• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
3  */
4 #include <linux/bpf.h>
5 #include <linux/bpf_trace.h>
6 #include <linux/bpf_lirc.h>
7 #include <linux/bpf_verifier.h>
8 #include <linux/btf.h>
9 #include <linux/syscalls.h>
10 #include <linux/slab.h>
11 #include <linux/sched/signal.h>
12 #include <linux/vmalloc.h>
13 #include <linux/mmzone.h>
14 #include <linux/anon_inodes.h>
15 #include <linux/fdtable.h>
16 #include <linux/file.h>
17 #include <linux/fs.h>
18 #include <linux/license.h>
19 #include <linux/filter.h>
20 #include <linux/version.h>
21 #include <linux/kernel.h>
22 #include <linux/idr.h>
23 #include <linux/cred.h>
24 #include <linux/timekeeping.h>
25 #include <linux/ctype.h>
26 #include <linux/nospec.h>
27 #include <linux/audit.h>
28 #include <uapi/linux/btf.h>
29 #include <linux/pgtable.h>
30 #include <linux/bpf_lsm.h>
31 #include <linux/poll.h>
32 #include <linux/bpf-netns.h>
33 #include <linux/rcupdate_trace.h>
34 
35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
36 			  (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
37 			  (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
38 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
41 			IS_FD_HASH(map))
42 
43 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
44 
45 DEFINE_PER_CPU(int, bpf_prog_active);
46 static DEFINE_IDR(prog_idr);
47 static DEFINE_SPINLOCK(prog_idr_lock);
48 static DEFINE_IDR(map_idr);
49 static DEFINE_SPINLOCK(map_idr_lock);
50 static DEFINE_IDR(link_idr);
51 static DEFINE_SPINLOCK(link_idr_lock);
52 
53 int sysctl_unprivileged_bpf_disabled __read_mostly =
54 	IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
55 
56 static const struct bpf_map_ops * const bpf_map_types[] = {
57 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
58 #define BPF_MAP_TYPE(_id, _ops) \
59 	[_id] = &_ops,
60 #define BPF_LINK_TYPE(_id, _name)
61 #include <linux/bpf_types.h>
62 #undef BPF_PROG_TYPE
63 #undef BPF_MAP_TYPE
64 #undef BPF_LINK_TYPE
65 };
66 
67 /*
68  * If we're handed a bigger struct than we know of, ensure all the unknown bits
69  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
70  * we don't know about yet.
71  *
72  * There is a ToCToU between this function call and the following
73  * copy_from_user() call. However, this is not a concern since this function is
74  * meant to be a future-proofing of bits.
75  */
bpf_check_uarg_tail_zero(void __user * uaddr,size_t expected_size,size_t actual_size)76 int bpf_check_uarg_tail_zero(void __user *uaddr,
77 			     size_t expected_size,
78 			     size_t actual_size)
79 {
80 	unsigned char __user *addr = uaddr + expected_size;
81 	int res;
82 
83 	if (unlikely(actual_size > PAGE_SIZE))	/* silly large */
84 		return -E2BIG;
85 
86 	if (actual_size <= expected_size)
87 		return 0;
88 
89 	res = check_zeroed_user(addr, actual_size - expected_size);
90 	if (res < 0)
91 		return res;
92 	return res ? 0 : -E2BIG;
93 }
94 
95 const struct bpf_map_ops bpf_map_offload_ops = {
96 	.map_meta_equal = bpf_map_meta_equal,
97 	.map_alloc = bpf_map_offload_map_alloc,
98 	.map_free = bpf_map_offload_map_free,
99 	.map_check_btf = map_check_no_btf,
100 };
101 
find_and_alloc_map(union bpf_attr * attr)102 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
103 {
104 	const struct bpf_map_ops *ops;
105 	u32 type = attr->map_type;
106 	struct bpf_map *map;
107 	int err;
108 
109 	if (type >= ARRAY_SIZE(bpf_map_types))
110 		return ERR_PTR(-EINVAL);
111 	type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
112 	ops = bpf_map_types[type];
113 	if (!ops)
114 		return ERR_PTR(-EINVAL);
115 
116 	if (ops->map_alloc_check) {
117 		err = ops->map_alloc_check(attr);
118 		if (err)
119 			return ERR_PTR(err);
120 	}
121 	if (attr->map_ifindex)
122 		ops = &bpf_map_offload_ops;
123 	map = ops->map_alloc(attr);
124 	if (IS_ERR(map))
125 		return map;
126 	map->ops = ops;
127 	map->map_type = type;
128 	return map;
129 }
130 
bpf_map_write_active_inc(struct bpf_map * map)131 static void bpf_map_write_active_inc(struct bpf_map *map)
132 {
133 	atomic64_inc(&map->writecnt);
134 }
135 
bpf_map_write_active_dec(struct bpf_map * map)136 static void bpf_map_write_active_dec(struct bpf_map *map)
137 {
138 	atomic64_dec(&map->writecnt);
139 }
140 
bpf_map_write_active(const struct bpf_map * map)141 bool bpf_map_write_active(const struct bpf_map *map)
142 {
143 	return atomic64_read(&map->writecnt) != 0;
144 }
145 
bpf_map_value_size(struct bpf_map * map)146 static u32 bpf_map_value_size(struct bpf_map *map)
147 {
148 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
149 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
150 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
151 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
152 		return round_up(map->value_size, 8) * num_possible_cpus();
153 	else if (IS_FD_MAP(map))
154 		return sizeof(u32);
155 	else
156 		return  map->value_size;
157 }
158 
maybe_wait_bpf_programs(struct bpf_map * map)159 static void maybe_wait_bpf_programs(struct bpf_map *map)
160 {
161 	/* Wait for any running BPF programs to complete so that
162 	 * userspace, when we return to it, knows that all programs
163 	 * that could be running use the new map value.
164 	 */
165 	if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
166 	    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
167 		synchronize_rcu();
168 }
169 
bpf_map_update_value(struct bpf_map * map,struct fd f,void * key,void * value,__u64 flags)170 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
171 				void *value, __u64 flags)
172 {
173 	int err;
174 
175 	/* Need to create a kthread, thus must support schedule */
176 	if (bpf_map_is_dev_bound(map)) {
177 		return bpf_map_offload_update_elem(map, key, value, flags);
178 	} else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
179 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
180 		return map->ops->map_update_elem(map, key, value, flags);
181 	} else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
182 		   map->map_type == BPF_MAP_TYPE_SOCKMAP) {
183 		return sock_map_update_elem_sys(map, key, value, flags);
184 	} else if (IS_FD_PROG_ARRAY(map)) {
185 		return bpf_fd_array_map_update_elem(map, f.file, key, value,
186 						    flags);
187 	}
188 
189 	bpf_disable_instrumentation();
190 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
191 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
192 		err = bpf_percpu_hash_update(map, key, value, flags);
193 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
194 		err = bpf_percpu_array_update(map, key, value, flags);
195 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
196 		err = bpf_percpu_cgroup_storage_update(map, key, value,
197 						       flags);
198 	} else if (IS_FD_ARRAY(map)) {
199 		rcu_read_lock();
200 		err = bpf_fd_array_map_update_elem(map, f.file, key, value,
201 						   flags);
202 		rcu_read_unlock();
203 	} else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
204 		rcu_read_lock();
205 		err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
206 						  flags);
207 		rcu_read_unlock();
208 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
209 		/* rcu_read_lock() is not needed */
210 		err = bpf_fd_reuseport_array_update_elem(map, key, value,
211 							 flags);
212 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
213 		   map->map_type == BPF_MAP_TYPE_STACK) {
214 		err = map->ops->map_push_elem(map, value, flags);
215 	} else {
216 		rcu_read_lock();
217 		err = map->ops->map_update_elem(map, key, value, flags);
218 		rcu_read_unlock();
219 	}
220 	bpf_enable_instrumentation();
221 	maybe_wait_bpf_programs(map);
222 
223 	return err;
224 }
225 
bpf_map_copy_value(struct bpf_map * map,void * key,void * value,__u64 flags)226 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
227 			      __u64 flags)
228 {
229 	void *ptr;
230 	int err;
231 
232 	if (bpf_map_is_dev_bound(map))
233 		return bpf_map_offload_lookup_elem(map, key, value);
234 
235 	bpf_disable_instrumentation();
236 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
237 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
238 		err = bpf_percpu_hash_copy(map, key, value);
239 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
240 		err = bpf_percpu_array_copy(map, key, value);
241 	} else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
242 		err = bpf_percpu_cgroup_storage_copy(map, key, value);
243 	} else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
244 		err = bpf_stackmap_copy(map, key, value);
245 	} else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
246 		err = bpf_fd_array_map_lookup_elem(map, key, value);
247 	} else if (IS_FD_HASH(map)) {
248 		err = bpf_fd_htab_map_lookup_elem(map, key, value);
249 	} else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
250 		err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
251 	} else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
252 		   map->map_type == BPF_MAP_TYPE_STACK) {
253 		err = map->ops->map_peek_elem(map, value);
254 	} else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
255 		/* struct_ops map requires directly updating "value" */
256 		err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
257 	} else {
258 		rcu_read_lock();
259 		if (map->ops->map_lookup_elem_sys_only)
260 			ptr = map->ops->map_lookup_elem_sys_only(map, key);
261 		else
262 			ptr = map->ops->map_lookup_elem(map, key);
263 		if (IS_ERR(ptr)) {
264 			err = PTR_ERR(ptr);
265 		} else if (!ptr) {
266 			err = -ENOENT;
267 		} else {
268 			err = 0;
269 			if (flags & BPF_F_LOCK)
270 				/* lock 'ptr' and copy everything but lock */
271 				copy_map_value_locked(map, value, ptr, true);
272 			else
273 				copy_map_value(map, value, ptr);
274 			/* mask lock, since value wasn't zero inited */
275 			check_and_init_map_lock(map, value);
276 		}
277 		rcu_read_unlock();
278 	}
279 
280 	bpf_enable_instrumentation();
281 	maybe_wait_bpf_programs(map);
282 
283 	return err;
284 }
285 
__bpf_map_area_alloc(u64 size,int numa_node,bool mmapable)286 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
287 {
288 	/* We really just want to fail instead of triggering OOM killer
289 	 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
290 	 * which is used for lower order allocation requests.
291 	 *
292 	 * It has been observed that higher order allocation requests done by
293 	 * vmalloc with __GFP_NORETRY being set might fail due to not trying
294 	 * to reclaim memory from the page cache, thus we set
295 	 * __GFP_RETRY_MAYFAIL to avoid such situations.
296 	 */
297 
298 	const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO;
299 	unsigned int flags = 0;
300 	unsigned long align = 1;
301 	void *area;
302 
303 	if (size >= SIZE_MAX)
304 		return NULL;
305 
306 	/* kmalloc()'ed memory can't be mmap()'ed */
307 	if (mmapable) {
308 		BUG_ON(!PAGE_ALIGNED(size));
309 		align = SHMLBA;
310 		flags = VM_USERMAP;
311 	} else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
312 		area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
313 				    numa_node);
314 		if (area != NULL)
315 			return area;
316 	}
317 
318 	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
319 			gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
320 			flags, numa_node, __builtin_return_address(0));
321 }
322 
bpf_map_area_alloc(u64 size,int numa_node)323 void *bpf_map_area_alloc(u64 size, int numa_node)
324 {
325 	return __bpf_map_area_alloc(size, numa_node, false);
326 }
327 
bpf_map_area_mmapable_alloc(u64 size,int numa_node)328 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
329 {
330 	return __bpf_map_area_alloc(size, numa_node, true);
331 }
332 
bpf_map_area_free(void * area)333 void bpf_map_area_free(void *area)
334 {
335 	kvfree(area);
336 }
337 
bpf_map_flags_retain_permanent(u32 flags)338 static u32 bpf_map_flags_retain_permanent(u32 flags)
339 {
340 	/* Some map creation flags are not tied to the map object but
341 	 * rather to the map fd instead, so they have no meaning upon
342 	 * map object inspection since multiple file descriptors with
343 	 * different (access) properties can exist here. Thus, given
344 	 * this has zero meaning for the map itself, lets clear these
345 	 * from here.
346 	 */
347 	return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
348 }
349 
bpf_map_init_from_attr(struct bpf_map * map,union bpf_attr * attr)350 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
351 {
352 	map->map_type = attr->map_type;
353 	map->key_size = attr->key_size;
354 	map->value_size = attr->value_size;
355 	map->max_entries = attr->max_entries;
356 	map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
357 	map->numa_node = bpf_map_attr_numa_node(attr);
358 }
359 
bpf_charge_memlock(struct user_struct * user,u32 pages)360 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
361 {
362 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
363 
364 	if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
365 		atomic_long_sub(pages, &user->locked_vm);
366 		return -EPERM;
367 	}
368 	return 0;
369 }
370 
bpf_uncharge_memlock(struct user_struct * user,u32 pages)371 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
372 {
373 	if (user)
374 		atomic_long_sub(pages, &user->locked_vm);
375 }
376 
bpf_map_charge_init(struct bpf_map_memory * mem,u64 size)377 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
378 {
379 	u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
380 	struct user_struct *user;
381 	int ret;
382 
383 	if (size >= U32_MAX - PAGE_SIZE)
384 		return -E2BIG;
385 
386 	user = get_current_user();
387 	ret = bpf_charge_memlock(user, pages);
388 	if (ret) {
389 		free_uid(user);
390 		return ret;
391 	}
392 
393 	mem->pages = pages;
394 	mem->user = user;
395 
396 	return 0;
397 }
398 
bpf_map_charge_finish(struct bpf_map_memory * mem)399 void bpf_map_charge_finish(struct bpf_map_memory *mem)
400 {
401 	bpf_uncharge_memlock(mem->user, mem->pages);
402 	free_uid(mem->user);
403 }
404 
bpf_map_charge_move(struct bpf_map_memory * dst,struct bpf_map_memory * src)405 void bpf_map_charge_move(struct bpf_map_memory *dst,
406 			 struct bpf_map_memory *src)
407 {
408 	*dst = *src;
409 
410 	/* Make sure src will not be used for the redundant uncharging. */
411 	memset(src, 0, sizeof(struct bpf_map_memory));
412 }
413 
bpf_map_charge_memlock(struct bpf_map * map,u32 pages)414 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
415 {
416 	int ret;
417 
418 	ret = bpf_charge_memlock(map->memory.user, pages);
419 	if (ret)
420 		return ret;
421 	map->memory.pages += pages;
422 	return ret;
423 }
424 
bpf_map_uncharge_memlock(struct bpf_map * map,u32 pages)425 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
426 {
427 	bpf_uncharge_memlock(map->memory.user, pages);
428 	map->memory.pages -= pages;
429 }
430 
bpf_map_alloc_id(struct bpf_map * map)431 static int bpf_map_alloc_id(struct bpf_map *map)
432 {
433 	int id;
434 
435 	idr_preload(GFP_KERNEL);
436 	spin_lock_bh(&map_idr_lock);
437 	id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
438 	if (id > 0)
439 		map->id = id;
440 	spin_unlock_bh(&map_idr_lock);
441 	idr_preload_end();
442 
443 	if (WARN_ON_ONCE(!id))
444 		return -ENOSPC;
445 
446 	return id > 0 ? 0 : id;
447 }
448 
bpf_map_free_id(struct bpf_map * map,bool do_idr_lock)449 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
450 {
451 	unsigned long flags;
452 
453 	/* Offloaded maps are removed from the IDR store when their device
454 	 * disappears - even if someone holds an fd to them they are unusable,
455 	 * the memory is gone, all ops will fail; they are simply waiting for
456 	 * refcnt to drop to be freed.
457 	 */
458 	if (!map->id)
459 		return;
460 
461 	if (do_idr_lock)
462 		spin_lock_irqsave(&map_idr_lock, flags);
463 	else
464 		__acquire(&map_idr_lock);
465 
466 	idr_remove(&map_idr, map->id);
467 	map->id = 0;
468 
469 	if (do_idr_lock)
470 		spin_unlock_irqrestore(&map_idr_lock, flags);
471 	else
472 		__release(&map_idr_lock);
473 }
474 
475 /* called from workqueue */
bpf_map_free_deferred(struct work_struct * work)476 static void bpf_map_free_deferred(struct work_struct *work)
477 {
478 	struct bpf_map *map = container_of(work, struct bpf_map, work);
479 	struct bpf_map_memory mem;
480 
481 	bpf_map_charge_move(&mem, &map->memory);
482 	security_bpf_map_free(map);
483 	/* implementation dependent freeing */
484 	map->ops->map_free(map);
485 	bpf_map_charge_finish(&mem);
486 }
487 
bpf_map_put_uref(struct bpf_map * map)488 static void bpf_map_put_uref(struct bpf_map *map)
489 {
490 	if (atomic64_dec_and_test(&map->usercnt)) {
491 		if (map->ops->map_release_uref)
492 			map->ops->map_release_uref(map);
493 	}
494 }
495 
496 /* decrement map refcnt and schedule it for freeing via workqueue
497  * (unrelying map implementation ops->map_free() might sleep)
498  */
__bpf_map_put(struct bpf_map * map,bool do_idr_lock)499 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
500 {
501 	if (atomic64_dec_and_test(&map->refcnt)) {
502 		/* bpf_map_free_id() must be called first */
503 		bpf_map_free_id(map, do_idr_lock);
504 		btf_put(map->btf);
505 		INIT_WORK(&map->work, bpf_map_free_deferred);
506 		schedule_work(&map->work);
507 	}
508 }
509 
bpf_map_put(struct bpf_map * map)510 void bpf_map_put(struct bpf_map *map)
511 {
512 	__bpf_map_put(map, true);
513 }
514 EXPORT_SYMBOL_GPL(bpf_map_put);
515 
bpf_map_put_with_uref(struct bpf_map * map)516 void bpf_map_put_with_uref(struct bpf_map *map)
517 {
518 	bpf_map_put_uref(map);
519 	bpf_map_put(map);
520 }
521 
bpf_map_release(struct inode * inode,struct file * filp)522 static int bpf_map_release(struct inode *inode, struct file *filp)
523 {
524 	struct bpf_map *map = filp->private_data;
525 
526 	if (map->ops->map_release)
527 		map->ops->map_release(map, filp);
528 
529 	bpf_map_put_with_uref(map);
530 	return 0;
531 }
532 
map_get_sys_perms(struct bpf_map * map,struct fd f)533 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
534 {
535 	fmode_t mode = f.file->f_mode;
536 
537 	/* Our file permissions may have been overridden by global
538 	 * map permissions facing syscall side.
539 	 */
540 	if (READ_ONCE(map->frozen))
541 		mode &= ~FMODE_CAN_WRITE;
542 	return mode;
543 }
544 
545 #ifdef CONFIG_PROC_FS
bpf_map_show_fdinfo(struct seq_file * m,struct file * filp)546 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
547 {
548 	const struct bpf_map *map = filp->private_data;
549 	const struct bpf_array *array;
550 	u32 type = 0, jited = 0;
551 
552 	if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
553 		array = container_of(map, struct bpf_array, map);
554 		spin_lock(&array->aux->owner.lock);
555 		type  = array->aux->owner.type;
556 		jited = array->aux->owner.jited;
557 		spin_unlock(&array->aux->owner.lock);
558 	}
559 
560 	seq_printf(m,
561 		   "map_type:\t%u\n"
562 		   "key_size:\t%u\n"
563 		   "value_size:\t%u\n"
564 		   "max_entries:\t%u\n"
565 		   "map_flags:\t%#x\n"
566 		   "memlock:\t%llu\n"
567 		   "map_id:\t%u\n"
568 		   "frozen:\t%u\n",
569 		   map->map_type,
570 		   map->key_size,
571 		   map->value_size,
572 		   map->max_entries,
573 		   map->map_flags,
574 		   map->memory.pages * 1ULL << PAGE_SHIFT,
575 		   map->id,
576 		   READ_ONCE(map->frozen));
577 	if (type) {
578 		seq_printf(m, "owner_prog_type:\t%u\n", type);
579 		seq_printf(m, "owner_jited:\t%u\n", jited);
580 	}
581 }
582 #endif
583 
bpf_dummy_read(struct file * filp,char __user * buf,size_t siz,loff_t * ppos)584 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
585 			      loff_t *ppos)
586 {
587 	/* We need this handler such that alloc_file() enables
588 	 * f_mode with FMODE_CAN_READ.
589 	 */
590 	return -EINVAL;
591 }
592 
bpf_dummy_write(struct file * filp,const char __user * buf,size_t siz,loff_t * ppos)593 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
594 			       size_t siz, loff_t *ppos)
595 {
596 	/* We need this handler such that alloc_file() enables
597 	 * f_mode with FMODE_CAN_WRITE.
598 	 */
599 	return -EINVAL;
600 }
601 
602 /* called for any extra memory-mapped regions (except initial) */
bpf_map_mmap_open(struct vm_area_struct * vma)603 static void bpf_map_mmap_open(struct vm_area_struct *vma)
604 {
605 	struct bpf_map *map = vma->vm_file->private_data;
606 
607 	if (vma->vm_flags & VM_MAYWRITE)
608 		bpf_map_write_active_inc(map);
609 }
610 
611 /* called for all unmapped memory region (including initial) */
bpf_map_mmap_close(struct vm_area_struct * vma)612 static void bpf_map_mmap_close(struct vm_area_struct *vma)
613 {
614 	struct bpf_map *map = vma->vm_file->private_data;
615 
616 	if (vma->vm_flags & VM_MAYWRITE)
617 		bpf_map_write_active_dec(map);
618 }
619 
620 static const struct vm_operations_struct bpf_map_default_vmops = {
621 	.open		= bpf_map_mmap_open,
622 	.close		= bpf_map_mmap_close,
623 };
624 
bpf_map_mmap(struct file * filp,struct vm_area_struct * vma)625 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
626 {
627 	struct bpf_map *map = filp->private_data;
628 	int err;
629 
630 	if (!map->ops->map_mmap || map_value_has_spin_lock(map))
631 		return -ENOTSUPP;
632 
633 	if (!(vma->vm_flags & VM_SHARED))
634 		return -EINVAL;
635 
636 	mutex_lock(&map->freeze_mutex);
637 
638 	if (vma->vm_flags & VM_WRITE) {
639 		if (map->frozen) {
640 			err = -EPERM;
641 			goto out;
642 		}
643 		/* map is meant to be read-only, so do not allow mapping as
644 		 * writable, because it's possible to leak a writable page
645 		 * reference and allows user-space to still modify it after
646 		 * freezing, while verifier will assume contents do not change
647 		 */
648 		if (map->map_flags & BPF_F_RDONLY_PROG) {
649 			err = -EACCES;
650 			goto out;
651 		}
652 	}
653 
654 	/* set default open/close callbacks */
655 	vma->vm_ops = &bpf_map_default_vmops;
656 	vma->vm_private_data = map;
657 	vma->vm_flags &= ~VM_MAYEXEC;
658 	if (!(vma->vm_flags & VM_WRITE))
659 		/* disallow re-mapping with PROT_WRITE */
660 		vma->vm_flags &= ~VM_MAYWRITE;
661 
662 	err = map->ops->map_mmap(map, vma);
663 	if (err)
664 		goto out;
665 
666 	if (vma->vm_flags & VM_MAYWRITE)
667 		bpf_map_write_active_inc(map);
668 out:
669 	mutex_unlock(&map->freeze_mutex);
670 	return err;
671 }
672 
bpf_map_poll(struct file * filp,struct poll_table_struct * pts)673 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
674 {
675 	struct bpf_map *map = filp->private_data;
676 
677 	if (map->ops->map_poll)
678 		return map->ops->map_poll(map, filp, pts);
679 
680 	return EPOLLERR;
681 }
682 
683 const struct file_operations bpf_map_fops = {
684 #ifdef CONFIG_PROC_FS
685 	.show_fdinfo	= bpf_map_show_fdinfo,
686 #endif
687 	.release	= bpf_map_release,
688 	.read		= bpf_dummy_read,
689 	.write		= bpf_dummy_write,
690 	.mmap		= bpf_map_mmap,
691 	.poll		= bpf_map_poll,
692 };
693 
bpf_map_new_fd(struct bpf_map * map,int flags)694 int bpf_map_new_fd(struct bpf_map *map, int flags)
695 {
696 	int ret;
697 
698 	ret = security_bpf_map(map, OPEN_FMODE(flags));
699 	if (ret < 0)
700 		return ret;
701 
702 	return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
703 				flags | O_CLOEXEC);
704 }
705 
bpf_get_file_flag(int flags)706 int bpf_get_file_flag(int flags)
707 {
708 	if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
709 		return -EINVAL;
710 	if (flags & BPF_F_RDONLY)
711 		return O_RDONLY;
712 	if (flags & BPF_F_WRONLY)
713 		return O_WRONLY;
714 	return O_RDWR;
715 }
716 
717 /* helper macro to check that unused fields 'union bpf_attr' are zero */
718 #define CHECK_ATTR(CMD) \
719 	memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
720 		   sizeof(attr->CMD##_LAST_FIELD), 0, \
721 		   sizeof(*attr) - \
722 		   offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
723 		   sizeof(attr->CMD##_LAST_FIELD)) != NULL
724 
725 /* dst and src must have at least "size" number of bytes.
726  * Return strlen on success and < 0 on error.
727  */
bpf_obj_name_cpy(char * dst,const char * src,unsigned int size)728 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
729 {
730 	const char *end = src + size;
731 	const char *orig_src = src;
732 
733 	memset(dst, 0, size);
734 	/* Copy all isalnum(), '_' and '.' chars. */
735 	while (src < end && *src) {
736 		if (!isalnum(*src) &&
737 		    *src != '_' && *src != '.')
738 			return -EINVAL;
739 		*dst++ = *src++;
740 	}
741 
742 	/* No '\0' found in "size" number of bytes */
743 	if (src == end)
744 		return -EINVAL;
745 
746 	return src - orig_src;
747 }
748 
map_check_no_btf(const struct bpf_map * map,const struct btf * btf,const struct btf_type * key_type,const struct btf_type * value_type)749 int map_check_no_btf(const struct bpf_map *map,
750 		     const struct btf *btf,
751 		     const struct btf_type *key_type,
752 		     const struct btf_type *value_type)
753 {
754 	return -ENOTSUPP;
755 }
756 
map_check_btf(struct bpf_map * map,const struct btf * btf,u32 btf_key_id,u32 btf_value_id)757 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
758 			 u32 btf_key_id, u32 btf_value_id)
759 {
760 	const struct btf_type *key_type, *value_type;
761 	u32 key_size, value_size;
762 	int ret = 0;
763 
764 	/* Some maps allow key to be unspecified. */
765 	if (btf_key_id) {
766 		key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
767 		if (!key_type || key_size != map->key_size)
768 			return -EINVAL;
769 	} else {
770 		key_type = btf_type_by_id(btf, 0);
771 		if (!map->ops->map_check_btf)
772 			return -EINVAL;
773 	}
774 
775 	value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
776 	if (!value_type || value_size != map->value_size)
777 		return -EINVAL;
778 
779 	map->spin_lock_off = btf_find_spin_lock(btf, value_type);
780 
781 	if (map_value_has_spin_lock(map)) {
782 		if (map->map_flags & BPF_F_RDONLY_PROG)
783 			return -EACCES;
784 		if (map->map_type != BPF_MAP_TYPE_HASH &&
785 		    map->map_type != BPF_MAP_TYPE_ARRAY &&
786 		    map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
787 		    map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
788 		    map->map_type != BPF_MAP_TYPE_INODE_STORAGE)
789 			return -ENOTSUPP;
790 		if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
791 		    map->value_size) {
792 			WARN_ONCE(1,
793 				  "verifier bug spin_lock_off %d value_size %d\n",
794 				  map->spin_lock_off, map->value_size);
795 			return -EFAULT;
796 		}
797 	}
798 
799 	if (map->ops->map_check_btf)
800 		ret = map->ops->map_check_btf(map, btf, key_type, value_type);
801 
802 	return ret;
803 }
804 
805 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
806 /* called via syscall */
map_create(union bpf_attr * attr)807 static int map_create(union bpf_attr *attr)
808 {
809 	int numa_node = bpf_map_attr_numa_node(attr);
810 	struct bpf_map_memory mem;
811 	struct bpf_map *map;
812 	int f_flags;
813 	int err;
814 
815 	err = CHECK_ATTR(BPF_MAP_CREATE);
816 	if (err)
817 		return -EINVAL;
818 
819 	if (attr->btf_vmlinux_value_type_id) {
820 		if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
821 		    attr->btf_key_type_id || attr->btf_value_type_id)
822 			return -EINVAL;
823 	} else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
824 		return -EINVAL;
825 	}
826 
827 	f_flags = bpf_get_file_flag(attr->map_flags);
828 	if (f_flags < 0)
829 		return f_flags;
830 
831 	if (numa_node != NUMA_NO_NODE &&
832 	    ((unsigned int)numa_node >= nr_node_ids ||
833 	     !node_online(numa_node)))
834 		return -EINVAL;
835 
836 	/* find map type and init map: hashtable vs rbtree vs bloom vs ... */
837 	map = find_and_alloc_map(attr);
838 	if (IS_ERR(map))
839 		return PTR_ERR(map);
840 
841 	err = bpf_obj_name_cpy(map->name, attr->map_name,
842 			       sizeof(attr->map_name));
843 	if (err < 0)
844 		goto free_map;
845 
846 	atomic64_set(&map->refcnt, 1);
847 	atomic64_set(&map->usercnt, 1);
848 	mutex_init(&map->freeze_mutex);
849 
850 	map->spin_lock_off = -EINVAL;
851 	if (attr->btf_key_type_id || attr->btf_value_type_id ||
852 	    /* Even the map's value is a kernel's struct,
853 	     * the bpf_prog.o must have BTF to begin with
854 	     * to figure out the corresponding kernel's
855 	     * counter part.  Thus, attr->btf_fd has
856 	     * to be valid also.
857 	     */
858 	    attr->btf_vmlinux_value_type_id) {
859 		struct btf *btf;
860 
861 		btf = btf_get_by_fd(attr->btf_fd);
862 		if (IS_ERR(btf)) {
863 			err = PTR_ERR(btf);
864 			goto free_map;
865 		}
866 		map->btf = btf;
867 
868 		if (attr->btf_value_type_id) {
869 			err = map_check_btf(map, btf, attr->btf_key_type_id,
870 					    attr->btf_value_type_id);
871 			if (err)
872 				goto free_map;
873 		}
874 
875 		map->btf_key_type_id = attr->btf_key_type_id;
876 		map->btf_value_type_id = attr->btf_value_type_id;
877 		map->btf_vmlinux_value_type_id =
878 			attr->btf_vmlinux_value_type_id;
879 	}
880 
881 	err = security_bpf_map_alloc(map);
882 	if (err)
883 		goto free_map;
884 
885 	err = bpf_map_alloc_id(map);
886 	if (err)
887 		goto free_map_sec;
888 
889 	err = bpf_map_new_fd(map, f_flags);
890 	if (err < 0) {
891 		/* failed to allocate fd.
892 		 * bpf_map_put_with_uref() is needed because the above
893 		 * bpf_map_alloc_id() has published the map
894 		 * to the userspace and the userspace may
895 		 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
896 		 */
897 		bpf_map_put_with_uref(map);
898 		return err;
899 	}
900 
901 	return err;
902 
903 free_map_sec:
904 	security_bpf_map_free(map);
905 free_map:
906 	btf_put(map->btf);
907 	bpf_map_charge_move(&mem, &map->memory);
908 	map->ops->map_free(map);
909 	bpf_map_charge_finish(&mem);
910 	return err;
911 }
912 
913 /* if error is returned, fd is released.
914  * On success caller should complete fd access with matching fdput()
915  */
__bpf_map_get(struct fd f)916 struct bpf_map *__bpf_map_get(struct fd f)
917 {
918 	if (!f.file)
919 		return ERR_PTR(-EBADF);
920 	if (f.file->f_op != &bpf_map_fops) {
921 		fdput(f);
922 		return ERR_PTR(-EINVAL);
923 	}
924 
925 	return f.file->private_data;
926 }
927 
bpf_map_inc(struct bpf_map * map)928 void bpf_map_inc(struct bpf_map *map)
929 {
930 	atomic64_inc(&map->refcnt);
931 }
932 EXPORT_SYMBOL_GPL(bpf_map_inc);
933 
bpf_map_inc_with_uref(struct bpf_map * map)934 void bpf_map_inc_with_uref(struct bpf_map *map)
935 {
936 	atomic64_inc(&map->refcnt);
937 	atomic64_inc(&map->usercnt);
938 }
939 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
940 
bpf_map_get(u32 ufd)941 struct bpf_map *bpf_map_get(u32 ufd)
942 {
943 	struct fd f = fdget(ufd);
944 	struct bpf_map *map;
945 
946 	map = __bpf_map_get(f);
947 	if (IS_ERR(map))
948 		return map;
949 
950 	bpf_map_inc(map);
951 	fdput(f);
952 
953 	return map;
954 }
955 
bpf_map_get_with_uref(u32 ufd)956 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
957 {
958 	struct fd f = fdget(ufd);
959 	struct bpf_map *map;
960 
961 	map = __bpf_map_get(f);
962 	if (IS_ERR(map))
963 		return map;
964 
965 	bpf_map_inc_with_uref(map);
966 	fdput(f);
967 
968 	return map;
969 }
970 
971 /* map_idr_lock should have been held */
__bpf_map_inc_not_zero(struct bpf_map * map,bool uref)972 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
973 {
974 	int refold;
975 
976 	refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
977 	if (!refold)
978 		return ERR_PTR(-ENOENT);
979 	if (uref)
980 		atomic64_inc(&map->usercnt);
981 
982 	return map;
983 }
984 
bpf_map_inc_not_zero(struct bpf_map * map)985 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
986 {
987 	spin_lock_bh(&map_idr_lock);
988 	map = __bpf_map_inc_not_zero(map, false);
989 	spin_unlock_bh(&map_idr_lock);
990 
991 	return map;
992 }
993 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
994 
bpf_stackmap_copy(struct bpf_map * map,void * key,void * value)995 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
996 {
997 	return -ENOTSUPP;
998 }
999 
__bpf_copy_key(void __user * ukey,u64 key_size)1000 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1001 {
1002 	if (key_size)
1003 		return memdup_user(ukey, key_size);
1004 
1005 	if (ukey)
1006 		return ERR_PTR(-EINVAL);
1007 
1008 	return NULL;
1009 }
1010 
1011 /* last field in 'union bpf_attr' used by this command */
1012 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1013 
map_lookup_elem(union bpf_attr * attr)1014 static int map_lookup_elem(union bpf_attr *attr)
1015 {
1016 	void __user *ukey = u64_to_user_ptr(attr->key);
1017 	void __user *uvalue = u64_to_user_ptr(attr->value);
1018 	int ufd = attr->map_fd;
1019 	struct bpf_map *map;
1020 	void *key, *value;
1021 	u32 value_size;
1022 	struct fd f;
1023 	int err;
1024 
1025 	if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1026 		return -EINVAL;
1027 
1028 	if (attr->flags & ~BPF_F_LOCK)
1029 		return -EINVAL;
1030 
1031 	f = fdget(ufd);
1032 	map = __bpf_map_get(f);
1033 	if (IS_ERR(map))
1034 		return PTR_ERR(map);
1035 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1036 		err = -EPERM;
1037 		goto err_put;
1038 	}
1039 
1040 	if ((attr->flags & BPF_F_LOCK) &&
1041 	    !map_value_has_spin_lock(map)) {
1042 		err = -EINVAL;
1043 		goto err_put;
1044 	}
1045 
1046 	key = __bpf_copy_key(ukey, map->key_size);
1047 	if (IS_ERR(key)) {
1048 		err = PTR_ERR(key);
1049 		goto err_put;
1050 	}
1051 
1052 	value_size = bpf_map_value_size(map);
1053 
1054 	err = -ENOMEM;
1055 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1056 	if (!value)
1057 		goto free_key;
1058 
1059 	err = bpf_map_copy_value(map, key, value, attr->flags);
1060 	if (err)
1061 		goto free_value;
1062 
1063 	err = -EFAULT;
1064 	if (copy_to_user(uvalue, value, value_size) != 0)
1065 		goto free_value;
1066 
1067 	err = 0;
1068 
1069 free_value:
1070 	kfree(value);
1071 free_key:
1072 	kfree(key);
1073 err_put:
1074 	fdput(f);
1075 	return err;
1076 }
1077 
1078 
1079 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1080 
map_update_elem(union bpf_attr * attr)1081 static int map_update_elem(union bpf_attr *attr)
1082 {
1083 	void __user *ukey = u64_to_user_ptr(attr->key);
1084 	void __user *uvalue = u64_to_user_ptr(attr->value);
1085 	int ufd = attr->map_fd;
1086 	struct bpf_map *map;
1087 	void *key, *value;
1088 	u32 value_size;
1089 	struct fd f;
1090 	int err;
1091 
1092 	if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1093 		return -EINVAL;
1094 
1095 	f = fdget(ufd);
1096 	map = __bpf_map_get(f);
1097 	if (IS_ERR(map))
1098 		return PTR_ERR(map);
1099 	bpf_map_write_active_inc(map);
1100 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1101 		err = -EPERM;
1102 		goto err_put;
1103 	}
1104 
1105 	if ((attr->flags & BPF_F_LOCK) &&
1106 	    !map_value_has_spin_lock(map)) {
1107 		err = -EINVAL;
1108 		goto err_put;
1109 	}
1110 
1111 	key = __bpf_copy_key(ukey, map->key_size);
1112 	if (IS_ERR(key)) {
1113 		err = PTR_ERR(key);
1114 		goto err_put;
1115 	}
1116 
1117 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1118 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1119 	    map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1120 	    map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1121 		value_size = round_up(map->value_size, 8) * num_possible_cpus();
1122 	else
1123 		value_size = map->value_size;
1124 
1125 	err = -ENOMEM;
1126 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1127 	if (!value)
1128 		goto free_key;
1129 
1130 	err = -EFAULT;
1131 	if (copy_from_user(value, uvalue, value_size) != 0)
1132 		goto free_value;
1133 
1134 	err = bpf_map_update_value(map, f, key, value, attr->flags);
1135 
1136 free_value:
1137 	kfree(value);
1138 free_key:
1139 	kfree(key);
1140 err_put:
1141 	bpf_map_write_active_dec(map);
1142 	fdput(f);
1143 	return err;
1144 }
1145 
1146 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1147 
map_delete_elem(union bpf_attr * attr)1148 static int map_delete_elem(union bpf_attr *attr)
1149 {
1150 	void __user *ukey = u64_to_user_ptr(attr->key);
1151 	int ufd = attr->map_fd;
1152 	struct bpf_map *map;
1153 	struct fd f;
1154 	void *key;
1155 	int err;
1156 
1157 	if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1158 		return -EINVAL;
1159 
1160 	f = fdget(ufd);
1161 	map = __bpf_map_get(f);
1162 	if (IS_ERR(map))
1163 		return PTR_ERR(map);
1164 	bpf_map_write_active_inc(map);
1165 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1166 		err = -EPERM;
1167 		goto err_put;
1168 	}
1169 
1170 	key = __bpf_copy_key(ukey, map->key_size);
1171 	if (IS_ERR(key)) {
1172 		err = PTR_ERR(key);
1173 		goto err_put;
1174 	}
1175 
1176 	if (bpf_map_is_dev_bound(map)) {
1177 		err = bpf_map_offload_delete_elem(map, key);
1178 		goto out;
1179 	} else if (IS_FD_PROG_ARRAY(map) ||
1180 		   map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1181 		/* These maps require sleepable context */
1182 		err = map->ops->map_delete_elem(map, key);
1183 		goto out;
1184 	}
1185 
1186 	bpf_disable_instrumentation();
1187 	rcu_read_lock();
1188 	err = map->ops->map_delete_elem(map, key);
1189 	rcu_read_unlock();
1190 	bpf_enable_instrumentation();
1191 	maybe_wait_bpf_programs(map);
1192 out:
1193 	kfree(key);
1194 err_put:
1195 	bpf_map_write_active_dec(map);
1196 	fdput(f);
1197 	return err;
1198 }
1199 
1200 /* last field in 'union bpf_attr' used by this command */
1201 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1202 
map_get_next_key(union bpf_attr * attr)1203 static int map_get_next_key(union bpf_attr *attr)
1204 {
1205 	void __user *ukey = u64_to_user_ptr(attr->key);
1206 	void __user *unext_key = u64_to_user_ptr(attr->next_key);
1207 	int ufd = attr->map_fd;
1208 	struct bpf_map *map;
1209 	void *key, *next_key;
1210 	struct fd f;
1211 	int err;
1212 
1213 	if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1214 		return -EINVAL;
1215 
1216 	f = fdget(ufd);
1217 	map = __bpf_map_get(f);
1218 	if (IS_ERR(map))
1219 		return PTR_ERR(map);
1220 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1221 		err = -EPERM;
1222 		goto err_put;
1223 	}
1224 
1225 	if (ukey) {
1226 		key = __bpf_copy_key(ukey, map->key_size);
1227 		if (IS_ERR(key)) {
1228 			err = PTR_ERR(key);
1229 			goto err_put;
1230 		}
1231 	} else {
1232 		key = NULL;
1233 	}
1234 
1235 	err = -ENOMEM;
1236 	next_key = kmalloc(map->key_size, GFP_USER);
1237 	if (!next_key)
1238 		goto free_key;
1239 
1240 	if (bpf_map_is_dev_bound(map)) {
1241 		err = bpf_map_offload_get_next_key(map, key, next_key);
1242 		goto out;
1243 	}
1244 
1245 	rcu_read_lock();
1246 	err = map->ops->map_get_next_key(map, key, next_key);
1247 	rcu_read_unlock();
1248 out:
1249 	if (err)
1250 		goto free_next_key;
1251 
1252 	err = -EFAULT;
1253 	if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1254 		goto free_next_key;
1255 
1256 	err = 0;
1257 
1258 free_next_key:
1259 	kfree(next_key);
1260 free_key:
1261 	kfree(key);
1262 err_put:
1263 	fdput(f);
1264 	return err;
1265 }
1266 
generic_map_delete_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1267 int generic_map_delete_batch(struct bpf_map *map,
1268 			     const union bpf_attr *attr,
1269 			     union bpf_attr __user *uattr)
1270 {
1271 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1272 	u32 cp, max_count;
1273 	int err = 0;
1274 	void *key;
1275 
1276 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1277 		return -EINVAL;
1278 
1279 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1280 	    !map_value_has_spin_lock(map)) {
1281 		return -EINVAL;
1282 	}
1283 
1284 	max_count = attr->batch.count;
1285 	if (!max_count)
1286 		return 0;
1287 
1288 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1289 	if (!key)
1290 		return -ENOMEM;
1291 
1292 	for (cp = 0; cp < max_count; cp++) {
1293 		err = -EFAULT;
1294 		if (copy_from_user(key, keys + cp * map->key_size,
1295 				   map->key_size))
1296 			break;
1297 
1298 		if (bpf_map_is_dev_bound(map)) {
1299 			err = bpf_map_offload_delete_elem(map, key);
1300 			break;
1301 		}
1302 
1303 		bpf_disable_instrumentation();
1304 		rcu_read_lock();
1305 		err = map->ops->map_delete_elem(map, key);
1306 		rcu_read_unlock();
1307 		bpf_enable_instrumentation();
1308 		maybe_wait_bpf_programs(map);
1309 		if (err)
1310 			break;
1311 		cond_resched();
1312 	}
1313 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1314 		err = -EFAULT;
1315 
1316 	kfree(key);
1317 	return err;
1318 }
1319 
generic_map_update_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1320 int generic_map_update_batch(struct bpf_map *map,
1321 			     const union bpf_attr *attr,
1322 			     union bpf_attr __user *uattr)
1323 {
1324 	void __user *values = u64_to_user_ptr(attr->batch.values);
1325 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1326 	u32 value_size, cp, max_count;
1327 	int ufd = attr->batch.map_fd;
1328 	void *key, *value;
1329 	struct fd f;
1330 	int err = 0;
1331 
1332 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1333 		return -EINVAL;
1334 
1335 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1336 	    !map_value_has_spin_lock(map)) {
1337 		return -EINVAL;
1338 	}
1339 
1340 	value_size = bpf_map_value_size(map);
1341 
1342 	max_count = attr->batch.count;
1343 	if (!max_count)
1344 		return 0;
1345 
1346 	key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1347 	if (!key)
1348 		return -ENOMEM;
1349 
1350 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1351 	if (!value) {
1352 		kfree(key);
1353 		return -ENOMEM;
1354 	}
1355 
1356 	f = fdget(ufd); /* bpf_map_do_batch() guarantees ufd is valid */
1357 	for (cp = 0; cp < max_count; cp++) {
1358 		err = -EFAULT;
1359 		if (copy_from_user(key, keys + cp * map->key_size,
1360 		    map->key_size) ||
1361 		    copy_from_user(value, values + cp * value_size, value_size))
1362 			break;
1363 
1364 		err = bpf_map_update_value(map, f, key, value,
1365 					   attr->batch.elem_flags);
1366 
1367 		if (err)
1368 			break;
1369 		cond_resched();
1370 	}
1371 
1372 	if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1373 		err = -EFAULT;
1374 
1375 	kfree(value);
1376 	kfree(key);
1377 	fdput(f);
1378 	return err;
1379 }
1380 
1381 #define MAP_LOOKUP_RETRIES 3
1382 
generic_map_lookup_batch(struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)1383 int generic_map_lookup_batch(struct bpf_map *map,
1384 				    const union bpf_attr *attr,
1385 				    union bpf_attr __user *uattr)
1386 {
1387 	void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1388 	void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1389 	void __user *values = u64_to_user_ptr(attr->batch.values);
1390 	void __user *keys = u64_to_user_ptr(attr->batch.keys);
1391 	void *buf, *buf_prevkey, *prev_key, *key, *value;
1392 	int err, retry = MAP_LOOKUP_RETRIES;
1393 	u32 value_size, cp, max_count;
1394 
1395 	if (attr->batch.elem_flags & ~BPF_F_LOCK)
1396 		return -EINVAL;
1397 
1398 	if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1399 	    !map_value_has_spin_lock(map))
1400 		return -EINVAL;
1401 
1402 	value_size = bpf_map_value_size(map);
1403 
1404 	max_count = attr->batch.count;
1405 	if (!max_count)
1406 		return 0;
1407 
1408 	if (put_user(0, &uattr->batch.count))
1409 		return -EFAULT;
1410 
1411 	buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1412 	if (!buf_prevkey)
1413 		return -ENOMEM;
1414 
1415 	buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1416 	if (!buf) {
1417 		kfree(buf_prevkey);
1418 		return -ENOMEM;
1419 	}
1420 
1421 	err = -EFAULT;
1422 	prev_key = NULL;
1423 	if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1424 		goto free_buf;
1425 	key = buf;
1426 	value = key + map->key_size;
1427 	if (ubatch)
1428 		prev_key = buf_prevkey;
1429 
1430 	for (cp = 0; cp < max_count;) {
1431 		rcu_read_lock();
1432 		err = map->ops->map_get_next_key(map, prev_key, key);
1433 		rcu_read_unlock();
1434 		if (err)
1435 			break;
1436 		err = bpf_map_copy_value(map, key, value,
1437 					 attr->batch.elem_flags);
1438 
1439 		if (err == -ENOENT) {
1440 			if (retry) {
1441 				retry--;
1442 				continue;
1443 			}
1444 			err = -EINTR;
1445 			break;
1446 		}
1447 
1448 		if (err)
1449 			goto free_buf;
1450 
1451 		if (copy_to_user(keys + cp * map->key_size, key,
1452 				 map->key_size)) {
1453 			err = -EFAULT;
1454 			goto free_buf;
1455 		}
1456 		if (copy_to_user(values + cp * value_size, value, value_size)) {
1457 			err = -EFAULT;
1458 			goto free_buf;
1459 		}
1460 
1461 		if (!prev_key)
1462 			prev_key = buf_prevkey;
1463 
1464 		swap(prev_key, key);
1465 		retry = MAP_LOOKUP_RETRIES;
1466 		cp++;
1467 		cond_resched();
1468 	}
1469 
1470 	if (err == -EFAULT)
1471 		goto free_buf;
1472 
1473 	if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1474 		    (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1475 		err = -EFAULT;
1476 
1477 free_buf:
1478 	kfree(buf_prevkey);
1479 	kfree(buf);
1480 	return err;
1481 }
1482 
1483 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1484 
map_lookup_and_delete_elem(union bpf_attr * attr)1485 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1486 {
1487 	void __user *ukey = u64_to_user_ptr(attr->key);
1488 	void __user *uvalue = u64_to_user_ptr(attr->value);
1489 	int ufd = attr->map_fd;
1490 	struct bpf_map *map;
1491 	void *key, *value;
1492 	u32 value_size;
1493 	struct fd f;
1494 	int err;
1495 
1496 	if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1497 		return -EINVAL;
1498 
1499 	f = fdget(ufd);
1500 	map = __bpf_map_get(f);
1501 	if (IS_ERR(map))
1502 		return PTR_ERR(map);
1503 	bpf_map_write_active_inc(map);
1504 	if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1505 	    !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1506 		err = -EPERM;
1507 		goto err_put;
1508 	}
1509 
1510 	key = __bpf_copy_key(ukey, map->key_size);
1511 	if (IS_ERR(key)) {
1512 		err = PTR_ERR(key);
1513 		goto err_put;
1514 	}
1515 
1516 	value_size = map->value_size;
1517 
1518 	err = -ENOMEM;
1519 	value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1520 	if (!value)
1521 		goto free_key;
1522 
1523 	if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1524 	    map->map_type == BPF_MAP_TYPE_STACK) {
1525 		err = map->ops->map_pop_elem(map, value);
1526 	} else {
1527 		err = -ENOTSUPP;
1528 	}
1529 
1530 	if (err)
1531 		goto free_value;
1532 
1533 	if (copy_to_user(uvalue, value, value_size) != 0) {
1534 		err = -EFAULT;
1535 		goto free_value;
1536 	}
1537 
1538 	err = 0;
1539 
1540 free_value:
1541 	kfree(value);
1542 free_key:
1543 	kfree(key);
1544 err_put:
1545 	bpf_map_write_active_dec(map);
1546 	fdput(f);
1547 	return err;
1548 }
1549 
1550 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1551 
map_freeze(const union bpf_attr * attr)1552 static int map_freeze(const union bpf_attr *attr)
1553 {
1554 	int err = 0, ufd = attr->map_fd;
1555 	struct bpf_map *map;
1556 	struct fd f;
1557 
1558 	if (CHECK_ATTR(BPF_MAP_FREEZE))
1559 		return -EINVAL;
1560 
1561 	f = fdget(ufd);
1562 	map = __bpf_map_get(f);
1563 	if (IS_ERR(map))
1564 		return PTR_ERR(map);
1565 
1566 	if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1567 		fdput(f);
1568 		return -ENOTSUPP;
1569 	}
1570 
1571 	mutex_lock(&map->freeze_mutex);
1572 	if (bpf_map_write_active(map)) {
1573 		err = -EBUSY;
1574 		goto err_put;
1575 	}
1576 	if (READ_ONCE(map->frozen)) {
1577 		err = -EBUSY;
1578 		goto err_put;
1579 	}
1580 	if (!bpf_capable()) {
1581 		err = -EPERM;
1582 		goto err_put;
1583 	}
1584 
1585 	WRITE_ONCE(map->frozen, true);
1586 err_put:
1587 	mutex_unlock(&map->freeze_mutex);
1588 	fdput(f);
1589 	return err;
1590 }
1591 
1592 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1593 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1594 	[_id] = & _name ## _prog_ops,
1595 #define BPF_MAP_TYPE(_id, _ops)
1596 #define BPF_LINK_TYPE(_id, _name)
1597 #include <linux/bpf_types.h>
1598 #undef BPF_PROG_TYPE
1599 #undef BPF_MAP_TYPE
1600 #undef BPF_LINK_TYPE
1601 };
1602 
find_prog_type(enum bpf_prog_type type,struct bpf_prog * prog)1603 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1604 {
1605 	const struct bpf_prog_ops *ops;
1606 
1607 	if (type >= ARRAY_SIZE(bpf_prog_types))
1608 		return -EINVAL;
1609 	type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1610 	ops = bpf_prog_types[type];
1611 	if (!ops)
1612 		return -EINVAL;
1613 
1614 	if (!bpf_prog_is_dev_bound(prog->aux))
1615 		prog->aux->ops = ops;
1616 	else
1617 		prog->aux->ops = &bpf_offload_prog_ops;
1618 	prog->type = type;
1619 	return 0;
1620 }
1621 
1622 enum bpf_audit {
1623 	BPF_AUDIT_LOAD,
1624 	BPF_AUDIT_UNLOAD,
1625 	BPF_AUDIT_MAX,
1626 };
1627 
1628 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1629 	[BPF_AUDIT_LOAD]   = "LOAD",
1630 	[BPF_AUDIT_UNLOAD] = "UNLOAD",
1631 };
1632 
bpf_audit_prog(const struct bpf_prog * prog,unsigned int op)1633 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1634 {
1635 	struct audit_context *ctx = NULL;
1636 	struct audit_buffer *ab;
1637 
1638 	if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1639 		return;
1640 	if (audit_enabled == AUDIT_OFF)
1641 		return;
1642 	if (op == BPF_AUDIT_LOAD)
1643 		ctx = audit_context();
1644 	ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1645 	if (unlikely(!ab))
1646 		return;
1647 	audit_log_format(ab, "prog-id=%u op=%s",
1648 			 prog->aux->id, bpf_audit_str[op]);
1649 	audit_log_end(ab);
1650 }
1651 
__bpf_prog_charge(struct user_struct * user,u32 pages)1652 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1653 {
1654 	unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1655 	unsigned long user_bufs;
1656 
1657 	if (user) {
1658 		user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1659 		if (user_bufs > memlock_limit) {
1660 			atomic_long_sub(pages, &user->locked_vm);
1661 			return -EPERM;
1662 		}
1663 	}
1664 
1665 	return 0;
1666 }
1667 
__bpf_prog_uncharge(struct user_struct * user,u32 pages)1668 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1669 {
1670 	if (user)
1671 		atomic_long_sub(pages, &user->locked_vm);
1672 }
1673 
bpf_prog_charge_memlock(struct bpf_prog * prog)1674 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1675 {
1676 	struct user_struct *user = get_current_user();
1677 	int ret;
1678 
1679 	ret = __bpf_prog_charge(user, prog->pages);
1680 	if (ret) {
1681 		free_uid(user);
1682 		return ret;
1683 	}
1684 
1685 	prog->aux->user = user;
1686 	return 0;
1687 }
1688 
bpf_prog_uncharge_memlock(struct bpf_prog * prog)1689 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1690 {
1691 	struct user_struct *user = prog->aux->user;
1692 
1693 	__bpf_prog_uncharge(user, prog->pages);
1694 	free_uid(user);
1695 }
1696 
bpf_prog_alloc_id(struct bpf_prog * prog)1697 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1698 {
1699 	int id;
1700 
1701 	idr_preload(GFP_KERNEL);
1702 	spin_lock_bh(&prog_idr_lock);
1703 	id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1704 	if (id > 0)
1705 		prog->aux->id = id;
1706 	spin_unlock_bh(&prog_idr_lock);
1707 	idr_preload_end();
1708 
1709 	/* id is in [1, INT_MAX) */
1710 	if (WARN_ON_ONCE(!id))
1711 		return -ENOSPC;
1712 
1713 	return id > 0 ? 0 : id;
1714 }
1715 
bpf_prog_free_id(struct bpf_prog * prog,bool do_idr_lock)1716 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1717 {
1718 	/* cBPF to eBPF migrations are currently not in the idr store.
1719 	 * Offloaded programs are removed from the store when their device
1720 	 * disappears - even if someone grabs an fd to them they are unusable,
1721 	 * simply waiting for refcnt to drop to be freed.
1722 	 */
1723 	if (!prog->aux->id)
1724 		return;
1725 
1726 	if (do_idr_lock)
1727 		spin_lock_bh(&prog_idr_lock);
1728 	else
1729 		__acquire(&prog_idr_lock);
1730 
1731 	idr_remove(&prog_idr, prog->aux->id);
1732 	prog->aux->id = 0;
1733 
1734 	if (do_idr_lock)
1735 		spin_unlock_bh(&prog_idr_lock);
1736 	else
1737 		__release(&prog_idr_lock);
1738 }
1739 
__bpf_prog_put_rcu(struct rcu_head * rcu)1740 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1741 {
1742 	struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1743 
1744 	kvfree(aux->func_info);
1745 	kfree(aux->func_info_aux);
1746 	bpf_prog_uncharge_memlock(aux->prog);
1747 	security_bpf_prog_free(aux);
1748 	bpf_prog_free(aux->prog);
1749 }
1750 
__bpf_prog_put_noref(struct bpf_prog * prog,bool deferred)1751 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1752 {
1753 	bpf_prog_kallsyms_del_all(prog);
1754 	btf_put(prog->aux->btf);
1755 	bpf_prog_free_linfo(prog);
1756 
1757 	if (deferred) {
1758 		if (prog->aux->sleepable)
1759 			call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1760 		else
1761 			call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1762 	} else {
1763 		__bpf_prog_put_rcu(&prog->aux->rcu);
1764 	}
1765 }
1766 
__bpf_prog_put(struct bpf_prog * prog,bool do_idr_lock)1767 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1768 {
1769 	if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1770 		perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1771 		bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1772 		/* bpf_prog_free_id() must be called first */
1773 		bpf_prog_free_id(prog, do_idr_lock);
1774 		__bpf_prog_put_noref(prog, true);
1775 	}
1776 }
1777 
bpf_prog_put(struct bpf_prog * prog)1778 void bpf_prog_put(struct bpf_prog *prog)
1779 {
1780 	__bpf_prog_put(prog, true);
1781 }
1782 EXPORT_SYMBOL_GPL(bpf_prog_put);
1783 
bpf_prog_release(struct inode * inode,struct file * filp)1784 static int bpf_prog_release(struct inode *inode, struct file *filp)
1785 {
1786 	struct bpf_prog *prog = filp->private_data;
1787 
1788 	bpf_prog_put(prog);
1789 	return 0;
1790 }
1791 
bpf_prog_get_stats(const struct bpf_prog * prog,struct bpf_prog_stats * stats)1792 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1793 			       struct bpf_prog_stats *stats)
1794 {
1795 	u64 nsecs = 0, cnt = 0;
1796 	int cpu;
1797 
1798 	for_each_possible_cpu(cpu) {
1799 		const struct bpf_prog_stats *st;
1800 		unsigned int start;
1801 		u64 tnsecs, tcnt;
1802 
1803 		st = per_cpu_ptr(prog->aux->stats, cpu);
1804 		do {
1805 			start = u64_stats_fetch_begin_irq(&st->syncp);
1806 			tnsecs = st->nsecs;
1807 			tcnt = st->cnt;
1808 		} while (u64_stats_fetch_retry_irq(&st->syncp, start));
1809 		nsecs += tnsecs;
1810 		cnt += tcnt;
1811 	}
1812 	stats->nsecs = nsecs;
1813 	stats->cnt = cnt;
1814 }
1815 
1816 #ifdef CONFIG_PROC_FS
bpf_prog_show_fdinfo(struct seq_file * m,struct file * filp)1817 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1818 {
1819 	const struct bpf_prog *prog = filp->private_data;
1820 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1821 	struct bpf_prog_stats stats;
1822 
1823 	bpf_prog_get_stats(prog, &stats);
1824 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1825 	seq_printf(m,
1826 		   "prog_type:\t%u\n"
1827 		   "prog_jited:\t%u\n"
1828 		   "prog_tag:\t%s\n"
1829 		   "memlock:\t%llu\n"
1830 		   "prog_id:\t%u\n"
1831 		   "run_time_ns:\t%llu\n"
1832 		   "run_cnt:\t%llu\n",
1833 		   prog->type,
1834 		   prog->jited,
1835 		   prog_tag,
1836 		   prog->pages * 1ULL << PAGE_SHIFT,
1837 		   prog->aux->id,
1838 		   stats.nsecs,
1839 		   stats.cnt);
1840 }
1841 #endif
1842 
1843 const struct file_operations bpf_prog_fops = {
1844 #ifdef CONFIG_PROC_FS
1845 	.show_fdinfo	= bpf_prog_show_fdinfo,
1846 #endif
1847 	.release	= bpf_prog_release,
1848 	.read		= bpf_dummy_read,
1849 	.write		= bpf_dummy_write,
1850 };
1851 
bpf_prog_new_fd(struct bpf_prog * prog)1852 int bpf_prog_new_fd(struct bpf_prog *prog)
1853 {
1854 	int ret;
1855 
1856 	ret = security_bpf_prog(prog);
1857 	if (ret < 0)
1858 		return ret;
1859 
1860 	return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1861 				O_RDWR | O_CLOEXEC);
1862 }
1863 
____bpf_prog_get(struct fd f)1864 static struct bpf_prog *____bpf_prog_get(struct fd f)
1865 {
1866 	if (!f.file)
1867 		return ERR_PTR(-EBADF);
1868 	if (f.file->f_op != &bpf_prog_fops) {
1869 		fdput(f);
1870 		return ERR_PTR(-EINVAL);
1871 	}
1872 
1873 	return f.file->private_data;
1874 }
1875 
bpf_prog_add(struct bpf_prog * prog,int i)1876 void bpf_prog_add(struct bpf_prog *prog, int i)
1877 {
1878 	atomic64_add(i, &prog->aux->refcnt);
1879 }
1880 EXPORT_SYMBOL_GPL(bpf_prog_add);
1881 
bpf_prog_sub(struct bpf_prog * prog,int i)1882 void bpf_prog_sub(struct bpf_prog *prog, int i)
1883 {
1884 	/* Only to be used for undoing previous bpf_prog_add() in some
1885 	 * error path. We still know that another entity in our call
1886 	 * path holds a reference to the program, thus atomic_sub() can
1887 	 * be safely used in such cases!
1888 	 */
1889 	WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1890 }
1891 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1892 
bpf_prog_inc(struct bpf_prog * prog)1893 void bpf_prog_inc(struct bpf_prog *prog)
1894 {
1895 	atomic64_inc(&prog->aux->refcnt);
1896 }
1897 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1898 
1899 /* prog_idr_lock should have been held */
bpf_prog_inc_not_zero(struct bpf_prog * prog)1900 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1901 {
1902 	int refold;
1903 
1904 	refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1905 
1906 	if (!refold)
1907 		return ERR_PTR(-ENOENT);
1908 
1909 	return prog;
1910 }
1911 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1912 
bpf_prog_get_ok(struct bpf_prog * prog,enum bpf_prog_type * attach_type,bool attach_drv)1913 bool bpf_prog_get_ok(struct bpf_prog *prog,
1914 			    enum bpf_prog_type *attach_type, bool attach_drv)
1915 {
1916 	/* not an attachment, just a refcount inc, always allow */
1917 	if (!attach_type)
1918 		return true;
1919 
1920 	if (prog->type != *attach_type)
1921 		return false;
1922 	if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1923 		return false;
1924 
1925 	return true;
1926 }
1927 
__bpf_prog_get(u32 ufd,enum bpf_prog_type * attach_type,bool attach_drv)1928 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1929 				       bool attach_drv)
1930 {
1931 	struct fd f = fdget(ufd);
1932 	struct bpf_prog *prog;
1933 
1934 	prog = ____bpf_prog_get(f);
1935 	if (IS_ERR(prog))
1936 		return prog;
1937 	if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1938 		prog = ERR_PTR(-EINVAL);
1939 		goto out;
1940 	}
1941 
1942 	bpf_prog_inc(prog);
1943 out:
1944 	fdput(f);
1945 	return prog;
1946 }
1947 
bpf_prog_get(u32 ufd)1948 struct bpf_prog *bpf_prog_get(u32 ufd)
1949 {
1950 	return __bpf_prog_get(ufd, NULL, false);
1951 }
1952 
bpf_prog_get_type_dev(u32 ufd,enum bpf_prog_type type,bool attach_drv)1953 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1954 				       bool attach_drv)
1955 {
1956 	return __bpf_prog_get(ufd, &type, attach_drv);
1957 }
1958 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1959 
1960 /* Initially all BPF programs could be loaded w/o specifying
1961  * expected_attach_type. Later for some of them specifying expected_attach_type
1962  * at load time became required so that program could be validated properly.
1963  * Programs of types that are allowed to be loaded both w/ and w/o (for
1964  * backward compatibility) expected_attach_type, should have the default attach
1965  * type assigned to expected_attach_type for the latter case, so that it can be
1966  * validated later at attach time.
1967  *
1968  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1969  * prog type requires it but has some attach types that have to be backward
1970  * compatible.
1971  */
bpf_prog_load_fixup_attach_type(union bpf_attr * attr)1972 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1973 {
1974 	switch (attr->prog_type) {
1975 	case BPF_PROG_TYPE_CGROUP_SOCK:
1976 		/* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1977 		 * exist so checking for non-zero is the way to go here.
1978 		 */
1979 		if (!attr->expected_attach_type)
1980 			attr->expected_attach_type =
1981 				BPF_CGROUP_INET_SOCK_CREATE;
1982 		break;
1983 	}
1984 }
1985 
1986 static int
bpf_prog_load_check_attach(enum bpf_prog_type prog_type,enum bpf_attach_type expected_attach_type,u32 btf_id,u32 prog_fd)1987 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1988 			   enum bpf_attach_type expected_attach_type,
1989 			   u32 btf_id, u32 prog_fd)
1990 {
1991 	if (btf_id) {
1992 		if (btf_id > BTF_MAX_TYPE)
1993 			return -EINVAL;
1994 
1995 		switch (prog_type) {
1996 		case BPF_PROG_TYPE_TRACING:
1997 		case BPF_PROG_TYPE_LSM:
1998 		case BPF_PROG_TYPE_STRUCT_OPS:
1999 		case BPF_PROG_TYPE_EXT:
2000 			break;
2001 		default:
2002 			return -EINVAL;
2003 		}
2004 	}
2005 
2006 	if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
2007 	    prog_type != BPF_PROG_TYPE_EXT)
2008 		return -EINVAL;
2009 
2010 	switch (prog_type) {
2011 	case BPF_PROG_TYPE_CGROUP_SOCK:
2012 		switch (expected_attach_type) {
2013 		case BPF_CGROUP_INET_SOCK_CREATE:
2014 		case BPF_CGROUP_INET_SOCK_RELEASE:
2015 		case BPF_CGROUP_INET4_POST_BIND:
2016 		case BPF_CGROUP_INET6_POST_BIND:
2017 			return 0;
2018 		default:
2019 			return -EINVAL;
2020 		}
2021 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2022 		switch (expected_attach_type) {
2023 		case BPF_CGROUP_INET4_BIND:
2024 		case BPF_CGROUP_INET6_BIND:
2025 		case BPF_CGROUP_INET4_CONNECT:
2026 		case BPF_CGROUP_INET6_CONNECT:
2027 		case BPF_CGROUP_INET4_GETPEERNAME:
2028 		case BPF_CGROUP_INET6_GETPEERNAME:
2029 		case BPF_CGROUP_INET4_GETSOCKNAME:
2030 		case BPF_CGROUP_INET6_GETSOCKNAME:
2031 		case BPF_CGROUP_UDP4_SENDMSG:
2032 		case BPF_CGROUP_UDP6_SENDMSG:
2033 		case BPF_CGROUP_UDP4_RECVMSG:
2034 		case BPF_CGROUP_UDP6_RECVMSG:
2035 			return 0;
2036 		default:
2037 			return -EINVAL;
2038 		}
2039 	case BPF_PROG_TYPE_CGROUP_SKB:
2040 		switch (expected_attach_type) {
2041 		case BPF_CGROUP_INET_INGRESS:
2042 		case BPF_CGROUP_INET_EGRESS:
2043 			return 0;
2044 		default:
2045 			return -EINVAL;
2046 		}
2047 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2048 		switch (expected_attach_type) {
2049 		case BPF_CGROUP_SETSOCKOPT:
2050 		case BPF_CGROUP_GETSOCKOPT:
2051 			return 0;
2052 		default:
2053 			return -EINVAL;
2054 		}
2055 	case BPF_PROG_TYPE_SK_LOOKUP:
2056 		if (expected_attach_type == BPF_SK_LOOKUP)
2057 			return 0;
2058 		return -EINVAL;
2059 	case BPF_PROG_TYPE_EXT:
2060 		if (expected_attach_type)
2061 			return -EINVAL;
2062 		fallthrough;
2063 	default:
2064 		return 0;
2065 	}
2066 }
2067 
is_net_admin_prog_type(enum bpf_prog_type prog_type)2068 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2069 {
2070 	switch (prog_type) {
2071 	case BPF_PROG_TYPE_SCHED_CLS:
2072 	case BPF_PROG_TYPE_SCHED_ACT:
2073 	case BPF_PROG_TYPE_XDP:
2074 	case BPF_PROG_TYPE_LWT_IN:
2075 	case BPF_PROG_TYPE_LWT_OUT:
2076 	case BPF_PROG_TYPE_LWT_XMIT:
2077 	case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2078 	case BPF_PROG_TYPE_SK_SKB:
2079 	case BPF_PROG_TYPE_SK_MSG:
2080 	case BPF_PROG_TYPE_LIRC_MODE2:
2081 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
2082 	case BPF_PROG_TYPE_CGROUP_DEVICE:
2083 	case BPF_PROG_TYPE_CGROUP_SOCK:
2084 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2085 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2086 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
2087 	case BPF_PROG_TYPE_SOCK_OPS:
2088 	case BPF_PROG_TYPE_EXT: /* extends any prog */
2089 		return true;
2090 	case BPF_PROG_TYPE_CGROUP_SKB:
2091 		/* always unpriv */
2092 	case BPF_PROG_TYPE_SK_REUSEPORT:
2093 		/* equivalent to SOCKET_FILTER. need CAP_BPF only */
2094 	default:
2095 		return false;
2096 	}
2097 }
2098 
is_perfmon_prog_type(enum bpf_prog_type prog_type)2099 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2100 {
2101 	switch (prog_type) {
2102 	case BPF_PROG_TYPE_KPROBE:
2103 	case BPF_PROG_TYPE_TRACEPOINT:
2104 	case BPF_PROG_TYPE_PERF_EVENT:
2105 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2106 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2107 	case BPF_PROG_TYPE_TRACING:
2108 	case BPF_PROG_TYPE_LSM:
2109 	case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2110 	case BPF_PROG_TYPE_EXT: /* extends any prog */
2111 		return true;
2112 	default:
2113 		return false;
2114 	}
2115 }
2116 
2117 /* last field in 'union bpf_attr' used by this command */
2118 #define	BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2119 
bpf_prog_load(union bpf_attr * attr,union bpf_attr __user * uattr)2120 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2121 {
2122 	enum bpf_prog_type type = attr->prog_type;
2123 	struct bpf_prog *prog;
2124 	int err;
2125 	char license[128];
2126 	bool is_gpl;
2127 
2128 	if (CHECK_ATTR(BPF_PROG_LOAD))
2129 		return -EINVAL;
2130 
2131 	if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2132 				 BPF_F_ANY_ALIGNMENT |
2133 				 BPF_F_TEST_STATE_FREQ |
2134 				 BPF_F_SLEEPABLE |
2135 				 BPF_F_TEST_RND_HI32))
2136 		return -EINVAL;
2137 
2138 	if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2139 	    (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2140 	    !bpf_capable())
2141 		return -EPERM;
2142 
2143 	/* copy eBPF program license from user space */
2144 	if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2145 			      sizeof(license) - 1) < 0)
2146 		return -EFAULT;
2147 	license[sizeof(license) - 1] = 0;
2148 
2149 	/* eBPF programs must be GPL compatible to use GPL-ed functions */
2150 	is_gpl = license_is_gpl_compatible(license);
2151 
2152 	if (attr->insn_cnt == 0 ||
2153 	    attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2154 		return -E2BIG;
2155 	if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2156 	    type != BPF_PROG_TYPE_CGROUP_SKB &&
2157 	    !bpf_capable())
2158 		return -EPERM;
2159 
2160 	if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2161 		return -EPERM;
2162 	if (is_perfmon_prog_type(type) && !perfmon_capable())
2163 		return -EPERM;
2164 
2165 	bpf_prog_load_fixup_attach_type(attr);
2166 	if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2167 				       attr->attach_btf_id,
2168 				       attr->attach_prog_fd))
2169 		return -EINVAL;
2170 
2171 	/* plain bpf_prog allocation */
2172 	prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2173 	if (!prog)
2174 		return -ENOMEM;
2175 
2176 	prog->expected_attach_type = attr->expected_attach_type;
2177 	prog->aux->attach_btf_id = attr->attach_btf_id;
2178 	if (attr->attach_prog_fd) {
2179 		struct bpf_prog *dst_prog;
2180 
2181 		dst_prog = bpf_prog_get(attr->attach_prog_fd);
2182 		if (IS_ERR(dst_prog)) {
2183 			err = PTR_ERR(dst_prog);
2184 			goto free_prog_nouncharge;
2185 		}
2186 		prog->aux->dst_prog = dst_prog;
2187 	}
2188 
2189 	prog->aux->offload_requested = !!attr->prog_ifindex;
2190 	prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2191 
2192 	err = security_bpf_prog_alloc(prog->aux);
2193 	if (err)
2194 		goto free_prog_nouncharge;
2195 
2196 	err = bpf_prog_charge_memlock(prog);
2197 	if (err)
2198 		goto free_prog_sec;
2199 
2200 	prog->len = attr->insn_cnt;
2201 
2202 	err = -EFAULT;
2203 	if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2204 			   bpf_prog_insn_size(prog)) != 0)
2205 		goto free_prog;
2206 
2207 	prog->orig_prog = NULL;
2208 	prog->jited = 0;
2209 
2210 	atomic64_set(&prog->aux->refcnt, 1);
2211 	prog->gpl_compatible = is_gpl ? 1 : 0;
2212 
2213 	if (bpf_prog_is_dev_bound(prog->aux)) {
2214 		err = bpf_prog_offload_init(prog, attr);
2215 		if (err)
2216 			goto free_prog;
2217 	}
2218 
2219 	/* find program type: socket_filter vs tracing_filter */
2220 	err = find_prog_type(type, prog);
2221 	if (err < 0)
2222 		goto free_prog;
2223 
2224 	prog->aux->load_time = ktime_get_boottime_ns();
2225 	err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2226 			       sizeof(attr->prog_name));
2227 	if (err < 0)
2228 		goto free_prog;
2229 
2230 	/* run eBPF verifier */
2231 	err = bpf_check(&prog, attr, uattr);
2232 	if (err < 0)
2233 		goto free_used_maps;
2234 
2235 	prog = bpf_prog_select_runtime(prog, &err);
2236 	if (err < 0)
2237 		goto free_used_maps;
2238 
2239 	err = bpf_prog_alloc_id(prog);
2240 	if (err)
2241 		goto free_used_maps;
2242 
2243 	/* Upon success of bpf_prog_alloc_id(), the BPF prog is
2244 	 * effectively publicly exposed. However, retrieving via
2245 	 * bpf_prog_get_fd_by_id() will take another reference,
2246 	 * therefore it cannot be gone underneath us.
2247 	 *
2248 	 * Only for the time /after/ successful bpf_prog_new_fd()
2249 	 * and before returning to userspace, we might just hold
2250 	 * one reference and any parallel close on that fd could
2251 	 * rip everything out. Hence, below notifications must
2252 	 * happen before bpf_prog_new_fd().
2253 	 *
2254 	 * Also, any failure handling from this point onwards must
2255 	 * be using bpf_prog_put() given the program is exposed.
2256 	 */
2257 	bpf_prog_kallsyms_add(prog);
2258 	perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2259 	bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2260 
2261 	err = bpf_prog_new_fd(prog);
2262 	if (err < 0)
2263 		bpf_prog_put(prog);
2264 	return err;
2265 
2266 free_used_maps:
2267 	/* In case we have subprogs, we need to wait for a grace
2268 	 * period before we can tear down JIT memory since symbols
2269 	 * are already exposed under kallsyms.
2270 	 */
2271 	__bpf_prog_put_noref(prog, prog->aux->func_cnt);
2272 	return err;
2273 free_prog:
2274 	bpf_prog_uncharge_memlock(prog);
2275 free_prog_sec:
2276 	security_bpf_prog_free(prog->aux);
2277 free_prog_nouncharge:
2278 	bpf_prog_free(prog);
2279 	return err;
2280 }
2281 
2282 #define BPF_OBJ_LAST_FIELD file_flags
2283 
bpf_obj_pin(const union bpf_attr * attr)2284 static int bpf_obj_pin(const union bpf_attr *attr)
2285 {
2286 	if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2287 		return -EINVAL;
2288 
2289 	return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2290 }
2291 
bpf_obj_get(const union bpf_attr * attr)2292 static int bpf_obj_get(const union bpf_attr *attr)
2293 {
2294 	if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2295 	    attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2296 		return -EINVAL;
2297 
2298 	return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2299 				attr->file_flags);
2300 }
2301 
bpf_link_init(struct bpf_link * link,enum bpf_link_type type,const struct bpf_link_ops * ops,struct bpf_prog * prog)2302 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2303 		   const struct bpf_link_ops *ops, struct bpf_prog *prog)
2304 {
2305 	atomic64_set(&link->refcnt, 1);
2306 	link->type = type;
2307 	link->id = 0;
2308 	link->ops = ops;
2309 	link->prog = prog;
2310 }
2311 
bpf_link_free_id(int id)2312 static void bpf_link_free_id(int id)
2313 {
2314 	if (!id)
2315 		return;
2316 
2317 	spin_lock_bh(&link_idr_lock);
2318 	idr_remove(&link_idr, id);
2319 	spin_unlock_bh(&link_idr_lock);
2320 }
2321 
2322 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2323  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2324  * anon_inode's release() call. This helper marksbpf_link as
2325  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2326  * is not decremented, it's the responsibility of a calling code that failed
2327  * to complete bpf_link initialization.
2328  */
bpf_link_cleanup(struct bpf_link_primer * primer)2329 void bpf_link_cleanup(struct bpf_link_primer *primer)
2330 {
2331 	primer->link->prog = NULL;
2332 	bpf_link_free_id(primer->id);
2333 	fput(primer->file);
2334 	put_unused_fd(primer->fd);
2335 }
2336 
bpf_link_inc(struct bpf_link * link)2337 void bpf_link_inc(struct bpf_link *link)
2338 {
2339 	atomic64_inc(&link->refcnt);
2340 }
2341 
2342 /* bpf_link_free is guaranteed to be called from process context */
bpf_link_free(struct bpf_link * link)2343 static void bpf_link_free(struct bpf_link *link)
2344 {
2345 	bpf_link_free_id(link->id);
2346 	if (link->prog) {
2347 		/* detach BPF program, clean up used resources */
2348 		link->ops->release(link);
2349 		bpf_prog_put(link->prog);
2350 	}
2351 	/* free bpf_link and its containing memory */
2352 	link->ops->dealloc(link);
2353 }
2354 
bpf_link_put_deferred(struct work_struct * work)2355 static void bpf_link_put_deferred(struct work_struct *work)
2356 {
2357 	struct bpf_link *link = container_of(work, struct bpf_link, work);
2358 
2359 	bpf_link_free(link);
2360 }
2361 
2362 /* bpf_link_put can be called from atomic context, but ensures that resources
2363  * are freed from process context
2364  */
bpf_link_put(struct bpf_link * link)2365 void bpf_link_put(struct bpf_link *link)
2366 {
2367 	if (!atomic64_dec_and_test(&link->refcnt))
2368 		return;
2369 
2370 	if (in_atomic()) {
2371 		INIT_WORK(&link->work, bpf_link_put_deferred);
2372 		schedule_work(&link->work);
2373 	} else {
2374 		bpf_link_free(link);
2375 	}
2376 }
2377 
bpf_link_release(struct inode * inode,struct file * filp)2378 static int bpf_link_release(struct inode *inode, struct file *filp)
2379 {
2380 	struct bpf_link *link = filp->private_data;
2381 
2382 	bpf_link_put(link);
2383 	return 0;
2384 }
2385 
2386 #ifdef CONFIG_PROC_FS
2387 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2388 #define BPF_MAP_TYPE(_id, _ops)
2389 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2390 static const char *bpf_link_type_strs[] = {
2391 	[BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2392 #include <linux/bpf_types.h>
2393 };
2394 #undef BPF_PROG_TYPE
2395 #undef BPF_MAP_TYPE
2396 #undef BPF_LINK_TYPE
2397 
bpf_link_show_fdinfo(struct seq_file * m,struct file * filp)2398 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2399 {
2400 	const struct bpf_link *link = filp->private_data;
2401 	const struct bpf_prog *prog = link->prog;
2402 	char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2403 
2404 	bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2405 	seq_printf(m,
2406 		   "link_type:\t%s\n"
2407 		   "link_id:\t%u\n"
2408 		   "prog_tag:\t%s\n"
2409 		   "prog_id:\t%u\n",
2410 		   bpf_link_type_strs[link->type],
2411 		   link->id,
2412 		   prog_tag,
2413 		   prog->aux->id);
2414 	if (link->ops->show_fdinfo)
2415 		link->ops->show_fdinfo(link, m);
2416 }
2417 #endif
2418 
2419 static const struct file_operations bpf_link_fops = {
2420 #ifdef CONFIG_PROC_FS
2421 	.show_fdinfo	= bpf_link_show_fdinfo,
2422 #endif
2423 	.release	= bpf_link_release,
2424 	.read		= bpf_dummy_read,
2425 	.write		= bpf_dummy_write,
2426 };
2427 
bpf_link_alloc_id(struct bpf_link * link)2428 static int bpf_link_alloc_id(struct bpf_link *link)
2429 {
2430 	int id;
2431 
2432 	idr_preload(GFP_KERNEL);
2433 	spin_lock_bh(&link_idr_lock);
2434 	id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2435 	spin_unlock_bh(&link_idr_lock);
2436 	idr_preload_end();
2437 
2438 	return id;
2439 }
2440 
2441 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2442  * reserving unused FD and allocating ID from link_idr. This is to be paired
2443  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2444  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2445  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2446  * transient state is passed around in struct bpf_link_primer.
2447  * This is preferred way to create and initialize bpf_link, especially when
2448  * there are complicated and expensive operations inbetween creating bpf_link
2449  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2450  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2451  * expensive (and potentially failing) roll back operations in a rare case
2452  * that file, FD, or ID can't be allocated.
2453  */
bpf_link_prime(struct bpf_link * link,struct bpf_link_primer * primer)2454 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2455 {
2456 	struct file *file;
2457 	int fd, id;
2458 
2459 	fd = get_unused_fd_flags(O_CLOEXEC);
2460 	if (fd < 0)
2461 		return fd;
2462 
2463 
2464 	id = bpf_link_alloc_id(link);
2465 	if (id < 0) {
2466 		put_unused_fd(fd);
2467 		return id;
2468 	}
2469 
2470 	file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2471 	if (IS_ERR(file)) {
2472 		bpf_link_free_id(id);
2473 		put_unused_fd(fd);
2474 		return PTR_ERR(file);
2475 	}
2476 
2477 	primer->link = link;
2478 	primer->file = file;
2479 	primer->fd = fd;
2480 	primer->id = id;
2481 	return 0;
2482 }
2483 
bpf_link_settle(struct bpf_link_primer * primer)2484 int bpf_link_settle(struct bpf_link_primer *primer)
2485 {
2486 	/* make bpf_link fetchable by ID */
2487 	spin_lock_bh(&link_idr_lock);
2488 	primer->link->id = primer->id;
2489 	spin_unlock_bh(&link_idr_lock);
2490 	/* make bpf_link fetchable by FD */
2491 	fd_install(primer->fd, primer->file);
2492 	/* pass through installed FD */
2493 	return primer->fd;
2494 }
2495 
bpf_link_new_fd(struct bpf_link * link)2496 int bpf_link_new_fd(struct bpf_link *link)
2497 {
2498 	return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2499 }
2500 
bpf_link_get_from_fd(u32 ufd)2501 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2502 {
2503 	struct fd f = fdget(ufd);
2504 	struct bpf_link *link;
2505 
2506 	if (!f.file)
2507 		return ERR_PTR(-EBADF);
2508 	if (f.file->f_op != &bpf_link_fops) {
2509 		fdput(f);
2510 		return ERR_PTR(-EINVAL);
2511 	}
2512 
2513 	link = f.file->private_data;
2514 	bpf_link_inc(link);
2515 	fdput(f);
2516 
2517 	return link;
2518 }
2519 
2520 struct bpf_tracing_link {
2521 	struct bpf_link link;
2522 	enum bpf_attach_type attach_type;
2523 	struct bpf_trampoline *trampoline;
2524 	struct bpf_prog *tgt_prog;
2525 };
2526 
bpf_tracing_link_release(struct bpf_link * link)2527 static void bpf_tracing_link_release(struct bpf_link *link)
2528 {
2529 	struct bpf_tracing_link *tr_link =
2530 		container_of(link, struct bpf_tracing_link, link);
2531 
2532 	WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2533 						tr_link->trampoline));
2534 
2535 	bpf_trampoline_put(tr_link->trampoline);
2536 
2537 	/* tgt_prog is NULL if target is a kernel function */
2538 	if (tr_link->tgt_prog)
2539 		bpf_prog_put(tr_link->tgt_prog);
2540 }
2541 
bpf_tracing_link_dealloc(struct bpf_link * link)2542 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2543 {
2544 	struct bpf_tracing_link *tr_link =
2545 		container_of(link, struct bpf_tracing_link, link);
2546 
2547 	kfree(tr_link);
2548 }
2549 
bpf_tracing_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)2550 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2551 					 struct seq_file *seq)
2552 {
2553 	struct bpf_tracing_link *tr_link =
2554 		container_of(link, struct bpf_tracing_link, link);
2555 
2556 	seq_printf(seq,
2557 		   "attach_type:\t%d\n",
2558 		   tr_link->attach_type);
2559 }
2560 
bpf_tracing_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)2561 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2562 					   struct bpf_link_info *info)
2563 {
2564 	struct bpf_tracing_link *tr_link =
2565 		container_of(link, struct bpf_tracing_link, link);
2566 
2567 	info->tracing.attach_type = tr_link->attach_type;
2568 
2569 	return 0;
2570 }
2571 
2572 static const struct bpf_link_ops bpf_tracing_link_lops = {
2573 	.release = bpf_tracing_link_release,
2574 	.dealloc = bpf_tracing_link_dealloc,
2575 	.show_fdinfo = bpf_tracing_link_show_fdinfo,
2576 	.fill_link_info = bpf_tracing_link_fill_link_info,
2577 };
2578 
bpf_tracing_prog_attach(struct bpf_prog * prog,int tgt_prog_fd,u32 btf_id)2579 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2580 				   int tgt_prog_fd,
2581 				   u32 btf_id)
2582 {
2583 	struct bpf_link_primer link_primer;
2584 	struct bpf_prog *tgt_prog = NULL;
2585 	struct bpf_trampoline *tr = NULL;
2586 	struct bpf_tracing_link *link;
2587 	u64 key = 0;
2588 	int err;
2589 
2590 	switch (prog->type) {
2591 	case BPF_PROG_TYPE_TRACING:
2592 		if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2593 		    prog->expected_attach_type != BPF_TRACE_FEXIT &&
2594 		    prog->expected_attach_type != BPF_MODIFY_RETURN) {
2595 			err = -EINVAL;
2596 			goto out_put_prog;
2597 		}
2598 		break;
2599 	case BPF_PROG_TYPE_EXT:
2600 		if (prog->expected_attach_type != 0) {
2601 			err = -EINVAL;
2602 			goto out_put_prog;
2603 		}
2604 		break;
2605 	case BPF_PROG_TYPE_LSM:
2606 		if (prog->expected_attach_type != BPF_LSM_MAC) {
2607 			err = -EINVAL;
2608 			goto out_put_prog;
2609 		}
2610 		break;
2611 	default:
2612 		err = -EINVAL;
2613 		goto out_put_prog;
2614 	}
2615 
2616 	if (!!tgt_prog_fd != !!btf_id) {
2617 		err = -EINVAL;
2618 		goto out_put_prog;
2619 	}
2620 
2621 	if (tgt_prog_fd) {
2622 		/* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2623 		if (prog->type != BPF_PROG_TYPE_EXT) {
2624 			err = -EINVAL;
2625 			goto out_put_prog;
2626 		}
2627 
2628 		tgt_prog = bpf_prog_get(tgt_prog_fd);
2629 		if (IS_ERR(tgt_prog)) {
2630 			err = PTR_ERR(tgt_prog);
2631 			tgt_prog = NULL;
2632 			goto out_put_prog;
2633 		}
2634 
2635 		key = bpf_trampoline_compute_key(tgt_prog, btf_id);
2636 	}
2637 
2638 	link = kzalloc(sizeof(*link), GFP_USER);
2639 	if (!link) {
2640 		err = -ENOMEM;
2641 		goto out_put_prog;
2642 	}
2643 	bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2644 		      &bpf_tracing_link_lops, prog);
2645 	link->attach_type = prog->expected_attach_type;
2646 
2647 	mutex_lock(&prog->aux->dst_mutex);
2648 
2649 	/* There are a few possible cases here:
2650 	 *
2651 	 * - if prog->aux->dst_trampoline is set, the program was just loaded
2652 	 *   and not yet attached to anything, so we can use the values stored
2653 	 *   in prog->aux
2654 	 *
2655 	 * - if prog->aux->dst_trampoline is NULL, the program has already been
2656          *   attached to a target and its initial target was cleared (below)
2657 	 *
2658 	 * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2659 	 *   target_btf_id using the link_create API.
2660 	 *
2661 	 * - if tgt_prog == NULL when this function was called using the old
2662          *   raw_tracepoint_open API, and we need a target from prog->aux
2663          *
2664          * The combination of no saved target in prog->aux, and no target
2665          * specified on load is illegal, and we reject that here.
2666 	 */
2667 	if (!prog->aux->dst_trampoline && !tgt_prog) {
2668 		err = -ENOENT;
2669 		goto out_unlock;
2670 	}
2671 
2672 	if (!prog->aux->dst_trampoline ||
2673 	    (key && key != prog->aux->dst_trampoline->key)) {
2674 		/* If there is no saved target, or the specified target is
2675 		 * different from the destination specified at load time, we
2676 		 * need a new trampoline and a check for compatibility
2677 		 */
2678 		struct bpf_attach_target_info tgt_info = {};
2679 
2680 		err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2681 					      &tgt_info);
2682 		if (err)
2683 			goto out_unlock;
2684 
2685 		tr = bpf_trampoline_get(key, &tgt_info);
2686 		if (!tr) {
2687 			err = -ENOMEM;
2688 			goto out_unlock;
2689 		}
2690 	} else {
2691 		/* The caller didn't specify a target, or the target was the
2692 		 * same as the destination supplied during program load. This
2693 		 * means we can reuse the trampoline and reference from program
2694 		 * load time, and there is no need to allocate a new one. This
2695 		 * can only happen once for any program, as the saved values in
2696 		 * prog->aux are cleared below.
2697 		 */
2698 		tr = prog->aux->dst_trampoline;
2699 		tgt_prog = prog->aux->dst_prog;
2700 	}
2701 
2702 	err = bpf_link_prime(&link->link, &link_primer);
2703 	if (err)
2704 		goto out_unlock;
2705 
2706 	err = bpf_trampoline_link_prog(prog, tr);
2707 	if (err) {
2708 		bpf_link_cleanup(&link_primer);
2709 		link = NULL;
2710 		goto out_unlock;
2711 	}
2712 
2713 	link->tgt_prog = tgt_prog;
2714 	link->trampoline = tr;
2715 
2716 	/* Always clear the trampoline and target prog from prog->aux to make
2717 	 * sure the original attach destination is not kept alive after a
2718 	 * program is (re-)attached to another target.
2719 	 */
2720 	if (prog->aux->dst_prog &&
2721 	    (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2722 		/* got extra prog ref from syscall, or attaching to different prog */
2723 		bpf_prog_put(prog->aux->dst_prog);
2724 	if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2725 		/* we allocated a new trampoline, so free the old one */
2726 		bpf_trampoline_put(prog->aux->dst_trampoline);
2727 
2728 	prog->aux->dst_prog = NULL;
2729 	prog->aux->dst_trampoline = NULL;
2730 	mutex_unlock(&prog->aux->dst_mutex);
2731 
2732 	return bpf_link_settle(&link_primer);
2733 out_unlock:
2734 	if (tr && tr != prog->aux->dst_trampoline)
2735 		bpf_trampoline_put(tr);
2736 	mutex_unlock(&prog->aux->dst_mutex);
2737 	kfree(link);
2738 out_put_prog:
2739 	if (tgt_prog_fd && tgt_prog)
2740 		bpf_prog_put(tgt_prog);
2741 	return err;
2742 }
2743 
2744 struct bpf_raw_tp_link {
2745 	struct bpf_link link;
2746 	struct bpf_raw_event_map *btp;
2747 };
2748 
bpf_raw_tp_link_release(struct bpf_link * link)2749 static void bpf_raw_tp_link_release(struct bpf_link *link)
2750 {
2751 	struct bpf_raw_tp_link *raw_tp =
2752 		container_of(link, struct bpf_raw_tp_link, link);
2753 
2754 	bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2755 	bpf_put_raw_tracepoint(raw_tp->btp);
2756 }
2757 
bpf_raw_tp_link_dealloc(struct bpf_link * link)2758 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2759 {
2760 	struct bpf_raw_tp_link *raw_tp =
2761 		container_of(link, struct bpf_raw_tp_link, link);
2762 
2763 	kfree(raw_tp);
2764 }
2765 
bpf_raw_tp_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)2766 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2767 					struct seq_file *seq)
2768 {
2769 	struct bpf_raw_tp_link *raw_tp_link =
2770 		container_of(link, struct bpf_raw_tp_link, link);
2771 
2772 	seq_printf(seq,
2773 		   "tp_name:\t%s\n",
2774 		   raw_tp_link->btp->tp->name);
2775 }
2776 
bpf_raw_tp_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)2777 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2778 					  struct bpf_link_info *info)
2779 {
2780 	struct bpf_raw_tp_link *raw_tp_link =
2781 		container_of(link, struct bpf_raw_tp_link, link);
2782 	char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2783 	const char *tp_name = raw_tp_link->btp->tp->name;
2784 	u32 ulen = info->raw_tracepoint.tp_name_len;
2785 	size_t tp_len = strlen(tp_name);
2786 
2787 	if (!ulen ^ !ubuf)
2788 		return -EINVAL;
2789 
2790 	info->raw_tracepoint.tp_name_len = tp_len + 1;
2791 
2792 	if (!ubuf)
2793 		return 0;
2794 
2795 	if (ulen >= tp_len + 1) {
2796 		if (copy_to_user(ubuf, tp_name, tp_len + 1))
2797 			return -EFAULT;
2798 	} else {
2799 		char zero = '\0';
2800 
2801 		if (copy_to_user(ubuf, tp_name, ulen - 1))
2802 			return -EFAULT;
2803 		if (put_user(zero, ubuf + ulen - 1))
2804 			return -EFAULT;
2805 		return -ENOSPC;
2806 	}
2807 
2808 	return 0;
2809 }
2810 
2811 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2812 	.release = bpf_raw_tp_link_release,
2813 	.dealloc = bpf_raw_tp_link_dealloc,
2814 	.show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2815 	.fill_link_info = bpf_raw_tp_link_fill_link_info,
2816 };
2817 
2818 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2819 
bpf_raw_tracepoint_open(const union bpf_attr * attr)2820 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2821 {
2822 	struct bpf_link_primer link_primer;
2823 	struct bpf_raw_tp_link *link;
2824 	struct bpf_raw_event_map *btp;
2825 	struct bpf_prog *prog;
2826 	const char *tp_name;
2827 	char buf[128];
2828 	int err;
2829 
2830 	if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2831 		return -EINVAL;
2832 
2833 	prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2834 	if (IS_ERR(prog))
2835 		return PTR_ERR(prog);
2836 
2837 	switch (prog->type) {
2838 	case BPF_PROG_TYPE_TRACING:
2839 	case BPF_PROG_TYPE_EXT:
2840 	case BPF_PROG_TYPE_LSM:
2841 		if (attr->raw_tracepoint.name) {
2842 			/* The attach point for this category of programs
2843 			 * should be specified via btf_id during program load.
2844 			 */
2845 			err = -EINVAL;
2846 			goto out_put_prog;
2847 		}
2848 		if (prog->type == BPF_PROG_TYPE_TRACING &&
2849 		    prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2850 			tp_name = prog->aux->attach_func_name;
2851 			break;
2852 		}
2853 		err = bpf_tracing_prog_attach(prog, 0, 0);
2854 		if (err >= 0)
2855 			return err;
2856 		goto out_put_prog;
2857 	case BPF_PROG_TYPE_RAW_TRACEPOINT:
2858 	case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2859 		if (strncpy_from_user(buf,
2860 				      u64_to_user_ptr(attr->raw_tracepoint.name),
2861 				      sizeof(buf) - 1) < 0) {
2862 			err = -EFAULT;
2863 			goto out_put_prog;
2864 		}
2865 		buf[sizeof(buf) - 1] = 0;
2866 		tp_name = buf;
2867 		break;
2868 	default:
2869 		err = -EINVAL;
2870 		goto out_put_prog;
2871 	}
2872 
2873 	btp = bpf_get_raw_tracepoint(tp_name);
2874 	if (!btp) {
2875 		err = -ENOENT;
2876 		goto out_put_prog;
2877 	}
2878 
2879 	link = kzalloc(sizeof(*link), GFP_USER);
2880 	if (!link) {
2881 		err = -ENOMEM;
2882 		goto out_put_btp;
2883 	}
2884 	bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2885 		      &bpf_raw_tp_link_lops, prog);
2886 	link->btp = btp;
2887 
2888 	err = bpf_link_prime(&link->link, &link_primer);
2889 	if (err) {
2890 		kfree(link);
2891 		goto out_put_btp;
2892 	}
2893 
2894 	err = bpf_probe_register(link->btp, prog);
2895 	if (err) {
2896 		bpf_link_cleanup(&link_primer);
2897 		goto out_put_btp;
2898 	}
2899 
2900 	return bpf_link_settle(&link_primer);
2901 
2902 out_put_btp:
2903 	bpf_put_raw_tracepoint(btp);
2904 out_put_prog:
2905 	bpf_prog_put(prog);
2906 	return err;
2907 }
2908 
bpf_prog_attach_check_attach_type(const struct bpf_prog * prog,enum bpf_attach_type attach_type)2909 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2910 					     enum bpf_attach_type attach_type)
2911 {
2912 	switch (prog->type) {
2913 	case BPF_PROG_TYPE_CGROUP_SOCK:
2914 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2915 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2916 	case BPF_PROG_TYPE_SK_LOOKUP:
2917 		return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2918 	case BPF_PROG_TYPE_CGROUP_SKB:
2919 		if (!capable(CAP_NET_ADMIN))
2920 			/* cg-skb progs can be loaded by unpriv user.
2921 			 * check permissions at attach time.
2922 			 */
2923 			return -EPERM;
2924 		return prog->enforce_expected_attach_type &&
2925 			prog->expected_attach_type != attach_type ?
2926 			-EINVAL : 0;
2927 	default:
2928 		return 0;
2929 	}
2930 }
2931 
2932 static enum bpf_prog_type
attach_type_to_prog_type(enum bpf_attach_type attach_type)2933 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2934 {
2935 	switch (attach_type) {
2936 	case BPF_CGROUP_INET_INGRESS:
2937 	case BPF_CGROUP_INET_EGRESS:
2938 		return BPF_PROG_TYPE_CGROUP_SKB;
2939 	case BPF_CGROUP_INET_SOCK_CREATE:
2940 	case BPF_CGROUP_INET_SOCK_RELEASE:
2941 	case BPF_CGROUP_INET4_POST_BIND:
2942 	case BPF_CGROUP_INET6_POST_BIND:
2943 		return BPF_PROG_TYPE_CGROUP_SOCK;
2944 	case BPF_CGROUP_INET4_BIND:
2945 	case BPF_CGROUP_INET6_BIND:
2946 	case BPF_CGROUP_INET4_CONNECT:
2947 	case BPF_CGROUP_INET6_CONNECT:
2948 	case BPF_CGROUP_INET4_GETPEERNAME:
2949 	case BPF_CGROUP_INET6_GETPEERNAME:
2950 	case BPF_CGROUP_INET4_GETSOCKNAME:
2951 	case BPF_CGROUP_INET6_GETSOCKNAME:
2952 	case BPF_CGROUP_UDP4_SENDMSG:
2953 	case BPF_CGROUP_UDP6_SENDMSG:
2954 	case BPF_CGROUP_UDP4_RECVMSG:
2955 	case BPF_CGROUP_UDP6_RECVMSG:
2956 		return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2957 	case BPF_CGROUP_SOCK_OPS:
2958 		return BPF_PROG_TYPE_SOCK_OPS;
2959 	case BPF_CGROUP_DEVICE:
2960 		return BPF_PROG_TYPE_CGROUP_DEVICE;
2961 	case BPF_SK_MSG_VERDICT:
2962 		return BPF_PROG_TYPE_SK_MSG;
2963 	case BPF_SK_SKB_STREAM_PARSER:
2964 	case BPF_SK_SKB_STREAM_VERDICT:
2965 		return BPF_PROG_TYPE_SK_SKB;
2966 	case BPF_LIRC_MODE2:
2967 		return BPF_PROG_TYPE_LIRC_MODE2;
2968 	case BPF_FLOW_DISSECTOR:
2969 		return BPF_PROG_TYPE_FLOW_DISSECTOR;
2970 	case BPF_CGROUP_SYSCTL:
2971 		return BPF_PROG_TYPE_CGROUP_SYSCTL;
2972 	case BPF_CGROUP_GETSOCKOPT:
2973 	case BPF_CGROUP_SETSOCKOPT:
2974 		return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2975 	case BPF_TRACE_ITER:
2976 		return BPF_PROG_TYPE_TRACING;
2977 	case BPF_SK_LOOKUP:
2978 		return BPF_PROG_TYPE_SK_LOOKUP;
2979 	case BPF_XDP:
2980 		return BPF_PROG_TYPE_XDP;
2981 	default:
2982 		return BPF_PROG_TYPE_UNSPEC;
2983 	}
2984 }
2985 
2986 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2987 
2988 #define BPF_F_ATTACH_MASK \
2989 	(BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2990 
bpf_prog_attach(const union bpf_attr * attr)2991 static int bpf_prog_attach(const union bpf_attr *attr)
2992 {
2993 	enum bpf_prog_type ptype;
2994 	struct bpf_prog *prog;
2995 	int ret;
2996 
2997 	if (CHECK_ATTR(BPF_PROG_ATTACH))
2998 		return -EINVAL;
2999 
3000 	if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3001 		return -EINVAL;
3002 
3003 	ptype = attach_type_to_prog_type(attr->attach_type);
3004 	if (ptype == BPF_PROG_TYPE_UNSPEC)
3005 		return -EINVAL;
3006 
3007 	prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3008 	if (IS_ERR(prog))
3009 		return PTR_ERR(prog);
3010 
3011 	if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3012 		bpf_prog_put(prog);
3013 		return -EINVAL;
3014 	}
3015 
3016 	switch (ptype) {
3017 	case BPF_PROG_TYPE_SK_SKB:
3018 	case BPF_PROG_TYPE_SK_MSG:
3019 		ret = sock_map_get_from_fd(attr, prog);
3020 		break;
3021 	case BPF_PROG_TYPE_LIRC_MODE2:
3022 		ret = lirc_prog_attach(attr, prog);
3023 		break;
3024 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3025 		ret = netns_bpf_prog_attach(attr, prog);
3026 		break;
3027 	case BPF_PROG_TYPE_CGROUP_DEVICE:
3028 	case BPF_PROG_TYPE_CGROUP_SKB:
3029 	case BPF_PROG_TYPE_CGROUP_SOCK:
3030 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3031 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3032 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3033 	case BPF_PROG_TYPE_SOCK_OPS:
3034 		ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3035 		break;
3036 	default:
3037 		ret = -EINVAL;
3038 	}
3039 
3040 	if (ret)
3041 		bpf_prog_put(prog);
3042 	return ret;
3043 }
3044 
3045 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3046 
bpf_prog_detach(const union bpf_attr * attr)3047 static int bpf_prog_detach(const union bpf_attr *attr)
3048 {
3049 	enum bpf_prog_type ptype;
3050 
3051 	if (CHECK_ATTR(BPF_PROG_DETACH))
3052 		return -EINVAL;
3053 
3054 	ptype = attach_type_to_prog_type(attr->attach_type);
3055 
3056 	switch (ptype) {
3057 	case BPF_PROG_TYPE_SK_MSG:
3058 	case BPF_PROG_TYPE_SK_SKB:
3059 		return sock_map_prog_detach(attr, ptype);
3060 	case BPF_PROG_TYPE_LIRC_MODE2:
3061 		return lirc_prog_detach(attr);
3062 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
3063 		return netns_bpf_prog_detach(attr, ptype);
3064 	case BPF_PROG_TYPE_CGROUP_DEVICE:
3065 	case BPF_PROG_TYPE_CGROUP_SKB:
3066 	case BPF_PROG_TYPE_CGROUP_SOCK:
3067 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3068 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3069 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
3070 	case BPF_PROG_TYPE_SOCK_OPS:
3071 		return cgroup_bpf_prog_detach(attr, ptype);
3072 	default:
3073 		return -EINVAL;
3074 	}
3075 }
3076 
3077 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3078 
bpf_prog_query(const union bpf_attr * attr,union bpf_attr __user * uattr)3079 static int bpf_prog_query(const union bpf_attr *attr,
3080 			  union bpf_attr __user *uattr)
3081 {
3082 	if (!capable(CAP_NET_ADMIN))
3083 		return -EPERM;
3084 	if (CHECK_ATTR(BPF_PROG_QUERY))
3085 		return -EINVAL;
3086 	if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3087 		return -EINVAL;
3088 
3089 	switch (attr->query.attach_type) {
3090 	case BPF_CGROUP_INET_INGRESS:
3091 	case BPF_CGROUP_INET_EGRESS:
3092 	case BPF_CGROUP_INET_SOCK_CREATE:
3093 	case BPF_CGROUP_INET_SOCK_RELEASE:
3094 	case BPF_CGROUP_INET4_BIND:
3095 	case BPF_CGROUP_INET6_BIND:
3096 	case BPF_CGROUP_INET4_POST_BIND:
3097 	case BPF_CGROUP_INET6_POST_BIND:
3098 	case BPF_CGROUP_INET4_CONNECT:
3099 	case BPF_CGROUP_INET6_CONNECT:
3100 	case BPF_CGROUP_INET4_GETPEERNAME:
3101 	case BPF_CGROUP_INET6_GETPEERNAME:
3102 	case BPF_CGROUP_INET4_GETSOCKNAME:
3103 	case BPF_CGROUP_INET6_GETSOCKNAME:
3104 	case BPF_CGROUP_UDP4_SENDMSG:
3105 	case BPF_CGROUP_UDP6_SENDMSG:
3106 	case BPF_CGROUP_UDP4_RECVMSG:
3107 	case BPF_CGROUP_UDP6_RECVMSG:
3108 	case BPF_CGROUP_SOCK_OPS:
3109 	case BPF_CGROUP_DEVICE:
3110 	case BPF_CGROUP_SYSCTL:
3111 	case BPF_CGROUP_GETSOCKOPT:
3112 	case BPF_CGROUP_SETSOCKOPT:
3113 		return cgroup_bpf_prog_query(attr, uattr);
3114 	case BPF_LIRC_MODE2:
3115 		return lirc_prog_query(attr, uattr);
3116 	case BPF_FLOW_DISSECTOR:
3117 	case BPF_SK_LOOKUP:
3118 		return netns_bpf_prog_query(attr, uattr);
3119 	default:
3120 		return -EINVAL;
3121 	}
3122 }
3123 
3124 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3125 
bpf_prog_test_run(const union bpf_attr * attr,union bpf_attr __user * uattr)3126 static int bpf_prog_test_run(const union bpf_attr *attr,
3127 			     union bpf_attr __user *uattr)
3128 {
3129 	struct bpf_prog *prog;
3130 	int ret = -ENOTSUPP;
3131 
3132 	if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3133 		return -EINVAL;
3134 
3135 	if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3136 	    (!attr->test.ctx_size_in && attr->test.ctx_in))
3137 		return -EINVAL;
3138 
3139 	if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3140 	    (!attr->test.ctx_size_out && attr->test.ctx_out))
3141 		return -EINVAL;
3142 
3143 	prog = bpf_prog_get(attr->test.prog_fd);
3144 	if (IS_ERR(prog))
3145 		return PTR_ERR(prog);
3146 
3147 	if (prog->aux->ops->test_run)
3148 		ret = prog->aux->ops->test_run(prog, attr, uattr);
3149 
3150 	bpf_prog_put(prog);
3151 	return ret;
3152 }
3153 
3154 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3155 
bpf_obj_get_next_id(const union bpf_attr * attr,union bpf_attr __user * uattr,struct idr * idr,spinlock_t * lock)3156 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3157 			       union bpf_attr __user *uattr,
3158 			       struct idr *idr,
3159 			       spinlock_t *lock)
3160 {
3161 	u32 next_id = attr->start_id;
3162 	int err = 0;
3163 
3164 	if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3165 		return -EINVAL;
3166 
3167 	if (!capable(CAP_SYS_ADMIN))
3168 		return -EPERM;
3169 
3170 	next_id++;
3171 	spin_lock_bh(lock);
3172 	if (!idr_get_next(idr, &next_id))
3173 		err = -ENOENT;
3174 	spin_unlock_bh(lock);
3175 
3176 	if (!err)
3177 		err = put_user(next_id, &uattr->next_id);
3178 
3179 	return err;
3180 }
3181 
bpf_map_get_curr_or_next(u32 * id)3182 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3183 {
3184 	struct bpf_map *map;
3185 
3186 	spin_lock_bh(&map_idr_lock);
3187 again:
3188 	map = idr_get_next(&map_idr, id);
3189 	if (map) {
3190 		map = __bpf_map_inc_not_zero(map, false);
3191 		if (IS_ERR(map)) {
3192 			(*id)++;
3193 			goto again;
3194 		}
3195 	}
3196 	spin_unlock_bh(&map_idr_lock);
3197 
3198 	return map;
3199 }
3200 
bpf_prog_get_curr_or_next(u32 * id)3201 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3202 {
3203 	struct bpf_prog *prog;
3204 
3205 	spin_lock_bh(&prog_idr_lock);
3206 again:
3207 	prog = idr_get_next(&prog_idr, id);
3208 	if (prog) {
3209 		prog = bpf_prog_inc_not_zero(prog);
3210 		if (IS_ERR(prog)) {
3211 			(*id)++;
3212 			goto again;
3213 		}
3214 	}
3215 	spin_unlock_bh(&prog_idr_lock);
3216 
3217 	return prog;
3218 }
3219 
3220 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3221 
bpf_prog_by_id(u32 id)3222 struct bpf_prog *bpf_prog_by_id(u32 id)
3223 {
3224 	struct bpf_prog *prog;
3225 
3226 	if (!id)
3227 		return ERR_PTR(-ENOENT);
3228 
3229 	spin_lock_bh(&prog_idr_lock);
3230 	prog = idr_find(&prog_idr, id);
3231 	if (prog)
3232 		prog = bpf_prog_inc_not_zero(prog);
3233 	else
3234 		prog = ERR_PTR(-ENOENT);
3235 	spin_unlock_bh(&prog_idr_lock);
3236 	return prog;
3237 }
3238 
bpf_prog_get_fd_by_id(const union bpf_attr * attr)3239 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3240 {
3241 	struct bpf_prog *prog;
3242 	u32 id = attr->prog_id;
3243 	int fd;
3244 
3245 	if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3246 		return -EINVAL;
3247 
3248 	if (!capable(CAP_SYS_ADMIN))
3249 		return -EPERM;
3250 
3251 	prog = bpf_prog_by_id(id);
3252 	if (IS_ERR(prog))
3253 		return PTR_ERR(prog);
3254 
3255 	fd = bpf_prog_new_fd(prog);
3256 	if (fd < 0)
3257 		bpf_prog_put(prog);
3258 
3259 	return fd;
3260 }
3261 
3262 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3263 
bpf_map_get_fd_by_id(const union bpf_attr * attr)3264 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3265 {
3266 	struct bpf_map *map;
3267 	u32 id = attr->map_id;
3268 	int f_flags;
3269 	int fd;
3270 
3271 	if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3272 	    attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3273 		return -EINVAL;
3274 
3275 	if (!capable(CAP_SYS_ADMIN))
3276 		return -EPERM;
3277 
3278 	f_flags = bpf_get_file_flag(attr->open_flags);
3279 	if (f_flags < 0)
3280 		return f_flags;
3281 
3282 	spin_lock_bh(&map_idr_lock);
3283 	map = idr_find(&map_idr, id);
3284 	if (map)
3285 		map = __bpf_map_inc_not_zero(map, true);
3286 	else
3287 		map = ERR_PTR(-ENOENT);
3288 	spin_unlock_bh(&map_idr_lock);
3289 
3290 	if (IS_ERR(map))
3291 		return PTR_ERR(map);
3292 
3293 	fd = bpf_map_new_fd(map, f_flags);
3294 	if (fd < 0)
3295 		bpf_map_put_with_uref(map);
3296 
3297 	return fd;
3298 }
3299 
bpf_map_from_imm(const struct bpf_prog * prog,unsigned long addr,u32 * off,u32 * type)3300 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3301 					      unsigned long addr, u32 *off,
3302 					      u32 *type)
3303 {
3304 	const struct bpf_map *map;
3305 	int i;
3306 
3307 	mutex_lock(&prog->aux->used_maps_mutex);
3308 	for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3309 		map = prog->aux->used_maps[i];
3310 		if (map == (void *)addr) {
3311 			*type = BPF_PSEUDO_MAP_FD;
3312 			goto out;
3313 		}
3314 		if (!map->ops->map_direct_value_meta)
3315 			continue;
3316 		if (!map->ops->map_direct_value_meta(map, addr, off)) {
3317 			*type = BPF_PSEUDO_MAP_VALUE;
3318 			goto out;
3319 		}
3320 	}
3321 	map = NULL;
3322 
3323 out:
3324 	mutex_unlock(&prog->aux->used_maps_mutex);
3325 	return map;
3326 }
3327 
bpf_insn_prepare_dump(const struct bpf_prog * prog,const struct cred * f_cred)3328 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3329 					      const struct cred *f_cred)
3330 {
3331 	const struct bpf_map *map;
3332 	struct bpf_insn *insns;
3333 	u32 off, type;
3334 	u64 imm;
3335 	u8 code;
3336 	int i;
3337 
3338 	insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3339 			GFP_USER);
3340 	if (!insns)
3341 		return insns;
3342 
3343 	for (i = 0; i < prog->len; i++) {
3344 		code = insns[i].code;
3345 
3346 		if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3347 			insns[i].code = BPF_JMP | BPF_CALL;
3348 			insns[i].imm = BPF_FUNC_tail_call;
3349 			/* fall-through */
3350 		}
3351 		if (code == (BPF_JMP | BPF_CALL) ||
3352 		    code == (BPF_JMP | BPF_CALL_ARGS)) {
3353 			if (code == (BPF_JMP | BPF_CALL_ARGS))
3354 				insns[i].code = BPF_JMP | BPF_CALL;
3355 			if (!bpf_dump_raw_ok(f_cred))
3356 				insns[i].imm = 0;
3357 			continue;
3358 		}
3359 		if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3360 			insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3361 			continue;
3362 		}
3363 
3364 		if (code != (BPF_LD | BPF_IMM | BPF_DW))
3365 			continue;
3366 
3367 		imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3368 		map = bpf_map_from_imm(prog, imm, &off, &type);
3369 		if (map) {
3370 			insns[i].src_reg = type;
3371 			insns[i].imm = map->id;
3372 			insns[i + 1].imm = off;
3373 			continue;
3374 		}
3375 	}
3376 
3377 	return insns;
3378 }
3379 
set_info_rec_size(struct bpf_prog_info * info)3380 static int set_info_rec_size(struct bpf_prog_info *info)
3381 {
3382 	/*
3383 	 * Ensure info.*_rec_size is the same as kernel expected size
3384 	 *
3385 	 * or
3386 	 *
3387 	 * Only allow zero *_rec_size if both _rec_size and _cnt are
3388 	 * zero.  In this case, the kernel will set the expected
3389 	 * _rec_size back to the info.
3390 	 */
3391 
3392 	if ((info->nr_func_info || info->func_info_rec_size) &&
3393 	    info->func_info_rec_size != sizeof(struct bpf_func_info))
3394 		return -EINVAL;
3395 
3396 	if ((info->nr_line_info || info->line_info_rec_size) &&
3397 	    info->line_info_rec_size != sizeof(struct bpf_line_info))
3398 		return -EINVAL;
3399 
3400 	if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3401 	    info->jited_line_info_rec_size != sizeof(__u64))
3402 		return -EINVAL;
3403 
3404 	info->func_info_rec_size = sizeof(struct bpf_func_info);
3405 	info->line_info_rec_size = sizeof(struct bpf_line_info);
3406 	info->jited_line_info_rec_size = sizeof(__u64);
3407 
3408 	return 0;
3409 }
3410 
bpf_prog_get_info_by_fd(struct file * file,struct bpf_prog * prog,const union bpf_attr * attr,union bpf_attr __user * uattr)3411 static int bpf_prog_get_info_by_fd(struct file *file,
3412 				   struct bpf_prog *prog,
3413 				   const union bpf_attr *attr,
3414 				   union bpf_attr __user *uattr)
3415 {
3416 	struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3417 	struct bpf_prog_info info;
3418 	u32 info_len = attr->info.info_len;
3419 	struct bpf_prog_stats stats;
3420 	char __user *uinsns;
3421 	u32 ulen;
3422 	int err;
3423 
3424 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3425 	if (err)
3426 		return err;
3427 	info_len = min_t(u32, sizeof(info), info_len);
3428 
3429 	memset(&info, 0, sizeof(info));
3430 	if (copy_from_user(&info, uinfo, info_len))
3431 		return -EFAULT;
3432 
3433 	info.type = prog->type;
3434 	info.id = prog->aux->id;
3435 	info.load_time = prog->aux->load_time;
3436 	info.created_by_uid = from_kuid_munged(current_user_ns(),
3437 					       prog->aux->user->uid);
3438 	info.gpl_compatible = prog->gpl_compatible;
3439 
3440 	memcpy(info.tag, prog->tag, sizeof(prog->tag));
3441 	memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3442 
3443 	mutex_lock(&prog->aux->used_maps_mutex);
3444 	ulen = info.nr_map_ids;
3445 	info.nr_map_ids = prog->aux->used_map_cnt;
3446 	ulen = min_t(u32, info.nr_map_ids, ulen);
3447 	if (ulen) {
3448 		u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3449 		u32 i;
3450 
3451 		for (i = 0; i < ulen; i++)
3452 			if (put_user(prog->aux->used_maps[i]->id,
3453 				     &user_map_ids[i])) {
3454 				mutex_unlock(&prog->aux->used_maps_mutex);
3455 				return -EFAULT;
3456 			}
3457 	}
3458 	mutex_unlock(&prog->aux->used_maps_mutex);
3459 
3460 	err = set_info_rec_size(&info);
3461 	if (err)
3462 		return err;
3463 
3464 	bpf_prog_get_stats(prog, &stats);
3465 	info.run_time_ns = stats.nsecs;
3466 	info.run_cnt = stats.cnt;
3467 
3468 	if (!bpf_capable()) {
3469 		info.jited_prog_len = 0;
3470 		info.xlated_prog_len = 0;
3471 		info.nr_jited_ksyms = 0;
3472 		info.nr_jited_func_lens = 0;
3473 		info.nr_func_info = 0;
3474 		info.nr_line_info = 0;
3475 		info.nr_jited_line_info = 0;
3476 		goto done;
3477 	}
3478 
3479 	ulen = info.xlated_prog_len;
3480 	info.xlated_prog_len = bpf_prog_insn_size(prog);
3481 	if (info.xlated_prog_len && ulen) {
3482 		struct bpf_insn *insns_sanitized;
3483 		bool fault;
3484 
3485 		if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3486 			info.xlated_prog_insns = 0;
3487 			goto done;
3488 		}
3489 		insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3490 		if (!insns_sanitized)
3491 			return -ENOMEM;
3492 		uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3493 		ulen = min_t(u32, info.xlated_prog_len, ulen);
3494 		fault = copy_to_user(uinsns, insns_sanitized, ulen);
3495 		kfree(insns_sanitized);
3496 		if (fault)
3497 			return -EFAULT;
3498 	}
3499 
3500 	if (bpf_prog_is_dev_bound(prog->aux)) {
3501 		err = bpf_prog_offload_info_fill(&info, prog);
3502 		if (err)
3503 			return err;
3504 		goto done;
3505 	}
3506 
3507 	/* NOTE: the following code is supposed to be skipped for offload.
3508 	 * bpf_prog_offload_info_fill() is the place to fill similar fields
3509 	 * for offload.
3510 	 */
3511 	ulen = info.jited_prog_len;
3512 	if (prog->aux->func_cnt) {
3513 		u32 i;
3514 
3515 		info.jited_prog_len = 0;
3516 		for (i = 0; i < prog->aux->func_cnt; i++)
3517 			info.jited_prog_len += prog->aux->func[i]->jited_len;
3518 	} else {
3519 		info.jited_prog_len = prog->jited_len;
3520 	}
3521 
3522 	if (info.jited_prog_len && ulen) {
3523 		if (bpf_dump_raw_ok(file->f_cred)) {
3524 			uinsns = u64_to_user_ptr(info.jited_prog_insns);
3525 			ulen = min_t(u32, info.jited_prog_len, ulen);
3526 
3527 			/* for multi-function programs, copy the JITed
3528 			 * instructions for all the functions
3529 			 */
3530 			if (prog->aux->func_cnt) {
3531 				u32 len, free, i;
3532 				u8 *img;
3533 
3534 				free = ulen;
3535 				for (i = 0; i < prog->aux->func_cnt; i++) {
3536 					len = prog->aux->func[i]->jited_len;
3537 					len = min_t(u32, len, free);
3538 					img = (u8 *) prog->aux->func[i]->bpf_func;
3539 					if (copy_to_user(uinsns, img, len))
3540 						return -EFAULT;
3541 					uinsns += len;
3542 					free -= len;
3543 					if (!free)
3544 						break;
3545 				}
3546 			} else {
3547 				if (copy_to_user(uinsns, prog->bpf_func, ulen))
3548 					return -EFAULT;
3549 			}
3550 		} else {
3551 			info.jited_prog_insns = 0;
3552 		}
3553 	}
3554 
3555 	ulen = info.nr_jited_ksyms;
3556 	info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3557 	if (ulen) {
3558 		if (bpf_dump_raw_ok(file->f_cred)) {
3559 			unsigned long ksym_addr;
3560 			u64 __user *user_ksyms;
3561 			u32 i;
3562 
3563 			/* copy the address of the kernel symbol
3564 			 * corresponding to each function
3565 			 */
3566 			ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3567 			user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3568 			if (prog->aux->func_cnt) {
3569 				for (i = 0; i < ulen; i++) {
3570 					ksym_addr = (unsigned long)
3571 						prog->aux->func[i]->bpf_func;
3572 					if (put_user((u64) ksym_addr,
3573 						     &user_ksyms[i]))
3574 						return -EFAULT;
3575 				}
3576 			} else {
3577 				ksym_addr = (unsigned long) prog->bpf_func;
3578 				if (put_user((u64) ksym_addr, &user_ksyms[0]))
3579 					return -EFAULT;
3580 			}
3581 		} else {
3582 			info.jited_ksyms = 0;
3583 		}
3584 	}
3585 
3586 	ulen = info.nr_jited_func_lens;
3587 	info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3588 	if (ulen) {
3589 		if (bpf_dump_raw_ok(file->f_cred)) {
3590 			u32 __user *user_lens;
3591 			u32 func_len, i;
3592 
3593 			/* copy the JITed image lengths for each function */
3594 			ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3595 			user_lens = u64_to_user_ptr(info.jited_func_lens);
3596 			if (prog->aux->func_cnt) {
3597 				for (i = 0; i < ulen; i++) {
3598 					func_len =
3599 						prog->aux->func[i]->jited_len;
3600 					if (put_user(func_len, &user_lens[i]))
3601 						return -EFAULT;
3602 				}
3603 			} else {
3604 				func_len = prog->jited_len;
3605 				if (put_user(func_len, &user_lens[0]))
3606 					return -EFAULT;
3607 			}
3608 		} else {
3609 			info.jited_func_lens = 0;
3610 		}
3611 	}
3612 
3613 	if (prog->aux->btf)
3614 		info.btf_id = btf_id(prog->aux->btf);
3615 
3616 	ulen = info.nr_func_info;
3617 	info.nr_func_info = prog->aux->func_info_cnt;
3618 	if (info.nr_func_info && ulen) {
3619 		char __user *user_finfo;
3620 
3621 		user_finfo = u64_to_user_ptr(info.func_info);
3622 		ulen = min_t(u32, info.nr_func_info, ulen);
3623 		if (copy_to_user(user_finfo, prog->aux->func_info,
3624 				 info.func_info_rec_size * ulen))
3625 			return -EFAULT;
3626 	}
3627 
3628 	ulen = info.nr_line_info;
3629 	info.nr_line_info = prog->aux->nr_linfo;
3630 	if (info.nr_line_info && ulen) {
3631 		__u8 __user *user_linfo;
3632 
3633 		user_linfo = u64_to_user_ptr(info.line_info);
3634 		ulen = min_t(u32, info.nr_line_info, ulen);
3635 		if (copy_to_user(user_linfo, prog->aux->linfo,
3636 				 info.line_info_rec_size * ulen))
3637 			return -EFAULT;
3638 	}
3639 
3640 	ulen = info.nr_jited_line_info;
3641 	if (prog->aux->jited_linfo)
3642 		info.nr_jited_line_info = prog->aux->nr_linfo;
3643 	else
3644 		info.nr_jited_line_info = 0;
3645 	if (info.nr_jited_line_info && ulen) {
3646 		if (bpf_dump_raw_ok(file->f_cred)) {
3647 			__u64 __user *user_linfo;
3648 			u32 i;
3649 
3650 			user_linfo = u64_to_user_ptr(info.jited_line_info);
3651 			ulen = min_t(u32, info.nr_jited_line_info, ulen);
3652 			for (i = 0; i < ulen; i++) {
3653 				if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3654 					     &user_linfo[i]))
3655 					return -EFAULT;
3656 			}
3657 		} else {
3658 			info.jited_line_info = 0;
3659 		}
3660 	}
3661 
3662 	ulen = info.nr_prog_tags;
3663 	info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3664 	if (ulen) {
3665 		__u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3666 		u32 i;
3667 
3668 		user_prog_tags = u64_to_user_ptr(info.prog_tags);
3669 		ulen = min_t(u32, info.nr_prog_tags, ulen);
3670 		if (prog->aux->func_cnt) {
3671 			for (i = 0; i < ulen; i++) {
3672 				if (copy_to_user(user_prog_tags[i],
3673 						 prog->aux->func[i]->tag,
3674 						 BPF_TAG_SIZE))
3675 					return -EFAULT;
3676 			}
3677 		} else {
3678 			if (copy_to_user(user_prog_tags[0],
3679 					 prog->tag, BPF_TAG_SIZE))
3680 				return -EFAULT;
3681 		}
3682 	}
3683 
3684 done:
3685 	if (copy_to_user(uinfo, &info, info_len) ||
3686 	    put_user(info_len, &uattr->info.info_len))
3687 		return -EFAULT;
3688 
3689 	return 0;
3690 }
3691 
bpf_map_get_info_by_fd(struct file * file,struct bpf_map * map,const union bpf_attr * attr,union bpf_attr __user * uattr)3692 static int bpf_map_get_info_by_fd(struct file *file,
3693 				  struct bpf_map *map,
3694 				  const union bpf_attr *attr,
3695 				  union bpf_attr __user *uattr)
3696 {
3697 	struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3698 	struct bpf_map_info info;
3699 	u32 info_len = attr->info.info_len;
3700 	int err;
3701 
3702 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3703 	if (err)
3704 		return err;
3705 	info_len = min_t(u32, sizeof(info), info_len);
3706 
3707 	memset(&info, 0, sizeof(info));
3708 	info.type = map->map_type;
3709 	info.id = map->id;
3710 	info.key_size = map->key_size;
3711 	info.value_size = map->value_size;
3712 	info.max_entries = map->max_entries;
3713 	info.map_flags = map->map_flags;
3714 	memcpy(info.name, map->name, sizeof(map->name));
3715 
3716 	if (map->btf) {
3717 		info.btf_id = btf_id(map->btf);
3718 		info.btf_key_type_id = map->btf_key_type_id;
3719 		info.btf_value_type_id = map->btf_value_type_id;
3720 	}
3721 	info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3722 
3723 	if (bpf_map_is_dev_bound(map)) {
3724 		err = bpf_map_offload_info_fill(&info, map);
3725 		if (err)
3726 			return err;
3727 	}
3728 
3729 	if (copy_to_user(uinfo, &info, info_len) ||
3730 	    put_user(info_len, &uattr->info.info_len))
3731 		return -EFAULT;
3732 
3733 	return 0;
3734 }
3735 
bpf_btf_get_info_by_fd(struct file * file,struct btf * btf,const union bpf_attr * attr,union bpf_attr __user * uattr)3736 static int bpf_btf_get_info_by_fd(struct file *file,
3737 				  struct btf *btf,
3738 				  const union bpf_attr *attr,
3739 				  union bpf_attr __user *uattr)
3740 {
3741 	struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3742 	u32 info_len = attr->info.info_len;
3743 	int err;
3744 
3745 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3746 	if (err)
3747 		return err;
3748 
3749 	return btf_get_info_by_fd(btf, attr, uattr);
3750 }
3751 
bpf_link_get_info_by_fd(struct file * file,struct bpf_link * link,const union bpf_attr * attr,union bpf_attr __user * uattr)3752 static int bpf_link_get_info_by_fd(struct file *file,
3753 				  struct bpf_link *link,
3754 				  const union bpf_attr *attr,
3755 				  union bpf_attr __user *uattr)
3756 {
3757 	struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3758 	struct bpf_link_info info;
3759 	u32 info_len = attr->info.info_len;
3760 	int err;
3761 
3762 	err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3763 	if (err)
3764 		return err;
3765 	info_len = min_t(u32, sizeof(info), info_len);
3766 
3767 	memset(&info, 0, sizeof(info));
3768 	if (copy_from_user(&info, uinfo, info_len))
3769 		return -EFAULT;
3770 
3771 	info.type = link->type;
3772 	info.id = link->id;
3773 	info.prog_id = link->prog->aux->id;
3774 
3775 	if (link->ops->fill_link_info) {
3776 		err = link->ops->fill_link_info(link, &info);
3777 		if (err)
3778 			return err;
3779 	}
3780 
3781 	if (copy_to_user(uinfo, &info, info_len) ||
3782 	    put_user(info_len, &uattr->info.info_len))
3783 		return -EFAULT;
3784 
3785 	return 0;
3786 }
3787 
3788 
3789 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3790 
bpf_obj_get_info_by_fd(const union bpf_attr * attr,union bpf_attr __user * uattr)3791 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3792 				  union bpf_attr __user *uattr)
3793 {
3794 	int ufd = attr->info.bpf_fd;
3795 	struct fd f;
3796 	int err;
3797 
3798 	if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3799 		return -EINVAL;
3800 
3801 	f = fdget(ufd);
3802 	if (!f.file)
3803 		return -EBADFD;
3804 
3805 	if (f.file->f_op == &bpf_prog_fops)
3806 		err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3807 					      uattr);
3808 	else if (f.file->f_op == &bpf_map_fops)
3809 		err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3810 					     uattr);
3811 	else if (f.file->f_op == &btf_fops)
3812 		err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3813 	else if (f.file->f_op == &bpf_link_fops)
3814 		err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3815 					      attr, uattr);
3816 	else
3817 		err = -EINVAL;
3818 
3819 	fdput(f);
3820 	return err;
3821 }
3822 
3823 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3824 
bpf_btf_load(const union bpf_attr * attr)3825 static int bpf_btf_load(const union bpf_attr *attr)
3826 {
3827 	if (CHECK_ATTR(BPF_BTF_LOAD))
3828 		return -EINVAL;
3829 
3830 	if (!bpf_capable())
3831 		return -EPERM;
3832 
3833 	return btf_new_fd(attr);
3834 }
3835 
3836 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3837 
bpf_btf_get_fd_by_id(const union bpf_attr * attr)3838 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3839 {
3840 	if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3841 		return -EINVAL;
3842 
3843 	if (!capable(CAP_SYS_ADMIN))
3844 		return -EPERM;
3845 
3846 	return btf_get_fd_by_id(attr->btf_id);
3847 }
3848 
bpf_task_fd_query_copy(const union bpf_attr * attr,union bpf_attr __user * uattr,u32 prog_id,u32 fd_type,const char * buf,u64 probe_offset,u64 probe_addr)3849 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3850 				    union bpf_attr __user *uattr,
3851 				    u32 prog_id, u32 fd_type,
3852 				    const char *buf, u64 probe_offset,
3853 				    u64 probe_addr)
3854 {
3855 	char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3856 	u32 len = buf ? strlen(buf) : 0, input_len;
3857 	int err = 0;
3858 
3859 	if (put_user(len, &uattr->task_fd_query.buf_len))
3860 		return -EFAULT;
3861 	input_len = attr->task_fd_query.buf_len;
3862 	if (input_len && ubuf) {
3863 		if (!len) {
3864 			/* nothing to copy, just make ubuf NULL terminated */
3865 			char zero = '\0';
3866 
3867 			if (put_user(zero, ubuf))
3868 				return -EFAULT;
3869 		} else if (input_len >= len + 1) {
3870 			/* ubuf can hold the string with NULL terminator */
3871 			if (copy_to_user(ubuf, buf, len + 1))
3872 				return -EFAULT;
3873 		} else {
3874 			/* ubuf cannot hold the string with NULL terminator,
3875 			 * do a partial copy with NULL terminator.
3876 			 */
3877 			char zero = '\0';
3878 
3879 			err = -ENOSPC;
3880 			if (copy_to_user(ubuf, buf, input_len - 1))
3881 				return -EFAULT;
3882 			if (put_user(zero, ubuf + input_len - 1))
3883 				return -EFAULT;
3884 		}
3885 	}
3886 
3887 	if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3888 	    put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3889 	    put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3890 	    put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3891 		return -EFAULT;
3892 
3893 	return err;
3894 }
3895 
3896 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3897 
bpf_task_fd_query(const union bpf_attr * attr,union bpf_attr __user * uattr)3898 static int bpf_task_fd_query(const union bpf_attr *attr,
3899 			     union bpf_attr __user *uattr)
3900 {
3901 	pid_t pid = attr->task_fd_query.pid;
3902 	u32 fd = attr->task_fd_query.fd;
3903 	const struct perf_event *event;
3904 	struct files_struct *files;
3905 	struct task_struct *task;
3906 	struct file *file;
3907 	int err;
3908 
3909 	if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3910 		return -EINVAL;
3911 
3912 	if (!capable(CAP_SYS_ADMIN))
3913 		return -EPERM;
3914 
3915 	if (attr->task_fd_query.flags != 0)
3916 		return -EINVAL;
3917 
3918 	rcu_read_lock();
3919 	task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3920 	rcu_read_unlock();
3921 	if (!task)
3922 		return -ENOENT;
3923 
3924 	files = get_files_struct(task);
3925 	put_task_struct(task);
3926 	if (!files)
3927 		return -ENOENT;
3928 
3929 	err = 0;
3930 	spin_lock(&files->file_lock);
3931 	file = fcheck_files(files, fd);
3932 	if (!file)
3933 		err = -EBADF;
3934 	else
3935 		get_file(file);
3936 	spin_unlock(&files->file_lock);
3937 	put_files_struct(files);
3938 
3939 	if (err)
3940 		goto out;
3941 
3942 	if (file->f_op == &bpf_link_fops) {
3943 		struct bpf_link *link = file->private_data;
3944 
3945 		if (link->ops == &bpf_raw_tp_link_lops) {
3946 			struct bpf_raw_tp_link *raw_tp =
3947 				container_of(link, struct bpf_raw_tp_link, link);
3948 			struct bpf_raw_event_map *btp = raw_tp->btp;
3949 
3950 			err = bpf_task_fd_query_copy(attr, uattr,
3951 						     raw_tp->link.prog->aux->id,
3952 						     BPF_FD_TYPE_RAW_TRACEPOINT,
3953 						     btp->tp->name, 0, 0);
3954 			goto put_file;
3955 		}
3956 		goto out_not_supp;
3957 	}
3958 
3959 	event = perf_get_event(file);
3960 	if (!IS_ERR(event)) {
3961 		u64 probe_offset, probe_addr;
3962 		u32 prog_id, fd_type;
3963 		const char *buf;
3964 
3965 		err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3966 					      &buf, &probe_offset,
3967 					      &probe_addr);
3968 		if (!err)
3969 			err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3970 						     fd_type, buf,
3971 						     probe_offset,
3972 						     probe_addr);
3973 		goto put_file;
3974 	}
3975 
3976 out_not_supp:
3977 	err = -ENOTSUPP;
3978 put_file:
3979 	fput(file);
3980 out:
3981 	return err;
3982 }
3983 
3984 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3985 
3986 #define BPF_DO_BATCH(fn)			\
3987 	do {					\
3988 		if (!fn) {			\
3989 			err = -ENOTSUPP;	\
3990 			goto err_put;		\
3991 		}				\
3992 		err = fn(map, attr, uattr);	\
3993 	} while (0)
3994 
bpf_map_do_batch(const union bpf_attr * attr,union bpf_attr __user * uattr,int cmd)3995 static int bpf_map_do_batch(const union bpf_attr *attr,
3996 			    union bpf_attr __user *uattr,
3997 			    int cmd)
3998 {
3999 	bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4000 			 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4001 	bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4002 	struct bpf_map *map;
4003 	int err, ufd;
4004 	struct fd f;
4005 
4006 	if (CHECK_ATTR(BPF_MAP_BATCH))
4007 		return -EINVAL;
4008 
4009 	ufd = attr->batch.map_fd;
4010 	f = fdget(ufd);
4011 	map = __bpf_map_get(f);
4012 	if (IS_ERR(map))
4013 		return PTR_ERR(map);
4014 	if (has_write)
4015 		bpf_map_write_active_inc(map);
4016 	if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4017 		err = -EPERM;
4018 		goto err_put;
4019 	}
4020 	if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4021 		err = -EPERM;
4022 		goto err_put;
4023 	}
4024 
4025 	if (cmd == BPF_MAP_LOOKUP_BATCH)
4026 		BPF_DO_BATCH(map->ops->map_lookup_batch);
4027 	else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4028 		BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4029 	else if (cmd == BPF_MAP_UPDATE_BATCH)
4030 		BPF_DO_BATCH(map->ops->map_update_batch);
4031 	else
4032 		BPF_DO_BATCH(map->ops->map_delete_batch);
4033 err_put:
4034 	if (has_write)
4035 		bpf_map_write_active_dec(map);
4036 	fdput(f);
4037 	return err;
4038 }
4039 
tracing_bpf_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)4040 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4041 {
4042 	if (attr->link_create.attach_type != prog->expected_attach_type)
4043 		return -EINVAL;
4044 
4045 	if (prog->expected_attach_type == BPF_TRACE_ITER)
4046 		return bpf_iter_link_attach(attr, prog);
4047 	else if (prog->type == BPF_PROG_TYPE_EXT)
4048 		return bpf_tracing_prog_attach(prog,
4049 					       attr->link_create.target_fd,
4050 					       attr->link_create.target_btf_id);
4051 	return -EINVAL;
4052 }
4053 
4054 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
link_create(union bpf_attr * attr)4055 static int link_create(union bpf_attr *attr)
4056 {
4057 	enum bpf_prog_type ptype;
4058 	struct bpf_prog *prog;
4059 	int ret;
4060 
4061 	if (CHECK_ATTR(BPF_LINK_CREATE))
4062 		return -EINVAL;
4063 
4064 	prog = bpf_prog_get(attr->link_create.prog_fd);
4065 	if (IS_ERR(prog))
4066 		return PTR_ERR(prog);
4067 
4068 	ret = bpf_prog_attach_check_attach_type(prog,
4069 						attr->link_create.attach_type);
4070 	if (ret)
4071 		goto out;
4072 
4073 	if (prog->type == BPF_PROG_TYPE_EXT) {
4074 		ret = tracing_bpf_link_attach(attr, prog);
4075 		goto out;
4076 	}
4077 
4078 	ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4079 	if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4080 		ret = -EINVAL;
4081 		goto out;
4082 	}
4083 
4084 	switch (ptype) {
4085 	case BPF_PROG_TYPE_CGROUP_SKB:
4086 	case BPF_PROG_TYPE_CGROUP_SOCK:
4087 	case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4088 	case BPF_PROG_TYPE_SOCK_OPS:
4089 	case BPF_PROG_TYPE_CGROUP_DEVICE:
4090 	case BPF_PROG_TYPE_CGROUP_SYSCTL:
4091 	case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4092 		ret = cgroup_bpf_link_attach(attr, prog);
4093 		break;
4094 	case BPF_PROG_TYPE_TRACING:
4095 		ret = tracing_bpf_link_attach(attr, prog);
4096 		break;
4097 	case BPF_PROG_TYPE_FLOW_DISSECTOR:
4098 	case BPF_PROG_TYPE_SK_LOOKUP:
4099 		ret = netns_bpf_link_create(attr, prog);
4100 		break;
4101 #ifdef CONFIG_NET
4102 	case BPF_PROG_TYPE_XDP:
4103 		ret = bpf_xdp_link_attach(attr, prog);
4104 		break;
4105 #endif
4106 	default:
4107 		ret = -EINVAL;
4108 	}
4109 
4110 out:
4111 	if (ret < 0)
4112 		bpf_prog_put(prog);
4113 	return ret;
4114 }
4115 
4116 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4117 
link_update(union bpf_attr * attr)4118 static int link_update(union bpf_attr *attr)
4119 {
4120 	struct bpf_prog *old_prog = NULL, *new_prog;
4121 	struct bpf_link *link;
4122 	u32 flags;
4123 	int ret;
4124 
4125 	if (CHECK_ATTR(BPF_LINK_UPDATE))
4126 		return -EINVAL;
4127 
4128 	flags = attr->link_update.flags;
4129 	if (flags & ~BPF_F_REPLACE)
4130 		return -EINVAL;
4131 
4132 	link = bpf_link_get_from_fd(attr->link_update.link_fd);
4133 	if (IS_ERR(link))
4134 		return PTR_ERR(link);
4135 
4136 	new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4137 	if (IS_ERR(new_prog)) {
4138 		ret = PTR_ERR(new_prog);
4139 		goto out_put_link;
4140 	}
4141 
4142 	if (flags & BPF_F_REPLACE) {
4143 		old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4144 		if (IS_ERR(old_prog)) {
4145 			ret = PTR_ERR(old_prog);
4146 			old_prog = NULL;
4147 			goto out_put_progs;
4148 		}
4149 	} else if (attr->link_update.old_prog_fd) {
4150 		ret = -EINVAL;
4151 		goto out_put_progs;
4152 	}
4153 
4154 	if (link->ops->update_prog)
4155 		ret = link->ops->update_prog(link, new_prog, old_prog);
4156 	else
4157 		ret = -EINVAL;
4158 
4159 out_put_progs:
4160 	if (old_prog)
4161 		bpf_prog_put(old_prog);
4162 	if (ret)
4163 		bpf_prog_put(new_prog);
4164 out_put_link:
4165 	bpf_link_put(link);
4166 	return ret;
4167 }
4168 
4169 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4170 
link_detach(union bpf_attr * attr)4171 static int link_detach(union bpf_attr *attr)
4172 {
4173 	struct bpf_link *link;
4174 	int ret;
4175 
4176 	if (CHECK_ATTR(BPF_LINK_DETACH))
4177 		return -EINVAL;
4178 
4179 	link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4180 	if (IS_ERR(link))
4181 		return PTR_ERR(link);
4182 
4183 	if (link->ops->detach)
4184 		ret = link->ops->detach(link);
4185 	else
4186 		ret = -EOPNOTSUPP;
4187 
4188 	bpf_link_put(link);
4189 	return ret;
4190 }
4191 
bpf_link_inc_not_zero(struct bpf_link * link)4192 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4193 {
4194 	return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4195 }
4196 
bpf_link_by_id(u32 id)4197 struct bpf_link *bpf_link_by_id(u32 id)
4198 {
4199 	struct bpf_link *link;
4200 
4201 	if (!id)
4202 		return ERR_PTR(-ENOENT);
4203 
4204 	spin_lock_bh(&link_idr_lock);
4205 	/* before link is "settled", ID is 0, pretend it doesn't exist yet */
4206 	link = idr_find(&link_idr, id);
4207 	if (link) {
4208 		if (link->id)
4209 			link = bpf_link_inc_not_zero(link);
4210 		else
4211 			link = ERR_PTR(-EAGAIN);
4212 	} else {
4213 		link = ERR_PTR(-ENOENT);
4214 	}
4215 	spin_unlock_bh(&link_idr_lock);
4216 	return link;
4217 }
4218 
4219 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4220 
bpf_link_get_fd_by_id(const union bpf_attr * attr)4221 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4222 {
4223 	struct bpf_link *link;
4224 	u32 id = attr->link_id;
4225 	int fd;
4226 
4227 	if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4228 		return -EINVAL;
4229 
4230 	if (!capable(CAP_SYS_ADMIN))
4231 		return -EPERM;
4232 
4233 	link = bpf_link_by_id(id);
4234 	if (IS_ERR(link))
4235 		return PTR_ERR(link);
4236 
4237 	fd = bpf_link_new_fd(link);
4238 	if (fd < 0)
4239 		bpf_link_put(link);
4240 
4241 	return fd;
4242 }
4243 
4244 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4245 
bpf_stats_release(struct inode * inode,struct file * file)4246 static int bpf_stats_release(struct inode *inode, struct file *file)
4247 {
4248 	mutex_lock(&bpf_stats_enabled_mutex);
4249 	static_key_slow_dec(&bpf_stats_enabled_key.key);
4250 	mutex_unlock(&bpf_stats_enabled_mutex);
4251 	return 0;
4252 }
4253 
4254 static const struct file_operations bpf_stats_fops = {
4255 	.release = bpf_stats_release,
4256 };
4257 
bpf_enable_runtime_stats(void)4258 static int bpf_enable_runtime_stats(void)
4259 {
4260 	int fd;
4261 
4262 	mutex_lock(&bpf_stats_enabled_mutex);
4263 
4264 	/* Set a very high limit to avoid overflow */
4265 	if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4266 		mutex_unlock(&bpf_stats_enabled_mutex);
4267 		return -EBUSY;
4268 	}
4269 
4270 	fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4271 	if (fd >= 0)
4272 		static_key_slow_inc(&bpf_stats_enabled_key.key);
4273 
4274 	mutex_unlock(&bpf_stats_enabled_mutex);
4275 	return fd;
4276 }
4277 
4278 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4279 
bpf_enable_stats(union bpf_attr * attr)4280 static int bpf_enable_stats(union bpf_attr *attr)
4281 {
4282 
4283 	if (CHECK_ATTR(BPF_ENABLE_STATS))
4284 		return -EINVAL;
4285 
4286 	if (!capable(CAP_SYS_ADMIN))
4287 		return -EPERM;
4288 
4289 	switch (attr->enable_stats.type) {
4290 	case BPF_STATS_RUN_TIME:
4291 		return bpf_enable_runtime_stats();
4292 	default:
4293 		break;
4294 	}
4295 	return -EINVAL;
4296 }
4297 
4298 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4299 
bpf_iter_create(union bpf_attr * attr)4300 static int bpf_iter_create(union bpf_attr *attr)
4301 {
4302 	struct bpf_link *link;
4303 	int err;
4304 
4305 	if (CHECK_ATTR(BPF_ITER_CREATE))
4306 		return -EINVAL;
4307 
4308 	if (attr->iter_create.flags)
4309 		return -EINVAL;
4310 
4311 	link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4312 	if (IS_ERR(link))
4313 		return PTR_ERR(link);
4314 
4315 	err = bpf_iter_new_fd(link);
4316 	bpf_link_put(link);
4317 
4318 	return err;
4319 }
4320 
4321 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4322 
bpf_prog_bind_map(union bpf_attr * attr)4323 static int bpf_prog_bind_map(union bpf_attr *attr)
4324 {
4325 	struct bpf_prog *prog;
4326 	struct bpf_map *map;
4327 	struct bpf_map **used_maps_old, **used_maps_new;
4328 	int i, ret = 0;
4329 
4330 	if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4331 		return -EINVAL;
4332 
4333 	if (attr->prog_bind_map.flags)
4334 		return -EINVAL;
4335 
4336 	prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4337 	if (IS_ERR(prog))
4338 		return PTR_ERR(prog);
4339 
4340 	map = bpf_map_get(attr->prog_bind_map.map_fd);
4341 	if (IS_ERR(map)) {
4342 		ret = PTR_ERR(map);
4343 		goto out_prog_put;
4344 	}
4345 
4346 	mutex_lock(&prog->aux->used_maps_mutex);
4347 
4348 	used_maps_old = prog->aux->used_maps;
4349 
4350 	for (i = 0; i < prog->aux->used_map_cnt; i++)
4351 		if (used_maps_old[i] == map) {
4352 			bpf_map_put(map);
4353 			goto out_unlock;
4354 		}
4355 
4356 	used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4357 				      sizeof(used_maps_new[0]),
4358 				      GFP_KERNEL);
4359 	if (!used_maps_new) {
4360 		ret = -ENOMEM;
4361 		goto out_unlock;
4362 	}
4363 
4364 	memcpy(used_maps_new, used_maps_old,
4365 	       sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4366 	used_maps_new[prog->aux->used_map_cnt] = map;
4367 
4368 	prog->aux->used_map_cnt++;
4369 	prog->aux->used_maps = used_maps_new;
4370 
4371 	kfree(used_maps_old);
4372 
4373 out_unlock:
4374 	mutex_unlock(&prog->aux->used_maps_mutex);
4375 
4376 	if (ret)
4377 		bpf_map_put(map);
4378 out_prog_put:
4379 	bpf_prog_put(prog);
4380 	return ret;
4381 }
4382 
SYSCALL_DEFINE3(bpf,int,cmd,union bpf_attr __user *,uattr,unsigned int,size)4383 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4384 {
4385 	union bpf_attr attr;
4386 	int err;
4387 
4388 	if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4389 		return -EPERM;
4390 
4391 	err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4392 	if (err)
4393 		return err;
4394 	size = min_t(u32, size, sizeof(attr));
4395 
4396 	/* copy attributes from user space, may be less than sizeof(bpf_attr) */
4397 	memset(&attr, 0, sizeof(attr));
4398 	if (copy_from_user(&attr, uattr, size) != 0)
4399 		return -EFAULT;
4400 
4401 	err = security_bpf(cmd, &attr, size);
4402 	if (err < 0)
4403 		return err;
4404 
4405 	switch (cmd) {
4406 	case BPF_MAP_CREATE:
4407 		err = map_create(&attr);
4408 		break;
4409 	case BPF_MAP_LOOKUP_ELEM:
4410 		err = map_lookup_elem(&attr);
4411 		break;
4412 	case BPF_MAP_UPDATE_ELEM:
4413 		err = map_update_elem(&attr);
4414 		break;
4415 	case BPF_MAP_DELETE_ELEM:
4416 		err = map_delete_elem(&attr);
4417 		break;
4418 	case BPF_MAP_GET_NEXT_KEY:
4419 		err = map_get_next_key(&attr);
4420 		break;
4421 	case BPF_MAP_FREEZE:
4422 		err = map_freeze(&attr);
4423 		break;
4424 	case BPF_PROG_LOAD:
4425 		err = bpf_prog_load(&attr, uattr);
4426 		break;
4427 	case BPF_OBJ_PIN:
4428 		err = bpf_obj_pin(&attr);
4429 		break;
4430 	case BPF_OBJ_GET:
4431 		err = bpf_obj_get(&attr);
4432 		break;
4433 	case BPF_PROG_ATTACH:
4434 		err = bpf_prog_attach(&attr);
4435 		break;
4436 	case BPF_PROG_DETACH:
4437 		err = bpf_prog_detach(&attr);
4438 		break;
4439 	case BPF_PROG_QUERY:
4440 		err = bpf_prog_query(&attr, uattr);
4441 		break;
4442 	case BPF_PROG_TEST_RUN:
4443 		err = bpf_prog_test_run(&attr, uattr);
4444 		break;
4445 	case BPF_PROG_GET_NEXT_ID:
4446 		err = bpf_obj_get_next_id(&attr, uattr,
4447 					  &prog_idr, &prog_idr_lock);
4448 		break;
4449 	case BPF_MAP_GET_NEXT_ID:
4450 		err = bpf_obj_get_next_id(&attr, uattr,
4451 					  &map_idr, &map_idr_lock);
4452 		break;
4453 	case BPF_BTF_GET_NEXT_ID:
4454 		err = bpf_obj_get_next_id(&attr, uattr,
4455 					  &btf_idr, &btf_idr_lock);
4456 		break;
4457 	case BPF_PROG_GET_FD_BY_ID:
4458 		err = bpf_prog_get_fd_by_id(&attr);
4459 		break;
4460 	case BPF_MAP_GET_FD_BY_ID:
4461 		err = bpf_map_get_fd_by_id(&attr);
4462 		break;
4463 	case BPF_OBJ_GET_INFO_BY_FD:
4464 		err = bpf_obj_get_info_by_fd(&attr, uattr);
4465 		break;
4466 	case BPF_RAW_TRACEPOINT_OPEN:
4467 		err = bpf_raw_tracepoint_open(&attr);
4468 		break;
4469 	case BPF_BTF_LOAD:
4470 		err = bpf_btf_load(&attr);
4471 		break;
4472 	case BPF_BTF_GET_FD_BY_ID:
4473 		err = bpf_btf_get_fd_by_id(&attr);
4474 		break;
4475 	case BPF_TASK_FD_QUERY:
4476 		err = bpf_task_fd_query(&attr, uattr);
4477 		break;
4478 	case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4479 		err = map_lookup_and_delete_elem(&attr);
4480 		break;
4481 	case BPF_MAP_LOOKUP_BATCH:
4482 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4483 		break;
4484 	case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4485 		err = bpf_map_do_batch(&attr, uattr,
4486 				       BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4487 		break;
4488 	case BPF_MAP_UPDATE_BATCH:
4489 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4490 		break;
4491 	case BPF_MAP_DELETE_BATCH:
4492 		err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4493 		break;
4494 	case BPF_LINK_CREATE:
4495 		err = link_create(&attr);
4496 		break;
4497 	case BPF_LINK_UPDATE:
4498 		err = link_update(&attr);
4499 		break;
4500 	case BPF_LINK_GET_FD_BY_ID:
4501 		err = bpf_link_get_fd_by_id(&attr);
4502 		break;
4503 	case BPF_LINK_GET_NEXT_ID:
4504 		err = bpf_obj_get_next_id(&attr, uattr,
4505 					  &link_idr, &link_idr_lock);
4506 		break;
4507 	case BPF_ENABLE_STATS:
4508 		err = bpf_enable_stats(&attr);
4509 		break;
4510 	case BPF_ITER_CREATE:
4511 		err = bpf_iter_create(&attr);
4512 		break;
4513 	case BPF_LINK_DETACH:
4514 		err = link_detach(&attr);
4515 		break;
4516 	case BPF_PROG_BIND_MAP:
4517 		err = bpf_prog_bind_map(&attr);
4518 		break;
4519 	default:
4520 		err = -EINVAL;
4521 		break;
4522 	}
4523 
4524 	return err;
4525 }
4526