INET_SOCK_CREATE;
1924 		break;
1925 	}
1926 }
1927 
1928 static int
1929 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1930 			   enum bpf_attach_type expected_attach_type,
1931 			   u32 btf_id, u32 prog_fd)
1932 {
1933 	if (btf_id) {
1934 		if (btf_id > BTF_MAX_TYPE)
1935 			return -EINVAL;
1936 
1937 		switch (prog_type) {
1938 		case BPF_PROG_TYPE_TRACING:
1939 		case BPF_PROG_TYPE_STRUCT_OPS:
1940 		case BPF_PROG_TYPE_EXT:
1941 			break;
1942 		default:
1943 			return -EINVAL;
1944 		}
1945 	}
1946 
1947 	if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
1948 	    prog_type != BPF_PROG_TYPE_EXT)
1949 		return -EINVAL;
1950 
1951 	switch (prog_type) {
1952 	case BPF_PROG_TYPE_CGROUP_SOCK:
1953 		switch (expected_attach_type) {
1954 		case BPF_CGROUP_INET_SOCK_CREATE:
1955 		case BPF_CGROUP_INET4_POST_BIND:
1956 		case BPF_CGROUP_INET6_POST_BIND:
1957 			return 0;
1958 		default:
1959 			return -EINVAL;
1960 		}
1961 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1962 		switch (expected_attach_type) {
1963 		case BPF_CGROUP_INET4_BIND:
1964 		case BPF_CGROUP_INET6_BIND:
1965 		case BPF_CGROUP_INET4_CONNECT:
1966 		case BPF_CGROUP_INET6_CONNECT:
1967 		case BPF_CGROUP_UDP4_SENDMSG:
1968 		case BPF_CGROUP_UDP6_SENDMSG:
1969 		case BPF_CGROUP_UDP4_RECVMSG:
1970 		case BPF_CGROUP_UDP6_RECVMSG:
1971 			return 0;
1972 		default:
1973 			return -EINVAL;
1974 		}
1975 	case BPF_PROG_TYPE_CGROUP_SKB:
1976 		switch (expected_attach_type) {
1977 		case BPF_CGROUP_INET_INGRESS:
1978 		case BPF_CGROUP_INET_EGRESS:
1979 			return 0;
1980 		default:
1981 			return -EINVAL;
1982 		}
1983 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1984 		switch (expected_attach_type) {
1985 		case BPF_CGROUP_SETSOCKOPT:
1986 		case BPF_CGROUP_GETSOCKOPT:
1987 			return 0;
1988 		default:
1989 			return -EINVAL;
1990 		}
1991 	case BPF_PROG_TYPE_EXT:
1992 		if (expected_attach_type)
1993 			return -EINVAL;
1994 		/* fallthrough */
1995 	default:
1996 		return 0;
1997 	}
1998 }
1999 
2000 /* last field in 'union bpf_attr' used by this command */
2001 #define	BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2002 
2003 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2004 {
2005 	enum bpf_prog_type type = attr->prog_type;
2006 	struct bpf_prog *prog;
2007 	int err;
2008 	char license[128];
2009 	bool is_gpl;
2010 
2011 	if (CHECK_ATTR(BPF_PROG_LOAD))
2012 		return -EINVAL;
2013 
2014 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2015 				 BPF_F_ANY_ALIGNMENT |
2016 				 BPF_F_TEST_STATE_FREQ |
2017 				 BPF_F_TEST_RND_HI32))
2018 		return -EINVAL;
2019 
2020 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2021 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2022 	    !capable(CAP_SYS_ADMIN))
2023 		return -EPERM;
2024 
2025 	/* copy eBPF program license from user space */
2026 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2027 			      sizeof(license) - 1) < 0)
2028 		return -EFAULT;
2029 	license[sizeof(license) - 1] = 0;
2030 
2031 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2032 	is_gpl = license_is_gpl_compatible(license);
2033 
2034 	if (attr->insn_cnt == 0 ||
2035 	    attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2036 		return -E2BIG;
2037 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2038 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2039 	    !capable(CAP_SYS_ADMIN))
2040 		return -EPERM;
2041 
2042 	bpf_prog_load_fixup_attach_type(attr);
2043 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2044 				       attr->attach_btf_id,
2045 				       attr->attach_prog_fd))
2046 		return -EINVAL;
2047 
2048 	/* plain bpf_prog allocation */
2049 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2050 	if (!prog)
2051 		return -ENOMEM;
2052 
2053 	prog->expected_attach_type = attr->expected_attach_type;
2054 	prog->aux->attach_btf_id = attr->attach_btf_id;
2055 	if (attr->attach_prog_fd) {
2056 		struct bpf_prog *tgt_prog;
2057 
2058 		tgt_prog = bpf_prog_get(attr->attach_prog_fd);
2059 		if (IS_ERR(tgt_prog)) {
2060 			err = PTR_ERR(tgt_prog);
2061 			goto free_prog_nouncharge;
2062 		}
2063 		prog->aux->linked_prog = tgt_prog;
2064 	}
2065 
2066 	prog->aux->offload_requested = !!attr->prog_ifindex;
2067 
2068 	err = security_bpf_prog_alloc(prog->aux);
2069 	if (err)
2070 		goto free_prog_nouncharge;
2071 
2072 	err = bpf_prog_charge_memlock(prog);
2073 	if (err)
2074 		goto free_prog_sec;
2075 
2076 	prog->len = attr->insn_cnt;
2077 
2078 	err = -EFAULT;
2079 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2080 			   bpf_prog_insn_size(prog)) != 0)
2081 		goto free_prog;
2082 
2083 	prog->orig_prog = NULL;
2084 	prog->jited = 0;
2085 
2086 	atomic64_set(&prog->aux->refcnt, 1);
2087 	prog->gpl_compatible = is_gpl ? 1 : 0;
2088 
2089 	if (bpf_prog_is_dev_bound(prog->aux)) {
2090 		err = bpf_prog_offload_init(prog, attr);
2091 		if (err)
2092 			goto free_prog;
2093 	}
2094 
2095 	/* find program type: socket_filter vs tracing_filter */
2096 	err = find_prog_type(type, prog);
2097 	if (err < 0)
2098 		goto free_prog;
2099 
2100 	prog->aux->load_time = ktime_get_boottime_ns();
2101 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
2102 	if (err)
2103 		goto free_prog;
2104 
2105 	/* run eBPF verifier */
2106 	err = bpf_check(&prog, attr, uattr);
2107 	if (err < 0)
2108 		goto free_used_maps;
2109 
2110 	prog = bpf_prog_select_runtime(prog, &err);
2111 	if (err < 0)
2112 		goto free_used_maps;
2113 
2114 	err = bpf_prog_alloc_id(prog);
2115 	if (err)
2116 		goto free_used_maps;
2117 
2118 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2119 	 * effectively publicly exposed. However, retrieving via
2120 	 * bpf_prog_get_fd_by_id() will take another reference,
2121 	 * therefore it cannot be gone underneath us.
2122 	 *
2123 	 * Only for the time /after/ successful bpf_prog_new_fd()
2124 	 * and before returning to userspace, we might just hold
2125 	 * one reference and any parallel close on that fd could
2126 	 * rip everything out. Hence, below notifications must
2127 	 * happen before bpf_prog_new_fd().
2128 	 *
2129 	 * Also, any failure handling from this point onwards must
2130 	 * be using bpf_prog_put() given the program is exposed.
2131 	 */
2132 	bpf_prog_kallsyms_add(prog);
2133 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2134 	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2135 
2136 	err = bpf_prog_new_fd(prog);
2137 	if (err < 0)
2138 		bpf_prog_put(prog);
2139 	return err;
2140 
2141 free_used_maps:
2142 	/* In case we have subprogs, we need to wait for a grace
2143 	 * period before we can tear down JIT memory since symbols
2144 	 * are already exposed under kallsyms.
2145 	 */
2146 	__bpf_prog_put_noref(prog, prog->aux->func_cnt);
2147 	return err;
2148 free_prog:
2149 	bpf_prog_uncharge_memlock(prog);
2150 free_prog_sec:
2151 	security_bpf_prog_free(prog->aux);
2152 free_prog_nouncharge:
2153 	bpf_prog_free(prog);
2154 	return err;
2155 }
2156 
2157 #define BPF_OBJ_LAST_FIELD file_flags
2158 
2159 static int bpf_obj_pin(const union bpf_attr *attr)
2160 {
2161 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2162 		return -EINVAL;
2163 
2164 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2165 }
2166 
2167 static int bpf_obj_get(const union bpf_attr *attr)
2168 {
2169 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2170 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2171 		return -EINVAL;
2172 
2173 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2174 				attr->file_flags);
2175 }
2176 
2177 static int bpf_tracing_prog_release(struct inode *inode, struct file *filp)
2178 {
2179 	struct bpf_prog *prog = filp->private_data;
2180 
2181 	WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
2182 	bpf_prog_put(prog);
2183 	return 0;
2184 }
2185 
2186 static const struct file_operations bpf_tracing_prog_fops = {
2187 	.release	= bpf_tracing_prog_release,
2188 	.read		= bpf_dummy_read,
2189 	.write		= bpf_dummy_write,
2190 };
2191 
2192 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
2193 {
2194 	int tr_fd, err;
2195 
2196 	if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2197 	    prog->expected_attach_type != BPF_TRACE_FEXIT &&
2198 	    prog->type != BPF_PROG_TYPE_EXT) {
2199 		err = -EINVAL;
2200 		goto out_put_prog;
2201 	}
2202 
2203 	err = bpf_trampoline_link_prog(prog);
2204 	if (err)
2205 		goto out_put_prog;
2206 
2207 	tr_fd = anon_inode_getfd("bpf-tracing-prog", &bpf_tracing_prog_fops,
2208 				 prog, O_CLOEXEC);
2209 	if (tr_fd < 0) {
2210 		WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
2211 		err = tr_fd;
2212 		goto out_put_prog;
2213 	}
2214 	return tr_fd;
2215 
2216 out_put_prog:
2217 	bpf_prog_put(prog);
2218 	return err;
2219 }
2220 
2221 struct bpf_raw_tracepoint {
2222 	struct bpf_raw_event_map *btp;
2223 	struct bpf_prog *prog;
2224 };
2225 
2226 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
2227 {
2228 	struct bpf_raw_tracepoint *raw_tp = filp->private_data;
2229 
2230 	if (raw_tp->prog) {
2231 		bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
2232 		bpf_prog_put(raw_tp->prog);
2233 	}
2234 	bpf_put_raw_tracepoint(raw_tp->btp);
2235 	kfree(raw_tp);
2236 	return 0;
2237 }
2238 
2239 static const struct file_operations bpf_raw_tp_fops = {
2240 	.release	= bpf_raw_tracepoint_release,
2241 	.read		= bpf_dummy_read,
2242 	.write		= bpf_dummy_write,
2243 };
2244 
2245 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2246 
2247 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2248 {
2249 	struct bpf_raw_tracepoint *raw_tp;
2250 	struct bpf_raw_event_map *btp;
2251 	struct bpf_prog *prog;
2252 	const char *tp_name;
2253 	char buf[128];
2254 	int tp_fd, err;
2255 
2256 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2257 		return -EINVAL;
2258 
2259 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2260 	if (IS_ERR(prog))
2261 		return PTR_ERR(prog);
2262 
2263 	if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
2264 	    prog->type != BPF_PROG_TYPE_TRACING &&
2265 	    prog->type != BPF_PROG_TYPE_EXT &&
2266 	    prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
2267 		err = -EINVAL;
2268 		goto out_put_prog;
2269 	}
2270 
2271 	if (prog->type == BPF_PROG_TYPE_TRACING ||
2272 	    prog->type == BPF_PROG_TYPE_EXT) {
2273 		if (attr->raw_tracepoint.name) {
2274 			/* The attach point for this category of programs
2275 			 * should be specified via btf_id during program load.
2276 			 */
2277 			err = -EINVAL;
2278 			goto out_put_prog;
2279 		}
2280 		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
2281 			tp_name = prog->aux->attach_func_name;
2282 		else
2283 			return bpf_tracing_prog_attach(prog);
2284 	} else {
2285 		if (strncpy_from_user(buf,
2286 				      u64_to_user_ptr(attr->raw_tracepoint.name),
2287 				      sizeof(buf) - 1) < 0) {
2288 			err = -EFAULT;
2289 			goto out_put_prog;
2290 		}
2291 		buf[sizeof(buf) - 1] = 0;
2292 		tp_name = buf;
2293 	}
2294 
2295 	btp = bpf_get_raw_tracepoint(tp_name);
2296 	if (!btp) {
2297 		err = -ENOENT;
2298 		goto out_put_prog;
2299 	}
2300 
2301 	raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
2302 	if (!raw_tp) {
2303 		err = -ENOMEM;
2304 		goto out_put_btp;
2305 	}
2306 	raw_tp->btp = btp;
2307 	raw_tp->prog = prog;
2308 
2309 	err = bpf_probe_register(raw_tp->btp, prog);
2310 	if (err)
2311 		goto out_free_tp;
2312 
2313 	tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
2314 				 O_CLOEXEC);
2315 	if (tp_fd < 0) {
2316 		bpf_probe_unregister(raw_tp->btp, prog);
2317 		err = tp_fd;
2318 		goto out_free_tp;
2319 	}
2320 	return tp_fd;
2321 
2322 out_free_tp:
2323 	kfree(raw_tp);
2324 out_put_btp:
2325 	bpf_put_raw_tracepoint(btp);
2326 out_put_prog:
2327 	bpf_prog_put(prog);
2328 	return err;
2329 }
2330 
2331 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2332 					     enum bpf_attach_type attach_type)
2333 {
2334 	switch (prog->type) {
2335 	case BPF_PROG_TYPE_CGROUP_SOCK:
2336 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2337 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2338 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2339 	case BPF_PROG_TYPE_CGROUP_SKB:
2340 		return prog->enforce_expected_attach_type &&
2341 			prog->expected_attach_type != attach_type ?
2342 			-EINVAL : 0;
2343 	default:
2344 		return 0;
2345 	}
2346 }
2347 
2348 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2349 
2350 #define BPF_F_ATTACH_MASK \
2351 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2352 
2353 static int bpf_prog_attach(const union bpf_attr *attr)
2354 {
2355 	enum bpf_prog_type ptype;
2356 	struct bpf_prog *prog;
2357 	int ret;
2358 
2359 	if (!capable(CAP_NET_ADMIN))
2360 		return -EPERM;
2361 
2362 	if (CHECK_ATTR(BPF_PROG_ATTACH))
2363 		return -EINVAL;
2364 
2365 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2366 		return -EINVAL;
2367 
2368 	switch (attr->attach_type) {
2369 	case BPF_CGROUP_INET_INGRESS:
2370 	case BPF_CGROUP_INET_EGRESS:
2371 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2372 		break;
2373 	case BPF_CGROUP_INET_SOCK_CREATE:
2374 	case BPF_CGROUP_INET4_POST_BIND:
2375 	case BPF_CGROUP_INET6_POST_BIND:
2376 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2377 		break;
2378 	case BPF_CGROUP_INET4_BIND:
2379 	case BPF_CGROUP_INET6_BIND:
2380 	case BPF_CGROUP_INET4_CONNECT:
2381 	case BPF_CGROUP_INET6_CONNECT:
2382 	case BPF_CGROUP_UDP4_SENDMSG:
2383 	case BPF_CGROUP_UDP6_SENDMSG:
2384 	case BPF_CGROUP_UDP4_RECVMSG:
2385 	case BPF_CGROUP_UDP6_RECVMSG:
2386 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2387 		break;
2388 	case BPF_CGROUP_SOCK_OPS:
2389 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2390 		break;
2391 	case BPF_CGROUP_DEVICE:
2392 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2393 		break;
2394 	case BPF_SK_MSG_VERDICT:
2395 		ptype = BPF_PROG_TYPE_SK_MSG;
2396 		break;
2397 	case BPF_SK_SKB_STREAM_PARSER:
2398 	case BPF_SK_SKB_STREAM_VERDICT:
2399 		ptype = BPF_PROG_TYPE_SK_SKB;
2400 		break;
2401 	case BPF_LIRC_MODE2:
2402 		ptype = BPF_PROG_TYPE_LIRC_MODE2;
2403 		break;
2404 	case BPF_FLOW_DISSECTOR:
2405 		ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2406 		break;
2407 	case BPF_CGROUP_SYSCTL:
2408 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2409 		break;
2410 	case BPF_CGROUP_GETSOCKOPT:
2411 	case BPF_CGROUP_SETSOCKOPT:
2412 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2413 		break;
2414 	default:
2415 		return -EINVAL;
2416 	}
2417 
2418 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2419 	if (IS_ERR(prog))
2420 		return PTR_ERR(prog);
2421 
2422 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2423 		bpf_prog_put(prog);
2424 		return -EINVAL;
2425 	}
2426 
2427 	switch (ptype) {
2428 	case BPF_PROG_TYPE_SK_SKB:
2429 	case BPF_PROG_TYPE_SK_MSG:
2430 		ret = sock_map_get_from_fd(attr, prog);
2431 		break;
2432 	case BPF_PROG_TYPE_LIRC_MODE2:
2433 		ret = lirc_prog_attach(attr, prog);
2434 		break;
2435 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2436 		ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2437 		break;
2438 	default:
2439 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2440 	}
2441 
2442 	if (ret)
2443 		bpf_prog_put(prog);
2444 	return ret;
2445 }
2446 
2447 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2448 
2449 static int bpf_prog_detach(const union bpf_attr *attr)
2450 {
2451 	enum bpf_prog_type ptype;
2452 
2453 	if (!capable(CAP_NET_ADMIN))
2454 		return -EPERM;
2455 
2456 	if (CHECK_ATTR(BPF_PROG_DETACH))
2457 		return -EINVAL;
2458 
2459 	switch (attr->attach_type) {
2460 	case BPF_CGROUP_INET_INGRESS:
2461 	case BPF_CGROUP_INET_EGRESS:
2462 		ptype = BPF_PROG_TYPE_CGROUP_SKB;
2463 		break;
2464 	case BPF_CGROUP_INET_SOCK_CREATE:
2465 	case BPF_CGROUP_INET4_POST_BIND:
2466 	case BPF_CGROUP_INET6_POST_BIND:
2467 		ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2468 		break;
2469 	case BPF_CGROUP_INET4_BIND:
2470 	case BPF_CGROUP_INET6_BIND:
2471 	case BPF_CGROUP_INET4_CONNECT:
2472 	case BPF_CGROUP_INET6_CONNECT:
2473 	case BPF_CGROUP_UDP4_SENDMSG:
2474 	case BPF_CGROUP_UDP6_SENDMSG:
2475 	case BPF_CGROUP_UDP4_RECVMSG:
2476 	case BPF_CGROUP_UDP6_RECVMSG:
2477 		ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2478 		break;
2479 	case BPF_CGROUP_SOCK_OPS:
2480 		ptype = BPF_PROG_TYPE_SOCK_OPS;
2481 		break;
2482 	case BPF_CGROUP_DEVICE:
2483 		ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2484 		break;
2485 	case BPF_SK_MSG_VERDICT:
2486 		return sock_map_get_from_fd(attr, NULL);
2487 	case BPF_SK_SKB_STREAM_PARSER:
2488 	case BPF_SK_SKB_STREAM_VERDICT:
2489 		return sock_map_get_from_fd(attr, NULL);
2490 	case BPF_LIRC_MODE2:
2491 		return lirc_prog_detach(attr);
2492 	case BPF_FLOW_DISSECTOR:
2493 		return skb_flow_dissector_bpf_prog_detach(attr);
2494 	case BPF_CGROUP_SYSCTL:
2495 		ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2496 		break;
2497 	case BPF_CGROUP_GETSOCKOPT:
2498 	case BPF_CGROUP_SETSOCKOPT:
2499 		ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2500 		break;
2501 	default:
2502 		return -EINVAL;
2503 	}
2504 
2505 	return cgroup_bpf_prog_detach(attr, ptype);
2506 }
2507 
2508 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2509 
2510 static int bpf_prog_query(const union bpf_attr *attr,
2511 			  union bpf_attr __user *uattr)
2512 {
2513 	if (!capable(CAP_NET_ADMIN))
2514 		return -EPERM;
2515 	if (CHECK_ATTR(BPF_PROG_QUERY))
2516 		return -EINVAL;
2517 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2518 		return -EINVAL;
2519 
2520 	switch (attr->query.attach_type) {
2521 	case BPF_CGROUP_INET_INGRESS:
2522 	case BPF_CGROUP_INET_EGRESS:
2523 	case BPF_CGROUP_INET_SOCK_CREATE:
2524 	case BPF_CGROUP_INET4_BIND:
2525 	case BPF_CGROUP_INET6_BIND:
2526 	case BPF_CGROUP_INET4_POST_BIND:
2527 	case BPF_CGROUP_INET6_POST_BIND:
2528 	case BPF_CGROUP_INET4_CONNECT:
2529 	case BPF_CGROUP_INET6_CONNECT:
2530 	case BPF_CGROUP_UDP4_SENDMSG:
2531 	case BPF_CGROUP_UDP6_SENDMSG:
2532 	case BPF_CGROUP_UDP4_RECVMSG:
2533 	case BPF_CGROUP_UDP6_RECVMSG:
2534 	case BPF_CGROUP_SOCK_OPS:
2535 	case BPF_CGROUP_DEVICE:
2536 	case BPF_CGROUP_SYSCTL:
2537 	case BPF_CGROUP_GETSOCKOPT:
2538 	case BPF_CGROUP_SETSOCKOPT:
2539 		break;
2540 	case BPF_LIRC_MODE2:
2541 		return lirc_prog_query(attr, uattr);
2542 	case BPF_FLOW_DISSECTOR:
2543 		return skb_flow_dissector_prog_query(attr, uattr);
2544 	default:
2545 		return -EINVAL;
2546 	}
2547 
2548 	return cgroup_bpf_prog_query(attr, uattr);
2549 }
2550 
2551 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2552 
2553 static int bpf_prog_test_run(const union bpf_attr *attr,
2554 			     union bpf_attr __user *uattr)
2555 {
2556 	struct bpf_prog *prog;
2557 	int ret = -ENOTSUPP;
2558 
2559 	if (!capable(CAP_SYS_ADMIN))
2560 		return -EPERM;
2561 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2562 		return -EINVAL;
2563 
2564 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2565 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
2566 		return -EINVAL;
2567 
2568 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2569 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
2570 		return -EINVAL;
2571 
2572 	prog = bpf_prog_get(attr->test.prog_fd);
2573 	if (IS_ERR(prog))
2574 		return PTR_ERR(prog);
2575 
2576 	if (prog->aux->ops->test_run)
2577 		ret = prog->aux->ops->test_run(prog, attr, uattr);
2578 
2579 	bpf_prog_put(prog);
2580 	return ret;
2581 }
2582 
2583 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2584 
2585 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2586 			       union bpf_attr __user *uattr,
2587 			       struct idr *idr,
2588 			       spinlock_t *lock)
2589 {
2590 	u32 next_id = attr->start_id;
2591 	int err = 0;
2592 
2593 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2594 		return -EINVAL;
2595 
2596 	if (!capable(CAP_SYS_ADMIN))
2597 		return -EPERM;
2598 
2599 	next_id++;
2600 	spin_lock_bh(lock);
2601 	if (!idr_get_next(idr, &next_id))
2602 		err = -ENOENT;
2603 	spin_unlock_bh(lock);
2604 
2605 	if (!err)
2606 		err = put_user(next_id, &uattr->next_id);
2607 
2608 	return err;
2609 }
2610 
2611 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2612 
2613 struct bpf_prog *bpf_prog_by_id(u32 id)
2614 {
2615 	struct bpf_prog *prog;
2616 
2617 	if (!id)
2618 		return ERR_PTR(-ENOENT);
2619 
2620 	spin_lock_bh(&prog_idr_lock);
2621 	prog = idr_find(&prog_idr, id);
2622 	if (prog)
2623 		prog = bpf_prog_inc_not_zero(prog);
2624 	else
2625 		prog = ERR_PTR(-ENOENT);
2626 	spin_unlock_bh(&prog_idr_lock);
2627 	return prog;
2628 }
2629 
2630 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2631 {
2632 	struct bpf_prog *prog;
2633 	u32 id = attr->prog_id;
2634 	int fd;
2635 
2636 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2637 		return -EINVAL;
2638 
2639 	if (!capable(CAP_SYS_ADMIN))
2640 		return -EPERM;
2641 
2642 	prog = bpf_prog_by_id(id);
2643 	if (IS_ERR(prog))
2644 		return PTR_ERR(prog);
2645 
2646 	fd = bpf_prog_new_fd(prog);
2647 	if (fd < 0)
2648 		bpf_prog_put(prog);
2649 
2650 	return fd;
2651 }
2652 
2653 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2654 
2655 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2656 {
2657 	struct bpf_map *map;
2658 	u32 id = attr->map_id;
2659 	int f_flags;
2660 	int fd;
2661 
2662 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2663 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2664 		return -EINVAL;
2665 
2666 	if (!capable(CAP_SYS_ADMIN))
2667 		return -EPERM;
2668 
2669 	f_flags = bpf_get_file_flag(attr->open_flags);
2670 	if (f_flags < 0)
2671 		return f_flags;
2672 
2673 	spin_lock_bh(&map_idr_lock);
2674 	map = idr_find(&map_idr, id);
2675 	if (map)
2676 		map = __bpf_map_inc_not_zero(map, true);
2677 	else
2678 		map = ERR_PTR(-ENOENT);
2679 	spin_unlock_bh(&map_idr_lock);
2680 
2681 	if (IS_ERR(map))
2682 		return PTR_ERR(map);
2683 
2684 	fd = bpf_map_new_fd(map, f_flags);
2685 	if (fd < 0)
2686 		bpf_map_put_with_uref(map);
2687 
2688 	return fd;
2689 }
2690 
2691 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2692 					      unsigned long addr, u32 *off,
2693 					      u32 *type)
2694 {
2695 	const struct bpf_map *map;
2696 	int i;
2697 
2698 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2699 		map = prog->aux->used_maps[i];
2700 		if (map == (void *)addr) {
2701 			*type = BPF_PSEUDO_MAP_FD;
2702 			return map;
2703 		}
2704 		if (!map->ops->map_direct_value_meta)
2705 			continue;
2706 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
2707 			*type = BPF_PSEUDO_MAP_VALUE;
2708 			return map;
2709 		}
2710 	}
2711 
2712 	return NULL;
2713 }
2714 
2715 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2716 {
2717 	const struct bpf_map *map;
2718 	struct bpf_insn *insns;
2719 	u32 off, type;
2720 	u64 imm;
2721 	int i;
2722 
2723 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2724 			GFP_USER);
2725 	if (!insns)
2726 		return insns;
2727 
2728 	for (i = 0; i < prog->len; i++) {
2729 		if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2730 			insns[i].code = BPF_JMP | BPF_CALL;
2731 			insns[i].imm = BPF_FUNC_tail_call;
2732 			/* fall-through */
2733 		}
2734 		if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2735 		    insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2736 			if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2737 				insns[i].code = BPF_JMP | BPF_CALL;
2738 			if (!bpf_dump_raw_ok())
2739 				insns[i].imm = 0;
2740 			continue;
2741 		}
2742 
2743 		if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2744 			continue;
2745 
2746 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2747 		map = bpf_map_from_imm(prog, imm, &off, &type);
2748 		if (map) {
2749 			insns[i].src_reg = type;
2750 			insns[i].imm = map->id;
2751 			insns[i + 1].imm = off;
2752 			continue;
2753 		}
2754 	}
2755 
2756 	return insns;
2757 }
2758 
2759 static int set_info_rec_size(struct bpf_prog_info *info)
2760 {
2761 	/*
2762 	 * Ensure info.*_rec_size is the same as kernel expected size
2763 	 *
2764 	 * or
2765 	 *
2766 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
2767 	 * zero.  In this case, the kernel will set the expected
2768 	 * _rec_size back to the info.
2769 	 */
2770 
2771 	if ((info->nr_func_info || info->func_info_rec_size) &&
2772 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
2773 		return -EINVAL;
2774 
2775 	if ((info->nr_line_info || info->line_info_rec_size) &&
2776 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
2777 		return -EINVAL;
2778 
2779 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2780 	    info->jited_line_info_rec_size != sizeof(__u64))
2781 		return -EINVAL;
2782 
2783 	info->func_info_rec_size = sizeof(struct bpf_func_info);
2784 	info->line_info_rec_size = sizeof(struct bpf_line_info);
2785 	info->jited_line_info_rec_size = sizeof(__u64);
2786 
2787 	return 0;
2788 }
2789 
2790 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2791 				   const union bpf_attr *attr,
2792 				   union bpf_attr __user *uattr)
2793 {
2794 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2795 	struct bpf_prog_info info = {};
2796 	u32 info_len = attr->info.info_len;
2797 	struct bpf_prog_stats stats;
2798 	char __user *uinsns;
2799 	u32 ulen;
2800 	int err;
2801 
2802 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2803 	if (err)
2804 		return err;
2805 	info_len = min_t(u32, sizeof(info), info_len);
2806 
2807 	if (copy_from_user(&info, uinfo, info_len))
2808 		return -EFAULT;
2809 
2810 	info.type = prog->type;
2811 	info.id = prog->aux->id;
2812 	info.load_time = prog->aux->load_time;
2813 	info.created_by_uid = from_kuid_munged(current_user_ns(),
2814 					       prog->aux->user->uid);
2815 	info.gpl_compatible = prog->gpl_compatible;
2816 
2817 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
2818 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2819 
2820 	ulen = info.nr_map_ids;
2821 	info.nr_map_ids = prog->aux->used_map_cnt;
2822 	ulen = min_t(u32, info.nr_map_ids, ulen);
2823 	if (ulen) {
2824 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2825 		u32 i;
2826 
2827 		for (i = 0; i < ulen; i++)
2828 			if (put_user(prog->aux->used_maps[i]->id,
2829 				     &user_map_ids[i]))
2830 				return -EFAULT;
2831 	}
2832 
2833 	err = set_info_rec_size(&info);
2834 	if (err)
2835 		return err;
2836 
2837 	bpf_prog_get_stats(prog, &stats);
2838 	info.run_time_ns = stats.nsecs;
2839 	info.run_cnt = stats.cnt;
2840 
2841 	if (!capable(CAP_SYS_ADMIN)) {
2842 		info.jited_prog_len = 0;
2843 		info.xlated_prog_len = 0;
2844 		info.nr_jited_ksyms = 0;
2845 		info.nr_jited_func_lens = 0;
2846 		info.nr_func_info = 0;
2847 		info.nr_line_info = 0;
2848 		info.nr_jited_line_info = 0;
2849 		goto done;
2850 	}
2851 
2852 	ulen = info.xlated_prog_len;
2853 	info.xlated_prog_len = bpf_prog_insn_size(prog);
2854 	if (info.xlated_prog_len && ulen) {
2855 		struct bpf_insn *insns_sanitized;
2856 		bool fault;
2857 
2858 		if (prog->blinded && !bpf_dump_raw_ok()) {
2859 			info.xlated_prog_insns = 0;
2860 			goto done;
2861 		}
2862 		insns_sanitized = bpf_insn_prepare_dump(prog);
2863 		if (!insns_sanitized)
2864 			return -ENOMEM;
2865 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2866 		ulen = min_t(u32, info.xlated_prog_len, ulen);
2867 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
2868 		kfree(insns_sanitized);
2869 		if (fault)
2870 			return -EFAULT;
2871 	}
2872 
2873 	if (bpf_prog_is_dev_bound(prog->aux)) {
2874 		err = bpf_prog_offload_info_fill(&info, prog);
2875 		if (err)
2876 			return err;
2877 		goto done;
2878 	}
2879 
2880 	/* NOTE: the following code is supposed to be skipped for offload.
2881 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
2882 	 * for offload.
2883 	 */
2884 	ulen = info.jited_prog_len;
2885 	if (prog->aux->func_cnt) {
2886 		u32 i;
2887 
2888 		info.jited_prog_len = 0;
2889 		for (i = 0; i < prog->aux->func_cnt; i++)
2890 			info.jited_prog_len += prog->aux->func[i]->jited_len;
2891 	} else {
2892 		info.jited_prog_len = prog->jited_len;
2893 	}
2894 
2895 	if (info.jited_prog_len && ulen) {
2896 		if (bpf_dump_raw_ok()) {
2897 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
2898 			ulen = min_t(u32, info.jited_prog_len, ulen);
2899 
2900 			/* for multi-function programs, copy the JITed
2901 			 * instructions for all the functions
2902 			 */
2903 			if (prog->aux->func_cnt) {
2904 				u32 len, free, i;
2905 				u8 *img;
2906 
2907 				free = ulen;
2908 				for (i = 0; i < prog->aux->func_cnt; i++) {
2909 					len = prog->aux->func[i]->jited_len;
2910 					len = min_t(u32, len, free);
2911 					img = (u8 *) prog->aux->func[i]->bpf_func;
2912 					if (copy_to_user(uinsns, img, len))
2913 						return -EFAULT;
2914 					uinsns += len;
2915 					free -= len;
2916 					if (!free)
2917 						break;
2918 				}
2919 			} else {
2920 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
2921 					return -EFAULT;
2922 			}
2923 		} else {
2924 			info.jited_prog_insns = 0;
2925 		}
2926 	}
2927 
2928 	ulen = info.nr_jited_ksyms;
2929 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2930 	if (ulen) {
2931 		if (bpf_dump_raw_ok()) {
2932 			unsigned long ksym_addr;
2933 			u64 __user *user_ksyms;
2934 			u32 i;
2935 
2936 			/* copy the address of the kernel symbol
2937 			 * corresponding to each function
2938 			 */
2939 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2940 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2941 			if (prog->aux->func_cnt) {
2942 				for (i = 0; i < ulen; i++) {
2943 					ksym_addr = (unsigned long)
2944 						prog->aux->func[i]->bpf_func;
2945 					if (put_user((u64) ksym_addr,
2946 						     &user_ksyms[i]))
2947 						return -EFAULT;
2948 				}
2949 			} else {
2950 				ksym_addr = (unsigned long) prog->bpf_func;
2951 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
2952 					return -EFAULT;
2953 			}
2954 		} else {
2955 			info.jited_ksyms = 0;
2956 		}
2957 	}
2958 
2959 	ulen = info.nr_jited_func_lens;
2960 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2961 	if (ulen) {
2962 		if (bpf_dump_raw_ok()) {
2963 			u32 __user *user_lens;
2964 			u32 func_len, i;
2965 
2966 			/* copy the JITed image lengths for each function */
2967 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2968 			user_lens = u64_to_user_ptr(info.jited_func_lens);
2969 			if (prog->aux->func_cnt) {
2970 				for (i = 0; i < ulen; i++) {
2971 					func_len =
2972 						prog->aux->func[i]->jited_len;
2973 					if (put_user(func_len, &user_lens[i]))
2974 						return -EFAULT;
2975 				}
2976 			} else {
2977 				func_len = prog->jited_len;
2978 				if (put_user(func_len, &user_lens[0]))
2979 					return -EFAULT;
2980 			}
2981 		} else {
2982 			info.jited_func_lens = 0;
2983 		}
2984 	}
2985 
2986 	if (prog->aux->btf)
2987 		info.btf_id = btf_id(prog->aux->btf);
2988 
2989 	ulen = info.nr_func_info;
2990 	info.nr_func_info = prog->aux->func_info_cnt;
2991 	if (info.nr_func_info && ulen) {
2992 		char __user *user_finfo;
2993 
2994 		user_finfo = u64_to_user_ptr(info.func_info);
2995 		ulen = min_t(u32, info.nr_func_info, ulen);
2996 		if (copy_to_user(user_finfo, prog->aux->func_info,
2997 				 info.func_info_rec_size * ulen))
2998 			return -EFAULT;
2999 	}
3000 
3001 	ulen = info.nr_line_info;
3002 	info.nr_line_info = prog->aux->nr_linfo;
3003 	if (info.nr_line_info && ulen) {
3004 		__u8 __user *user_linfo;
3005 
3006 		user_linfo = u64_to_user_ptr(info.line_info);
3007 		ulen = min_t(u32, info.nr_line_info, ulen);
3008 		if (copy_to_user(user_linfo, prog->aux->linfo,
3009 				 info.line_info_rec_size * ulen))
3010 			return -EFAULT;
3011 	}
3012 
3013 	ulen = info.nr_jited_line_info;
3014 	if (prog->aux->jited_linfo)
3015 		info.nr_jited_line_info = prog->aux->nr_linfo;
3016 	else
3017 		info.nr_jited_line_info = 0;
3018 	if (info.nr_jited_line_info && ulen) {
3019 		if (bpf_dump_raw_ok()) {
3020 			__u64 __user *user_linfo;
3021 			u32 i;
3022 
3023 			user_linfo = u64_to_user_ptr(info.jited_line_info);
3024 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
3025 			for (i = 0; i < ulen; i++) {
3026 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3027 					     &user_linfo[i]))
3028 					return -EFAULT;
3029 			}
3030 		} else {
3031 			info.jited_line_info = 0;
3032 		}
3033 	}
3034 
3035 	ulen = info.nr_prog_tags;
3036 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3037 	if (ulen) {
3038 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3039 		u32 i;
3040 
3041 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
3042 		ulen = min_t(u32, info.nr_prog_tags, ulen);
3043 		if (prog->aux->func_cnt) {
3044 			for (i = 0; i < ulen; i++) {
3045 				if (copy_to_user(user_prog_tags[i],
3046 						 prog->aux->func[i]->tag,
3047 						 BPF_TAG_SIZE))
3048 					return -EFAULT;
3049 			}
3050 		} else {
3051 			if (copy_to_user(user_prog_tags[0],
3052 					 prog->tag, BPF_TAG_SIZE))
3053 				return -EFAULT;
3054 		}
3055 	}
3056 
3057 done:
3058 	if (copy_to_user(uinfo, &info, info_len) ||
3059 	    put_user(info_len, &uattr->info.info_len))
3060 		return -EFAULT;
3061 
3062 	return 0;
3063 }
3064 
3065 static int bpf_map_get_info_by_fd(struct bpf_map *map,
3066 				  const union bpf_attr *attr,
3067 				  union bpf_attr __user *uattr)
3068 {
3069 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3070 	struct bpf_map_info info = {};
3071 	u32 info_len = attr->info.info_len;
3072 	int err;
3073 
3074 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3075 	if (err)
3076 		return err;
3077 	info_len = min_t(u32, sizeof(info), info_len);
3078 
3079 	info.type = map->map_type;
3080 	info.id = map->id;
3081 	info.key_size = map->key_size;
3082 	info.value_size = map->value_size;
3083 	info.max_entries = map->max_entries;
3084 	info.map_flags = map->map_flags;
3085 	memcpy(info.name, map->name, sizeof(map->name));
3086 
3087 	if (map->btf) {
3088 		info.btf_id = btf_id(map->btf);
3089 		info.btf_key_type_id = map->btf_key_type_id;
3090 		info.btf_value_type_id = map->btf_value_type_id;
3091 	}
3092 	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3093 
3094 	if (bpf_map_is_dev_bound(map)) {
3095 		err = bpf_map_offload_info_fill(&info, map);
3096 		if (err)
3097 			return err;
3098 	}
3099 
3100 	if (copy_to_user(uinfo, &info, info_len) ||
3101 	    put_user(info_len, &uattr->info.info_len))
3102 		return -EFAULT;
3103 
3104 	return 0;
3105 }
3106 
3107 static int bpf_btf_get_info_by_fd(struct btf *btf,
3108 				  const union bpf_attr *attr,
3109 				  union bpf_attr __user *uattr)
3110 {
3111 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3112 	u32 info_len = attr->info.info_len;
3113 	int err;
3114 
3115 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3116 	if (err)
3117 		return err;
3118 
3119 	return btf_get_info_by_fd(btf, attr, uattr);
3120 }
3121 
3122 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3123 
3124 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3125 				  union bpf_attr __user *uattr)
3126 {
3127 	int ufd = attr->info.bpf_fd;
3128 	struct fd f;
3129 	int err;
3130 
3131 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3132 		return -EINVAL;
3133 
3134 	f = fdget(ufd);
3135 	if (!f.file)
3136 		return -EBADFD;
3137 
3138 	if (f.file->f_op == &bpf_prog_fops)
3139 		err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
3140 					      uattr);
3141 	else if (f.file->f_op == &bpf_map_fops)
3142 		err = bpf_map_get_info_by_fd(f.file->private_data, attr,
3143 					     uattr);
3144 	else if (f.file->f_op == &btf_fops)
3145 		err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
3146 	else
3147 		err = -EINVAL;
3148 
3149 	fdput(f);
3150 	return err;
3151 }
3152 
3153 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3154 
3155 static int bpf_btf_load(const union bpf_attr *attr)
3156 {
3157 	if (CHECK_ATTR(BPF_BTF_LOAD))
3158 		return -EINVAL;
3159 
3160 	if (!capable(CAP_SYS_ADMIN))
3161 		return -EPERM;
3162 
3163 	return btf_new_fd(attr);
3164 }
3165 
3166 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3167 
3168 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3169 {
3170 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3171 		return -EINVAL;
3172 
3173 	if (!capable(CAP_SYS_ADMIN))
3174 		return -EPERM;
3175 
3176 	return btf_get_fd_by_id(attr->btf_id);
3177 }
3178 
3179 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3180 				    union bpf_attr __user *uattr,
3181 				    u32 prog_id, u32 fd_type,
3182 				    const char *buf, u64 probe_offset,
3183 				    u64 probe_addr)
3184 {
3185 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3186 	u32 len = buf ? strlen(buf) : 0, input_len;
3187 	int err = 0;
3188 
3189 	if (put_user(len, &uattr->task_fd_query.buf_len))
3190 		return -EFAULT;
3191 	input_len = attr->task_fd_query.buf_len;
3192 	if (input_len && ubuf) {
3193 		if (!len) {
3194 			/* nothing to copy, just make ubuf NULL terminated */
3195 			char zero = '\0';
3196 
3197 			if (put_user(zero, ubuf))
3198 				return -EFAULT;
3199 		} else if (input_len >= len + 1) {
3200 			/* ubuf can hold the string with NULL terminator */
3201 			if (copy_to_user(ubuf, buf, len + 1))
3202 				return -EFAULT;
3203 		} else {
3204 			/* ubuf cannot hold the string with NULL terminator,
3205 			 * do a partial copy with NULL terminator.
3206 			 */
3207 			char zero = '\0';
3208 
3209 			err = -ENOSPC;
3210 			if (copy_to_user(ubuf, buf, input_len - 1))
3211 				return -EFAULT;
3212 			if (put_user(zero, ubuf + input_len - 1))
3213 				return -EFAULT;
3214 		}
3215 	}
3216 
3217 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3218 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3219 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3220 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3221 		return -EFAULT;
3222 
3223 	return err;
3224 }
3225 
3226 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3227 
3228 static int bpf_task_fd_query(const union bpf_attr *attr,
3229 			     union bpf_attr __user *uattr)
3230 {
3231 	pid_t pid = attr->task_fd_query.pid;
3232 	u32 fd = attr->task_fd_query.fd;
3233 	const struct perf_event *event;
3234 	struct files_struct *files;
3235 	struct task_struct *task;
3236 	struct file *file;
3237 	int err;
3238 
3239 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3240 		return -EINVAL;
3241 
3242 	if (!capable(CAP_SYS_ADMIN))
3243 		return -EPERM;
3244 
3245 	if (attr->task_fd_query.flags != 0)
3246 		return -EINVAL;
3247 
3248 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3249 	if (!task)
3250 		return -ENOENT;
3251 
3252 	files = get_files_struct(task);
3253 	put_task_struct(task);
3254 	if (!files)
3255 		return -ENOENT;
3256 
3257 	err = 0;
3258 	spin_lock(&files->file_lock);
3259 	file = fcheck_files(files, fd);
3260 	if (!file)
3261 		err = -EBADF;
3262 	else
3263 		get_file(file);
3264 	spin_unlock(&files->file_lock);
3265 	put_files_struct(files);
3266 
3267 	if (err)
3268 		goto out;
3269 
3270 	if (file->f_op == &bpf_raw_tp_fops) {
3271 		struct bpf_raw_tracepoint *raw_tp = file->private_data;
3272 		struct bpf_raw_event_map *btp = raw_tp->btp;
3273 
3274 		err = bpf_task_fd_query_copy(attr, uattr,
3275 					     raw_tp->prog->aux->id,
3276 					     BPF_FD_TYPE_RAW_TRACEPOINT,
3277 					     btp->tp->name, 0, 0);
3278 		goto put_file;
3279 	}
3280 
3281 	event = perf_get_event(file);
3282 	if (!IS_ERR(event)) {
3283 		u64 probe_offset, probe_addr;
3284 		u32 prog_id, fd_type;
3285 		const char *buf;
3286 
3287 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3288 					      &buf, &probe_offset,
3289 					      &probe_addr);
3290 		if (!err)
3291 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3292 						     fd_type, buf,
3293 						     probe_offset,
3294 						     probe_addr);
3295 		goto put_file;
3296 	}
3297 
3298 	err = -ENOTSUPP;
3299 put_file:
3300 	fput(file);
3301 out:
3302 	return err;
3303 }
3304 
3305 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3306 
3307 #define BPF_DO_BATCH(fn)			\
3308 	do {					\
3309 		if (!fn) {			\
3310 			err = -ENOTSUPP;	\
3311 			goto err_put;		\
3312 		}				\
3313 		err = fn(map, attr, uattr);	\
3314 	} while (0)
3315 
3316 static int bpf_map_do_batch(const union bpf_attr *attr,
3317 			    union bpf_attr __user *uattr,
3318 			    int cmd)
3319 {
3320 	struct bpf_map *map;
3321 	int err, ufd;
3322 	struct fd f;
3323 
3324 	if (CHECK_ATTR(BPF_MAP_BATCH))
3325 		return -EINVAL;
3326 
3327 	ufd = attr->batch.map_fd;
3328 	f = fdget(ufd);
3329 	map = __bpf_map_get(f);
3330 	if (IS_ERR(map))
3331 		return PTR_ERR(map);
3332 
3333 	if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3334 	     cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3335 	    !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3336 		err = -EPERM;
3337 		goto err_put;
3338 	}
3339 
3340 	if (cmd != BPF_MAP_LOOKUP_BATCH &&
3341 	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3342 		err = -EPERM;
3343 		goto err_put;
3344 	}
3345 
3346 	if (cmd == BPF_MAP_LOOKUP_BATCH)
3347 		BPF_DO_BATCH(map->ops->map_lookup_batch);
3348 	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3349 		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3350 	else if (cmd == BPF_MAP_UPDATE_BATCH)
3351 		BPF_DO_BATCH(map->ops->map_update_batch);
3352 	else
3353 		BPF_DO_BATCH(map->ops->map_delete_batch);
3354 
3355 err_put:
3356 	fdput(f);
3357 	return err;
3358 }
3359 
3360 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3361 {
3362 	union bpf_attr attr = {};
3363 	int err;
3364 
3365 	if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3366 		return -EPERM;
3367 
3368 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3369 	if (err)
3370 		return err;
3371 	size = min_t(u32, size, sizeof(attr));
3372 
3373 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
3374 	if (copy_from_user(&attr, uattr, size) != 0)
3375 		return -EFAULT;
3376 
3377 	err = security_bpf(cmd, &attr, size);
3378 	if (err < 0)
3379 		return err;
3380 
3381 	switch (cmd) {
3382 	case BPF_MAP_CREATE:
3383 		err = map_create(&attr);
3384 		break;
3385 	case BPF_MAP_LOOKUP_ELEM:
3386 		err = map_lookup_elem(&attr);
3387 		break;
3388 	case BPF_MAP_UPDATE_ELEM:
3389 		err = map_update_elem(&attr);
3390 		break;
3391 	case BPF_MAP_DELETE_ELEM:
3392 		err = map_delete_elem(&attr);
3393 		break;
3394 	case BPF_MAP_GET_NEXT_KEY:
3395 		err = map_get_next_key(&attr);
3396 		break;
3397 	case BPF_MAP_FREEZE:
3398 		err = map_freeze(&attr);
3399 		break;
3400 	case BPF_PROG_LOAD:
3401 		err = bpf_prog_load(&attr, uattr);
3402 		break;
3403 	case BPF_OBJ_PIN:
3404 		err = bpf_obj_pin(&attr);
3405 		break;
3406 	case BPF_OBJ_GET:
3407 		err = bpf_obj_get(&attr);
3408 		break;
3409 	case BPF_PROG_ATTACH:
3410 		err = bpf_prog_attach(&attr);
3411 		break;
3412 	case BPF_PROG_DETACH:
3413 		err = bpf_prog_detach(&attr);
3414 		break;
3415 	case BPF_PROG_QUERY:
3416 		err = bpf_prog_query(&attr, uattr);
3417 		break;
3418 	case BPF_PROG_TEST_RUN:
3419 		err = bpf_prog_test_run(&attr, uattr);
3420 		break;
3421 	case BPF_PROG_GET_NEXT_ID:
3422 		err = bpf_obj_get_next_id(&attr, uattr,
3423 					  &prog_idr, &prog_idr_lock);
3424 		break;
3425 	case BPF_MAP_GET_NEXT_ID:
3426 		err = bpf_obj_get_next_id(&attr, uattr,
3427 					  &map_idr, &map_idr_lock);
3428 		break;
3429 	case BPF_BTF_GET_NEXT_ID:
3430 		err = bpf_obj_get_next_id(&attr, uattr,
3431 					  &btf_idr, &btf_idr_lock);
3432 		break;
3433 	case BPF_PROG_GET_FD_BY_ID:
3434 		err = bpf_prog_get_fd_by_id(&attr);
3435 		break;
3436 	case BPF_MAP_GET_FD_BY_ID:
3437 		err = bpf_map_get_fd_by_id(&attr);
3438 		break;
3439 	case BPF_OBJ_GET_INFO_BY_FD:
3440 		err = bpf_obj_get_info_by_fd(&attr, uattr);
3441 		break;
3442 	case BPF_RAW_TRACEPOINT_OPEN:
3443 		err = bpf_raw_tracepoint_open(&attr);
3444 		break;
3445 	case BPF_BTF_LOAD:
3446 		err = bpf_btf_load(&attr);
3447 		break;
3448 	case BPF_BTF_GET_FD_BY_ID:
3449 		err = bpf_btf_get_fd_by_id(&attr);
3450 		break;
3451 	case BPF_TASK_FD_QUERY:
3452 		err = bpf_task_fd_query(&attr, uattr);
3453 		break;
3454 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3455 		err = map_lookup_and_delete_elem(&attr);
3456 		break;
3457 	case BPF_MAP_LOOKUP_BATCH:
3458 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
3459 		break;
3460 	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
3461 		err = bpf_map_do_batch(&attr, uattr,
3462 				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
3463 		break;
3464 	case BPF_MAP_UPDATE_BATCH:
3465 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
3466 		break;
3467 	case BPF_MAP_DELETE_BATCH:
3468 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
3469 		break;
3470 	default:
3471 		err = -EINVAL;
3472 		break;
3473 	}
3474 
3475 	return err;
3476 }
3477