Loading...
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#include <linux/kasan.h>
26
27#include "../../lib/kstrtox.h"
28
29/* If kernel subsystem is allowing eBPF programs to call this function,
30 * inside its own verifier_ops->get_func_proto() callback it should return
31 * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
32 *
33 * Different map implementations will rely on rcu in map methods
34 * lookup/update/delete, therefore eBPF programs must run under rcu lock
35 * if program is allowed to access maps, so check rcu_read_lock_held() or
36 * rcu_read_lock_trace_held() in all three functions.
37 */
38BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
39{
40 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
41 !rcu_read_lock_bh_held());
42 return (unsigned long) map->ops->map_lookup_elem(map, key);
43}
44
45const struct bpf_func_proto bpf_map_lookup_elem_proto = {
46 .func = bpf_map_lookup_elem,
47 .gpl_only = false,
48 .pkt_access = true,
49 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
50 .arg1_type = ARG_CONST_MAP_PTR,
51 .arg2_type = ARG_PTR_TO_MAP_KEY,
52};
53
54BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
55 void *, value, u64, flags)
56{
57 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
58 !rcu_read_lock_bh_held());
59 return map->ops->map_update_elem(map, key, value, flags);
60}
61
62const struct bpf_func_proto bpf_map_update_elem_proto = {
63 .func = bpf_map_update_elem,
64 .gpl_only = false,
65 .pkt_access = true,
66 .ret_type = RET_INTEGER,
67 .arg1_type = ARG_CONST_MAP_PTR,
68 .arg2_type = ARG_PTR_TO_MAP_KEY,
69 .arg3_type = ARG_PTR_TO_MAP_VALUE,
70 .arg4_type = ARG_ANYTHING,
71};
72
73BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
74{
75 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
76 !rcu_read_lock_bh_held());
77 return map->ops->map_delete_elem(map, key);
78}
79
80const struct bpf_func_proto bpf_map_delete_elem_proto = {
81 .func = bpf_map_delete_elem,
82 .gpl_only = false,
83 .pkt_access = true,
84 .ret_type = RET_INTEGER,
85 .arg1_type = ARG_CONST_MAP_PTR,
86 .arg2_type = ARG_PTR_TO_MAP_KEY,
87};
88
89BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
90{
91 return map->ops->map_push_elem(map, value, flags);
92}
93
94const struct bpf_func_proto bpf_map_push_elem_proto = {
95 .func = bpf_map_push_elem,
96 .gpl_only = false,
97 .pkt_access = true,
98 .ret_type = RET_INTEGER,
99 .arg1_type = ARG_CONST_MAP_PTR,
100 .arg2_type = ARG_PTR_TO_MAP_VALUE,
101 .arg3_type = ARG_ANYTHING,
102};
103
104BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
105{
106 return map->ops->map_pop_elem(map, value);
107}
108
109const struct bpf_func_proto bpf_map_pop_elem_proto = {
110 .func = bpf_map_pop_elem,
111 .gpl_only = false,
112 .ret_type = RET_INTEGER,
113 .arg1_type = ARG_CONST_MAP_PTR,
114 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
115};
116
117BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
118{
119 return map->ops->map_peek_elem(map, value);
120}
121
122const struct bpf_func_proto bpf_map_peek_elem_proto = {
123 .func = bpf_map_peek_elem,
124 .gpl_only = false,
125 .ret_type = RET_INTEGER,
126 .arg1_type = ARG_CONST_MAP_PTR,
127 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
128};
129
130BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
131{
132 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
133 return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu);
134}
135
136const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
137 .func = bpf_map_lookup_percpu_elem,
138 .gpl_only = false,
139 .pkt_access = true,
140 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
141 .arg1_type = ARG_CONST_MAP_PTR,
142 .arg2_type = ARG_PTR_TO_MAP_KEY,
143 .arg3_type = ARG_ANYTHING,
144};
145
146const struct bpf_func_proto bpf_get_prandom_u32_proto = {
147 .func = bpf_user_rnd_u32,
148 .gpl_only = false,
149 .ret_type = RET_INTEGER,
150};
151
152BPF_CALL_0(bpf_get_smp_processor_id)
153{
154 return smp_processor_id();
155}
156
157const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
158 .func = bpf_get_smp_processor_id,
159 .gpl_only = false,
160 .ret_type = RET_INTEGER,
161};
162
163BPF_CALL_0(bpf_get_numa_node_id)
164{
165 return numa_node_id();
166}
167
168const struct bpf_func_proto bpf_get_numa_node_id_proto = {
169 .func = bpf_get_numa_node_id,
170 .gpl_only = false,
171 .ret_type = RET_INTEGER,
172};
173
174BPF_CALL_0(bpf_ktime_get_ns)
175{
176 /* NMI safe access to clock monotonic */
177 return ktime_get_mono_fast_ns();
178}
179
180const struct bpf_func_proto bpf_ktime_get_ns_proto = {
181 .func = bpf_ktime_get_ns,
182 .gpl_only = false,
183 .ret_type = RET_INTEGER,
184};
185
186BPF_CALL_0(bpf_ktime_get_boot_ns)
187{
188 /* NMI safe access to clock boottime */
189 return ktime_get_boot_fast_ns();
190}
191
192const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
193 .func = bpf_ktime_get_boot_ns,
194 .gpl_only = false,
195 .ret_type = RET_INTEGER,
196};
197
198BPF_CALL_0(bpf_ktime_get_coarse_ns)
199{
200 return ktime_get_coarse_ns();
201}
202
203const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
204 .func = bpf_ktime_get_coarse_ns,
205 .gpl_only = false,
206 .ret_type = RET_INTEGER,
207};
208
209BPF_CALL_0(bpf_ktime_get_tai_ns)
210{
211 /* NMI safe access to clock tai */
212 return ktime_get_tai_fast_ns();
213}
214
215const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
216 .func = bpf_ktime_get_tai_ns,
217 .gpl_only = false,
218 .ret_type = RET_INTEGER,
219};
220
221BPF_CALL_0(bpf_get_current_pid_tgid)
222{
223 struct task_struct *task = current;
224
225 if (unlikely(!task))
226 return -EINVAL;
227
228 return (u64) task->tgid << 32 | task->pid;
229}
230
231const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
232 .func = bpf_get_current_pid_tgid,
233 .gpl_only = false,
234 .ret_type = RET_INTEGER,
235};
236
237BPF_CALL_0(bpf_get_current_uid_gid)
238{
239 struct task_struct *task = current;
240 kuid_t uid;
241 kgid_t gid;
242
243 if (unlikely(!task))
244 return -EINVAL;
245
246 current_uid_gid(&uid, &gid);
247 return (u64) from_kgid(&init_user_ns, gid) << 32 |
248 from_kuid(&init_user_ns, uid);
249}
250
251const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
252 .func = bpf_get_current_uid_gid,
253 .gpl_only = false,
254 .ret_type = RET_INTEGER,
255};
256
257BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
258{
259 struct task_struct *task = current;
260
261 if (unlikely(!task))
262 goto err_clear;
263
264 /* Verifier guarantees that size > 0 */
265 strscpy_pad(buf, task->comm, size);
266 return 0;
267err_clear:
268 memset(buf, 0, size);
269 return -EINVAL;
270}
271
272const struct bpf_func_proto bpf_get_current_comm_proto = {
273 .func = bpf_get_current_comm,
274 .gpl_only = false,
275 .ret_type = RET_INTEGER,
276 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
277 .arg2_type = ARG_CONST_SIZE,
278};
279
280#if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
281
282static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
283{
284 arch_spinlock_t *l = (void *)lock;
285 union {
286 __u32 val;
287 arch_spinlock_t lock;
288 } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
289
290 compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
291 BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
292 BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
293 preempt_disable();
294 arch_spin_lock(l);
295}
296
297static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
298{
299 arch_spinlock_t *l = (void *)lock;
300
301 arch_spin_unlock(l);
302 preempt_enable();
303}
304
305#else
306
307static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
308{
309 atomic_t *l = (void *)lock;
310
311 BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
312 do {
313 atomic_cond_read_relaxed(l, !VAL);
314 } while (atomic_xchg(l, 1));
315}
316
317static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
318{
319 atomic_t *l = (void *)lock;
320
321 atomic_set_release(l, 0);
322}
323
324#endif
325
326static DEFINE_PER_CPU(unsigned long, irqsave_flags);
327
328static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
329{
330 unsigned long flags;
331
332 local_irq_save(flags);
333 __bpf_spin_lock(lock);
334 __this_cpu_write(irqsave_flags, flags);
335}
336
337notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
338{
339 __bpf_spin_lock_irqsave(lock);
340 return 0;
341}
342
343const struct bpf_func_proto bpf_spin_lock_proto = {
344 .func = bpf_spin_lock,
345 .gpl_only = false,
346 .ret_type = RET_VOID,
347 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
348 .arg1_btf_id = BPF_PTR_POISON,
349};
350
351static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
352{
353 unsigned long flags;
354
355 flags = __this_cpu_read(irqsave_flags);
356 __bpf_spin_unlock(lock);
357 local_irq_restore(flags);
358}
359
360notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
361{
362 __bpf_spin_unlock_irqrestore(lock);
363 return 0;
364}
365
366const struct bpf_func_proto bpf_spin_unlock_proto = {
367 .func = bpf_spin_unlock,
368 .gpl_only = false,
369 .ret_type = RET_VOID,
370 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
371 .arg1_btf_id = BPF_PTR_POISON,
372};
373
374void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
375 bool lock_src)
376{
377 struct bpf_spin_lock *lock;
378
379 if (lock_src)
380 lock = src + map->record->spin_lock_off;
381 else
382 lock = dst + map->record->spin_lock_off;
383 preempt_disable();
384 __bpf_spin_lock_irqsave(lock);
385 copy_map_value(map, dst, src);
386 __bpf_spin_unlock_irqrestore(lock);
387 preempt_enable();
388}
389
390BPF_CALL_0(bpf_jiffies64)
391{
392 return get_jiffies_64();
393}
394
395const struct bpf_func_proto bpf_jiffies64_proto = {
396 .func = bpf_jiffies64,
397 .gpl_only = false,
398 .ret_type = RET_INTEGER,
399};
400
401#ifdef CONFIG_CGROUPS
402BPF_CALL_0(bpf_get_current_cgroup_id)
403{
404 struct cgroup *cgrp;
405 u64 cgrp_id;
406
407 rcu_read_lock();
408 cgrp = task_dfl_cgroup(current);
409 cgrp_id = cgroup_id(cgrp);
410 rcu_read_unlock();
411
412 return cgrp_id;
413}
414
415const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
416 .func = bpf_get_current_cgroup_id,
417 .gpl_only = false,
418 .ret_type = RET_INTEGER,
419};
420
421BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
422{
423 struct cgroup *cgrp;
424 struct cgroup *ancestor;
425 u64 cgrp_id;
426
427 rcu_read_lock();
428 cgrp = task_dfl_cgroup(current);
429 ancestor = cgroup_ancestor(cgrp, ancestor_level);
430 cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
431 rcu_read_unlock();
432
433 return cgrp_id;
434}
435
436const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
437 .func = bpf_get_current_ancestor_cgroup_id,
438 .gpl_only = false,
439 .ret_type = RET_INTEGER,
440 .arg1_type = ARG_ANYTHING,
441};
442#endif /* CONFIG_CGROUPS */
443
444#define BPF_STRTOX_BASE_MASK 0x1F
445
446static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
447 unsigned long long *res, bool *is_negative)
448{
449 unsigned int base = flags & BPF_STRTOX_BASE_MASK;
450 const char *cur_buf = buf;
451 size_t cur_len = buf_len;
452 unsigned int consumed;
453 size_t val_len;
454 char str[64];
455
456 if (!buf || !buf_len || !res || !is_negative)
457 return -EINVAL;
458
459 if (base != 0 && base != 8 && base != 10 && base != 16)
460 return -EINVAL;
461
462 if (flags & ~BPF_STRTOX_BASE_MASK)
463 return -EINVAL;
464
465 while (cur_buf < buf + buf_len && isspace(*cur_buf))
466 ++cur_buf;
467
468 *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
469 if (*is_negative)
470 ++cur_buf;
471
472 consumed = cur_buf - buf;
473 cur_len -= consumed;
474 if (!cur_len)
475 return -EINVAL;
476
477 cur_len = min(cur_len, sizeof(str) - 1);
478 memcpy(str, cur_buf, cur_len);
479 str[cur_len] = '\0';
480 cur_buf = str;
481
482 cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
483 val_len = _parse_integer(cur_buf, base, res);
484
485 if (val_len & KSTRTOX_OVERFLOW)
486 return -ERANGE;
487
488 if (val_len == 0)
489 return -EINVAL;
490
491 cur_buf += val_len;
492 consumed += cur_buf - str;
493
494 return consumed;
495}
496
497static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
498 long long *res)
499{
500 unsigned long long _res;
501 bool is_negative;
502 int err;
503
504 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
505 if (err < 0)
506 return err;
507 if (is_negative) {
508 if ((long long)-_res > 0)
509 return -ERANGE;
510 *res = -_res;
511 } else {
512 if ((long long)_res < 0)
513 return -ERANGE;
514 *res = _res;
515 }
516 return err;
517}
518
519BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
520 long *, res)
521{
522 long long _res;
523 int err;
524
525 err = __bpf_strtoll(buf, buf_len, flags, &_res);
526 if (err < 0)
527 return err;
528 if (_res != (long)_res)
529 return -ERANGE;
530 *res = _res;
531 return err;
532}
533
534const struct bpf_func_proto bpf_strtol_proto = {
535 .func = bpf_strtol,
536 .gpl_only = false,
537 .ret_type = RET_INTEGER,
538 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
539 .arg2_type = ARG_CONST_SIZE,
540 .arg3_type = ARG_ANYTHING,
541 .arg4_type = ARG_PTR_TO_LONG,
542};
543
544BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
545 unsigned long *, res)
546{
547 unsigned long long _res;
548 bool is_negative;
549 int err;
550
551 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
552 if (err < 0)
553 return err;
554 if (is_negative)
555 return -EINVAL;
556 if (_res != (unsigned long)_res)
557 return -ERANGE;
558 *res = _res;
559 return err;
560}
561
562const struct bpf_func_proto bpf_strtoul_proto = {
563 .func = bpf_strtoul,
564 .gpl_only = false,
565 .ret_type = RET_INTEGER,
566 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
567 .arg2_type = ARG_CONST_SIZE,
568 .arg3_type = ARG_ANYTHING,
569 .arg4_type = ARG_PTR_TO_LONG,
570};
571
572BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
573{
574 return strncmp(s1, s2, s1_sz);
575}
576
577static const struct bpf_func_proto bpf_strncmp_proto = {
578 .func = bpf_strncmp,
579 .gpl_only = false,
580 .ret_type = RET_INTEGER,
581 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
582 .arg2_type = ARG_CONST_SIZE,
583 .arg3_type = ARG_PTR_TO_CONST_STR,
584};
585
586BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
587 struct bpf_pidns_info *, nsdata, u32, size)
588{
589 struct task_struct *task = current;
590 struct pid_namespace *pidns;
591 int err = -EINVAL;
592
593 if (unlikely(size != sizeof(struct bpf_pidns_info)))
594 goto clear;
595
596 if (unlikely((u64)(dev_t)dev != dev))
597 goto clear;
598
599 if (unlikely(!task))
600 goto clear;
601
602 pidns = task_active_pid_ns(task);
603 if (unlikely(!pidns)) {
604 err = -ENOENT;
605 goto clear;
606 }
607
608 if (!ns_match(&pidns->ns, (dev_t)dev, ino))
609 goto clear;
610
611 nsdata->pid = task_pid_nr_ns(task, pidns);
612 nsdata->tgid = task_tgid_nr_ns(task, pidns);
613 return 0;
614clear:
615 memset((void *)nsdata, 0, (size_t) size);
616 return err;
617}
618
619const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
620 .func = bpf_get_ns_current_pid_tgid,
621 .gpl_only = false,
622 .ret_type = RET_INTEGER,
623 .arg1_type = ARG_ANYTHING,
624 .arg2_type = ARG_ANYTHING,
625 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
626 .arg4_type = ARG_CONST_SIZE,
627};
628
629static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
630 .func = bpf_get_raw_cpu_id,
631 .gpl_only = false,
632 .ret_type = RET_INTEGER,
633};
634
635BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
636 u64, flags, void *, data, u64, size)
637{
638 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
639 return -EINVAL;
640
641 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
642}
643
644const struct bpf_func_proto bpf_event_output_data_proto = {
645 .func = bpf_event_output_data,
646 .gpl_only = true,
647 .ret_type = RET_INTEGER,
648 .arg1_type = ARG_PTR_TO_CTX,
649 .arg2_type = ARG_CONST_MAP_PTR,
650 .arg3_type = ARG_ANYTHING,
651 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
652 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
653};
654
655BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
656 const void __user *, user_ptr)
657{
658 int ret = copy_from_user(dst, user_ptr, size);
659
660 if (unlikely(ret)) {
661 memset(dst, 0, size);
662 ret = -EFAULT;
663 }
664
665 return ret;
666}
667
668const struct bpf_func_proto bpf_copy_from_user_proto = {
669 .func = bpf_copy_from_user,
670 .gpl_only = false,
671 .might_sleep = true,
672 .ret_type = RET_INTEGER,
673 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
674 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
675 .arg3_type = ARG_ANYTHING,
676};
677
678BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
679 const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
680{
681 int ret;
682
683 /* flags is not used yet */
684 if (unlikely(flags))
685 return -EINVAL;
686
687 if (unlikely(!size))
688 return 0;
689
690 ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
691 if (ret == size)
692 return 0;
693
694 memset(dst, 0, size);
695 /* Return -EFAULT for partial read */
696 return ret < 0 ? ret : -EFAULT;
697}
698
699const struct bpf_func_proto bpf_copy_from_user_task_proto = {
700 .func = bpf_copy_from_user_task,
701 .gpl_only = true,
702 .might_sleep = true,
703 .ret_type = RET_INTEGER,
704 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
705 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
706 .arg3_type = ARG_ANYTHING,
707 .arg4_type = ARG_PTR_TO_BTF_ID,
708 .arg4_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
709 .arg5_type = ARG_ANYTHING
710};
711
712BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
713{
714 if (cpu >= nr_cpu_ids)
715 return (unsigned long)NULL;
716
717 return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
718}
719
720const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
721 .func = bpf_per_cpu_ptr,
722 .gpl_only = false,
723 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
724 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
725 .arg2_type = ARG_ANYTHING,
726};
727
728BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
729{
730 return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
731}
732
733const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
734 .func = bpf_this_cpu_ptr,
735 .gpl_only = false,
736 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
737 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
738};
739
740static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
741 size_t bufsz)
742{
743 void __user *user_ptr = (__force void __user *)unsafe_ptr;
744
745 buf[0] = 0;
746
747 switch (fmt_ptype) {
748 case 's':
749#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
750 if ((unsigned long)unsafe_ptr < TASK_SIZE)
751 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
752 fallthrough;
753#endif
754 case 'k':
755 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
756 case 'u':
757 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
758 }
759
760 return -EINVAL;
761}
762
763/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
764 * arguments representation.
765 */
766#define MAX_BPRINTF_BIN_ARGS 512
767
768/* Support executing three nested bprintf helper calls on a given CPU */
769#define MAX_BPRINTF_NEST_LEVEL 3
770struct bpf_bprintf_buffers {
771 char bin_args[MAX_BPRINTF_BIN_ARGS];
772 char buf[MAX_BPRINTF_BUF];
773};
774
775static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
776static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
777
778static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
779{
780 int nest_level;
781
782 preempt_disable();
783 nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
784 if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
785 this_cpu_dec(bpf_bprintf_nest_level);
786 preempt_enable();
787 return -EBUSY;
788 }
789 *bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
790
791 return 0;
792}
793
794void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
795{
796 if (!data->bin_args && !data->buf)
797 return;
798 if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
799 return;
800 this_cpu_dec(bpf_bprintf_nest_level);
801 preempt_enable();
802}
803
804/*
805 * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
806 *
807 * Returns a negative value if fmt is an invalid format string or 0 otherwise.
808 *
809 * This can be used in two ways:
810 * - Format string verification only: when data->get_bin_args is false
811 * - Arguments preparation: in addition to the above verification, it writes in
812 * data->bin_args a binary representation of arguments usable by bstr_printf
813 * where pointers from BPF have been sanitized.
814 *
815 * In argument preparation mode, if 0 is returned, safe temporary buffers are
816 * allocated and bpf_bprintf_cleanup should be called to free them after use.
817 */
818int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
819 u32 num_args, struct bpf_bprintf_data *data)
820{
821 bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
822 char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
823 struct bpf_bprintf_buffers *buffers = NULL;
824 size_t sizeof_cur_arg, sizeof_cur_ip;
825 int err, i, num_spec = 0;
826 u64 cur_arg;
827 char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
828
829 fmt_end = strnchr(fmt, fmt_size, 0);
830 if (!fmt_end)
831 return -EINVAL;
832 fmt_size = fmt_end - fmt;
833
834 if (get_buffers && try_get_buffers(&buffers))
835 return -EBUSY;
836
837 if (data->get_bin_args) {
838 if (num_args)
839 tmp_buf = buffers->bin_args;
840 tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
841 data->bin_args = (u32 *)tmp_buf;
842 }
843
844 if (data->get_buf)
845 data->buf = buffers->buf;
846
847 for (i = 0; i < fmt_size; i++) {
848 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
849 err = -EINVAL;
850 goto out;
851 }
852
853 if (fmt[i] != '%')
854 continue;
855
856 if (fmt[i + 1] == '%') {
857 i++;
858 continue;
859 }
860
861 if (num_spec >= num_args) {
862 err = -EINVAL;
863 goto out;
864 }
865
866 /* The string is zero-terminated so if fmt[i] != 0, we can
867 * always access fmt[i + 1], in the worst case it will be a 0
868 */
869 i++;
870
871 /* skip optional "[0 +-][num]" width formatting field */
872 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
873 fmt[i] == ' ')
874 i++;
875 if (fmt[i] >= '1' && fmt[i] <= '9') {
876 i++;
877 while (fmt[i] >= '0' && fmt[i] <= '9')
878 i++;
879 }
880
881 if (fmt[i] == 'p') {
882 sizeof_cur_arg = sizeof(long);
883
884 if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
885 fmt[i + 2] == 's') {
886 fmt_ptype = fmt[i + 1];
887 i += 2;
888 goto fmt_str;
889 }
890
891 if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
892 ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
893 fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
894 fmt[i + 1] == 'S') {
895 /* just kernel pointers */
896 if (tmp_buf)
897 cur_arg = raw_args[num_spec];
898 i++;
899 goto nocopy_fmt;
900 }
901
902 if (fmt[i + 1] == 'B') {
903 if (tmp_buf) {
904 err = snprintf(tmp_buf,
905 (tmp_buf_end - tmp_buf),
906 "%pB",
907 (void *)(long)raw_args[num_spec]);
908 tmp_buf += (err + 1);
909 }
910
911 i++;
912 num_spec++;
913 continue;
914 }
915
916 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
917 if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
918 (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
919 err = -EINVAL;
920 goto out;
921 }
922
923 i += 2;
924 if (!tmp_buf)
925 goto nocopy_fmt;
926
927 sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
928 if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
929 err = -ENOSPC;
930 goto out;
931 }
932
933 unsafe_ptr = (char *)(long)raw_args[num_spec];
934 err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
935 sizeof_cur_ip);
936 if (err < 0)
937 memset(cur_ip, 0, sizeof_cur_ip);
938
939 /* hack: bstr_printf expects IP addresses to be
940 * pre-formatted as strings, ironically, the easiest way
941 * to do that is to call snprintf.
942 */
943 ip_spec[2] = fmt[i - 1];
944 ip_spec[3] = fmt[i];
945 err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
946 ip_spec, &cur_ip);
947
948 tmp_buf += err + 1;
949 num_spec++;
950
951 continue;
952 } else if (fmt[i] == 's') {
953 fmt_ptype = fmt[i];
954fmt_str:
955 if (fmt[i + 1] != 0 &&
956 !isspace(fmt[i + 1]) &&
957 !ispunct(fmt[i + 1])) {
958 err = -EINVAL;
959 goto out;
960 }
961
962 if (!tmp_buf)
963 goto nocopy_fmt;
964
965 if (tmp_buf_end == tmp_buf) {
966 err = -ENOSPC;
967 goto out;
968 }
969
970 unsafe_ptr = (char *)(long)raw_args[num_spec];
971 err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
972 fmt_ptype,
973 tmp_buf_end - tmp_buf);
974 if (err < 0) {
975 tmp_buf[0] = '\0';
976 err = 1;
977 }
978
979 tmp_buf += err;
980 num_spec++;
981
982 continue;
983 } else if (fmt[i] == 'c') {
984 if (!tmp_buf)
985 goto nocopy_fmt;
986
987 if (tmp_buf_end == tmp_buf) {
988 err = -ENOSPC;
989 goto out;
990 }
991
992 *tmp_buf = raw_args[num_spec];
993 tmp_buf++;
994 num_spec++;
995
996 continue;
997 }
998
999 sizeof_cur_arg = sizeof(int);
1000
1001 if (fmt[i] == 'l') {
1002 sizeof_cur_arg = sizeof(long);
1003 i++;
1004 }
1005 if (fmt[i] == 'l') {
1006 sizeof_cur_arg = sizeof(long long);
1007 i++;
1008 }
1009
1010 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1011 fmt[i] != 'x' && fmt[i] != 'X') {
1012 err = -EINVAL;
1013 goto out;
1014 }
1015
1016 if (tmp_buf)
1017 cur_arg = raw_args[num_spec];
1018nocopy_fmt:
1019 if (tmp_buf) {
1020 tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1021 if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1022 err = -ENOSPC;
1023 goto out;
1024 }
1025
1026 if (sizeof_cur_arg == 8) {
1027 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
1028 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1029 } else {
1030 *(u32 *)tmp_buf = (u32)(long)cur_arg;
1031 }
1032 tmp_buf += sizeof_cur_arg;
1033 }
1034 num_spec++;
1035 }
1036
1037 err = 0;
1038out:
1039 if (err)
1040 bpf_bprintf_cleanup(data);
1041 return err;
1042}
1043
1044BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1045 const void *, args, u32, data_len)
1046{
1047 struct bpf_bprintf_data data = {
1048 .get_bin_args = true,
1049 };
1050 int err, num_args;
1051
1052 if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1053 (data_len && !args))
1054 return -EINVAL;
1055 num_args = data_len / 8;
1056
1057 /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1058 * can safely give an unbounded size.
1059 */
1060 err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1061 if (err < 0)
1062 return err;
1063
1064 err = bstr_printf(str, str_size, fmt, data.bin_args);
1065
1066 bpf_bprintf_cleanup(&data);
1067
1068 return err + 1;
1069}
1070
1071const struct bpf_func_proto bpf_snprintf_proto = {
1072 .func = bpf_snprintf,
1073 .gpl_only = true,
1074 .ret_type = RET_INTEGER,
1075 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1076 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1077 .arg3_type = ARG_PTR_TO_CONST_STR,
1078 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1079 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1080};
1081
1082/* BPF map elements can contain 'struct bpf_timer'.
1083 * Such map owns all of its BPF timers.
1084 * 'struct bpf_timer' is allocated as part of map element allocation
1085 * and it's zero initialized.
1086 * That space is used to keep 'struct bpf_timer_kern'.
1087 * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1088 * remembers 'struct bpf_map *' pointer it's part of.
1089 * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1090 * bpf_timer_start() arms the timer.
1091 * If user space reference to a map goes to zero at this point
1092 * ops->map_release_uref callback is responsible for cancelling the timers,
1093 * freeing their memory, and decrementing prog's refcnts.
1094 * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1095 * Inner maps can contain bpf timers as well. ops->map_release_uref is
1096 * freeing the timers when inner map is replaced or deleted by user space.
1097 */
1098struct bpf_hrtimer {
1099 struct hrtimer timer;
1100 struct bpf_map *map;
1101 struct bpf_prog *prog;
1102 void __rcu *callback_fn;
1103 void *value;
1104 struct rcu_head rcu;
1105};
1106
1107/* the actual struct hidden inside uapi struct bpf_timer */
1108struct bpf_timer_kern {
1109 struct bpf_hrtimer *timer;
1110 /* bpf_spin_lock is used here instead of spinlock_t to make
1111 * sure that it always fits into space reserved by struct bpf_timer
1112 * regardless of LOCKDEP and spinlock debug flags.
1113 */
1114 struct bpf_spin_lock lock;
1115} __attribute__((aligned(8)));
1116
1117static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1118
1119static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1120{
1121 struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1122 struct bpf_map *map = t->map;
1123 void *value = t->value;
1124 bpf_callback_t callback_fn;
1125 void *key;
1126 u32 idx;
1127
1128 BTF_TYPE_EMIT(struct bpf_timer);
1129 callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1130 if (!callback_fn)
1131 goto out;
1132
1133 /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1134 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1135 * Remember the timer this callback is servicing to prevent
1136 * deadlock if callback_fn() calls bpf_timer_cancel() or
1137 * bpf_map_delete_elem() on the same timer.
1138 */
1139 this_cpu_write(hrtimer_running, t);
1140 if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1141 struct bpf_array *array = container_of(map, struct bpf_array, map);
1142
1143 /* compute the key */
1144 idx = ((char *)value - array->value) / array->elem_size;
1145 key = &idx;
1146 } else { /* hash or lru */
1147 key = value - round_up(map->key_size, 8);
1148 }
1149
1150 callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1151 /* The verifier checked that return value is zero. */
1152
1153 this_cpu_write(hrtimer_running, NULL);
1154out:
1155 return HRTIMER_NORESTART;
1156}
1157
1158BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1159 u64, flags)
1160{
1161 clockid_t clockid = flags & (MAX_CLOCKS - 1);
1162 struct bpf_hrtimer *t;
1163 int ret = 0;
1164
1165 BUILD_BUG_ON(MAX_CLOCKS != 16);
1166 BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1167 BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1168
1169 if (in_nmi())
1170 return -EOPNOTSUPP;
1171
1172 if (flags >= MAX_CLOCKS ||
1173 /* similar to timerfd except _ALARM variants are not supported */
1174 (clockid != CLOCK_MONOTONIC &&
1175 clockid != CLOCK_REALTIME &&
1176 clockid != CLOCK_BOOTTIME))
1177 return -EINVAL;
1178 __bpf_spin_lock_irqsave(&timer->lock);
1179 t = timer->timer;
1180 if (t) {
1181 ret = -EBUSY;
1182 goto out;
1183 }
1184 /* allocate hrtimer via map_kmalloc to use memcg accounting */
1185 t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1186 if (!t) {
1187 ret = -ENOMEM;
1188 goto out;
1189 }
1190 t->value = (void *)timer - map->record->timer_off;
1191 t->map = map;
1192 t->prog = NULL;
1193 rcu_assign_pointer(t->callback_fn, NULL);
1194 hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1195 t->timer.function = bpf_timer_cb;
1196 WRITE_ONCE(timer->timer, t);
1197 /* Guarantee the order between timer->timer and map->usercnt. So
1198 * when there are concurrent uref release and bpf timer init, either
1199 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1200 * timer or atomic64_read() below returns a zero usercnt.
1201 */
1202 smp_mb();
1203 if (!atomic64_read(&map->usercnt)) {
1204 /* maps with timers must be either held by user space
1205 * or pinned in bpffs.
1206 */
1207 WRITE_ONCE(timer->timer, NULL);
1208 kfree(t);
1209 ret = -EPERM;
1210 }
1211out:
1212 __bpf_spin_unlock_irqrestore(&timer->lock);
1213 return ret;
1214}
1215
1216static const struct bpf_func_proto bpf_timer_init_proto = {
1217 .func = bpf_timer_init,
1218 .gpl_only = true,
1219 .ret_type = RET_INTEGER,
1220 .arg1_type = ARG_PTR_TO_TIMER,
1221 .arg2_type = ARG_CONST_MAP_PTR,
1222 .arg3_type = ARG_ANYTHING,
1223};
1224
1225BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1226 struct bpf_prog_aux *, aux)
1227{
1228 struct bpf_prog *prev, *prog = aux->prog;
1229 struct bpf_hrtimer *t;
1230 int ret = 0;
1231
1232 if (in_nmi())
1233 return -EOPNOTSUPP;
1234 __bpf_spin_lock_irqsave(&timer->lock);
1235 t = timer->timer;
1236 if (!t) {
1237 ret = -EINVAL;
1238 goto out;
1239 }
1240 if (!atomic64_read(&t->map->usercnt)) {
1241 /* maps with timers must be either held by user space
1242 * or pinned in bpffs. Otherwise timer might still be
1243 * running even when bpf prog is detached and user space
1244 * is gone, since map_release_uref won't ever be called.
1245 */
1246 ret = -EPERM;
1247 goto out;
1248 }
1249 prev = t->prog;
1250 if (prev != prog) {
1251 /* Bump prog refcnt once. Every bpf_timer_set_callback()
1252 * can pick different callback_fn-s within the same prog.
1253 */
1254 prog = bpf_prog_inc_not_zero(prog);
1255 if (IS_ERR(prog)) {
1256 ret = PTR_ERR(prog);
1257 goto out;
1258 }
1259 if (prev)
1260 /* Drop prev prog refcnt when swapping with new prog */
1261 bpf_prog_put(prev);
1262 t->prog = prog;
1263 }
1264 rcu_assign_pointer(t->callback_fn, callback_fn);
1265out:
1266 __bpf_spin_unlock_irqrestore(&timer->lock);
1267 return ret;
1268}
1269
1270static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1271 .func = bpf_timer_set_callback,
1272 .gpl_only = true,
1273 .ret_type = RET_INTEGER,
1274 .arg1_type = ARG_PTR_TO_TIMER,
1275 .arg2_type = ARG_PTR_TO_FUNC,
1276};
1277
1278BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1279{
1280 struct bpf_hrtimer *t;
1281 int ret = 0;
1282 enum hrtimer_mode mode;
1283
1284 if (in_nmi())
1285 return -EOPNOTSUPP;
1286 if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
1287 return -EINVAL;
1288 __bpf_spin_lock_irqsave(&timer->lock);
1289 t = timer->timer;
1290 if (!t || !t->prog) {
1291 ret = -EINVAL;
1292 goto out;
1293 }
1294
1295 if (flags & BPF_F_TIMER_ABS)
1296 mode = HRTIMER_MODE_ABS_SOFT;
1297 else
1298 mode = HRTIMER_MODE_REL_SOFT;
1299
1300 if (flags & BPF_F_TIMER_CPU_PIN)
1301 mode |= HRTIMER_MODE_PINNED;
1302
1303 hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
1304out:
1305 __bpf_spin_unlock_irqrestore(&timer->lock);
1306 return ret;
1307}
1308
1309static const struct bpf_func_proto bpf_timer_start_proto = {
1310 .func = bpf_timer_start,
1311 .gpl_only = true,
1312 .ret_type = RET_INTEGER,
1313 .arg1_type = ARG_PTR_TO_TIMER,
1314 .arg2_type = ARG_ANYTHING,
1315 .arg3_type = ARG_ANYTHING,
1316};
1317
1318static void drop_prog_refcnt(struct bpf_hrtimer *t)
1319{
1320 struct bpf_prog *prog = t->prog;
1321
1322 if (prog) {
1323 bpf_prog_put(prog);
1324 t->prog = NULL;
1325 rcu_assign_pointer(t->callback_fn, NULL);
1326 }
1327}
1328
1329BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1330{
1331 struct bpf_hrtimer *t;
1332 int ret = 0;
1333
1334 if (in_nmi())
1335 return -EOPNOTSUPP;
1336 rcu_read_lock();
1337 __bpf_spin_lock_irqsave(&timer->lock);
1338 t = timer->timer;
1339 if (!t) {
1340 ret = -EINVAL;
1341 goto out;
1342 }
1343 if (this_cpu_read(hrtimer_running) == t) {
1344 /* If bpf callback_fn is trying to bpf_timer_cancel()
1345 * its own timer the hrtimer_cancel() will deadlock
1346 * since it waits for callback_fn to finish
1347 */
1348 ret = -EDEADLK;
1349 goto out;
1350 }
1351 drop_prog_refcnt(t);
1352out:
1353 __bpf_spin_unlock_irqrestore(&timer->lock);
1354 /* Cancel the timer and wait for associated callback to finish
1355 * if it was running.
1356 */
1357 ret = ret ?: hrtimer_cancel(&t->timer);
1358 rcu_read_unlock();
1359 return ret;
1360}
1361
1362static const struct bpf_func_proto bpf_timer_cancel_proto = {
1363 .func = bpf_timer_cancel,
1364 .gpl_only = true,
1365 .ret_type = RET_INTEGER,
1366 .arg1_type = ARG_PTR_TO_TIMER,
1367};
1368
1369/* This function is called by map_delete/update_elem for individual element and
1370 * by ops->map_release_uref when the user space reference to a map reaches zero.
1371 */
1372void bpf_timer_cancel_and_free(void *val)
1373{
1374 struct bpf_timer_kern *timer = val;
1375 struct bpf_hrtimer *t;
1376
1377 /* Performance optimization: read timer->timer without lock first. */
1378 if (!READ_ONCE(timer->timer))
1379 return;
1380
1381 __bpf_spin_lock_irqsave(&timer->lock);
1382 /* re-read it under lock */
1383 t = timer->timer;
1384 if (!t)
1385 goto out;
1386 drop_prog_refcnt(t);
1387 /* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1388 * this timer, since it won't be initialized.
1389 */
1390 WRITE_ONCE(timer->timer, NULL);
1391out:
1392 __bpf_spin_unlock_irqrestore(&timer->lock);
1393 if (!t)
1394 return;
1395 /* Cancel the timer and wait for callback to complete if it was running.
1396 * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1397 * right after for both preallocated and non-preallocated maps.
1398 * The timer->timer = NULL was already done and no code path can
1399 * see address 't' anymore.
1400 *
1401 * Check that bpf_map_delete/update_elem() wasn't called from timer
1402 * callback_fn. In such case don't call hrtimer_cancel() (since it will
1403 * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1404 * return -1). Though callback_fn is still running on this cpu it's
1405 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1406 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1407 * since timer->timer = NULL was already done. The timer will be
1408 * effectively cancelled because bpf_timer_cb() will return
1409 * HRTIMER_NORESTART.
1410 */
1411 if (this_cpu_read(hrtimer_running) != t)
1412 hrtimer_cancel(&t->timer);
1413 kfree_rcu(t, rcu);
1414}
1415
1416BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
1417{
1418 unsigned long *kptr = map_value;
1419
1420 return xchg(kptr, (unsigned long)ptr);
1421}
1422
1423/* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1424 * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1425 * denote type that verifier will determine.
1426 */
1427static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1428 .func = bpf_kptr_xchg,
1429 .gpl_only = false,
1430 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1431 .ret_btf_id = BPF_PTR_POISON,
1432 .arg1_type = ARG_PTR_TO_KPTR,
1433 .arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1434 .arg2_btf_id = BPF_PTR_POISON,
1435};
1436
1437/* Since the upper 8 bits of dynptr->size is reserved, the
1438 * maximum supported size is 2^24 - 1.
1439 */
1440#define DYNPTR_MAX_SIZE ((1UL << 24) - 1)
1441#define DYNPTR_TYPE_SHIFT 28
1442#define DYNPTR_SIZE_MASK 0xFFFFFF
1443#define DYNPTR_RDONLY_BIT BIT(31)
1444
1445static bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1446{
1447 return ptr->size & DYNPTR_RDONLY_BIT;
1448}
1449
1450void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1451{
1452 ptr->size |= DYNPTR_RDONLY_BIT;
1453}
1454
1455static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1456{
1457 ptr->size |= type << DYNPTR_TYPE_SHIFT;
1458}
1459
1460static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1461{
1462 return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1463}
1464
1465u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1466{
1467 return ptr->size & DYNPTR_SIZE_MASK;
1468}
1469
1470static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1471{
1472 u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1473
1474 ptr->size = new_size | metadata;
1475}
1476
1477int bpf_dynptr_check_size(u32 size)
1478{
1479 return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1480}
1481
1482void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1483 enum bpf_dynptr_type type, u32 offset, u32 size)
1484{
1485 ptr->data = data;
1486 ptr->offset = offset;
1487 ptr->size = size;
1488 bpf_dynptr_set_type(ptr, type);
1489}
1490
1491void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1492{
1493 memset(ptr, 0, sizeof(*ptr));
1494}
1495
1496static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1497{
1498 u32 size = __bpf_dynptr_size(ptr);
1499
1500 if (len > size || offset > size - len)
1501 return -E2BIG;
1502
1503 return 0;
1504}
1505
1506BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1507{
1508 int err;
1509
1510 BTF_TYPE_EMIT(struct bpf_dynptr);
1511
1512 err = bpf_dynptr_check_size(size);
1513 if (err)
1514 goto error;
1515
1516 /* flags is currently unsupported */
1517 if (flags) {
1518 err = -EINVAL;
1519 goto error;
1520 }
1521
1522 bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1523
1524 return 0;
1525
1526error:
1527 bpf_dynptr_set_null(ptr);
1528 return err;
1529}
1530
1531static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1532 .func = bpf_dynptr_from_mem,
1533 .gpl_only = false,
1534 .ret_type = RET_INTEGER,
1535 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1536 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1537 .arg3_type = ARG_ANYTHING,
1538 .arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT,
1539};
1540
1541BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1542 u32, offset, u64, flags)
1543{
1544 enum bpf_dynptr_type type;
1545 int err;
1546
1547 if (!src->data || flags)
1548 return -EINVAL;
1549
1550 err = bpf_dynptr_check_off_len(src, offset, len);
1551 if (err)
1552 return err;
1553
1554 type = bpf_dynptr_get_type(src);
1555
1556 switch (type) {
1557 case BPF_DYNPTR_TYPE_LOCAL:
1558 case BPF_DYNPTR_TYPE_RINGBUF:
1559 /* Source and destination may possibly overlap, hence use memmove to
1560 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1561 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1562 */
1563 memmove(dst, src->data + src->offset + offset, len);
1564 return 0;
1565 case BPF_DYNPTR_TYPE_SKB:
1566 return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len);
1567 case BPF_DYNPTR_TYPE_XDP:
1568 return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len);
1569 default:
1570 WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1571 return -EFAULT;
1572 }
1573}
1574
1575static const struct bpf_func_proto bpf_dynptr_read_proto = {
1576 .func = bpf_dynptr_read,
1577 .gpl_only = false,
1578 .ret_type = RET_INTEGER,
1579 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1580 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1581 .arg3_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1582 .arg4_type = ARG_ANYTHING,
1583 .arg5_type = ARG_ANYTHING,
1584};
1585
1586BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1587 u32, len, u64, flags)
1588{
1589 enum bpf_dynptr_type type;
1590 int err;
1591
1592 if (!dst->data || __bpf_dynptr_is_rdonly(dst))
1593 return -EINVAL;
1594
1595 err = bpf_dynptr_check_off_len(dst, offset, len);
1596 if (err)
1597 return err;
1598
1599 type = bpf_dynptr_get_type(dst);
1600
1601 switch (type) {
1602 case BPF_DYNPTR_TYPE_LOCAL:
1603 case BPF_DYNPTR_TYPE_RINGBUF:
1604 if (flags)
1605 return -EINVAL;
1606 /* Source and destination may possibly overlap, hence use memmove to
1607 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1608 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1609 */
1610 memmove(dst->data + dst->offset + offset, src, len);
1611 return 0;
1612 case BPF_DYNPTR_TYPE_SKB:
1613 return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len,
1614 flags);
1615 case BPF_DYNPTR_TYPE_XDP:
1616 if (flags)
1617 return -EINVAL;
1618 return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len);
1619 default:
1620 WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1621 return -EFAULT;
1622 }
1623}
1624
1625static const struct bpf_func_proto bpf_dynptr_write_proto = {
1626 .func = bpf_dynptr_write,
1627 .gpl_only = false,
1628 .ret_type = RET_INTEGER,
1629 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1630 .arg2_type = ARG_ANYTHING,
1631 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1632 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1633 .arg5_type = ARG_ANYTHING,
1634};
1635
1636BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1637{
1638 enum bpf_dynptr_type type;
1639 int err;
1640
1641 if (!ptr->data)
1642 return 0;
1643
1644 err = bpf_dynptr_check_off_len(ptr, offset, len);
1645 if (err)
1646 return 0;
1647
1648 if (__bpf_dynptr_is_rdonly(ptr))
1649 return 0;
1650
1651 type = bpf_dynptr_get_type(ptr);
1652
1653 switch (type) {
1654 case BPF_DYNPTR_TYPE_LOCAL:
1655 case BPF_DYNPTR_TYPE_RINGBUF:
1656 return (unsigned long)(ptr->data + ptr->offset + offset);
1657 case BPF_DYNPTR_TYPE_SKB:
1658 case BPF_DYNPTR_TYPE_XDP:
1659 /* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1660 return 0;
1661 default:
1662 WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1663 return 0;
1664 }
1665}
1666
1667static const struct bpf_func_proto bpf_dynptr_data_proto = {
1668 .func = bpf_dynptr_data,
1669 .gpl_only = false,
1670 .ret_type = RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1671 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1672 .arg2_type = ARG_ANYTHING,
1673 .arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO,
1674};
1675
1676const struct bpf_func_proto bpf_get_current_task_proto __weak;
1677const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1678const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1679const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1680const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1681const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1682const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1683
1684const struct bpf_func_proto *
1685bpf_base_func_proto(enum bpf_func_id func_id)
1686{
1687 switch (func_id) {
1688 case BPF_FUNC_map_lookup_elem:
1689 return &bpf_map_lookup_elem_proto;
1690 case BPF_FUNC_map_update_elem:
1691 return &bpf_map_update_elem_proto;
1692 case BPF_FUNC_map_delete_elem:
1693 return &bpf_map_delete_elem_proto;
1694 case BPF_FUNC_map_push_elem:
1695 return &bpf_map_push_elem_proto;
1696 case BPF_FUNC_map_pop_elem:
1697 return &bpf_map_pop_elem_proto;
1698 case BPF_FUNC_map_peek_elem:
1699 return &bpf_map_peek_elem_proto;
1700 case BPF_FUNC_map_lookup_percpu_elem:
1701 return &bpf_map_lookup_percpu_elem_proto;
1702 case BPF_FUNC_get_prandom_u32:
1703 return &bpf_get_prandom_u32_proto;
1704 case BPF_FUNC_get_smp_processor_id:
1705 return &bpf_get_raw_smp_processor_id_proto;
1706 case BPF_FUNC_get_numa_node_id:
1707 return &bpf_get_numa_node_id_proto;
1708 case BPF_FUNC_tail_call:
1709 return &bpf_tail_call_proto;
1710 case BPF_FUNC_ktime_get_ns:
1711 return &bpf_ktime_get_ns_proto;
1712 case BPF_FUNC_ktime_get_boot_ns:
1713 return &bpf_ktime_get_boot_ns_proto;
1714 case BPF_FUNC_ktime_get_tai_ns:
1715 return &bpf_ktime_get_tai_ns_proto;
1716 case BPF_FUNC_ringbuf_output:
1717 return &bpf_ringbuf_output_proto;
1718 case BPF_FUNC_ringbuf_reserve:
1719 return &bpf_ringbuf_reserve_proto;
1720 case BPF_FUNC_ringbuf_submit:
1721 return &bpf_ringbuf_submit_proto;
1722 case BPF_FUNC_ringbuf_discard:
1723 return &bpf_ringbuf_discard_proto;
1724 case BPF_FUNC_ringbuf_query:
1725 return &bpf_ringbuf_query_proto;
1726 case BPF_FUNC_strncmp:
1727 return &bpf_strncmp_proto;
1728 case BPF_FUNC_strtol:
1729 return &bpf_strtol_proto;
1730 case BPF_FUNC_strtoul:
1731 return &bpf_strtoul_proto;
1732 default:
1733 break;
1734 }
1735
1736 if (!bpf_capable())
1737 return NULL;
1738
1739 switch (func_id) {
1740 case BPF_FUNC_spin_lock:
1741 return &bpf_spin_lock_proto;
1742 case BPF_FUNC_spin_unlock:
1743 return &bpf_spin_unlock_proto;
1744 case BPF_FUNC_jiffies64:
1745 return &bpf_jiffies64_proto;
1746 case BPF_FUNC_per_cpu_ptr:
1747 return &bpf_per_cpu_ptr_proto;
1748 case BPF_FUNC_this_cpu_ptr:
1749 return &bpf_this_cpu_ptr_proto;
1750 case BPF_FUNC_timer_init:
1751 return &bpf_timer_init_proto;
1752 case BPF_FUNC_timer_set_callback:
1753 return &bpf_timer_set_callback_proto;
1754 case BPF_FUNC_timer_start:
1755 return &bpf_timer_start_proto;
1756 case BPF_FUNC_timer_cancel:
1757 return &bpf_timer_cancel_proto;
1758 case BPF_FUNC_kptr_xchg:
1759 return &bpf_kptr_xchg_proto;
1760 case BPF_FUNC_for_each_map_elem:
1761 return &bpf_for_each_map_elem_proto;
1762 case BPF_FUNC_loop:
1763 return &bpf_loop_proto;
1764 case BPF_FUNC_user_ringbuf_drain:
1765 return &bpf_user_ringbuf_drain_proto;
1766 case BPF_FUNC_ringbuf_reserve_dynptr:
1767 return &bpf_ringbuf_reserve_dynptr_proto;
1768 case BPF_FUNC_ringbuf_submit_dynptr:
1769 return &bpf_ringbuf_submit_dynptr_proto;
1770 case BPF_FUNC_ringbuf_discard_dynptr:
1771 return &bpf_ringbuf_discard_dynptr_proto;
1772 case BPF_FUNC_dynptr_from_mem:
1773 return &bpf_dynptr_from_mem_proto;
1774 case BPF_FUNC_dynptr_read:
1775 return &bpf_dynptr_read_proto;
1776 case BPF_FUNC_dynptr_write:
1777 return &bpf_dynptr_write_proto;
1778 case BPF_FUNC_dynptr_data:
1779 return &bpf_dynptr_data_proto;
1780#ifdef CONFIG_CGROUPS
1781 case BPF_FUNC_cgrp_storage_get:
1782 return &bpf_cgrp_storage_get_proto;
1783 case BPF_FUNC_cgrp_storage_delete:
1784 return &bpf_cgrp_storage_delete_proto;
1785 case BPF_FUNC_get_current_cgroup_id:
1786 return &bpf_get_current_cgroup_id_proto;
1787 case BPF_FUNC_get_current_ancestor_cgroup_id:
1788 return &bpf_get_current_ancestor_cgroup_id_proto;
1789#endif
1790 default:
1791 break;
1792 }
1793
1794 if (!perfmon_capable())
1795 return NULL;
1796
1797 switch (func_id) {
1798 case BPF_FUNC_trace_printk:
1799 return bpf_get_trace_printk_proto();
1800 case BPF_FUNC_get_current_task:
1801 return &bpf_get_current_task_proto;
1802 case BPF_FUNC_get_current_task_btf:
1803 return &bpf_get_current_task_btf_proto;
1804 case BPF_FUNC_probe_read_user:
1805 return &bpf_probe_read_user_proto;
1806 case BPF_FUNC_probe_read_kernel:
1807 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1808 NULL : &bpf_probe_read_kernel_proto;
1809 case BPF_FUNC_probe_read_user_str:
1810 return &bpf_probe_read_user_str_proto;
1811 case BPF_FUNC_probe_read_kernel_str:
1812 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1813 NULL : &bpf_probe_read_kernel_str_proto;
1814 case BPF_FUNC_snprintf_btf:
1815 return &bpf_snprintf_btf_proto;
1816 case BPF_FUNC_snprintf:
1817 return &bpf_snprintf_proto;
1818 case BPF_FUNC_task_pt_regs:
1819 return &bpf_task_pt_regs_proto;
1820 case BPF_FUNC_trace_vprintk:
1821 return bpf_get_trace_vprintk_proto();
1822 default:
1823 return NULL;
1824 }
1825}
1826
1827void bpf_list_head_free(const struct btf_field *field, void *list_head,
1828 struct bpf_spin_lock *spin_lock)
1829{
1830 struct list_head *head = list_head, *orig_head = list_head;
1831
1832 BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
1833 BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
1834
1835 /* Do the actual list draining outside the lock to not hold the lock for
1836 * too long, and also prevent deadlocks if tracing programs end up
1837 * executing on entry/exit of functions called inside the critical
1838 * section, and end up doing map ops that call bpf_list_head_free for
1839 * the same map value again.
1840 */
1841 __bpf_spin_lock_irqsave(spin_lock);
1842 if (!head->next || list_empty(head))
1843 goto unlock;
1844 head = head->next;
1845unlock:
1846 INIT_LIST_HEAD(orig_head);
1847 __bpf_spin_unlock_irqrestore(spin_lock);
1848
1849 while (head != orig_head) {
1850 void *obj = head;
1851
1852 obj -= field->graph_root.node_offset;
1853 head = head->next;
1854 /* The contained type can also have resources, including a
1855 * bpf_list_head which needs to be freed.
1856 */
1857 migrate_disable();
1858 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
1859 migrate_enable();
1860 }
1861}
1862
1863/* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
1864 * 'rb_node *', so field name of rb_node within containing struct is not
1865 * needed.
1866 *
1867 * Since bpf_rb_tree's node type has a corresponding struct btf_field with
1868 * graph_root.node_offset, it's not necessary to know field name
1869 * or type of node struct
1870 */
1871#define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
1872 for (pos = rb_first_postorder(root); \
1873 pos && ({ n = rb_next_postorder(pos); 1; }); \
1874 pos = n)
1875
1876void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
1877 struct bpf_spin_lock *spin_lock)
1878{
1879 struct rb_root_cached orig_root, *root = rb_root;
1880 struct rb_node *pos, *n;
1881 void *obj;
1882
1883 BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
1884 BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
1885
1886 __bpf_spin_lock_irqsave(spin_lock);
1887 orig_root = *root;
1888 *root = RB_ROOT_CACHED;
1889 __bpf_spin_unlock_irqrestore(spin_lock);
1890
1891 bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
1892 obj = pos;
1893 obj -= field->graph_root.node_offset;
1894
1895
1896 migrate_disable();
1897 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
1898 migrate_enable();
1899 }
1900}
1901
1902__bpf_kfunc_start_defs();
1903
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__bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1919{
1920 u64 size = local_type_id__k;
1921
1922 /* The verifier has ensured that meta__ign must be NULL */
1923 return bpf_mem_alloc(&bpf_global_percpu_ma, size);
1924}
1925
1926/* Must be called under migrate_disable(), as required by bpf_mem_free */
1927void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
1928{
1929 struct bpf_mem_alloc *ma;
1930
1931 if (rec && rec->refcount_off >= 0 &&
1932 !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
1933 /* Object is refcounted and refcount_dec didn't result in 0
1934 * refcount. Return without freeing the object
1935 */
1936 return;
1937 }
1938
1939 if (rec)
1940 bpf_obj_free_fields(rec, p);
1941
1942 if (percpu)
1943 ma = &bpf_global_percpu_ma;
1944 else
1945 ma = &bpf_global_ma;
1946 bpf_mem_free_rcu(ma, p);
1947}
1948
1949__bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
1950{
1951 struct btf_struct_meta *meta = meta__ign;
1952 void *p = p__alloc;
1953
1954 __bpf_obj_drop_impl(p, meta ? meta->record : NULL, false);
1955}
1956
1957__bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
1958{
1959 /* The verifier has ensured that meta__ign must be NULL */
1960 bpf_mem_free_rcu(&bpf_global_percpu_ma, p__alloc);
1961}
1962
1963__bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
1964{
1965 struct btf_struct_meta *meta = meta__ign;
1966 struct bpf_refcount *ref;
1967
1968 /* Could just cast directly to refcount_t *, but need some code using
1969 * bpf_refcount type so that it is emitted in vmlinux BTF
1970 */
1971 ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
1972 if (!refcount_inc_not_zero((refcount_t *)ref))
1973 return NULL;
1974
1975 /* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
1976 * in verifier.c
1977 */
1978 return (void *)p__refcounted_kptr;
1979}
1980
1981static int __bpf_list_add(struct bpf_list_node_kern *node,
1982 struct bpf_list_head *head,
1983 bool tail, struct btf_record *rec, u64 off)
1984{
1985 struct list_head *n = &node->list_head, *h = (void *)head;
1986
1987 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
1988 * called on its fields, so init here
1989 */
1990 if (unlikely(!h->next))
1991 INIT_LIST_HEAD(h);
1992
1993 /* node->owner != NULL implies !list_empty(n), no need to separately
1994 * check the latter
1995 */
1996 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
1997 /* Only called from BPF prog, no need to migrate_disable */
1998 __bpf_obj_drop_impl((void *)n - off, rec, false);
1999 return -EINVAL;
2000 }
2001
2002 tail ? list_add_tail(n, h) : list_add(n, h);
2003 WRITE_ONCE(node->owner, head);
2004
2005 return 0;
2006}
2007
2008__bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
2009 struct bpf_list_node *node,
2010 void *meta__ign, u64 off)
2011{
2012 struct bpf_list_node_kern *n = (void *)node;
2013 struct btf_struct_meta *meta = meta__ign;
2014
2015 return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off);
2016}
2017
2018__bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2019 struct bpf_list_node *node,
2020 void *meta__ign, u64 off)
2021{
2022 struct bpf_list_node_kern *n = (void *)node;
2023 struct btf_struct_meta *meta = meta__ign;
2024
2025 return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off);
2026}
2027
2028static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2029{
2030 struct list_head *n, *h = (void *)head;
2031 struct bpf_list_node_kern *node;
2032
2033 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2034 * called on its fields, so init here
2035 */
2036 if (unlikely(!h->next))
2037 INIT_LIST_HEAD(h);
2038 if (list_empty(h))
2039 return NULL;
2040
2041 n = tail ? h->prev : h->next;
2042 node = container_of(n, struct bpf_list_node_kern, list_head);
2043 if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2044 return NULL;
2045
2046 list_del_init(n);
2047 WRITE_ONCE(node->owner, NULL);
2048 return (struct bpf_list_node *)n;
2049}
2050
2051__bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2052{
2053 return __bpf_list_del(head, false);
2054}
2055
2056__bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2057{
2058 return __bpf_list_del(head, true);
2059}
2060
2061__bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2062 struct bpf_rb_node *node)
2063{
2064 struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2065 struct rb_root_cached *r = (struct rb_root_cached *)root;
2066 struct rb_node *n = &node_internal->rb_node;
2067
2068 /* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2069 * n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2070 */
2071 if (READ_ONCE(node_internal->owner) != root)
2072 return NULL;
2073
2074 rb_erase_cached(n, r);
2075 RB_CLEAR_NODE(n);
2076 WRITE_ONCE(node_internal->owner, NULL);
2077 return (struct bpf_rb_node *)n;
2078}
2079
2080/* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2081 * program
2082 */
2083static int __bpf_rbtree_add(struct bpf_rb_root *root,
2084 struct bpf_rb_node_kern *node,
2085 void *less, struct btf_record *rec, u64 off)
2086{
2087 struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2088 struct rb_node *parent = NULL, *n = &node->rb_node;
2089 bpf_callback_t cb = (bpf_callback_t)less;
2090 bool leftmost = true;
2091
2092 /* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2093 * check the latter
2094 */
2095 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2096 /* Only called from BPF prog, no need to migrate_disable */
2097 __bpf_obj_drop_impl((void *)n - off, rec, false);
2098 return -EINVAL;
2099 }
2100
2101 while (*link) {
2102 parent = *link;
2103 if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2104 link = &parent->rb_left;
2105 } else {
2106 link = &parent->rb_right;
2107 leftmost = false;
2108 }
2109 }
2110
2111 rb_link_node(n, parent, link);
2112 rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost);
2113 WRITE_ONCE(node->owner, root);
2114 return 0;
2115}
2116
2117__bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2118 bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2119 void *meta__ign, u64 off)
2120{
2121 struct btf_struct_meta *meta = meta__ign;
2122 struct bpf_rb_node_kern *n = (void *)node;
2123
2124 return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off);
2125}
2126
2127__bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2128{
2129 struct rb_root_cached *r = (struct rb_root_cached *)root;
2130
2131 return (struct bpf_rb_node *)rb_first_cached(r);
2132}
2133
2134/**
2135 * bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2136 * kfunc which is not stored in a map as a kptr, must be released by calling
2137 * bpf_task_release().
2138 * @p: The task on which a reference is being acquired.
2139 */
2140__bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2141{
2142 if (refcount_inc_not_zero(&p->rcu_users))
2143 return p;
2144 return NULL;
2145}
2146
2147/**
2148 * bpf_task_release - Release the reference acquired on a task.
2149 * @p: The task on which a reference is being released.
2150 */
2151__bpf_kfunc void bpf_task_release(struct task_struct *p)
2152{
2153 put_task_struct_rcu_user(p);
2154}
2155
2156__bpf_kfunc void bpf_task_release_dtor(void *p)
2157{
2158 put_task_struct_rcu_user(p);
2159}
2160CFI_NOSEAL(bpf_task_release_dtor);
2161
2162#ifdef CONFIG_CGROUPS
2163/**
2164 * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2165 * this kfunc which is not stored in a map as a kptr, must be released by
2166 * calling bpf_cgroup_release().
2167 * @cgrp: The cgroup on which a reference is being acquired.
2168 */
2169__bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2170{
2171 return cgroup_tryget(cgrp) ? cgrp : NULL;
2172}
2173
2174/**
2175 * bpf_cgroup_release - Release the reference acquired on a cgroup.
2176 * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2177 * not be freed until the current grace period has ended, even if its refcount
2178 * drops to 0.
2179 * @cgrp: The cgroup on which a reference is being released.
2180 */
2181__bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2182{
2183 cgroup_put(cgrp);
2184}
2185
2186__bpf_kfunc void bpf_cgroup_release_dtor(void *cgrp)
2187{
2188 cgroup_put(cgrp);
2189}
2190CFI_NOSEAL(bpf_cgroup_release_dtor);
2191
2192/**
2193 * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2194 * array. A cgroup returned by this kfunc which is not subsequently stored in a
2195 * map, must be released by calling bpf_cgroup_release().
2196 * @cgrp: The cgroup for which we're performing a lookup.
2197 * @level: The level of ancestor to look up.
2198 */
2199__bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2200{
2201 struct cgroup *ancestor;
2202
2203 if (level > cgrp->level || level < 0)
2204 return NULL;
2205
2206 /* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2207 ancestor = cgrp->ancestors[level];
2208 if (!cgroup_tryget(ancestor))
2209 return NULL;
2210 return ancestor;
2211}
2212
2213/**
2214 * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2215 * kfunc which is not subsequently stored in a map, must be released by calling
2216 * bpf_cgroup_release().
2217 * @cgid: cgroup id.
2218 */
2219__bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2220{
2221 struct cgroup *cgrp;
2222
2223 cgrp = cgroup_get_from_id(cgid);
2224 if (IS_ERR(cgrp))
2225 return NULL;
2226 return cgrp;
2227}
2228
2229/**
2230 * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2231 * task's membership of cgroup ancestry.
2232 * @task: the task to be tested
2233 * @ancestor: possible ancestor of @task's cgroup
2234 *
2235 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2236 * It follows all the same rules as cgroup_is_descendant, and only applies
2237 * to the default hierarchy.
2238 */
2239__bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2240 struct cgroup *ancestor)
2241{
2242 long ret;
2243
2244 rcu_read_lock();
2245 ret = task_under_cgroup_hierarchy(task, ancestor);
2246 rcu_read_unlock();
2247 return ret;
2248}
2249
2250/**
2251 * bpf_task_get_cgroup1 - Acquires the associated cgroup of a task within a
2252 * specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
2253 * hierarchy ID.
2254 * @task: The target task
2255 * @hierarchy_id: The ID of a cgroup1 hierarchy
2256 *
2257 * On success, the cgroup is returen. On failure, NULL is returned.
2258 */
2259__bpf_kfunc struct cgroup *
2260bpf_task_get_cgroup1(struct task_struct *task, int hierarchy_id)
2261{
2262 struct cgroup *cgrp = task_get_cgroup1(task, hierarchy_id);
2263
2264 if (IS_ERR(cgrp))
2265 return NULL;
2266 return cgrp;
2267}
2268#endif /* CONFIG_CGROUPS */
2269
2270/**
2271 * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2272 * in the root pid namespace idr. If a task is returned, it must either be
2273 * stored in a map, or released with bpf_task_release().
2274 * @pid: The pid of the task being looked up.
2275 */
2276__bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2277{
2278 struct task_struct *p;
2279
2280 rcu_read_lock();
2281 p = find_task_by_pid_ns(pid, &init_pid_ns);
2282 if (p)
2283 p = bpf_task_acquire(p);
2284 rcu_read_unlock();
2285
2286 return p;
2287}
2288
2289/**
2290 * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2291 * @ptr: The dynptr whose data slice to retrieve
2292 * @offset: Offset into the dynptr
2293 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2294 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2295 * length of the requested slice. This must be a constant.
2296 *
2297 * For non-skb and non-xdp type dynptrs, there is no difference between
2298 * bpf_dynptr_slice and bpf_dynptr_data.
2299 *
2300 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2301 *
2302 * If the intention is to write to the data slice, please use
2303 * bpf_dynptr_slice_rdwr.
2304 *
2305 * The user must check that the returned pointer is not null before using it.
2306 *
2307 * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2308 * does not change the underlying packet data pointers, so a call to
2309 * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2310 * the bpf program.
2311 *
2312 * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2313 * data slice (can be either direct pointer to the data or a pointer to the user
2314 * provided buffer, with its contents containing the data, if unable to obtain
2315 * direct pointer)
2316 */
2317__bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset,
2318 void *buffer__opt, u32 buffer__szk)
2319{
2320 enum bpf_dynptr_type type;
2321 u32 len = buffer__szk;
2322 int err;
2323
2324 if (!ptr->data)
2325 return NULL;
2326
2327 err = bpf_dynptr_check_off_len(ptr, offset, len);
2328 if (err)
2329 return NULL;
2330
2331 type = bpf_dynptr_get_type(ptr);
2332
2333 switch (type) {
2334 case BPF_DYNPTR_TYPE_LOCAL:
2335 case BPF_DYNPTR_TYPE_RINGBUF:
2336 return ptr->data + ptr->offset + offset;
2337 case BPF_DYNPTR_TYPE_SKB:
2338 if (buffer__opt)
2339 return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt);
2340 else
2341 return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len);
2342 case BPF_DYNPTR_TYPE_XDP:
2343 {
2344 void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len);
2345 if (!IS_ERR_OR_NULL(xdp_ptr))
2346 return xdp_ptr;
2347
2348 if (!buffer__opt)
2349 return NULL;
2350 bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false);
2351 return buffer__opt;
2352 }
2353 default:
2354 WARN_ONCE(true, "unknown dynptr type %d\n", type);
2355 return NULL;
2356 }
2357}
2358
2359/**
2360 * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2361 * @ptr: The dynptr whose data slice to retrieve
2362 * @offset: Offset into the dynptr
2363 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2364 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2365 * length of the requested slice. This must be a constant.
2366 *
2367 * For non-skb and non-xdp type dynptrs, there is no difference between
2368 * bpf_dynptr_slice and bpf_dynptr_data.
2369 *
2370 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2371 *
2372 * The returned pointer is writable and may point to either directly the dynptr
2373 * data at the requested offset or to the buffer if unable to obtain a direct
2374 * data pointer to (example: the requested slice is to the paged area of an skb
2375 * packet). In the case where the returned pointer is to the buffer, the user
2376 * is responsible for persisting writes through calling bpf_dynptr_write(). This
2377 * usually looks something like this pattern:
2378 *
2379 * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2380 * if (!eth)
2381 * return TC_ACT_SHOT;
2382 *
2383 * // mutate eth header //
2384 *
2385 * if (eth == buffer)
2386 * bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2387 *
2388 * Please note that, as in the example above, the user must check that the
2389 * returned pointer is not null before using it.
2390 *
2391 * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2392 * does not change the underlying packet data pointers, so a call to
2393 * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2394 * the bpf program.
2395 *
2396 * Return: NULL if the call failed (eg invalid dynptr), pointer to a
2397 * data slice (can be either direct pointer to the data or a pointer to the user
2398 * provided buffer, with its contents containing the data, if unable to obtain
2399 * direct pointer)
2400 */
2401__bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset,
2402 void *buffer__opt, u32 buffer__szk)
2403{
2404 if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2405 return NULL;
2406
2407 /* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2408 *
2409 * For skb-type dynptrs, it is safe to write into the returned pointer
2410 * if the bpf program allows skb data writes. There are two possiblities
2411 * that may occur when calling bpf_dynptr_slice_rdwr:
2412 *
2413 * 1) The requested slice is in the head of the skb. In this case, the
2414 * returned pointer is directly to skb data, and if the skb is cloned, the
2415 * verifier will have uncloned it (see bpf_unclone_prologue()) already.
2416 * The pointer can be directly written into.
2417 *
2418 * 2) Some portion of the requested slice is in the paged buffer area.
2419 * In this case, the requested data will be copied out into the buffer
2420 * and the returned pointer will be a pointer to the buffer. The skb
2421 * will not be pulled. To persist the write, the user will need to call
2422 * bpf_dynptr_write(), which will pull the skb and commit the write.
2423 *
2424 * Similarly for xdp programs, if the requested slice is not across xdp
2425 * fragments, then a direct pointer will be returned, otherwise the data
2426 * will be copied out into the buffer and the user will need to call
2427 * bpf_dynptr_write() to commit changes.
2428 */
2429 return bpf_dynptr_slice(ptr, offset, buffer__opt, buffer__szk);
2430}
2431
2432__bpf_kfunc int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end)
2433{
2434 u32 size;
2435
2436 if (!ptr->data || start > end)
2437 return -EINVAL;
2438
2439 size = __bpf_dynptr_size(ptr);
2440
2441 if (start > size || end > size)
2442 return -ERANGE;
2443
2444 ptr->offset += start;
2445 bpf_dynptr_set_size(ptr, end - start);
2446
2447 return 0;
2448}
2449
2450__bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr)
2451{
2452 return !ptr->data;
2453}
2454
2455__bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr)
2456{
2457 if (!ptr->data)
2458 return false;
2459
2460 return __bpf_dynptr_is_rdonly(ptr);
2461}
2462
2463__bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
2464{
2465 if (!ptr->data)
2466 return -EINVAL;
2467
2468 return __bpf_dynptr_size(ptr);
2469}
2470
2471__bpf_kfunc int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr,
2472 struct bpf_dynptr_kern *clone__uninit)
2473{
2474 if (!ptr->data) {
2475 bpf_dynptr_set_null(clone__uninit);
2476 return -EINVAL;
2477 }
2478
2479 *clone__uninit = *ptr;
2480
2481 return 0;
2482}
2483
2484__bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2485{
2486 return obj;
2487}
2488
2489__bpf_kfunc void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k)
2490{
2491 return obj__ign;
2492}
2493
2494__bpf_kfunc void bpf_rcu_read_lock(void)
2495{
2496 rcu_read_lock();
2497}
2498
2499__bpf_kfunc void bpf_rcu_read_unlock(void)
2500{
2501 rcu_read_unlock();
2502}
2503
2504struct bpf_throw_ctx {
2505 struct bpf_prog_aux *aux;
2506 u64 sp;
2507 u64 bp;
2508 int cnt;
2509};
2510
2511static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
2512{
2513 struct bpf_throw_ctx *ctx = cookie;
2514 struct bpf_prog *prog;
2515
2516 if (!is_bpf_text_address(ip))
2517 return !ctx->cnt;
2518 prog = bpf_prog_ksym_find(ip);
2519 ctx->cnt++;
2520 if (bpf_is_subprog(prog))
2521 return true;
2522 ctx->aux = prog->aux;
2523 ctx->sp = sp;
2524 ctx->bp = bp;
2525 return false;
2526}
2527
2528__bpf_kfunc void bpf_throw(u64 cookie)
2529{
2530 struct bpf_throw_ctx ctx = {};
2531
2532 arch_bpf_stack_walk(bpf_stack_walker, &ctx);
2533 WARN_ON_ONCE(!ctx.aux);
2534 if (ctx.aux)
2535 WARN_ON_ONCE(!ctx.aux->exception_boundary);
2536 WARN_ON_ONCE(!ctx.bp);
2537 WARN_ON_ONCE(!ctx.cnt);
2538 /* Prevent KASAN false positives for CONFIG_KASAN_STACK by unpoisoning
2539 * deeper stack depths than ctx.sp as we do not return from bpf_throw,
2540 * which skips compiler generated instrumentation to do the same.
2541 */
2542 kasan_unpoison_task_stack_below((void *)(long)ctx.sp);
2543 ctx.aux->bpf_exception_cb(cookie, ctx.sp, ctx.bp, 0, 0);
2544 WARN(1, "A call to BPF exception callback should never return\n");
2545}
2546
2547__bpf_kfunc_end_defs();
2548
2549BTF_SET8_START(generic_btf_ids)
2550#ifdef CONFIG_KEXEC_CORE
2551BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
2552#endif
2553BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2554BTF_ID_FLAGS(func, bpf_percpu_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2555BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
2556BTF_ID_FLAGS(func, bpf_percpu_obj_drop_impl, KF_RELEASE)
2557BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL | KF_RCU)
2558BTF_ID_FLAGS(func, bpf_list_push_front_impl)
2559BTF_ID_FLAGS(func, bpf_list_push_back_impl)
2560BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
2561BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
2562BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2563BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
2564BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
2565BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
2566BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
2567
2568#ifdef CONFIG_CGROUPS
2569BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2570BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
2571BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2572BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
2573BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
2574BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2575#endif
2576BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
2577BTF_ID_FLAGS(func, bpf_throw)
2578BTF_SET8_END(generic_btf_ids)
2579
2580static const struct btf_kfunc_id_set generic_kfunc_set = {
2581 .owner = THIS_MODULE,
2582 .set = &generic_btf_ids,
2583};
2584
2585
2586BTF_ID_LIST(generic_dtor_ids)
2587BTF_ID(struct, task_struct)
2588BTF_ID(func, bpf_task_release_dtor)
2589#ifdef CONFIG_CGROUPS
2590BTF_ID(struct, cgroup)
2591BTF_ID(func, bpf_cgroup_release_dtor)
2592#endif
2593
2594BTF_SET8_START(common_btf_ids)
2595BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx)
2596BTF_ID_FLAGS(func, bpf_rdonly_cast)
2597BTF_ID_FLAGS(func, bpf_rcu_read_lock)
2598BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
2599BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
2600BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
2601BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
2602BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
2603BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
2604BTF_ID_FLAGS(func, bpf_iter_task_vma_new, KF_ITER_NEW | KF_RCU)
2605BTF_ID_FLAGS(func, bpf_iter_task_vma_next, KF_ITER_NEXT | KF_RET_NULL)
2606BTF_ID_FLAGS(func, bpf_iter_task_vma_destroy, KF_ITER_DESTROY)
2607#ifdef CONFIG_CGROUPS
2608BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS)
2609BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
2610BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
2611BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2612BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
2613BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
2614#endif
2615BTF_ID_FLAGS(func, bpf_iter_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2616BTF_ID_FLAGS(func, bpf_iter_task_next, KF_ITER_NEXT | KF_RET_NULL)
2617BTF_ID_FLAGS(func, bpf_iter_task_destroy, KF_ITER_DESTROY)
2618BTF_ID_FLAGS(func, bpf_dynptr_adjust)
2619BTF_ID_FLAGS(func, bpf_dynptr_is_null)
2620BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
2621BTF_ID_FLAGS(func, bpf_dynptr_size)
2622BTF_ID_FLAGS(func, bpf_dynptr_clone)
2623BTF_SET8_END(common_btf_ids)
2624
2625static const struct btf_kfunc_id_set common_kfunc_set = {
2626 .owner = THIS_MODULE,
2627 .set = &common_btf_ids,
2628};
2629
2630static int __init kfunc_init(void)
2631{
2632 int ret;
2633 const struct btf_id_dtor_kfunc generic_dtors[] = {
2634 {
2635 .btf_id = generic_dtor_ids[0],
2636 .kfunc_btf_id = generic_dtor_ids[1]
2637 },
2638#ifdef CONFIG_CGROUPS
2639 {
2640 .btf_id = generic_dtor_ids[2],
2641 .kfunc_btf_id = generic_dtor_ids[3]
2642 },
2643#endif
2644 };
2645
2646 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set);
2647 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set);
2648 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &generic_kfunc_set);
2649 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set);
2650 ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors,
2651 ARRAY_SIZE(generic_dtors),
2652 THIS_MODULE);
2653 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set);
2654}
2655
2656late_initcall(kfunc_init);
2657
2658/* Get a pointer to dynptr data up to len bytes for read only access. If
2659 * the dynptr doesn't have continuous data up to len bytes, return NULL.
2660 */
2661const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len)
2662{
2663 return bpf_dynptr_slice(ptr, 0, NULL, len);
2664}
2665
2666/* Get a pointer to dynptr data up to len bytes for read write access. If
2667 * the dynptr doesn't have continuous data up to len bytes, or the dynptr
2668 * is read only, return NULL.
2669 */
2670void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len)
2671{
2672 if (__bpf_dynptr_is_rdonly(ptr))
2673 return NULL;
2674 return (void *)__bpf_dynptr_data(ptr, len);
2675}
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#include <linux/kasan.h>
26
27#include "../../lib/kstrtox.h"
28
29/* If kernel subsystem is allowing eBPF programs to call this function,
30 * inside its own verifier_ops->get_func_proto() callback it should return
31 * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
32 *
33 * Different map implementations will rely on rcu in map methods
34 * lookup/update/delete, therefore eBPF programs must run under rcu lock
35 * if program is allowed to access maps, so check rcu_read_lock_held() or
36 * rcu_read_lock_trace_held() in all three functions.
37 */
38BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
39{
40 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
41 !rcu_read_lock_bh_held());
42 return (unsigned long) map->ops->map_lookup_elem(map, key);
43}
44
45const struct bpf_func_proto bpf_map_lookup_elem_proto = {
46 .func = bpf_map_lookup_elem,
47 .gpl_only = false,
48 .pkt_access = true,
49 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
50 .arg1_type = ARG_CONST_MAP_PTR,
51 .arg2_type = ARG_PTR_TO_MAP_KEY,
52};
53
54BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
55 void *, value, u64, flags)
56{
57 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
58 !rcu_read_lock_bh_held());
59 return map->ops->map_update_elem(map, key, value, flags);
60}
61
62const struct bpf_func_proto bpf_map_update_elem_proto = {
63 .func = bpf_map_update_elem,
64 .gpl_only = false,
65 .pkt_access = true,
66 .ret_type = RET_INTEGER,
67 .arg1_type = ARG_CONST_MAP_PTR,
68 .arg2_type = ARG_PTR_TO_MAP_KEY,
69 .arg3_type = ARG_PTR_TO_MAP_VALUE,
70 .arg4_type = ARG_ANYTHING,
71};
72
73BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
74{
75 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_trace_held() &&
76 !rcu_read_lock_bh_held());
77 return map->ops->map_delete_elem(map, key);
78}
79
80const struct bpf_func_proto bpf_map_delete_elem_proto = {
81 .func = bpf_map_delete_elem,
82 .gpl_only = false,
83 .pkt_access = true,
84 .ret_type = RET_INTEGER,
85 .arg1_type = ARG_CONST_MAP_PTR,
86 .arg2_type = ARG_PTR_TO_MAP_KEY,
87};
88
89BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
90{
91 return map->ops->map_push_elem(map, value, flags);
92}
93
94const struct bpf_func_proto bpf_map_push_elem_proto = {
95 .func = bpf_map_push_elem,
96 .gpl_only = false,
97 .pkt_access = true,
98 .ret_type = RET_INTEGER,
99 .arg1_type = ARG_CONST_MAP_PTR,
100 .arg2_type = ARG_PTR_TO_MAP_VALUE,
101 .arg3_type = ARG_ANYTHING,
102};
103
104BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
105{
106 return map->ops->map_pop_elem(map, value);
107}
108
109const struct bpf_func_proto bpf_map_pop_elem_proto = {
110 .func = bpf_map_pop_elem,
111 .gpl_only = false,
112 .ret_type = RET_INTEGER,
113 .arg1_type = ARG_CONST_MAP_PTR,
114 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT | MEM_WRITE,
115};
116
117BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
118{
119 return map->ops->map_peek_elem(map, value);
120}
121
122const struct bpf_func_proto bpf_map_peek_elem_proto = {
123 .func = bpf_map_peek_elem,
124 .gpl_only = false,
125 .ret_type = RET_INTEGER,
126 .arg1_type = ARG_CONST_MAP_PTR,
127 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT | MEM_WRITE,
128};
129
130BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
131{
132 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
133 return (unsigned long) map->ops->map_lookup_percpu_elem(map, key, cpu);
134}
135
136const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
137 .func = bpf_map_lookup_percpu_elem,
138 .gpl_only = false,
139 .pkt_access = true,
140 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
141 .arg1_type = ARG_CONST_MAP_PTR,
142 .arg2_type = ARG_PTR_TO_MAP_KEY,
143 .arg3_type = ARG_ANYTHING,
144};
145
146const struct bpf_func_proto bpf_get_prandom_u32_proto = {
147 .func = bpf_user_rnd_u32,
148 .gpl_only = false,
149 .ret_type = RET_INTEGER,
150};
151
152BPF_CALL_0(bpf_get_smp_processor_id)
153{
154 return smp_processor_id();
155}
156
157const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
158 .func = bpf_get_smp_processor_id,
159 .gpl_only = false,
160 .ret_type = RET_INTEGER,
161 .allow_fastcall = true,
162};
163
164BPF_CALL_0(bpf_get_numa_node_id)
165{
166 return numa_node_id();
167}
168
169const struct bpf_func_proto bpf_get_numa_node_id_proto = {
170 .func = bpf_get_numa_node_id,
171 .gpl_only = false,
172 .ret_type = RET_INTEGER,
173};
174
175BPF_CALL_0(bpf_ktime_get_ns)
176{
177 /* NMI safe access to clock monotonic */
178 return ktime_get_mono_fast_ns();
179}
180
181const struct bpf_func_proto bpf_ktime_get_ns_proto = {
182 .func = bpf_ktime_get_ns,
183 .gpl_only = false,
184 .ret_type = RET_INTEGER,
185};
186
187BPF_CALL_0(bpf_ktime_get_boot_ns)
188{
189 /* NMI safe access to clock boottime */
190 return ktime_get_boot_fast_ns();
191}
192
193const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
194 .func = bpf_ktime_get_boot_ns,
195 .gpl_only = false,
196 .ret_type = RET_INTEGER,
197};
198
199BPF_CALL_0(bpf_ktime_get_coarse_ns)
200{
201 return ktime_get_coarse_ns();
202}
203
204const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
205 .func = bpf_ktime_get_coarse_ns,
206 .gpl_only = false,
207 .ret_type = RET_INTEGER,
208};
209
210BPF_CALL_0(bpf_ktime_get_tai_ns)
211{
212 /* NMI safe access to clock tai */
213 return ktime_get_tai_fast_ns();
214}
215
216const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
217 .func = bpf_ktime_get_tai_ns,
218 .gpl_only = false,
219 .ret_type = RET_INTEGER,
220};
221
222BPF_CALL_0(bpf_get_current_pid_tgid)
223{
224 struct task_struct *task = current;
225
226 if (unlikely(!task))
227 return -EINVAL;
228
229 return (u64) task->tgid << 32 | task->pid;
230}
231
232const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
233 .func = bpf_get_current_pid_tgid,
234 .gpl_only = false,
235 .ret_type = RET_INTEGER,
236};
237
238BPF_CALL_0(bpf_get_current_uid_gid)
239{
240 struct task_struct *task = current;
241 kuid_t uid;
242 kgid_t gid;
243
244 if (unlikely(!task))
245 return -EINVAL;
246
247 current_uid_gid(&uid, &gid);
248 return (u64) from_kgid(&init_user_ns, gid) << 32 |
249 from_kuid(&init_user_ns, uid);
250}
251
252const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
253 .func = bpf_get_current_uid_gid,
254 .gpl_only = false,
255 .ret_type = RET_INTEGER,
256};
257
258BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
259{
260 struct task_struct *task = current;
261
262 if (unlikely(!task))
263 goto err_clear;
264
265 /* Verifier guarantees that size > 0 */
266 strscpy_pad(buf, task->comm, size);
267 return 0;
268err_clear:
269 memset(buf, 0, size);
270 return -EINVAL;
271}
272
273const struct bpf_func_proto bpf_get_current_comm_proto = {
274 .func = bpf_get_current_comm,
275 .gpl_only = false,
276 .ret_type = RET_INTEGER,
277 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
278 .arg2_type = ARG_CONST_SIZE,
279};
280
281#if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
282
283static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
284{
285 arch_spinlock_t *l = (void *)lock;
286 union {
287 __u32 val;
288 arch_spinlock_t lock;
289 } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
290
291 compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
292 BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
293 BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
294 preempt_disable();
295 arch_spin_lock(l);
296}
297
298static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
299{
300 arch_spinlock_t *l = (void *)lock;
301
302 arch_spin_unlock(l);
303 preempt_enable();
304}
305
306#else
307
308static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
309{
310 atomic_t *l = (void *)lock;
311
312 BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
313 do {
314 atomic_cond_read_relaxed(l, !VAL);
315 } while (atomic_xchg(l, 1));
316}
317
318static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
319{
320 atomic_t *l = (void *)lock;
321
322 atomic_set_release(l, 0);
323}
324
325#endif
326
327static DEFINE_PER_CPU(unsigned long, irqsave_flags);
328
329static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
330{
331 unsigned long flags;
332
333 local_irq_save(flags);
334 __bpf_spin_lock(lock);
335 __this_cpu_write(irqsave_flags, flags);
336}
337
338NOTRACE_BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
339{
340 __bpf_spin_lock_irqsave(lock);
341 return 0;
342}
343
344const struct bpf_func_proto bpf_spin_lock_proto = {
345 .func = bpf_spin_lock,
346 .gpl_only = false,
347 .ret_type = RET_VOID,
348 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
349 .arg1_btf_id = BPF_PTR_POISON,
350};
351
352static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
353{
354 unsigned long flags;
355
356 flags = __this_cpu_read(irqsave_flags);
357 __bpf_spin_unlock(lock);
358 local_irq_restore(flags);
359}
360
361NOTRACE_BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
362{
363 __bpf_spin_unlock_irqrestore(lock);
364 return 0;
365}
366
367const struct bpf_func_proto bpf_spin_unlock_proto = {
368 .func = bpf_spin_unlock,
369 .gpl_only = false,
370 .ret_type = RET_VOID,
371 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
372 .arg1_btf_id = BPF_PTR_POISON,
373};
374
375void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
376 bool lock_src)
377{
378 struct bpf_spin_lock *lock;
379
380 if (lock_src)
381 lock = src + map->record->spin_lock_off;
382 else
383 lock = dst + map->record->spin_lock_off;
384 preempt_disable();
385 __bpf_spin_lock_irqsave(lock);
386 copy_map_value(map, dst, src);
387 __bpf_spin_unlock_irqrestore(lock);
388 preempt_enable();
389}
390
391BPF_CALL_0(bpf_jiffies64)
392{
393 return get_jiffies_64();
394}
395
396const struct bpf_func_proto bpf_jiffies64_proto = {
397 .func = bpf_jiffies64,
398 .gpl_only = false,
399 .ret_type = RET_INTEGER,
400};
401
402#ifdef CONFIG_CGROUPS
403BPF_CALL_0(bpf_get_current_cgroup_id)
404{
405 struct cgroup *cgrp;
406 u64 cgrp_id;
407
408 rcu_read_lock();
409 cgrp = task_dfl_cgroup(current);
410 cgrp_id = cgroup_id(cgrp);
411 rcu_read_unlock();
412
413 return cgrp_id;
414}
415
416const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
417 .func = bpf_get_current_cgroup_id,
418 .gpl_only = false,
419 .ret_type = RET_INTEGER,
420};
421
422BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
423{
424 struct cgroup *cgrp;
425 struct cgroup *ancestor;
426 u64 cgrp_id;
427
428 rcu_read_lock();
429 cgrp = task_dfl_cgroup(current);
430 ancestor = cgroup_ancestor(cgrp, ancestor_level);
431 cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
432 rcu_read_unlock();
433
434 return cgrp_id;
435}
436
437const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
438 .func = bpf_get_current_ancestor_cgroup_id,
439 .gpl_only = false,
440 .ret_type = RET_INTEGER,
441 .arg1_type = ARG_ANYTHING,
442};
443#endif /* CONFIG_CGROUPS */
444
445#define BPF_STRTOX_BASE_MASK 0x1F
446
447static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
448 unsigned long long *res, bool *is_negative)
449{
450 unsigned int base = flags & BPF_STRTOX_BASE_MASK;
451 const char *cur_buf = buf;
452 size_t cur_len = buf_len;
453 unsigned int consumed;
454 size_t val_len;
455 char str[64];
456
457 if (!buf || !buf_len || !res || !is_negative)
458 return -EINVAL;
459
460 if (base != 0 && base != 8 && base != 10 && base != 16)
461 return -EINVAL;
462
463 if (flags & ~BPF_STRTOX_BASE_MASK)
464 return -EINVAL;
465
466 while (cur_buf < buf + buf_len && isspace(*cur_buf))
467 ++cur_buf;
468
469 *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
470 if (*is_negative)
471 ++cur_buf;
472
473 consumed = cur_buf - buf;
474 cur_len -= consumed;
475 if (!cur_len)
476 return -EINVAL;
477
478 cur_len = min(cur_len, sizeof(str) - 1);
479 memcpy(str, cur_buf, cur_len);
480 str[cur_len] = '\0';
481 cur_buf = str;
482
483 cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
484 val_len = _parse_integer(cur_buf, base, res);
485
486 if (val_len & KSTRTOX_OVERFLOW)
487 return -ERANGE;
488
489 if (val_len == 0)
490 return -EINVAL;
491
492 cur_buf += val_len;
493 consumed += cur_buf - str;
494
495 return consumed;
496}
497
498static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
499 long long *res)
500{
501 unsigned long long _res;
502 bool is_negative;
503 int err;
504
505 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
506 if (err < 0)
507 return err;
508 if (is_negative) {
509 if ((long long)-_res > 0)
510 return -ERANGE;
511 *res = -_res;
512 } else {
513 if ((long long)_res < 0)
514 return -ERANGE;
515 *res = _res;
516 }
517 return err;
518}
519
520BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
521 s64 *, res)
522{
523 long long _res;
524 int err;
525
526 *res = 0;
527 err = __bpf_strtoll(buf, buf_len, flags, &_res);
528 if (err < 0)
529 return err;
530 *res = _res;
531 return err;
532}
533
534const struct bpf_func_proto bpf_strtol_proto = {
535 .func = bpf_strtol,
536 .gpl_only = false,
537 .ret_type = RET_INTEGER,
538 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
539 .arg2_type = ARG_CONST_SIZE,
540 .arg3_type = ARG_ANYTHING,
541 .arg4_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
542 .arg4_size = sizeof(s64),
543};
544
545BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
546 u64 *, res)
547{
548 unsigned long long _res;
549 bool is_negative;
550 int err;
551
552 *res = 0;
553 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
554 if (err < 0)
555 return err;
556 if (is_negative)
557 return -EINVAL;
558 *res = _res;
559 return err;
560}
561
562const struct bpf_func_proto bpf_strtoul_proto = {
563 .func = bpf_strtoul,
564 .gpl_only = false,
565 .ret_type = RET_INTEGER,
566 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
567 .arg2_type = ARG_CONST_SIZE,
568 .arg3_type = ARG_ANYTHING,
569 .arg4_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
570 .arg4_size = sizeof(u64),
571};
572
573BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
574{
575 return strncmp(s1, s2, s1_sz);
576}
577
578static const struct bpf_func_proto bpf_strncmp_proto = {
579 .func = bpf_strncmp,
580 .gpl_only = false,
581 .ret_type = RET_INTEGER,
582 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
583 .arg2_type = ARG_CONST_SIZE,
584 .arg3_type = ARG_PTR_TO_CONST_STR,
585};
586
587BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
588 struct bpf_pidns_info *, nsdata, u32, size)
589{
590 struct task_struct *task = current;
591 struct pid_namespace *pidns;
592 int err = -EINVAL;
593
594 if (unlikely(size != sizeof(struct bpf_pidns_info)))
595 goto clear;
596
597 if (unlikely((u64)(dev_t)dev != dev))
598 goto clear;
599
600 if (unlikely(!task))
601 goto clear;
602
603 pidns = task_active_pid_ns(task);
604 if (unlikely(!pidns)) {
605 err = -ENOENT;
606 goto clear;
607 }
608
609 if (!ns_match(&pidns->ns, (dev_t)dev, ino))
610 goto clear;
611
612 nsdata->pid = task_pid_nr_ns(task, pidns);
613 nsdata->tgid = task_tgid_nr_ns(task, pidns);
614 return 0;
615clear:
616 memset((void *)nsdata, 0, (size_t) size);
617 return err;
618}
619
620const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
621 .func = bpf_get_ns_current_pid_tgid,
622 .gpl_only = false,
623 .ret_type = RET_INTEGER,
624 .arg1_type = ARG_ANYTHING,
625 .arg2_type = ARG_ANYTHING,
626 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
627 .arg4_type = ARG_CONST_SIZE,
628};
629
630static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
631 .func = bpf_get_raw_cpu_id,
632 .gpl_only = false,
633 .ret_type = RET_INTEGER,
634};
635
636BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
637 u64, flags, void *, data, u64, size)
638{
639 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
640 return -EINVAL;
641
642 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
643}
644
645const struct bpf_func_proto bpf_event_output_data_proto = {
646 .func = bpf_event_output_data,
647 .gpl_only = true,
648 .ret_type = RET_INTEGER,
649 .arg1_type = ARG_PTR_TO_CTX,
650 .arg2_type = ARG_CONST_MAP_PTR,
651 .arg3_type = ARG_ANYTHING,
652 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
653 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
654};
655
656BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
657 const void __user *, user_ptr)
658{
659 int ret = copy_from_user(dst, user_ptr, size);
660
661 if (unlikely(ret)) {
662 memset(dst, 0, size);
663 ret = -EFAULT;
664 }
665
666 return ret;
667}
668
669const struct bpf_func_proto bpf_copy_from_user_proto = {
670 .func = bpf_copy_from_user,
671 .gpl_only = false,
672 .might_sleep = true,
673 .ret_type = RET_INTEGER,
674 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
675 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
676 .arg3_type = ARG_ANYTHING,
677};
678
679BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
680 const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
681{
682 int ret;
683
684 /* flags is not used yet */
685 if (unlikely(flags))
686 return -EINVAL;
687
688 if (unlikely(!size))
689 return 0;
690
691 ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
692 if (ret == size)
693 return 0;
694
695 memset(dst, 0, size);
696 /* Return -EFAULT for partial read */
697 return ret < 0 ? ret : -EFAULT;
698}
699
700const struct bpf_func_proto bpf_copy_from_user_task_proto = {
701 .func = bpf_copy_from_user_task,
702 .gpl_only = true,
703 .might_sleep = true,
704 .ret_type = RET_INTEGER,
705 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
706 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
707 .arg3_type = ARG_ANYTHING,
708 .arg4_type = ARG_PTR_TO_BTF_ID,
709 .arg4_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
710 .arg5_type = ARG_ANYTHING
711};
712
713BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
714{
715 if (cpu >= nr_cpu_ids)
716 return (unsigned long)NULL;
717
718 return (unsigned long)per_cpu_ptr((const void __percpu *)(const uintptr_t)ptr, cpu);
719}
720
721const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
722 .func = bpf_per_cpu_ptr,
723 .gpl_only = false,
724 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | PTR_MAYBE_NULL | MEM_RDONLY,
725 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
726 .arg2_type = ARG_ANYTHING,
727};
728
729BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
730{
731 return (unsigned long)this_cpu_ptr((const void __percpu *)(const uintptr_t)percpu_ptr);
732}
733
734const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
735 .func = bpf_this_cpu_ptr,
736 .gpl_only = false,
737 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
738 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
739};
740
741static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
742 size_t bufsz)
743{
744 void __user *user_ptr = (__force void __user *)unsafe_ptr;
745
746 buf[0] = 0;
747
748 switch (fmt_ptype) {
749 case 's':
750#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
751 if ((unsigned long)unsafe_ptr < TASK_SIZE)
752 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
753 fallthrough;
754#endif
755 case 'k':
756 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
757 case 'u':
758 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
759 }
760
761 return -EINVAL;
762}
763
764/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
765 * arguments representation.
766 */
767#define MAX_BPRINTF_BIN_ARGS 512
768
769/* Support executing three nested bprintf helper calls on a given CPU */
770#define MAX_BPRINTF_NEST_LEVEL 3
771struct bpf_bprintf_buffers {
772 char bin_args[MAX_BPRINTF_BIN_ARGS];
773 char buf[MAX_BPRINTF_BUF];
774};
775
776static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
777static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
778
779static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
780{
781 int nest_level;
782
783 preempt_disable();
784 nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
785 if (WARN_ON_ONCE(nest_level > MAX_BPRINTF_NEST_LEVEL)) {
786 this_cpu_dec(bpf_bprintf_nest_level);
787 preempt_enable();
788 return -EBUSY;
789 }
790 *bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
791
792 return 0;
793}
794
795void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
796{
797 if (!data->bin_args && !data->buf)
798 return;
799 if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
800 return;
801 this_cpu_dec(bpf_bprintf_nest_level);
802 preempt_enable();
803}
804
805/*
806 * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
807 *
808 * Returns a negative value if fmt is an invalid format string or 0 otherwise.
809 *
810 * This can be used in two ways:
811 * - Format string verification only: when data->get_bin_args is false
812 * - Arguments preparation: in addition to the above verification, it writes in
813 * data->bin_args a binary representation of arguments usable by bstr_printf
814 * where pointers from BPF have been sanitized.
815 *
816 * In argument preparation mode, if 0 is returned, safe temporary buffers are
817 * allocated and bpf_bprintf_cleanup should be called to free them after use.
818 */
819int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
820 u32 num_args, struct bpf_bprintf_data *data)
821{
822 bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
823 char *unsafe_ptr = NULL, *tmp_buf = NULL, *tmp_buf_end, *fmt_end;
824 struct bpf_bprintf_buffers *buffers = NULL;
825 size_t sizeof_cur_arg, sizeof_cur_ip;
826 int err, i, num_spec = 0;
827 u64 cur_arg;
828 char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
829
830 fmt_end = strnchr(fmt, fmt_size, 0);
831 if (!fmt_end)
832 return -EINVAL;
833 fmt_size = fmt_end - fmt;
834
835 if (get_buffers && try_get_buffers(&buffers))
836 return -EBUSY;
837
838 if (data->get_bin_args) {
839 if (num_args)
840 tmp_buf = buffers->bin_args;
841 tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
842 data->bin_args = (u32 *)tmp_buf;
843 }
844
845 if (data->get_buf)
846 data->buf = buffers->buf;
847
848 for (i = 0; i < fmt_size; i++) {
849 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
850 err = -EINVAL;
851 goto out;
852 }
853
854 if (fmt[i] != '%')
855 continue;
856
857 if (fmt[i + 1] == '%') {
858 i++;
859 continue;
860 }
861
862 if (num_spec >= num_args) {
863 err = -EINVAL;
864 goto out;
865 }
866
867 /* The string is zero-terminated so if fmt[i] != 0, we can
868 * always access fmt[i + 1], in the worst case it will be a 0
869 */
870 i++;
871
872 /* skip optional "[0 +-][num]" width formatting field */
873 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
874 fmt[i] == ' ')
875 i++;
876 if (fmt[i] >= '1' && fmt[i] <= '9') {
877 i++;
878 while (fmt[i] >= '0' && fmt[i] <= '9')
879 i++;
880 }
881
882 if (fmt[i] == 'p') {
883 sizeof_cur_arg = sizeof(long);
884
885 if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
886 fmt[i + 2] == 's') {
887 fmt_ptype = fmt[i + 1];
888 i += 2;
889 goto fmt_str;
890 }
891
892 if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
893 ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
894 fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
895 fmt[i + 1] == 'S') {
896 /* just kernel pointers */
897 if (tmp_buf)
898 cur_arg = raw_args[num_spec];
899 i++;
900 goto nocopy_fmt;
901 }
902
903 if (fmt[i + 1] == 'B') {
904 if (tmp_buf) {
905 err = snprintf(tmp_buf,
906 (tmp_buf_end - tmp_buf),
907 "%pB",
908 (void *)(long)raw_args[num_spec]);
909 tmp_buf += (err + 1);
910 }
911
912 i++;
913 num_spec++;
914 continue;
915 }
916
917 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
918 if ((fmt[i + 1] != 'i' && fmt[i + 1] != 'I') ||
919 (fmt[i + 2] != '4' && fmt[i + 2] != '6')) {
920 err = -EINVAL;
921 goto out;
922 }
923
924 i += 2;
925 if (!tmp_buf)
926 goto nocopy_fmt;
927
928 sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
929 if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
930 err = -ENOSPC;
931 goto out;
932 }
933
934 unsafe_ptr = (char *)(long)raw_args[num_spec];
935 err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
936 sizeof_cur_ip);
937 if (err < 0)
938 memset(cur_ip, 0, sizeof_cur_ip);
939
940 /* hack: bstr_printf expects IP addresses to be
941 * pre-formatted as strings, ironically, the easiest way
942 * to do that is to call snprintf.
943 */
944 ip_spec[2] = fmt[i - 1];
945 ip_spec[3] = fmt[i];
946 err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
947 ip_spec, &cur_ip);
948
949 tmp_buf += err + 1;
950 num_spec++;
951
952 continue;
953 } else if (fmt[i] == 's') {
954 fmt_ptype = fmt[i];
955fmt_str:
956 if (fmt[i + 1] != 0 &&
957 !isspace(fmt[i + 1]) &&
958 !ispunct(fmt[i + 1])) {
959 err = -EINVAL;
960 goto out;
961 }
962
963 if (!tmp_buf)
964 goto nocopy_fmt;
965
966 if (tmp_buf_end == tmp_buf) {
967 err = -ENOSPC;
968 goto out;
969 }
970
971 unsafe_ptr = (char *)(long)raw_args[num_spec];
972 err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
973 fmt_ptype,
974 tmp_buf_end - tmp_buf);
975 if (err < 0) {
976 tmp_buf[0] = '\0';
977 err = 1;
978 }
979
980 tmp_buf += err;
981 num_spec++;
982
983 continue;
984 } else if (fmt[i] == 'c') {
985 if (!tmp_buf)
986 goto nocopy_fmt;
987
988 if (tmp_buf_end == tmp_buf) {
989 err = -ENOSPC;
990 goto out;
991 }
992
993 *tmp_buf = raw_args[num_spec];
994 tmp_buf++;
995 num_spec++;
996
997 continue;
998 }
999
1000 sizeof_cur_arg = sizeof(int);
1001
1002 if (fmt[i] == 'l') {
1003 sizeof_cur_arg = sizeof(long);
1004 i++;
1005 }
1006 if (fmt[i] == 'l') {
1007 sizeof_cur_arg = sizeof(long long);
1008 i++;
1009 }
1010
1011 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1012 fmt[i] != 'x' && fmt[i] != 'X') {
1013 err = -EINVAL;
1014 goto out;
1015 }
1016
1017 if (tmp_buf)
1018 cur_arg = raw_args[num_spec];
1019nocopy_fmt:
1020 if (tmp_buf) {
1021 tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1022 if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1023 err = -ENOSPC;
1024 goto out;
1025 }
1026
1027 if (sizeof_cur_arg == 8) {
1028 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
1029 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1030 } else {
1031 *(u32 *)tmp_buf = (u32)(long)cur_arg;
1032 }
1033 tmp_buf += sizeof_cur_arg;
1034 }
1035 num_spec++;
1036 }
1037
1038 err = 0;
1039out:
1040 if (err)
1041 bpf_bprintf_cleanup(data);
1042 return err;
1043}
1044
1045BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1046 const void *, args, u32, data_len)
1047{
1048 struct bpf_bprintf_data data = {
1049 .get_bin_args = true,
1050 };
1051 int err, num_args;
1052
1053 if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1054 (data_len && !args))
1055 return -EINVAL;
1056 num_args = data_len / 8;
1057
1058 /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1059 * can safely give an unbounded size.
1060 */
1061 err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1062 if (err < 0)
1063 return err;
1064
1065 err = bstr_printf(str, str_size, fmt, data.bin_args);
1066
1067 bpf_bprintf_cleanup(&data);
1068
1069 return err + 1;
1070}
1071
1072const struct bpf_func_proto bpf_snprintf_proto = {
1073 .func = bpf_snprintf,
1074 .gpl_only = true,
1075 .ret_type = RET_INTEGER,
1076 .arg1_type = ARG_PTR_TO_MEM_OR_NULL,
1077 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1078 .arg3_type = ARG_PTR_TO_CONST_STR,
1079 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
1080 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1081};
1082
1083struct bpf_async_cb {
1084 struct bpf_map *map;
1085 struct bpf_prog *prog;
1086 void __rcu *callback_fn;
1087 void *value;
1088 union {
1089 struct rcu_head rcu;
1090 struct work_struct delete_work;
1091 };
1092 u64 flags;
1093};
1094
1095/* BPF map elements can contain 'struct bpf_timer'.
1096 * Such map owns all of its BPF timers.
1097 * 'struct bpf_timer' is allocated as part of map element allocation
1098 * and it's zero initialized.
1099 * That space is used to keep 'struct bpf_async_kern'.
1100 * bpf_timer_init() allocates 'struct bpf_hrtimer', inits hrtimer, and
1101 * remembers 'struct bpf_map *' pointer it's part of.
1102 * bpf_timer_set_callback() increments prog refcnt and assign bpf callback_fn.
1103 * bpf_timer_start() arms the timer.
1104 * If user space reference to a map goes to zero at this point
1105 * ops->map_release_uref callback is responsible for cancelling the timers,
1106 * freeing their memory, and decrementing prog's refcnts.
1107 * bpf_timer_cancel() cancels the timer and decrements prog's refcnt.
1108 * Inner maps can contain bpf timers as well. ops->map_release_uref is
1109 * freeing the timers when inner map is replaced or deleted by user space.
1110 */
1111struct bpf_hrtimer {
1112 struct bpf_async_cb cb;
1113 struct hrtimer timer;
1114 atomic_t cancelling;
1115};
1116
1117struct bpf_work {
1118 struct bpf_async_cb cb;
1119 struct work_struct work;
1120 struct work_struct delete_work;
1121};
1122
1123/* the actual struct hidden inside uapi struct bpf_timer and bpf_wq */
1124struct bpf_async_kern {
1125 union {
1126 struct bpf_async_cb *cb;
1127 struct bpf_hrtimer *timer;
1128 struct bpf_work *work;
1129 };
1130 /* bpf_spin_lock is used here instead of spinlock_t to make
1131 * sure that it always fits into space reserved by struct bpf_timer
1132 * regardless of LOCKDEP and spinlock debug flags.
1133 */
1134 struct bpf_spin_lock lock;
1135} __attribute__((aligned(8)));
1136
1137enum bpf_async_type {
1138 BPF_ASYNC_TYPE_TIMER = 0,
1139 BPF_ASYNC_TYPE_WQ,
1140};
1141
1142static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1143
1144static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1145{
1146 struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1147 struct bpf_map *map = t->cb.map;
1148 void *value = t->cb.value;
1149 bpf_callback_t callback_fn;
1150 void *key;
1151 u32 idx;
1152
1153 BTF_TYPE_EMIT(struct bpf_timer);
1154 callback_fn = rcu_dereference_check(t->cb.callback_fn, rcu_read_lock_bh_held());
1155 if (!callback_fn)
1156 goto out;
1157
1158 /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1159 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1160 * Remember the timer this callback is servicing to prevent
1161 * deadlock if callback_fn() calls bpf_timer_cancel() or
1162 * bpf_map_delete_elem() on the same timer.
1163 */
1164 this_cpu_write(hrtimer_running, t);
1165 if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1166 struct bpf_array *array = container_of(map, struct bpf_array, map);
1167
1168 /* compute the key */
1169 idx = ((char *)value - array->value) / array->elem_size;
1170 key = &idx;
1171 } else { /* hash or lru */
1172 key = value - round_up(map->key_size, 8);
1173 }
1174
1175 callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1176 /* The verifier checked that return value is zero. */
1177
1178 this_cpu_write(hrtimer_running, NULL);
1179out:
1180 return HRTIMER_NORESTART;
1181}
1182
1183static void bpf_wq_work(struct work_struct *work)
1184{
1185 struct bpf_work *w = container_of(work, struct bpf_work, work);
1186 struct bpf_async_cb *cb = &w->cb;
1187 struct bpf_map *map = cb->map;
1188 bpf_callback_t callback_fn;
1189 void *value = cb->value;
1190 void *key;
1191 u32 idx;
1192
1193 BTF_TYPE_EMIT(struct bpf_wq);
1194
1195 callback_fn = READ_ONCE(cb->callback_fn);
1196 if (!callback_fn)
1197 return;
1198
1199 if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1200 struct bpf_array *array = container_of(map, struct bpf_array, map);
1201
1202 /* compute the key */
1203 idx = ((char *)value - array->value) / array->elem_size;
1204 key = &idx;
1205 } else { /* hash or lru */
1206 key = value - round_up(map->key_size, 8);
1207 }
1208
1209 rcu_read_lock_trace();
1210 migrate_disable();
1211
1212 callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1213
1214 migrate_enable();
1215 rcu_read_unlock_trace();
1216}
1217
1218static void bpf_wq_delete_work(struct work_struct *work)
1219{
1220 struct bpf_work *w = container_of(work, struct bpf_work, delete_work);
1221
1222 cancel_work_sync(&w->work);
1223
1224 kfree_rcu(w, cb.rcu);
1225}
1226
1227static void bpf_timer_delete_work(struct work_struct *work)
1228{
1229 struct bpf_hrtimer *t = container_of(work, struct bpf_hrtimer, cb.delete_work);
1230
1231 /* Cancel the timer and wait for callback to complete if it was running.
1232 * If hrtimer_cancel() can be safely called it's safe to call
1233 * kfree_rcu(t) right after for both preallocated and non-preallocated
1234 * maps. The async->cb = NULL was already done and no code path can see
1235 * address 't' anymore. Timer if armed for existing bpf_hrtimer before
1236 * bpf_timer_cancel_and_free will have been cancelled.
1237 */
1238 hrtimer_cancel(&t->timer);
1239 kfree_rcu(t, cb.rcu);
1240}
1241
1242static int __bpf_async_init(struct bpf_async_kern *async, struct bpf_map *map, u64 flags,
1243 enum bpf_async_type type)
1244{
1245 struct bpf_async_cb *cb;
1246 struct bpf_hrtimer *t;
1247 struct bpf_work *w;
1248 clockid_t clockid;
1249 size_t size;
1250 int ret = 0;
1251
1252 if (in_nmi())
1253 return -EOPNOTSUPP;
1254
1255 switch (type) {
1256 case BPF_ASYNC_TYPE_TIMER:
1257 size = sizeof(struct bpf_hrtimer);
1258 break;
1259 case BPF_ASYNC_TYPE_WQ:
1260 size = sizeof(struct bpf_work);
1261 break;
1262 default:
1263 return -EINVAL;
1264 }
1265
1266 __bpf_spin_lock_irqsave(&async->lock);
1267 t = async->timer;
1268 if (t) {
1269 ret = -EBUSY;
1270 goto out;
1271 }
1272
1273 /* allocate hrtimer via map_kmalloc to use memcg accounting */
1274 cb = bpf_map_kmalloc_node(map, size, GFP_ATOMIC, map->numa_node);
1275 if (!cb) {
1276 ret = -ENOMEM;
1277 goto out;
1278 }
1279
1280 switch (type) {
1281 case BPF_ASYNC_TYPE_TIMER:
1282 clockid = flags & (MAX_CLOCKS - 1);
1283 t = (struct bpf_hrtimer *)cb;
1284
1285 atomic_set(&t->cancelling, 0);
1286 INIT_WORK(&t->cb.delete_work, bpf_timer_delete_work);
1287 hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1288 t->timer.function = bpf_timer_cb;
1289 cb->value = (void *)async - map->record->timer_off;
1290 break;
1291 case BPF_ASYNC_TYPE_WQ:
1292 w = (struct bpf_work *)cb;
1293
1294 INIT_WORK(&w->work, bpf_wq_work);
1295 INIT_WORK(&w->delete_work, bpf_wq_delete_work);
1296 cb->value = (void *)async - map->record->wq_off;
1297 break;
1298 }
1299 cb->map = map;
1300 cb->prog = NULL;
1301 cb->flags = flags;
1302 rcu_assign_pointer(cb->callback_fn, NULL);
1303
1304 WRITE_ONCE(async->cb, cb);
1305 /* Guarantee the order between async->cb and map->usercnt. So
1306 * when there are concurrent uref release and bpf timer init, either
1307 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1308 * timer or atomic64_read() below returns a zero usercnt.
1309 */
1310 smp_mb();
1311 if (!atomic64_read(&map->usercnt)) {
1312 /* maps with timers must be either held by user space
1313 * or pinned in bpffs.
1314 */
1315 WRITE_ONCE(async->cb, NULL);
1316 kfree(cb);
1317 ret = -EPERM;
1318 }
1319out:
1320 __bpf_spin_unlock_irqrestore(&async->lock);
1321 return ret;
1322}
1323
1324BPF_CALL_3(bpf_timer_init, struct bpf_async_kern *, timer, struct bpf_map *, map,
1325 u64, flags)
1326{
1327 clock_t clockid = flags & (MAX_CLOCKS - 1);
1328
1329 BUILD_BUG_ON(MAX_CLOCKS != 16);
1330 BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_timer));
1331 BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_timer));
1332
1333 if (flags >= MAX_CLOCKS ||
1334 /* similar to timerfd except _ALARM variants are not supported */
1335 (clockid != CLOCK_MONOTONIC &&
1336 clockid != CLOCK_REALTIME &&
1337 clockid != CLOCK_BOOTTIME))
1338 return -EINVAL;
1339
1340 return __bpf_async_init(timer, map, flags, BPF_ASYNC_TYPE_TIMER);
1341}
1342
1343static const struct bpf_func_proto bpf_timer_init_proto = {
1344 .func = bpf_timer_init,
1345 .gpl_only = true,
1346 .ret_type = RET_INTEGER,
1347 .arg1_type = ARG_PTR_TO_TIMER,
1348 .arg2_type = ARG_CONST_MAP_PTR,
1349 .arg3_type = ARG_ANYTHING,
1350};
1351
1352static int __bpf_async_set_callback(struct bpf_async_kern *async, void *callback_fn,
1353 struct bpf_prog_aux *aux, unsigned int flags,
1354 enum bpf_async_type type)
1355{
1356 struct bpf_prog *prev, *prog = aux->prog;
1357 struct bpf_async_cb *cb;
1358 int ret = 0;
1359
1360 if (in_nmi())
1361 return -EOPNOTSUPP;
1362 __bpf_spin_lock_irqsave(&async->lock);
1363 cb = async->cb;
1364 if (!cb) {
1365 ret = -EINVAL;
1366 goto out;
1367 }
1368 if (!atomic64_read(&cb->map->usercnt)) {
1369 /* maps with timers must be either held by user space
1370 * or pinned in bpffs. Otherwise timer might still be
1371 * running even when bpf prog is detached and user space
1372 * is gone, since map_release_uref won't ever be called.
1373 */
1374 ret = -EPERM;
1375 goto out;
1376 }
1377 prev = cb->prog;
1378 if (prev != prog) {
1379 /* Bump prog refcnt once. Every bpf_timer_set_callback()
1380 * can pick different callback_fn-s within the same prog.
1381 */
1382 prog = bpf_prog_inc_not_zero(prog);
1383 if (IS_ERR(prog)) {
1384 ret = PTR_ERR(prog);
1385 goto out;
1386 }
1387 if (prev)
1388 /* Drop prev prog refcnt when swapping with new prog */
1389 bpf_prog_put(prev);
1390 cb->prog = prog;
1391 }
1392 rcu_assign_pointer(cb->callback_fn, callback_fn);
1393out:
1394 __bpf_spin_unlock_irqrestore(&async->lock);
1395 return ret;
1396}
1397
1398BPF_CALL_3(bpf_timer_set_callback, struct bpf_async_kern *, timer, void *, callback_fn,
1399 struct bpf_prog_aux *, aux)
1400{
1401 return __bpf_async_set_callback(timer, callback_fn, aux, 0, BPF_ASYNC_TYPE_TIMER);
1402}
1403
1404static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1405 .func = bpf_timer_set_callback,
1406 .gpl_only = true,
1407 .ret_type = RET_INTEGER,
1408 .arg1_type = ARG_PTR_TO_TIMER,
1409 .arg2_type = ARG_PTR_TO_FUNC,
1410};
1411
1412BPF_CALL_3(bpf_timer_start, struct bpf_async_kern *, timer, u64, nsecs, u64, flags)
1413{
1414 struct bpf_hrtimer *t;
1415 int ret = 0;
1416 enum hrtimer_mode mode;
1417
1418 if (in_nmi())
1419 return -EOPNOTSUPP;
1420 if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
1421 return -EINVAL;
1422 __bpf_spin_lock_irqsave(&timer->lock);
1423 t = timer->timer;
1424 if (!t || !t->cb.prog) {
1425 ret = -EINVAL;
1426 goto out;
1427 }
1428
1429 if (flags & BPF_F_TIMER_ABS)
1430 mode = HRTIMER_MODE_ABS_SOFT;
1431 else
1432 mode = HRTIMER_MODE_REL_SOFT;
1433
1434 if (flags & BPF_F_TIMER_CPU_PIN)
1435 mode |= HRTIMER_MODE_PINNED;
1436
1437 hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
1438out:
1439 __bpf_spin_unlock_irqrestore(&timer->lock);
1440 return ret;
1441}
1442
1443static const struct bpf_func_proto bpf_timer_start_proto = {
1444 .func = bpf_timer_start,
1445 .gpl_only = true,
1446 .ret_type = RET_INTEGER,
1447 .arg1_type = ARG_PTR_TO_TIMER,
1448 .arg2_type = ARG_ANYTHING,
1449 .arg3_type = ARG_ANYTHING,
1450};
1451
1452static void drop_prog_refcnt(struct bpf_async_cb *async)
1453{
1454 struct bpf_prog *prog = async->prog;
1455
1456 if (prog) {
1457 bpf_prog_put(prog);
1458 async->prog = NULL;
1459 rcu_assign_pointer(async->callback_fn, NULL);
1460 }
1461}
1462
1463BPF_CALL_1(bpf_timer_cancel, struct bpf_async_kern *, timer)
1464{
1465 struct bpf_hrtimer *t, *cur_t;
1466 bool inc = false;
1467 int ret = 0;
1468
1469 if (in_nmi())
1470 return -EOPNOTSUPP;
1471 rcu_read_lock();
1472 __bpf_spin_lock_irqsave(&timer->lock);
1473 t = timer->timer;
1474 if (!t) {
1475 ret = -EINVAL;
1476 goto out;
1477 }
1478
1479 cur_t = this_cpu_read(hrtimer_running);
1480 if (cur_t == t) {
1481 /* If bpf callback_fn is trying to bpf_timer_cancel()
1482 * its own timer the hrtimer_cancel() will deadlock
1483 * since it waits for callback_fn to finish.
1484 */
1485 ret = -EDEADLK;
1486 goto out;
1487 }
1488
1489 /* Only account in-flight cancellations when invoked from a timer
1490 * callback, since we want to avoid waiting only if other _callbacks_
1491 * are waiting on us, to avoid introducing lockups. Non-callback paths
1492 * are ok, since nobody would synchronously wait for their completion.
1493 */
1494 if (!cur_t)
1495 goto drop;
1496 atomic_inc(&t->cancelling);
1497 /* Need full barrier after relaxed atomic_inc */
1498 smp_mb__after_atomic();
1499 inc = true;
1500 if (atomic_read(&cur_t->cancelling)) {
1501 /* We're cancelling timer t, while some other timer callback is
1502 * attempting to cancel us. In such a case, it might be possible
1503 * that timer t belongs to the other callback, or some other
1504 * callback waiting upon it (creating transitive dependencies
1505 * upon us), and we will enter a deadlock if we continue
1506 * cancelling and waiting for it synchronously, since it might
1507 * do the same. Bail!
1508 */
1509 ret = -EDEADLK;
1510 goto out;
1511 }
1512drop:
1513 drop_prog_refcnt(&t->cb);
1514out:
1515 __bpf_spin_unlock_irqrestore(&timer->lock);
1516 /* Cancel the timer and wait for associated callback to finish
1517 * if it was running.
1518 */
1519 ret = ret ?: hrtimer_cancel(&t->timer);
1520 if (inc)
1521 atomic_dec(&t->cancelling);
1522 rcu_read_unlock();
1523 return ret;
1524}
1525
1526static const struct bpf_func_proto bpf_timer_cancel_proto = {
1527 .func = bpf_timer_cancel,
1528 .gpl_only = true,
1529 .ret_type = RET_INTEGER,
1530 .arg1_type = ARG_PTR_TO_TIMER,
1531};
1532
1533static struct bpf_async_cb *__bpf_async_cancel_and_free(struct bpf_async_kern *async)
1534{
1535 struct bpf_async_cb *cb;
1536
1537 /* Performance optimization: read async->cb without lock first. */
1538 if (!READ_ONCE(async->cb))
1539 return NULL;
1540
1541 __bpf_spin_lock_irqsave(&async->lock);
1542 /* re-read it under lock */
1543 cb = async->cb;
1544 if (!cb)
1545 goto out;
1546 drop_prog_refcnt(cb);
1547 /* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1548 * this timer, since it won't be initialized.
1549 */
1550 WRITE_ONCE(async->cb, NULL);
1551out:
1552 __bpf_spin_unlock_irqrestore(&async->lock);
1553 return cb;
1554}
1555
1556/* This function is called by map_delete/update_elem for individual element and
1557 * by ops->map_release_uref when the user space reference to a map reaches zero.
1558 */
1559void bpf_timer_cancel_and_free(void *val)
1560{
1561 struct bpf_hrtimer *t;
1562
1563 t = (struct bpf_hrtimer *)__bpf_async_cancel_and_free(val);
1564
1565 if (!t)
1566 return;
1567 /* We check that bpf_map_delete/update_elem() was called from timer
1568 * callback_fn. In such case we don't call hrtimer_cancel() (since it
1569 * will deadlock) and don't call hrtimer_try_to_cancel() (since it will
1570 * just return -1). Though callback_fn is still running on this cpu it's
1571 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1572 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1573 * since async->cb = NULL was already done. The timer will be
1574 * effectively cancelled because bpf_timer_cb() will return
1575 * HRTIMER_NORESTART.
1576 *
1577 * However, it is possible the timer callback_fn calling us armed the
1578 * timer _before_ calling us, such that failing to cancel it here will
1579 * cause it to possibly use struct hrtimer after freeing bpf_hrtimer.
1580 * Therefore, we _need_ to cancel any outstanding timers before we do
1581 * kfree_rcu, even though no more timers can be armed.
1582 *
1583 * Moreover, we need to schedule work even if timer does not belong to
1584 * the calling callback_fn, as on two different CPUs, we can end up in a
1585 * situation where both sides run in parallel, try to cancel one
1586 * another, and we end up waiting on both sides in hrtimer_cancel
1587 * without making forward progress, since timer1 depends on time2
1588 * callback to finish, and vice versa.
1589 *
1590 * CPU 1 (timer1_cb) CPU 2 (timer2_cb)
1591 * bpf_timer_cancel_and_free(timer2) bpf_timer_cancel_and_free(timer1)
1592 *
1593 * To avoid these issues, punt to workqueue context when we are in a
1594 * timer callback.
1595 */
1596 if (this_cpu_read(hrtimer_running)) {
1597 queue_work(system_unbound_wq, &t->cb.delete_work);
1598 return;
1599 }
1600
1601 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1602 /* If the timer is running on other CPU, also use a kworker to
1603 * wait for the completion of the timer instead of trying to
1604 * acquire a sleepable lock in hrtimer_cancel() to wait for its
1605 * completion.
1606 */
1607 if (hrtimer_try_to_cancel(&t->timer) >= 0)
1608 kfree_rcu(t, cb.rcu);
1609 else
1610 queue_work(system_unbound_wq, &t->cb.delete_work);
1611 } else {
1612 bpf_timer_delete_work(&t->cb.delete_work);
1613 }
1614}
1615
1616/* This function is called by map_delete/update_elem for individual element and
1617 * by ops->map_release_uref when the user space reference to a map reaches zero.
1618 */
1619void bpf_wq_cancel_and_free(void *val)
1620{
1621 struct bpf_work *work;
1622
1623 BTF_TYPE_EMIT(struct bpf_wq);
1624
1625 work = (struct bpf_work *)__bpf_async_cancel_and_free(val);
1626 if (!work)
1627 return;
1628 /* Trigger cancel of the sleepable work, but *do not* wait for
1629 * it to finish if it was running as we might not be in a
1630 * sleepable context.
1631 * kfree will be called once the work has finished.
1632 */
1633 schedule_work(&work->delete_work);
1634}
1635
1636BPF_CALL_2(bpf_kptr_xchg, void *, dst, void *, ptr)
1637{
1638 unsigned long *kptr = dst;
1639
1640 /* This helper may be inlined by verifier. */
1641 return xchg(kptr, (unsigned long)ptr);
1642}
1643
1644/* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1645 * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1646 * denote type that verifier will determine.
1647 */
1648static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1649 .func = bpf_kptr_xchg,
1650 .gpl_only = false,
1651 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1652 .ret_btf_id = BPF_PTR_POISON,
1653 .arg1_type = ARG_KPTR_XCHG_DEST,
1654 .arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1655 .arg2_btf_id = BPF_PTR_POISON,
1656};
1657
1658/* Since the upper 8 bits of dynptr->size is reserved, the
1659 * maximum supported size is 2^24 - 1.
1660 */
1661#define DYNPTR_MAX_SIZE ((1UL << 24) - 1)
1662#define DYNPTR_TYPE_SHIFT 28
1663#define DYNPTR_SIZE_MASK 0xFFFFFF
1664#define DYNPTR_RDONLY_BIT BIT(31)
1665
1666bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1667{
1668 return ptr->size & DYNPTR_RDONLY_BIT;
1669}
1670
1671void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1672{
1673 ptr->size |= DYNPTR_RDONLY_BIT;
1674}
1675
1676static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1677{
1678 ptr->size |= type << DYNPTR_TYPE_SHIFT;
1679}
1680
1681static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1682{
1683 return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1684}
1685
1686u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1687{
1688 return ptr->size & DYNPTR_SIZE_MASK;
1689}
1690
1691static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1692{
1693 u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1694
1695 ptr->size = new_size | metadata;
1696}
1697
1698int bpf_dynptr_check_size(u32 size)
1699{
1700 return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1701}
1702
1703void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1704 enum bpf_dynptr_type type, u32 offset, u32 size)
1705{
1706 ptr->data = data;
1707 ptr->offset = offset;
1708 ptr->size = size;
1709 bpf_dynptr_set_type(ptr, type);
1710}
1711
1712void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1713{
1714 memset(ptr, 0, sizeof(*ptr));
1715}
1716
1717static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1718{
1719 u32 size = __bpf_dynptr_size(ptr);
1720
1721 if (len > size || offset > size - len)
1722 return -E2BIG;
1723
1724 return 0;
1725}
1726
1727BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1728{
1729 int err;
1730
1731 BTF_TYPE_EMIT(struct bpf_dynptr);
1732
1733 err = bpf_dynptr_check_size(size);
1734 if (err)
1735 goto error;
1736
1737 /* flags is currently unsupported */
1738 if (flags) {
1739 err = -EINVAL;
1740 goto error;
1741 }
1742
1743 bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1744
1745 return 0;
1746
1747error:
1748 bpf_dynptr_set_null(ptr);
1749 return err;
1750}
1751
1752static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1753 .func = bpf_dynptr_from_mem,
1754 .gpl_only = false,
1755 .ret_type = RET_INTEGER,
1756 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1757 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1758 .arg3_type = ARG_ANYTHING,
1759 .arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT | MEM_WRITE,
1760};
1761
1762BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1763 u32, offset, u64, flags)
1764{
1765 enum bpf_dynptr_type type;
1766 int err;
1767
1768 if (!src->data || flags)
1769 return -EINVAL;
1770
1771 err = bpf_dynptr_check_off_len(src, offset, len);
1772 if (err)
1773 return err;
1774
1775 type = bpf_dynptr_get_type(src);
1776
1777 switch (type) {
1778 case BPF_DYNPTR_TYPE_LOCAL:
1779 case BPF_DYNPTR_TYPE_RINGBUF:
1780 /* Source and destination may possibly overlap, hence use memmove to
1781 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1782 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1783 */
1784 memmove(dst, src->data + src->offset + offset, len);
1785 return 0;
1786 case BPF_DYNPTR_TYPE_SKB:
1787 return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len);
1788 case BPF_DYNPTR_TYPE_XDP:
1789 return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len);
1790 default:
1791 WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1792 return -EFAULT;
1793 }
1794}
1795
1796static const struct bpf_func_proto bpf_dynptr_read_proto = {
1797 .func = bpf_dynptr_read,
1798 .gpl_only = false,
1799 .ret_type = RET_INTEGER,
1800 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1801 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1802 .arg3_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1803 .arg4_type = ARG_ANYTHING,
1804 .arg5_type = ARG_ANYTHING,
1805};
1806
1807BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1808 u32, len, u64, flags)
1809{
1810 enum bpf_dynptr_type type;
1811 int err;
1812
1813 if (!dst->data || __bpf_dynptr_is_rdonly(dst))
1814 return -EINVAL;
1815
1816 err = bpf_dynptr_check_off_len(dst, offset, len);
1817 if (err)
1818 return err;
1819
1820 type = bpf_dynptr_get_type(dst);
1821
1822 switch (type) {
1823 case BPF_DYNPTR_TYPE_LOCAL:
1824 case BPF_DYNPTR_TYPE_RINGBUF:
1825 if (flags)
1826 return -EINVAL;
1827 /* Source and destination may possibly overlap, hence use memmove to
1828 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1829 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1830 */
1831 memmove(dst->data + dst->offset + offset, src, len);
1832 return 0;
1833 case BPF_DYNPTR_TYPE_SKB:
1834 return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len,
1835 flags);
1836 case BPF_DYNPTR_TYPE_XDP:
1837 if (flags)
1838 return -EINVAL;
1839 return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len);
1840 default:
1841 WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1842 return -EFAULT;
1843 }
1844}
1845
1846static const struct bpf_func_proto bpf_dynptr_write_proto = {
1847 .func = bpf_dynptr_write,
1848 .gpl_only = false,
1849 .ret_type = RET_INTEGER,
1850 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1851 .arg2_type = ARG_ANYTHING,
1852 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1853 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1854 .arg5_type = ARG_ANYTHING,
1855};
1856
1857BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1858{
1859 enum bpf_dynptr_type type;
1860 int err;
1861
1862 if (!ptr->data)
1863 return 0;
1864
1865 err = bpf_dynptr_check_off_len(ptr, offset, len);
1866 if (err)
1867 return 0;
1868
1869 if (__bpf_dynptr_is_rdonly(ptr))
1870 return 0;
1871
1872 type = bpf_dynptr_get_type(ptr);
1873
1874 switch (type) {
1875 case BPF_DYNPTR_TYPE_LOCAL:
1876 case BPF_DYNPTR_TYPE_RINGBUF:
1877 return (unsigned long)(ptr->data + ptr->offset + offset);
1878 case BPF_DYNPTR_TYPE_SKB:
1879 case BPF_DYNPTR_TYPE_XDP:
1880 /* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1881 return 0;
1882 default:
1883 WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1884 return 0;
1885 }
1886}
1887
1888static const struct bpf_func_proto bpf_dynptr_data_proto = {
1889 .func = bpf_dynptr_data,
1890 .gpl_only = false,
1891 .ret_type = RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1892 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1893 .arg2_type = ARG_ANYTHING,
1894 .arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO,
1895};
1896
1897const struct bpf_func_proto bpf_get_current_task_proto __weak;
1898const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1899const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1900const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1901const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1902const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1903const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1904
1905const struct bpf_func_proto *
1906bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1907{
1908 switch (func_id) {
1909 case BPF_FUNC_map_lookup_elem:
1910 return &bpf_map_lookup_elem_proto;
1911 case BPF_FUNC_map_update_elem:
1912 return &bpf_map_update_elem_proto;
1913 case BPF_FUNC_map_delete_elem:
1914 return &bpf_map_delete_elem_proto;
1915 case BPF_FUNC_map_push_elem:
1916 return &bpf_map_push_elem_proto;
1917 case BPF_FUNC_map_pop_elem:
1918 return &bpf_map_pop_elem_proto;
1919 case BPF_FUNC_map_peek_elem:
1920 return &bpf_map_peek_elem_proto;
1921 case BPF_FUNC_map_lookup_percpu_elem:
1922 return &bpf_map_lookup_percpu_elem_proto;
1923 case BPF_FUNC_get_prandom_u32:
1924 return &bpf_get_prandom_u32_proto;
1925 case BPF_FUNC_get_smp_processor_id:
1926 return &bpf_get_raw_smp_processor_id_proto;
1927 case BPF_FUNC_get_numa_node_id:
1928 return &bpf_get_numa_node_id_proto;
1929 case BPF_FUNC_tail_call:
1930 return &bpf_tail_call_proto;
1931 case BPF_FUNC_ktime_get_ns:
1932 return &bpf_ktime_get_ns_proto;
1933 case BPF_FUNC_ktime_get_boot_ns:
1934 return &bpf_ktime_get_boot_ns_proto;
1935 case BPF_FUNC_ktime_get_tai_ns:
1936 return &bpf_ktime_get_tai_ns_proto;
1937 case BPF_FUNC_ringbuf_output:
1938 return &bpf_ringbuf_output_proto;
1939 case BPF_FUNC_ringbuf_reserve:
1940 return &bpf_ringbuf_reserve_proto;
1941 case BPF_FUNC_ringbuf_submit:
1942 return &bpf_ringbuf_submit_proto;
1943 case BPF_FUNC_ringbuf_discard:
1944 return &bpf_ringbuf_discard_proto;
1945 case BPF_FUNC_ringbuf_query:
1946 return &bpf_ringbuf_query_proto;
1947 case BPF_FUNC_strncmp:
1948 return &bpf_strncmp_proto;
1949 case BPF_FUNC_strtol:
1950 return &bpf_strtol_proto;
1951 case BPF_FUNC_strtoul:
1952 return &bpf_strtoul_proto;
1953 case BPF_FUNC_get_current_pid_tgid:
1954 return &bpf_get_current_pid_tgid_proto;
1955 case BPF_FUNC_get_ns_current_pid_tgid:
1956 return &bpf_get_ns_current_pid_tgid_proto;
1957 default:
1958 break;
1959 }
1960
1961 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
1962 return NULL;
1963
1964 switch (func_id) {
1965 case BPF_FUNC_spin_lock:
1966 return &bpf_spin_lock_proto;
1967 case BPF_FUNC_spin_unlock:
1968 return &bpf_spin_unlock_proto;
1969 case BPF_FUNC_jiffies64:
1970 return &bpf_jiffies64_proto;
1971 case BPF_FUNC_per_cpu_ptr:
1972 return &bpf_per_cpu_ptr_proto;
1973 case BPF_FUNC_this_cpu_ptr:
1974 return &bpf_this_cpu_ptr_proto;
1975 case BPF_FUNC_timer_init:
1976 return &bpf_timer_init_proto;
1977 case BPF_FUNC_timer_set_callback:
1978 return &bpf_timer_set_callback_proto;
1979 case BPF_FUNC_timer_start:
1980 return &bpf_timer_start_proto;
1981 case BPF_FUNC_timer_cancel:
1982 return &bpf_timer_cancel_proto;
1983 case BPF_FUNC_kptr_xchg:
1984 return &bpf_kptr_xchg_proto;
1985 case BPF_FUNC_for_each_map_elem:
1986 return &bpf_for_each_map_elem_proto;
1987 case BPF_FUNC_loop:
1988 return &bpf_loop_proto;
1989 case BPF_FUNC_user_ringbuf_drain:
1990 return &bpf_user_ringbuf_drain_proto;
1991 case BPF_FUNC_ringbuf_reserve_dynptr:
1992 return &bpf_ringbuf_reserve_dynptr_proto;
1993 case BPF_FUNC_ringbuf_submit_dynptr:
1994 return &bpf_ringbuf_submit_dynptr_proto;
1995 case BPF_FUNC_ringbuf_discard_dynptr:
1996 return &bpf_ringbuf_discard_dynptr_proto;
1997 case BPF_FUNC_dynptr_from_mem:
1998 return &bpf_dynptr_from_mem_proto;
1999 case BPF_FUNC_dynptr_read:
2000 return &bpf_dynptr_read_proto;
2001 case BPF_FUNC_dynptr_write:
2002 return &bpf_dynptr_write_proto;
2003 case BPF_FUNC_dynptr_data:
2004 return &bpf_dynptr_data_proto;
2005#ifdef CONFIG_CGROUPS
2006 case BPF_FUNC_cgrp_storage_get:
2007 return &bpf_cgrp_storage_get_proto;
2008 case BPF_FUNC_cgrp_storage_delete:
2009 return &bpf_cgrp_storage_delete_proto;
2010 case BPF_FUNC_get_current_cgroup_id:
2011 return &bpf_get_current_cgroup_id_proto;
2012 case BPF_FUNC_get_current_ancestor_cgroup_id:
2013 return &bpf_get_current_ancestor_cgroup_id_proto;
2014#endif
2015 default:
2016 break;
2017 }
2018
2019 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
2020 return NULL;
2021
2022 switch (func_id) {
2023 case BPF_FUNC_trace_printk:
2024 return bpf_get_trace_printk_proto();
2025 case BPF_FUNC_get_current_task:
2026 return &bpf_get_current_task_proto;
2027 case BPF_FUNC_get_current_task_btf:
2028 return &bpf_get_current_task_btf_proto;
2029 case BPF_FUNC_probe_read_user:
2030 return &bpf_probe_read_user_proto;
2031 case BPF_FUNC_probe_read_kernel:
2032 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
2033 NULL : &bpf_probe_read_kernel_proto;
2034 case BPF_FUNC_probe_read_user_str:
2035 return &bpf_probe_read_user_str_proto;
2036 case BPF_FUNC_probe_read_kernel_str:
2037 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
2038 NULL : &bpf_probe_read_kernel_str_proto;
2039 case BPF_FUNC_snprintf_btf:
2040 return &bpf_snprintf_btf_proto;
2041 case BPF_FUNC_snprintf:
2042 return &bpf_snprintf_proto;
2043 case BPF_FUNC_task_pt_regs:
2044 return &bpf_task_pt_regs_proto;
2045 case BPF_FUNC_trace_vprintk:
2046 return bpf_get_trace_vprintk_proto();
2047 default:
2048 return NULL;
2049 }
2050}
2051EXPORT_SYMBOL_GPL(bpf_base_func_proto);
2052
2053void bpf_list_head_free(const struct btf_field *field, void *list_head,
2054 struct bpf_spin_lock *spin_lock)
2055{
2056 struct list_head *head = list_head, *orig_head = list_head;
2057
2058 BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
2059 BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
2060
2061 /* Do the actual list draining outside the lock to not hold the lock for
2062 * too long, and also prevent deadlocks if tracing programs end up
2063 * executing on entry/exit of functions called inside the critical
2064 * section, and end up doing map ops that call bpf_list_head_free for
2065 * the same map value again.
2066 */
2067 __bpf_spin_lock_irqsave(spin_lock);
2068 if (!head->next || list_empty(head))
2069 goto unlock;
2070 head = head->next;
2071unlock:
2072 INIT_LIST_HEAD(orig_head);
2073 __bpf_spin_unlock_irqrestore(spin_lock);
2074
2075 while (head != orig_head) {
2076 void *obj = head;
2077
2078 obj -= field->graph_root.node_offset;
2079 head = head->next;
2080 /* The contained type can also have resources, including a
2081 * bpf_list_head which needs to be freed.
2082 */
2083 migrate_disable();
2084 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
2085 migrate_enable();
2086 }
2087}
2088
2089/* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
2090 * 'rb_node *', so field name of rb_node within containing struct is not
2091 * needed.
2092 *
2093 * Since bpf_rb_tree's node type has a corresponding struct btf_field with
2094 * graph_root.node_offset, it's not necessary to know field name
2095 * or type of node struct
2096 */
2097#define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
2098 for (pos = rb_first_postorder(root); \
2099 pos && ({ n = rb_next_postorder(pos); 1; }); \
2100 pos = n)
2101
2102void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
2103 struct bpf_spin_lock *spin_lock)
2104{
2105 struct rb_root_cached orig_root, *root = rb_root;
2106 struct rb_node *pos, *n;
2107 void *obj;
2108
2109 BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
2110 BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
2111
2112 __bpf_spin_lock_irqsave(spin_lock);
2113 orig_root = *root;
2114 *root = RB_ROOT_CACHED;
2115 __bpf_spin_unlock_irqrestore(spin_lock);
2116
2117 bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
2118 obj = pos;
2119 obj -= field->graph_root.node_offset;
2120
2121
2122 migrate_disable();
2123 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
2124 migrate_enable();
2125 }
2126}
2127
2128__bpf_kfunc_start_defs();
2129
2130__bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign)
2131{
2132 struct btf_struct_meta *meta = meta__ign;
2133 u64 size = local_type_id__k;
2134 void *p;
2135
2136 p = bpf_mem_alloc(&bpf_global_ma, size);
2137 if (!p)
2138 return NULL;
2139 if (meta)
2140 bpf_obj_init(meta->record, p);
2141 return p;
2142}
2143
2144__bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
2145{
2146 u64 size = local_type_id__k;
2147
2148 /* The verifier has ensured that meta__ign must be NULL */
2149 return bpf_mem_alloc(&bpf_global_percpu_ma, size);
2150}
2151
2152/* Must be called under migrate_disable(), as required by bpf_mem_free */
2153void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
2154{
2155 struct bpf_mem_alloc *ma;
2156
2157 if (rec && rec->refcount_off >= 0 &&
2158 !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
2159 /* Object is refcounted and refcount_dec didn't result in 0
2160 * refcount. Return without freeing the object
2161 */
2162 return;
2163 }
2164
2165 if (rec)
2166 bpf_obj_free_fields(rec, p);
2167
2168 if (percpu)
2169 ma = &bpf_global_percpu_ma;
2170 else
2171 ma = &bpf_global_ma;
2172 bpf_mem_free_rcu(ma, p);
2173}
2174
2175__bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
2176{
2177 struct btf_struct_meta *meta = meta__ign;
2178 void *p = p__alloc;
2179
2180 __bpf_obj_drop_impl(p, meta ? meta->record : NULL, false);
2181}
2182
2183__bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
2184{
2185 /* The verifier has ensured that meta__ign must be NULL */
2186 bpf_mem_free_rcu(&bpf_global_percpu_ma, p__alloc);
2187}
2188
2189__bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
2190{
2191 struct btf_struct_meta *meta = meta__ign;
2192 struct bpf_refcount *ref;
2193
2194 /* Could just cast directly to refcount_t *, but need some code using
2195 * bpf_refcount type so that it is emitted in vmlinux BTF
2196 */
2197 ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
2198 if (!refcount_inc_not_zero((refcount_t *)ref))
2199 return NULL;
2200
2201 /* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
2202 * in verifier.c
2203 */
2204 return (void *)p__refcounted_kptr;
2205}
2206
2207static int __bpf_list_add(struct bpf_list_node_kern *node,
2208 struct bpf_list_head *head,
2209 bool tail, struct btf_record *rec, u64 off)
2210{
2211 struct list_head *n = &node->list_head, *h = (void *)head;
2212
2213 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2214 * called on its fields, so init here
2215 */
2216 if (unlikely(!h->next))
2217 INIT_LIST_HEAD(h);
2218
2219 /* node->owner != NULL implies !list_empty(n), no need to separately
2220 * check the latter
2221 */
2222 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2223 /* Only called from BPF prog, no need to migrate_disable */
2224 __bpf_obj_drop_impl((void *)n - off, rec, false);
2225 return -EINVAL;
2226 }
2227
2228 tail ? list_add_tail(n, h) : list_add(n, h);
2229 WRITE_ONCE(node->owner, head);
2230
2231 return 0;
2232}
2233
2234__bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
2235 struct bpf_list_node *node,
2236 void *meta__ign, u64 off)
2237{
2238 struct bpf_list_node_kern *n = (void *)node;
2239 struct btf_struct_meta *meta = meta__ign;
2240
2241 return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off);
2242}
2243
2244__bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2245 struct bpf_list_node *node,
2246 void *meta__ign, u64 off)
2247{
2248 struct bpf_list_node_kern *n = (void *)node;
2249 struct btf_struct_meta *meta = meta__ign;
2250
2251 return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off);
2252}
2253
2254static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2255{
2256 struct list_head *n, *h = (void *)head;
2257 struct bpf_list_node_kern *node;
2258
2259 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2260 * called on its fields, so init here
2261 */
2262 if (unlikely(!h->next))
2263 INIT_LIST_HEAD(h);
2264 if (list_empty(h))
2265 return NULL;
2266
2267 n = tail ? h->prev : h->next;
2268 node = container_of(n, struct bpf_list_node_kern, list_head);
2269 if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2270 return NULL;
2271
2272 list_del_init(n);
2273 WRITE_ONCE(node->owner, NULL);
2274 return (struct bpf_list_node *)n;
2275}
2276
2277__bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2278{
2279 return __bpf_list_del(head, false);
2280}
2281
2282__bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2283{
2284 return __bpf_list_del(head, true);
2285}
2286
2287__bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2288 struct bpf_rb_node *node)
2289{
2290 struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2291 struct rb_root_cached *r = (struct rb_root_cached *)root;
2292 struct rb_node *n = &node_internal->rb_node;
2293
2294 /* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2295 * n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2296 */
2297 if (READ_ONCE(node_internal->owner) != root)
2298 return NULL;
2299
2300 rb_erase_cached(n, r);
2301 RB_CLEAR_NODE(n);
2302 WRITE_ONCE(node_internal->owner, NULL);
2303 return (struct bpf_rb_node *)n;
2304}
2305
2306/* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2307 * program
2308 */
2309static int __bpf_rbtree_add(struct bpf_rb_root *root,
2310 struct bpf_rb_node_kern *node,
2311 void *less, struct btf_record *rec, u64 off)
2312{
2313 struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2314 struct rb_node *parent = NULL, *n = &node->rb_node;
2315 bpf_callback_t cb = (bpf_callback_t)less;
2316 bool leftmost = true;
2317
2318 /* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2319 * check the latter
2320 */
2321 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2322 /* Only called from BPF prog, no need to migrate_disable */
2323 __bpf_obj_drop_impl((void *)n - off, rec, false);
2324 return -EINVAL;
2325 }
2326
2327 while (*link) {
2328 parent = *link;
2329 if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2330 link = &parent->rb_left;
2331 } else {
2332 link = &parent->rb_right;
2333 leftmost = false;
2334 }
2335 }
2336
2337 rb_link_node(n, parent, link);
2338 rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost);
2339 WRITE_ONCE(node->owner, root);
2340 return 0;
2341}
2342
2343__bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2344 bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2345 void *meta__ign, u64 off)
2346{
2347 struct btf_struct_meta *meta = meta__ign;
2348 struct bpf_rb_node_kern *n = (void *)node;
2349
2350 return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off);
2351}
2352
2353__bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2354{
2355 struct rb_root_cached *r = (struct rb_root_cached *)root;
2356
2357 return (struct bpf_rb_node *)rb_first_cached(r);
2358}
2359
2360/**
2361 * bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2362 * kfunc which is not stored in a map as a kptr, must be released by calling
2363 * bpf_task_release().
2364 * @p: The task on which a reference is being acquired.
2365 */
2366__bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2367{
2368 if (refcount_inc_not_zero(&p->rcu_users))
2369 return p;
2370 return NULL;
2371}
2372
2373/**
2374 * bpf_task_release - Release the reference acquired on a task.
2375 * @p: The task on which a reference is being released.
2376 */
2377__bpf_kfunc void bpf_task_release(struct task_struct *p)
2378{
2379 put_task_struct_rcu_user(p);
2380}
2381
2382__bpf_kfunc void bpf_task_release_dtor(void *p)
2383{
2384 put_task_struct_rcu_user(p);
2385}
2386CFI_NOSEAL(bpf_task_release_dtor);
2387
2388#ifdef CONFIG_CGROUPS
2389/**
2390 * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2391 * this kfunc which is not stored in a map as a kptr, must be released by
2392 * calling bpf_cgroup_release().
2393 * @cgrp: The cgroup on which a reference is being acquired.
2394 */
2395__bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2396{
2397 return cgroup_tryget(cgrp) ? cgrp : NULL;
2398}
2399
2400/**
2401 * bpf_cgroup_release - Release the reference acquired on a cgroup.
2402 * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2403 * not be freed until the current grace period has ended, even if its refcount
2404 * drops to 0.
2405 * @cgrp: The cgroup on which a reference is being released.
2406 */
2407__bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2408{
2409 cgroup_put(cgrp);
2410}
2411
2412__bpf_kfunc void bpf_cgroup_release_dtor(void *cgrp)
2413{
2414 cgroup_put(cgrp);
2415}
2416CFI_NOSEAL(bpf_cgroup_release_dtor);
2417
2418/**
2419 * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2420 * array. A cgroup returned by this kfunc which is not subsequently stored in a
2421 * map, must be released by calling bpf_cgroup_release().
2422 * @cgrp: The cgroup for which we're performing a lookup.
2423 * @level: The level of ancestor to look up.
2424 */
2425__bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2426{
2427 struct cgroup *ancestor;
2428
2429 if (level > cgrp->level || level < 0)
2430 return NULL;
2431
2432 /* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2433 ancestor = cgrp->ancestors[level];
2434 if (!cgroup_tryget(ancestor))
2435 return NULL;
2436 return ancestor;
2437}
2438
2439/**
2440 * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2441 * kfunc which is not subsequently stored in a map, must be released by calling
2442 * bpf_cgroup_release().
2443 * @cgid: cgroup id.
2444 */
2445__bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2446{
2447 struct cgroup *cgrp;
2448
2449 cgrp = cgroup_get_from_id(cgid);
2450 if (IS_ERR(cgrp))
2451 return NULL;
2452 return cgrp;
2453}
2454
2455/**
2456 * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2457 * task's membership of cgroup ancestry.
2458 * @task: the task to be tested
2459 * @ancestor: possible ancestor of @task's cgroup
2460 *
2461 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2462 * It follows all the same rules as cgroup_is_descendant, and only applies
2463 * to the default hierarchy.
2464 */
2465__bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2466 struct cgroup *ancestor)
2467{
2468 long ret;
2469
2470 rcu_read_lock();
2471 ret = task_under_cgroup_hierarchy(task, ancestor);
2472 rcu_read_unlock();
2473 return ret;
2474}
2475
2476BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
2477{
2478 struct bpf_array *array = container_of(map, struct bpf_array, map);
2479 struct cgroup *cgrp;
2480
2481 if (unlikely(idx >= array->map.max_entries))
2482 return -E2BIG;
2483
2484 cgrp = READ_ONCE(array->ptrs[idx]);
2485 if (unlikely(!cgrp))
2486 return -EAGAIN;
2487
2488 return task_under_cgroup_hierarchy(current, cgrp);
2489}
2490
2491const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
2492 .func = bpf_current_task_under_cgroup,
2493 .gpl_only = false,
2494 .ret_type = RET_INTEGER,
2495 .arg1_type = ARG_CONST_MAP_PTR,
2496 .arg2_type = ARG_ANYTHING,
2497};
2498
2499/**
2500 * bpf_task_get_cgroup1 - Acquires the associated cgroup of a task within a
2501 * specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
2502 * hierarchy ID.
2503 * @task: The target task
2504 * @hierarchy_id: The ID of a cgroup1 hierarchy
2505 *
2506 * On success, the cgroup is returen. On failure, NULL is returned.
2507 */
2508__bpf_kfunc struct cgroup *
2509bpf_task_get_cgroup1(struct task_struct *task, int hierarchy_id)
2510{
2511 struct cgroup *cgrp = task_get_cgroup1(task, hierarchy_id);
2512
2513 if (IS_ERR(cgrp))
2514 return NULL;
2515 return cgrp;
2516}
2517#endif /* CONFIG_CGROUPS */
2518
2519/**
2520 * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2521 * in the root pid namespace idr. If a task is returned, it must either be
2522 * stored in a map, or released with bpf_task_release().
2523 * @pid: The pid of the task being looked up.
2524 */
2525__bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2526{
2527 struct task_struct *p;
2528
2529 rcu_read_lock();
2530 p = find_task_by_pid_ns(pid, &init_pid_ns);
2531 if (p)
2532 p = bpf_task_acquire(p);
2533 rcu_read_unlock();
2534
2535 return p;
2536}
2537
2538/**
2539 * bpf_task_from_vpid - Find a struct task_struct from its vpid by looking it up
2540 * in the pid namespace of the current task. If a task is returned, it must
2541 * either be stored in a map, or released with bpf_task_release().
2542 * @vpid: The vpid of the task being looked up.
2543 */
2544__bpf_kfunc struct task_struct *bpf_task_from_vpid(s32 vpid)
2545{
2546 struct task_struct *p;
2547
2548 rcu_read_lock();
2549 p = find_task_by_vpid(vpid);
2550 if (p)
2551 p = bpf_task_acquire(p);
2552 rcu_read_unlock();
2553
2554 return p;
2555}
2556
2557/**
2558 * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2559 * @p: The dynptr whose data slice to retrieve
2560 * @offset: Offset into the dynptr
2561 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2562 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2563 * length of the requested slice. This must be a constant.
2564 *
2565 * For non-skb and non-xdp type dynptrs, there is no difference between
2566 * bpf_dynptr_slice and bpf_dynptr_data.
2567 *
2568 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2569 *
2570 * If the intention is to write to the data slice, please use
2571 * bpf_dynptr_slice_rdwr.
2572 *
2573 * The user must check that the returned pointer is not null before using it.
2574 *
2575 * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2576 * does not change the underlying packet data pointers, so a call to
2577 * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2578 * the bpf program.
2579 *
2580 * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2581 * data slice (can be either direct pointer to the data or a pointer to the user
2582 * provided buffer, with its contents containing the data, if unable to obtain
2583 * direct pointer)
2584 */
2585__bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr *p, u32 offset,
2586 void *buffer__opt, u32 buffer__szk)
2587{
2588 const struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2589 enum bpf_dynptr_type type;
2590 u32 len = buffer__szk;
2591 int err;
2592
2593 if (!ptr->data)
2594 return NULL;
2595
2596 err = bpf_dynptr_check_off_len(ptr, offset, len);
2597 if (err)
2598 return NULL;
2599
2600 type = bpf_dynptr_get_type(ptr);
2601
2602 switch (type) {
2603 case BPF_DYNPTR_TYPE_LOCAL:
2604 case BPF_DYNPTR_TYPE_RINGBUF:
2605 return ptr->data + ptr->offset + offset;
2606 case BPF_DYNPTR_TYPE_SKB:
2607 if (buffer__opt)
2608 return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt);
2609 else
2610 return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len);
2611 case BPF_DYNPTR_TYPE_XDP:
2612 {
2613 void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len);
2614 if (!IS_ERR_OR_NULL(xdp_ptr))
2615 return xdp_ptr;
2616
2617 if (!buffer__opt)
2618 return NULL;
2619 bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false);
2620 return buffer__opt;
2621 }
2622 default:
2623 WARN_ONCE(true, "unknown dynptr type %d\n", type);
2624 return NULL;
2625 }
2626}
2627
2628/**
2629 * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2630 * @p: The dynptr whose data slice to retrieve
2631 * @offset: Offset into the dynptr
2632 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2633 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2634 * length of the requested slice. This must be a constant.
2635 *
2636 * For non-skb and non-xdp type dynptrs, there is no difference between
2637 * bpf_dynptr_slice and bpf_dynptr_data.
2638 *
2639 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2640 *
2641 * The returned pointer is writable and may point to either directly the dynptr
2642 * data at the requested offset or to the buffer if unable to obtain a direct
2643 * data pointer to (example: the requested slice is to the paged area of an skb
2644 * packet). In the case where the returned pointer is to the buffer, the user
2645 * is responsible for persisting writes through calling bpf_dynptr_write(). This
2646 * usually looks something like this pattern:
2647 *
2648 * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2649 * if (!eth)
2650 * return TC_ACT_SHOT;
2651 *
2652 * // mutate eth header //
2653 *
2654 * if (eth == buffer)
2655 * bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2656 *
2657 * Please note that, as in the example above, the user must check that the
2658 * returned pointer is not null before using it.
2659 *
2660 * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2661 * does not change the underlying packet data pointers, so a call to
2662 * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2663 * the bpf program.
2664 *
2665 * Return: NULL if the call failed (eg invalid dynptr), pointer to a
2666 * data slice (can be either direct pointer to the data or a pointer to the user
2667 * provided buffer, with its contents containing the data, if unable to obtain
2668 * direct pointer)
2669 */
2670__bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr *p, u32 offset,
2671 void *buffer__opt, u32 buffer__szk)
2672{
2673 const struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2674
2675 if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2676 return NULL;
2677
2678 /* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2679 *
2680 * For skb-type dynptrs, it is safe to write into the returned pointer
2681 * if the bpf program allows skb data writes. There are two possibilities
2682 * that may occur when calling bpf_dynptr_slice_rdwr:
2683 *
2684 * 1) The requested slice is in the head of the skb. In this case, the
2685 * returned pointer is directly to skb data, and if the skb is cloned, the
2686 * verifier will have uncloned it (see bpf_unclone_prologue()) already.
2687 * The pointer can be directly written into.
2688 *
2689 * 2) Some portion of the requested slice is in the paged buffer area.
2690 * In this case, the requested data will be copied out into the buffer
2691 * and the returned pointer will be a pointer to the buffer. The skb
2692 * will not be pulled. To persist the write, the user will need to call
2693 * bpf_dynptr_write(), which will pull the skb and commit the write.
2694 *
2695 * Similarly for xdp programs, if the requested slice is not across xdp
2696 * fragments, then a direct pointer will be returned, otherwise the data
2697 * will be copied out into the buffer and the user will need to call
2698 * bpf_dynptr_write() to commit changes.
2699 */
2700 return bpf_dynptr_slice(p, offset, buffer__opt, buffer__szk);
2701}
2702
2703__bpf_kfunc int bpf_dynptr_adjust(const struct bpf_dynptr *p, u32 start, u32 end)
2704{
2705 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2706 u32 size;
2707
2708 if (!ptr->data || start > end)
2709 return -EINVAL;
2710
2711 size = __bpf_dynptr_size(ptr);
2712
2713 if (start > size || end > size)
2714 return -ERANGE;
2715
2716 ptr->offset += start;
2717 bpf_dynptr_set_size(ptr, end - start);
2718
2719 return 0;
2720}
2721
2722__bpf_kfunc bool bpf_dynptr_is_null(const struct bpf_dynptr *p)
2723{
2724 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2725
2726 return !ptr->data;
2727}
2728
2729__bpf_kfunc bool bpf_dynptr_is_rdonly(const struct bpf_dynptr *p)
2730{
2731 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2732
2733 if (!ptr->data)
2734 return false;
2735
2736 return __bpf_dynptr_is_rdonly(ptr);
2737}
2738
2739__bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr *p)
2740{
2741 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2742
2743 if (!ptr->data)
2744 return -EINVAL;
2745
2746 return __bpf_dynptr_size(ptr);
2747}
2748
2749__bpf_kfunc int bpf_dynptr_clone(const struct bpf_dynptr *p,
2750 struct bpf_dynptr *clone__uninit)
2751{
2752 struct bpf_dynptr_kern *clone = (struct bpf_dynptr_kern *)clone__uninit;
2753 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)p;
2754
2755 if (!ptr->data) {
2756 bpf_dynptr_set_null(clone);
2757 return -EINVAL;
2758 }
2759
2760 *clone = *ptr;
2761
2762 return 0;
2763}
2764
2765__bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2766{
2767 return obj;
2768}
2769
2770__bpf_kfunc void *bpf_rdonly_cast(const void *obj__ign, u32 btf_id__k)
2771{
2772 return (void *)obj__ign;
2773}
2774
2775__bpf_kfunc void bpf_rcu_read_lock(void)
2776{
2777 rcu_read_lock();
2778}
2779
2780__bpf_kfunc void bpf_rcu_read_unlock(void)
2781{
2782 rcu_read_unlock();
2783}
2784
2785struct bpf_throw_ctx {
2786 struct bpf_prog_aux *aux;
2787 u64 sp;
2788 u64 bp;
2789 int cnt;
2790};
2791
2792static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
2793{
2794 struct bpf_throw_ctx *ctx = cookie;
2795 struct bpf_prog *prog;
2796
2797 if (!is_bpf_text_address(ip))
2798 return !ctx->cnt;
2799 prog = bpf_prog_ksym_find(ip);
2800 ctx->cnt++;
2801 if (bpf_is_subprog(prog))
2802 return true;
2803 ctx->aux = prog->aux;
2804 ctx->sp = sp;
2805 ctx->bp = bp;
2806 return false;
2807}
2808
2809__bpf_kfunc void bpf_throw(u64 cookie)
2810{
2811 struct bpf_throw_ctx ctx = {};
2812
2813 arch_bpf_stack_walk(bpf_stack_walker, &ctx);
2814 WARN_ON_ONCE(!ctx.aux);
2815 if (ctx.aux)
2816 WARN_ON_ONCE(!ctx.aux->exception_boundary);
2817 WARN_ON_ONCE(!ctx.bp);
2818 WARN_ON_ONCE(!ctx.cnt);
2819 /* Prevent KASAN false positives for CONFIG_KASAN_STACK by unpoisoning
2820 * deeper stack depths than ctx.sp as we do not return from bpf_throw,
2821 * which skips compiler generated instrumentation to do the same.
2822 */
2823 kasan_unpoison_task_stack_below((void *)(long)ctx.sp);
2824 ctx.aux->bpf_exception_cb(cookie, ctx.sp, ctx.bp, 0, 0);
2825 WARN(1, "A call to BPF exception callback should never return\n");
2826}
2827
2828__bpf_kfunc int bpf_wq_init(struct bpf_wq *wq, void *p__map, unsigned int flags)
2829{
2830 struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2831 struct bpf_map *map = p__map;
2832
2833 BUILD_BUG_ON(sizeof(struct bpf_async_kern) > sizeof(struct bpf_wq));
2834 BUILD_BUG_ON(__alignof__(struct bpf_async_kern) != __alignof__(struct bpf_wq));
2835
2836 if (flags)
2837 return -EINVAL;
2838
2839 return __bpf_async_init(async, map, flags, BPF_ASYNC_TYPE_WQ);
2840}
2841
2842__bpf_kfunc int bpf_wq_start(struct bpf_wq *wq, unsigned int flags)
2843{
2844 struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2845 struct bpf_work *w;
2846
2847 if (in_nmi())
2848 return -EOPNOTSUPP;
2849 if (flags)
2850 return -EINVAL;
2851 w = READ_ONCE(async->work);
2852 if (!w || !READ_ONCE(w->cb.prog))
2853 return -EINVAL;
2854
2855 schedule_work(&w->work);
2856 return 0;
2857}
2858
2859__bpf_kfunc int bpf_wq_set_callback_impl(struct bpf_wq *wq,
2860 int (callback_fn)(void *map, int *key, void *value),
2861 unsigned int flags,
2862 void *aux__ign)
2863{
2864 struct bpf_prog_aux *aux = (struct bpf_prog_aux *)aux__ign;
2865 struct bpf_async_kern *async = (struct bpf_async_kern *)wq;
2866
2867 if (flags)
2868 return -EINVAL;
2869
2870 return __bpf_async_set_callback(async, callback_fn, aux, flags, BPF_ASYNC_TYPE_WQ);
2871}
2872
2873__bpf_kfunc void bpf_preempt_disable(void)
2874{
2875 preempt_disable();
2876}
2877
2878__bpf_kfunc void bpf_preempt_enable(void)
2879{
2880 preempt_enable();
2881}
2882
2883struct bpf_iter_bits {
2884 __u64 __opaque[2];
2885} __aligned(8);
2886
2887#define BITS_ITER_NR_WORDS_MAX 511
2888
2889struct bpf_iter_bits_kern {
2890 union {
2891 __u64 *bits;
2892 __u64 bits_copy;
2893 };
2894 int nr_bits;
2895 int bit;
2896} __aligned(8);
2897
2898/* On 64-bit hosts, unsigned long and u64 have the same size, so passing
2899 * a u64 pointer and an unsigned long pointer to find_next_bit() will
2900 * return the same result, as both point to the same 8-byte area.
2901 *
2902 * For 32-bit little-endian hosts, using a u64 pointer or unsigned long
2903 * pointer also makes no difference. This is because the first iterated
2904 * unsigned long is composed of bits 0-31 of the u64 and the second unsigned
2905 * long is composed of bits 32-63 of the u64.
2906 *
2907 * However, for 32-bit big-endian hosts, this is not the case. The first
2908 * iterated unsigned long will be bits 32-63 of the u64, so swap these two
2909 * ulong values within the u64.
2910 */
2911static void swap_ulong_in_u64(u64 *bits, unsigned int nr)
2912{
2913#if (BITS_PER_LONG == 32) && defined(__BIG_ENDIAN)
2914 unsigned int i;
2915
2916 for (i = 0; i < nr; i++)
2917 bits[i] = (bits[i] >> 32) | ((u64)(u32)bits[i] << 32);
2918#endif
2919}
2920
2921/**
2922 * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area
2923 * @it: The new bpf_iter_bits to be created
2924 * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over
2925 * @nr_words: The size of the specified memory area, measured in 8-byte units.
2926 * The maximum value of @nr_words is @BITS_ITER_NR_WORDS_MAX. This limit may be
2927 * further reduced by the BPF memory allocator implementation.
2928 *
2929 * This function initializes a new bpf_iter_bits structure for iterating over
2930 * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It
2931 * copies the data of the memory area to the newly created bpf_iter_bits @it for
2932 * subsequent iteration operations.
2933 *
2934 * On success, 0 is returned. On failure, ERR is returned.
2935 */
2936__bpf_kfunc int
2937bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words)
2938{
2939 struct bpf_iter_bits_kern *kit = (void *)it;
2940 u32 nr_bytes = nr_words * sizeof(u64);
2941 u32 nr_bits = BYTES_TO_BITS(nr_bytes);
2942 int err;
2943
2944 BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits));
2945 BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) !=
2946 __alignof__(struct bpf_iter_bits));
2947
2948 kit->nr_bits = 0;
2949 kit->bits_copy = 0;
2950 kit->bit = -1;
2951
2952 if (!unsafe_ptr__ign || !nr_words)
2953 return -EINVAL;
2954 if (nr_words > BITS_ITER_NR_WORDS_MAX)
2955 return -E2BIG;
2956
2957 /* Optimization for u64 mask */
2958 if (nr_bits == 64) {
2959 err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign);
2960 if (err)
2961 return -EFAULT;
2962
2963 swap_ulong_in_u64(&kit->bits_copy, nr_words);
2964
2965 kit->nr_bits = nr_bits;
2966 return 0;
2967 }
2968
2969 if (bpf_mem_alloc_check_size(false, nr_bytes))
2970 return -E2BIG;
2971
2972 /* Fallback to memalloc */
2973 kit->bits = bpf_mem_alloc(&bpf_global_ma, nr_bytes);
2974 if (!kit->bits)
2975 return -ENOMEM;
2976
2977 err = bpf_probe_read_kernel_common(kit->bits, nr_bytes, unsafe_ptr__ign);
2978 if (err) {
2979 bpf_mem_free(&bpf_global_ma, kit->bits);
2980 return err;
2981 }
2982
2983 swap_ulong_in_u64(kit->bits, nr_words);
2984
2985 kit->nr_bits = nr_bits;
2986 return 0;
2987}
2988
2989/**
2990 * bpf_iter_bits_next() - Get the next bit in a bpf_iter_bits
2991 * @it: The bpf_iter_bits to be checked
2992 *
2993 * This function returns a pointer to a number representing the value of the
2994 * next bit in the bits.
2995 *
2996 * If there are no further bits available, it returns NULL.
2997 */
2998__bpf_kfunc int *bpf_iter_bits_next(struct bpf_iter_bits *it)
2999{
3000 struct bpf_iter_bits_kern *kit = (void *)it;
3001 int bit = kit->bit, nr_bits = kit->nr_bits;
3002 const void *bits;
3003
3004 if (!nr_bits || bit >= nr_bits)
3005 return NULL;
3006
3007 bits = nr_bits == 64 ? &kit->bits_copy : kit->bits;
3008 bit = find_next_bit(bits, nr_bits, bit + 1);
3009 if (bit >= nr_bits) {
3010 kit->bit = bit;
3011 return NULL;
3012 }
3013
3014 kit->bit = bit;
3015 return &kit->bit;
3016}
3017
3018/**
3019 * bpf_iter_bits_destroy() - Destroy a bpf_iter_bits
3020 * @it: The bpf_iter_bits to be destroyed
3021 *
3022 * Destroy the resource associated with the bpf_iter_bits.
3023 */
3024__bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it)
3025{
3026 struct bpf_iter_bits_kern *kit = (void *)it;
3027
3028 if (kit->nr_bits <= 64)
3029 return;
3030 bpf_mem_free(&bpf_global_ma, kit->bits);
3031}
3032
3033/**
3034 * bpf_copy_from_user_str() - Copy a string from an unsafe user address
3035 * @dst: Destination address, in kernel space. This buffer must be
3036 * at least @dst__sz bytes long.
3037 * @dst__sz: Maximum number of bytes to copy, includes the trailing NUL.
3038 * @unsafe_ptr__ign: Source address, in user space.
3039 * @flags: The only supported flag is BPF_F_PAD_ZEROS
3040 *
3041 * Copies a NUL-terminated string from userspace to BPF space. If user string is
3042 * too long this will still ensure zero termination in the dst buffer unless
3043 * buffer size is 0.
3044 *
3045 * If BPF_F_PAD_ZEROS flag is set, memset the tail of @dst to 0 on success and
3046 * memset all of @dst on failure.
3047 */
3048__bpf_kfunc int bpf_copy_from_user_str(void *dst, u32 dst__sz, const void __user *unsafe_ptr__ign, u64 flags)
3049{
3050 int ret;
3051
3052 if (unlikely(flags & ~BPF_F_PAD_ZEROS))
3053 return -EINVAL;
3054
3055 if (unlikely(!dst__sz))
3056 return 0;
3057
3058 ret = strncpy_from_user(dst, unsafe_ptr__ign, dst__sz - 1);
3059 if (ret < 0) {
3060 if (flags & BPF_F_PAD_ZEROS)
3061 memset((char *)dst, 0, dst__sz);
3062
3063 return ret;
3064 }
3065
3066 if (flags & BPF_F_PAD_ZEROS)
3067 memset((char *)dst + ret, 0, dst__sz - ret);
3068 else
3069 ((char *)dst)[ret] = '\0';
3070
3071 return ret + 1;
3072}
3073
3074__bpf_kfunc_end_defs();
3075
3076BTF_KFUNCS_START(generic_btf_ids)
3077#ifdef CONFIG_CRASH_DUMP
3078BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
3079#endif
3080BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
3081BTF_ID_FLAGS(func, bpf_percpu_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
3082BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
3083BTF_ID_FLAGS(func, bpf_percpu_obj_drop_impl, KF_RELEASE)
3084BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL | KF_RCU)
3085BTF_ID_FLAGS(func, bpf_list_push_front_impl)
3086BTF_ID_FLAGS(func, bpf_list_push_back_impl)
3087BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
3088BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
3089BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3090BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
3091BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
3092BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
3093BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
3094
3095#ifdef CONFIG_CGROUPS
3096BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3097BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
3098BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3099BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
3100BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
3101BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
3102#endif
3103BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
3104BTF_ID_FLAGS(func, bpf_task_from_vpid, KF_ACQUIRE | KF_RET_NULL)
3105BTF_ID_FLAGS(func, bpf_throw)
3106BTF_ID_FLAGS(func, bpf_send_signal_task, KF_TRUSTED_ARGS)
3107BTF_KFUNCS_END(generic_btf_ids)
3108
3109static const struct btf_kfunc_id_set generic_kfunc_set = {
3110 .owner = THIS_MODULE,
3111 .set = &generic_btf_ids,
3112};
3113
3114
3115BTF_ID_LIST(generic_dtor_ids)
3116BTF_ID(struct, task_struct)
3117BTF_ID(func, bpf_task_release_dtor)
3118#ifdef CONFIG_CGROUPS
3119BTF_ID(struct, cgroup)
3120BTF_ID(func, bpf_cgroup_release_dtor)
3121#endif
3122
3123BTF_KFUNCS_START(common_btf_ids)
3124BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx, KF_FASTCALL)
3125BTF_ID_FLAGS(func, bpf_rdonly_cast, KF_FASTCALL)
3126BTF_ID_FLAGS(func, bpf_rcu_read_lock)
3127BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
3128BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
3129BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
3130BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
3131BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
3132BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
3133BTF_ID_FLAGS(func, bpf_iter_task_vma_new, KF_ITER_NEW | KF_RCU)
3134BTF_ID_FLAGS(func, bpf_iter_task_vma_next, KF_ITER_NEXT | KF_RET_NULL)
3135BTF_ID_FLAGS(func, bpf_iter_task_vma_destroy, KF_ITER_DESTROY)
3136#ifdef CONFIG_CGROUPS
3137BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS)
3138BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
3139BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
3140BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
3141BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
3142BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
3143#endif
3144BTF_ID_FLAGS(func, bpf_iter_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
3145BTF_ID_FLAGS(func, bpf_iter_task_next, KF_ITER_NEXT | KF_RET_NULL)
3146BTF_ID_FLAGS(func, bpf_iter_task_destroy, KF_ITER_DESTROY)
3147BTF_ID_FLAGS(func, bpf_dynptr_adjust)
3148BTF_ID_FLAGS(func, bpf_dynptr_is_null)
3149BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
3150BTF_ID_FLAGS(func, bpf_dynptr_size)
3151BTF_ID_FLAGS(func, bpf_dynptr_clone)
3152BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
3153BTF_ID_FLAGS(func, bpf_wq_init)
3154BTF_ID_FLAGS(func, bpf_wq_set_callback_impl)
3155BTF_ID_FLAGS(func, bpf_wq_start)
3156BTF_ID_FLAGS(func, bpf_preempt_disable)
3157BTF_ID_FLAGS(func, bpf_preempt_enable)
3158BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
3159BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
3160BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
3161BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
3162BTF_ID_FLAGS(func, bpf_get_kmem_cache)
3163BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
3164BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
3165BTF_ID_FLAGS(func, bpf_iter_kmem_cache_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
3166BTF_KFUNCS_END(common_btf_ids)
3167
3168static const struct btf_kfunc_id_set common_kfunc_set = {
3169 .owner = THIS_MODULE,
3170 .set = &common_btf_ids,
3171};
3172
3173static int __init kfunc_init(void)
3174{
3175 int ret;
3176 const struct btf_id_dtor_kfunc generic_dtors[] = {
3177 {
3178 .btf_id = generic_dtor_ids[0],
3179 .kfunc_btf_id = generic_dtor_ids[1]
3180 },
3181#ifdef CONFIG_CGROUPS
3182 {
3183 .btf_id = generic_dtor_ids[2],
3184 .kfunc_btf_id = generic_dtor_ids[3]
3185 },
3186#endif
3187 };
3188
3189 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set);
3190 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set);
3191 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &generic_kfunc_set);
3192 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set);
3193 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &generic_kfunc_set);
3194 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &generic_kfunc_set);
3195 ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors,
3196 ARRAY_SIZE(generic_dtors),
3197 THIS_MODULE);
3198 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set);
3199}
3200
3201late_initcall(kfunc_init);
3202
3203/* Get a pointer to dynptr data up to len bytes for read only access. If
3204 * the dynptr doesn't have continuous data up to len bytes, return NULL.
3205 */
3206const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len)
3207{
3208 const struct bpf_dynptr *p = (struct bpf_dynptr *)ptr;
3209
3210 return bpf_dynptr_slice(p, 0, NULL, len);
3211}
3212
3213/* Get a pointer to dynptr data up to len bytes for read write access. If
3214 * the dynptr doesn't have continuous data up to len bytes, or the dynptr
3215 * is read only, return NULL.
3216 */
3217void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len)
3218{
3219 if (__bpf_dynptr_is_rdonly(ptr))
3220 return NULL;
3221 return (void *)__bpf_dynptr_data(ptr, len);
3222}