• 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/rcupdate.h>
6 #include <linux/random.h>
7 #include <linux/smp.h>
8 #include <linux/topology.h>
9 #include <linux/ktime.h>
10 #include <linux/sched.h>
11 #include <linux/uidgid.h>
12 #include <linux/filter.h>
13 #include <linux/ctype.h>
14 #include <linux/jiffies.h>
15 #include <linux/pid_namespace.h>
16 #include <linux/proc_ns.h>
17 #include <linux/security.h>
18 
19 #include "../../lib/kstrtox.h"
20 
21 /* If kernel subsystem is allowing eBPF programs to call this function,
22  * inside its own verifier_ops->get_func_proto() callback it should return
23  * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
24  *
25  * Different map implementations will rely on rcu in map methods
26  * lookup/update/delete, therefore eBPF programs must run under rcu lock
27  * if program is allowed to access maps, so check rcu_read_lock_held in
28  * all three functions.
29  */
BPF_CALL_2(bpf_map_lookup_elem,struct bpf_map *,map,void *,key)30 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
31 {
32 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
33 	return (unsigned long) map->ops->map_lookup_elem(map, key);
34 }
35 
36 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
37 	.func		= bpf_map_lookup_elem,
38 	.gpl_only	= false,
39 	.pkt_access	= true,
40 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
41 	.arg1_type	= ARG_CONST_MAP_PTR,
42 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
43 };
44 
BPF_CALL_4(bpf_map_update_elem,struct bpf_map *,map,void *,key,void *,value,u64,flags)45 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
46 	   void *, value, u64, flags)
47 {
48 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
49 	return map->ops->map_update_elem(map, key, value, flags);
50 }
51 
52 const struct bpf_func_proto bpf_map_update_elem_proto = {
53 	.func		= bpf_map_update_elem,
54 	.gpl_only	= false,
55 	.pkt_access	= true,
56 	.ret_type	= RET_INTEGER,
57 	.arg1_type	= ARG_CONST_MAP_PTR,
58 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
59 	.arg3_type	= ARG_PTR_TO_MAP_VALUE,
60 	.arg4_type	= ARG_ANYTHING,
61 };
62 
BPF_CALL_2(bpf_map_delete_elem,struct bpf_map *,map,void *,key)63 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
64 {
65 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
66 	return map->ops->map_delete_elem(map, key);
67 }
68 
69 const struct bpf_func_proto bpf_map_delete_elem_proto = {
70 	.func		= bpf_map_delete_elem,
71 	.gpl_only	= false,
72 	.pkt_access	= true,
73 	.ret_type	= RET_INTEGER,
74 	.arg1_type	= ARG_CONST_MAP_PTR,
75 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
76 };
77 
BPF_CALL_3(bpf_map_push_elem,struct bpf_map *,map,void *,value,u64,flags)78 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
79 {
80 	return map->ops->map_push_elem(map, value, flags);
81 }
82 
83 const struct bpf_func_proto bpf_map_push_elem_proto = {
84 	.func		= bpf_map_push_elem,
85 	.gpl_only	= false,
86 	.pkt_access	= true,
87 	.ret_type	= RET_INTEGER,
88 	.arg1_type	= ARG_CONST_MAP_PTR,
89 	.arg2_type	= ARG_PTR_TO_MAP_VALUE,
90 	.arg3_type	= ARG_ANYTHING,
91 };
92 
BPF_CALL_2(bpf_map_pop_elem,struct bpf_map *,map,void *,value)93 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
94 {
95 	return map->ops->map_pop_elem(map, value);
96 }
97 
98 const struct bpf_func_proto bpf_map_pop_elem_proto = {
99 	.func		= bpf_map_pop_elem,
100 	.gpl_only	= false,
101 	.ret_type	= RET_INTEGER,
102 	.arg1_type	= ARG_CONST_MAP_PTR,
103 	.arg2_type	= ARG_PTR_TO_UNINIT_MAP_VALUE,
104 };
105 
BPF_CALL_2(bpf_map_peek_elem,struct bpf_map *,map,void *,value)106 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
107 {
108 	return map->ops->map_peek_elem(map, value);
109 }
110 
111 const struct bpf_func_proto bpf_map_peek_elem_proto = {
112 	.func		= bpf_map_peek_elem,
113 	.gpl_only	= false,
114 	.ret_type	= RET_INTEGER,
115 	.arg1_type	= ARG_CONST_MAP_PTR,
116 	.arg2_type	= ARG_PTR_TO_UNINIT_MAP_VALUE,
117 };
118 
119 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
120 	.func		= bpf_user_rnd_u32,
121 	.gpl_only	= false,
122 	.ret_type	= RET_INTEGER,
123 };
124 
BPF_CALL_0(bpf_get_smp_processor_id)125 BPF_CALL_0(bpf_get_smp_processor_id)
126 {
127 	return smp_processor_id();
128 }
129 
130 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
131 	.func		= bpf_get_smp_processor_id,
132 	.gpl_only	= false,
133 	.ret_type	= RET_INTEGER,
134 };
135 
BPF_CALL_0(bpf_get_numa_node_id)136 BPF_CALL_0(bpf_get_numa_node_id)
137 {
138 	return numa_node_id();
139 }
140 
141 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
142 	.func		= bpf_get_numa_node_id,
143 	.gpl_only	= false,
144 	.ret_type	= RET_INTEGER,
145 };
146 
BPF_CALL_0(bpf_ktime_get_ns)147 BPF_CALL_0(bpf_ktime_get_ns)
148 {
149 	/* NMI safe access to clock monotonic */
150 	return ktime_get_mono_fast_ns();
151 }
152 
153 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
154 	.func		= bpf_ktime_get_ns,
155 	.gpl_only	= false,
156 	.ret_type	= RET_INTEGER,
157 };
158 
BPF_CALL_0(bpf_ktime_get_boot_ns)159 BPF_CALL_0(bpf_ktime_get_boot_ns)
160 {
161 	/* NMI safe access to clock boottime */
162 	return ktime_get_boot_fast_ns();
163 }
164 
165 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
166 	.func		= bpf_ktime_get_boot_ns,
167 	.gpl_only	= false,
168 	.ret_type	= RET_INTEGER,
169 };
170 
BPF_CALL_0(bpf_ktime_get_coarse_ns)171 BPF_CALL_0(bpf_ktime_get_coarse_ns)
172 {
173 	return ktime_get_coarse_ns();
174 }
175 
176 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
177 	.func		= bpf_ktime_get_coarse_ns,
178 	.gpl_only	= false,
179 	.ret_type	= RET_INTEGER,
180 };
181 
BPF_CALL_0(bpf_get_current_pid_tgid)182 BPF_CALL_0(bpf_get_current_pid_tgid)
183 {
184 	struct task_struct *task = current;
185 
186 	if (unlikely(!task))
187 		return -EINVAL;
188 
189 	return (u64) task->tgid << 32 | task->pid;
190 }
191 
192 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
193 	.func		= bpf_get_current_pid_tgid,
194 	.gpl_only	= false,
195 	.ret_type	= RET_INTEGER,
196 };
197 
BPF_CALL_0(bpf_get_current_uid_gid)198 BPF_CALL_0(bpf_get_current_uid_gid)
199 {
200 	struct task_struct *task = current;
201 	kuid_t uid;
202 	kgid_t gid;
203 
204 	if (unlikely(!task))
205 		return -EINVAL;
206 
207 	current_uid_gid(&uid, &gid);
208 	return (u64) from_kgid(&init_user_ns, gid) << 32 |
209 		     from_kuid(&init_user_ns, uid);
210 }
211 
212 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
213 	.func		= bpf_get_current_uid_gid,
214 	.gpl_only	= false,
215 	.ret_type	= RET_INTEGER,
216 };
217 
BPF_CALL_2(bpf_get_current_comm,char *,buf,u32,size)218 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
219 {
220 	struct task_struct *task = current;
221 
222 	if (unlikely(!task))
223 		goto err_clear;
224 
225 	strncpy(buf, task->comm, size);
226 
227 	/* Verifier guarantees that size > 0. For task->comm exceeding
228 	 * size, guarantee that buf is %NUL-terminated. Unconditionally
229 	 * done here to save the size test.
230 	 */
231 	buf[size - 1] = 0;
232 	return 0;
233 err_clear:
234 	memset(buf, 0, size);
235 	return -EINVAL;
236 }
237 
238 const struct bpf_func_proto bpf_get_current_comm_proto = {
239 	.func		= bpf_get_current_comm,
240 	.gpl_only	= false,
241 	.ret_type	= RET_INTEGER,
242 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
243 	.arg2_type	= ARG_CONST_SIZE,
244 };
245 
246 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
247 
__bpf_spin_lock(struct bpf_spin_lock * lock)248 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
249 {
250 	arch_spinlock_t *l = (void *)lock;
251 	union {
252 		__u32 val;
253 		arch_spinlock_t lock;
254 	} u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
255 
256 	compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
257 	BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
258 	BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
259 	arch_spin_lock(l);
260 }
261 
__bpf_spin_unlock(struct bpf_spin_lock * lock)262 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
263 {
264 	arch_spinlock_t *l = (void *)lock;
265 
266 	arch_spin_unlock(l);
267 }
268 
269 #else
270 
__bpf_spin_lock(struct bpf_spin_lock * lock)271 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
272 {
273 	atomic_t *l = (void *)lock;
274 
275 	BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
276 	do {
277 		atomic_cond_read_relaxed(l, !VAL);
278 	} while (atomic_xchg(l, 1));
279 }
280 
__bpf_spin_unlock(struct bpf_spin_lock * lock)281 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
282 {
283 	atomic_t *l = (void *)lock;
284 
285 	atomic_set_release(l, 0);
286 }
287 
288 #endif
289 
290 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
291 
__bpf_spin_lock_irqsave(struct bpf_spin_lock * lock)292 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
293 {
294 	unsigned long flags;
295 
296 	local_irq_save(flags);
297 	__bpf_spin_lock(lock);
298 	__this_cpu_write(irqsave_flags, flags);
299 }
300 
BPF_CALL_1(bpf_spin_lock,struct bpf_spin_lock *,lock)301 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
302 {
303 	__bpf_spin_lock_irqsave(lock);
304 	return 0;
305 }
306 
307 const struct bpf_func_proto bpf_spin_lock_proto = {
308 	.func		= bpf_spin_lock,
309 	.gpl_only	= false,
310 	.ret_type	= RET_VOID,
311 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
312 };
313 
__bpf_spin_unlock_irqrestore(struct bpf_spin_lock * lock)314 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
315 {
316 	unsigned long flags;
317 
318 	flags = __this_cpu_read(irqsave_flags);
319 	__bpf_spin_unlock(lock);
320 	local_irq_restore(flags);
321 }
322 
BPF_CALL_1(bpf_spin_unlock,struct bpf_spin_lock *,lock)323 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
324 {
325 	__bpf_spin_unlock_irqrestore(lock);
326 	return 0;
327 }
328 
329 const struct bpf_func_proto bpf_spin_unlock_proto = {
330 	.func		= bpf_spin_unlock,
331 	.gpl_only	= false,
332 	.ret_type	= RET_VOID,
333 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
334 };
335 
copy_map_value_locked(struct bpf_map * map,void * dst,void * src,bool lock_src)336 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
337 			   bool lock_src)
338 {
339 	struct bpf_spin_lock *lock;
340 
341 	if (lock_src)
342 		lock = src + map->spin_lock_off;
343 	else
344 		lock = dst + map->spin_lock_off;
345 	preempt_disable();
346 	__bpf_spin_lock_irqsave(lock);
347 	copy_map_value(map, dst, src);
348 	__bpf_spin_unlock_irqrestore(lock);
349 	preempt_enable();
350 }
351 
BPF_CALL_0(bpf_jiffies64)352 BPF_CALL_0(bpf_jiffies64)
353 {
354 	return get_jiffies_64();
355 }
356 
357 const struct bpf_func_proto bpf_jiffies64_proto = {
358 	.func		= bpf_jiffies64,
359 	.gpl_only	= false,
360 	.ret_type	= RET_INTEGER,
361 };
362 
363 #ifdef CONFIG_CGROUPS
BPF_CALL_0(bpf_get_current_cgroup_id)364 BPF_CALL_0(bpf_get_current_cgroup_id)
365 {
366 	struct cgroup *cgrp;
367 	u64 cgrp_id;
368 
369 	rcu_read_lock();
370 	cgrp = task_dfl_cgroup(current);
371 	cgrp_id = cgroup_id(cgrp);
372 	rcu_read_unlock();
373 
374 	return cgrp_id;
375 }
376 
377 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
378 	.func		= bpf_get_current_cgroup_id,
379 	.gpl_only	= false,
380 	.ret_type	= RET_INTEGER,
381 };
382 
BPF_CALL_1(bpf_get_current_ancestor_cgroup_id,int,ancestor_level)383 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
384 {
385 	struct cgroup *cgrp;
386 	struct cgroup *ancestor;
387 	u64 cgrp_id;
388 
389 	rcu_read_lock();
390 	cgrp = task_dfl_cgroup(current);
391 	ancestor = cgroup_ancestor(cgrp, ancestor_level);
392 	cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
393 	rcu_read_unlock();
394 
395 	return cgrp_id;
396 }
397 
398 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
399 	.func		= bpf_get_current_ancestor_cgroup_id,
400 	.gpl_only	= false,
401 	.ret_type	= RET_INTEGER,
402 	.arg1_type	= ARG_ANYTHING,
403 };
404 
405 #ifdef CONFIG_CGROUP_BPF
406 
BPF_CALL_2(bpf_get_local_storage,struct bpf_map *,map,u64,flags)407 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
408 {
409 	/* flags argument is not used now,
410 	 * but provides an ability to extend the API.
411 	 * verifier checks that its value is correct.
412 	 */
413 	enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
414 	struct bpf_cgroup_storage *storage;
415 	struct bpf_cg_run_ctx *ctx;
416 	void *ptr;
417 
418 	/* get current cgroup storage from BPF run context */
419 	ctx = container_of(current->bpf_ctx, struct bpf_cg_run_ctx, run_ctx);
420 	storage = ctx->prog_item->cgroup_storage[stype];
421 
422 	if (stype == BPF_CGROUP_STORAGE_SHARED)
423 		ptr = &READ_ONCE(storage->buf)->data[0];
424 	else
425 		ptr = this_cpu_ptr(storage->percpu_buf);
426 
427 	return (unsigned long)ptr;
428 }
429 
430 const struct bpf_func_proto bpf_get_local_storage_proto = {
431 	.func		= bpf_get_local_storage,
432 	.gpl_only	= false,
433 	.ret_type	= RET_PTR_TO_MAP_VALUE,
434 	.arg1_type	= ARG_CONST_MAP_PTR,
435 	.arg2_type	= ARG_ANYTHING,
436 };
437 #endif
438 
439 #define BPF_STRTOX_BASE_MASK 0x1F
440 
__bpf_strtoull(const char * buf,size_t buf_len,u64 flags,unsigned long long * res,bool * is_negative)441 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
442 			  unsigned long long *res, bool *is_negative)
443 {
444 	unsigned int base = flags & BPF_STRTOX_BASE_MASK;
445 	const char *cur_buf = buf;
446 	size_t cur_len = buf_len;
447 	unsigned int consumed;
448 	size_t val_len;
449 	char str[64];
450 
451 	if (!buf || !buf_len || !res || !is_negative)
452 		return -EINVAL;
453 
454 	if (base != 0 && base != 8 && base != 10 && base != 16)
455 		return -EINVAL;
456 
457 	if (flags & ~BPF_STRTOX_BASE_MASK)
458 		return -EINVAL;
459 
460 	while (cur_buf < buf + buf_len && isspace(*cur_buf))
461 		++cur_buf;
462 
463 	*is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
464 	if (*is_negative)
465 		++cur_buf;
466 
467 	consumed = cur_buf - buf;
468 	cur_len -= consumed;
469 	if (!cur_len)
470 		return -EINVAL;
471 
472 	cur_len = min(cur_len, sizeof(str) - 1);
473 	memcpy(str, cur_buf, cur_len);
474 	str[cur_len] = '\0';
475 	cur_buf = str;
476 
477 	cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
478 	val_len = _parse_integer(cur_buf, base, res);
479 
480 	if (val_len & KSTRTOX_OVERFLOW)
481 		return -ERANGE;
482 
483 	if (val_len == 0)
484 		return -EINVAL;
485 
486 	cur_buf += val_len;
487 	consumed += cur_buf - str;
488 
489 	return consumed;
490 }
491 
__bpf_strtoll(const char * buf,size_t buf_len,u64 flags,long long * res)492 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
493 			 long long *res)
494 {
495 	unsigned long long _res;
496 	bool is_negative;
497 	int err;
498 
499 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
500 	if (err < 0)
501 		return err;
502 	if (is_negative) {
503 		if ((long long)-_res > 0)
504 			return -ERANGE;
505 		*res = -_res;
506 	} else {
507 		if ((long long)_res < 0)
508 			return -ERANGE;
509 		*res = _res;
510 	}
511 	return err;
512 }
513 
BPF_CALL_4(bpf_strtol,const char *,buf,size_t,buf_len,u64,flags,long *,res)514 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
515 	   long *, res)
516 {
517 	long long _res;
518 	int err;
519 
520 	err = __bpf_strtoll(buf, buf_len, flags, &_res);
521 	if (err < 0)
522 		return err;
523 	if (_res != (long)_res)
524 		return -ERANGE;
525 	*res = _res;
526 	return err;
527 }
528 
529 const struct bpf_func_proto bpf_strtol_proto = {
530 	.func		= bpf_strtol,
531 	.gpl_only	= false,
532 	.ret_type	= RET_INTEGER,
533 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
534 	.arg2_type	= ARG_CONST_SIZE,
535 	.arg3_type	= ARG_ANYTHING,
536 	.arg4_type	= ARG_PTR_TO_LONG,
537 };
538 
BPF_CALL_4(bpf_strtoul,const char *,buf,size_t,buf_len,u64,flags,unsigned long *,res)539 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
540 	   unsigned long *, res)
541 {
542 	unsigned long long _res;
543 	bool is_negative;
544 	int err;
545 
546 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
547 	if (err < 0)
548 		return err;
549 	if (is_negative)
550 		return -EINVAL;
551 	if (_res != (unsigned long)_res)
552 		return -ERANGE;
553 	*res = _res;
554 	return err;
555 }
556 
557 const struct bpf_func_proto bpf_strtoul_proto = {
558 	.func		= bpf_strtoul,
559 	.gpl_only	= false,
560 	.ret_type	= RET_INTEGER,
561 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
562 	.arg2_type	= ARG_CONST_SIZE,
563 	.arg3_type	= ARG_ANYTHING,
564 	.arg4_type	= ARG_PTR_TO_LONG,
565 };
566 #endif
567 
BPF_CALL_4(bpf_get_ns_current_pid_tgid,u64,dev,u64,ino,struct bpf_pidns_info *,nsdata,u32,size)568 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
569 	   struct bpf_pidns_info *, nsdata, u32, size)
570 {
571 	struct task_struct *task = current;
572 	struct pid_namespace *pidns;
573 	int err = -EINVAL;
574 
575 	if (unlikely(size != sizeof(struct bpf_pidns_info)))
576 		goto clear;
577 
578 	if (unlikely((u64)(dev_t)dev != dev))
579 		goto clear;
580 
581 	if (unlikely(!task))
582 		goto clear;
583 
584 	pidns = task_active_pid_ns(task);
585 	if (unlikely(!pidns)) {
586 		err = -ENOENT;
587 		goto clear;
588 	}
589 
590 	if (!ns_match(&pidns->ns, (dev_t)dev, ino))
591 		goto clear;
592 
593 	nsdata->pid = task_pid_nr_ns(task, pidns);
594 	nsdata->tgid = task_tgid_nr_ns(task, pidns);
595 	return 0;
596 clear:
597 	memset((void *)nsdata, 0, (size_t) size);
598 	return err;
599 }
600 
601 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
602 	.func		= bpf_get_ns_current_pid_tgid,
603 	.gpl_only	= false,
604 	.ret_type	= RET_INTEGER,
605 	.arg1_type	= ARG_ANYTHING,
606 	.arg2_type	= ARG_ANYTHING,
607 	.arg3_type      = ARG_PTR_TO_UNINIT_MEM,
608 	.arg4_type      = ARG_CONST_SIZE,
609 };
610 
611 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
612 	.func		= bpf_get_raw_cpu_id,
613 	.gpl_only	= false,
614 	.ret_type	= RET_INTEGER,
615 };
616 
BPF_CALL_5(bpf_event_output_data,void *,ctx,struct bpf_map *,map,u64,flags,void *,data,u64,size)617 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
618 	   u64, flags, void *, data, u64, size)
619 {
620 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
621 		return -EINVAL;
622 
623 	return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
624 }
625 
626 const struct bpf_func_proto bpf_event_output_data_proto =  {
627 	.func		= bpf_event_output_data,
628 	.gpl_only       = true,
629 	.ret_type       = RET_INTEGER,
630 	.arg1_type      = ARG_PTR_TO_CTX,
631 	.arg2_type      = ARG_CONST_MAP_PTR,
632 	.arg3_type      = ARG_ANYTHING,
633 	.arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
634 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
635 };
636 
BPF_CALL_3(bpf_copy_from_user,void *,dst,u32,size,const void __user *,user_ptr)637 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
638 	   const void __user *, user_ptr)
639 {
640 	int ret = copy_from_user(dst, user_ptr, size);
641 
642 	if (unlikely(ret)) {
643 		memset(dst, 0, size);
644 		ret = -EFAULT;
645 	}
646 
647 	return ret;
648 }
649 
650 const struct bpf_func_proto bpf_copy_from_user_proto = {
651 	.func		= bpf_copy_from_user,
652 	.gpl_only	= false,
653 	.ret_type	= RET_INTEGER,
654 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
655 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
656 	.arg3_type	= ARG_ANYTHING,
657 };
658 
BPF_CALL_2(bpf_per_cpu_ptr,const void *,ptr,u32,cpu)659 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
660 {
661 	if (cpu >= nr_cpu_ids)
662 		return (unsigned long)NULL;
663 
664 	return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
665 }
666 
667 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
668 	.func		= bpf_per_cpu_ptr,
669 	.gpl_only	= false,
670 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
671 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
672 	.arg2_type	= ARG_ANYTHING,
673 };
674 
BPF_CALL_1(bpf_this_cpu_ptr,const void *,percpu_ptr)675 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
676 {
677 	return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
678 }
679 
680 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
681 	.func		= bpf_this_cpu_ptr,
682 	.gpl_only	= false,
683 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
684 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
685 };
686 
bpf_trace_copy_string(char * buf,void * unsafe_ptr,char fmt_ptype,size_t bufsz)687 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
688 		size_t bufsz)
689 {
690 	void __user *user_ptr = (__force void __user *)unsafe_ptr;
691 
692 	buf[0] = 0;
693 
694 	switch (fmt_ptype) {
695 	case 's':
696 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
697 		if ((unsigned long)unsafe_ptr < TASK_SIZE)
698 			return strncpy_from_user_nofault(buf, user_ptr, bufsz);
699 		fallthrough;
700 #endif
701 	case 'k':
702 		return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
703 	case 'u':
704 		return strncpy_from_user_nofault(buf, user_ptr, bufsz);
705 	}
706 
707 	return -EINVAL;
708 }
709 
710 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
711  * arguments representation.
712  */
713 #define MAX_BPRINTF_BIN_ARGS	512
714 
715 /* Support executing three nested bprintf helper calls on a given CPU */
716 #define MAX_BPRINTF_NEST_LEVEL	3
717 struct bpf_bprintf_buffers {
718 	char bin_args[MAX_BPRINTF_BIN_ARGS];
719 	char buf[MAX_BPRINTF_BUF];
720 };
721 
722 static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
723 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
724 
try_get_buffers(struct bpf_bprintf_buffers ** bufs)725 static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
726 {
727 	int nest_level;
728 
729 	preempt_disable();
730 	nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
731 	if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
732 		this_cpu_dec(bpf_bprintf_nest_level);
733 		preempt_enable();
734 		return -EBUSY;
735 	}
736 	*bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
737 
738 	return 0;
739 }
740 
bpf_bprintf_cleanup(struct bpf_bprintf_data * data)741 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
742 {
743 	if (!data->bin_args && !data->buf)
744 		return;
745 	if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
746 		return;
747 	this_cpu_dec(bpf_bprintf_nest_level);
748 	preempt_enable();
749 }
750 
751 /*
752  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
753  *
754  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
755  *
756  * This can be used in two ways:
757  * - Format string verification only: when data->get_bin_args is false
758  * - Arguments preparation: in addition to the above verification, it writes in
759  *   data->bin_args a binary representation of arguments usable by bstr_printf
760  *   where pointers from BPF have been sanitized.
761  *
762  * In argument preparation mode, if 0 is returned, safe temporary buffers are
763  * allocated and bpf_bprintf_cleanup should be called to free them after use.
764  */
bpf_bprintf_prepare(char * fmt,u32 fmt_size,const u64 * raw_args,u32 num_args,struct bpf_bprintf_data * data)765 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
766 			u32 num_args, struct bpf_bprintf_data *data)
767 {
768 	bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
769 	char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
770 	struct bpf_bprintf_buffers *buffers = NULL;
771 	size_t sizeof_cur_arg, sizeof_cur_ip;
772 	int err, i, num_spec = 0;
773 	u64 cur_arg;
774 	char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
775 
776 	fmt_end = strnchr(fmt, fmt_size, 0);
777 	if (!fmt_end)
778 		return -EINVAL;
779 	fmt_size = fmt_end - fmt;
780 
781 	if (get_buffers && try_get_buffers(&buffers))
782 		return -EBUSY;
783 
784 	if (data->get_bin_args) {
785 		if (num_args)
786 			tmp_buf = buffers->bin_args;
787 		tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
788 		data->bin_args = (u32 *)tmp_buf;
789 	}
790 
791 	if (data->get_buf)
792 		data->buf = buffers->buf;
793 
794 	for (i = 0; i < fmt_size; i++) {
795 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
796 			err = -EINVAL;
797 			goto out;
798 		}
799 
800 		if (fmt[i] != '%')
801 			continue;
802 
803 		if (fmt[i + 1] == '%') {
804 			i++;
805 			continue;
806 		}
807 
808 		if (num_spec >= num_args) {
809 			err = -EINVAL;
810 			goto out;
811 		}
812 
813 		/* The string is zero-terminated so if fmt[i] != 0, we can
814 		 * always access fmt[i + 1], in the worst case it will be a 0
815 		 */
816 		i++;
817 
818 		/* skip optional "[0 +-][num]" width formatting field */
819 		while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
820 		       fmt[i] == ' ')
821 			i++;
822 		if (fmt[i] >= '1' && fmt[i] <= '9') {
823 			i++;
824 			while (fmt[i] >= '0' && fmt[i] <= '9')
825 				i++;
826 		}
827 
828 		if (fmt[i] == 'p') {
829 			sizeof_cur_arg = sizeof(long);
830 
831 			if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
832 			    fmt[i + 2] == 's') {
833 				fmt_ptype = fmt[i + 1];
834 				i += 2;
835 				goto fmt_str;
836 			}
837 
838 			if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
839 			    ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
840 			    fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
841 			    fmt[i + 1] == 'S') {
842 				/* just kernel pointers */
843 				if (tmp_buf)
844 					cur_arg = raw_args[num_spec];
845 				i++;
846 				goto nocopy_fmt;
847 			}
848 
849 			if (fmt[i + 1] == 'B') {
850 				if (tmp_buf)  {
851 					err = snprintf(tmp_buf,
852 						       (tmp_buf_end - tmp_buf),
853 						       "%pB",
854 						       (void *)(long)raw_args[num_spec]);
855 					tmp_buf += (err + 1);
856 				}
857 
858 				i++;
859 				num_spec++;
860 				continue;
861 			}
862 
863 			/* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
864 			if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
865 			    (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
866 				err = -EINVAL;
867 				goto out;
868 			}
869 
870 			i += 2;
871 			if (!tmp_buf)
872 				goto nocopy_fmt;
873 
874 			sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
875 			if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
876 				err = -ENOSPC;
877 				goto out;
878 			}
879 
880 			unsafe_ptr = (char *)(long)raw_args[num_spec];
881 			err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
882 						       sizeof_cur_ip);
883 			if (err < 0)
884 				memset(cur_ip, 0, sizeof_cur_ip);
885 
886 			/* hack: bstr_printf expects IP addresses to be
887 			 * pre-formatted as strings, ironically, the easiest way
888 			 * to do that is to call snprintf.
889 			 */
890 			ip_spec[2] = fmt[i - 1];
891 			ip_spec[3] = fmt[i];
892 			err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
893 				       ip_spec, &cur_ip);
894 
895 			tmp_buf += err + 1;
896 			num_spec++;
897 
898 			continue;
899 		} else if (fmt[i] == 's') {
900 			fmt_ptype = fmt[i];
901 fmt_str:
902 			if (fmt[i + 1] != 0 &&
903 			    !isspace(fmt[i + 1]) &&
904 			    !ispunct(fmt[i + 1])) {
905 				err = -EINVAL;
906 				goto out;
907 			}
908 
909 			if (!tmp_buf)
910 				goto nocopy_fmt;
911 
912 			if (tmp_buf_end == tmp_buf) {
913 				err = -ENOSPC;
914 				goto out;
915 			}
916 
917 			unsafe_ptr = (char *)(long)raw_args[num_spec];
918 			err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
919 						    fmt_ptype,
920 						    tmp_buf_end - tmp_buf);
921 			if (err < 0) {
922 				tmp_buf[0] = '\0';
923 				err = 1;
924 			}
925 
926 			tmp_buf += err;
927 			num_spec++;
928 
929 			continue;
930 		} else if (fmt[i] == 'c') {
931 			if (!tmp_buf)
932 				goto nocopy_fmt;
933 
934 			if (tmp_buf_end == tmp_buf) {
935 				err = -ENOSPC;
936 				goto out;
937 			}
938 
939 			*tmp_buf = raw_args[num_spec];
940 			tmp_buf++;
941 			num_spec++;
942 
943 			continue;
944 		}
945 
946 		sizeof_cur_arg = sizeof(int);
947 
948 		if (fmt[i] == 'l') {
949 			sizeof_cur_arg = sizeof(long);
950 			i++;
951 		}
952 		if (fmt[i] == 'l') {
953 			sizeof_cur_arg = sizeof(long long);
954 			i++;
955 		}
956 
957 		if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
958 		    fmt[i] != 'x' && fmt[i] != 'X') {
959 			err = -EINVAL;
960 			goto out;
961 		}
962 
963 		if (tmp_buf)
964 			cur_arg = raw_args[num_spec];
965 nocopy_fmt:
966 		if (tmp_buf) {
967 			tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
968 			if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
969 				err = -ENOSPC;
970 				goto out;
971 			}
972 
973 			if (sizeof_cur_arg == 8) {
974 				*(u32 *)tmp_buf = *(u32 *)&cur_arg;
975 				*(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
976 			} else {
977 				*(u32 *)tmp_buf = (u32)(long)cur_arg;
978 			}
979 			tmp_buf += sizeof_cur_arg;
980 		}
981 		num_spec++;
982 	}
983 
984 	err = 0;
985 out:
986 	if (err)
987 		bpf_bprintf_cleanup(data);
988 	return err;
989 }
990 
BPF_CALL_5(bpf_snprintf,char *,str,u32,str_size,char *,fmt,const void *,args,u32,data_len)991 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
992 	   const void *, args, u32, data_len)
993 {
994 	struct bpf_bprintf_data data = {
995 		.get_bin_args	= true,
996 	};
997 	int err, num_args;
998 
999 	if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1000 	    (data_len && !args))
1001 		return -EINVAL;
1002 	num_args = data_len / 8;
1003 
1004 	/* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1005 	 * can safely give an unbounded size.
1006 	 */
1007 	err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1008 	if (err < 0)
1009 		return err;
1010 
1011 	err = bstr_printf(str, str_size, fmt, data.bin_args);
1012 
1013 	bpf_bprintf_cleanup(&data);
1014 
1015 	return err + 1;
1016 }
1017 
1018 const struct bpf_func_proto bpf_snprintf_proto = {
1019 	.func		= bpf_snprintf,
1020 	.gpl_only	= true,
1021 	.ret_type	= RET_INTEGER,
1022 	.arg1_type	= ARG_PTR_TO_MEM_OR_NULL,
1023 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1024 	.arg3_type	= ARG_PTR_TO_CONST_STR,
1025 	.arg4_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1026 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1027 };
1028 
1029 /* BPF map elements can contain 'struct bpf_timer'.
1030  * Such map owns all of its BPF timers.
1031  * 'struct bpf_timer' is allocated as part of map element allocation
1032  * and it's zero initialized.
1033  * That space is used to keep 'struct bpf_timer_kern'.
1034  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1035  * remembers 'struct bpf_map *' pointer it's part of.
1036  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1037  * bpf_timer_start() arms the timer.
1038  * If user space reference to a map goes to zero at this point
1039  * ops->map_release_uref callback is responsible for cancelling the timers,
1040  * freeing their memory, and decrementing prog's refcnts.
1041  * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1042  * Inner maps can contain bpf timers as well. ops->map_release_uref is
1043  * freeing the timers when inner map is replaced or deleted by user space.
1044  */
1045 struct bpf_hrtimer {
1046 	struct hrtimer timer;
1047 	struct bpf_map *map;
1048 	struct bpf_prog *prog;
1049 	void __rcu *callback_fn;
1050 	void *value;
1051 	struct rcu_head rcu;
1052 };
1053 
1054 /* the actual struct hidden inside uapi struct bpf_timer */
1055 struct bpf_timer_kern {
1056 	struct bpf_hrtimer *timer;
1057 	/* bpf_spin_lock is used here instead of spinlock_t to make
1058 	 * sure that it always fits into space resereved by struct bpf_timer
1059 	 * regardless of LOCKDEP and spinlock debug flags.
1060 	 */
1061 	struct bpf_spin_lock lock;
1062 } __attribute__((aligned(8)));
1063 
1064 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1065 
bpf_timer_cb(struct hrtimer * hrtimer)1066 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1067 {
1068 	struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1069 	struct bpf_map *map = t->map;
1070 	void *value = t->value;
1071 	void *callback_fn;
1072 	void *key;
1073 	u32 idx;
1074 
1075 	callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1076 	if (!callback_fn)
1077 		goto out;
1078 
1079 	/* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1080 	 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1081 	 * Remember the timer this callback is servicing to prevent
1082 	 * deadlock if callback_fn() calls bpf_timer_cancel() or
1083 	 * bpf_map_delete_elem() on the same timer.
1084 	 */
1085 	this_cpu_write(hrtimer_running, t);
1086 	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1087 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1088 
1089 		/* compute the key */
1090 		idx = ((char *)value - array->value) / array->elem_size;
1091 		key = &idx;
1092 	} else { /* hash or lru */
1093 		key = value - round_up(map->key_size, 8);
1094 	}
1095 
1096 	BPF_CAST_CALL(callback_fn)((u64)(long)map, (u64)(long)key,
1097 				   (u64)(long)value, 0, 0);
1098 	/* The verifier checked that return value is zero. */
1099 
1100 	this_cpu_write(hrtimer_running, NULL);
1101 out:
1102 	return HRTIMER_NORESTART;
1103 }
1104 
BPF_CALL_3(bpf_timer_init,struct bpf_timer_kern *,timer,struct bpf_map *,map,u64,flags)1105 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1106 	   u64, flags)
1107 {
1108 	clockid_t clockid = flags & (MAX_CLOCKS - 1);
1109 	struct bpf_hrtimer *t;
1110 	int ret = 0;
1111 
1112 	BUILD_BUG_ON(MAX_CLOCKS != 16);
1113 	BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1114 	BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1115 
1116 	if (in_nmi())
1117 		return -EOPNOTSUPP;
1118 
1119 	if (flags >= MAX_CLOCKS ||
1120 	    /* similar to timerfd except _ALARM variants are not supported */
1121 	    (clockid != CLOCK_MONOTONIC &&
1122 	     clockid != CLOCK_REALTIME &&
1123 	     clockid != CLOCK_BOOTTIME))
1124 		return -EINVAL;
1125 	__bpf_spin_lock_irqsave(&timer->lock);
1126 	t = timer->timer;
1127 	if (t) {
1128 		ret = -EBUSY;
1129 		goto out;
1130 	}
1131 	/* allocate hrtimer via map_kmalloc to use memcg accounting */
1132 	t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1133 	if (!t) {
1134 		ret = -ENOMEM;
1135 		goto out;
1136 	}
1137 	t->value = (void *)timer - map->timer_off;
1138 	t->map = map;
1139 	t->prog = NULL;
1140 	rcu_assign_pointer(t->callback_fn, NULL);
1141 	hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1142 	t->timer.function = bpf_timer_cb;
1143 	WRITE_ONCE(timer->timer, t);
1144 	/* Guarantee the order between timer->timer and map->usercnt. So
1145 	 * when there are concurrent uref release and bpf timer init, either
1146 	 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1147 	 * timer or atomic64_read() below returns a zero usercnt.
1148 	 */
1149 	smp_mb();
1150 	if (!atomic64_read(&map->usercnt)) {
1151 		/* maps with timers must be either held by user space
1152 		 * or pinned in bpffs.
1153 		 */
1154 		WRITE_ONCE(timer->timer, NULL);
1155 		kfree(t);
1156 		ret = -EPERM;
1157 	}
1158 out:
1159 	__bpf_spin_unlock_irqrestore(&timer->lock);
1160 	return ret;
1161 }
1162 
1163 static const struct bpf_func_proto bpf_timer_init_proto = {
1164 	.func		= bpf_timer_init,
1165 	.gpl_only	= true,
1166 	.ret_type	= RET_INTEGER,
1167 	.arg1_type	= ARG_PTR_TO_TIMER,
1168 	.arg2_type	= ARG_CONST_MAP_PTR,
1169 	.arg3_type	= ARG_ANYTHING,
1170 };
1171 
BPF_CALL_3(bpf_timer_set_callback,struct bpf_timer_kern *,timer,void *,callback_fn,struct bpf_prog_aux *,aux)1172 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1173 	   struct bpf_prog_aux *, aux)
1174 {
1175 	struct bpf_prog *prev, *prog = aux->prog;
1176 	struct bpf_hrtimer *t;
1177 	int ret = 0;
1178 
1179 	if (in_nmi())
1180 		return -EOPNOTSUPP;
1181 	__bpf_spin_lock_irqsave(&timer->lock);
1182 	t = timer->timer;
1183 	if (!t) {
1184 		ret = -EINVAL;
1185 		goto out;
1186 	}
1187 	if (!atomic64_read(&t->map->usercnt)) {
1188 		/* maps with timers must be either held by user space
1189 		 * or pinned in bpffs. Otherwise timer might still be
1190 		 * running even when bpf prog is detached and user space
1191 		 * is gone, since map_release_uref won't ever be called.
1192 		 */
1193 		ret = -EPERM;
1194 		goto out;
1195 	}
1196 	prev = t->prog;
1197 	if (prev != prog) {
1198 		/* Bump prog refcnt once. Every bpf_timer_set_callback()
1199 		 * can pick different callback_fn-s within the same prog.
1200 		 */
1201 		prog = bpf_prog_inc_not_zero(prog);
1202 		if (IS_ERR(prog)) {
1203 			ret = PTR_ERR(prog);
1204 			goto out;
1205 		}
1206 		if (prev)
1207 			/* Drop prev prog refcnt when swapping with new prog */
1208 			bpf_prog_put(prev);
1209 		t->prog = prog;
1210 	}
1211 	rcu_assign_pointer(t->callback_fn, callback_fn);
1212 out:
1213 	__bpf_spin_unlock_irqrestore(&timer->lock);
1214 	return ret;
1215 }
1216 
1217 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1218 	.func		= bpf_timer_set_callback,
1219 	.gpl_only	= true,
1220 	.ret_type	= RET_INTEGER,
1221 	.arg1_type	= ARG_PTR_TO_TIMER,
1222 	.arg2_type	= ARG_PTR_TO_FUNC,
1223 };
1224 
BPF_CALL_3(bpf_timer_start,struct bpf_timer_kern *,timer,u64,nsecs,u64,flags)1225 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1226 {
1227 	struct bpf_hrtimer *t;
1228 	int ret = 0;
1229 
1230 	if (in_nmi())
1231 		return -EOPNOTSUPP;
1232 	if (flags)
1233 		return -EINVAL;
1234 	__bpf_spin_lock_irqsave(&timer->lock);
1235 	t = timer->timer;
1236 	if (!t || !t->prog) {
1237 		ret = -EINVAL;
1238 		goto out;
1239 	}
1240 	hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
1241 out:
1242 	__bpf_spin_unlock_irqrestore(&timer->lock);
1243 	return ret;
1244 }
1245 
1246 static const struct bpf_func_proto bpf_timer_start_proto = {
1247 	.func		= bpf_timer_start,
1248 	.gpl_only	= true,
1249 	.ret_type	= RET_INTEGER,
1250 	.arg1_type	= ARG_PTR_TO_TIMER,
1251 	.arg2_type	= ARG_ANYTHING,
1252 	.arg3_type	= ARG_ANYTHING,
1253 };
1254 
drop_prog_refcnt(struct bpf_hrtimer * t)1255 static void drop_prog_refcnt(struct bpf_hrtimer *t)
1256 {
1257 	struct bpf_prog *prog = t->prog;
1258 
1259 	if (prog) {
1260 		bpf_prog_put(prog);
1261 		t->prog = NULL;
1262 		rcu_assign_pointer(t->callback_fn, NULL);
1263 	}
1264 }
1265 
BPF_CALL_1(bpf_timer_cancel,struct bpf_timer_kern *,timer)1266 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1267 {
1268 	struct bpf_hrtimer *t;
1269 	int ret = 0;
1270 
1271 	if (in_nmi())
1272 		return -EOPNOTSUPP;
1273 	rcu_read_lock();
1274 	__bpf_spin_lock_irqsave(&timer->lock);
1275 	t = timer->timer;
1276 	if (!t) {
1277 		ret = -EINVAL;
1278 		goto out;
1279 	}
1280 	if (this_cpu_read(hrtimer_running) == t) {
1281 		/* If bpf callback_fn is trying to bpf_timer_cancel()
1282 		 * its own timer the hrtimer_cancel() will deadlock
1283 		 * since it waits for callback_fn to finish
1284 		 */
1285 		ret = -EDEADLK;
1286 		goto out;
1287 	}
1288 	drop_prog_refcnt(t);
1289 out:
1290 	__bpf_spin_unlock_irqrestore(&timer->lock);
1291 	/* Cancel the timer and wait for associated callback to finish
1292 	 * if it was running.
1293 	 */
1294 	ret = ret ?: hrtimer_cancel(&t->timer);
1295 	rcu_read_unlock();
1296 	return ret;
1297 }
1298 
1299 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1300 	.func		= bpf_timer_cancel,
1301 	.gpl_only	= true,
1302 	.ret_type	= RET_INTEGER,
1303 	.arg1_type	= ARG_PTR_TO_TIMER,
1304 };
1305 
1306 /* This function is called by map_delete/update_elem for individual element and
1307  * by ops->map_release_uref when the user space reference to a map reaches zero.
1308  */
bpf_timer_cancel_and_free(void * val)1309 void bpf_timer_cancel_and_free(void *val)
1310 {
1311 	struct bpf_timer_kern *timer = val;
1312 	struct bpf_hrtimer *t;
1313 
1314 	/* Performance optimization: read timer->timer without lock first. */
1315 	if (!READ_ONCE(timer->timer))
1316 		return;
1317 
1318 	__bpf_spin_lock_irqsave(&timer->lock);
1319 	/* re-read it under lock */
1320 	t = timer->timer;
1321 	if (!t)
1322 		goto out;
1323 	drop_prog_refcnt(t);
1324 	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1325 	 * this timer, since it won't be initialized.
1326 	 */
1327 	WRITE_ONCE(timer->timer, NULL);
1328 out:
1329 	__bpf_spin_unlock_irqrestore(&timer->lock);
1330 	if (!t)
1331 		return;
1332 	/* Cancel the timer and wait for callback to complete if it was running.
1333 	 * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1334 	 * right after for both preallocated and non-preallocated maps.
1335 	 * The timer->timer = NULL was already done and no code path can
1336 	 * see address 't' anymore.
1337 	 *
1338 	 * Check that bpf_map_delete/update_elem() wasn't called from timer
1339 	 * callback_fn. In such case don't call hrtimer_cancel() (since it will
1340 	 * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1341 	 * return -1). Though callback_fn is still running on this cpu it's
1342 	 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1343 	 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1344 	 * since timer->timer = NULL was already done. The timer will be
1345 	 * effectively cancelled because bpf_timer_cb() will return
1346 	 * HRTIMER_NORESTART.
1347 	 */
1348 	if (this_cpu_read(hrtimer_running) != t)
1349 		hrtimer_cancel(&t->timer);
1350 	kfree_rcu(t, rcu);
1351 }
1352 
1353 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1354 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1355 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1356 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1357 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1358 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1359 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1360 
1361 const struct bpf_func_proto *
bpf_base_func_proto(enum bpf_func_id func_id)1362 bpf_base_func_proto(enum bpf_func_id func_id)
1363 {
1364 	switch (func_id) {
1365 	case BPF_FUNC_map_lookup_elem:
1366 		return &bpf_map_lookup_elem_proto;
1367 	case BPF_FUNC_map_update_elem:
1368 		return &bpf_map_update_elem_proto;
1369 	case BPF_FUNC_map_delete_elem:
1370 		return &bpf_map_delete_elem_proto;
1371 	case BPF_FUNC_map_push_elem:
1372 		return &bpf_map_push_elem_proto;
1373 	case BPF_FUNC_map_pop_elem:
1374 		return &bpf_map_pop_elem_proto;
1375 	case BPF_FUNC_map_peek_elem:
1376 		return &bpf_map_peek_elem_proto;
1377 	case BPF_FUNC_get_prandom_u32:
1378 		return &bpf_get_prandom_u32_proto;
1379 	case BPF_FUNC_get_smp_processor_id:
1380 		return &bpf_get_raw_smp_processor_id_proto;
1381 	case BPF_FUNC_get_numa_node_id:
1382 		return &bpf_get_numa_node_id_proto;
1383 	case BPF_FUNC_tail_call:
1384 		return &bpf_tail_call_proto;
1385 	case BPF_FUNC_ktime_get_ns:
1386 		return &bpf_ktime_get_ns_proto;
1387 	case BPF_FUNC_ktime_get_boot_ns:
1388 		return &bpf_ktime_get_boot_ns_proto;
1389 	case BPF_FUNC_ringbuf_output:
1390 		return &bpf_ringbuf_output_proto;
1391 	case BPF_FUNC_ringbuf_reserve:
1392 		return &bpf_ringbuf_reserve_proto;
1393 	case BPF_FUNC_ringbuf_submit:
1394 		return &bpf_ringbuf_submit_proto;
1395 	case BPF_FUNC_ringbuf_discard:
1396 		return &bpf_ringbuf_discard_proto;
1397 	case BPF_FUNC_ringbuf_query:
1398 		return &bpf_ringbuf_query_proto;
1399 	case BPF_FUNC_for_each_map_elem:
1400 		return &bpf_for_each_map_elem_proto;
1401 	default:
1402 		break;
1403 	}
1404 
1405 	if (!bpf_capable())
1406 		return NULL;
1407 
1408 	switch (func_id) {
1409 	case BPF_FUNC_spin_lock:
1410 		return &bpf_spin_lock_proto;
1411 	case BPF_FUNC_spin_unlock:
1412 		return &bpf_spin_unlock_proto;
1413 	case BPF_FUNC_jiffies64:
1414 		return &bpf_jiffies64_proto;
1415 	case BPF_FUNC_per_cpu_ptr:
1416 		return &bpf_per_cpu_ptr_proto;
1417 	case BPF_FUNC_this_cpu_ptr:
1418 		return &bpf_this_cpu_ptr_proto;
1419 	case BPF_FUNC_timer_init:
1420 		return &bpf_timer_init_proto;
1421 	case BPF_FUNC_timer_set_callback:
1422 		return &bpf_timer_set_callback_proto;
1423 	case BPF_FUNC_timer_start:
1424 		return &bpf_timer_start_proto;
1425 	case BPF_FUNC_timer_cancel:
1426 		return &bpf_timer_cancel_proto;
1427 	default:
1428 		break;
1429 	}
1430 
1431 	if (!perfmon_capable())
1432 		return NULL;
1433 
1434 	switch (func_id) {
1435 	case BPF_FUNC_trace_printk:
1436 		return bpf_get_trace_printk_proto();
1437 	case BPF_FUNC_get_current_task:
1438 		return &bpf_get_current_task_proto;
1439 	case BPF_FUNC_get_current_task_btf:
1440 		return &bpf_get_current_task_btf_proto;
1441 	case BPF_FUNC_probe_read_user:
1442 		return &bpf_probe_read_user_proto;
1443 	case BPF_FUNC_probe_read_kernel:
1444 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1445 		       NULL : &bpf_probe_read_kernel_proto;
1446 	case BPF_FUNC_probe_read_user_str:
1447 		return &bpf_probe_read_user_str_proto;
1448 	case BPF_FUNC_probe_read_kernel_str:
1449 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1450 		       NULL : &bpf_probe_read_kernel_str_proto;
1451 	case BPF_FUNC_snprintf_btf:
1452 		return &bpf_snprintf_btf_proto;
1453 	case BPF_FUNC_snprintf:
1454 		return &bpf_snprintf_proto;
1455 	case BPF_FUNC_task_pt_regs:
1456 		return &bpf_task_pt_regs_proto;
1457 	default:
1458 		return NULL;
1459 	}
1460 }
1461