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/btf.h>
6 #include <linux/bpf-cgroup.h>
7 #include <linux/cgroup.h>
8 #include <linux/rcupdate.h>
9 #include <linux/random.h>
10 #include <linux/smp.h>
11 #include <linux/topology.h>
12 #include <linux/ktime.h>
13 #include <linux/sched.h>
14 #include <linux/uidgid.h>
15 #include <linux/filter.h>
16 #include <linux/ctype.h>
17 #include <linux/jiffies.h>
18 #include <linux/pid_namespace.h>
19 #include <linux/poison.h>
20 #include <linux/proc_ns.h>
21 #include <linux/sched/task.h>
22 #include <linux/security.h>
23 #include <linux/btf_ids.h>
24 #include <linux/bpf_mem_alloc.h>
25 
26 #include "../../lib/kstrtox.h"
27 
28 /* If kernel subsystem is allowing eBPF programs to call this function,
29  * inside its own verifier_ops->get_func_proto() callback it should return
30  * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
31  *
32  * Different map implementations will rely on rcu in map methods
33  * lookup/update/delete, therefore eBPF programs must run under rcu lock
34  * if program is allowed to access maps, so check rcu_read_lock_held() or
35  * rcu_read_lock_trace_held() in all three functions.
36  */
BPF_CALL_2(bpf_map_lookup_elem,struct bpf_map *,map,void *,key)37 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
38 {
39 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
40 		     !rcu_read_lock_bh_held());
41 	return (unsigned long) map->ops->map_lookup_elem(map, key);
42 }
43 
44 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
45 	.func		= bpf_map_lookup_elem,
46 	.gpl_only	= false,
47 	.pkt_access	= true,
48 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
49 	.arg1_type	= ARG_CONST_MAP_PTR,
50 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
51 };
52 
BPF_CALL_4(bpf_map_update_elem,struct bpf_map *,map,void *,key,void *,value,u64,flags)53 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
54 	   void *, value, u64, flags)
55 {
56 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
57 		     !rcu_read_lock_bh_held());
58 	return map->ops->map_update_elem(map, key, value, flags);
59 }
60 
61 const struct bpf_func_proto bpf_map_update_elem_proto = {
62 	.func		= bpf_map_update_elem,
63 	.gpl_only	= false,
64 	.pkt_access	= true,
65 	.ret_type	= RET_INTEGER,
66 	.arg1_type	= ARG_CONST_MAP_PTR,
67 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
68 	.arg3_type	= ARG_PTR_TO_MAP_VALUE,
69 	.arg4_type	= ARG_ANYTHING,
70 };
71 
BPF_CALL_2(bpf_map_delete_elem,struct bpf_map *,map,void *,key)72 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
73 {
74 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
75 		     !rcu_read_lock_bh_held());
76 	return map->ops->map_delete_elem(map, key);
77 }
78 
79 const struct bpf_func_proto bpf_map_delete_elem_proto = {
80 	.func		= bpf_map_delete_elem,
81 	.gpl_only	= false,
82 	.pkt_access	= true,
83 	.ret_type	= RET_INTEGER,
84 	.arg1_type	= ARG_CONST_MAP_PTR,
85 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
86 };
87 
BPF_CALL_3(bpf_map_push_elem,struct bpf_map *,map,void *,value,u64,flags)88 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
89 {
90 	return map->ops->map_push_elem(map, value, flags);
91 }
92 
93 const struct bpf_func_proto bpf_map_push_elem_proto = {
94 	.func		= bpf_map_push_elem,
95 	.gpl_only	= false,
96 	.pkt_access	= true,
97 	.ret_type	= RET_INTEGER,
98 	.arg1_type	= ARG_CONST_MAP_PTR,
99 	.arg2_type	= ARG_PTR_TO_MAP_VALUE,
100 	.arg3_type	= ARG_ANYTHING,
101 };
102 
BPF_CALL_2(bpf_map_pop_elem,struct bpf_map *,map,void *,value)103 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
104 {
105 	return map->ops->map_pop_elem(map, value);
106 }
107 
108 const struct bpf_func_proto bpf_map_pop_elem_proto = {
109 	.func		= bpf_map_pop_elem,
110 	.gpl_only	= false,
111 	.ret_type	= RET_INTEGER,
112 	.arg1_type	= ARG_CONST_MAP_PTR,
113 	.arg2_type	= ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
114 };
115 
BPF_CALL_2(bpf_map_peek_elem,struct bpf_map *,map,void *,value)116 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
117 {
118 	return map->ops->map_peek_elem(map, value);
119 }
120 
121 const struct bpf_func_proto bpf_map_peek_elem_proto = {
122 	.func		= bpf_map_peek_elem,
123 	.gpl_only	= false,
124 	.ret_type	= RET_INTEGER,
125 	.arg1_type	= ARG_CONST_MAP_PTR,
126 	.arg2_type	= ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
127 };
128 
BPF_CALL_3(bpf_map_lookup_percpu_elem,struct bpf_map *,map,void *,key,u32,cpu)129 BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
130 {
131 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
132 	return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu);
133 }
134 
135 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
136 	.func		= bpf_map_lookup_percpu_elem,
137 	.gpl_only	= false,
138 	.pkt_access	= true,
139 	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
140 	.arg1_type	= ARG_CONST_MAP_PTR,
141 	.arg2_type	= ARG_PTR_TO_MAP_KEY,
142 	.arg3_type	= ARG_ANYTHING,
143 };
144 
145 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
146 	.func		= bpf_user_rnd_u32,
147 	.gpl_only	= false,
148 	.ret_type	= RET_INTEGER,
149 };
150 
BPF_CALL_0(bpf_get_smp_processor_id)151 BPF_CALL_0(bpf_get_smp_processor_id)
152 {
153 	return smp_processor_id();
154 }
155 
156 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
157 	.func		= bpf_get_smp_processor_id,
158 	.gpl_only	= false,
159 	.ret_type	= RET_INTEGER,
160 };
161 
BPF_CALL_0(bpf_get_numa_node_id)162 BPF_CALL_0(bpf_get_numa_node_id)
163 {
164 	return numa_node_id();
165 }
166 
167 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
168 	.func		= bpf_get_numa_node_id,
169 	.gpl_only	= false,
170 	.ret_type	= RET_INTEGER,
171 };
172 
BPF_CALL_0(bpf_ktime_get_ns)173 BPF_CALL_0(bpf_ktime_get_ns)
174 {
175 	/* NMI safe access to clock monotonic */
176 	return ktime_get_mono_fast_ns();
177 }
178 
179 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
180 	.func		= bpf_ktime_get_ns,
181 	.gpl_only	= false,
182 	.ret_type	= RET_INTEGER,
183 };
184 
BPF_CALL_0(bpf_ktime_get_boot_ns)185 BPF_CALL_0(bpf_ktime_get_boot_ns)
186 {
187 	/* NMI safe access to clock boottime */
188 	return ktime_get_boot_fast_ns();
189 }
190 
191 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
192 	.func		= bpf_ktime_get_boot_ns,
193 	.gpl_only	= false,
194 	.ret_type	= RET_INTEGER,
195 };
196 
BPF_CALL_0(bpf_ktime_get_coarse_ns)197 BPF_CALL_0(bpf_ktime_get_coarse_ns)
198 {
199 	return ktime_get_coarse_ns();
200 }
201 
202 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
203 	.func		= bpf_ktime_get_coarse_ns,
204 	.gpl_only	= false,
205 	.ret_type	= RET_INTEGER,
206 };
207 
BPF_CALL_0(bpf_ktime_get_tai_ns)208 BPF_CALL_0(bpf_ktime_get_tai_ns)
209 {
210 	/* NMI safe access to clock tai */
211 	return ktime_get_tai_fast_ns();
212 }
213 
214 const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
215 	.func		= bpf_ktime_get_tai_ns,
216 	.gpl_only	= false,
217 	.ret_type	= RET_INTEGER,
218 };
219 
BPF_CALL_0(bpf_get_current_pid_tgid)220 BPF_CALL_0(bpf_get_current_pid_tgid)
221 {
222 	struct task_struct *task = current;
223 
224 	if (unlikely(!task))
225 		return -EINVAL;
226 
227 	return (u64) task->tgid << 32 | task->pid;
228 }
229 
230 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
231 	.func		= bpf_get_current_pid_tgid,
232 	.gpl_only	= false,
233 	.ret_type	= RET_INTEGER,
234 };
235 
BPF_CALL_0(bpf_get_current_uid_gid)236 BPF_CALL_0(bpf_get_current_uid_gid)
237 {
238 	struct task_struct *task = current;
239 	kuid_t uid;
240 	kgid_t gid;
241 
242 	if (unlikely(!task))
243 		return -EINVAL;
244 
245 	current_uid_gid(&uid, &gid);
246 	return (u64) from_kgid(&init_user_ns, gid) << 32 |
247 		     from_kuid(&init_user_ns, uid);
248 }
249 
250 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
251 	.func		= bpf_get_current_uid_gid,
252 	.gpl_only	= false,
253 	.ret_type	= RET_INTEGER,
254 };
255 
BPF_CALL_2(bpf_get_current_comm,char *,buf,u32,size)256 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
257 {
258 	struct task_struct *task = current;
259 
260 	if (unlikely(!task))
261 		goto err_clear;
262 
263 	/* Verifier guarantees that size > 0 */
264 	strscpy_pad(buf, task->comm, size);
265 	return 0;
266 err_clear:
267 	memset(buf, 0, size);
268 	return -EINVAL;
269 }
270 
271 const struct bpf_func_proto bpf_get_current_comm_proto = {
272 	.func		= bpf_get_current_comm,
273 	.gpl_only	= false,
274 	.ret_type	= RET_INTEGER,
275 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
276 	.arg2_type	= ARG_CONST_SIZE,
277 };
278 
279 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
280 
__bpf_spin_lock(struct bpf_spin_lock * lock)281 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
282 {
283 	arch_spinlock_t *l = (void *)lock;
284 	union {
285 		__u32 val;
286 		arch_spinlock_t lock;
287 	} u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
288 
289 	compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
290 	BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
291 	BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
292 	preempt_disable();
293 	arch_spin_lock(l);
294 }
295 
__bpf_spin_unlock(struct bpf_spin_lock * lock)296 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
297 {
298 	arch_spinlock_t *l = (void *)lock;
299 
300 	arch_spin_unlock(l);
301 	preempt_enable();
302 }
303 
304 #else
305 
__bpf_spin_lock(struct bpf_spin_lock * lock)306 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
307 {
308 	atomic_t *l = (void *)lock;
309 
310 	BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
311 	do {
312 		atomic_cond_read_relaxed(l, !VAL);
313 	} while (atomic_xchg(l, 1));
314 }
315 
__bpf_spin_unlock(struct bpf_spin_lock * lock)316 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
317 {
318 	atomic_t *l = (void *)lock;
319 
320 	atomic_set_release(l, 0);
321 }
322 
323 #endif
324 
325 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
326 
__bpf_spin_lock_irqsave(struct bpf_spin_lock * lock)327 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
328 {
329 	unsigned long flags;
330 
331 	local_irq_save(flags);
332 	__bpf_spin_lock(lock);
333 	__this_cpu_write(irqsave_flags, flags);
334 }
335 
BPF_CALL_1(bpf_spin_lock,struct bpf_spin_lock *,lock)336 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
337 {
338 	__bpf_spin_lock_irqsave(lock);
339 	return 0;
340 }
341 
342 const struct bpf_func_proto bpf_spin_lock_proto = {
343 	.func		= bpf_spin_lock,
344 	.gpl_only	= false,
345 	.ret_type	= RET_VOID,
346 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
347 	.arg1_btf_id    = BPF_PTR_POISON,
348 };
349 
__bpf_spin_unlock_irqrestore(struct bpf_spin_lock * lock)350 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
351 {
352 	unsigned long flags;
353 
354 	flags = __this_cpu_read(irqsave_flags);
355 	__bpf_spin_unlock(lock);
356 	local_irq_restore(flags);
357 }
358 
BPF_CALL_1(bpf_spin_unlock,struct bpf_spin_lock *,lock)359 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
360 {
361 	__bpf_spin_unlock_irqrestore(lock);
362 	return 0;
363 }
364 
365 const struct bpf_func_proto bpf_spin_unlock_proto = {
366 	.func		= bpf_spin_unlock,
367 	.gpl_only	= false,
368 	.ret_type	= RET_VOID,
369 	.arg1_type	= ARG_PTR_TO_SPIN_LOCK,
370 	.arg1_btf_id    = BPF_PTR_POISON,
371 };
372 
copy_map_value_locked(struct bpf_map * map,void * dst,void * src,bool lock_src)373 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
374 			   bool lock_src)
375 {
376 	struct bpf_spin_lock *lock;
377 
378 	if (lock_src)
379 		lock = src + map->record->spin_lock_off;
380 	else
381 		lock = dst + map->record->spin_lock_off;
382 	preempt_disable();
383 	__bpf_spin_lock_irqsave(lock);
384 	copy_map_value(map, dst, src);
385 	__bpf_spin_unlock_irqrestore(lock);
386 	preempt_enable();
387 }
388 
BPF_CALL_0(bpf_jiffies64)389 BPF_CALL_0(bpf_jiffies64)
390 {
391 	return get_jiffies_64();
392 }
393 
394 const struct bpf_func_proto bpf_jiffies64_proto = {
395 	.func		= bpf_jiffies64,
396 	.gpl_only	= false,
397 	.ret_type	= RET_INTEGER,
398 };
399 
400 #ifdef CONFIG_CGROUPS
BPF_CALL_0(bpf_get_current_cgroup_id)401 BPF_CALL_0(bpf_get_current_cgroup_id)
402 {
403 	struct cgroup *cgrp;
404 	u64 cgrp_id;
405 
406 	rcu_read_lock();
407 	cgrp = task_dfl_cgroup(current);
408 	cgrp_id = cgroup_id(cgrp);
409 	rcu_read_unlock();
410 
411 	return cgrp_id;
412 }
413 
414 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
415 	.func		= bpf_get_current_cgroup_id,
416 	.gpl_only	= false,
417 	.ret_type	= RET_INTEGER,
418 };
419 
BPF_CALL_1(bpf_get_current_ancestor_cgroup_id,int,ancestor_level)420 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
421 {
422 	struct cgroup *cgrp;
423 	struct cgroup *ancestor;
424 	u64 cgrp_id;
425 
426 	rcu_read_lock();
427 	cgrp = task_dfl_cgroup(current);
428 	ancestor = cgroup_ancestor(cgrp, ancestor_level);
429 	cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
430 	rcu_read_unlock();
431 
432 	return cgrp_id;
433 }
434 
435 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
436 	.func		= bpf_get_current_ancestor_cgroup_id,
437 	.gpl_only	= false,
438 	.ret_type	= RET_INTEGER,
439 	.arg1_type	= ARG_ANYTHING,
440 };
441 #endif /* CONFIG_CGROUPS */
442 
443 #define BPF_STRTOX_BASE_MASK 0x1F
444 
__bpf_strtoull(const char * buf,size_t buf_len,u64 flags,unsigned long long * res,bool * is_negative)445 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
446 			  unsigned long long *res, bool *is_negative)
447 {
448 	unsigned int base = flags & BPF_STRTOX_BASE_MASK;
449 	const char *cur_buf = buf;
450 	size_t cur_len = buf_len;
451 	unsigned int consumed;
452 	size_t val_len;
453 	char str[64];
454 
455 	if (!buf || !buf_len || !res || !is_negative)
456 		return -EINVAL;
457 
458 	if (base != 0 && base != 8 && base != 10 && base != 16)
459 		return -EINVAL;
460 
461 	if (flags & ~BPF_STRTOX_BASE_MASK)
462 		return -EINVAL;
463 
464 	while (cur_buf < buf + buf_len && isspace(*cur_buf))
465 		++cur_buf;
466 
467 	*is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
468 	if (*is_negative)
469 		++cur_buf;
470 
471 	consumed = cur_buf - buf;
472 	cur_len -= consumed;
473 	if (!cur_len)
474 		return -EINVAL;
475 
476 	cur_len = min(cur_len, sizeof(str) - 1);
477 	memcpy(str, cur_buf, cur_len);
478 	str[cur_len] = '\0';
479 	cur_buf = str;
480 
481 	cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
482 	val_len = _parse_integer(cur_buf, base, res);
483 
484 	if (val_len & KSTRTOX_OVERFLOW)
485 		return -ERANGE;
486 
487 	if (val_len == 0)
488 		return -EINVAL;
489 
490 	cur_buf += val_len;
491 	consumed += cur_buf - str;
492 
493 	return consumed;
494 }
495 
__bpf_strtoll(const char * buf,size_t buf_len,u64 flags,long long * res)496 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
497 			 long long *res)
498 {
499 	unsigned long long _res;
500 	bool is_negative;
501 	int err;
502 
503 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
504 	if (err < 0)
505 		return err;
506 	if (is_negative) {
507 		if ((long long)-_res > 0)
508 			return -ERANGE;
509 		*res = -_res;
510 	} else {
511 		if ((long long)_res < 0)
512 			return -ERANGE;
513 		*res = _res;
514 	}
515 	return err;
516 }
517 
BPF_CALL_4(bpf_strtol,const char *,buf,size_t,buf_len,u64,flags,long *,res)518 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
519 	   long *, res)
520 {
521 	long long _res;
522 	int err;
523 
524 	err = __bpf_strtoll(buf, buf_len, flags, &_res);
525 	if (err < 0)
526 		return err;
527 	if (_res != (long)_res)
528 		return -ERANGE;
529 	*res = _res;
530 	return err;
531 }
532 
533 const struct bpf_func_proto bpf_strtol_proto = {
534 	.func		= bpf_strtol,
535 	.gpl_only	= false,
536 	.ret_type	= RET_INTEGER,
537 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
538 	.arg2_type	= ARG_CONST_SIZE,
539 	.arg3_type	= ARG_ANYTHING,
540 	.arg4_type	= ARG_PTR_TO_LONG,
541 };
542 
BPF_CALL_4(bpf_strtoul,const char *,buf,size_t,buf_len,u64,flags,unsigned long *,res)543 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
544 	   unsigned long *, res)
545 {
546 	unsigned long long _res;
547 	bool is_negative;
548 	int err;
549 
550 	err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
551 	if (err < 0)
552 		return err;
553 	if (is_negative)
554 		return -EINVAL;
555 	if (_res != (unsigned long)_res)
556 		return -ERANGE;
557 	*res = _res;
558 	return err;
559 }
560 
561 const struct bpf_func_proto bpf_strtoul_proto = {
562 	.func		= bpf_strtoul,
563 	.gpl_only	= false,
564 	.ret_type	= RET_INTEGER,
565 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
566 	.arg2_type	= ARG_CONST_SIZE,
567 	.arg3_type	= ARG_ANYTHING,
568 	.arg4_type	= ARG_PTR_TO_LONG,
569 };
570 
BPF_CALL_3(bpf_strncmp,const char *,s1,u32,s1_sz,const char *,s2)571 BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
572 {
573 	return strncmp(s1, s2, s1_sz);
574 }
575 
576 static const struct bpf_func_proto bpf_strncmp_proto = {
577 	.func		= bpf_strncmp,
578 	.gpl_only	= false,
579 	.ret_type	= RET_INTEGER,
580 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
581 	.arg2_type	= ARG_CONST_SIZE,
582 	.arg3_type	= ARG_PTR_TO_CONST_STR,
583 };
584 
BPF_CALL_4(bpf_get_ns_current_pid_tgid,u64,dev,u64,ino,struct bpf_pidns_info *,nsdata,u32,size)585 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
586 	   struct bpf_pidns_info *, nsdata, u32, size)
587 {
588 	struct task_struct *task = current;
589 	struct pid_namespace *pidns;
590 	int err = -EINVAL;
591 
592 	if (unlikely(size != sizeof(struct bpf_pidns_info)))
593 		goto clear;
594 
595 	if (unlikely((u64)(dev_t)dev != dev))
596 		goto clear;
597 
598 	if (unlikely(!task))
599 		goto clear;
600 
601 	pidns = task_active_pid_ns(task);
602 	if (unlikely(!pidns)) {
603 		err = -ENOENT;
604 		goto clear;
605 	}
606 
607 	if (!ns_match(&pidns->ns, (dev_t)dev, ino))
608 		goto clear;
609 
610 	nsdata->pid = task_pid_nr_ns(task, pidns);
611 	nsdata->tgid = task_tgid_nr_ns(task, pidns);
612 	return 0;
613 clear:
614 	memset((void *)nsdata, 0, (size_t) size);
615 	return err;
616 }
617 
618 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
619 	.func		= bpf_get_ns_current_pid_tgid,
620 	.gpl_only	= false,
621 	.ret_type	= RET_INTEGER,
622 	.arg1_type	= ARG_ANYTHING,
623 	.arg2_type	= ARG_ANYTHING,
624 	.arg3_type      = ARG_PTR_TO_UNINIT_MEM,
625 	.arg4_type      = ARG_CONST_SIZE,
626 };
627 
628 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
629 	.func		= bpf_get_raw_cpu_id,
630 	.gpl_only	= false,
631 	.ret_type	= RET_INTEGER,
632 };
633 
BPF_CALL_5(bpf_event_output_data,void *,ctx,struct bpf_map *,map,u64,flags,void *,data,u64,size)634 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
635 	   u64, flags, void *, data, u64, size)
636 {
637 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
638 		return -EINVAL;
639 
640 	return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
641 }
642 
643 const struct bpf_func_proto bpf_event_output_data_proto =  {
644 	.func		= bpf_event_output_data,
645 	.gpl_only       = true,
646 	.ret_type       = RET_INTEGER,
647 	.arg1_type      = ARG_PTR_TO_CTX,
648 	.arg2_type      = ARG_CONST_MAP_PTR,
649 	.arg3_type      = ARG_ANYTHING,
650 	.arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
651 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
652 };
653 
BPF_CALL_3(bpf_copy_from_user,void *,dst,u32,size,const void __user *,user_ptr)654 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
655 	   const void __user *, user_ptr)
656 {
657 	int ret = copy_from_user(dst, user_ptr, size);
658 
659 	if (unlikely(ret)) {
660 		memset(dst, 0, size);
661 		ret = -EFAULT;
662 	}
663 
664 	return ret;
665 }
666 
667 const struct bpf_func_proto bpf_copy_from_user_proto = {
668 	.func		= bpf_copy_from_user,
669 	.gpl_only	= false,
670 	.might_sleep	= true,
671 	.ret_type	= RET_INTEGER,
672 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
673 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
674 	.arg3_type	= ARG_ANYTHING,
675 };
676 
BPF_CALL_5(bpf_copy_from_user_task,void *,dst,u32,size,const void __user *,user_ptr,struct task_struct *,tsk,u64,flags)677 BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
678 	   const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
679 {
680 	int ret;
681 
682 	/* flags is not used yet */
683 	if (unlikely(flags))
684 		return -EINVAL;
685 
686 	if (unlikely(!size))
687 		return 0;
688 
689 	ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
690 	if (ret == size)
691 		return 0;
692 
693 	memset(dst, 0, size);
694 	/* Return -EFAULT for partial read */
695 	return ret < 0 ? ret : -EFAULT;
696 }
697 
698 const struct bpf_func_proto bpf_copy_from_user_task_proto = {
699 	.func		= bpf_copy_from_user_task,
700 	.gpl_only	= true,
701 	.might_sleep	= true,
702 	.ret_type	= RET_INTEGER,
703 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
704 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
705 	.arg3_type	= ARG_ANYTHING,
706 	.arg4_type	= ARG_PTR_TO_BTF_ID,
707 	.arg4_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
708 	.arg5_type	= ARG_ANYTHING
709 };
710 
BPF_CALL_2(bpf_per_cpu_ptr,const void *,ptr,u32,cpu)711 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
712 {
713 	if (cpu >= nr_cpu_ids)
714 		return (unsigned long)NULL;
715 
716 	return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
717 }
718 
719 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
720 	.func		= bpf_per_cpu_ptr,
721 	.gpl_only	= false,
722 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
723 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
724 	.arg2_type	= ARG_ANYTHING,
725 };
726 
BPF_CALL_1(bpf_this_cpu_ptr,const void *,percpu_ptr)727 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
728 {
729 	return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
730 }
731 
732 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
733 	.func		= bpf_this_cpu_ptr,
734 	.gpl_only	= false,
735 	.ret_type	= RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
736 	.arg1_type	= ARG_PTR_TO_PERCPU_BTF_ID,
737 };
738 
bpf_trace_copy_string(char * buf,void * unsafe_ptr,char fmt_ptype,size_t bufsz)739 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
740 		size_t bufsz)
741 {
742 	void __user *user_ptr = (__force void __user *)unsafe_ptr;
743 
744 	buf[0] = 0;
745 
746 	switch (fmt_ptype) {
747 	case 's':
748 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
749 		if ((unsigned long)unsafe_ptr < TASK_SIZE)
750 			return strncpy_from_user_nofault(buf, user_ptr, bufsz);
751 		fallthrough;
752 #endif
753 	case 'k':
754 		return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
755 	case 'u':
756 		return strncpy_from_user_nofault(buf, user_ptr, bufsz);
757 	}
758 
759 	return -EINVAL;
760 }
761 
762 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
763  * arguments representation.
764  */
765 #define MAX_BPRINTF_BIN_ARGS	512
766 
767 /* Support executing three nested bprintf helper calls on a given CPU */
768 #define MAX_BPRINTF_NEST_LEVEL	3
769 struct bpf_bprintf_buffers {
770 	char bin_args[MAX_BPRINTF_BIN_ARGS];
771 	char buf[MAX_BPRINTF_BUF];
772 };
773 
774 static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
775 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
776 
try_get_buffers(struct bpf_bprintf_buffers ** bufs)777 static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
778 {
779 	int nest_level;
780 
781 	preempt_disable();
782 	nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
783 	if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
784 		this_cpu_dec(bpf_bprintf_nest_level);
785 		preempt_enable();
786 		return -EBUSY;
787 	}
788 	*bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
789 
790 	return 0;
791 }
792 
bpf_bprintf_cleanup(struct bpf_bprintf_data * data)793 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
794 {
795 	if (!data->bin_args && !data->buf)
796 		return;
797 	if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
798 		return;
799 	this_cpu_dec(bpf_bprintf_nest_level);
800 	preempt_enable();
801 }
802 
803 /*
804  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
805  *
806  * Returns a negative value if fmt is an invalid format string or 0 otherwise.
807  *
808  * This can be used in two ways:
809  * - Format string verification only: when data->get_bin_args is false
810  * - Arguments preparation: in addition to the above verification, it writes in
811  *   data->bin_args a binary representation of arguments usable by bstr_printf
812  *   where pointers from BPF have been sanitized.
813  *
814  * In argument preparation mode, if 0 is returned, safe temporary buffers are
815  * allocated and bpf_bprintf_cleanup should be called to free them after use.
816  */
bpf_bprintf_prepare(char * fmt,u32 fmt_size,const u64 * raw_args,u32 num_args,struct bpf_bprintf_data * data)817 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
818 			u32 num_args, struct bpf_bprintf_data *data)
819 {
820 	bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
821 	char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
822 	struct bpf_bprintf_buffers *buffers = NULL;
823 	size_t sizeof_cur_arg, sizeof_cur_ip;
824 	int err, i, num_spec = 0;
825 	u64 cur_arg;
826 	char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
827 
828 	fmt_end = strnchr(fmt, fmt_size, 0);
829 	if (!fmt_end)
830 		return -EINVAL;
831 	fmt_size = fmt_end - fmt;
832 
833 	if (get_buffers && try_get_buffers(&buffers))
834 		return -EBUSY;
835 
836 	if (data->get_bin_args) {
837 		if (num_args)
838 			tmp_buf = buffers->bin_args;
839 		tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
840 		data->bin_args = (u32 *)tmp_buf;
841 	}
842 
843 	if (data->get_buf)
844 		data->buf = buffers->buf;
845 
846 	for (i = 0; i < fmt_size; i++) {
847 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
848 			err = -EINVAL;
849 			goto out;
850 		}
851 
852 		if (fmt[i] != '%')
853 			continue;
854 
855 		if (fmt[i + 1] == '%') {
856 			i++;
857 			continue;
858 		}
859 
860 		if (num_spec >= num_args) {
861 			err = -EINVAL;
862 			goto out;
863 		}
864 
865 		/* The string is zero-terminated so if fmt[i] != 0, we can
866 		 * always access fmt[i + 1], in the worst case it will be a 0
867 		 */
868 		i++;
869 
870 		/* skip optional "[0 +-][num]" width formatting field */
871 		while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
872 		       fmt[i] == ' ')
873 			i++;
874 		if (fmt[i] >= '1' && fmt[i] <= '9') {
875 			i++;
876 			while (fmt[i] >= '0' && fmt[i] <= '9')
877 				i++;
878 		}
879 
880 		if (fmt[i] == 'p') {
881 			sizeof_cur_arg = sizeof(long);
882 
883 			if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
884 			    fmt[i + 2] == 's') {
885 				fmt_ptype = fmt[i + 1];
886 				i += 2;
887 				goto fmt_str;
888 			}
889 
890 			if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
891 			    ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
892 			    fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
893 			    fmt[i + 1] == 'S') {
894 				/* just kernel pointers */
895 				if (tmp_buf)
896 					cur_arg = raw_args[num_spec];
897 				i++;
898 				goto nocopy_fmt;
899 			}
900 
901 			if (fmt[i + 1] == 'B') {
902 				if (tmp_buf)  {
903 					err = snprintf(tmp_buf,
904 						       (tmp_buf_end - tmp_buf),
905 						       "%pB",
906 						       (void *)(long)raw_args[num_spec]);
907 					tmp_buf += (err + 1);
908 				}
909 
910 				i++;
911 				num_spec++;
912 				continue;
913 			}
914 
915 			/* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
916 			if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
917 			    (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
918 				err = -EINVAL;
919 				goto out;
920 			}
921 
922 			i += 2;
923 			if (!tmp_buf)
924 				goto nocopy_fmt;
925 
926 			sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
927 			if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
928 				err = -ENOSPC;
929 				goto out;
930 			}
931 
932 			unsafe_ptr = (char *)(long)raw_args[num_spec];
933 			err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
934 						       sizeof_cur_ip);
935 			if (err < 0)
936 				memset(cur_ip, 0, sizeof_cur_ip);
937 
938 			/* hack: bstr_printf expects IP addresses to be
939 			 * pre-formatted as strings, ironically, the easiest way
940 			 * to do that is to call snprintf.
941 			 */
942 			ip_spec[2] = fmt[i - 1];
943 			ip_spec[3] = fmt[i];
944 			err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
945 				       ip_spec, &cur_ip);
946 
947 			tmp_buf += err + 1;
948 			num_spec++;
949 
950 			continue;
951 		} else if (fmt[i] == 's') {
952 			fmt_ptype = fmt[i];
953 fmt_str:
954 			if (fmt[i + 1] != 0 &&
955 			    !isspace(fmt[i + 1]) &&
956 			    !ispunct(fmt[i + 1])) {
957 				err = -EINVAL;
958 				goto out;
959 			}
960 
961 			if (!tmp_buf)
962 				goto nocopy_fmt;
963 
964 			if (tmp_buf_end == tmp_buf) {
965 				err = -ENOSPC;
966 				goto out;
967 			}
968 
969 			unsafe_ptr = (char *)(long)raw_args[num_spec];
970 			err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
971 						    fmt_ptype,
972 						    tmp_buf_end - tmp_buf);
973 			if (err < 0) {
974 				tmp_buf[0] = '\0';
975 				err = 1;
976 			}
977 
978 			tmp_buf += err;
979 			num_spec++;
980 
981 			continue;
982 		} else if (fmt[i] == 'c') {
983 			if (!tmp_buf)
984 				goto nocopy_fmt;
985 
986 			if (tmp_buf_end == tmp_buf) {
987 				err = -ENOSPC;
988 				goto out;
989 			}
990 
991 			*tmp_buf = raw_args[num_spec];
992 			tmp_buf++;
993 			num_spec++;
994 
995 			continue;
996 		}
997 
998 		sizeof_cur_arg = sizeof(int);
999 
1000 		if (fmt[i] == 'l') {
1001 			sizeof_cur_arg = sizeof(long);
1002 			i++;
1003 		}
1004 		if (fmt[i] == 'l') {
1005 			sizeof_cur_arg = sizeof(long long);
1006 			i++;
1007 		}
1008 
1009 		if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1010 		    fmt[i] != 'x' && fmt[i] != 'X') {
1011 			err = -EINVAL;
1012 			goto out;
1013 		}
1014 
1015 		if (tmp_buf)
1016 			cur_arg = raw_args[num_spec];
1017 nocopy_fmt:
1018 		if (tmp_buf) {
1019 			tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1020 			if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1021 				err = -ENOSPC;
1022 				goto out;
1023 			}
1024 
1025 			if (sizeof_cur_arg == 8) {
1026 				*(u32 *)tmp_buf = *(u32 *)&cur_arg;
1027 				*(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1028 			} else {
1029 				*(u32 *)tmp_buf = (u32)(long)cur_arg;
1030 			}
1031 			tmp_buf += sizeof_cur_arg;
1032 		}
1033 		num_spec++;
1034 	}
1035 
1036 	err = 0;
1037 out:
1038 	if (err)
1039 		bpf_bprintf_cleanup(data);
1040 	return err;
1041 }
1042 
BPF_CALL_5(bpf_snprintf,char *,str,u32,str_size,char *,fmt,const void *,args,u32,data_len)1043 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1044 	   const void *, args, u32, data_len)
1045 {
1046 	struct bpf_bprintf_data data = {
1047 		.get_bin_args	= true,
1048 	};
1049 	int err, num_args;
1050 
1051 	if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1052 	    (data_len && !args))
1053 		return -EINVAL;
1054 	num_args = data_len / 8;
1055 
1056 	/* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1057 	 * can safely give an unbounded size.
1058 	 */
1059 	err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1060 	if (err < 0)
1061 		return err;
1062 
1063 	err = bstr_printf(str, str_size, fmt, data.bin_args);
1064 
1065 	bpf_bprintf_cleanup(&data);
1066 
1067 	return err + 1;
1068 }
1069 
1070 const struct bpf_func_proto bpf_snprintf_proto = {
1071 	.func		= bpf_snprintf,
1072 	.gpl_only	= true,
1073 	.ret_type	= RET_INTEGER,
1074 	.arg1_type	= ARG_PTR_TO_MEM_OR_NULL,
1075 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1076 	.arg3_type	= ARG_PTR_TO_CONST_STR,
1077 	.arg4_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1078 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1079 };
1080 
1081 /* BPF map elements can contain 'struct bpf_timer'.
1082  * Such map owns all of its BPF timers.
1083  * 'struct bpf_timer' is allocated as part of map element allocation
1084  * and it's zero initialized.
1085  * That space is used to keep 'struct bpf_timer_kern'.
1086  * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1087  * remembers 'struct bpf_map *' pointer it's part of.
1088  * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1089  * bpf_timer_start() arms the timer.
1090  * If user space reference to a map goes to zero at this point
1091  * ops->map_release_uref callback is responsible for cancelling the timers,
1092  * freeing their memory, and decrementing prog's refcnts.
1093  * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1094  * Inner maps can contain bpf timers as well. ops->map_release_uref is
1095  * freeing the timers when inner map is replaced or deleted by user space.
1096  */
1097 struct bpf_hrtimer {
1098 	struct hrtimer timer;
1099 	struct bpf_map *map;
1100 	struct bpf_prog *prog;
1101 	void __rcu *callback_fn;
1102 	void *value;
1103 	struct rcu_head rcu;
1104 };
1105 
1106 /* the actual struct hidden inside uapi struct bpf_timer */
1107 struct bpf_timer_kern {
1108 	struct bpf_hrtimer *timer;
1109 	/* bpf_spin_lock is used here instead of spinlock_t to make
1110 	 * sure that it always fits into space reserved by struct bpf_timer
1111 	 * regardless of LOCKDEP and spinlock debug flags.
1112 	 */
1113 	struct bpf_spin_lock lock;
1114 } __attribute__((aligned(8)));
1115 
1116 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1117 
bpf_timer_cb(struct hrtimer * hrtimer)1118 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1119 {
1120 	struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1121 	struct bpf_map *map = t->map;
1122 	void *value = t->value;
1123 	bpf_callback_t callback_fn;
1124 	void *key;
1125 	u32 idx;
1126 
1127 	BTF_TYPE_EMIT(struct bpf_timer);
1128 	callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1129 	if (!callback_fn)
1130 		goto out;
1131 
1132 	/* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1133 	 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1134 	 * Remember the timer this callback is servicing to prevent
1135 	 * deadlock if callback_fn() calls bpf_timer_cancel() or
1136 	 * bpf_map_delete_elem() on the same timer.
1137 	 */
1138 	this_cpu_write(hrtimer_running, t);
1139 	if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1140 		struct bpf_array *array = container_of(map, struct bpf_array, map);
1141 
1142 		/* compute the key */
1143 		idx = ((char *)value - array->value) / array->elem_size;
1144 		key = &idx;
1145 	} else { /* hash or lru */
1146 		key = value - round_up(map->key_size, 8);
1147 	}
1148 
1149 	callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1150 	/* The verifier checked that return value is zero. */
1151 
1152 	this_cpu_write(hrtimer_running, NULL);
1153 out:
1154 	return HRTIMER_NORESTART;
1155 }
1156 
BPF_CALL_3(bpf_timer_init,struct bpf_timer_kern *,timer,struct bpf_map *,map,u64,flags)1157 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1158 	   u64, flags)
1159 {
1160 	clockid_t clockid = flags & (MAX_CLOCKS - 1);
1161 	struct bpf_hrtimer *t;
1162 	int ret = 0;
1163 
1164 	BUILD_BUG_ON(MAX_CLOCKS != 16);
1165 	BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1166 	BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1167 
1168 	if (in_nmi())
1169 		return -EOPNOTSUPP;
1170 
1171 	if (flags >= MAX_CLOCKS ||
1172 	    /* similar to timerfd except _ALARM variants are not supported */
1173 	    (clockid != CLOCK_MONOTONIC &&
1174 	     clockid != CLOCK_REALTIME &&
1175 	     clockid != CLOCK_BOOTTIME))
1176 		return -EINVAL;
1177 	__bpf_spin_lock_irqsave(&timer->lock);
1178 	t = timer->timer;
1179 	if (t) {
1180 		ret = -EBUSY;
1181 		goto out;
1182 	}
1183 	/* allocate hrtimer via map_kmalloc to use memcg accounting */
1184 	t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1185 	if (!t) {
1186 		ret = -ENOMEM;
1187 		goto out;
1188 	}
1189 	t->value = (void *)timer - map->record->timer_off;
1190 	t->map = map;
1191 	t->prog = NULL;
1192 	rcu_assign_pointer(t->callback_fn, NULL);
1193 	hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1194 	t->timer.function = bpf_timer_cb;
1195 	WRITE_ONCE(timer->timer, t);
1196 	/* Guarantee the order between timer->timer and map->usercnt. So
1197 	 * when there are concurrent uref release and bpf timer init, either
1198 	 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1199 	 * timer or atomic64_read() below returns a zero usercnt.
1200 	 */
1201 	smp_mb();
1202 	if (!atomic64_read(&map->usercnt)) {
1203 		/* maps with timers must be either held by user space
1204 		 * or pinned in bpffs.
1205 		 */
1206 		WRITE_ONCE(timer->timer, NULL);
1207 		kfree(t);
1208 		ret = -EPERM;
1209 	}
1210 out:
1211 	__bpf_spin_unlock_irqrestore(&timer->lock);
1212 	return ret;
1213 }
1214 
1215 static const struct bpf_func_proto bpf_timer_init_proto = {
1216 	.func		= bpf_timer_init,
1217 	.gpl_only	= true,
1218 	.ret_type	= RET_INTEGER,
1219 	.arg1_type	= ARG_PTR_TO_TIMER,
1220 	.arg2_type	= ARG_CONST_MAP_PTR,
1221 	.arg3_type	= ARG_ANYTHING,
1222 };
1223 
BPF_CALL_3(bpf_timer_set_callback,struct bpf_timer_kern *,timer,void *,callback_fn,struct bpf_prog_aux *,aux)1224 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1225 	   struct bpf_prog_aux *, aux)
1226 {
1227 	struct bpf_prog *prev, *prog = aux->prog;
1228 	struct bpf_hrtimer *t;
1229 	int ret = 0;
1230 
1231 	if (in_nmi())
1232 		return -EOPNOTSUPP;
1233 	__bpf_spin_lock_irqsave(&timer->lock);
1234 	t = timer->timer;
1235 	if (!t) {
1236 		ret = -EINVAL;
1237 		goto out;
1238 	}
1239 	if (!atomic64_read(&t->map->usercnt)) {
1240 		/* maps with timers must be either held by user space
1241 		 * or pinned in bpffs. Otherwise timer might still be
1242 		 * running even when bpf prog is detached and user space
1243 		 * is gone, since map_release_uref won't ever be called.
1244 		 */
1245 		ret = -EPERM;
1246 		goto out;
1247 	}
1248 	prev = t->prog;
1249 	if (prev != prog) {
1250 		/* Bump prog refcnt once. Every bpf_timer_set_callback()
1251 		 * can pick different callback_fn-s within the same prog.
1252 		 */
1253 		prog = bpf_prog_inc_not_zero(prog);
1254 		if (IS_ERR(prog)) {
1255 			ret = PTR_ERR(prog);
1256 			goto out;
1257 		}
1258 		if (prev)
1259 			/* Drop prev prog refcnt when swapping with new prog */
1260 			bpf_prog_put(prev);
1261 		t->prog = prog;
1262 	}
1263 	rcu_assign_pointer(t->callback_fn, callback_fn);
1264 out:
1265 	__bpf_spin_unlock_irqrestore(&timer->lock);
1266 	return ret;
1267 }
1268 
1269 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1270 	.func		= bpf_timer_set_callback,
1271 	.gpl_only	= true,
1272 	.ret_type	= RET_INTEGER,
1273 	.arg1_type	= ARG_PTR_TO_TIMER,
1274 	.arg2_type	= ARG_PTR_TO_FUNC,
1275 };
1276 
BPF_CALL_3(bpf_timer_start,struct bpf_timer_kern *,timer,u64,nsecs,u64,flags)1277 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1278 {
1279 	struct bpf_hrtimer *t;
1280 	int ret = 0;
1281 	enum hrtimer_mode mode;
1282 
1283 	if (in_nmi())
1284 		return -EOPNOTSUPP;
1285 	if (flags > BPF_F_TIMER_ABS)
1286 		return -EINVAL;
1287 	__bpf_spin_lock_irqsave(&timer->lock);
1288 	t = timer->timer;
1289 	if (!t || !t->prog) {
1290 		ret = -EINVAL;
1291 		goto out;
1292 	}
1293 
1294 	if (flags & BPF_F_TIMER_ABS)
1295 		mode = HRTIMER_MODE_ABS_SOFT;
1296 	else
1297 		mode = HRTIMER_MODE_REL_SOFT;
1298 
1299 	hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
1300 out:
1301 	__bpf_spin_unlock_irqrestore(&timer->lock);
1302 	return ret;
1303 }
1304 
1305 static const struct bpf_func_proto bpf_timer_start_proto = {
1306 	.func		= bpf_timer_start,
1307 	.gpl_only	= true,
1308 	.ret_type	= RET_INTEGER,
1309 	.arg1_type	= ARG_PTR_TO_TIMER,
1310 	.arg2_type	= ARG_ANYTHING,
1311 	.arg3_type	= ARG_ANYTHING,
1312 };
1313 
drop_prog_refcnt(struct bpf_hrtimer * t)1314 static void drop_prog_refcnt(struct bpf_hrtimer *t)
1315 {
1316 	struct bpf_prog *prog = t->prog;
1317 
1318 	if (prog) {
1319 		bpf_prog_put(prog);
1320 		t->prog = NULL;
1321 		rcu_assign_pointer(t->callback_fn, NULL);
1322 	}
1323 }
1324 
BPF_CALL_1(bpf_timer_cancel,struct bpf_timer_kern *,timer)1325 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1326 {
1327 	struct bpf_hrtimer *t;
1328 	int ret = 0;
1329 
1330 	if (in_nmi())
1331 		return -EOPNOTSUPP;
1332 	rcu_read_lock();
1333 	__bpf_spin_lock_irqsave(&timer->lock);
1334 	t = timer->timer;
1335 	if (!t) {
1336 		ret = -EINVAL;
1337 		goto out;
1338 	}
1339 	if (this_cpu_read(hrtimer_running) == t) {
1340 		/* If bpf callback_fn is trying to bpf_timer_cancel()
1341 		 * its own timer the hrtimer_cancel() will deadlock
1342 		 * since it waits for callback_fn to finish
1343 		 */
1344 		ret = -EDEADLK;
1345 		goto out;
1346 	}
1347 	drop_prog_refcnt(t);
1348 out:
1349 	__bpf_spin_unlock_irqrestore(&timer->lock);
1350 	/* Cancel the timer and wait for associated callback to finish
1351 	 * if it was running.
1352 	 */
1353 	ret = ret ?: hrtimer_cancel(&t->timer);
1354 	rcu_read_unlock();
1355 	return ret;
1356 }
1357 
1358 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1359 	.func		= bpf_timer_cancel,
1360 	.gpl_only	= true,
1361 	.ret_type	= RET_INTEGER,
1362 	.arg1_type	= ARG_PTR_TO_TIMER,
1363 };
1364 
1365 /* This function is called by map_delete/update_elem for individual element and
1366  * by ops->map_release_uref when the user space reference to a map reaches zero.
1367  */
bpf_timer_cancel_and_free(void * val)1368 void bpf_timer_cancel_and_free(void *val)
1369 {
1370 	struct bpf_timer_kern *timer = val;
1371 	struct bpf_hrtimer *t;
1372 
1373 	/* Performance optimization: read timer->timer without lock first. */
1374 	if (!READ_ONCE(timer->timer))
1375 		return;
1376 
1377 	__bpf_spin_lock_irqsave(&timer->lock);
1378 	/* re-read it under lock */
1379 	t = timer->timer;
1380 	if (!t)
1381 		goto out;
1382 	drop_prog_refcnt(t);
1383 	/* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1384 	 * this timer, since it won't be initialized.
1385 	 */
1386 	WRITE_ONCE(timer->timer, NULL);
1387 out:
1388 	__bpf_spin_unlock_irqrestore(&timer->lock);
1389 	if (!t)
1390 		return;
1391 	/* Cancel the timer and wait for callback to complete if it was running.
1392 	 * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1393 	 * right after for both preallocated and non-preallocated maps.
1394 	 * The timer->timer = NULL was already done and no code path can
1395 	 * see address 't' anymore.
1396 	 *
1397 	 * Check that bpf_map_delete/update_elem() wasn't called from timer
1398 	 * callback_fn. In such case don't call hrtimer_cancel() (since it will
1399 	 * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1400 	 * return -1). Though callback_fn is still running on this cpu it's
1401 	 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1402 	 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1403 	 * since timer->timer = NULL was already done. The timer will be
1404 	 * effectively cancelled because bpf_timer_cb() will return
1405 	 * HRTIMER_NORESTART.
1406 	 */
1407 	if (this_cpu_read(hrtimer_running) != t)
1408 		hrtimer_cancel(&t->timer);
1409 	kfree_rcu(t, rcu);
1410 }
1411 
BPF_CALL_2(bpf_kptr_xchg,void *,map_value,void *,ptr)1412 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
1413 {
1414 	unsigned long *kptr = map_value;
1415 
1416 	return xchg(kptr, (unsigned long)ptr);
1417 }
1418 
1419 /* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1420  * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1421  * denote type that verifier will determine.
1422  */
1423 static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1424 	.func         = bpf_kptr_xchg,
1425 	.gpl_only     = false,
1426 	.ret_type     = RET_PTR_TO_BTF_ID_OR_NULL,
1427 	.ret_btf_id   = BPF_PTR_POISON,
1428 	.arg1_type    = ARG_PTR_TO_KPTR,
1429 	.arg2_type    = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1430 	.arg2_btf_id  = BPF_PTR_POISON,
1431 };
1432 
1433 /* Since the upper 8 bits of dynptr->size is reserved, the
1434  * maximum supported size is 2^24 - 1.
1435  */
1436 #define DYNPTR_MAX_SIZE	((1UL << 24) - 1)
1437 #define DYNPTR_TYPE_SHIFT	28
1438 #define DYNPTR_SIZE_MASK	0xFFFFFF
1439 #define DYNPTR_RDONLY_BIT	BIT(31)
1440 
__bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern * ptr)1441 static bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1442 {
1443 	return ptr->size & DYNPTR_RDONLY_BIT;
1444 }
1445 
bpf_dynptr_set_rdonly(struct bpf_dynptr_kern * ptr)1446 void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1447 {
1448 	ptr->size |= DYNPTR_RDONLY_BIT;
1449 }
1450 
bpf_dynptr_set_type(struct bpf_dynptr_kern * ptr,enum bpf_dynptr_type type)1451 static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1452 {
1453 	ptr->size |= type << DYNPTR_TYPE_SHIFT;
1454 }
1455 
bpf_dynptr_get_type(const struct bpf_dynptr_kern * ptr)1456 static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1457 {
1458 	return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1459 }
1460 
__bpf_dynptr_size(const struct bpf_dynptr_kern * ptr)1461 u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1462 {
1463 	return ptr->size & DYNPTR_SIZE_MASK;
1464 }
1465 
bpf_dynptr_set_size(struct bpf_dynptr_kern * ptr,u32 new_size)1466 static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1467 {
1468 	u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1469 
1470 	ptr->size = new_size | metadata;
1471 }
1472 
bpf_dynptr_check_size(u32 size)1473 int bpf_dynptr_check_size(u32 size)
1474 {
1475 	return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1476 }
1477 
bpf_dynptr_init(struct bpf_dynptr_kern * ptr,void * data,enum bpf_dynptr_type type,u32 offset,u32 size)1478 void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1479 		     enum bpf_dynptr_type type, u32 offset, u32 size)
1480 {
1481 	ptr->data = data;
1482 	ptr->offset = offset;
1483 	ptr->size = size;
1484 	bpf_dynptr_set_type(ptr, type);
1485 }
1486 
bpf_dynptr_set_null(struct bpf_dynptr_kern * ptr)1487 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1488 {
1489 	memset(ptr, 0, sizeof(*ptr));
1490 }
1491 
bpf_dynptr_check_off_len(const struct bpf_dynptr_kern * ptr,u32 offset,u32 len)1492 static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1493 {
1494 	u32 size = __bpf_dynptr_size(ptr);
1495 
1496 	if (len > size || offset > size - len)
1497 		return -E2BIG;
1498 
1499 	return 0;
1500 }
1501 
BPF_CALL_4(bpf_dynptr_from_mem,void *,data,u32,size,u64,flags,struct bpf_dynptr_kern *,ptr)1502 BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1503 {
1504 	int err;
1505 
1506 	BTF_TYPE_EMIT(struct bpf_dynptr);
1507 
1508 	err = bpf_dynptr_check_size(size);
1509 	if (err)
1510 		goto error;
1511 
1512 	/* flags is currently unsupported */
1513 	if (flags) {
1514 		err = -EINVAL;
1515 		goto error;
1516 	}
1517 
1518 	bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1519 
1520 	return 0;
1521 
1522 error:
1523 	bpf_dynptr_set_null(ptr);
1524 	return err;
1525 }
1526 
1527 static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1528 	.func		= bpf_dynptr_from_mem,
1529 	.gpl_only	= false,
1530 	.ret_type	= RET_INTEGER,
1531 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1532 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1533 	.arg3_type	= ARG_ANYTHING,
1534 	.arg4_type	= ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT,
1535 };
1536 
BPF_CALL_5(bpf_dynptr_read,void *,dst,u32,len,const struct bpf_dynptr_kern *,src,u32,offset,u64,flags)1537 BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1538 	   u32, offset, u64, flags)
1539 {
1540 	enum bpf_dynptr_type type;
1541 	int err;
1542 
1543 	if (!src->data || flags)
1544 		return -EINVAL;
1545 
1546 	err = bpf_dynptr_check_off_len(src, offset, len);
1547 	if (err)
1548 		return err;
1549 
1550 	type = bpf_dynptr_get_type(src);
1551 
1552 	switch (type) {
1553 	case BPF_DYNPTR_TYPE_LOCAL:
1554 	case BPF_DYNPTR_TYPE_RINGBUF:
1555 		/* Source and destination may possibly overlap, hence use memmove to
1556 		 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1557 		 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1558 		 */
1559 		memmove(dst, src->data + src->offset + offset, len);
1560 		return 0;
1561 	case BPF_DYNPTR_TYPE_SKB:
1562 		return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len);
1563 	case BPF_DYNPTR_TYPE_XDP:
1564 		return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len);
1565 	default:
1566 		WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1567 		return -EFAULT;
1568 	}
1569 }
1570 
1571 static const struct bpf_func_proto bpf_dynptr_read_proto = {
1572 	.func		= bpf_dynptr_read,
1573 	.gpl_only	= false,
1574 	.ret_type	= RET_INTEGER,
1575 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1576 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1577 	.arg3_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1578 	.arg4_type	= ARG_ANYTHING,
1579 	.arg5_type	= ARG_ANYTHING,
1580 };
1581 
BPF_CALL_5(bpf_dynptr_write,const struct bpf_dynptr_kern *,dst,u32,offset,void *,src,u32,len,u64,flags)1582 BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1583 	   u32, len, u64, flags)
1584 {
1585 	enum bpf_dynptr_type type;
1586 	int err;
1587 
1588 	if (!dst->data || __bpf_dynptr_is_rdonly(dst))
1589 		return -EINVAL;
1590 
1591 	err = bpf_dynptr_check_off_len(dst, offset, len);
1592 	if (err)
1593 		return err;
1594 
1595 	type = bpf_dynptr_get_type(dst);
1596 
1597 	switch (type) {
1598 	case BPF_DYNPTR_TYPE_LOCAL:
1599 	case BPF_DYNPTR_TYPE_RINGBUF:
1600 		if (flags)
1601 			return -EINVAL;
1602 		/* Source and destination may possibly overlap, hence use memmove to
1603 		 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1604 		 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1605 		 */
1606 		memmove(dst->data + dst->offset + offset, src, len);
1607 		return 0;
1608 	case BPF_DYNPTR_TYPE_SKB:
1609 		return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len,
1610 					     flags);
1611 	case BPF_DYNPTR_TYPE_XDP:
1612 		if (flags)
1613 			return -EINVAL;
1614 		return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len);
1615 	default:
1616 		WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1617 		return -EFAULT;
1618 	}
1619 }
1620 
1621 static const struct bpf_func_proto bpf_dynptr_write_proto = {
1622 	.func		= bpf_dynptr_write,
1623 	.gpl_only	= false,
1624 	.ret_type	= RET_INTEGER,
1625 	.arg1_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1626 	.arg2_type	= ARG_ANYTHING,
1627 	.arg3_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1628 	.arg4_type	= ARG_CONST_SIZE_OR_ZERO,
1629 	.arg5_type	= ARG_ANYTHING,
1630 };
1631 
BPF_CALL_3(bpf_dynptr_data,const struct bpf_dynptr_kern *,ptr,u32,offset,u32,len)1632 BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1633 {
1634 	enum bpf_dynptr_type type;
1635 	int err;
1636 
1637 	if (!ptr->data)
1638 		return 0;
1639 
1640 	err = bpf_dynptr_check_off_len(ptr, offset, len);
1641 	if (err)
1642 		return 0;
1643 
1644 	if (__bpf_dynptr_is_rdonly(ptr))
1645 		return 0;
1646 
1647 	type = bpf_dynptr_get_type(ptr);
1648 
1649 	switch (type) {
1650 	case BPF_DYNPTR_TYPE_LOCAL:
1651 	case BPF_DYNPTR_TYPE_RINGBUF:
1652 		return (unsigned long)(ptr->data + ptr->offset + offset);
1653 	case BPF_DYNPTR_TYPE_SKB:
1654 	case BPF_DYNPTR_TYPE_XDP:
1655 		/* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1656 		return 0;
1657 	default:
1658 		WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1659 		return 0;
1660 	}
1661 }
1662 
1663 static const struct bpf_func_proto bpf_dynptr_data_proto = {
1664 	.func		= bpf_dynptr_data,
1665 	.gpl_only	= false,
1666 	.ret_type	= RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1667 	.arg1_type	= ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1668 	.arg2_type	= ARG_ANYTHING,
1669 	.arg3_type	= ARG_CONST_ALLOC_SIZE_OR_ZERO,
1670 };
1671 
1672 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1673 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1674 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1675 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1676 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1677 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1678 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1679 
1680 const struct bpf_func_proto *
bpf_base_func_proto(enum bpf_func_id func_id)1681 bpf_base_func_proto(enum bpf_func_id func_id)
1682 {
1683 	switch (func_id) {
1684 	case BPF_FUNC_map_lookup_elem:
1685 		return &bpf_map_lookup_elem_proto;
1686 	case BPF_FUNC_map_update_elem:
1687 		return &bpf_map_update_elem_proto;
1688 	case BPF_FUNC_map_delete_elem:
1689 		return &bpf_map_delete_elem_proto;
1690 	case BPF_FUNC_map_push_elem:
1691 		return &bpf_map_push_elem_proto;
1692 	case BPF_FUNC_map_pop_elem:
1693 		return &bpf_map_pop_elem_proto;
1694 	case BPF_FUNC_map_peek_elem:
1695 		return &bpf_map_peek_elem_proto;
1696 	case BPF_FUNC_map_lookup_percpu_elem:
1697 		return &bpf_map_lookup_percpu_elem_proto;
1698 	case BPF_FUNC_get_prandom_u32:
1699 		return &bpf_get_prandom_u32_proto;
1700 	case BPF_FUNC_get_smp_processor_id:
1701 		return &bpf_get_raw_smp_processor_id_proto;
1702 	case BPF_FUNC_get_numa_node_id:
1703 		return &bpf_get_numa_node_id_proto;
1704 	case BPF_FUNC_tail_call:
1705 		return &bpf_tail_call_proto;
1706 	case BPF_FUNC_ktime_get_ns:
1707 		return &bpf_ktime_get_ns_proto;
1708 	case BPF_FUNC_ktime_get_boot_ns:
1709 		return &bpf_ktime_get_boot_ns_proto;
1710 	case BPF_FUNC_ktime_get_tai_ns:
1711 		return &bpf_ktime_get_tai_ns_proto;
1712 	case BPF_FUNC_ringbuf_output:
1713 		return &bpf_ringbuf_output_proto;
1714 	case BPF_FUNC_ringbuf_reserve:
1715 		return &bpf_ringbuf_reserve_proto;
1716 	case BPF_FUNC_ringbuf_submit:
1717 		return &bpf_ringbuf_submit_proto;
1718 	case BPF_FUNC_ringbuf_discard:
1719 		return &bpf_ringbuf_discard_proto;
1720 	case BPF_FUNC_ringbuf_query:
1721 		return &bpf_ringbuf_query_proto;
1722 	case BPF_FUNC_strncmp:
1723 		return &bpf_strncmp_proto;
1724 	case BPF_FUNC_strtol:
1725 		return &bpf_strtol_proto;
1726 	case BPF_FUNC_strtoul:
1727 		return &bpf_strtoul_proto;
1728 	default:
1729 		break;
1730 	}
1731 
1732 	if (!bpf_capable())
1733 		return NULL;
1734 
1735 	switch (func_id) {
1736 	case BPF_FUNC_spin_lock:
1737 		return &bpf_spin_lock_proto;
1738 	case BPF_FUNC_spin_unlock:
1739 		return &bpf_spin_unlock_proto;
1740 	case BPF_FUNC_jiffies64:
1741 		return &bpf_jiffies64_proto;
1742 	case BPF_FUNC_per_cpu_ptr:
1743 		return &bpf_per_cpu_ptr_proto;
1744 	case BPF_FUNC_this_cpu_ptr:
1745 		return &bpf_this_cpu_ptr_proto;
1746 	case BPF_FUNC_timer_init:
1747 		return &bpf_timer_init_proto;
1748 	case BPF_FUNC_timer_set_callback:
1749 		return &bpf_timer_set_callback_proto;
1750 	case BPF_FUNC_timer_start:
1751 		return &bpf_timer_start_proto;
1752 	case BPF_FUNC_timer_cancel:
1753 		return &bpf_timer_cancel_proto;
1754 	case BPF_FUNC_kptr_xchg:
1755 		return &bpf_kptr_xchg_proto;
1756 	case BPF_FUNC_for_each_map_elem:
1757 		return &bpf_for_each_map_elem_proto;
1758 	case BPF_FUNC_loop:
1759 		return &bpf_loop_proto;
1760 	case BPF_FUNC_user_ringbuf_drain:
1761 		return &bpf_user_ringbuf_drain_proto;
1762 	case BPF_FUNC_ringbuf_reserve_dynptr:
1763 		return &bpf_ringbuf_reserve_dynptr_proto;
1764 	case BPF_FUNC_ringbuf_submit_dynptr:
1765 		return &bpf_ringbuf_submit_dynptr_proto;
1766 	case BPF_FUNC_ringbuf_discard_dynptr:
1767 		return &bpf_ringbuf_discard_dynptr_proto;
1768 	case BPF_FUNC_dynptr_from_mem:
1769 		return &bpf_dynptr_from_mem_proto;
1770 	case BPF_FUNC_dynptr_read:
1771 		return &bpf_dynptr_read_proto;
1772 	case BPF_FUNC_dynptr_write:
1773 		return &bpf_dynptr_write_proto;
1774 	case BPF_FUNC_dynptr_data:
1775 		return &bpf_dynptr_data_proto;
1776 #ifdef CONFIG_CGROUPS
1777 	case BPF_FUNC_cgrp_storage_get:
1778 		return &bpf_cgrp_storage_get_proto;
1779 	case BPF_FUNC_cgrp_storage_delete:
1780 		return &bpf_cgrp_storage_delete_proto;
1781 	case BPF_FUNC_get_current_cgroup_id:
1782 		return &bpf_get_current_cgroup_id_proto;
1783 	case BPF_FUNC_get_current_ancestor_cgroup_id:
1784 		return &bpf_get_current_ancestor_cgroup_id_proto;
1785 #endif
1786 	default:
1787 		break;
1788 	}
1789 
1790 	if (!perfmon_capable())
1791 		return NULL;
1792 
1793 	switch (func_id) {
1794 	case BPF_FUNC_trace_printk:
1795 		return bpf_get_trace_printk_proto();
1796 	case BPF_FUNC_get_current_task:
1797 		return &bpf_get_current_task_proto;
1798 	case BPF_FUNC_get_current_task_btf:
1799 		return &bpf_get_current_task_btf_proto;
1800 	case BPF_FUNC_probe_read_user:
1801 		return &bpf_probe_read_user_proto;
1802 	case BPF_FUNC_probe_read_kernel:
1803 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1804 		       NULL : &bpf_probe_read_kernel_proto;
1805 	case BPF_FUNC_probe_read_user_str:
1806 		return &bpf_probe_read_user_str_proto;
1807 	case BPF_FUNC_probe_read_kernel_str:
1808 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1809 		       NULL : &bpf_probe_read_kernel_str_proto;
1810 	case BPF_FUNC_snprintf_btf:
1811 		return &bpf_snprintf_btf_proto;
1812 	case BPF_FUNC_snprintf:
1813 		return &bpf_snprintf_proto;
1814 	case BPF_FUNC_task_pt_regs:
1815 		return &bpf_task_pt_regs_proto;
1816 	case BPF_FUNC_trace_vprintk:
1817 		return bpf_get_trace_vprintk_proto();
1818 	default:
1819 		return NULL;
1820 	}
1821 }
1822 
1823 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec);
1824 
bpf_list_head_free(const struct btf_field * field,void * list_head,struct bpf_spin_lock * spin_lock)1825 void bpf_list_head_free(const struct btf_field *field, void *list_head,
1826 			struct bpf_spin_lock *spin_lock)
1827 {
1828 	struct list_head *head = list_head, *orig_head = list_head;
1829 
1830 	BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
1831 	BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
1832 
1833 	/* Do the actual list draining outside the lock to not hold the lock for
1834 	 * too long, and also prevent deadlocks if tracing programs end up
1835 	 * executing on entry/exit of functions called inside the critical
1836 	 * section, and end up doing map ops that call bpf_list_head_free for
1837 	 * the same map value again.
1838 	 */
1839 	__bpf_spin_lock_irqsave(spin_lock);
1840 	if (!head->next || list_empty(head))
1841 		goto unlock;
1842 	head = head->next;
1843 unlock:
1844 	INIT_LIST_HEAD(orig_head);
1845 	__bpf_spin_unlock_irqrestore(spin_lock);
1846 
1847 	while (head != orig_head) {
1848 		void *obj = head;
1849 
1850 		obj -= field->graph_root.node_offset;
1851 		head = head->next;
1852 		/* The contained type can also have resources, including a
1853 		 * bpf_list_head which needs to be freed.
1854 		 */
1855 		migrate_disable();
1856 		__bpf_obj_drop_impl(obj, field->graph_root.value_rec);
1857 		migrate_enable();
1858 	}
1859 }
1860 
1861 /* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
1862  * 'rb_node *', so field name of rb_node within containing struct is not
1863  * needed.
1864  *
1865  * Since bpf_rb_tree's node type has a corresponding struct btf_field with
1866  * graph_root.node_offset, it's not necessary to know field name
1867  * or type of node struct
1868  */
1869 #define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
1870 	for (pos = rb_first_postorder(root); \
1871 	    pos && ({ n = rb_next_postorder(pos); 1; }); \
1872 	    pos = n)
1873 
bpf_rb_root_free(const struct btf_field * field,void * rb_root,struct bpf_spin_lock * spin_lock)1874 void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
1875 		      struct bpf_spin_lock *spin_lock)
1876 {
1877 	struct rb_root_cached orig_root, *root = rb_root;
1878 	struct rb_node *pos, *n;
1879 	void *obj;
1880 
1881 	BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
1882 	BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
1883 
1884 	__bpf_spin_lock_irqsave(spin_lock);
1885 	orig_root = *root;
1886 	*root = RB_ROOT_CACHED;
1887 	__bpf_spin_unlock_irqrestore(spin_lock);
1888 
1889 	bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
1890 		obj = pos;
1891 		obj -= field->graph_root.node_offset;
1892 
1893 
1894 		migrate_disable();
1895 		__bpf_obj_drop_impl(obj, field->graph_root.value_rec);
1896 		migrate_enable();
1897 	}
1898 }
1899 
1900 __diag_push();
1901 __diag_ignore_all("-Wmissing-prototypes",
1902 		  "Global functions as their definitions will be in vmlinux BTF");
1903 
bpf_obj_new_impl(u64 local_type_id__k,void * meta__ign)1904 __bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1905 {
1906 	struct btf_struct_meta *meta = meta__ign;
1907 	u64 size = local_type_id__k;
1908 	void *p;
1909 
1910 	p = bpf_mem_alloc(&bpf_global_ma, size);
1911 	if (!p)
1912 		return NULL;
1913 	if (meta)
1914 		bpf_obj_init(meta->record, p);
1915 	return p;
1916 }
1917 
1918 /* Must be called under migrate_disable(), as required by bpf_mem_free */
__bpf_obj_drop_impl(void * p,const struct btf_record * rec)1919 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec)
1920 {
1921 	if (rec && rec->refcount_off >= 0 &&
1922 	    !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
1923 		/* Object is refcounted and refcount_dec didn't result in 0
1924 		 * refcount. Return without freeing the object
1925 		 */
1926 		return;
1927 	}
1928 
1929 	if (rec)
1930 		bpf_obj_free_fields(rec, p);
1931 
1932 	if (rec && rec->refcount_off >= 0)
1933 		bpf_mem_free_rcu(&bpf_global_ma, p);
1934 	else
1935 		bpf_mem_free(&bpf_global_ma, p);
1936 }
1937 
bpf_obj_drop_impl(void * p__alloc,void * meta__ign)1938 __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
1939 {
1940 	struct btf_struct_meta *meta = meta__ign;
1941 	void *p = p__alloc;
1942 
1943 	__bpf_obj_drop_impl(p, meta ? meta->record : NULL);
1944 }
1945 
bpf_refcount_acquire_impl(void * p__refcounted_kptr,void * meta__ign)1946 __bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
1947 {
1948 	struct btf_struct_meta *meta = meta__ign;
1949 	struct bpf_refcount *ref;
1950 
1951 	/* Could just cast directly to refcount_t *, but need some code using
1952 	 * bpf_refcount type so that it is emitted in vmlinux BTF
1953 	 */
1954 	ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
1955 	if (!refcount_inc_not_zero((refcount_t *)ref))
1956 		return NULL;
1957 
1958 	/* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
1959 	 * in verifier.c
1960 	 */
1961 	return (void *)p__refcounted_kptr;
1962 }
1963 
__bpf_list_add(struct bpf_list_node_kern * node,struct bpf_list_head * head,bool tail,struct btf_record * rec,u64 off)1964 static int __bpf_list_add(struct bpf_list_node_kern *node,
1965 			  struct bpf_list_head *head,
1966 			  bool tail, struct btf_record *rec, u64 off)
1967 {
1968 	struct list_head *n = &node->list_head, *h = (void *)head;
1969 
1970 	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
1971 	 * called on its fields, so init here
1972 	 */
1973 	if (unlikely(!h->next))
1974 		INIT_LIST_HEAD(h);
1975 
1976 	/* node->owner != NULL implies !list_empty(n), no need to separately
1977 	 * check the latter
1978 	 */
1979 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
1980 		/* Only called from BPF prog, no need to migrate_disable */
1981 		__bpf_obj_drop_impl((void *)n - off, rec);
1982 		return -EINVAL;
1983 	}
1984 
1985 	tail ? list_add_tail(n, h) : list_add(n, h);
1986 	WRITE_ONCE(node->owner, head);
1987 
1988 	return 0;
1989 }
1990 
bpf_list_push_front_impl(struct bpf_list_head * head,struct bpf_list_node * node,void * meta__ign,u64 off)1991 __bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
1992 					 struct bpf_list_node *node,
1993 					 void *meta__ign, u64 off)
1994 {
1995 	struct bpf_list_node_kern *n = (void *)node;
1996 	struct btf_struct_meta *meta = meta__ign;
1997 
1998 	return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off);
1999 }
2000 
bpf_list_push_back_impl(struct bpf_list_head * head,struct bpf_list_node * node,void * meta__ign,u64 off)2001 __bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2002 					struct bpf_list_node *node,
2003 					void *meta__ign, u64 off)
2004 {
2005 	struct bpf_list_node_kern *n = (void *)node;
2006 	struct btf_struct_meta *meta = meta__ign;
2007 
2008 	return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off);
2009 }
2010 
__bpf_list_del(struct bpf_list_head * head,bool tail)2011 static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2012 {
2013 	struct list_head *n, *h = (void *)head;
2014 	struct bpf_list_node_kern *node;
2015 
2016 	/* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2017 	 * called on its fields, so init here
2018 	 */
2019 	if (unlikely(!h->next))
2020 		INIT_LIST_HEAD(h);
2021 	if (list_empty(h))
2022 		return NULL;
2023 
2024 	n = tail ? h->prev : h->next;
2025 	node = container_of(n, struct bpf_list_node_kern, list_head);
2026 	if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2027 		return NULL;
2028 
2029 	list_del_init(n);
2030 	WRITE_ONCE(node->owner, NULL);
2031 	return (struct bpf_list_node *)n;
2032 }
2033 
bpf_list_pop_front(struct bpf_list_head * head)2034 __bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2035 {
2036 	return __bpf_list_del(head, false);
2037 }
2038 
bpf_list_pop_back(struct bpf_list_head * head)2039 __bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2040 {
2041 	return __bpf_list_del(head, true);
2042 }
2043 
bpf_rbtree_remove(struct bpf_rb_root * root,struct bpf_rb_node * node)2044 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2045 						  struct bpf_rb_node *node)
2046 {
2047 	struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2048 	struct rb_root_cached *r = (struct rb_root_cached *)root;
2049 	struct rb_node *n = &node_internal->rb_node;
2050 
2051 	/* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2052 	 * n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2053 	 */
2054 	if (READ_ONCE(node_internal->owner) != root)
2055 		return NULL;
2056 
2057 	rb_erase_cached(n, r);
2058 	RB_CLEAR_NODE(n);
2059 	WRITE_ONCE(node_internal->owner, NULL);
2060 	return (struct bpf_rb_node *)n;
2061 }
2062 
2063 /* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2064  * program
2065  */
__bpf_rbtree_add(struct bpf_rb_root * root,struct bpf_rb_node_kern * node,void * less,struct btf_record * rec,u64 off)2066 static int __bpf_rbtree_add(struct bpf_rb_root *root,
2067 			    struct bpf_rb_node_kern *node,
2068 			    void *less, struct btf_record *rec, u64 off)
2069 {
2070 	struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2071 	struct rb_node *parent = NULL, *n = &node->rb_node;
2072 	bpf_callback_t cb = (bpf_callback_t)less;
2073 	bool leftmost = true;
2074 
2075 	/* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2076 	 * check the latter
2077 	 */
2078 	if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2079 		/* Only called from BPF prog, no need to migrate_disable */
2080 		__bpf_obj_drop_impl((void *)n - off, rec);
2081 		return -EINVAL;
2082 	}
2083 
2084 	while (*link) {
2085 		parent = *link;
2086 		if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2087 			link = &parent->rb_left;
2088 		} else {
2089 			link = &parent->rb_right;
2090 			leftmost = false;
2091 		}
2092 	}
2093 
2094 	rb_link_node(n, parent, link);
2095 	rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost);
2096 	WRITE_ONCE(node->owner, root);
2097 	return 0;
2098 }
2099 
bpf_rbtree_add_impl(struct bpf_rb_root * root,struct bpf_rb_node * node,bool (less)(struct bpf_rb_node * a,const struct bpf_rb_node * b),void * meta__ign,u64 off)2100 __bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2101 				    bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2102 				    void *meta__ign, u64 off)
2103 {
2104 	struct btf_struct_meta *meta = meta__ign;
2105 	struct bpf_rb_node_kern *n = (void *)node;
2106 
2107 	return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off);
2108 }
2109 
bpf_rbtree_first(struct bpf_rb_root * root)2110 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2111 {
2112 	struct rb_root_cached *r = (struct rb_root_cached *)root;
2113 
2114 	return (struct bpf_rb_node *)rb_first_cached(r);
2115 }
2116 
2117 /**
2118  * bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2119  * kfunc which is not stored in a map as a kptr, must be released by calling
2120  * bpf_task_release().
2121  * @p: The task on which a reference is being acquired.
2122  */
bpf_task_acquire(struct task_struct * p)2123 __bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2124 {
2125 	if (refcount_inc_not_zero(&p->rcu_users))
2126 		return p;
2127 	return NULL;
2128 }
2129 
2130 /**
2131  * bpf_task_release - Release the reference acquired on a task.
2132  * @p: The task on which a reference is being released.
2133  */
bpf_task_release(struct task_struct * p)2134 __bpf_kfunc void bpf_task_release(struct task_struct *p)
2135 {
2136 	put_task_struct_rcu_user(p);
2137 }
2138 
2139 #ifdef CONFIG_CGROUPS
2140 /**
2141  * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2142  * this kfunc which is not stored in a map as a kptr, must be released by
2143  * calling bpf_cgroup_release().
2144  * @cgrp: The cgroup on which a reference is being acquired.
2145  */
bpf_cgroup_acquire(struct cgroup * cgrp)2146 __bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2147 {
2148 	return cgroup_tryget(cgrp) ? cgrp : NULL;
2149 }
2150 
2151 /**
2152  * bpf_cgroup_release - Release the reference acquired on a cgroup.
2153  * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2154  * not be freed until the current grace period has ended, even if its refcount
2155  * drops to 0.
2156  * @cgrp: The cgroup on which a reference is being released.
2157  */
bpf_cgroup_release(struct cgroup * cgrp)2158 __bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2159 {
2160 	cgroup_put(cgrp);
2161 }
2162 
2163 /**
2164  * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2165  * array. A cgroup returned by this kfunc which is not subsequently stored in a
2166  * map, must be released by calling bpf_cgroup_release().
2167  * @cgrp: The cgroup for which we're performing a lookup.
2168  * @level: The level of ancestor to look up.
2169  */
bpf_cgroup_ancestor(struct cgroup * cgrp,int level)2170 __bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2171 {
2172 	struct cgroup *ancestor;
2173 
2174 	if (level > cgrp->level || level < 0)
2175 		return NULL;
2176 
2177 	/* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2178 	ancestor = cgrp->ancestors[level];
2179 	if (!cgroup_tryget(ancestor))
2180 		return NULL;
2181 	return ancestor;
2182 }
2183 
2184 /**
2185  * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2186  * kfunc which is not subsequently stored in a map, must be released by calling
2187  * bpf_cgroup_release().
2188  * @cgid: cgroup id.
2189  */
bpf_cgroup_from_id(u64 cgid)2190 __bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2191 {
2192 	struct cgroup *cgrp;
2193 
2194 	cgrp = cgroup_get_from_id(cgid);
2195 	if (IS_ERR(cgrp))
2196 		return NULL;
2197 	return cgrp;
2198 }
2199 
2200 /**
2201  * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2202  * task's membership of cgroup ancestry.
2203  * @task: the task to be tested
2204  * @ancestor: possible ancestor of @task's cgroup
2205  *
2206  * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2207  * It follows all the same rules as cgroup_is_descendant, and only applies
2208  * to the default hierarchy.
2209  */
bpf_task_under_cgroup(struct task_struct * task,struct cgroup * ancestor)2210 __bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2211 				       struct cgroup *ancestor)
2212 {
2213 	long ret;
2214 
2215 	rcu_read_lock();
2216 	ret = task_under_cgroup_hierarchy(task, ancestor);
2217 	rcu_read_unlock();
2218 	return ret;
2219 }
2220 #endif /* CONFIG_CGROUPS */
2221 
2222 /**
2223  * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2224  * in the root pid namespace idr. If a task is returned, it must either be
2225  * stored in a map, or released with bpf_task_release().
2226  * @pid: The pid of the task being looked up.
2227  */
bpf_task_from_pid(s32 pid)2228 __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2229 {
2230 	struct task_struct *p;
2231 
2232 	rcu_read_lock();
2233 	p = find_task_by_pid_ns(pid, &init_pid_ns);
2234 	if (p)
2235 		p = bpf_task_acquire(p);
2236 	rcu_read_unlock();
2237 
2238 	return p;
2239 }
2240 
2241 /**
2242  * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2243  * @ptr: The dynptr whose data slice to retrieve
2244  * @offset: Offset into the dynptr
2245  * @buffer__opt: User-provided buffer to copy contents into.  May be NULL
2246  * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2247  *               length of the requested slice. This must be a constant.
2248  *
2249  * For non-skb and non-xdp type dynptrs, there is no difference between
2250  * bpf_dynptr_slice and bpf_dynptr_data.
2251  *
2252  *  If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2253  *
2254  * If the intention is to write to the data slice, please use
2255  * bpf_dynptr_slice_rdwr.
2256  *
2257  * The user must check that the returned pointer is not null before using it.
2258  *
2259  * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2260  * does not change the underlying packet data pointers, so a call to
2261  * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2262  * the bpf program.
2263  *
2264  * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2265  * data slice (can be either direct pointer to the data or a pointer to the user
2266  * provided buffer, with its contents containing the data, if unable to obtain
2267  * direct pointer)
2268  */
bpf_dynptr_slice(const struct bpf_dynptr_kern * ptr,u32 offset,void * buffer__opt,u32 buffer__szk)2269 __bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset,
2270 				   void *buffer__opt, u32 buffer__szk)
2271 {
2272 	enum bpf_dynptr_type type;
2273 	u32 len = buffer__szk;
2274 	int err;
2275 
2276 	if (!ptr->data)
2277 		return NULL;
2278 
2279 	err = bpf_dynptr_check_off_len(ptr, offset, len);
2280 	if (err)
2281 		return NULL;
2282 
2283 	type = bpf_dynptr_get_type(ptr);
2284 
2285 	switch (type) {
2286 	case BPF_DYNPTR_TYPE_LOCAL:
2287 	case BPF_DYNPTR_TYPE_RINGBUF:
2288 		return ptr->data + ptr->offset + offset;
2289 	case BPF_DYNPTR_TYPE_SKB:
2290 		if (buffer__opt)
2291 			return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt);
2292 		else
2293 			return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len);
2294 	case BPF_DYNPTR_TYPE_XDP:
2295 	{
2296 		void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len);
2297 		if (!IS_ERR_OR_NULL(xdp_ptr))
2298 			return xdp_ptr;
2299 
2300 		if (!buffer__opt)
2301 			return NULL;
2302 		bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false);
2303 		return buffer__opt;
2304 	}
2305 	default:
2306 		WARN_ONCE(true, "unknown dynptr type %d\n", type);
2307 		return NULL;
2308 	}
2309 }
2310 
2311 /**
2312  * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2313  * @ptr: The dynptr whose data slice to retrieve
2314  * @offset: Offset into the dynptr
2315  * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2316  * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2317  *               length of the requested slice. This must be a constant.
2318  *
2319  * For non-skb and non-xdp type dynptrs, there is no difference between
2320  * bpf_dynptr_slice and bpf_dynptr_data.
2321  *
2322  * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2323  *
2324  * The returned pointer is writable and may point to either directly the dynptr
2325  * data at the requested offset or to the buffer if unable to obtain a direct
2326  * data pointer to (example: the requested slice is to the paged area of an skb
2327  * packet). In the case where the returned pointer is to the buffer, the user
2328  * is responsible for persisting writes through calling bpf_dynptr_write(). This
2329  * usually looks something like this pattern:
2330  *
2331  * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2332  * if (!eth)
2333  *	return TC_ACT_SHOT;
2334  *
2335  * // mutate eth header //
2336  *
2337  * if (eth == buffer)
2338  *	bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2339  *
2340  * Please note that, as in the example above, the user must check that the
2341  * returned pointer is not null before using it.
2342  *
2343  * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2344  * does not change the underlying packet data pointers, so a call to
2345  * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2346  * the bpf program.
2347  *
2348  * Return: NULL if the call failed (eg invalid dynptr), pointer to a
2349  * data slice (can be either direct pointer to the data or a pointer to the user
2350  * provided buffer, with its contents containing the data, if unable to obtain
2351  * direct pointer)
2352  */
bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern * ptr,u32 offset,void * buffer__opt,u32 buffer__szk)2353 __bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset,
2354 					void *buffer__opt, u32 buffer__szk)
2355 {
2356 	if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2357 		return NULL;
2358 
2359 	/* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2360 	 *
2361 	 * For skb-type dynptrs, it is safe to write into the returned pointer
2362 	 * if the bpf program allows skb data writes. There are two possiblities
2363 	 * that may occur when calling bpf_dynptr_slice_rdwr:
2364 	 *
2365 	 * 1) The requested slice is in the head of the skb. In this case, the
2366 	 * returned pointer is directly to skb data, and if the skb is cloned, the
2367 	 * verifier will have uncloned it (see bpf_unclone_prologue()) already.
2368 	 * The pointer can be directly written into.
2369 	 *
2370 	 * 2) Some portion of the requested slice is in the paged buffer area.
2371 	 * In this case, the requested data will be copied out into the buffer
2372 	 * and the returned pointer will be a pointer to the buffer. The skb
2373 	 * will not be pulled. To persist the write, the user will need to call
2374 	 * bpf_dynptr_write(), which will pull the skb and commit the write.
2375 	 *
2376 	 * Similarly for xdp programs, if the requested slice is not across xdp
2377 	 * fragments, then a direct pointer will be returned, otherwise the data
2378 	 * will be copied out into the buffer and the user will need to call
2379 	 * bpf_dynptr_write() to commit changes.
2380 	 */
2381 	return bpf_dynptr_slice(ptr, offset, buffer__opt, buffer__szk);
2382 }
2383 
bpf_dynptr_adjust(struct bpf_dynptr_kern * ptr,u32 start,u32 end)2384 __bpf_kfunc int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end)
2385 {
2386 	u32 size;
2387 
2388 	if (!ptr->data || start > end)
2389 		return -EINVAL;
2390 
2391 	size = __bpf_dynptr_size(ptr);
2392 
2393 	if (start > size || end > size)
2394 		return -ERANGE;
2395 
2396 	ptr->offset += start;
2397 	bpf_dynptr_set_size(ptr, end - start);
2398 
2399 	return 0;
2400 }
2401 
bpf_dynptr_is_null(struct bpf_dynptr_kern * ptr)2402 __bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr)
2403 {
2404 	return !ptr->data;
2405 }
2406 
bpf_dynptr_is_rdonly(struct bpf_dynptr_kern * ptr)2407 __bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr)
2408 {
2409 	if (!ptr->data)
2410 		return false;
2411 
2412 	return __bpf_dynptr_is_rdonly(ptr);
2413 }
2414 
bpf_dynptr_size(const struct bpf_dynptr_kern * ptr)2415 __bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
2416 {
2417 	if (!ptr->data)
2418 		return -EINVAL;
2419 
2420 	return __bpf_dynptr_size(ptr);
2421 }
2422 
bpf_dynptr_clone(struct bpf_dynptr_kern * ptr,struct bpf_dynptr_kern * clone__uninit)2423 __bpf_kfunc int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr,
2424 				 struct bpf_dynptr_kern *clone__uninit)
2425 {
2426 	if (!ptr->data) {
2427 		bpf_dynptr_set_null(clone__uninit);
2428 		return -EINVAL;
2429 	}
2430 
2431 	*clone__uninit = *ptr;
2432 
2433 	return 0;
2434 }
2435 
bpf_cast_to_kern_ctx(void * obj)2436 __bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2437 {
2438 	return obj;
2439 }
2440 
bpf_rdonly_cast(void * obj__ign,u32 btf_id__k)2441 __bpf_kfunc void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k)
2442 {
2443 	return obj__ign;
2444 }
2445 
bpf_rcu_read_lock(void)2446 __bpf_kfunc void bpf_rcu_read_lock(void)
2447 {
2448 	rcu_read_lock();
2449 }
2450 
bpf_rcu_read_unlock(void)2451 __bpf_kfunc void bpf_rcu_read_unlock(void)
2452 {
2453 	rcu_read_unlock();
2454 }
2455 
2456 __diag_pop();
2457 
2458 BTF_SET8_START(generic_btf_ids)
2459 #ifdef CONFIG_KEXEC_CORE
2460 BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
2461 #endif
2462 BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2463 BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
2464 BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL)
2465 BTF_ID_FLAGS(func, bpf_list_push_front_impl)
2466 BTF_ID_FLAGS(func, bpf_list_push_back_impl)
2467 BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
2468 BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
2469 BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2470 BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
2471 BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
2472 BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
2473 BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
2474 
2475 #ifdef CONFIG_CGROUPS
2476 BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2477 BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
2478 BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2479 BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
2480 BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
2481 #endif
2482 BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
2483 BTF_SET8_END(generic_btf_ids)
2484 
2485 static const struct btf_kfunc_id_set generic_kfunc_set = {
2486 	.owner = THIS_MODULE,
2487 	.set   = &generic_btf_ids,
2488 };
2489 
2490 
2491 BTF_ID_LIST(generic_dtor_ids)
2492 BTF_ID(struct, task_struct)
2493 BTF_ID(func, bpf_task_release)
2494 #ifdef CONFIG_CGROUPS
2495 BTF_ID(struct, cgroup)
2496 BTF_ID(func, bpf_cgroup_release)
2497 #endif
2498 
2499 BTF_SET8_START(common_btf_ids)
2500 BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx)
2501 BTF_ID_FLAGS(func, bpf_rdonly_cast)
2502 BTF_ID_FLAGS(func, bpf_rcu_read_lock)
2503 BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
2504 BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
2505 BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
2506 BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
2507 BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
2508 BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
2509 BTF_ID_FLAGS(func, bpf_dynptr_adjust)
2510 BTF_ID_FLAGS(func, bpf_dynptr_is_null)
2511 BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
2512 BTF_ID_FLAGS(func, bpf_dynptr_size)
2513 BTF_ID_FLAGS(func, bpf_dynptr_clone)
2514 BTF_SET8_END(common_btf_ids)
2515 
2516 static const struct btf_kfunc_id_set common_kfunc_set = {
2517 	.owner = THIS_MODULE,
2518 	.set   = &common_btf_ids,
2519 };
2520 
kfunc_init(void)2521 static int __init kfunc_init(void)
2522 {
2523 	int ret;
2524 	const struct btf_id_dtor_kfunc generic_dtors[] = {
2525 		{
2526 			.btf_id       = generic_dtor_ids[0],
2527 			.kfunc_btf_id = generic_dtor_ids[1]
2528 		},
2529 #ifdef CONFIG_CGROUPS
2530 		{
2531 			.btf_id       = generic_dtor_ids[2],
2532 			.kfunc_btf_id = generic_dtor_ids[3]
2533 		},
2534 #endif
2535 	};
2536 
2537 	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set);
2538 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set);
2539 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set);
2540 	ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors,
2541 						  ARRAY_SIZE(generic_dtors),
2542 						  THIS_MODULE);
2543 	return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set);
2544 }
2545 
2546 late_initcall(kfunc_init);
2547