xref: /linux/kernel/bpf/syscall.c (revision a4a755c422242c27cb0f7900ac00cf33ac17b1ce)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  */
4 #include <linux/bpf.h>
5 #include <linux/bpf-cgroup.h>
6 #include <linux/bpf_trace.h>
7 #include <linux/bpf_lirc.h>
8 #include <linux/bpf_verifier.h>
9 #include <linux/bsearch.h>
10 #include <linux/btf.h>
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/vmalloc.h>
15 #include <linux/mmzone.h>
16 #include <linux/anon_inodes.h>
17 #include <linux/fdtable.h>
18 #include <linux/file.h>
19 #include <linux/fs.h>
20 #include <linux/license.h>
21 #include <linux/filter.h>
22 #include <linux/kernel.h>
23 #include <linux/idr.h>
24 #include <linux/cred.h>
25 #include <linux/timekeeping.h>
26 #include <linux/ctype.h>
27 #include <linux/nospec.h>
28 #include <linux/audit.h>
29 #include <uapi/linux/btf.h>
30 #include <linux/pgtable.h>
31 #include <linux/bpf_lsm.h>
32 #include <linux/poll.h>
33 #include <linux/sort.h>
34 #include <linux/bpf-netns.h>
35 #include <linux/rcupdate_trace.h>
36 #include <linux/memcontrol.h>
37 #include <linux/trace_events.h>
38 
39 #include <net/netfilter/nf_bpf_link.h>
40 #include <net/netkit.h>
41 #include <net/tcx.h>
42 
43 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
44 			  (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
45 			  (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
46 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
47 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
48 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
49 			IS_FD_HASH(map))
50 
51 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
52 
53 DEFINE_PER_CPU(int, bpf_prog_active);
54 static DEFINE_IDR(prog_idr);
55 static DEFINE_SPINLOCK(prog_idr_lock);
56 static DEFINE_IDR(map_idr);
57 static DEFINE_SPINLOCK(map_idr_lock);
58 static DEFINE_IDR(link_idr);
59 static DEFINE_SPINLOCK(link_idr_lock);
60 
61 int sysctl_unprivileged_bpf_disabled __read_mostly =
62 	IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
63 
64 static const struct bpf_map_ops * const bpf_map_types[] = {
65 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
66 #define BPF_MAP_TYPE(_id, _ops) \
67 	[_id] = &_ops,
68 #define BPF_LINK_TYPE(_id, _name)
69 #include <linux/bpf_types.h>
70 #undef BPF_PROG_TYPE
71 #undef BPF_MAP_TYPE
72 #undef BPF_LINK_TYPE
73 };
74 
75 /*
76  * If we're handed a bigger struct than we know of, ensure all the unknown bits
77  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
78  * we don't know about yet.
79  *
80  * There is a ToCToU between this function call and the following
81  * copy_from_user() call. However, this is not a concern since this function is
82  * meant to be a future-proofing of bits.
83  */
84 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
85 			     size_t expected_size,
86 			     size_t actual_size)
87 {
88 	int res;
89 
90 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
91 		return -E2BIG;
92 
93 	if (actual_size <= expected_size)
94 		return 0;
95 
96 	if (uaddr.is_kernel)
97 		res = memchr_inv(uaddr.kernel + expected_size, 0,
98 				 actual_size - expected_size) == NULL;
99 	else
100 		res = check_zeroed_user(uaddr.user + expected_size,
101 					actual_size - expected_size);
102 	if (res < 0)
103 		return res;
104 	return res ? 0 : -E2BIG;
105 }
106 
107 const struct bpf_map_ops bpf_map_offload_ops = {
108 	.map_meta_equal = bpf_map_meta_equal,
109 	.map_alloc = bpf_map_offload_map_alloc,
110 	.map_free = bpf_map_offload_map_free,
111 	.map_check_btf = map_check_no_btf,
112 	.map_mem_usage = bpf_map_offload_map_mem_usage,
113 };
114 
115 static void bpf_map_write_active_inc(struct bpf_map *map)
116 {
117 	atomic64_inc(&map->writecnt);
118 }
119 
120 static void bpf_map_write_active_dec(struct bpf_map *map)
121 {
122 	atomic64_dec(&map->writecnt);
123 }
124 
125 bool bpf_map_write_active(const struct bpf_map *map)
126 {
127 	return atomic64_read(&map->writecnt) != 0;
128 }
129 
130 static u32 bpf_map_value_size(const struct bpf_map *map)
131 {
132 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
133 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
134 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
135 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
136 		return round_up(map->value_size, 8) * num_possible_cpus();
137 	else if (IS_FD_MAP(map))
138 		return sizeof(u32);
139 	else
140 		return  map->value_size;
141 }
142 
143 static void maybe_wait_bpf_programs(struct bpf_map *map)
144 {
145 	/* Wait for any running non-sleepable BPF programs to complete so that
146 	 * userspace, when we return to it, knows that all non-sleepable
147 	 * programs that could be running use the new map value. For sleepable
148 	 * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
149 	 * for the completions of these programs, but considering the waiting
150 	 * time can be very long and userspace may think it will hang forever,
151 	 * so don't handle sleepable BPF programs now.
152 	 */
153 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
154 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
155 		synchronize_rcu();
156 }
157 
158 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
159 				void *key, void *value, __u64 flags)
160 {
161 	int err;
162 
163 	/* Need to create a kthread, thus must support schedule */
164 	if (bpf_map_is_offloaded(map)) {
165 		return bpf_map_offload_update_elem(map, key, value, flags);
166 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
167 		   map->map_type == BPF_MAP_TYPE_ARENA ||
168 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
169 		return map->ops->map_update_elem(map, key, value, flags);
170 	} else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
171 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
172 		return sock_map_update_elem_sys(map, key, value, flags);
173 	} else if (IS_FD_PROG_ARRAY(map)) {
174 		return bpf_fd_array_map_update_elem(map, map_file, key, value,
175 						    flags);
176 	}
177 
178 	bpf_disable_instrumentation();
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 		err = bpf_fd_array_map_update_elem(map, map_file, key, value,
189 						   flags);
190 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
191 		err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
192 						  flags);
193 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
194 		/* rcu_read_lock() is not needed */
195 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
196 							 flags);
197 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
198 		   map->map_type == BPF_MAP_TYPE_STACK ||
199 		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
200 		err = map->ops->map_push_elem(map, value, flags);
201 	} else {
202 		rcu_read_lock();
203 		err = map->ops->map_update_elem(map, key, value, flags);
204 		rcu_read_unlock();
205 	}
206 	bpf_enable_instrumentation();
207 
208 	return err;
209 }
210 
211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
212 			      __u64 flags)
213 {
214 	void *ptr;
215 	int err;
216 
217 	if (bpf_map_is_offloaded(map))
218 		return bpf_map_offload_lookup_elem(map, key, value);
219 
220 	bpf_disable_instrumentation();
221 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
222 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
223 		err = bpf_percpu_hash_copy(map, key, value);
224 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
225 		err = bpf_percpu_array_copy(map, key, value);
226 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
227 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
228 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
229 		err = bpf_stackmap_copy(map, key, value);
230 	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
231 		err = bpf_fd_array_map_lookup_elem(map, key, value);
232 	} else if (IS_FD_HASH(map)) {
233 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
234 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
235 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
236 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
237 		   map->map_type == BPF_MAP_TYPE_STACK ||
238 		   map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
239 		err = map->ops->map_peek_elem(map, value);
240 	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
241 		/* struct_ops map requires directly updating "value" */
242 		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
243 	} else {
244 		rcu_read_lock();
245 		if (map->ops->map_lookup_elem_sys_only)
246 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
247 		else
248 			ptr = map->ops->map_lookup_elem(map, key);
249 		if (IS_ERR(ptr)) {
250 			err = PTR_ERR(ptr);
251 		} else if (!ptr) {
252 			err = -ENOENT;
253 		} else {
254 			err = 0;
255 			if (flags & BPF_F_LOCK)
256 				/* lock 'ptr' and copy everything but lock */
257 				copy_map_value_locked(map, value, ptr, true);
258 			else
259 				copy_map_value(map, value, ptr);
260 			/* mask lock and timer, since value wasn't zero inited */
261 			check_and_init_map_value(map, value);
262 		}
263 		rcu_read_unlock();
264 	}
265 
266 	bpf_enable_instrumentation();
267 
268 	return err;
269 }
270 
271 /* Please, do not use this function outside from the map creation path
272  * (e.g. in map update path) without taking care of setting the active
273  * memory cgroup (see at bpf_map_kmalloc_node() for example).
274  */
275 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
276 {
277 	/* We really just want to fail instead of triggering OOM killer
278 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
279 	 * which is used for lower order allocation requests.
280 	 *
281 	 * It has been observed that higher order allocation requests done by
282 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
283 	 * to reclaim memory from the page cache, thus we set
284 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
285 	 */
286 
287 	gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
288 	unsigned int flags = 0;
289 	unsigned long align = 1;
290 	void *area;
291 
292 	if (size >= SIZE_MAX)
293 		return NULL;
294 
295 	/* kmalloc()'ed memory can't be mmap()'ed */
296 	if (mmapable) {
297 		BUG_ON(!PAGE_ALIGNED(size));
298 		align = SHMLBA;
299 		flags = VM_USERMAP;
300 	} else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
301 		area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
302 				    numa_node);
303 		if (area != NULL)
304 			return area;
305 	}
306 
307 	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
308 			gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
309 			flags, numa_node, __builtin_return_address(0));
310 }
311 
312 void *bpf_map_area_alloc(u64 size, int numa_node)
313 {
314 	return __bpf_map_area_alloc(size, numa_node, false);
315 }
316 
317 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
318 {
319 	return __bpf_map_area_alloc(size, numa_node, true);
320 }
321 
322 void bpf_map_area_free(void *area)
323 {
324 	kvfree(area);
325 }
326 
327 static u32 bpf_map_flags_retain_permanent(u32 flags)
328 {
329 	/* Some map creation flags are not tied to the map object but
330 	 * rather to the map fd instead, so they have no meaning upon
331 	 * map object inspection since multiple file descriptors with
332 	 * different (access) properties can exist here. Thus, given
333 	 * this has zero meaning for the map itself, lets clear these
334 	 * from here.
335 	 */
336 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
337 }
338 
339 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
340 {
341 	map->map_type = attr->map_type;
342 	map->key_size = attr->key_size;
343 	map->value_size = attr->value_size;
344 	map->max_entries = attr->max_entries;
345 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
346 	map->numa_node = bpf_map_attr_numa_node(attr);
347 	map->map_extra = attr->map_extra;
348 }
349 
350 static int bpf_map_alloc_id(struct bpf_map *map)
351 {
352 	int id;
353 
354 	idr_preload(GFP_KERNEL);
355 	spin_lock_bh(&map_idr_lock);
356 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
357 	if (id > 0)
358 		map->id = id;
359 	spin_unlock_bh(&map_idr_lock);
360 	idr_preload_end();
361 
362 	if (WARN_ON_ONCE(!id))
363 		return -ENOSPC;
364 
365 	return id > 0 ? 0 : id;
366 }
367 
368 void bpf_map_free_id(struct bpf_map *map)
369 {
370 	unsigned long flags;
371 
372 	/* Offloaded maps are removed from the IDR store when their device
373 	 * disappears - even if someone holds an fd to them they are unusable,
374 	 * the memory is gone, all ops will fail; they are simply waiting for
375 	 * refcnt to drop to be freed.
376 	 */
377 	if (!map->id)
378 		return;
379 
380 	spin_lock_irqsave(&map_idr_lock, flags);
381 
382 	idr_remove(&map_idr, map->id);
383 	map->id = 0;
384 
385 	spin_unlock_irqrestore(&map_idr_lock, flags);
386 }
387 
388 #ifdef CONFIG_MEMCG_KMEM
389 static void bpf_map_save_memcg(struct bpf_map *map)
390 {
391 	/* Currently if a map is created by a process belonging to the root
392 	 * memory cgroup, get_obj_cgroup_from_current() will return NULL.
393 	 * So we have to check map->objcg for being NULL each time it's
394 	 * being used.
395 	 */
396 	if (memcg_bpf_enabled())
397 		map->objcg = get_obj_cgroup_from_current();
398 }
399 
400 static void bpf_map_release_memcg(struct bpf_map *map)
401 {
402 	if (map->objcg)
403 		obj_cgroup_put(map->objcg);
404 }
405 
406 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
407 {
408 	if (map->objcg)
409 		return get_mem_cgroup_from_objcg(map->objcg);
410 
411 	return root_mem_cgroup;
412 }
413 
414 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
415 			   int node)
416 {
417 	struct mem_cgroup *memcg, *old_memcg;
418 	void *ptr;
419 
420 	memcg = bpf_map_get_memcg(map);
421 	old_memcg = set_active_memcg(memcg);
422 	ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
423 	set_active_memcg(old_memcg);
424 	mem_cgroup_put(memcg);
425 
426 	return ptr;
427 }
428 
429 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
430 {
431 	struct mem_cgroup *memcg, *old_memcg;
432 	void *ptr;
433 
434 	memcg = bpf_map_get_memcg(map);
435 	old_memcg = set_active_memcg(memcg);
436 	ptr = kzalloc(size, flags | __GFP_ACCOUNT);
437 	set_active_memcg(old_memcg);
438 	mem_cgroup_put(memcg);
439 
440 	return ptr;
441 }
442 
443 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
444 		       gfp_t flags)
445 {
446 	struct mem_cgroup *memcg, *old_memcg;
447 	void *ptr;
448 
449 	memcg = bpf_map_get_memcg(map);
450 	old_memcg = set_active_memcg(memcg);
451 	ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
452 	set_active_memcg(old_memcg);
453 	mem_cgroup_put(memcg);
454 
455 	return ptr;
456 }
457 
458 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
459 				    size_t align, gfp_t flags)
460 {
461 	struct mem_cgroup *memcg, *old_memcg;
462 	void __percpu *ptr;
463 
464 	memcg = bpf_map_get_memcg(map);
465 	old_memcg = set_active_memcg(memcg);
466 	ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
467 	set_active_memcg(old_memcg);
468 	mem_cgroup_put(memcg);
469 
470 	return ptr;
471 }
472 
473 #else
474 static void bpf_map_save_memcg(struct bpf_map *map)
475 {
476 }
477 
478 static void bpf_map_release_memcg(struct bpf_map *map)
479 {
480 }
481 #endif
482 
483 int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
484 			unsigned long nr_pages, struct page **pages)
485 {
486 	unsigned long i, j;
487 	struct page *pg;
488 	int ret = 0;
489 #ifdef CONFIG_MEMCG_KMEM
490 	struct mem_cgroup *memcg, *old_memcg;
491 
492 	memcg = bpf_map_get_memcg(map);
493 	old_memcg = set_active_memcg(memcg);
494 #endif
495 	for (i = 0; i < nr_pages; i++) {
496 		pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
497 
498 		if (pg) {
499 			pages[i] = pg;
500 			continue;
501 		}
502 		for (j = 0; j < i; j++)
503 			__free_page(pages[j]);
504 		ret = -ENOMEM;
505 		break;
506 	}
507 
508 #ifdef CONFIG_MEMCG_KMEM
509 	set_active_memcg(old_memcg);
510 	mem_cgroup_put(memcg);
511 #endif
512 	return ret;
513 }
514 
515 
516 static int btf_field_cmp(const void *a, const void *b)
517 {
518 	const struct btf_field *f1 = a, *f2 = b;
519 
520 	if (f1->offset < f2->offset)
521 		return -1;
522 	else if (f1->offset > f2->offset)
523 		return 1;
524 	return 0;
525 }
526 
527 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
528 				  u32 field_mask)
529 {
530 	struct btf_field *field;
531 
532 	if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
533 		return NULL;
534 	field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
535 	if (!field || !(field->type & field_mask))
536 		return NULL;
537 	return field;
538 }
539 
540 void btf_record_free(struct btf_record *rec)
541 {
542 	int i;
543 
544 	if (IS_ERR_OR_NULL(rec))
545 		return;
546 	for (i = 0; i < rec->cnt; i++) {
547 		switch (rec->fields[i].type) {
548 		case BPF_KPTR_UNREF:
549 		case BPF_KPTR_REF:
550 		case BPF_KPTR_PERCPU:
551 			if (rec->fields[i].kptr.module)
552 				module_put(rec->fields[i].kptr.module);
553 			btf_put(rec->fields[i].kptr.btf);
554 			break;
555 		case BPF_LIST_HEAD:
556 		case BPF_LIST_NODE:
557 		case BPF_RB_ROOT:
558 		case BPF_RB_NODE:
559 		case BPF_SPIN_LOCK:
560 		case BPF_TIMER:
561 		case BPF_REFCOUNT:
562 			/* Nothing to release */
563 			break;
564 		default:
565 			WARN_ON_ONCE(1);
566 			continue;
567 		}
568 	}
569 	kfree(rec);
570 }
571 
572 void bpf_map_free_record(struct bpf_map *map)
573 {
574 	btf_record_free(map->record);
575 	map->record = NULL;
576 }
577 
578 struct btf_record *btf_record_dup(const struct btf_record *rec)
579 {
580 	const struct btf_field *fields;
581 	struct btf_record *new_rec;
582 	int ret, size, i;
583 
584 	if (IS_ERR_OR_NULL(rec))
585 		return NULL;
586 	size = offsetof(struct btf_record, fields[rec->cnt]);
587 	new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
588 	if (!new_rec)
589 		return ERR_PTR(-ENOMEM);
590 	/* Do a deep copy of the btf_record */
591 	fields = rec->fields;
592 	new_rec->cnt = 0;
593 	for (i = 0; i < rec->cnt; i++) {
594 		switch (fields[i].type) {
595 		case BPF_KPTR_UNREF:
596 		case BPF_KPTR_REF:
597 		case BPF_KPTR_PERCPU:
598 			btf_get(fields[i].kptr.btf);
599 			if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
600 				ret = -ENXIO;
601 				goto free;
602 			}
603 			break;
604 		case BPF_LIST_HEAD:
605 		case BPF_LIST_NODE:
606 		case BPF_RB_ROOT:
607 		case BPF_RB_NODE:
608 		case BPF_SPIN_LOCK:
609 		case BPF_TIMER:
610 		case BPF_REFCOUNT:
611 			/* Nothing to acquire */
612 			break;
613 		default:
614 			ret = -EFAULT;
615 			WARN_ON_ONCE(1);
616 			goto free;
617 		}
618 		new_rec->cnt++;
619 	}
620 	return new_rec;
621 free:
622 	btf_record_free(new_rec);
623 	return ERR_PTR(ret);
624 }
625 
626 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
627 {
628 	bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
629 	int size;
630 
631 	if (!a_has_fields && !b_has_fields)
632 		return true;
633 	if (a_has_fields != b_has_fields)
634 		return false;
635 	if (rec_a->cnt != rec_b->cnt)
636 		return false;
637 	size = offsetof(struct btf_record, fields[rec_a->cnt]);
638 	/* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
639 	 * members are zeroed out. So memcmp is safe to do without worrying
640 	 * about padding/unused fields.
641 	 *
642 	 * While spin_lock, timer, and kptr have no relation to map BTF,
643 	 * list_head metadata is specific to map BTF, the btf and value_rec
644 	 * members in particular. btf is the map BTF, while value_rec points to
645 	 * btf_record in that map BTF.
646 	 *
647 	 * So while by default, we don't rely on the map BTF (which the records
648 	 * were parsed from) matching for both records, which is not backwards
649 	 * compatible, in case list_head is part of it, we implicitly rely on
650 	 * that by way of depending on memcmp succeeding for it.
651 	 */
652 	return !memcmp(rec_a, rec_b, size);
653 }
654 
655 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
656 {
657 	if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
658 		return;
659 	bpf_timer_cancel_and_free(obj + rec->timer_off);
660 }
661 
662 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
663 {
664 	const struct btf_field *fields;
665 	int i;
666 
667 	if (IS_ERR_OR_NULL(rec))
668 		return;
669 	fields = rec->fields;
670 	for (i = 0; i < rec->cnt; i++) {
671 		struct btf_struct_meta *pointee_struct_meta;
672 		const struct btf_field *field = &fields[i];
673 		void *field_ptr = obj + field->offset;
674 		void *xchgd_field;
675 
676 		switch (fields[i].type) {
677 		case BPF_SPIN_LOCK:
678 			break;
679 		case BPF_TIMER:
680 			bpf_timer_cancel_and_free(field_ptr);
681 			break;
682 		case BPF_KPTR_UNREF:
683 			WRITE_ONCE(*(u64 *)field_ptr, 0);
684 			break;
685 		case BPF_KPTR_REF:
686 		case BPF_KPTR_PERCPU:
687 			xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
688 			if (!xchgd_field)
689 				break;
690 
691 			if (!btf_is_kernel(field->kptr.btf)) {
692 				pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
693 									   field->kptr.btf_id);
694 				migrate_disable();
695 				__bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
696 								 pointee_struct_meta->record : NULL,
697 								 fields[i].type == BPF_KPTR_PERCPU);
698 				migrate_enable();
699 			} else {
700 				field->kptr.dtor(xchgd_field);
701 			}
702 			break;
703 		case BPF_LIST_HEAD:
704 			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
705 				continue;
706 			bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
707 			break;
708 		case BPF_RB_ROOT:
709 			if (WARN_ON_ONCE(rec->spin_lock_off < 0))
710 				continue;
711 			bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
712 			break;
713 		case BPF_LIST_NODE:
714 		case BPF_RB_NODE:
715 		case BPF_REFCOUNT:
716 			break;
717 		default:
718 			WARN_ON_ONCE(1);
719 			continue;
720 		}
721 	}
722 }
723 
724 /* called from workqueue */
725 static void bpf_map_free_deferred(struct work_struct *work)
726 {
727 	struct bpf_map *map = container_of(work, struct bpf_map, work);
728 	struct btf_record *rec = map->record;
729 	struct btf *btf = map->btf;
730 
731 	security_bpf_map_free(map);
732 	bpf_map_release_memcg(map);
733 	/* implementation dependent freeing */
734 	map->ops->map_free(map);
735 	/* Delay freeing of btf_record for maps, as map_free
736 	 * callback usually needs access to them. It is better to do it here
737 	 * than require each callback to do the free itself manually.
738 	 *
739 	 * Note that the btf_record stashed in map->inner_map_meta->record was
740 	 * already freed using the map_free callback for map in map case which
741 	 * eventually calls bpf_map_free_meta, since inner_map_meta is only a
742 	 * template bpf_map struct used during verification.
743 	 */
744 	btf_record_free(rec);
745 	/* Delay freeing of btf for maps, as map_free callback may need
746 	 * struct_meta info which will be freed with btf_put().
747 	 */
748 	btf_put(btf);
749 }
750 
751 static void bpf_map_put_uref(struct bpf_map *map)
752 {
753 	if (atomic64_dec_and_test(&map->usercnt)) {
754 		if (map->ops->map_release_uref)
755 			map->ops->map_release_uref(map);
756 	}
757 }
758 
759 static void bpf_map_free_in_work(struct bpf_map *map)
760 {
761 	INIT_WORK(&map->work, bpf_map_free_deferred);
762 	/* Avoid spawning kworkers, since they all might contend
763 	 * for the same mutex like slab_mutex.
764 	 */
765 	queue_work(system_unbound_wq, &map->work);
766 }
767 
768 static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
769 {
770 	bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
771 }
772 
773 static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
774 {
775 	if (rcu_trace_implies_rcu_gp())
776 		bpf_map_free_rcu_gp(rcu);
777 	else
778 		call_rcu(rcu, bpf_map_free_rcu_gp);
779 }
780 
781 /* decrement map refcnt and schedule it for freeing via workqueue
782  * (underlying map implementation ops->map_free() might sleep)
783  */
784 void bpf_map_put(struct bpf_map *map)
785 {
786 	if (atomic64_dec_and_test(&map->refcnt)) {
787 		/* bpf_map_free_id() must be called first */
788 		bpf_map_free_id(map);
789 
790 		WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
791 		if (READ_ONCE(map->free_after_mult_rcu_gp))
792 			call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
793 		else if (READ_ONCE(map->free_after_rcu_gp))
794 			call_rcu(&map->rcu, bpf_map_free_rcu_gp);
795 		else
796 			bpf_map_free_in_work(map);
797 	}
798 }
799 EXPORT_SYMBOL_GPL(bpf_map_put);
800 
801 void bpf_map_put_with_uref(struct bpf_map *map)
802 {
803 	bpf_map_put_uref(map);
804 	bpf_map_put(map);
805 }
806 
807 static int bpf_map_release(struct inode *inode, struct file *filp)
808 {
809 	struct bpf_map *map = filp->private_data;
810 
811 	if (map->ops->map_release)
812 		map->ops->map_release(map, filp);
813 
814 	bpf_map_put_with_uref(map);
815 	return 0;
816 }
817 
818 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
819 {
820 	fmode_t mode = f.file->f_mode;
821 
822 	/* Our file permissions may have been overridden by global
823 	 * map permissions facing syscall side.
824 	 */
825 	if (READ_ONCE(map->frozen))
826 		mode &= ~FMODE_CAN_WRITE;
827 	return mode;
828 }
829 
830 #ifdef CONFIG_PROC_FS
831 /* Show the memory usage of a bpf map */
832 static u64 bpf_map_memory_usage(const struct bpf_map *map)
833 {
834 	return map->ops->map_mem_usage(map);
835 }
836 
837 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
838 {
839 	struct bpf_map *map = filp->private_data;
840 	u32 type = 0, jited = 0;
841 
842 	if (map_type_contains_progs(map)) {
843 		spin_lock(&map->owner.lock);
844 		type  = map->owner.type;
845 		jited = map->owner.jited;
846 		spin_unlock(&map->owner.lock);
847 	}
848 
849 	seq_printf(m,
850 		   "map_type:\t%u\n"
851 		   "key_size:\t%u\n"
852 		   "value_size:\t%u\n"
853 		   "max_entries:\t%u\n"
854 		   "map_flags:\t%#x\n"
855 		   "map_extra:\t%#llx\n"
856 		   "memlock:\t%llu\n"
857 		   "map_id:\t%u\n"
858 		   "frozen:\t%u\n",
859 		   map->map_type,
860 		   map->key_size,
861 		   map->value_size,
862 		   map->max_entries,
863 		   map->map_flags,
864 		   (unsigned long long)map->map_extra,
865 		   bpf_map_memory_usage(map),
866 		   map->id,
867 		   READ_ONCE(map->frozen));
868 	if (type) {
869 		seq_printf(m, "owner_prog_type:\t%u\n", type);
870 		seq_printf(m, "owner_jited:\t%u\n", jited);
871 	}
872 }
873 #endif
874 
875 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
876 			      loff_t *ppos)
877 {
878 	/* We need this handler such that alloc_file() enables
879 	 * f_mode with FMODE_CAN_READ.
880 	 */
881 	return -EINVAL;
882 }
883 
884 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
885 			       size_t siz, loff_t *ppos)
886 {
887 	/* We need this handler such that alloc_file() enables
888 	 * f_mode with FMODE_CAN_WRITE.
889 	 */
890 	return -EINVAL;
891 }
892 
893 /* called for any extra memory-mapped regions (except initial) */
894 static void bpf_map_mmap_open(struct vm_area_struct *vma)
895 {
896 	struct bpf_map *map = vma->vm_file->private_data;
897 
898 	if (vma->vm_flags & VM_MAYWRITE)
899 		bpf_map_write_active_inc(map);
900 }
901 
902 /* called for all unmapped memory region (including initial) */
903 static void bpf_map_mmap_close(struct vm_area_struct *vma)
904 {
905 	struct bpf_map *map = vma->vm_file->private_data;
906 
907 	if (vma->vm_flags & VM_MAYWRITE)
908 		bpf_map_write_active_dec(map);
909 }
910 
911 static const struct vm_operations_struct bpf_map_default_vmops = {
912 	.open		= bpf_map_mmap_open,
913 	.close		= bpf_map_mmap_close,
914 };
915 
916 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
917 {
918 	struct bpf_map *map = filp->private_data;
919 	int err;
920 
921 	if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
922 		return -ENOTSUPP;
923 
924 	if (!(vma->vm_flags & VM_SHARED))
925 		return -EINVAL;
926 
927 	mutex_lock(&map->freeze_mutex);
928 
929 	if (vma->vm_flags & VM_WRITE) {
930 		if (map->frozen) {
931 			err = -EPERM;
932 			goto out;
933 		}
934 		/* map is meant to be read-only, so do not allow mapping as
935 		 * writable, because it's possible to leak a writable page
936 		 * reference and allows user-space to still modify it after
937 		 * freezing, while verifier will assume contents do not change
938 		 */
939 		if (map->map_flags & BPF_F_RDONLY_PROG) {
940 			err = -EACCES;
941 			goto out;
942 		}
943 	}
944 
945 	/* set default open/close callbacks */
946 	vma->vm_ops = &bpf_map_default_vmops;
947 	vma->vm_private_data = map;
948 	vm_flags_clear(vma, VM_MAYEXEC);
949 	if (!(vma->vm_flags & VM_WRITE))
950 		/* disallow re-mapping with PROT_WRITE */
951 		vm_flags_clear(vma, VM_MAYWRITE);
952 
953 	err = map->ops->map_mmap(map, vma);
954 	if (err)
955 		goto out;
956 
957 	if (vma->vm_flags & VM_MAYWRITE)
958 		bpf_map_write_active_inc(map);
959 out:
960 	mutex_unlock(&map->freeze_mutex);
961 	return err;
962 }
963 
964 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
965 {
966 	struct bpf_map *map = filp->private_data;
967 
968 	if (map->ops->map_poll)
969 		return map->ops->map_poll(map, filp, pts);
970 
971 	return EPOLLERR;
972 }
973 
974 static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
975 					   unsigned long len, unsigned long pgoff,
976 					   unsigned long flags)
977 {
978 	struct bpf_map *map = filp->private_data;
979 
980 	if (map->ops->map_get_unmapped_area)
981 		return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
982 #ifdef CONFIG_MMU
983 	return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
984 #else
985 	return addr;
986 #endif
987 }
988 
989 const struct file_operations bpf_map_fops = {
990 #ifdef CONFIG_PROC_FS
991 	.show_fdinfo	= bpf_map_show_fdinfo,
992 #endif
993 	.release	= bpf_map_release,
994 	.read		= bpf_dummy_read,
995 	.write		= bpf_dummy_write,
996 	.mmap		= bpf_map_mmap,
997 	.poll		= bpf_map_poll,
998 	.get_unmapped_area = bpf_get_unmapped_area,
999 };
1000 
1001 int bpf_map_new_fd(struct bpf_map *map, int flags)
1002 {
1003 	int ret;
1004 
1005 	ret = security_bpf_map(map, OPEN_FMODE(flags));
1006 	if (ret < 0)
1007 		return ret;
1008 
1009 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1010 				flags | O_CLOEXEC);
1011 }
1012 
1013 int bpf_get_file_flag(int flags)
1014 {
1015 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1016 		return -EINVAL;
1017 	if (flags & BPF_F_RDONLY)
1018 		return O_RDONLY;
1019 	if (flags & BPF_F_WRONLY)
1020 		return O_WRONLY;
1021 	return O_RDWR;
1022 }
1023 
1024 /* helper macro to check that unused fields 'union bpf_attr' are zero */
1025 #define CHECK_ATTR(CMD) \
1026 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1027 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
1028 		   sizeof(*attr) - \
1029 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1030 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
1031 
1032 /* dst and src must have at least "size" number of bytes.
1033  * Return strlen on success and < 0 on error.
1034  */
1035 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1036 {
1037 	const char *end = src + size;
1038 	const char *orig_src = src;
1039 
1040 	memset(dst, 0, size);
1041 	/* Copy all isalnum(), '_' and '.' chars. */
1042 	while (src < end && *src) {
1043 		if (!isalnum(*src) &&
1044 		    *src != '_' && *src != '.')
1045 			return -EINVAL;
1046 		*dst++ = *src++;
1047 	}
1048 
1049 	/* No '\0' found in "size" number of bytes */
1050 	if (src == end)
1051 		return -EINVAL;
1052 
1053 	return src - orig_src;
1054 }
1055 
1056 int map_check_no_btf(const struct bpf_map *map,
1057 		     const struct btf *btf,
1058 		     const struct btf_type *key_type,
1059 		     const struct btf_type *value_type)
1060 {
1061 	return -ENOTSUPP;
1062 }
1063 
1064 static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1065 			 const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1066 {
1067 	const struct btf_type *key_type, *value_type;
1068 	u32 key_size, value_size;
1069 	int ret = 0;
1070 
1071 	/* Some maps allow key to be unspecified. */
1072 	if (btf_key_id) {
1073 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1074 		if (!key_type || key_size != map->key_size)
1075 			return -EINVAL;
1076 	} else {
1077 		key_type = btf_type_by_id(btf, 0);
1078 		if (!map->ops->map_check_btf)
1079 			return -EINVAL;
1080 	}
1081 
1082 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1083 	if (!value_type || value_size != map->value_size)
1084 		return -EINVAL;
1085 
1086 	map->record = btf_parse_fields(btf, value_type,
1087 				       BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1088 				       BPF_RB_ROOT | BPF_REFCOUNT,
1089 				       map->value_size);
1090 	if (!IS_ERR_OR_NULL(map->record)) {
1091 		int i;
1092 
1093 		if (!bpf_token_capable(token, CAP_BPF)) {
1094 			ret = -EPERM;
1095 			goto free_map_tab;
1096 		}
1097 		if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1098 			ret = -EACCES;
1099 			goto free_map_tab;
1100 		}
1101 		for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1102 			switch (map->record->field_mask & (1 << i)) {
1103 			case 0:
1104 				continue;
1105 			case BPF_SPIN_LOCK:
1106 				if (map->map_type != BPF_MAP_TYPE_HASH &&
1107 				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1108 				    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1109 				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1110 				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1111 				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1112 				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1113 					ret = -EOPNOTSUPP;
1114 					goto free_map_tab;
1115 				}
1116 				break;
1117 			case BPF_TIMER:
1118 				if (map->map_type != BPF_MAP_TYPE_HASH &&
1119 				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1120 				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1121 					ret = -EOPNOTSUPP;
1122 					goto free_map_tab;
1123 				}
1124 				break;
1125 			case BPF_KPTR_UNREF:
1126 			case BPF_KPTR_REF:
1127 			case BPF_KPTR_PERCPU:
1128 			case BPF_REFCOUNT:
1129 				if (map->map_type != BPF_MAP_TYPE_HASH &&
1130 				    map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1131 				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1132 				    map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1133 				    map->map_type != BPF_MAP_TYPE_ARRAY &&
1134 				    map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1135 				    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1136 				    map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1137 				    map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1138 				    map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1139 					ret = -EOPNOTSUPP;
1140 					goto free_map_tab;
1141 				}
1142 				break;
1143 			case BPF_LIST_HEAD:
1144 			case BPF_RB_ROOT:
1145 				if (map->map_type != BPF_MAP_TYPE_HASH &&
1146 				    map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1147 				    map->map_type != BPF_MAP_TYPE_ARRAY) {
1148 					ret = -EOPNOTSUPP;
1149 					goto free_map_tab;
1150 				}
1151 				break;
1152 			default:
1153 				/* Fail if map_type checks are missing for a field type */
1154 				ret = -EOPNOTSUPP;
1155 				goto free_map_tab;
1156 			}
1157 		}
1158 	}
1159 
1160 	ret = btf_check_and_fixup_fields(btf, map->record);
1161 	if (ret < 0)
1162 		goto free_map_tab;
1163 
1164 	if (map->ops->map_check_btf) {
1165 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1166 		if (ret < 0)
1167 			goto free_map_tab;
1168 	}
1169 
1170 	return ret;
1171 free_map_tab:
1172 	bpf_map_free_record(map);
1173 	return ret;
1174 }
1175 
1176 static bool bpf_net_capable(void)
1177 {
1178 	return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1179 }
1180 
1181 #define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1182 /* called via syscall */
1183 static int map_create(union bpf_attr *attr)
1184 {
1185 	const struct bpf_map_ops *ops;
1186 	struct bpf_token *token = NULL;
1187 	int numa_node = bpf_map_attr_numa_node(attr);
1188 	u32 map_type = attr->map_type;
1189 	struct bpf_map *map;
1190 	bool token_flag;
1191 	int f_flags;
1192 	int err;
1193 
1194 	err = CHECK_ATTR(BPF_MAP_CREATE);
1195 	if (err)
1196 		return -EINVAL;
1197 
1198 	/* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1199 	 * to avoid per-map type checks tripping on unknown flag
1200 	 */
1201 	token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1202 	attr->map_flags &= ~BPF_F_TOKEN_FD;
1203 
1204 	if (attr->btf_vmlinux_value_type_id) {
1205 		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1206 		    attr->btf_key_type_id || attr->btf_value_type_id)
1207 			return -EINVAL;
1208 	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1209 		return -EINVAL;
1210 	}
1211 
1212 	if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1213 	    attr->map_type != BPF_MAP_TYPE_ARENA &&
1214 	    attr->map_extra != 0)
1215 		return -EINVAL;
1216 
1217 	f_flags = bpf_get_file_flag(attr->map_flags);
1218 	if (f_flags < 0)
1219 		return f_flags;
1220 
1221 	if (numa_node != NUMA_NO_NODE &&
1222 	    ((unsigned int)numa_node >= nr_node_ids ||
1223 	     !node_online(numa_node)))
1224 		return -EINVAL;
1225 
1226 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1227 	map_type = attr->map_type;
1228 	if (map_type >= ARRAY_SIZE(bpf_map_types))
1229 		return -EINVAL;
1230 	map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1231 	ops = bpf_map_types[map_type];
1232 	if (!ops)
1233 		return -EINVAL;
1234 
1235 	if (ops->map_alloc_check) {
1236 		err = ops->map_alloc_check(attr);
1237 		if (err)
1238 			return err;
1239 	}
1240 	if (attr->map_ifindex)
1241 		ops = &bpf_map_offload_ops;
1242 	if (!ops->map_mem_usage)
1243 		return -EINVAL;
1244 
1245 	if (token_flag) {
1246 		token = bpf_token_get_from_fd(attr->map_token_fd);
1247 		if (IS_ERR(token))
1248 			return PTR_ERR(token);
1249 
1250 		/* if current token doesn't grant map creation permissions,
1251 		 * then we can't use this token, so ignore it and rely on
1252 		 * system-wide capabilities checks
1253 		 */
1254 		if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1255 		    !bpf_token_allow_map_type(token, attr->map_type)) {
1256 			bpf_token_put(token);
1257 			token = NULL;
1258 		}
1259 	}
1260 
1261 	err = -EPERM;
1262 
1263 	/* Intent here is for unprivileged_bpf_disabled to block BPF map
1264 	 * creation for unprivileged users; other actions depend
1265 	 * on fd availability and access to bpffs, so are dependent on
1266 	 * object creation success. Even with unprivileged BPF disabled,
1267 	 * capability checks are still carried out.
1268 	 */
1269 	if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1270 		goto put_token;
1271 
1272 	/* check privileged map type permissions */
1273 	switch (map_type) {
1274 	case BPF_MAP_TYPE_ARRAY:
1275 	case BPF_MAP_TYPE_PERCPU_ARRAY:
1276 	case BPF_MAP_TYPE_PROG_ARRAY:
1277 	case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1278 	case BPF_MAP_TYPE_CGROUP_ARRAY:
1279 	case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1280 	case BPF_MAP_TYPE_HASH:
1281 	case BPF_MAP_TYPE_PERCPU_HASH:
1282 	case BPF_MAP_TYPE_HASH_OF_MAPS:
1283 	case BPF_MAP_TYPE_RINGBUF:
1284 	case BPF_MAP_TYPE_USER_RINGBUF:
1285 	case BPF_MAP_TYPE_CGROUP_STORAGE:
1286 	case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1287 		/* unprivileged */
1288 		break;
1289 	case BPF_MAP_TYPE_SK_STORAGE:
1290 	case BPF_MAP_TYPE_INODE_STORAGE:
1291 	case BPF_MAP_TYPE_TASK_STORAGE:
1292 	case BPF_MAP_TYPE_CGRP_STORAGE:
1293 	case BPF_MAP_TYPE_BLOOM_FILTER:
1294 	case BPF_MAP_TYPE_LPM_TRIE:
1295 	case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1296 	case BPF_MAP_TYPE_STACK_TRACE:
1297 	case BPF_MAP_TYPE_QUEUE:
1298 	case BPF_MAP_TYPE_STACK:
1299 	case BPF_MAP_TYPE_LRU_HASH:
1300 	case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1301 	case BPF_MAP_TYPE_STRUCT_OPS:
1302 	case BPF_MAP_TYPE_CPUMAP:
1303 	case BPF_MAP_TYPE_ARENA:
1304 		if (!bpf_token_capable(token, CAP_BPF))
1305 			goto put_token;
1306 		break;
1307 	case BPF_MAP_TYPE_SOCKMAP:
1308 	case BPF_MAP_TYPE_SOCKHASH:
1309 	case BPF_MAP_TYPE_DEVMAP:
1310 	case BPF_MAP_TYPE_DEVMAP_HASH:
1311 	case BPF_MAP_TYPE_XSKMAP:
1312 		if (!bpf_token_capable(token, CAP_NET_ADMIN))
1313 			goto put_token;
1314 		break;
1315 	default:
1316 		WARN(1, "unsupported map type %d", map_type);
1317 		goto put_token;
1318 	}
1319 
1320 	map = ops->map_alloc(attr);
1321 	if (IS_ERR(map)) {
1322 		err = PTR_ERR(map);
1323 		goto put_token;
1324 	}
1325 	map->ops = ops;
1326 	map->map_type = map_type;
1327 
1328 	err = bpf_obj_name_cpy(map->name, attr->map_name,
1329 			       sizeof(attr->map_name));
1330 	if (err < 0)
1331 		goto free_map;
1332 
1333 	atomic64_set(&map->refcnt, 1);
1334 	atomic64_set(&map->usercnt, 1);
1335 	mutex_init(&map->freeze_mutex);
1336 	spin_lock_init(&map->owner.lock);
1337 
1338 	if (attr->btf_key_type_id || attr->btf_value_type_id ||
1339 	    /* Even the map's value is a kernel's struct,
1340 	     * the bpf_prog.o must have BTF to begin with
1341 	     * to figure out the corresponding kernel's
1342 	     * counter part.  Thus, attr->btf_fd has
1343 	     * to be valid also.
1344 	     */
1345 	    attr->btf_vmlinux_value_type_id) {
1346 		struct btf *btf;
1347 
1348 		btf = btf_get_by_fd(attr->btf_fd);
1349 		if (IS_ERR(btf)) {
1350 			err = PTR_ERR(btf);
1351 			goto free_map;
1352 		}
1353 		if (btf_is_kernel(btf)) {
1354 			btf_put(btf);
1355 			err = -EACCES;
1356 			goto free_map;
1357 		}
1358 		map->btf = btf;
1359 
1360 		if (attr->btf_value_type_id) {
1361 			err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1362 					    attr->btf_value_type_id);
1363 			if (err)
1364 				goto free_map;
1365 		}
1366 
1367 		map->btf_key_type_id = attr->btf_key_type_id;
1368 		map->btf_value_type_id = attr->btf_value_type_id;
1369 		map->btf_vmlinux_value_type_id =
1370 			attr->btf_vmlinux_value_type_id;
1371 	}
1372 
1373 	err = security_bpf_map_create(map, attr, token);
1374 	if (err)
1375 		goto free_map_sec;
1376 
1377 	err = bpf_map_alloc_id(map);
1378 	if (err)
1379 		goto free_map_sec;
1380 
1381 	bpf_map_save_memcg(map);
1382 	bpf_token_put(token);
1383 
1384 	err = bpf_map_new_fd(map, f_flags);
1385 	if (err < 0) {
1386 		/* failed to allocate fd.
1387 		 * bpf_map_put_with_uref() is needed because the above
1388 		 * bpf_map_alloc_id() has published the map
1389 		 * to the userspace and the userspace may
1390 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1391 		 */
1392 		bpf_map_put_with_uref(map);
1393 		return err;
1394 	}
1395 
1396 	return err;
1397 
1398 free_map_sec:
1399 	security_bpf_map_free(map);
1400 free_map:
1401 	btf_put(map->btf);
1402 	map->ops->map_free(map);
1403 put_token:
1404 	bpf_token_put(token);
1405 	return err;
1406 }
1407 
1408 /* if error is returned, fd is released.
1409  * On success caller should complete fd access with matching fdput()
1410  */
1411 struct bpf_map *__bpf_map_get(struct fd f)
1412 {
1413 	if (!f.file)
1414 		return ERR_PTR(-EBADF);
1415 	if (f.file->f_op != &bpf_map_fops) {
1416 		fdput(f);
1417 		return ERR_PTR(-EINVAL);
1418 	}
1419 
1420 	return f.file->private_data;
1421 }
1422 
1423 void bpf_map_inc(struct bpf_map *map)
1424 {
1425 	atomic64_inc(&map->refcnt);
1426 }
1427 EXPORT_SYMBOL_GPL(bpf_map_inc);
1428 
1429 void bpf_map_inc_with_uref(struct bpf_map *map)
1430 {
1431 	atomic64_inc(&map->refcnt);
1432 	atomic64_inc(&map->usercnt);
1433 }
1434 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1435 
1436 struct bpf_map *bpf_map_get(u32 ufd)
1437 {
1438 	struct fd f = fdget(ufd);
1439 	struct bpf_map *map;
1440 
1441 	map = __bpf_map_get(f);
1442 	if (IS_ERR(map))
1443 		return map;
1444 
1445 	bpf_map_inc(map);
1446 	fdput(f);
1447 
1448 	return map;
1449 }
1450 EXPORT_SYMBOL(bpf_map_get);
1451 
1452 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1453 {
1454 	struct fd f = fdget(ufd);
1455 	struct bpf_map *map;
1456 
1457 	map = __bpf_map_get(f);
1458 	if (IS_ERR(map))
1459 		return map;
1460 
1461 	bpf_map_inc_with_uref(map);
1462 	fdput(f);
1463 
1464 	return map;
1465 }
1466 
1467 /* map_idr_lock should have been held or the map should have been
1468  * protected by rcu read lock.
1469  */
1470 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1471 {
1472 	int refold;
1473 
1474 	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1475 	if (!refold)
1476 		return ERR_PTR(-ENOENT);
1477 	if (uref)
1478 		atomic64_inc(&map->usercnt);
1479 
1480 	return map;
1481 }
1482 
1483 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1484 {
1485 	spin_lock_bh(&map_idr_lock);
1486 	map = __bpf_map_inc_not_zero(map, false);
1487 	spin_unlock_bh(&map_idr_lock);
1488 
1489 	return map;
1490 }
1491 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1492 
1493 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1494 {
1495 	return -ENOTSUPP;
1496 }
1497 
1498 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1499 {
1500 	if (key_size)
1501 		return vmemdup_user(ukey, key_size);
1502 
1503 	if (ukey)
1504 		return ERR_PTR(-EINVAL);
1505 
1506 	return NULL;
1507 }
1508 
1509 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1510 {
1511 	if (key_size)
1512 		return kvmemdup_bpfptr(ukey, key_size);
1513 
1514 	if (!bpfptr_is_null(ukey))
1515 		return ERR_PTR(-EINVAL);
1516 
1517 	return NULL;
1518 }
1519 
1520 /* last field in 'union bpf_attr' used by this command */
1521 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1522 
1523 static int map_lookup_elem(union bpf_attr *attr)
1524 {
1525 	void __user *ukey = u64_to_user_ptr(attr->key);
1526 	void __user *uvalue = u64_to_user_ptr(attr->value);
1527 	int ufd = attr->map_fd;
1528 	struct bpf_map *map;
1529 	void *key, *value;
1530 	u32 value_size;
1531 	struct fd f;
1532 	int err;
1533 
1534 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1535 		return -EINVAL;
1536 
1537 	if (attr->flags & ~BPF_F_LOCK)
1538 		return -EINVAL;
1539 
1540 	f = fdget(ufd);
1541 	map = __bpf_map_get(f);
1542 	if (IS_ERR(map))
1543 		return PTR_ERR(map);
1544 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1545 		err = -EPERM;
1546 		goto err_put;
1547 	}
1548 
1549 	if ((attr->flags & BPF_F_LOCK) &&
1550 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1551 		err = -EINVAL;
1552 		goto err_put;
1553 	}
1554 
1555 	key = __bpf_copy_key(ukey, map->key_size);
1556 	if (IS_ERR(key)) {
1557 		err = PTR_ERR(key);
1558 		goto err_put;
1559 	}
1560 
1561 	value_size = bpf_map_value_size(map);
1562 
1563 	err = -ENOMEM;
1564 	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1565 	if (!value)
1566 		goto free_key;
1567 
1568 	if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1569 		if (copy_from_user(value, uvalue, value_size))
1570 			err = -EFAULT;
1571 		else
1572 			err = bpf_map_copy_value(map, key, value, attr->flags);
1573 		goto free_value;
1574 	}
1575 
1576 	err = bpf_map_copy_value(map, key, value, attr->flags);
1577 	if (err)
1578 		goto free_value;
1579 
1580 	err = -EFAULT;
1581 	if (copy_to_user(uvalue, value, value_size) != 0)
1582 		goto free_value;
1583 
1584 	err = 0;
1585 
1586 free_value:
1587 	kvfree(value);
1588 free_key:
1589 	kvfree(key);
1590 err_put:
1591 	fdput(f);
1592 	return err;
1593 }
1594 
1595 
1596 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1597 
1598 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1599 {
1600 	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1601 	bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1602 	int ufd = attr->map_fd;
1603 	struct bpf_map *map;
1604 	void *key, *value;
1605 	u32 value_size;
1606 	struct fd f;
1607 	int err;
1608 
1609 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1610 		return -EINVAL;
1611 
1612 	f = fdget(ufd);
1613 	map = __bpf_map_get(f);
1614 	if (IS_ERR(map))
1615 		return PTR_ERR(map);
1616 	bpf_map_write_active_inc(map);
1617 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1618 		err = -EPERM;
1619 		goto err_put;
1620 	}
1621 
1622 	if ((attr->flags & BPF_F_LOCK) &&
1623 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1624 		err = -EINVAL;
1625 		goto err_put;
1626 	}
1627 
1628 	key = ___bpf_copy_key(ukey, map->key_size);
1629 	if (IS_ERR(key)) {
1630 		err = PTR_ERR(key);
1631 		goto err_put;
1632 	}
1633 
1634 	value_size = bpf_map_value_size(map);
1635 	value = kvmemdup_bpfptr(uvalue, value_size);
1636 	if (IS_ERR(value)) {
1637 		err = PTR_ERR(value);
1638 		goto free_key;
1639 	}
1640 
1641 	err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1642 	if (!err)
1643 		maybe_wait_bpf_programs(map);
1644 
1645 	kvfree(value);
1646 free_key:
1647 	kvfree(key);
1648 err_put:
1649 	bpf_map_write_active_dec(map);
1650 	fdput(f);
1651 	return err;
1652 }
1653 
1654 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1655 
1656 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1657 {
1658 	bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1659 	int ufd = attr->map_fd;
1660 	struct bpf_map *map;
1661 	struct fd f;
1662 	void *key;
1663 	int err;
1664 
1665 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1666 		return -EINVAL;
1667 
1668 	f = fdget(ufd);
1669 	map = __bpf_map_get(f);
1670 	if (IS_ERR(map))
1671 		return PTR_ERR(map);
1672 	bpf_map_write_active_inc(map);
1673 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1674 		err = -EPERM;
1675 		goto err_put;
1676 	}
1677 
1678 	key = ___bpf_copy_key(ukey, map->key_size);
1679 	if (IS_ERR(key)) {
1680 		err = PTR_ERR(key);
1681 		goto err_put;
1682 	}
1683 
1684 	if (bpf_map_is_offloaded(map)) {
1685 		err = bpf_map_offload_delete_elem(map, key);
1686 		goto out;
1687 	} else if (IS_FD_PROG_ARRAY(map) ||
1688 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1689 		/* These maps require sleepable context */
1690 		err = map->ops->map_delete_elem(map, key);
1691 		goto out;
1692 	}
1693 
1694 	bpf_disable_instrumentation();
1695 	rcu_read_lock();
1696 	err = map->ops->map_delete_elem(map, key);
1697 	rcu_read_unlock();
1698 	bpf_enable_instrumentation();
1699 	if (!err)
1700 		maybe_wait_bpf_programs(map);
1701 out:
1702 	kvfree(key);
1703 err_put:
1704 	bpf_map_write_active_dec(map);
1705 	fdput(f);
1706 	return err;
1707 }
1708 
1709 /* last field in 'union bpf_attr' used by this command */
1710 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1711 
1712 static int map_get_next_key(union bpf_attr *attr)
1713 {
1714 	void __user *ukey = u64_to_user_ptr(attr->key);
1715 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1716 	int ufd = attr->map_fd;
1717 	struct bpf_map *map;
1718 	void *key, *next_key;
1719 	struct fd f;
1720 	int err;
1721 
1722 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1723 		return -EINVAL;
1724 
1725 	f = fdget(ufd);
1726 	map = __bpf_map_get(f);
1727 	if (IS_ERR(map))
1728 		return PTR_ERR(map);
1729 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1730 		err = -EPERM;
1731 		goto err_put;
1732 	}
1733 
1734 	if (ukey) {
1735 		key = __bpf_copy_key(ukey, map->key_size);
1736 		if (IS_ERR(key)) {
1737 			err = PTR_ERR(key);
1738 			goto err_put;
1739 		}
1740 	} else {
1741 		key = NULL;
1742 	}
1743 
1744 	err = -ENOMEM;
1745 	next_key = kvmalloc(map->key_size, GFP_USER);
1746 	if (!next_key)
1747 		goto free_key;
1748 
1749 	if (bpf_map_is_offloaded(map)) {
1750 		err = bpf_map_offload_get_next_key(map, key, next_key);
1751 		goto out;
1752 	}
1753 
1754 	rcu_read_lock();
1755 	err = map->ops->map_get_next_key(map, key, next_key);
1756 	rcu_read_unlock();
1757 out:
1758 	if (err)
1759 		goto free_next_key;
1760 
1761 	err = -EFAULT;
1762 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1763 		goto free_next_key;
1764 
1765 	err = 0;
1766 
1767 free_next_key:
1768 	kvfree(next_key);
1769 free_key:
1770 	kvfree(key);
1771 err_put:
1772 	fdput(f);
1773 	return err;
1774 }
1775 
1776 int generic_map_delete_batch(struct bpf_map *map,
1777 			     const union bpf_attr *attr,
1778 			     union bpf_attr __user *uattr)
1779 {
1780 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1781 	u32 cp, max_count;
1782 	int err = 0;
1783 	void *key;
1784 
1785 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1786 		return -EINVAL;
1787 
1788 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1789 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1790 		return -EINVAL;
1791 	}
1792 
1793 	max_count = attr->batch.count;
1794 	if (!max_count)
1795 		return 0;
1796 
1797 	if (put_user(0, &uattr->batch.count))
1798 		return -EFAULT;
1799 
1800 	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1801 	if (!key)
1802 		return -ENOMEM;
1803 
1804 	for (cp = 0; cp < max_count; cp++) {
1805 		err = -EFAULT;
1806 		if (copy_from_user(key, keys + cp * map->key_size,
1807 				   map->key_size))
1808 			break;
1809 
1810 		if (bpf_map_is_offloaded(map)) {
1811 			err = bpf_map_offload_delete_elem(map, key);
1812 			break;
1813 		}
1814 
1815 		bpf_disable_instrumentation();
1816 		rcu_read_lock();
1817 		err = map->ops->map_delete_elem(map, key);
1818 		rcu_read_unlock();
1819 		bpf_enable_instrumentation();
1820 		if (err)
1821 			break;
1822 		cond_resched();
1823 	}
1824 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1825 		err = -EFAULT;
1826 
1827 	kvfree(key);
1828 
1829 	return err;
1830 }
1831 
1832 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1833 			     const union bpf_attr *attr,
1834 			     union bpf_attr __user *uattr)
1835 {
1836 	void __user *values = u64_to_user_ptr(attr->batch.values);
1837 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1838 	u32 value_size, cp, max_count;
1839 	void *key, *value;
1840 	int err = 0;
1841 
1842 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1843 		return -EINVAL;
1844 
1845 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1846 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1847 		return -EINVAL;
1848 	}
1849 
1850 	value_size = bpf_map_value_size(map);
1851 
1852 	max_count = attr->batch.count;
1853 	if (!max_count)
1854 		return 0;
1855 
1856 	if (put_user(0, &uattr->batch.count))
1857 		return -EFAULT;
1858 
1859 	key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1860 	if (!key)
1861 		return -ENOMEM;
1862 
1863 	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1864 	if (!value) {
1865 		kvfree(key);
1866 		return -ENOMEM;
1867 	}
1868 
1869 	for (cp = 0; cp < max_count; cp++) {
1870 		err = -EFAULT;
1871 		if (copy_from_user(key, keys + cp * map->key_size,
1872 		    map->key_size) ||
1873 		    copy_from_user(value, values + cp * value_size, value_size))
1874 			break;
1875 
1876 		err = bpf_map_update_value(map, map_file, key, value,
1877 					   attr->batch.elem_flags);
1878 
1879 		if (err)
1880 			break;
1881 		cond_resched();
1882 	}
1883 
1884 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1885 		err = -EFAULT;
1886 
1887 	kvfree(value);
1888 	kvfree(key);
1889 
1890 	return err;
1891 }
1892 
1893 #define MAP_LOOKUP_RETRIES 3
1894 
1895 int generic_map_lookup_batch(struct bpf_map *map,
1896 				    const union bpf_attr *attr,
1897 				    union bpf_attr __user *uattr)
1898 {
1899 	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1900 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1901 	void __user *values = u64_to_user_ptr(attr->batch.values);
1902 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1903 	void *buf, *buf_prevkey, *prev_key, *key, *value;
1904 	int err, retry = MAP_LOOKUP_RETRIES;
1905 	u32 value_size, cp, max_count;
1906 
1907 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1908 		return -EINVAL;
1909 
1910 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1911 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1912 		return -EINVAL;
1913 
1914 	value_size = bpf_map_value_size(map);
1915 
1916 	max_count = attr->batch.count;
1917 	if (!max_count)
1918 		return 0;
1919 
1920 	if (put_user(0, &uattr->batch.count))
1921 		return -EFAULT;
1922 
1923 	buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1924 	if (!buf_prevkey)
1925 		return -ENOMEM;
1926 
1927 	buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1928 	if (!buf) {
1929 		kvfree(buf_prevkey);
1930 		return -ENOMEM;
1931 	}
1932 
1933 	err = -EFAULT;
1934 	prev_key = NULL;
1935 	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1936 		goto free_buf;
1937 	key = buf;
1938 	value = key + map->key_size;
1939 	if (ubatch)
1940 		prev_key = buf_prevkey;
1941 
1942 	for (cp = 0; cp < max_count;) {
1943 		rcu_read_lock();
1944 		err = map->ops->map_get_next_key(map, prev_key, key);
1945 		rcu_read_unlock();
1946 		if (err)
1947 			break;
1948 		err = bpf_map_copy_value(map, key, value,
1949 					 attr->batch.elem_flags);
1950 
1951 		if (err == -ENOENT) {
1952 			if (retry) {
1953 				retry--;
1954 				continue;
1955 			}
1956 			err = -EINTR;
1957 			break;
1958 		}
1959 
1960 		if (err)
1961 			goto free_buf;
1962 
1963 		if (copy_to_user(keys + cp * map->key_size, key,
1964 				 map->key_size)) {
1965 			err = -EFAULT;
1966 			goto free_buf;
1967 		}
1968 		if (copy_to_user(values + cp * value_size, value, value_size)) {
1969 			err = -EFAULT;
1970 			goto free_buf;
1971 		}
1972 
1973 		if (!prev_key)
1974 			prev_key = buf_prevkey;
1975 
1976 		swap(prev_key, key);
1977 		retry = MAP_LOOKUP_RETRIES;
1978 		cp++;
1979 		cond_resched();
1980 	}
1981 
1982 	if (err == -EFAULT)
1983 		goto free_buf;
1984 
1985 	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1986 		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1987 		err = -EFAULT;
1988 
1989 free_buf:
1990 	kvfree(buf_prevkey);
1991 	kvfree(buf);
1992 	return err;
1993 }
1994 
1995 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1996 
1997 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1998 {
1999 	void __user *ukey = u64_to_user_ptr(attr->key);
2000 	void __user *uvalue = u64_to_user_ptr(attr->value);
2001 	int ufd = attr->map_fd;
2002 	struct bpf_map *map;
2003 	void *key, *value;
2004 	u32 value_size;
2005 	struct fd f;
2006 	int err;
2007 
2008 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
2009 		return -EINVAL;
2010 
2011 	if (attr->flags & ~BPF_F_LOCK)
2012 		return -EINVAL;
2013 
2014 	f = fdget(ufd);
2015 	map = __bpf_map_get(f);
2016 	if (IS_ERR(map))
2017 		return PTR_ERR(map);
2018 	bpf_map_write_active_inc(map);
2019 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
2020 	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2021 		err = -EPERM;
2022 		goto err_put;
2023 	}
2024 
2025 	if (attr->flags &&
2026 	    (map->map_type == BPF_MAP_TYPE_QUEUE ||
2027 	     map->map_type == BPF_MAP_TYPE_STACK)) {
2028 		err = -EINVAL;
2029 		goto err_put;
2030 	}
2031 
2032 	if ((attr->flags & BPF_F_LOCK) &&
2033 	    !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2034 		err = -EINVAL;
2035 		goto err_put;
2036 	}
2037 
2038 	key = __bpf_copy_key(ukey, map->key_size);
2039 	if (IS_ERR(key)) {
2040 		err = PTR_ERR(key);
2041 		goto err_put;
2042 	}
2043 
2044 	value_size = bpf_map_value_size(map);
2045 
2046 	err = -ENOMEM;
2047 	value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2048 	if (!value)
2049 		goto free_key;
2050 
2051 	err = -ENOTSUPP;
2052 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2053 	    map->map_type == BPF_MAP_TYPE_STACK) {
2054 		err = map->ops->map_pop_elem(map, value);
2055 	} else if (map->map_type == BPF_MAP_TYPE_HASH ||
2056 		   map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2057 		   map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2058 		   map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2059 		if (!bpf_map_is_offloaded(map)) {
2060 			bpf_disable_instrumentation();
2061 			rcu_read_lock();
2062 			err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2063 			rcu_read_unlock();
2064 			bpf_enable_instrumentation();
2065 		}
2066 	}
2067 
2068 	if (err)
2069 		goto free_value;
2070 
2071 	if (copy_to_user(uvalue, value, value_size) != 0) {
2072 		err = -EFAULT;
2073 		goto free_value;
2074 	}
2075 
2076 	err = 0;
2077 
2078 free_value:
2079 	kvfree(value);
2080 free_key:
2081 	kvfree(key);
2082 err_put:
2083 	bpf_map_write_active_dec(map);
2084 	fdput(f);
2085 	return err;
2086 }
2087 
2088 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
2089 
2090 static int map_freeze(const union bpf_attr *attr)
2091 {
2092 	int err = 0, ufd = attr->map_fd;
2093 	struct bpf_map *map;
2094 	struct fd f;
2095 
2096 	if (CHECK_ATTR(BPF_MAP_FREEZE))
2097 		return -EINVAL;
2098 
2099 	f = fdget(ufd);
2100 	map = __bpf_map_get(f);
2101 	if (IS_ERR(map))
2102 		return PTR_ERR(map);
2103 
2104 	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
2105 		fdput(f);
2106 		return -ENOTSUPP;
2107 	}
2108 
2109 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2110 		fdput(f);
2111 		return -EPERM;
2112 	}
2113 
2114 	mutex_lock(&map->freeze_mutex);
2115 	if (bpf_map_write_active(map)) {
2116 		err = -EBUSY;
2117 		goto err_put;
2118 	}
2119 	if (READ_ONCE(map->frozen)) {
2120 		err = -EBUSY;
2121 		goto err_put;
2122 	}
2123 
2124 	WRITE_ONCE(map->frozen, true);
2125 err_put:
2126 	mutex_unlock(&map->freeze_mutex);
2127 	fdput(f);
2128 	return err;
2129 }
2130 
2131 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2132 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2133 	[_id] = & _name ## _prog_ops,
2134 #define BPF_MAP_TYPE(_id, _ops)
2135 #define BPF_LINK_TYPE(_id, _name)
2136 #include <linux/bpf_types.h>
2137 #undef BPF_PROG_TYPE
2138 #undef BPF_MAP_TYPE
2139 #undef BPF_LINK_TYPE
2140 };
2141 
2142 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2143 {
2144 	const struct bpf_prog_ops *ops;
2145 
2146 	if (type >= ARRAY_SIZE(bpf_prog_types))
2147 		return -EINVAL;
2148 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2149 	ops = bpf_prog_types[type];
2150 	if (!ops)
2151 		return -EINVAL;
2152 
2153 	if (!bpf_prog_is_offloaded(prog->aux))
2154 		prog->aux->ops = ops;
2155 	else
2156 		prog->aux->ops = &bpf_offload_prog_ops;
2157 	prog->type = type;
2158 	return 0;
2159 }
2160 
2161 enum bpf_audit {
2162 	BPF_AUDIT_LOAD,
2163 	BPF_AUDIT_UNLOAD,
2164 	BPF_AUDIT_MAX,
2165 };
2166 
2167 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2168 	[BPF_AUDIT_LOAD]   = "LOAD",
2169 	[BPF_AUDIT_UNLOAD] = "UNLOAD",
2170 };
2171 
2172 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2173 {
2174 	struct audit_context *ctx = NULL;
2175 	struct audit_buffer *ab;
2176 
2177 	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2178 		return;
2179 	if (audit_enabled == AUDIT_OFF)
2180 		return;
2181 	if (!in_irq() && !irqs_disabled())
2182 		ctx = audit_context();
2183 	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2184 	if (unlikely(!ab))
2185 		return;
2186 	audit_log_format(ab, "prog-id=%u op=%s",
2187 			 prog->aux->id, bpf_audit_str[op]);
2188 	audit_log_end(ab);
2189 }
2190 
2191 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2192 {
2193 	int id;
2194 
2195 	idr_preload(GFP_KERNEL);
2196 	spin_lock_bh(&prog_idr_lock);
2197 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2198 	if (id > 0)
2199 		prog->aux->id = id;
2200 	spin_unlock_bh(&prog_idr_lock);
2201 	idr_preload_end();
2202 
2203 	/* id is in [1, INT_MAX) */
2204 	if (WARN_ON_ONCE(!id))
2205 		return -ENOSPC;
2206 
2207 	return id > 0 ? 0 : id;
2208 }
2209 
2210 void bpf_prog_free_id(struct bpf_prog *prog)
2211 {
2212 	unsigned long flags;
2213 
2214 	/* cBPF to eBPF migrations are currently not in the idr store.
2215 	 * Offloaded programs are removed from the store when their device
2216 	 * disappears - even if someone grabs an fd to them they are unusable,
2217 	 * simply waiting for refcnt to drop to be freed.
2218 	 */
2219 	if (!prog->aux->id)
2220 		return;
2221 
2222 	spin_lock_irqsave(&prog_idr_lock, flags);
2223 	idr_remove(&prog_idr, prog->aux->id);
2224 	prog->aux->id = 0;
2225 	spin_unlock_irqrestore(&prog_idr_lock, flags);
2226 }
2227 
2228 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2229 {
2230 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2231 
2232 	kvfree(aux->func_info);
2233 	kfree(aux->func_info_aux);
2234 	free_uid(aux->user);
2235 	security_bpf_prog_free(aux->prog);
2236 	bpf_prog_free(aux->prog);
2237 }
2238 
2239 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2240 {
2241 	bpf_prog_kallsyms_del_all(prog);
2242 	btf_put(prog->aux->btf);
2243 	module_put(prog->aux->mod);
2244 	kvfree(prog->aux->jited_linfo);
2245 	kvfree(prog->aux->linfo);
2246 	kfree(prog->aux->kfunc_tab);
2247 	if (prog->aux->attach_btf)
2248 		btf_put(prog->aux->attach_btf);
2249 
2250 	if (deferred) {
2251 		if (prog->sleepable)
2252 			call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2253 		else
2254 			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2255 	} else {
2256 		__bpf_prog_put_rcu(&prog->aux->rcu);
2257 	}
2258 }
2259 
2260 static void bpf_prog_put_deferred(struct work_struct *work)
2261 {
2262 	struct bpf_prog_aux *aux;
2263 	struct bpf_prog *prog;
2264 
2265 	aux = container_of(work, struct bpf_prog_aux, work);
2266 	prog = aux->prog;
2267 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2268 	bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2269 	bpf_prog_free_id(prog);
2270 	__bpf_prog_put_noref(prog, true);
2271 }
2272 
2273 static void __bpf_prog_put(struct bpf_prog *prog)
2274 {
2275 	struct bpf_prog_aux *aux = prog->aux;
2276 
2277 	if (atomic64_dec_and_test(&aux->refcnt)) {
2278 		if (in_irq() || irqs_disabled()) {
2279 			INIT_WORK(&aux->work, bpf_prog_put_deferred);
2280 			schedule_work(&aux->work);
2281 		} else {
2282 			bpf_prog_put_deferred(&aux->work);
2283 		}
2284 	}
2285 }
2286 
2287 void bpf_prog_put(struct bpf_prog *prog)
2288 {
2289 	__bpf_prog_put(prog);
2290 }
2291 EXPORT_SYMBOL_GPL(bpf_prog_put);
2292 
2293 static int bpf_prog_release(struct inode *inode, struct file *filp)
2294 {
2295 	struct bpf_prog *prog = filp->private_data;
2296 
2297 	bpf_prog_put(prog);
2298 	return 0;
2299 }
2300 
2301 struct bpf_prog_kstats {
2302 	u64 nsecs;
2303 	u64 cnt;
2304 	u64 misses;
2305 };
2306 
2307 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2308 {
2309 	struct bpf_prog_stats *stats;
2310 	unsigned int flags;
2311 
2312 	stats = this_cpu_ptr(prog->stats);
2313 	flags = u64_stats_update_begin_irqsave(&stats->syncp);
2314 	u64_stats_inc(&stats->misses);
2315 	u64_stats_update_end_irqrestore(&stats->syncp, flags);
2316 }
2317 
2318 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2319 			       struct bpf_prog_kstats *stats)
2320 {
2321 	u64 nsecs = 0, cnt = 0, misses = 0;
2322 	int cpu;
2323 
2324 	for_each_possible_cpu(cpu) {
2325 		const struct bpf_prog_stats *st;
2326 		unsigned int start;
2327 		u64 tnsecs, tcnt, tmisses;
2328 
2329 		st = per_cpu_ptr(prog->stats, cpu);
2330 		do {
2331 			start = u64_stats_fetch_begin(&st->syncp);
2332 			tnsecs = u64_stats_read(&st->nsecs);
2333 			tcnt = u64_stats_read(&st->cnt);
2334 			tmisses = u64_stats_read(&st->misses);
2335 		} while (u64_stats_fetch_retry(&st->syncp, start));
2336 		nsecs += tnsecs;
2337 		cnt += tcnt;
2338 		misses += tmisses;
2339 	}
2340 	stats->nsecs = nsecs;
2341 	stats->cnt = cnt;
2342 	stats->misses = misses;
2343 }
2344 
2345 #ifdef CONFIG_PROC_FS
2346 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2347 {
2348 	const struct bpf_prog *prog = filp->private_data;
2349 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2350 	struct bpf_prog_kstats stats;
2351 
2352 	bpf_prog_get_stats(prog, &stats);
2353 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2354 	seq_printf(m,
2355 		   "prog_type:\t%u\n"
2356 		   "prog_jited:\t%u\n"
2357 		   "prog_tag:\t%s\n"
2358 		   "memlock:\t%llu\n"
2359 		   "prog_id:\t%u\n"
2360 		   "run_time_ns:\t%llu\n"
2361 		   "run_cnt:\t%llu\n"
2362 		   "recursion_misses:\t%llu\n"
2363 		   "verified_insns:\t%u\n",
2364 		   prog->type,
2365 		   prog->jited,
2366 		   prog_tag,
2367 		   prog->pages * 1ULL << PAGE_SHIFT,
2368 		   prog->aux->id,
2369 		   stats.nsecs,
2370 		   stats.cnt,
2371 		   stats.misses,
2372 		   prog->aux->verified_insns);
2373 }
2374 #endif
2375 
2376 const struct file_operations bpf_prog_fops = {
2377 #ifdef CONFIG_PROC_FS
2378 	.show_fdinfo	= bpf_prog_show_fdinfo,
2379 #endif
2380 	.release	= bpf_prog_release,
2381 	.read		= bpf_dummy_read,
2382 	.write		= bpf_dummy_write,
2383 };
2384 
2385 int bpf_prog_new_fd(struct bpf_prog *prog)
2386 {
2387 	int ret;
2388 
2389 	ret = security_bpf_prog(prog);
2390 	if (ret < 0)
2391 		return ret;
2392 
2393 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2394 				O_RDWR | O_CLOEXEC);
2395 }
2396 
2397 static struct bpf_prog *____bpf_prog_get(struct fd f)
2398 {
2399 	if (!f.file)
2400 		return ERR_PTR(-EBADF);
2401 	if (f.file->f_op != &bpf_prog_fops) {
2402 		fdput(f);
2403 		return ERR_PTR(-EINVAL);
2404 	}
2405 
2406 	return f.file->private_data;
2407 }
2408 
2409 void bpf_prog_add(struct bpf_prog *prog, int i)
2410 {
2411 	atomic64_add(i, &prog->aux->refcnt);
2412 }
2413 EXPORT_SYMBOL_GPL(bpf_prog_add);
2414 
2415 void bpf_prog_sub(struct bpf_prog *prog, int i)
2416 {
2417 	/* Only to be used for undoing previous bpf_prog_add() in some
2418 	 * error path. We still know that another entity in our call
2419 	 * path holds a reference to the program, thus atomic_sub() can
2420 	 * be safely used in such cases!
2421 	 */
2422 	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2423 }
2424 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2425 
2426 void bpf_prog_inc(struct bpf_prog *prog)
2427 {
2428 	atomic64_inc(&prog->aux->refcnt);
2429 }
2430 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2431 
2432 /* prog_idr_lock should have been held */
2433 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2434 {
2435 	int refold;
2436 
2437 	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2438 
2439 	if (!refold)
2440 		return ERR_PTR(-ENOENT);
2441 
2442 	return prog;
2443 }
2444 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2445 
2446 bool bpf_prog_get_ok(struct bpf_prog *prog,
2447 			    enum bpf_prog_type *attach_type, bool attach_drv)
2448 {
2449 	/* not an attachment, just a refcount inc, always allow */
2450 	if (!attach_type)
2451 		return true;
2452 
2453 	if (prog->type != *attach_type)
2454 		return false;
2455 	if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2456 		return false;
2457 
2458 	return true;
2459 }
2460 
2461 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2462 				       bool attach_drv)
2463 {
2464 	struct fd f = fdget(ufd);
2465 	struct bpf_prog *prog;
2466 
2467 	prog = ____bpf_prog_get(f);
2468 	if (IS_ERR(prog))
2469 		return prog;
2470 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2471 		prog = ERR_PTR(-EINVAL);
2472 		goto out;
2473 	}
2474 
2475 	bpf_prog_inc(prog);
2476 out:
2477 	fdput(f);
2478 	return prog;
2479 }
2480 
2481 struct bpf_prog *bpf_prog_get(u32 ufd)
2482 {
2483 	return __bpf_prog_get(ufd, NULL, false);
2484 }
2485 
2486 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2487 				       bool attach_drv)
2488 {
2489 	return __bpf_prog_get(ufd, &type, attach_drv);
2490 }
2491 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2492 
2493 /* Initially all BPF programs could be loaded w/o specifying
2494  * expected_attach_type. Later for some of them specifying expected_attach_type
2495  * at load time became required so that program could be validated properly.
2496  * Programs of types that are allowed to be loaded both w/ and w/o (for
2497  * backward compatibility) expected_attach_type, should have the default attach
2498  * type assigned to expected_attach_type for the latter case, so that it can be
2499  * validated later at attach time.
2500  *
2501  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2502  * prog type requires it but has some attach types that have to be backward
2503  * compatible.
2504  */
2505 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2506 {
2507 	switch (attr->prog_type) {
2508 	case BPF_PROG_TYPE_CGROUP_SOCK:
2509 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2510 		 * exist so checking for non-zero is the way to go here.
2511 		 */
2512 		if (!attr->expected_attach_type)
2513 			attr->expected_attach_type =
2514 				BPF_CGROUP_INET_SOCK_CREATE;
2515 		break;
2516 	case BPF_PROG_TYPE_SK_REUSEPORT:
2517 		if (!attr->expected_attach_type)
2518 			attr->expected_attach_type =
2519 				BPF_SK_REUSEPORT_SELECT;
2520 		break;
2521 	}
2522 }
2523 
2524 static int
2525 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2526 			   enum bpf_attach_type expected_attach_type,
2527 			   struct btf *attach_btf, u32 btf_id,
2528 			   struct bpf_prog *dst_prog)
2529 {
2530 	if (btf_id) {
2531 		if (btf_id > BTF_MAX_TYPE)
2532 			return -EINVAL;
2533 
2534 		if (!attach_btf && !dst_prog)
2535 			return -EINVAL;
2536 
2537 		switch (prog_type) {
2538 		case BPF_PROG_TYPE_TRACING:
2539 		case BPF_PROG_TYPE_LSM:
2540 		case BPF_PROG_TYPE_STRUCT_OPS:
2541 		case BPF_PROG_TYPE_EXT:
2542 			break;
2543 		default:
2544 			return -EINVAL;
2545 		}
2546 	}
2547 
2548 	if (attach_btf && (!btf_id || dst_prog))
2549 		return -EINVAL;
2550 
2551 	if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2552 	    prog_type != BPF_PROG_TYPE_EXT)
2553 		return -EINVAL;
2554 
2555 	switch (prog_type) {
2556 	case BPF_PROG_TYPE_CGROUP_SOCK:
2557 		switch (expected_attach_type) {
2558 		case BPF_CGROUP_INET_SOCK_CREATE:
2559 		case BPF_CGROUP_INET_SOCK_RELEASE:
2560 		case BPF_CGROUP_INET4_POST_BIND:
2561 		case BPF_CGROUP_INET6_POST_BIND:
2562 			return 0;
2563 		default:
2564 			return -EINVAL;
2565 		}
2566 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2567 		switch (expected_attach_type) {
2568 		case BPF_CGROUP_INET4_BIND:
2569 		case BPF_CGROUP_INET6_BIND:
2570 		case BPF_CGROUP_INET4_CONNECT:
2571 		case BPF_CGROUP_INET6_CONNECT:
2572 		case BPF_CGROUP_UNIX_CONNECT:
2573 		case BPF_CGROUP_INET4_GETPEERNAME:
2574 		case BPF_CGROUP_INET6_GETPEERNAME:
2575 		case BPF_CGROUP_UNIX_GETPEERNAME:
2576 		case BPF_CGROUP_INET4_GETSOCKNAME:
2577 		case BPF_CGROUP_INET6_GETSOCKNAME:
2578 		case BPF_CGROUP_UNIX_GETSOCKNAME:
2579 		case BPF_CGROUP_UDP4_SENDMSG:
2580 		case BPF_CGROUP_UDP6_SENDMSG:
2581 		case BPF_CGROUP_UNIX_SENDMSG:
2582 		case BPF_CGROUP_UDP4_RECVMSG:
2583 		case BPF_CGROUP_UDP6_RECVMSG:
2584 		case BPF_CGROUP_UNIX_RECVMSG:
2585 			return 0;
2586 		default:
2587 			return -EINVAL;
2588 		}
2589 	case BPF_PROG_TYPE_CGROUP_SKB:
2590 		switch (expected_attach_type) {
2591 		case BPF_CGROUP_INET_INGRESS:
2592 		case BPF_CGROUP_INET_EGRESS:
2593 			return 0;
2594 		default:
2595 			return -EINVAL;
2596 		}
2597 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2598 		switch (expected_attach_type) {
2599 		case BPF_CGROUP_SETSOCKOPT:
2600 		case BPF_CGROUP_GETSOCKOPT:
2601 			return 0;
2602 		default:
2603 			return -EINVAL;
2604 		}
2605 	case BPF_PROG_TYPE_SK_LOOKUP:
2606 		if (expected_attach_type == BPF_SK_LOOKUP)
2607 			return 0;
2608 		return -EINVAL;
2609 	case BPF_PROG_TYPE_SK_REUSEPORT:
2610 		switch (expected_attach_type) {
2611 		case BPF_SK_REUSEPORT_SELECT:
2612 		case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2613 			return 0;
2614 		default:
2615 			return -EINVAL;
2616 		}
2617 	case BPF_PROG_TYPE_NETFILTER:
2618 		if (expected_attach_type == BPF_NETFILTER)
2619 			return 0;
2620 		return -EINVAL;
2621 	case BPF_PROG_TYPE_SYSCALL:
2622 	case BPF_PROG_TYPE_EXT:
2623 		if (expected_attach_type)
2624 			return -EINVAL;
2625 		fallthrough;
2626 	default:
2627 		return 0;
2628 	}
2629 }
2630 
2631 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2632 {
2633 	switch (prog_type) {
2634 	case BPF_PROG_TYPE_SCHED_CLS:
2635 	case BPF_PROG_TYPE_SCHED_ACT:
2636 	case BPF_PROG_TYPE_XDP:
2637 	case BPF_PROG_TYPE_LWT_IN:
2638 	case BPF_PROG_TYPE_LWT_OUT:
2639 	case BPF_PROG_TYPE_LWT_XMIT:
2640 	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2641 	case BPF_PROG_TYPE_SK_SKB:
2642 	case BPF_PROG_TYPE_SK_MSG:
2643 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2644 	case BPF_PROG_TYPE_CGROUP_DEVICE:
2645 	case BPF_PROG_TYPE_CGROUP_SOCK:
2646 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2647 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2648 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
2649 	case BPF_PROG_TYPE_SOCK_OPS:
2650 	case BPF_PROG_TYPE_EXT: /* extends any prog */
2651 	case BPF_PROG_TYPE_NETFILTER:
2652 		return true;
2653 	case BPF_PROG_TYPE_CGROUP_SKB:
2654 		/* always unpriv */
2655 	case BPF_PROG_TYPE_SK_REUSEPORT:
2656 		/* equivalent to SOCKET_FILTER. need CAP_BPF only */
2657 	default:
2658 		return false;
2659 	}
2660 }
2661 
2662 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2663 {
2664 	switch (prog_type) {
2665 	case BPF_PROG_TYPE_KPROBE:
2666 	case BPF_PROG_TYPE_TRACEPOINT:
2667 	case BPF_PROG_TYPE_PERF_EVENT:
2668 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2669 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2670 	case BPF_PROG_TYPE_TRACING:
2671 	case BPF_PROG_TYPE_LSM:
2672 	case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2673 	case BPF_PROG_TYPE_EXT: /* extends any prog */
2674 		return true;
2675 	default:
2676 		return false;
2677 	}
2678 }
2679 
2680 /* last field in 'union bpf_attr' used by this command */
2681 #define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2682 
2683 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2684 {
2685 	enum bpf_prog_type type = attr->prog_type;
2686 	struct bpf_prog *prog, *dst_prog = NULL;
2687 	struct btf *attach_btf = NULL;
2688 	struct bpf_token *token = NULL;
2689 	bool bpf_cap;
2690 	int err;
2691 	char license[128];
2692 
2693 	if (CHECK_ATTR(BPF_PROG_LOAD))
2694 		return -EINVAL;
2695 
2696 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2697 				 BPF_F_ANY_ALIGNMENT |
2698 				 BPF_F_TEST_STATE_FREQ |
2699 				 BPF_F_SLEEPABLE |
2700 				 BPF_F_TEST_RND_HI32 |
2701 				 BPF_F_XDP_HAS_FRAGS |
2702 				 BPF_F_XDP_DEV_BOUND_ONLY |
2703 				 BPF_F_TEST_REG_INVARIANTS |
2704 				 BPF_F_TOKEN_FD))
2705 		return -EINVAL;
2706 
2707 	bpf_prog_load_fixup_attach_type(attr);
2708 
2709 	if (attr->prog_flags & BPF_F_TOKEN_FD) {
2710 		token = bpf_token_get_from_fd(attr->prog_token_fd);
2711 		if (IS_ERR(token))
2712 			return PTR_ERR(token);
2713 		/* if current token doesn't grant prog loading permissions,
2714 		 * then we can't use this token, so ignore it and rely on
2715 		 * system-wide capabilities checks
2716 		 */
2717 		if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2718 		    !bpf_token_allow_prog_type(token, attr->prog_type,
2719 					       attr->expected_attach_type)) {
2720 			bpf_token_put(token);
2721 			token = NULL;
2722 		}
2723 	}
2724 
2725 	bpf_cap = bpf_token_capable(token, CAP_BPF);
2726 	err = -EPERM;
2727 
2728 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2729 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2730 	    !bpf_cap)
2731 		goto put_token;
2732 
2733 	/* Intent here is for unprivileged_bpf_disabled to block BPF program
2734 	 * creation for unprivileged users; other actions depend
2735 	 * on fd availability and access to bpffs, so are dependent on
2736 	 * object creation success. Even with unprivileged BPF disabled,
2737 	 * capability checks are still carried out for these
2738 	 * and other operations.
2739 	 */
2740 	if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2741 		goto put_token;
2742 
2743 	if (attr->insn_cnt == 0 ||
2744 	    attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2745 		err = -E2BIG;
2746 		goto put_token;
2747 	}
2748 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2749 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2750 	    !bpf_cap)
2751 		goto put_token;
2752 
2753 	if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2754 		goto put_token;
2755 	if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2756 		goto put_token;
2757 
2758 	/* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2759 	 * or btf, we need to check which one it is
2760 	 */
2761 	if (attr->attach_prog_fd) {
2762 		dst_prog = bpf_prog_get(attr->attach_prog_fd);
2763 		if (IS_ERR(dst_prog)) {
2764 			dst_prog = NULL;
2765 			attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2766 			if (IS_ERR(attach_btf)) {
2767 				err = -EINVAL;
2768 				goto put_token;
2769 			}
2770 			if (!btf_is_kernel(attach_btf)) {
2771 				/* attaching through specifying bpf_prog's BTF
2772 				 * objects directly might be supported eventually
2773 				 */
2774 				btf_put(attach_btf);
2775 				err = -ENOTSUPP;
2776 				goto put_token;
2777 			}
2778 		}
2779 	} else if (attr->attach_btf_id) {
2780 		/* fall back to vmlinux BTF, if BTF type ID is specified */
2781 		attach_btf = bpf_get_btf_vmlinux();
2782 		if (IS_ERR(attach_btf)) {
2783 			err = PTR_ERR(attach_btf);
2784 			goto put_token;
2785 		}
2786 		if (!attach_btf) {
2787 			err = -EINVAL;
2788 			goto put_token;
2789 		}
2790 		btf_get(attach_btf);
2791 	}
2792 
2793 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2794 				       attach_btf, attr->attach_btf_id,
2795 				       dst_prog)) {
2796 		if (dst_prog)
2797 			bpf_prog_put(dst_prog);
2798 		if (attach_btf)
2799 			btf_put(attach_btf);
2800 		err = -EINVAL;
2801 		goto put_token;
2802 	}
2803 
2804 	/* plain bpf_prog allocation */
2805 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2806 	if (!prog) {
2807 		if (dst_prog)
2808 			bpf_prog_put(dst_prog);
2809 		if (attach_btf)
2810 			btf_put(attach_btf);
2811 		err = -EINVAL;
2812 		goto put_token;
2813 	}
2814 
2815 	prog->expected_attach_type = attr->expected_attach_type;
2816 	prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2817 	prog->aux->attach_btf = attach_btf;
2818 	prog->aux->attach_btf_id = attr->attach_btf_id;
2819 	prog->aux->dst_prog = dst_prog;
2820 	prog->aux->dev_bound = !!attr->prog_ifindex;
2821 	prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2822 
2823 	/* move token into prog->aux, reuse taken refcnt */
2824 	prog->aux->token = token;
2825 	token = NULL;
2826 
2827 	prog->aux->user = get_current_user();
2828 	prog->len = attr->insn_cnt;
2829 
2830 	err = -EFAULT;
2831 	if (copy_from_bpfptr(prog->insns,
2832 			     make_bpfptr(attr->insns, uattr.is_kernel),
2833 			     bpf_prog_insn_size(prog)) != 0)
2834 		goto free_prog;
2835 	/* copy eBPF program license from user space */
2836 	if (strncpy_from_bpfptr(license,
2837 				make_bpfptr(attr->license, uattr.is_kernel),
2838 				sizeof(license) - 1) < 0)
2839 		goto free_prog;
2840 	license[sizeof(license) - 1] = 0;
2841 
2842 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2843 	prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2844 
2845 	prog->orig_prog = NULL;
2846 	prog->jited = 0;
2847 
2848 	atomic64_set(&prog->aux->refcnt, 1);
2849 
2850 	if (bpf_prog_is_dev_bound(prog->aux)) {
2851 		err = bpf_prog_dev_bound_init(prog, attr);
2852 		if (err)
2853 			goto free_prog;
2854 	}
2855 
2856 	if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2857 	    bpf_prog_is_dev_bound(dst_prog->aux)) {
2858 		err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2859 		if (err)
2860 			goto free_prog;
2861 	}
2862 
2863 	/*
2864 	 * Bookkeeping for managing the program attachment chain.
2865 	 *
2866 	 * It might be tempting to set attach_tracing_prog flag at the attachment
2867 	 * time, but this will not prevent from loading bunch of tracing prog
2868 	 * first, then attach them one to another.
2869 	 *
2870 	 * The flag attach_tracing_prog is set for the whole program lifecycle, and
2871 	 * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2872 	 * programs cannot change attachment target.
2873 	 */
2874 	if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2875 	    dst_prog->type == BPF_PROG_TYPE_TRACING) {
2876 		prog->aux->attach_tracing_prog = true;
2877 	}
2878 
2879 	/* find program type: socket_filter vs tracing_filter */
2880 	err = find_prog_type(type, prog);
2881 	if (err < 0)
2882 		goto free_prog;
2883 
2884 	prog->aux->load_time = ktime_get_boottime_ns();
2885 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2886 			       sizeof(attr->prog_name));
2887 	if (err < 0)
2888 		goto free_prog;
2889 
2890 	err = security_bpf_prog_load(prog, attr, token);
2891 	if (err)
2892 		goto free_prog_sec;
2893 
2894 	/* run eBPF verifier */
2895 	err = bpf_check(&prog, attr, uattr, uattr_size);
2896 	if (err < 0)
2897 		goto free_used_maps;
2898 
2899 	prog = bpf_prog_select_runtime(prog, &err);
2900 	if (err < 0)
2901 		goto free_used_maps;
2902 
2903 	err = bpf_prog_alloc_id(prog);
2904 	if (err)
2905 		goto free_used_maps;
2906 
2907 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2908 	 * effectively publicly exposed. However, retrieving via
2909 	 * bpf_prog_get_fd_by_id() will take another reference,
2910 	 * therefore it cannot be gone underneath us.
2911 	 *
2912 	 * Only for the time /after/ successful bpf_prog_new_fd()
2913 	 * and before returning to userspace, we might just hold
2914 	 * one reference and any parallel close on that fd could
2915 	 * rip everything out. Hence, below notifications must
2916 	 * happen before bpf_prog_new_fd().
2917 	 *
2918 	 * Also, any failure handling from this point onwards must
2919 	 * be using bpf_prog_put() given the program is exposed.
2920 	 */
2921 	bpf_prog_kallsyms_add(prog);
2922 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2923 	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2924 
2925 	err = bpf_prog_new_fd(prog);
2926 	if (err < 0)
2927 		bpf_prog_put(prog);
2928 	return err;
2929 
2930 free_used_maps:
2931 	/* In case we have subprogs, we need to wait for a grace
2932 	 * period before we can tear down JIT memory since symbols
2933 	 * are already exposed under kallsyms.
2934 	 */
2935 	__bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2936 	return err;
2937 
2938 free_prog_sec:
2939 	security_bpf_prog_free(prog);
2940 free_prog:
2941 	free_uid(prog->aux->user);
2942 	if (prog->aux->attach_btf)
2943 		btf_put(prog->aux->attach_btf);
2944 	bpf_prog_free(prog);
2945 put_token:
2946 	bpf_token_put(token);
2947 	return err;
2948 }
2949 
2950 #define BPF_OBJ_LAST_FIELD path_fd
2951 
2952 static int bpf_obj_pin(const union bpf_attr *attr)
2953 {
2954 	int path_fd;
2955 
2956 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2957 		return -EINVAL;
2958 
2959 	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2960 	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2961 		return -EINVAL;
2962 
2963 	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2964 	return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2965 				u64_to_user_ptr(attr->pathname));
2966 }
2967 
2968 static int bpf_obj_get(const union bpf_attr *attr)
2969 {
2970 	int path_fd;
2971 
2972 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2973 	    attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2974 		return -EINVAL;
2975 
2976 	/* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2977 	if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2978 		return -EINVAL;
2979 
2980 	path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2981 	return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2982 				attr->file_flags);
2983 }
2984 
2985 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2986 		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
2987 {
2988 	atomic64_set(&link->refcnt, 1);
2989 	link->type = type;
2990 	link->id = 0;
2991 	link->ops = ops;
2992 	link->prog = prog;
2993 }
2994 
2995 static void bpf_link_free_id(int id)
2996 {
2997 	if (!id)
2998 		return;
2999 
3000 	spin_lock_bh(&link_idr_lock);
3001 	idr_remove(&link_idr, id);
3002 	spin_unlock_bh(&link_idr_lock);
3003 }
3004 
3005 /* Clean up bpf_link and corresponding anon_inode file and FD. After
3006  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
3007  * anon_inode's release() call. This helper marks bpf_link as
3008  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
3009  * is not decremented, it's the responsibility of a calling code that failed
3010  * to complete bpf_link initialization.
3011  * This helper eventually calls link's dealloc callback, but does not call
3012  * link's release callback.
3013  */
3014 void bpf_link_cleanup(struct bpf_link_primer *primer)
3015 {
3016 	primer->link->prog = NULL;
3017 	bpf_link_free_id(primer->id);
3018 	fput(primer->file);
3019 	put_unused_fd(primer->fd);
3020 }
3021 
3022 void bpf_link_inc(struct bpf_link *link)
3023 {
3024 	atomic64_inc(&link->refcnt);
3025 }
3026 
3027 static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
3028 {
3029 	struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
3030 
3031 	/* free bpf_link and its containing memory */
3032 	link->ops->dealloc_deferred(link);
3033 }
3034 
3035 static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
3036 {
3037 	if (rcu_trace_implies_rcu_gp())
3038 		bpf_link_defer_dealloc_rcu_gp(rcu);
3039 	else
3040 		call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
3041 }
3042 
3043 /* bpf_link_free is guaranteed to be called from process context */
3044 static void bpf_link_free(struct bpf_link *link)
3045 {
3046 	bool sleepable = false;
3047 
3048 	bpf_link_free_id(link->id);
3049 	if (link->prog) {
3050 		sleepable = link->prog->sleepable;
3051 		/* detach BPF program, clean up used resources */
3052 		link->ops->release(link);
3053 		bpf_prog_put(link->prog);
3054 	}
3055 	if (link->ops->dealloc_deferred) {
3056 		/* schedule BPF link deallocation; if underlying BPF program
3057 		 * is sleepable, we need to first wait for RCU tasks trace
3058 		 * sync, then go through "classic" RCU grace period
3059 		 */
3060 		if (sleepable)
3061 			call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3062 		else
3063 			call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3064 	}
3065 	if (link->ops->dealloc)
3066 		link->ops->dealloc(link);
3067 }
3068 
3069 static void bpf_link_put_deferred(struct work_struct *work)
3070 {
3071 	struct bpf_link *link = container_of(work, struct bpf_link, work);
3072 
3073 	bpf_link_free(link);
3074 }
3075 
3076 /* bpf_link_put might be called from atomic context. It needs to be called
3077  * from sleepable context in order to acquire sleeping locks during the process.
3078  */
3079 void bpf_link_put(struct bpf_link *link)
3080 {
3081 	if (!atomic64_dec_and_test(&link->refcnt))
3082 		return;
3083 
3084 	INIT_WORK(&link->work, bpf_link_put_deferred);
3085 	schedule_work(&link->work);
3086 }
3087 EXPORT_SYMBOL(bpf_link_put);
3088 
3089 static void bpf_link_put_direct(struct bpf_link *link)
3090 {
3091 	if (!atomic64_dec_and_test(&link->refcnt))
3092 		return;
3093 	bpf_link_free(link);
3094 }
3095 
3096 static int bpf_link_release(struct inode *inode, struct file *filp)
3097 {
3098 	struct bpf_link *link = filp->private_data;
3099 
3100 	bpf_link_put_direct(link);
3101 	return 0;
3102 }
3103 
3104 #ifdef CONFIG_PROC_FS
3105 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3106 #define BPF_MAP_TYPE(_id, _ops)
3107 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3108 static const char *bpf_link_type_strs[] = {
3109 	[BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3110 #include <linux/bpf_types.h>
3111 };
3112 #undef BPF_PROG_TYPE
3113 #undef BPF_MAP_TYPE
3114 #undef BPF_LINK_TYPE
3115 
3116 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3117 {
3118 	const struct bpf_link *link = filp->private_data;
3119 	const struct bpf_prog *prog = link->prog;
3120 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3121 
3122 	seq_printf(m,
3123 		   "link_type:\t%s\n"
3124 		   "link_id:\t%u\n",
3125 		   bpf_link_type_strs[link->type],
3126 		   link->id);
3127 	if (prog) {
3128 		bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3129 		seq_printf(m,
3130 			   "prog_tag:\t%s\n"
3131 			   "prog_id:\t%u\n",
3132 			   prog_tag,
3133 			   prog->aux->id);
3134 	}
3135 	if (link->ops->show_fdinfo)
3136 		link->ops->show_fdinfo(link, m);
3137 }
3138 #endif
3139 
3140 static const struct file_operations bpf_link_fops = {
3141 #ifdef CONFIG_PROC_FS
3142 	.show_fdinfo	= bpf_link_show_fdinfo,
3143 #endif
3144 	.release	= bpf_link_release,
3145 	.read		= bpf_dummy_read,
3146 	.write		= bpf_dummy_write,
3147 };
3148 
3149 static int bpf_link_alloc_id(struct bpf_link *link)
3150 {
3151 	int id;
3152 
3153 	idr_preload(GFP_KERNEL);
3154 	spin_lock_bh(&link_idr_lock);
3155 	id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3156 	spin_unlock_bh(&link_idr_lock);
3157 	idr_preload_end();
3158 
3159 	return id;
3160 }
3161 
3162 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3163  * reserving unused FD and allocating ID from link_idr. This is to be paired
3164  * with bpf_link_settle() to install FD and ID and expose bpf_link to
3165  * user-space, if bpf_link is successfully attached. If not, bpf_link and
3166  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3167  * transient state is passed around in struct bpf_link_primer.
3168  * This is preferred way to create and initialize bpf_link, especially when
3169  * there are complicated and expensive operations in between creating bpf_link
3170  * itself and attaching it to BPF hook. By using bpf_link_prime() and
3171  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3172  * expensive (and potentially failing) roll back operations in a rare case
3173  * that file, FD, or ID can't be allocated.
3174  */
3175 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3176 {
3177 	struct file *file;
3178 	int fd, id;
3179 
3180 	fd = get_unused_fd_flags(O_CLOEXEC);
3181 	if (fd < 0)
3182 		return fd;
3183 
3184 
3185 	id = bpf_link_alloc_id(link);
3186 	if (id < 0) {
3187 		put_unused_fd(fd);
3188 		return id;
3189 	}
3190 
3191 	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
3192 	if (IS_ERR(file)) {
3193 		bpf_link_free_id(id);
3194 		put_unused_fd(fd);
3195 		return PTR_ERR(file);
3196 	}
3197 
3198 	primer->link = link;
3199 	primer->file = file;
3200 	primer->fd = fd;
3201 	primer->id = id;
3202 	return 0;
3203 }
3204 
3205 int bpf_link_settle(struct bpf_link_primer *primer)
3206 {
3207 	/* make bpf_link fetchable by ID */
3208 	spin_lock_bh(&link_idr_lock);
3209 	primer->link->id = primer->id;
3210 	spin_unlock_bh(&link_idr_lock);
3211 	/* make bpf_link fetchable by FD */
3212 	fd_install(primer->fd, primer->file);
3213 	/* pass through installed FD */
3214 	return primer->fd;
3215 }
3216 
3217 int bpf_link_new_fd(struct bpf_link *link)
3218 {
3219 	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3220 }
3221 
3222 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3223 {
3224 	struct fd f = fdget(ufd);
3225 	struct bpf_link *link;
3226 
3227 	if (!f.file)
3228 		return ERR_PTR(-EBADF);
3229 	if (f.file->f_op != &bpf_link_fops) {
3230 		fdput(f);
3231 		return ERR_PTR(-EINVAL);
3232 	}
3233 
3234 	link = f.file->private_data;
3235 	bpf_link_inc(link);
3236 	fdput(f);
3237 
3238 	return link;
3239 }
3240 EXPORT_SYMBOL(bpf_link_get_from_fd);
3241 
3242 static void bpf_tracing_link_release(struct bpf_link *link)
3243 {
3244 	struct bpf_tracing_link *tr_link =
3245 		container_of(link, struct bpf_tracing_link, link.link);
3246 
3247 	WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3248 						tr_link->trampoline));
3249 
3250 	bpf_trampoline_put(tr_link->trampoline);
3251 
3252 	/* tgt_prog is NULL if target is a kernel function */
3253 	if (tr_link->tgt_prog)
3254 		bpf_prog_put(tr_link->tgt_prog);
3255 }
3256 
3257 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3258 {
3259 	struct bpf_tracing_link *tr_link =
3260 		container_of(link, struct bpf_tracing_link, link.link);
3261 
3262 	kfree(tr_link);
3263 }
3264 
3265 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3266 					 struct seq_file *seq)
3267 {
3268 	struct bpf_tracing_link *tr_link =
3269 		container_of(link, struct bpf_tracing_link, link.link);
3270 	u32 target_btf_id, target_obj_id;
3271 
3272 	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3273 				  &target_obj_id, &target_btf_id);
3274 	seq_printf(seq,
3275 		   "attach_type:\t%d\n"
3276 		   "target_obj_id:\t%u\n"
3277 		   "target_btf_id:\t%u\n",
3278 		   tr_link->attach_type,
3279 		   target_obj_id,
3280 		   target_btf_id);
3281 }
3282 
3283 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3284 					   struct bpf_link_info *info)
3285 {
3286 	struct bpf_tracing_link *tr_link =
3287 		container_of(link, struct bpf_tracing_link, link.link);
3288 
3289 	info->tracing.attach_type = tr_link->attach_type;
3290 	bpf_trampoline_unpack_key(tr_link->trampoline->key,
3291 				  &info->tracing.target_obj_id,
3292 				  &info->tracing.target_btf_id);
3293 
3294 	return 0;
3295 }
3296 
3297 static const struct bpf_link_ops bpf_tracing_link_lops = {
3298 	.release = bpf_tracing_link_release,
3299 	.dealloc = bpf_tracing_link_dealloc,
3300 	.show_fdinfo = bpf_tracing_link_show_fdinfo,
3301 	.fill_link_info = bpf_tracing_link_fill_link_info,
3302 };
3303 
3304 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3305 				   int tgt_prog_fd,
3306 				   u32 btf_id,
3307 				   u64 bpf_cookie)
3308 {
3309 	struct bpf_link_primer link_primer;
3310 	struct bpf_prog *tgt_prog = NULL;
3311 	struct bpf_trampoline *tr = NULL;
3312 	struct bpf_tracing_link *link;
3313 	u64 key = 0;
3314 	int err;
3315 
3316 	switch (prog->type) {
3317 	case BPF_PROG_TYPE_TRACING:
3318 		if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3319 		    prog->expected_attach_type != BPF_TRACE_FEXIT &&
3320 		    prog->expected_attach_type != BPF_MODIFY_RETURN) {
3321 			err = -EINVAL;
3322 			goto out_put_prog;
3323 		}
3324 		break;
3325 	case BPF_PROG_TYPE_EXT:
3326 		if (prog->expected_attach_type != 0) {
3327 			err = -EINVAL;
3328 			goto out_put_prog;
3329 		}
3330 		break;
3331 	case BPF_PROG_TYPE_LSM:
3332 		if (prog->expected_attach_type != BPF_LSM_MAC) {
3333 			err = -EINVAL;
3334 			goto out_put_prog;
3335 		}
3336 		break;
3337 	default:
3338 		err = -EINVAL;
3339 		goto out_put_prog;
3340 	}
3341 
3342 	if (!!tgt_prog_fd != !!btf_id) {
3343 		err = -EINVAL;
3344 		goto out_put_prog;
3345 	}
3346 
3347 	if (tgt_prog_fd) {
3348 		/*
3349 		 * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3350 		 * part would be changed to implement the same for
3351 		 * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3352 		 * attach_tracing_prog flag is set.
3353 		 */
3354 		if (prog->type != BPF_PROG_TYPE_EXT) {
3355 			err = -EINVAL;
3356 			goto out_put_prog;
3357 		}
3358 
3359 		tgt_prog = bpf_prog_get(tgt_prog_fd);
3360 		if (IS_ERR(tgt_prog)) {
3361 			err = PTR_ERR(tgt_prog);
3362 			tgt_prog = NULL;
3363 			goto out_put_prog;
3364 		}
3365 
3366 		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3367 	}
3368 
3369 	link = kzalloc(sizeof(*link), GFP_USER);
3370 	if (!link) {
3371 		err = -ENOMEM;
3372 		goto out_put_prog;
3373 	}
3374 	bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3375 		      &bpf_tracing_link_lops, prog);
3376 	link->attach_type = prog->expected_attach_type;
3377 	link->link.cookie = bpf_cookie;
3378 
3379 	mutex_lock(&prog->aux->dst_mutex);
3380 
3381 	/* There are a few possible cases here:
3382 	 *
3383 	 * - if prog->aux->dst_trampoline is set, the program was just loaded
3384 	 *   and not yet attached to anything, so we can use the values stored
3385 	 *   in prog->aux
3386 	 *
3387 	 * - if prog->aux->dst_trampoline is NULL, the program has already been
3388          *   attached to a target and its initial target was cleared (below)
3389 	 *
3390 	 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3391 	 *   target_btf_id using the link_create API.
3392 	 *
3393 	 * - if tgt_prog == NULL when this function was called using the old
3394 	 *   raw_tracepoint_open API, and we need a target from prog->aux
3395 	 *
3396 	 * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3397 	 *   was detached and is going for re-attachment.
3398 	 *
3399 	 * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3400 	 *   are NULL, then program was already attached and user did not provide
3401 	 *   tgt_prog_fd so we have no way to find out or create trampoline
3402 	 */
3403 	if (!prog->aux->dst_trampoline && !tgt_prog) {
3404 		/*
3405 		 * Allow re-attach for TRACING and LSM programs. If it's
3406 		 * currently linked, bpf_trampoline_link_prog will fail.
3407 		 * EXT programs need to specify tgt_prog_fd, so they
3408 		 * re-attach in separate code path.
3409 		 */
3410 		if (prog->type != BPF_PROG_TYPE_TRACING &&
3411 		    prog->type != BPF_PROG_TYPE_LSM) {
3412 			err = -EINVAL;
3413 			goto out_unlock;
3414 		}
3415 		/* We can allow re-attach only if we have valid attach_btf. */
3416 		if (!prog->aux->attach_btf) {
3417 			err = -EINVAL;
3418 			goto out_unlock;
3419 		}
3420 		btf_id = prog->aux->attach_btf_id;
3421 		key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3422 	}
3423 
3424 	if (!prog->aux->dst_trampoline ||
3425 	    (key && key != prog->aux->dst_trampoline->key)) {
3426 		/* If there is no saved target, or the specified target is
3427 		 * different from the destination specified at load time, we
3428 		 * need a new trampoline and a check for compatibility
3429 		 */
3430 		struct bpf_attach_target_info tgt_info = {};
3431 
3432 		err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3433 					      &tgt_info);
3434 		if (err)
3435 			goto out_unlock;
3436 
3437 		if (tgt_info.tgt_mod) {
3438 			module_put(prog->aux->mod);
3439 			prog->aux->mod = tgt_info.tgt_mod;
3440 		}
3441 
3442 		tr = bpf_trampoline_get(key, &tgt_info);
3443 		if (!tr) {
3444 			err = -ENOMEM;
3445 			goto out_unlock;
3446 		}
3447 	} else {
3448 		/* The caller didn't specify a target, or the target was the
3449 		 * same as the destination supplied during program load. This
3450 		 * means we can reuse the trampoline and reference from program
3451 		 * load time, and there is no need to allocate a new one. This
3452 		 * can only happen once for any program, as the saved values in
3453 		 * prog->aux are cleared below.
3454 		 */
3455 		tr = prog->aux->dst_trampoline;
3456 		tgt_prog = prog->aux->dst_prog;
3457 	}
3458 
3459 	err = bpf_link_prime(&link->link.link, &link_primer);
3460 	if (err)
3461 		goto out_unlock;
3462 
3463 	err = bpf_trampoline_link_prog(&link->link, tr);
3464 	if (err) {
3465 		bpf_link_cleanup(&link_primer);
3466 		link = NULL;
3467 		goto out_unlock;
3468 	}
3469 
3470 	link->tgt_prog = tgt_prog;
3471 	link->trampoline = tr;
3472 
3473 	/* Always clear the trampoline and target prog from prog->aux to make
3474 	 * sure the original attach destination is not kept alive after a
3475 	 * program is (re-)attached to another target.
3476 	 */
3477 	if (prog->aux->dst_prog &&
3478 	    (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3479 		/* got extra prog ref from syscall, or attaching to different prog */
3480 		bpf_prog_put(prog->aux->dst_prog);
3481 	if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3482 		/* we allocated a new trampoline, so free the old one */
3483 		bpf_trampoline_put(prog->aux->dst_trampoline);
3484 
3485 	prog->aux->dst_prog = NULL;
3486 	prog->aux->dst_trampoline = NULL;
3487 	mutex_unlock(&prog->aux->dst_mutex);
3488 
3489 	return bpf_link_settle(&link_primer);
3490 out_unlock:
3491 	if (tr && tr != prog->aux->dst_trampoline)
3492 		bpf_trampoline_put(tr);
3493 	mutex_unlock(&prog->aux->dst_mutex);
3494 	kfree(link);
3495 out_put_prog:
3496 	if (tgt_prog_fd && tgt_prog)
3497 		bpf_prog_put(tgt_prog);
3498 	return err;
3499 }
3500 
3501 struct bpf_raw_tp_link {
3502 	struct bpf_link link;
3503 	struct bpf_raw_event_map *btp;
3504 };
3505 
3506 static void bpf_raw_tp_link_release(struct bpf_link *link)
3507 {
3508 	struct bpf_raw_tp_link *raw_tp =
3509 		container_of(link, struct bpf_raw_tp_link, link);
3510 
3511 	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3512 	bpf_put_raw_tracepoint(raw_tp->btp);
3513 }
3514 
3515 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3516 {
3517 	struct bpf_raw_tp_link *raw_tp =
3518 		container_of(link, struct bpf_raw_tp_link, link);
3519 
3520 	kfree(raw_tp);
3521 }
3522 
3523 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3524 					struct seq_file *seq)
3525 {
3526 	struct bpf_raw_tp_link *raw_tp_link =
3527 		container_of(link, struct bpf_raw_tp_link, link);
3528 
3529 	seq_printf(seq,
3530 		   "tp_name:\t%s\n",
3531 		   raw_tp_link->btp->tp->name);
3532 }
3533 
3534 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3535 			    u32 len)
3536 {
3537 	if (ulen >= len + 1) {
3538 		if (copy_to_user(ubuf, buf, len + 1))
3539 			return -EFAULT;
3540 	} else {
3541 		char zero = '\0';
3542 
3543 		if (copy_to_user(ubuf, buf, ulen - 1))
3544 			return -EFAULT;
3545 		if (put_user(zero, ubuf + ulen - 1))
3546 			return -EFAULT;
3547 		return -ENOSPC;
3548 	}
3549 
3550 	return 0;
3551 }
3552 
3553 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3554 					  struct bpf_link_info *info)
3555 {
3556 	struct bpf_raw_tp_link *raw_tp_link =
3557 		container_of(link, struct bpf_raw_tp_link, link);
3558 	char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3559 	const char *tp_name = raw_tp_link->btp->tp->name;
3560 	u32 ulen = info->raw_tracepoint.tp_name_len;
3561 	size_t tp_len = strlen(tp_name);
3562 
3563 	if (!ulen ^ !ubuf)
3564 		return -EINVAL;
3565 
3566 	info->raw_tracepoint.tp_name_len = tp_len + 1;
3567 
3568 	if (!ubuf)
3569 		return 0;
3570 
3571 	return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3572 }
3573 
3574 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3575 	.release = bpf_raw_tp_link_release,
3576 	.dealloc_deferred = bpf_raw_tp_link_dealloc,
3577 	.show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3578 	.fill_link_info = bpf_raw_tp_link_fill_link_info,
3579 };
3580 
3581 #ifdef CONFIG_PERF_EVENTS
3582 struct bpf_perf_link {
3583 	struct bpf_link link;
3584 	struct file *perf_file;
3585 };
3586 
3587 static void bpf_perf_link_release(struct bpf_link *link)
3588 {
3589 	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3590 	struct perf_event *event = perf_link->perf_file->private_data;
3591 
3592 	perf_event_free_bpf_prog(event);
3593 	fput(perf_link->perf_file);
3594 }
3595 
3596 static void bpf_perf_link_dealloc(struct bpf_link *link)
3597 {
3598 	struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3599 
3600 	kfree(perf_link);
3601 }
3602 
3603 static int bpf_perf_link_fill_common(const struct perf_event *event,
3604 				     char __user *uname, u32 ulen,
3605 				     u64 *probe_offset, u64 *probe_addr,
3606 				     u32 *fd_type, unsigned long *missed)
3607 {
3608 	const char *buf;
3609 	u32 prog_id;
3610 	size_t len;
3611 	int err;
3612 
3613 	if (!ulen ^ !uname)
3614 		return -EINVAL;
3615 
3616 	err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3617 				      probe_offset, probe_addr, missed);
3618 	if (err)
3619 		return err;
3620 	if (!uname)
3621 		return 0;
3622 	if (buf) {
3623 		len = strlen(buf);
3624 		err = bpf_copy_to_user(uname, buf, ulen, len);
3625 		if (err)
3626 			return err;
3627 	} else {
3628 		char zero = '\0';
3629 
3630 		if (put_user(zero, uname))
3631 			return -EFAULT;
3632 	}
3633 	return 0;
3634 }
3635 
3636 #ifdef CONFIG_KPROBE_EVENTS
3637 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3638 				     struct bpf_link_info *info)
3639 {
3640 	unsigned long missed;
3641 	char __user *uname;
3642 	u64 addr, offset;
3643 	u32 ulen, type;
3644 	int err;
3645 
3646 	uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3647 	ulen = info->perf_event.kprobe.name_len;
3648 	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3649 					&type, &missed);
3650 	if (err)
3651 		return err;
3652 	if (type == BPF_FD_TYPE_KRETPROBE)
3653 		info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3654 	else
3655 		info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3656 
3657 	info->perf_event.kprobe.offset = offset;
3658 	info->perf_event.kprobe.missed = missed;
3659 	if (!kallsyms_show_value(current_cred()))
3660 		addr = 0;
3661 	info->perf_event.kprobe.addr = addr;
3662 	info->perf_event.kprobe.cookie = event->bpf_cookie;
3663 	return 0;
3664 }
3665 #endif
3666 
3667 #ifdef CONFIG_UPROBE_EVENTS
3668 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3669 				     struct bpf_link_info *info)
3670 {
3671 	char __user *uname;
3672 	u64 addr, offset;
3673 	u32 ulen, type;
3674 	int err;
3675 
3676 	uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3677 	ulen = info->perf_event.uprobe.name_len;
3678 	err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3679 					&type, NULL);
3680 	if (err)
3681 		return err;
3682 
3683 	if (type == BPF_FD_TYPE_URETPROBE)
3684 		info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3685 	else
3686 		info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3687 	info->perf_event.uprobe.offset = offset;
3688 	info->perf_event.uprobe.cookie = event->bpf_cookie;
3689 	return 0;
3690 }
3691 #endif
3692 
3693 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3694 				    struct bpf_link_info *info)
3695 {
3696 #ifdef CONFIG_KPROBE_EVENTS
3697 	if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3698 		return bpf_perf_link_fill_kprobe(event, info);
3699 #endif
3700 #ifdef CONFIG_UPROBE_EVENTS
3701 	if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3702 		return bpf_perf_link_fill_uprobe(event, info);
3703 #endif
3704 	return -EOPNOTSUPP;
3705 }
3706 
3707 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3708 					 struct bpf_link_info *info)
3709 {
3710 	char __user *uname;
3711 	u32 ulen;
3712 
3713 	uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3714 	ulen = info->perf_event.tracepoint.name_len;
3715 	info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3716 	info->perf_event.tracepoint.cookie = event->bpf_cookie;
3717 	return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3718 }
3719 
3720 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3721 					 struct bpf_link_info *info)
3722 {
3723 	info->perf_event.event.type = event->attr.type;
3724 	info->perf_event.event.config = event->attr.config;
3725 	info->perf_event.event.cookie = event->bpf_cookie;
3726 	info->perf_event.type = BPF_PERF_EVENT_EVENT;
3727 	return 0;
3728 }
3729 
3730 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3731 					struct bpf_link_info *info)
3732 {
3733 	struct bpf_perf_link *perf_link;
3734 	const struct perf_event *event;
3735 
3736 	perf_link = container_of(link, struct bpf_perf_link, link);
3737 	event = perf_get_event(perf_link->perf_file);
3738 	if (IS_ERR(event))
3739 		return PTR_ERR(event);
3740 
3741 	switch (event->prog->type) {
3742 	case BPF_PROG_TYPE_PERF_EVENT:
3743 		return bpf_perf_link_fill_perf_event(event, info);
3744 	case BPF_PROG_TYPE_TRACEPOINT:
3745 		return bpf_perf_link_fill_tracepoint(event, info);
3746 	case BPF_PROG_TYPE_KPROBE:
3747 		return bpf_perf_link_fill_probe(event, info);
3748 	default:
3749 		return -EOPNOTSUPP;
3750 	}
3751 }
3752 
3753 static const struct bpf_link_ops bpf_perf_link_lops = {
3754 	.release = bpf_perf_link_release,
3755 	.dealloc = bpf_perf_link_dealloc,
3756 	.fill_link_info = bpf_perf_link_fill_link_info,
3757 };
3758 
3759 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3760 {
3761 	struct bpf_link_primer link_primer;
3762 	struct bpf_perf_link *link;
3763 	struct perf_event *event;
3764 	struct file *perf_file;
3765 	int err;
3766 
3767 	if (attr->link_create.flags)
3768 		return -EINVAL;
3769 
3770 	perf_file = perf_event_get(attr->link_create.target_fd);
3771 	if (IS_ERR(perf_file))
3772 		return PTR_ERR(perf_file);
3773 
3774 	link = kzalloc(sizeof(*link), GFP_USER);
3775 	if (!link) {
3776 		err = -ENOMEM;
3777 		goto out_put_file;
3778 	}
3779 	bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3780 	link->perf_file = perf_file;
3781 
3782 	err = bpf_link_prime(&link->link, &link_primer);
3783 	if (err) {
3784 		kfree(link);
3785 		goto out_put_file;
3786 	}
3787 
3788 	event = perf_file->private_data;
3789 	err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3790 	if (err) {
3791 		bpf_link_cleanup(&link_primer);
3792 		goto out_put_file;
3793 	}
3794 	/* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3795 	bpf_prog_inc(prog);
3796 
3797 	return bpf_link_settle(&link_primer);
3798 
3799 out_put_file:
3800 	fput(perf_file);
3801 	return err;
3802 }
3803 #else
3804 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3805 {
3806 	return -EOPNOTSUPP;
3807 }
3808 #endif /* CONFIG_PERF_EVENTS */
3809 
3810 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3811 				  const char __user *user_tp_name)
3812 {
3813 	struct bpf_link_primer link_primer;
3814 	struct bpf_raw_tp_link *link;
3815 	struct bpf_raw_event_map *btp;
3816 	const char *tp_name;
3817 	char buf[128];
3818 	int err;
3819 
3820 	switch (prog->type) {
3821 	case BPF_PROG_TYPE_TRACING:
3822 	case BPF_PROG_TYPE_EXT:
3823 	case BPF_PROG_TYPE_LSM:
3824 		if (user_tp_name)
3825 			/* The attach point for this category of programs
3826 			 * should be specified via btf_id during program load.
3827 			 */
3828 			return -EINVAL;
3829 		if (prog->type == BPF_PROG_TYPE_TRACING &&
3830 		    prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3831 			tp_name = prog->aux->attach_func_name;
3832 			break;
3833 		}
3834 		return bpf_tracing_prog_attach(prog, 0, 0, 0);
3835 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
3836 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3837 		if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3838 			return -EFAULT;
3839 		buf[sizeof(buf) - 1] = 0;
3840 		tp_name = buf;
3841 		break;
3842 	default:
3843 		return -EINVAL;
3844 	}
3845 
3846 	btp = bpf_get_raw_tracepoint(tp_name);
3847 	if (!btp)
3848 		return -ENOENT;
3849 
3850 	link = kzalloc(sizeof(*link), GFP_USER);
3851 	if (!link) {
3852 		err = -ENOMEM;
3853 		goto out_put_btp;
3854 	}
3855 	bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3856 		      &bpf_raw_tp_link_lops, prog);
3857 	link->btp = btp;
3858 
3859 	err = bpf_link_prime(&link->link, &link_primer);
3860 	if (err) {
3861 		kfree(link);
3862 		goto out_put_btp;
3863 	}
3864 
3865 	err = bpf_probe_register(link->btp, prog);
3866 	if (err) {
3867 		bpf_link_cleanup(&link_primer);
3868 		goto out_put_btp;
3869 	}
3870 
3871 	return bpf_link_settle(&link_primer);
3872 
3873 out_put_btp:
3874 	bpf_put_raw_tracepoint(btp);
3875 	return err;
3876 }
3877 
3878 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3879 
3880 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3881 {
3882 	struct bpf_prog *prog;
3883 	int fd;
3884 
3885 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3886 		return -EINVAL;
3887 
3888 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3889 	if (IS_ERR(prog))
3890 		return PTR_ERR(prog);
3891 
3892 	fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3893 	if (fd < 0)
3894 		bpf_prog_put(prog);
3895 	return fd;
3896 }
3897 
3898 static enum bpf_prog_type
3899 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3900 {
3901 	switch (attach_type) {
3902 	case BPF_CGROUP_INET_INGRESS:
3903 	case BPF_CGROUP_INET_EGRESS:
3904 		return BPF_PROG_TYPE_CGROUP_SKB;
3905 	case BPF_CGROUP_INET_SOCK_CREATE:
3906 	case BPF_CGROUP_INET_SOCK_RELEASE:
3907 	case BPF_CGROUP_INET4_POST_BIND:
3908 	case BPF_CGROUP_INET6_POST_BIND:
3909 		return BPF_PROG_TYPE_CGROUP_SOCK;
3910 	case BPF_CGROUP_INET4_BIND:
3911 	case BPF_CGROUP_INET6_BIND:
3912 	case BPF_CGROUP_INET4_CONNECT:
3913 	case BPF_CGROUP_INET6_CONNECT:
3914 	case BPF_CGROUP_UNIX_CONNECT:
3915 	case BPF_CGROUP_INET4_GETPEERNAME:
3916 	case BPF_CGROUP_INET6_GETPEERNAME:
3917 	case BPF_CGROUP_UNIX_GETPEERNAME:
3918 	case BPF_CGROUP_INET4_GETSOCKNAME:
3919 	case BPF_CGROUP_INET6_GETSOCKNAME:
3920 	case BPF_CGROUP_UNIX_GETSOCKNAME:
3921 	case BPF_CGROUP_UDP4_SENDMSG:
3922 	case BPF_CGROUP_UDP6_SENDMSG:
3923 	case BPF_CGROUP_UNIX_SENDMSG:
3924 	case BPF_CGROUP_UDP4_RECVMSG:
3925 	case BPF_CGROUP_UDP6_RECVMSG:
3926 	case BPF_CGROUP_UNIX_RECVMSG:
3927 		return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3928 	case BPF_CGROUP_SOCK_OPS:
3929 		return BPF_PROG_TYPE_SOCK_OPS;
3930 	case BPF_CGROUP_DEVICE:
3931 		return BPF_PROG_TYPE_CGROUP_DEVICE;
3932 	case BPF_SK_MSG_VERDICT:
3933 		return BPF_PROG_TYPE_SK_MSG;
3934 	case BPF_SK_SKB_STREAM_PARSER:
3935 	case BPF_SK_SKB_STREAM_VERDICT:
3936 	case BPF_SK_SKB_VERDICT:
3937 		return BPF_PROG_TYPE_SK_SKB;
3938 	case BPF_LIRC_MODE2:
3939 		return BPF_PROG_TYPE_LIRC_MODE2;
3940 	case BPF_FLOW_DISSECTOR:
3941 		return BPF_PROG_TYPE_FLOW_DISSECTOR;
3942 	case BPF_CGROUP_SYSCTL:
3943 		return BPF_PROG_TYPE_CGROUP_SYSCTL;
3944 	case BPF_CGROUP_GETSOCKOPT:
3945 	case BPF_CGROUP_SETSOCKOPT:
3946 		return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3947 	case BPF_TRACE_ITER:
3948 	case BPF_TRACE_RAW_TP:
3949 	case BPF_TRACE_FENTRY:
3950 	case BPF_TRACE_FEXIT:
3951 	case BPF_MODIFY_RETURN:
3952 		return BPF_PROG_TYPE_TRACING;
3953 	case BPF_LSM_MAC:
3954 		return BPF_PROG_TYPE_LSM;
3955 	case BPF_SK_LOOKUP:
3956 		return BPF_PROG_TYPE_SK_LOOKUP;
3957 	case BPF_XDP:
3958 		return BPF_PROG_TYPE_XDP;
3959 	case BPF_LSM_CGROUP:
3960 		return BPF_PROG_TYPE_LSM;
3961 	case BPF_TCX_INGRESS:
3962 	case BPF_TCX_EGRESS:
3963 	case BPF_NETKIT_PRIMARY:
3964 	case BPF_NETKIT_PEER:
3965 		return BPF_PROG_TYPE_SCHED_CLS;
3966 	default:
3967 		return BPF_PROG_TYPE_UNSPEC;
3968 	}
3969 }
3970 
3971 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3972 					     enum bpf_attach_type attach_type)
3973 {
3974 	enum bpf_prog_type ptype;
3975 
3976 	switch (prog->type) {
3977 	case BPF_PROG_TYPE_CGROUP_SOCK:
3978 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3979 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3980 	case BPF_PROG_TYPE_SK_LOOKUP:
3981 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3982 	case BPF_PROG_TYPE_CGROUP_SKB:
3983 		if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
3984 			/* cg-skb progs can be loaded by unpriv user.
3985 			 * check permissions at attach time.
3986 			 */
3987 			return -EPERM;
3988 		return prog->enforce_expected_attach_type &&
3989 			prog->expected_attach_type != attach_type ?
3990 			-EINVAL : 0;
3991 	case BPF_PROG_TYPE_EXT:
3992 		return 0;
3993 	case BPF_PROG_TYPE_NETFILTER:
3994 		if (attach_type != BPF_NETFILTER)
3995 			return -EINVAL;
3996 		return 0;
3997 	case BPF_PROG_TYPE_PERF_EVENT:
3998 	case BPF_PROG_TYPE_TRACEPOINT:
3999 		if (attach_type != BPF_PERF_EVENT)
4000 			return -EINVAL;
4001 		return 0;
4002 	case BPF_PROG_TYPE_KPROBE:
4003 		if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
4004 		    attach_type != BPF_TRACE_KPROBE_MULTI)
4005 			return -EINVAL;
4006 		if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
4007 		    attach_type != BPF_TRACE_UPROBE_MULTI)
4008 			return -EINVAL;
4009 		if (attach_type != BPF_PERF_EVENT &&
4010 		    attach_type != BPF_TRACE_KPROBE_MULTI &&
4011 		    attach_type != BPF_TRACE_UPROBE_MULTI)
4012 			return -EINVAL;
4013 		return 0;
4014 	case BPF_PROG_TYPE_SCHED_CLS:
4015 		if (attach_type != BPF_TCX_INGRESS &&
4016 		    attach_type != BPF_TCX_EGRESS &&
4017 		    attach_type != BPF_NETKIT_PRIMARY &&
4018 		    attach_type != BPF_NETKIT_PEER)
4019 			return -EINVAL;
4020 		return 0;
4021 	default:
4022 		ptype = attach_type_to_prog_type(attach_type);
4023 		if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4024 			return -EINVAL;
4025 		return 0;
4026 	}
4027 }
4028 
4029 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4030 
4031 #define BPF_F_ATTACH_MASK_BASE	\
4032 	(BPF_F_ALLOW_OVERRIDE |	\
4033 	 BPF_F_ALLOW_MULTI |	\
4034 	 BPF_F_REPLACE)
4035 
4036 #define BPF_F_ATTACH_MASK_MPROG	\
4037 	(BPF_F_REPLACE |	\
4038 	 BPF_F_BEFORE |		\
4039 	 BPF_F_AFTER |		\
4040 	 BPF_F_ID |		\
4041 	 BPF_F_LINK)
4042 
4043 static int bpf_prog_attach(const union bpf_attr *attr)
4044 {
4045 	enum bpf_prog_type ptype;
4046 	struct bpf_prog *prog;
4047 	int ret;
4048 
4049 	if (CHECK_ATTR(BPF_PROG_ATTACH))
4050 		return -EINVAL;
4051 
4052 	ptype = attach_type_to_prog_type(attr->attach_type);
4053 	if (ptype == BPF_PROG_TYPE_UNSPEC)
4054 		return -EINVAL;
4055 	if (bpf_mprog_supported(ptype)) {
4056 		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4057 			return -EINVAL;
4058 	} else {
4059 		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4060 			return -EINVAL;
4061 		if (attr->relative_fd ||
4062 		    attr->expected_revision)
4063 			return -EINVAL;
4064 	}
4065 
4066 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4067 	if (IS_ERR(prog))
4068 		return PTR_ERR(prog);
4069 
4070 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4071 		bpf_prog_put(prog);
4072 		return -EINVAL;
4073 	}
4074 
4075 	switch (ptype) {
4076 	case BPF_PROG_TYPE_SK_SKB:
4077 	case BPF_PROG_TYPE_SK_MSG:
4078 		ret = sock_map_get_from_fd(attr, prog);
4079 		break;
4080 	case BPF_PROG_TYPE_LIRC_MODE2:
4081 		ret = lirc_prog_attach(attr, prog);
4082 		break;
4083 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4084 		ret = netns_bpf_prog_attach(attr, prog);
4085 		break;
4086 	case BPF_PROG_TYPE_CGROUP_DEVICE:
4087 	case BPF_PROG_TYPE_CGROUP_SKB:
4088 	case BPF_PROG_TYPE_CGROUP_SOCK:
4089 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4090 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4091 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4092 	case BPF_PROG_TYPE_SOCK_OPS:
4093 	case BPF_PROG_TYPE_LSM:
4094 		if (ptype == BPF_PROG_TYPE_LSM &&
4095 		    prog->expected_attach_type != BPF_LSM_CGROUP)
4096 			ret = -EINVAL;
4097 		else
4098 			ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4099 		break;
4100 	case BPF_PROG_TYPE_SCHED_CLS:
4101 		if (attr->attach_type == BPF_TCX_INGRESS ||
4102 		    attr->attach_type == BPF_TCX_EGRESS)
4103 			ret = tcx_prog_attach(attr, prog);
4104 		else
4105 			ret = netkit_prog_attach(attr, prog);
4106 		break;
4107 	default:
4108 		ret = -EINVAL;
4109 	}
4110 
4111 	if (ret)
4112 		bpf_prog_put(prog);
4113 	return ret;
4114 }
4115 
4116 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
4117 
4118 static int bpf_prog_detach(const union bpf_attr *attr)
4119 {
4120 	struct bpf_prog *prog = NULL;
4121 	enum bpf_prog_type ptype;
4122 	int ret;
4123 
4124 	if (CHECK_ATTR(BPF_PROG_DETACH))
4125 		return -EINVAL;
4126 
4127 	ptype = attach_type_to_prog_type(attr->attach_type);
4128 	if (bpf_mprog_supported(ptype)) {
4129 		if (ptype == BPF_PROG_TYPE_UNSPEC)
4130 			return -EINVAL;
4131 		if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4132 			return -EINVAL;
4133 		if (attr->attach_bpf_fd) {
4134 			prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4135 			if (IS_ERR(prog))
4136 				return PTR_ERR(prog);
4137 		}
4138 	} else if (attr->attach_flags ||
4139 		   attr->relative_fd ||
4140 		   attr->expected_revision) {
4141 		return -EINVAL;
4142 	}
4143 
4144 	switch (ptype) {
4145 	case BPF_PROG_TYPE_SK_MSG:
4146 	case BPF_PROG_TYPE_SK_SKB:
4147 		ret = sock_map_prog_detach(attr, ptype);
4148 		break;
4149 	case BPF_PROG_TYPE_LIRC_MODE2:
4150 		ret = lirc_prog_detach(attr);
4151 		break;
4152 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4153 		ret = netns_bpf_prog_detach(attr, ptype);
4154 		break;
4155 	case BPF_PROG_TYPE_CGROUP_DEVICE:
4156 	case BPF_PROG_TYPE_CGROUP_SKB:
4157 	case BPF_PROG_TYPE_CGROUP_SOCK:
4158 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4159 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4160 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4161 	case BPF_PROG_TYPE_SOCK_OPS:
4162 	case BPF_PROG_TYPE_LSM:
4163 		ret = cgroup_bpf_prog_detach(attr, ptype);
4164 		break;
4165 	case BPF_PROG_TYPE_SCHED_CLS:
4166 		if (attr->attach_type == BPF_TCX_INGRESS ||
4167 		    attr->attach_type == BPF_TCX_EGRESS)
4168 			ret = tcx_prog_detach(attr, prog);
4169 		else
4170 			ret = netkit_prog_detach(attr, prog);
4171 		break;
4172 	default:
4173 		ret = -EINVAL;
4174 	}
4175 
4176 	if (prog)
4177 		bpf_prog_put(prog);
4178 	return ret;
4179 }
4180 
4181 #define BPF_PROG_QUERY_LAST_FIELD query.revision
4182 
4183 static int bpf_prog_query(const union bpf_attr *attr,
4184 			  union bpf_attr __user *uattr)
4185 {
4186 	if (!bpf_net_capable())
4187 		return -EPERM;
4188 	if (CHECK_ATTR(BPF_PROG_QUERY))
4189 		return -EINVAL;
4190 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4191 		return -EINVAL;
4192 
4193 	switch (attr->query.attach_type) {
4194 	case BPF_CGROUP_INET_INGRESS:
4195 	case BPF_CGROUP_INET_EGRESS:
4196 	case BPF_CGROUP_INET_SOCK_CREATE:
4197 	case BPF_CGROUP_INET_SOCK_RELEASE:
4198 	case BPF_CGROUP_INET4_BIND:
4199 	case BPF_CGROUP_INET6_BIND:
4200 	case BPF_CGROUP_INET4_POST_BIND:
4201 	case BPF_CGROUP_INET6_POST_BIND:
4202 	case BPF_CGROUP_INET4_CONNECT:
4203 	case BPF_CGROUP_INET6_CONNECT:
4204 	case BPF_CGROUP_UNIX_CONNECT:
4205 	case BPF_CGROUP_INET4_GETPEERNAME:
4206 	case BPF_CGROUP_INET6_GETPEERNAME:
4207 	case BPF_CGROUP_UNIX_GETPEERNAME:
4208 	case BPF_CGROUP_INET4_GETSOCKNAME:
4209 	case BPF_CGROUP_INET6_GETSOCKNAME:
4210 	case BPF_CGROUP_UNIX_GETSOCKNAME:
4211 	case BPF_CGROUP_UDP4_SENDMSG:
4212 	case BPF_CGROUP_UDP6_SENDMSG:
4213 	case BPF_CGROUP_UNIX_SENDMSG:
4214 	case BPF_CGROUP_UDP4_RECVMSG:
4215 	case BPF_CGROUP_UDP6_RECVMSG:
4216 	case BPF_CGROUP_UNIX_RECVMSG:
4217 	case BPF_CGROUP_SOCK_OPS:
4218 	case BPF_CGROUP_DEVICE:
4219 	case BPF_CGROUP_SYSCTL:
4220 	case BPF_CGROUP_GETSOCKOPT:
4221 	case BPF_CGROUP_SETSOCKOPT:
4222 	case BPF_LSM_CGROUP:
4223 		return cgroup_bpf_prog_query(attr, uattr);
4224 	case BPF_LIRC_MODE2:
4225 		return lirc_prog_query(attr, uattr);
4226 	case BPF_FLOW_DISSECTOR:
4227 	case BPF_SK_LOOKUP:
4228 		return netns_bpf_prog_query(attr, uattr);
4229 	case BPF_SK_SKB_STREAM_PARSER:
4230 	case BPF_SK_SKB_STREAM_VERDICT:
4231 	case BPF_SK_MSG_VERDICT:
4232 	case BPF_SK_SKB_VERDICT:
4233 		return sock_map_bpf_prog_query(attr, uattr);
4234 	case BPF_TCX_INGRESS:
4235 	case BPF_TCX_EGRESS:
4236 		return tcx_prog_query(attr, uattr);
4237 	case BPF_NETKIT_PRIMARY:
4238 	case BPF_NETKIT_PEER:
4239 		return netkit_prog_query(attr, uattr);
4240 	default:
4241 		return -EINVAL;
4242 	}
4243 }
4244 
4245 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4246 
4247 static int bpf_prog_test_run(const union bpf_attr *attr,
4248 			     union bpf_attr __user *uattr)
4249 {
4250 	struct bpf_prog *prog;
4251 	int ret = -ENOTSUPP;
4252 
4253 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4254 		return -EINVAL;
4255 
4256 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4257 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
4258 		return -EINVAL;
4259 
4260 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4261 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
4262 		return -EINVAL;
4263 
4264 	prog = bpf_prog_get(attr->test.prog_fd);
4265 	if (IS_ERR(prog))
4266 		return PTR_ERR(prog);
4267 
4268 	if (prog->aux->ops->test_run)
4269 		ret = prog->aux->ops->test_run(prog, attr, uattr);
4270 
4271 	bpf_prog_put(prog);
4272 	return ret;
4273 }
4274 
4275 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4276 
4277 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4278 			       union bpf_attr __user *uattr,
4279 			       struct idr *idr,
4280 			       spinlock_t *lock)
4281 {
4282 	u32 next_id = attr->start_id;
4283 	int err = 0;
4284 
4285 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4286 		return -EINVAL;
4287 
4288 	if (!capable(CAP_SYS_ADMIN))
4289 		return -EPERM;
4290 
4291 	next_id++;
4292 	spin_lock_bh(lock);
4293 	if (!idr_get_next(idr, &next_id))
4294 		err = -ENOENT;
4295 	spin_unlock_bh(lock);
4296 
4297 	if (!err)
4298 		err = put_user(next_id, &uattr->next_id);
4299 
4300 	return err;
4301 }
4302 
4303 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4304 {
4305 	struct bpf_map *map;
4306 
4307 	spin_lock_bh(&map_idr_lock);
4308 again:
4309 	map = idr_get_next(&map_idr, id);
4310 	if (map) {
4311 		map = __bpf_map_inc_not_zero(map, false);
4312 		if (IS_ERR(map)) {
4313 			(*id)++;
4314 			goto again;
4315 		}
4316 	}
4317 	spin_unlock_bh(&map_idr_lock);
4318 
4319 	return map;
4320 }
4321 
4322 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4323 {
4324 	struct bpf_prog *prog;
4325 
4326 	spin_lock_bh(&prog_idr_lock);
4327 again:
4328 	prog = idr_get_next(&prog_idr, id);
4329 	if (prog) {
4330 		prog = bpf_prog_inc_not_zero(prog);
4331 		if (IS_ERR(prog)) {
4332 			(*id)++;
4333 			goto again;
4334 		}
4335 	}
4336 	spin_unlock_bh(&prog_idr_lock);
4337 
4338 	return prog;
4339 }
4340 
4341 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4342 
4343 struct bpf_prog *bpf_prog_by_id(u32 id)
4344 {
4345 	struct bpf_prog *prog;
4346 
4347 	if (!id)
4348 		return ERR_PTR(-ENOENT);
4349 
4350 	spin_lock_bh(&prog_idr_lock);
4351 	prog = idr_find(&prog_idr, id);
4352 	if (prog)
4353 		prog = bpf_prog_inc_not_zero(prog);
4354 	else
4355 		prog = ERR_PTR(-ENOENT);
4356 	spin_unlock_bh(&prog_idr_lock);
4357 	return prog;
4358 }
4359 
4360 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4361 {
4362 	struct bpf_prog *prog;
4363 	u32 id = attr->prog_id;
4364 	int fd;
4365 
4366 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4367 		return -EINVAL;
4368 
4369 	if (!capable(CAP_SYS_ADMIN))
4370 		return -EPERM;
4371 
4372 	prog = bpf_prog_by_id(id);
4373 	if (IS_ERR(prog))
4374 		return PTR_ERR(prog);
4375 
4376 	fd = bpf_prog_new_fd(prog);
4377 	if (fd < 0)
4378 		bpf_prog_put(prog);
4379 
4380 	return fd;
4381 }
4382 
4383 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4384 
4385 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4386 {
4387 	struct bpf_map *map;
4388 	u32 id = attr->map_id;
4389 	int f_flags;
4390 	int fd;
4391 
4392 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4393 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4394 		return -EINVAL;
4395 
4396 	if (!capable(CAP_SYS_ADMIN))
4397 		return -EPERM;
4398 
4399 	f_flags = bpf_get_file_flag(attr->open_flags);
4400 	if (f_flags < 0)
4401 		return f_flags;
4402 
4403 	spin_lock_bh(&map_idr_lock);
4404 	map = idr_find(&map_idr, id);
4405 	if (map)
4406 		map = __bpf_map_inc_not_zero(map, true);
4407 	else
4408 		map = ERR_PTR(-ENOENT);
4409 	spin_unlock_bh(&map_idr_lock);
4410 
4411 	if (IS_ERR(map))
4412 		return PTR_ERR(map);
4413 
4414 	fd = bpf_map_new_fd(map, f_flags);
4415 	if (fd < 0)
4416 		bpf_map_put_with_uref(map);
4417 
4418 	return fd;
4419 }
4420 
4421 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4422 					      unsigned long addr, u32 *off,
4423 					      u32 *type)
4424 {
4425 	const struct bpf_map *map;
4426 	int i;
4427 
4428 	mutex_lock(&prog->aux->used_maps_mutex);
4429 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4430 		map = prog->aux->used_maps[i];
4431 		if (map == (void *)addr) {
4432 			*type = BPF_PSEUDO_MAP_FD;
4433 			goto out;
4434 		}
4435 		if (!map->ops->map_direct_value_meta)
4436 			continue;
4437 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
4438 			*type = BPF_PSEUDO_MAP_VALUE;
4439 			goto out;
4440 		}
4441 	}
4442 	map = NULL;
4443 
4444 out:
4445 	mutex_unlock(&prog->aux->used_maps_mutex);
4446 	return map;
4447 }
4448 
4449 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4450 					      const struct cred *f_cred)
4451 {
4452 	const struct bpf_map *map;
4453 	struct bpf_insn *insns;
4454 	u32 off, type;
4455 	u64 imm;
4456 	u8 code;
4457 	int i;
4458 
4459 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4460 			GFP_USER);
4461 	if (!insns)
4462 		return insns;
4463 
4464 	for (i = 0; i < prog->len; i++) {
4465 		code = insns[i].code;
4466 
4467 		if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4468 			insns[i].code = BPF_JMP | BPF_CALL;
4469 			insns[i].imm = BPF_FUNC_tail_call;
4470 			/* fall-through */
4471 		}
4472 		if (code == (BPF_JMP | BPF_CALL) ||
4473 		    code == (BPF_JMP | BPF_CALL_ARGS)) {
4474 			if (code == (BPF_JMP | BPF_CALL_ARGS))
4475 				insns[i].code = BPF_JMP | BPF_CALL;
4476 			if (!bpf_dump_raw_ok(f_cred))
4477 				insns[i].imm = 0;
4478 			continue;
4479 		}
4480 		if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4481 			insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4482 			continue;
4483 		}
4484 
4485 		if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4486 		     BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4487 			insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4488 			continue;
4489 		}
4490 
4491 		if (code != (BPF_LD | BPF_IMM | BPF_DW))
4492 			continue;
4493 
4494 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4495 		map = bpf_map_from_imm(prog, imm, &off, &type);
4496 		if (map) {
4497 			insns[i].src_reg = type;
4498 			insns[i].imm = map->id;
4499 			insns[i + 1].imm = off;
4500 			continue;
4501 		}
4502 	}
4503 
4504 	return insns;
4505 }
4506 
4507 static int set_info_rec_size(struct bpf_prog_info *info)
4508 {
4509 	/*
4510 	 * Ensure info.*_rec_size is the same as kernel expected size
4511 	 *
4512 	 * or
4513 	 *
4514 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
4515 	 * zero.  In this case, the kernel will set the expected
4516 	 * _rec_size back to the info.
4517 	 */
4518 
4519 	if ((info->nr_func_info || info->func_info_rec_size) &&
4520 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
4521 		return -EINVAL;
4522 
4523 	if ((info->nr_line_info || info->line_info_rec_size) &&
4524 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
4525 		return -EINVAL;
4526 
4527 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4528 	    info->jited_line_info_rec_size != sizeof(__u64))
4529 		return -EINVAL;
4530 
4531 	info->func_info_rec_size = sizeof(struct bpf_func_info);
4532 	info->line_info_rec_size = sizeof(struct bpf_line_info);
4533 	info->jited_line_info_rec_size = sizeof(__u64);
4534 
4535 	return 0;
4536 }
4537 
4538 static int bpf_prog_get_info_by_fd(struct file *file,
4539 				   struct bpf_prog *prog,
4540 				   const union bpf_attr *attr,
4541 				   union bpf_attr __user *uattr)
4542 {
4543 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4544 	struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4545 	struct bpf_prog_info info;
4546 	u32 info_len = attr->info.info_len;
4547 	struct bpf_prog_kstats stats;
4548 	char __user *uinsns;
4549 	u32 ulen;
4550 	int err;
4551 
4552 	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4553 	if (err)
4554 		return err;
4555 	info_len = min_t(u32, sizeof(info), info_len);
4556 
4557 	memset(&info, 0, sizeof(info));
4558 	if (copy_from_user(&info, uinfo, info_len))
4559 		return -EFAULT;
4560 
4561 	info.type = prog->type;
4562 	info.id = prog->aux->id;
4563 	info.load_time = prog->aux->load_time;
4564 	info.created_by_uid = from_kuid_munged(current_user_ns(),
4565 					       prog->aux->user->uid);
4566 	info.gpl_compatible = prog->gpl_compatible;
4567 
4568 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
4569 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4570 
4571 	mutex_lock(&prog->aux->used_maps_mutex);
4572 	ulen = info.nr_map_ids;
4573 	info.nr_map_ids = prog->aux->used_map_cnt;
4574 	ulen = min_t(u32, info.nr_map_ids, ulen);
4575 	if (ulen) {
4576 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4577 		u32 i;
4578 
4579 		for (i = 0; i < ulen; i++)
4580 			if (put_user(prog->aux->used_maps[i]->id,
4581 				     &user_map_ids[i])) {
4582 				mutex_unlock(&prog->aux->used_maps_mutex);
4583 				return -EFAULT;
4584 			}
4585 	}
4586 	mutex_unlock(&prog->aux->used_maps_mutex);
4587 
4588 	err = set_info_rec_size(&info);
4589 	if (err)
4590 		return err;
4591 
4592 	bpf_prog_get_stats(prog, &stats);
4593 	info.run_time_ns = stats.nsecs;
4594 	info.run_cnt = stats.cnt;
4595 	info.recursion_misses = stats.misses;
4596 
4597 	info.verified_insns = prog->aux->verified_insns;
4598 
4599 	if (!bpf_capable()) {
4600 		info.jited_prog_len = 0;
4601 		info.xlated_prog_len = 0;
4602 		info.nr_jited_ksyms = 0;
4603 		info.nr_jited_func_lens = 0;
4604 		info.nr_func_info = 0;
4605 		info.nr_line_info = 0;
4606 		info.nr_jited_line_info = 0;
4607 		goto done;
4608 	}
4609 
4610 	ulen = info.xlated_prog_len;
4611 	info.xlated_prog_len = bpf_prog_insn_size(prog);
4612 	if (info.xlated_prog_len && ulen) {
4613 		struct bpf_insn *insns_sanitized;
4614 		bool fault;
4615 
4616 		if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4617 			info.xlated_prog_insns = 0;
4618 			goto done;
4619 		}
4620 		insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4621 		if (!insns_sanitized)
4622 			return -ENOMEM;
4623 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4624 		ulen = min_t(u32, info.xlated_prog_len, ulen);
4625 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
4626 		kfree(insns_sanitized);
4627 		if (fault)
4628 			return -EFAULT;
4629 	}
4630 
4631 	if (bpf_prog_is_offloaded(prog->aux)) {
4632 		err = bpf_prog_offload_info_fill(&info, prog);
4633 		if (err)
4634 			return err;
4635 		goto done;
4636 	}
4637 
4638 	/* NOTE: the following code is supposed to be skipped for offload.
4639 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
4640 	 * for offload.
4641 	 */
4642 	ulen = info.jited_prog_len;
4643 	if (prog->aux->func_cnt) {
4644 		u32 i;
4645 
4646 		info.jited_prog_len = 0;
4647 		for (i = 0; i < prog->aux->func_cnt; i++)
4648 			info.jited_prog_len += prog->aux->func[i]->jited_len;
4649 	} else {
4650 		info.jited_prog_len = prog->jited_len;
4651 	}
4652 
4653 	if (info.jited_prog_len && ulen) {
4654 		if (bpf_dump_raw_ok(file->f_cred)) {
4655 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
4656 			ulen = min_t(u32, info.jited_prog_len, ulen);
4657 
4658 			/* for multi-function programs, copy the JITed
4659 			 * instructions for all the functions
4660 			 */
4661 			if (prog->aux->func_cnt) {
4662 				u32 len, free, i;
4663 				u8 *img;
4664 
4665 				free = ulen;
4666 				for (i = 0; i < prog->aux->func_cnt; i++) {
4667 					len = prog->aux->func[i]->jited_len;
4668 					len = min_t(u32, len, free);
4669 					img = (u8 *) prog->aux->func[i]->bpf_func;
4670 					if (copy_to_user(uinsns, img, len))
4671 						return -EFAULT;
4672 					uinsns += len;
4673 					free -= len;
4674 					if (!free)
4675 						break;
4676 				}
4677 			} else {
4678 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
4679 					return -EFAULT;
4680 			}
4681 		} else {
4682 			info.jited_prog_insns = 0;
4683 		}
4684 	}
4685 
4686 	ulen = info.nr_jited_ksyms;
4687 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4688 	if (ulen) {
4689 		if (bpf_dump_raw_ok(file->f_cred)) {
4690 			unsigned long ksym_addr;
4691 			u64 __user *user_ksyms;
4692 			u32 i;
4693 
4694 			/* copy the address of the kernel symbol
4695 			 * corresponding to each function
4696 			 */
4697 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4698 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4699 			if (prog->aux->func_cnt) {
4700 				for (i = 0; i < ulen; i++) {
4701 					ksym_addr = (unsigned long)
4702 						prog->aux->func[i]->bpf_func;
4703 					if (put_user((u64) ksym_addr,
4704 						     &user_ksyms[i]))
4705 						return -EFAULT;
4706 				}
4707 			} else {
4708 				ksym_addr = (unsigned long) prog->bpf_func;
4709 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
4710 					return -EFAULT;
4711 			}
4712 		} else {
4713 			info.jited_ksyms = 0;
4714 		}
4715 	}
4716 
4717 	ulen = info.nr_jited_func_lens;
4718 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4719 	if (ulen) {
4720 		if (bpf_dump_raw_ok(file->f_cred)) {
4721 			u32 __user *user_lens;
4722 			u32 func_len, i;
4723 
4724 			/* copy the JITed image lengths for each function */
4725 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4726 			user_lens = u64_to_user_ptr(info.jited_func_lens);
4727 			if (prog->aux->func_cnt) {
4728 				for (i = 0; i < ulen; i++) {
4729 					func_len =
4730 						prog->aux->func[i]->jited_len;
4731 					if (put_user(func_len, &user_lens[i]))
4732 						return -EFAULT;
4733 				}
4734 			} else {
4735 				func_len = prog->jited_len;
4736 				if (put_user(func_len, &user_lens[0]))
4737 					return -EFAULT;
4738 			}
4739 		} else {
4740 			info.jited_func_lens = 0;
4741 		}
4742 	}
4743 
4744 	if (prog->aux->btf)
4745 		info.btf_id = btf_obj_id(prog->aux->btf);
4746 	info.attach_btf_id = prog->aux->attach_btf_id;
4747 	if (attach_btf)
4748 		info.attach_btf_obj_id = btf_obj_id(attach_btf);
4749 
4750 	ulen = info.nr_func_info;
4751 	info.nr_func_info = prog->aux->func_info_cnt;
4752 	if (info.nr_func_info && ulen) {
4753 		char __user *user_finfo;
4754 
4755 		user_finfo = u64_to_user_ptr(info.func_info);
4756 		ulen = min_t(u32, info.nr_func_info, ulen);
4757 		if (copy_to_user(user_finfo, prog->aux->func_info,
4758 				 info.func_info_rec_size * ulen))
4759 			return -EFAULT;
4760 	}
4761 
4762 	ulen = info.nr_line_info;
4763 	info.nr_line_info = prog->aux->nr_linfo;
4764 	if (info.nr_line_info && ulen) {
4765 		__u8 __user *user_linfo;
4766 
4767 		user_linfo = u64_to_user_ptr(info.line_info);
4768 		ulen = min_t(u32, info.nr_line_info, ulen);
4769 		if (copy_to_user(user_linfo, prog->aux->linfo,
4770 				 info.line_info_rec_size * ulen))
4771 			return -EFAULT;
4772 	}
4773 
4774 	ulen = info.nr_jited_line_info;
4775 	if (prog->aux->jited_linfo)
4776 		info.nr_jited_line_info = prog->aux->nr_linfo;
4777 	else
4778 		info.nr_jited_line_info = 0;
4779 	if (info.nr_jited_line_info && ulen) {
4780 		if (bpf_dump_raw_ok(file->f_cred)) {
4781 			unsigned long line_addr;
4782 			__u64 __user *user_linfo;
4783 			u32 i;
4784 
4785 			user_linfo = u64_to_user_ptr(info.jited_line_info);
4786 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
4787 			for (i = 0; i < ulen; i++) {
4788 				line_addr = (unsigned long)prog->aux->jited_linfo[i];
4789 				if (put_user((__u64)line_addr, &user_linfo[i]))
4790 					return -EFAULT;
4791 			}
4792 		} else {
4793 			info.jited_line_info = 0;
4794 		}
4795 	}
4796 
4797 	ulen = info.nr_prog_tags;
4798 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4799 	if (ulen) {
4800 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4801 		u32 i;
4802 
4803 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
4804 		ulen = min_t(u32, info.nr_prog_tags, ulen);
4805 		if (prog->aux->func_cnt) {
4806 			for (i = 0; i < ulen; i++) {
4807 				if (copy_to_user(user_prog_tags[i],
4808 						 prog->aux->func[i]->tag,
4809 						 BPF_TAG_SIZE))
4810 					return -EFAULT;
4811 			}
4812 		} else {
4813 			if (copy_to_user(user_prog_tags[0],
4814 					 prog->tag, BPF_TAG_SIZE))
4815 				return -EFAULT;
4816 		}
4817 	}
4818 
4819 done:
4820 	if (copy_to_user(uinfo, &info, info_len) ||
4821 	    put_user(info_len, &uattr->info.info_len))
4822 		return -EFAULT;
4823 
4824 	return 0;
4825 }
4826 
4827 static int bpf_map_get_info_by_fd(struct file *file,
4828 				  struct bpf_map *map,
4829 				  const union bpf_attr *attr,
4830 				  union bpf_attr __user *uattr)
4831 {
4832 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4833 	struct bpf_map_info info;
4834 	u32 info_len = attr->info.info_len;
4835 	int err;
4836 
4837 	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4838 	if (err)
4839 		return err;
4840 	info_len = min_t(u32, sizeof(info), info_len);
4841 
4842 	memset(&info, 0, sizeof(info));
4843 	info.type = map->map_type;
4844 	info.id = map->id;
4845 	info.key_size = map->key_size;
4846 	info.value_size = map->value_size;
4847 	info.max_entries = map->max_entries;
4848 	info.map_flags = map->map_flags;
4849 	info.map_extra = map->map_extra;
4850 	memcpy(info.name, map->name, sizeof(map->name));
4851 
4852 	if (map->btf) {
4853 		info.btf_id = btf_obj_id(map->btf);
4854 		info.btf_key_type_id = map->btf_key_type_id;
4855 		info.btf_value_type_id = map->btf_value_type_id;
4856 	}
4857 	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4858 	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4859 		bpf_map_struct_ops_info_fill(&info, map);
4860 
4861 	if (bpf_map_is_offloaded(map)) {
4862 		err = bpf_map_offload_info_fill(&info, map);
4863 		if (err)
4864 			return err;
4865 	}
4866 
4867 	if (copy_to_user(uinfo, &info, info_len) ||
4868 	    put_user(info_len, &uattr->info.info_len))
4869 		return -EFAULT;
4870 
4871 	return 0;
4872 }
4873 
4874 static int bpf_btf_get_info_by_fd(struct file *file,
4875 				  struct btf *btf,
4876 				  const union bpf_attr *attr,
4877 				  union bpf_attr __user *uattr)
4878 {
4879 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4880 	u32 info_len = attr->info.info_len;
4881 	int err;
4882 
4883 	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4884 	if (err)
4885 		return err;
4886 
4887 	return btf_get_info_by_fd(btf, attr, uattr);
4888 }
4889 
4890 static int bpf_link_get_info_by_fd(struct file *file,
4891 				  struct bpf_link *link,
4892 				  const union bpf_attr *attr,
4893 				  union bpf_attr __user *uattr)
4894 {
4895 	struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4896 	struct bpf_link_info info;
4897 	u32 info_len = attr->info.info_len;
4898 	int err;
4899 
4900 	err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4901 	if (err)
4902 		return err;
4903 	info_len = min_t(u32, sizeof(info), info_len);
4904 
4905 	memset(&info, 0, sizeof(info));
4906 	if (copy_from_user(&info, uinfo, info_len))
4907 		return -EFAULT;
4908 
4909 	info.type = link->type;
4910 	info.id = link->id;
4911 	if (link->prog)
4912 		info.prog_id = link->prog->aux->id;
4913 
4914 	if (link->ops->fill_link_info) {
4915 		err = link->ops->fill_link_info(link, &info);
4916 		if (err)
4917 			return err;
4918 	}
4919 
4920 	if (copy_to_user(uinfo, &info, info_len) ||
4921 	    put_user(info_len, &uattr->info.info_len))
4922 		return -EFAULT;
4923 
4924 	return 0;
4925 }
4926 
4927 
4928 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4929 
4930 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4931 				  union bpf_attr __user *uattr)
4932 {
4933 	int ufd = attr->info.bpf_fd;
4934 	struct fd f;
4935 	int err;
4936 
4937 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4938 		return -EINVAL;
4939 
4940 	f = fdget(ufd);
4941 	if (!f.file)
4942 		return -EBADFD;
4943 
4944 	if (f.file->f_op == &bpf_prog_fops)
4945 		err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4946 					      uattr);
4947 	else if (f.file->f_op == &bpf_map_fops)
4948 		err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4949 					     uattr);
4950 	else if (f.file->f_op == &btf_fops)
4951 		err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4952 	else if (f.file->f_op == &bpf_link_fops)
4953 		err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4954 					      attr, uattr);
4955 	else
4956 		err = -EINVAL;
4957 
4958 	fdput(f);
4959 	return err;
4960 }
4961 
4962 #define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
4963 
4964 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4965 {
4966 	struct bpf_token *token = NULL;
4967 
4968 	if (CHECK_ATTR(BPF_BTF_LOAD))
4969 		return -EINVAL;
4970 
4971 	if (attr->btf_flags & ~BPF_F_TOKEN_FD)
4972 		return -EINVAL;
4973 
4974 	if (attr->btf_flags & BPF_F_TOKEN_FD) {
4975 		token = bpf_token_get_from_fd(attr->btf_token_fd);
4976 		if (IS_ERR(token))
4977 			return PTR_ERR(token);
4978 		if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
4979 			bpf_token_put(token);
4980 			token = NULL;
4981 		}
4982 	}
4983 
4984 	if (!bpf_token_capable(token, CAP_BPF)) {
4985 		bpf_token_put(token);
4986 		return -EPERM;
4987 	}
4988 
4989 	bpf_token_put(token);
4990 
4991 	return btf_new_fd(attr, uattr, uattr_size);
4992 }
4993 
4994 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4995 
4996 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4997 {
4998 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4999 		return -EINVAL;
5000 
5001 	if (!capable(CAP_SYS_ADMIN))
5002 		return -EPERM;
5003 
5004 	return btf_get_fd_by_id(attr->btf_id);
5005 }
5006 
5007 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
5008 				    union bpf_attr __user *uattr,
5009 				    u32 prog_id, u32 fd_type,
5010 				    const char *buf, u64 probe_offset,
5011 				    u64 probe_addr)
5012 {
5013 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
5014 	u32 len = buf ? strlen(buf) : 0, input_len;
5015 	int err = 0;
5016 
5017 	if (put_user(len, &uattr->task_fd_query.buf_len))
5018 		return -EFAULT;
5019 	input_len = attr->task_fd_query.buf_len;
5020 	if (input_len && ubuf) {
5021 		if (!len) {
5022 			/* nothing to copy, just make ubuf NULL terminated */
5023 			char zero = '\0';
5024 
5025 			if (put_user(zero, ubuf))
5026 				return -EFAULT;
5027 		} else if (input_len >= len + 1) {
5028 			/* ubuf can hold the string with NULL terminator */
5029 			if (copy_to_user(ubuf, buf, len + 1))
5030 				return -EFAULT;
5031 		} else {
5032 			/* ubuf cannot hold the string with NULL terminator,
5033 			 * do a partial copy with NULL terminator.
5034 			 */
5035 			char zero = '\0';
5036 
5037 			err = -ENOSPC;
5038 			if (copy_to_user(ubuf, buf, input_len - 1))
5039 				return -EFAULT;
5040 			if (put_user(zero, ubuf + input_len - 1))
5041 				return -EFAULT;
5042 		}
5043 	}
5044 
5045 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5046 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5047 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5048 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5049 		return -EFAULT;
5050 
5051 	return err;
5052 }
5053 
5054 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5055 
5056 static int bpf_task_fd_query(const union bpf_attr *attr,
5057 			     union bpf_attr __user *uattr)
5058 {
5059 	pid_t pid = attr->task_fd_query.pid;
5060 	u32 fd = attr->task_fd_query.fd;
5061 	const struct perf_event *event;
5062 	struct task_struct *task;
5063 	struct file *file;
5064 	int err;
5065 
5066 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5067 		return -EINVAL;
5068 
5069 	if (!capable(CAP_SYS_ADMIN))
5070 		return -EPERM;
5071 
5072 	if (attr->task_fd_query.flags != 0)
5073 		return -EINVAL;
5074 
5075 	rcu_read_lock();
5076 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5077 	rcu_read_unlock();
5078 	if (!task)
5079 		return -ENOENT;
5080 
5081 	err = 0;
5082 	file = fget_task(task, fd);
5083 	put_task_struct(task);
5084 	if (!file)
5085 		return -EBADF;
5086 
5087 	if (file->f_op == &bpf_link_fops) {
5088 		struct bpf_link *link = file->private_data;
5089 
5090 		if (link->ops == &bpf_raw_tp_link_lops) {
5091 			struct bpf_raw_tp_link *raw_tp =
5092 				container_of(link, struct bpf_raw_tp_link, link);
5093 			struct bpf_raw_event_map *btp = raw_tp->btp;
5094 
5095 			err = bpf_task_fd_query_copy(attr, uattr,
5096 						     raw_tp->link.prog->aux->id,
5097 						     BPF_FD_TYPE_RAW_TRACEPOINT,
5098 						     btp->tp->name, 0, 0);
5099 			goto put_file;
5100 		}
5101 		goto out_not_supp;
5102 	}
5103 
5104 	event = perf_get_event(file);
5105 	if (!IS_ERR(event)) {
5106 		u64 probe_offset, probe_addr;
5107 		u32 prog_id, fd_type;
5108 		const char *buf;
5109 
5110 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5111 					      &buf, &probe_offset,
5112 					      &probe_addr, NULL);
5113 		if (!err)
5114 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5115 						     fd_type, buf,
5116 						     probe_offset,
5117 						     probe_addr);
5118 		goto put_file;
5119 	}
5120 
5121 out_not_supp:
5122 	err = -ENOTSUPP;
5123 put_file:
5124 	fput(file);
5125 	return err;
5126 }
5127 
5128 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
5129 
5130 #define BPF_DO_BATCH(fn, ...)			\
5131 	do {					\
5132 		if (!fn) {			\
5133 			err = -ENOTSUPP;	\
5134 			goto err_put;		\
5135 		}				\
5136 		err = fn(__VA_ARGS__);		\
5137 	} while (0)
5138 
5139 static int bpf_map_do_batch(const union bpf_attr *attr,
5140 			    union bpf_attr __user *uattr,
5141 			    int cmd)
5142 {
5143 	bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
5144 			 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5145 	bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5146 	struct bpf_map *map;
5147 	int err, ufd;
5148 	struct fd f;
5149 
5150 	if (CHECK_ATTR(BPF_MAP_BATCH))
5151 		return -EINVAL;
5152 
5153 	ufd = attr->batch.map_fd;
5154 	f = fdget(ufd);
5155 	map = __bpf_map_get(f);
5156 	if (IS_ERR(map))
5157 		return PTR_ERR(map);
5158 	if (has_write)
5159 		bpf_map_write_active_inc(map);
5160 	if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5161 		err = -EPERM;
5162 		goto err_put;
5163 	}
5164 	if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5165 		err = -EPERM;
5166 		goto err_put;
5167 	}
5168 
5169 	if (cmd == BPF_MAP_LOOKUP_BATCH)
5170 		BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5171 	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5172 		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5173 	else if (cmd == BPF_MAP_UPDATE_BATCH)
5174 		BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
5175 	else
5176 		BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5177 err_put:
5178 	if (has_write) {
5179 		maybe_wait_bpf_programs(map);
5180 		bpf_map_write_active_dec(map);
5181 	}
5182 	fdput(f);
5183 	return err;
5184 }
5185 
5186 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5187 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5188 {
5189 	struct bpf_prog *prog;
5190 	int ret;
5191 
5192 	if (CHECK_ATTR(BPF_LINK_CREATE))
5193 		return -EINVAL;
5194 
5195 	if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5196 		return bpf_struct_ops_link_create(attr);
5197 
5198 	prog = bpf_prog_get(attr->link_create.prog_fd);
5199 	if (IS_ERR(prog))
5200 		return PTR_ERR(prog);
5201 
5202 	ret = bpf_prog_attach_check_attach_type(prog,
5203 						attr->link_create.attach_type);
5204 	if (ret)
5205 		goto out;
5206 
5207 	switch (prog->type) {
5208 	case BPF_PROG_TYPE_CGROUP_SKB:
5209 	case BPF_PROG_TYPE_CGROUP_SOCK:
5210 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5211 	case BPF_PROG_TYPE_SOCK_OPS:
5212 	case BPF_PROG_TYPE_CGROUP_DEVICE:
5213 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
5214 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5215 		ret = cgroup_bpf_link_attach(attr, prog);
5216 		break;
5217 	case BPF_PROG_TYPE_EXT:
5218 		ret = bpf_tracing_prog_attach(prog,
5219 					      attr->link_create.target_fd,
5220 					      attr->link_create.target_btf_id,
5221 					      attr->link_create.tracing.cookie);
5222 		break;
5223 	case BPF_PROG_TYPE_LSM:
5224 	case BPF_PROG_TYPE_TRACING:
5225 		if (attr->link_create.attach_type != prog->expected_attach_type) {
5226 			ret = -EINVAL;
5227 			goto out;
5228 		}
5229 		if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5230 			ret = bpf_raw_tp_link_attach(prog, NULL);
5231 		else if (prog->expected_attach_type == BPF_TRACE_ITER)
5232 			ret = bpf_iter_link_attach(attr, uattr, prog);
5233 		else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5234 			ret = cgroup_bpf_link_attach(attr, prog);
5235 		else
5236 			ret = bpf_tracing_prog_attach(prog,
5237 						      attr->link_create.target_fd,
5238 						      attr->link_create.target_btf_id,
5239 						      attr->link_create.tracing.cookie);
5240 		break;
5241 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
5242 	case BPF_PROG_TYPE_SK_LOOKUP:
5243 		ret = netns_bpf_link_create(attr, prog);
5244 		break;
5245 #ifdef CONFIG_NET
5246 	case BPF_PROG_TYPE_XDP:
5247 		ret = bpf_xdp_link_attach(attr, prog);
5248 		break;
5249 	case BPF_PROG_TYPE_SCHED_CLS:
5250 		if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5251 		    attr->link_create.attach_type == BPF_TCX_EGRESS)
5252 			ret = tcx_link_attach(attr, prog);
5253 		else
5254 			ret = netkit_link_attach(attr, prog);
5255 		break;
5256 	case BPF_PROG_TYPE_NETFILTER:
5257 		ret = bpf_nf_link_attach(attr, prog);
5258 		break;
5259 #endif
5260 	case BPF_PROG_TYPE_PERF_EVENT:
5261 	case BPF_PROG_TYPE_TRACEPOINT:
5262 		ret = bpf_perf_link_attach(attr, prog);
5263 		break;
5264 	case BPF_PROG_TYPE_KPROBE:
5265 		if (attr->link_create.attach_type == BPF_PERF_EVENT)
5266 			ret = bpf_perf_link_attach(attr, prog);
5267 		else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5268 			ret = bpf_kprobe_multi_link_attach(attr, prog);
5269 		else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5270 			ret = bpf_uprobe_multi_link_attach(attr, prog);
5271 		break;
5272 	default:
5273 		ret = -EINVAL;
5274 	}
5275 
5276 out:
5277 	if (ret < 0)
5278 		bpf_prog_put(prog);
5279 	return ret;
5280 }
5281 
5282 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5283 {
5284 	struct bpf_map *new_map, *old_map = NULL;
5285 	int ret;
5286 
5287 	new_map = bpf_map_get(attr->link_update.new_map_fd);
5288 	if (IS_ERR(new_map))
5289 		return PTR_ERR(new_map);
5290 
5291 	if (attr->link_update.flags & BPF_F_REPLACE) {
5292 		old_map = bpf_map_get(attr->link_update.old_map_fd);
5293 		if (IS_ERR(old_map)) {
5294 			ret = PTR_ERR(old_map);
5295 			goto out_put;
5296 		}
5297 	} else if (attr->link_update.old_map_fd) {
5298 		ret = -EINVAL;
5299 		goto out_put;
5300 	}
5301 
5302 	ret = link->ops->update_map(link, new_map, old_map);
5303 
5304 	if (old_map)
5305 		bpf_map_put(old_map);
5306 out_put:
5307 	bpf_map_put(new_map);
5308 	return ret;
5309 }
5310 
5311 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5312 
5313 static int link_update(union bpf_attr *attr)
5314 {
5315 	struct bpf_prog *old_prog = NULL, *new_prog;
5316 	struct bpf_link *link;
5317 	u32 flags;
5318 	int ret;
5319 
5320 	if (CHECK_ATTR(BPF_LINK_UPDATE))
5321 		return -EINVAL;
5322 
5323 	flags = attr->link_update.flags;
5324 	if (flags & ~BPF_F_REPLACE)
5325 		return -EINVAL;
5326 
5327 	link = bpf_link_get_from_fd(attr->link_update.link_fd);
5328 	if (IS_ERR(link))
5329 		return PTR_ERR(link);
5330 
5331 	if (link->ops->update_map) {
5332 		ret = link_update_map(link, attr);
5333 		goto out_put_link;
5334 	}
5335 
5336 	new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5337 	if (IS_ERR(new_prog)) {
5338 		ret = PTR_ERR(new_prog);
5339 		goto out_put_link;
5340 	}
5341 
5342 	if (flags & BPF_F_REPLACE) {
5343 		old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5344 		if (IS_ERR(old_prog)) {
5345 			ret = PTR_ERR(old_prog);
5346 			old_prog = NULL;
5347 			goto out_put_progs;
5348 		}
5349 	} else if (attr->link_update.old_prog_fd) {
5350 		ret = -EINVAL;
5351 		goto out_put_progs;
5352 	}
5353 
5354 	if (link->ops->update_prog)
5355 		ret = link->ops->update_prog(link, new_prog, old_prog);
5356 	else
5357 		ret = -EINVAL;
5358 
5359 out_put_progs:
5360 	if (old_prog)
5361 		bpf_prog_put(old_prog);
5362 	if (ret)
5363 		bpf_prog_put(new_prog);
5364 out_put_link:
5365 	bpf_link_put_direct(link);
5366 	return ret;
5367 }
5368 
5369 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5370 
5371 static int link_detach(union bpf_attr *attr)
5372 {
5373 	struct bpf_link *link;
5374 	int ret;
5375 
5376 	if (CHECK_ATTR(BPF_LINK_DETACH))
5377 		return -EINVAL;
5378 
5379 	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5380 	if (IS_ERR(link))
5381 		return PTR_ERR(link);
5382 
5383 	if (link->ops->detach)
5384 		ret = link->ops->detach(link);
5385 	else
5386 		ret = -EOPNOTSUPP;
5387 
5388 	bpf_link_put_direct(link);
5389 	return ret;
5390 }
5391 
5392 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5393 {
5394 	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5395 }
5396 
5397 struct bpf_link *bpf_link_by_id(u32 id)
5398 {
5399 	struct bpf_link *link;
5400 
5401 	if (!id)
5402 		return ERR_PTR(-ENOENT);
5403 
5404 	spin_lock_bh(&link_idr_lock);
5405 	/* before link is "settled", ID is 0, pretend it doesn't exist yet */
5406 	link = idr_find(&link_idr, id);
5407 	if (link) {
5408 		if (link->id)
5409 			link = bpf_link_inc_not_zero(link);
5410 		else
5411 			link = ERR_PTR(-EAGAIN);
5412 	} else {
5413 		link = ERR_PTR(-ENOENT);
5414 	}
5415 	spin_unlock_bh(&link_idr_lock);
5416 	return link;
5417 }
5418 
5419 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5420 {
5421 	struct bpf_link *link;
5422 
5423 	spin_lock_bh(&link_idr_lock);
5424 again:
5425 	link = idr_get_next(&link_idr, id);
5426 	if (link) {
5427 		link = bpf_link_inc_not_zero(link);
5428 		if (IS_ERR(link)) {
5429 			(*id)++;
5430 			goto again;
5431 		}
5432 	}
5433 	spin_unlock_bh(&link_idr_lock);
5434 
5435 	return link;
5436 }
5437 
5438 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5439 
5440 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5441 {
5442 	struct bpf_link *link;
5443 	u32 id = attr->link_id;
5444 	int fd;
5445 
5446 	if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5447 		return -EINVAL;
5448 
5449 	if (!capable(CAP_SYS_ADMIN))
5450 		return -EPERM;
5451 
5452 	link = bpf_link_by_id(id);
5453 	if (IS_ERR(link))
5454 		return PTR_ERR(link);
5455 
5456 	fd = bpf_link_new_fd(link);
5457 	if (fd < 0)
5458 		bpf_link_put_direct(link);
5459 
5460 	return fd;
5461 }
5462 
5463 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5464 
5465 static int bpf_stats_release(struct inode *inode, struct file *file)
5466 {
5467 	mutex_lock(&bpf_stats_enabled_mutex);
5468 	static_key_slow_dec(&bpf_stats_enabled_key.key);
5469 	mutex_unlock(&bpf_stats_enabled_mutex);
5470 	return 0;
5471 }
5472 
5473 static const struct file_operations bpf_stats_fops = {
5474 	.release = bpf_stats_release,
5475 };
5476 
5477 static int bpf_enable_runtime_stats(void)
5478 {
5479 	int fd;
5480 
5481 	mutex_lock(&bpf_stats_enabled_mutex);
5482 
5483 	/* Set a very high limit to avoid overflow */
5484 	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5485 		mutex_unlock(&bpf_stats_enabled_mutex);
5486 		return -EBUSY;
5487 	}
5488 
5489 	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5490 	if (fd >= 0)
5491 		static_key_slow_inc(&bpf_stats_enabled_key.key);
5492 
5493 	mutex_unlock(&bpf_stats_enabled_mutex);
5494 	return fd;
5495 }
5496 
5497 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5498 
5499 static int bpf_enable_stats(union bpf_attr *attr)
5500 {
5501 
5502 	if (CHECK_ATTR(BPF_ENABLE_STATS))
5503 		return -EINVAL;
5504 
5505 	if (!capable(CAP_SYS_ADMIN))
5506 		return -EPERM;
5507 
5508 	switch (attr->enable_stats.type) {
5509 	case BPF_STATS_RUN_TIME:
5510 		return bpf_enable_runtime_stats();
5511 	default:
5512 		break;
5513 	}
5514 	return -EINVAL;
5515 }
5516 
5517 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5518 
5519 static int bpf_iter_create(union bpf_attr *attr)
5520 {
5521 	struct bpf_link *link;
5522 	int err;
5523 
5524 	if (CHECK_ATTR(BPF_ITER_CREATE))
5525 		return -EINVAL;
5526 
5527 	if (attr->iter_create.flags)
5528 		return -EINVAL;
5529 
5530 	link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5531 	if (IS_ERR(link))
5532 		return PTR_ERR(link);
5533 
5534 	err = bpf_iter_new_fd(link);
5535 	bpf_link_put_direct(link);
5536 
5537 	return err;
5538 }
5539 
5540 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5541 
5542 static int bpf_prog_bind_map(union bpf_attr *attr)
5543 {
5544 	struct bpf_prog *prog;
5545 	struct bpf_map *map;
5546 	struct bpf_map **used_maps_old, **used_maps_new;
5547 	int i, ret = 0;
5548 
5549 	if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5550 		return -EINVAL;
5551 
5552 	if (attr->prog_bind_map.flags)
5553 		return -EINVAL;
5554 
5555 	prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5556 	if (IS_ERR(prog))
5557 		return PTR_ERR(prog);
5558 
5559 	map = bpf_map_get(attr->prog_bind_map.map_fd);
5560 	if (IS_ERR(map)) {
5561 		ret = PTR_ERR(map);
5562 		goto out_prog_put;
5563 	}
5564 
5565 	mutex_lock(&prog->aux->used_maps_mutex);
5566 
5567 	used_maps_old = prog->aux->used_maps;
5568 
5569 	for (i = 0; i < prog->aux->used_map_cnt; i++)
5570 		if (used_maps_old[i] == map) {
5571 			bpf_map_put(map);
5572 			goto out_unlock;
5573 		}
5574 
5575 	used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5576 				      sizeof(used_maps_new[0]),
5577 				      GFP_KERNEL);
5578 	if (!used_maps_new) {
5579 		ret = -ENOMEM;
5580 		goto out_unlock;
5581 	}
5582 
5583 	/* The bpf program will not access the bpf map, but for the sake of
5584 	 * simplicity, increase sleepable_refcnt for sleepable program as well.
5585 	 */
5586 	if (prog->sleepable)
5587 		atomic64_inc(&map->sleepable_refcnt);
5588 	memcpy(used_maps_new, used_maps_old,
5589 	       sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5590 	used_maps_new[prog->aux->used_map_cnt] = map;
5591 
5592 	prog->aux->used_map_cnt++;
5593 	prog->aux->used_maps = used_maps_new;
5594 
5595 	kfree(used_maps_old);
5596 
5597 out_unlock:
5598 	mutex_unlock(&prog->aux->used_maps_mutex);
5599 
5600 	if (ret)
5601 		bpf_map_put(map);
5602 out_prog_put:
5603 	bpf_prog_put(prog);
5604 	return ret;
5605 }
5606 
5607 #define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5608 
5609 static int token_create(union bpf_attr *attr)
5610 {
5611 	if (CHECK_ATTR(BPF_TOKEN_CREATE))
5612 		return -EINVAL;
5613 
5614 	/* no flags are supported yet */
5615 	if (attr->token_create.flags)
5616 		return -EINVAL;
5617 
5618 	return bpf_token_create(attr);
5619 }
5620 
5621 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5622 {
5623 	union bpf_attr attr;
5624 	int err;
5625 
5626 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5627 	if (err)
5628 		return err;
5629 	size = min_t(u32, size, sizeof(attr));
5630 
5631 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
5632 	memset(&attr, 0, sizeof(attr));
5633 	if (copy_from_bpfptr(&attr, uattr, size) != 0)
5634 		return -EFAULT;
5635 
5636 	err = security_bpf(cmd, &attr, size);
5637 	if (err < 0)
5638 		return err;
5639 
5640 	switch (cmd) {
5641 	case BPF_MAP_CREATE:
5642 		err = map_create(&attr);
5643 		break;
5644 	case BPF_MAP_LOOKUP_ELEM:
5645 		err = map_lookup_elem(&attr);
5646 		break;
5647 	case BPF_MAP_UPDATE_ELEM:
5648 		err = map_update_elem(&attr, uattr);
5649 		break;
5650 	case BPF_MAP_DELETE_ELEM:
5651 		err = map_delete_elem(&attr, uattr);
5652 		break;
5653 	case BPF_MAP_GET_NEXT_KEY:
5654 		err = map_get_next_key(&attr);
5655 		break;
5656 	case BPF_MAP_FREEZE:
5657 		err = map_freeze(&attr);
5658 		break;
5659 	case BPF_PROG_LOAD:
5660 		err = bpf_prog_load(&attr, uattr, size);
5661 		break;
5662 	case BPF_OBJ_PIN:
5663 		err = bpf_obj_pin(&attr);
5664 		break;
5665 	case BPF_OBJ_GET:
5666 		err = bpf_obj_get(&attr);
5667 		break;
5668 	case BPF_PROG_ATTACH:
5669 		err = bpf_prog_attach(&attr);
5670 		break;
5671 	case BPF_PROG_DETACH:
5672 		err = bpf_prog_detach(&attr);
5673 		break;
5674 	case BPF_PROG_QUERY:
5675 		err = bpf_prog_query(&attr, uattr.user);
5676 		break;
5677 	case BPF_PROG_TEST_RUN:
5678 		err = bpf_prog_test_run(&attr, uattr.user);
5679 		break;
5680 	case BPF_PROG_GET_NEXT_ID:
5681 		err = bpf_obj_get_next_id(&attr, uattr.user,
5682 					  &prog_idr, &prog_idr_lock);
5683 		break;
5684 	case BPF_MAP_GET_NEXT_ID:
5685 		err = bpf_obj_get_next_id(&attr, uattr.user,
5686 					  &map_idr, &map_idr_lock);
5687 		break;
5688 	case BPF_BTF_GET_NEXT_ID:
5689 		err = bpf_obj_get_next_id(&attr, uattr.user,
5690 					  &btf_idr, &btf_idr_lock);
5691 		break;
5692 	case BPF_PROG_GET_FD_BY_ID:
5693 		err = bpf_prog_get_fd_by_id(&attr);
5694 		break;
5695 	case BPF_MAP_GET_FD_BY_ID:
5696 		err = bpf_map_get_fd_by_id(&attr);
5697 		break;
5698 	case BPF_OBJ_GET_INFO_BY_FD:
5699 		err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5700 		break;
5701 	case BPF_RAW_TRACEPOINT_OPEN:
5702 		err = bpf_raw_tracepoint_open(&attr);
5703 		break;
5704 	case BPF_BTF_LOAD:
5705 		err = bpf_btf_load(&attr, uattr, size);
5706 		break;
5707 	case BPF_BTF_GET_FD_BY_ID:
5708 		err = bpf_btf_get_fd_by_id(&attr);
5709 		break;
5710 	case BPF_TASK_FD_QUERY:
5711 		err = bpf_task_fd_query(&attr, uattr.user);
5712 		break;
5713 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5714 		err = map_lookup_and_delete_elem(&attr);
5715 		break;
5716 	case BPF_MAP_LOOKUP_BATCH:
5717 		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5718 		break;
5719 	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5720 		err = bpf_map_do_batch(&attr, uattr.user,
5721 				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5722 		break;
5723 	case BPF_MAP_UPDATE_BATCH:
5724 		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5725 		break;
5726 	case BPF_MAP_DELETE_BATCH:
5727 		err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5728 		break;
5729 	case BPF_LINK_CREATE:
5730 		err = link_create(&attr, uattr);
5731 		break;
5732 	case BPF_LINK_UPDATE:
5733 		err = link_update(&attr);
5734 		break;
5735 	case BPF_LINK_GET_FD_BY_ID:
5736 		err = bpf_link_get_fd_by_id(&attr);
5737 		break;
5738 	case BPF_LINK_GET_NEXT_ID:
5739 		err = bpf_obj_get_next_id(&attr, uattr.user,
5740 					  &link_idr, &link_idr_lock);
5741 		break;
5742 	case BPF_ENABLE_STATS:
5743 		err = bpf_enable_stats(&attr);
5744 		break;
5745 	case BPF_ITER_CREATE:
5746 		err = bpf_iter_create(&attr);
5747 		break;
5748 	case BPF_LINK_DETACH:
5749 		err = link_detach(&attr);
5750 		break;
5751 	case BPF_PROG_BIND_MAP:
5752 		err = bpf_prog_bind_map(&attr);
5753 		break;
5754 	case BPF_TOKEN_CREATE:
5755 		err = token_create(&attr);
5756 		break;
5757 	default:
5758 		err = -EINVAL;
5759 		break;
5760 	}
5761 
5762 	return err;
5763 }
5764 
5765 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5766 {
5767 	return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5768 }
5769 
5770 static bool syscall_prog_is_valid_access(int off, int size,
5771 					 enum bpf_access_type type,
5772 					 const struct bpf_prog *prog,
5773 					 struct bpf_insn_access_aux *info)
5774 {
5775 	if (off < 0 || off >= U16_MAX)
5776 		return false;
5777 	if (off % size != 0)
5778 		return false;
5779 	return true;
5780 }
5781 
5782 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5783 {
5784 	switch (cmd) {
5785 	case BPF_MAP_CREATE:
5786 	case BPF_MAP_DELETE_ELEM:
5787 	case BPF_MAP_UPDATE_ELEM:
5788 	case BPF_MAP_FREEZE:
5789 	case BPF_MAP_GET_FD_BY_ID:
5790 	case BPF_PROG_LOAD:
5791 	case BPF_BTF_LOAD:
5792 	case BPF_LINK_CREATE:
5793 	case BPF_RAW_TRACEPOINT_OPEN:
5794 		break;
5795 	default:
5796 		return -EINVAL;
5797 	}
5798 	return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5799 }
5800 
5801 
5802 /* To shut up -Wmissing-prototypes.
5803  * This function is used by the kernel light skeleton
5804  * to load bpf programs when modules are loaded or during kernel boot.
5805  * See tools/lib/bpf/skel_internal.h
5806  */
5807 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5808 
5809 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5810 {
5811 	struct bpf_prog * __maybe_unused prog;
5812 	struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5813 
5814 	switch (cmd) {
5815 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5816 	case BPF_PROG_TEST_RUN:
5817 		if (attr->test.data_in || attr->test.data_out ||
5818 		    attr->test.ctx_out || attr->test.duration ||
5819 		    attr->test.repeat || attr->test.flags)
5820 			return -EINVAL;
5821 
5822 		prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5823 		if (IS_ERR(prog))
5824 			return PTR_ERR(prog);
5825 
5826 		if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5827 		    attr->test.ctx_size_in > U16_MAX) {
5828 			bpf_prog_put(prog);
5829 			return -EINVAL;
5830 		}
5831 
5832 		run_ctx.bpf_cookie = 0;
5833 		if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5834 			/* recursion detected */
5835 			__bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5836 			bpf_prog_put(prog);
5837 			return -EBUSY;
5838 		}
5839 		attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5840 		__bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5841 						&run_ctx);
5842 		bpf_prog_put(prog);
5843 		return 0;
5844 #endif
5845 	default:
5846 		return ____bpf_sys_bpf(cmd, attr, size);
5847 	}
5848 }
5849 EXPORT_SYMBOL(kern_sys_bpf);
5850 
5851 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5852 	.func		= bpf_sys_bpf,
5853 	.gpl_only	= false,
5854 	.ret_type	= RET_INTEGER,
5855 	.arg1_type	= ARG_ANYTHING,
5856 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
5857 	.arg3_type	= ARG_CONST_SIZE,
5858 };
5859 
5860 const struct bpf_func_proto * __weak
5861 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5862 {
5863 	return bpf_base_func_proto(func_id, prog);
5864 }
5865 
5866 BPF_CALL_1(bpf_sys_close, u32, fd)
5867 {
5868 	/* When bpf program calls this helper there should not be
5869 	 * an fdget() without matching completed fdput().
5870 	 * This helper is allowed in the following callchain only:
5871 	 * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5872 	 */
5873 	return close_fd(fd);
5874 }
5875 
5876 static const struct bpf_func_proto bpf_sys_close_proto = {
5877 	.func		= bpf_sys_close,
5878 	.gpl_only	= false,
5879 	.ret_type	= RET_INTEGER,
5880 	.arg1_type	= ARG_ANYTHING,
5881 };
5882 
5883 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5884 {
5885 	if (flags)
5886 		return -EINVAL;
5887 
5888 	if (name_sz <= 1 || name[name_sz - 1])
5889 		return -EINVAL;
5890 
5891 	if (!bpf_dump_raw_ok(current_cred()))
5892 		return -EPERM;
5893 
5894 	*res = kallsyms_lookup_name(name);
5895 	return *res ? 0 : -ENOENT;
5896 }
5897 
5898 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5899 	.func		= bpf_kallsyms_lookup_name,
5900 	.gpl_only	= false,
5901 	.ret_type	= RET_INTEGER,
5902 	.arg1_type	= ARG_PTR_TO_MEM,
5903 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
5904 	.arg3_type	= ARG_ANYTHING,
5905 	.arg4_type	= ARG_PTR_TO_LONG,
5906 };
5907 
5908 static const struct bpf_func_proto *
5909 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5910 {
5911 	switch (func_id) {
5912 	case BPF_FUNC_sys_bpf:
5913 		return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
5914 		       ? NULL : &bpf_sys_bpf_proto;
5915 	case BPF_FUNC_btf_find_by_name_kind:
5916 		return &bpf_btf_find_by_name_kind_proto;
5917 	case BPF_FUNC_sys_close:
5918 		return &bpf_sys_close_proto;
5919 	case BPF_FUNC_kallsyms_lookup_name:
5920 		return &bpf_kallsyms_lookup_name_proto;
5921 	default:
5922 		return tracing_prog_func_proto(func_id, prog);
5923 	}
5924 }
5925 
5926 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5927 	.get_func_proto  = syscall_prog_func_proto,
5928 	.is_valid_access = syscall_prog_is_valid_access,
5929 };
5930 
5931 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5932 	.test_run = bpf_prog_test_run_syscall,
5933 };
5934 
5935 #ifdef CONFIG_SYSCTL
5936 static int bpf_stats_handler(struct ctl_table *table, int write,
5937 			     void *buffer, size_t *lenp, loff_t *ppos)
5938 {
5939 	struct static_key *key = (struct static_key *)table->data;
5940 	static int saved_val;
5941 	int val, ret;
5942 	struct ctl_table tmp = {
5943 		.data   = &val,
5944 		.maxlen = sizeof(val),
5945 		.mode   = table->mode,
5946 		.extra1 = SYSCTL_ZERO,
5947 		.extra2 = SYSCTL_ONE,
5948 	};
5949 
5950 	if (write && !capable(CAP_SYS_ADMIN))
5951 		return -EPERM;
5952 
5953 	mutex_lock(&bpf_stats_enabled_mutex);
5954 	val = saved_val;
5955 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5956 	if (write && !ret && val != saved_val) {
5957 		if (val)
5958 			static_key_slow_inc(key);
5959 		else
5960 			static_key_slow_dec(key);
5961 		saved_val = val;
5962 	}
5963 	mutex_unlock(&bpf_stats_enabled_mutex);
5964 	return ret;
5965 }
5966 
5967 void __weak unpriv_ebpf_notify(int new_state)
5968 {
5969 }
5970 
5971 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5972 			      void *buffer, size_t *lenp, loff_t *ppos)
5973 {
5974 	int ret, unpriv_enable = *(int *)table->data;
5975 	bool locked_state = unpriv_enable == 1;
5976 	struct ctl_table tmp = *table;
5977 
5978 	if (write && !capable(CAP_SYS_ADMIN))
5979 		return -EPERM;
5980 
5981 	tmp.data = &unpriv_enable;
5982 	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5983 	if (write && !ret) {
5984 		if (locked_state && unpriv_enable != 1)
5985 			return -EPERM;
5986 		*(int *)table->data = unpriv_enable;
5987 	}
5988 
5989 	if (write)
5990 		unpriv_ebpf_notify(unpriv_enable);
5991 
5992 	return ret;
5993 }
5994 
5995 static struct ctl_table bpf_syscall_table[] = {
5996 	{
5997 		.procname	= "unprivileged_bpf_disabled",
5998 		.data		= &sysctl_unprivileged_bpf_disabled,
5999 		.maxlen		= sizeof(sysctl_unprivileged_bpf_disabled),
6000 		.mode		= 0644,
6001 		.proc_handler	= bpf_unpriv_handler,
6002 		.extra1		= SYSCTL_ZERO,
6003 		.extra2		= SYSCTL_TWO,
6004 	},
6005 	{
6006 		.procname	= "bpf_stats_enabled",
6007 		.data		= &bpf_stats_enabled_key.key,
6008 		.mode		= 0644,
6009 		.proc_handler	= bpf_stats_handler,
6010 	},
6011 	{ }
6012 };
6013 
6014 static int __init bpf_syscall_sysctl_init(void)
6015 {
6016 	register_sysctl_init("kernel", bpf_syscall_table);
6017 	return 0;
6018 }
6019 late_initcall(bpf_syscall_sysctl_init);
6020 #endif /* CONFIG_SYSCTL */
6021