1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
5 #include <linux/rcupdate.h>
6 #include <linux/random.h>
8 #include <linux/topology.h>
9 #include <linux/ktime.h>
10 #include <linux/sched.h>
11 #include <linux/uidgid.h>
12 #include <linux/filter.h>
13 #include <linux/ctype.h>
14 #include <linux/jiffies.h>
15 #include <linux/pid_namespace.h>
16 #include <linux/proc_ns.h>
18 #include "../../lib/kstrtox.h"
20 /* If kernel subsystem is allowing eBPF programs to call this function,
21 * inside its own verifier_ops->get_func_proto() callback it should return
22 * bpf_map_lookup_elem_proto, so that verifier can properly check the arguments
24 * Different map implementations will rely on rcu in map methods
25 * lookup/update/delete, therefore eBPF programs must run under rcu lock
26 * if program is allowed to access maps, so check rcu_read_lock_held in
27 * all three functions.
29 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
31 WARN_ON_ONCE(!rcu_read_lock_held());
32 return (unsigned long) map->ops->map_lookup_elem(map, key);
35 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
36 .func = bpf_map_lookup_elem,
39 .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL,
40 .arg1_type = ARG_CONST_MAP_PTR,
41 .arg2_type = ARG_PTR_TO_MAP_KEY,
44 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
45 void *, value, u64, flags)
47 WARN_ON_ONCE(!rcu_read_lock_held());
48 return map->ops->map_update_elem(map, key, value, flags);
51 const struct bpf_func_proto bpf_map_update_elem_proto = {
52 .func = bpf_map_update_elem,
55 .ret_type = RET_INTEGER,
56 .arg1_type = ARG_CONST_MAP_PTR,
57 .arg2_type = ARG_PTR_TO_MAP_KEY,
58 .arg3_type = ARG_PTR_TO_MAP_VALUE,
59 .arg4_type = ARG_ANYTHING,
62 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
64 WARN_ON_ONCE(!rcu_read_lock_held());
65 return map->ops->map_delete_elem(map, key);
68 const struct bpf_func_proto bpf_map_delete_elem_proto = {
69 .func = bpf_map_delete_elem,
72 .ret_type = RET_INTEGER,
73 .arg1_type = ARG_CONST_MAP_PTR,
74 .arg2_type = ARG_PTR_TO_MAP_KEY,
77 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
79 return map->ops->map_push_elem(map, value, flags);
82 const struct bpf_func_proto bpf_map_push_elem_proto = {
83 .func = bpf_map_push_elem,
86 .ret_type = RET_INTEGER,
87 .arg1_type = ARG_CONST_MAP_PTR,
88 .arg2_type = ARG_PTR_TO_MAP_VALUE,
89 .arg3_type = ARG_ANYTHING,
92 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
94 return map->ops->map_pop_elem(map, value);
97 const struct bpf_func_proto bpf_map_pop_elem_proto = {
98 .func = bpf_map_pop_elem,
100 .ret_type = RET_INTEGER,
101 .arg1_type = ARG_CONST_MAP_PTR,
102 .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE,
105 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
107 return map->ops->map_peek_elem(map, value);
110 const struct bpf_func_proto bpf_map_peek_elem_proto = {
111 .func = bpf_map_pop_elem,
113 .ret_type = RET_INTEGER,
114 .arg1_type = ARG_CONST_MAP_PTR,
115 .arg2_type = ARG_PTR_TO_UNINIT_MAP_VALUE,
118 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
119 .func = bpf_user_rnd_u32,
121 .ret_type = RET_INTEGER,
124 BPF_CALL_0(bpf_get_smp_processor_id)
126 return smp_processor_id();
129 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
130 .func = bpf_get_smp_processor_id,
132 .ret_type = RET_INTEGER,
135 BPF_CALL_0(bpf_get_numa_node_id)
137 return numa_node_id();
140 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
141 .func = bpf_get_numa_node_id,
143 .ret_type = RET_INTEGER,
146 BPF_CALL_0(bpf_ktime_get_ns)
148 /* NMI safe access to clock monotonic */
149 return ktime_get_mono_fast_ns();
152 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
153 .func = bpf_ktime_get_ns,
155 .ret_type = RET_INTEGER,
158 BPF_CALL_0(bpf_ktime_get_boot_ns)
160 /* NMI safe access to clock boottime */
161 return ktime_get_boot_fast_ns();
164 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
165 .func = bpf_ktime_get_boot_ns,
167 .ret_type = RET_INTEGER,
170 BPF_CALL_0(bpf_get_current_pid_tgid)
172 struct task_struct *task = current;
177 return (u64) task->tgid << 32 | task->pid;
180 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
181 .func = bpf_get_current_pid_tgid,
183 .ret_type = RET_INTEGER,
186 BPF_CALL_0(bpf_get_current_uid_gid)
188 struct task_struct *task = current;
195 current_uid_gid(&uid, &gid);
196 return (u64) from_kgid(&init_user_ns, gid) << 32 |
197 from_kuid(&init_user_ns, uid);
200 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
201 .func = bpf_get_current_uid_gid,
203 .ret_type = RET_INTEGER,
206 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
208 struct task_struct *task = current;
213 strncpy(buf, task->comm, size);
215 /* Verifier guarantees that size > 0. For task->comm exceeding
216 * size, guarantee that buf is %NUL-terminated. Unconditionally
217 * done here to save the size test.
222 memset(buf, 0, size);
226 const struct bpf_func_proto bpf_get_current_comm_proto = {
227 .func = bpf_get_current_comm,
229 .ret_type = RET_INTEGER,
230 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
231 .arg2_type = ARG_CONST_SIZE,
234 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
236 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
238 arch_spinlock_t *l = (void *)lock;
241 arch_spinlock_t lock;
242 } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
244 compiletime_assert(u.val == 0, "__ARCH_SPIN_LOCK_UNLOCKED not 0");
245 BUILD_BUG_ON(sizeof(*l) != sizeof(__u32));
246 BUILD_BUG_ON(sizeof(*lock) != sizeof(__u32));
250 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
252 arch_spinlock_t *l = (void *)lock;
259 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
261 atomic_t *l = (void *)lock;
263 BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
265 atomic_cond_read_relaxed(l, !VAL);
266 } while (atomic_xchg(l, 1));
269 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
271 atomic_t *l = (void *)lock;
273 atomic_set_release(l, 0);
278 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
280 notrace BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
284 local_irq_save(flags);
285 __bpf_spin_lock(lock);
286 __this_cpu_write(irqsave_flags, flags);
290 const struct bpf_func_proto bpf_spin_lock_proto = {
291 .func = bpf_spin_lock,
293 .ret_type = RET_VOID,
294 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
297 notrace BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
301 flags = __this_cpu_read(irqsave_flags);
302 __bpf_spin_unlock(lock);
303 local_irq_restore(flags);
307 const struct bpf_func_proto bpf_spin_unlock_proto = {
308 .func = bpf_spin_unlock,
310 .ret_type = RET_VOID,
311 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
314 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
317 struct bpf_spin_lock *lock;
320 lock = src + map->spin_lock_off;
322 lock = dst + map->spin_lock_off;
324 ____bpf_spin_lock(lock);
325 copy_map_value(map, dst, src);
326 ____bpf_spin_unlock(lock);
330 BPF_CALL_0(bpf_jiffies64)
332 return get_jiffies_64();
335 const struct bpf_func_proto bpf_jiffies64_proto = {
336 .func = bpf_jiffies64,
338 .ret_type = RET_INTEGER,
341 #ifdef CONFIG_CGROUPS
342 BPF_CALL_0(bpf_get_current_cgroup_id)
344 struct cgroup *cgrp = task_dfl_cgroup(current);
346 return cgroup_id(cgrp);
349 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
350 .func = bpf_get_current_cgroup_id,
352 .ret_type = RET_INTEGER,
355 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
357 struct cgroup *cgrp = task_dfl_cgroup(current);
358 struct cgroup *ancestor;
360 ancestor = cgroup_ancestor(cgrp, ancestor_level);
363 return cgroup_id(ancestor);
366 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
367 .func = bpf_get_current_ancestor_cgroup_id,
369 .ret_type = RET_INTEGER,
370 .arg1_type = ARG_ANYTHING,
373 #ifdef CONFIG_CGROUP_BPF
374 DECLARE_PER_CPU(struct bpf_cgroup_storage*,
375 bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]);
377 BPF_CALL_2(bpf_get_local_storage, struct bpf_map *, map, u64, flags)
379 /* flags argument is not used now,
380 * but provides an ability to extend the API.
381 * verifier checks that its value is correct.
383 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map);
384 struct bpf_cgroup_storage *storage;
387 storage = this_cpu_read(bpf_cgroup_storage[stype]);
389 if (stype == BPF_CGROUP_STORAGE_SHARED)
390 ptr = &READ_ONCE(storage->buf)->data[0];
392 ptr = this_cpu_ptr(storage->percpu_buf);
394 return (unsigned long)ptr;
397 const struct bpf_func_proto bpf_get_local_storage_proto = {
398 .func = bpf_get_local_storage,
400 .ret_type = RET_PTR_TO_MAP_VALUE,
401 .arg1_type = ARG_CONST_MAP_PTR,
402 .arg2_type = ARG_ANYTHING,
406 #define BPF_STRTOX_BASE_MASK 0x1F
408 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
409 unsigned long long *res, bool *is_negative)
411 unsigned int base = flags & BPF_STRTOX_BASE_MASK;
412 const char *cur_buf = buf;
413 size_t cur_len = buf_len;
414 unsigned int consumed;
418 if (!buf || !buf_len || !res || !is_negative)
421 if (base != 0 && base != 8 && base != 10 && base != 16)
424 if (flags & ~BPF_STRTOX_BASE_MASK)
427 while (cur_buf < buf + buf_len && isspace(*cur_buf))
430 *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
434 consumed = cur_buf - buf;
439 cur_len = min(cur_len, sizeof(str) - 1);
440 memcpy(str, cur_buf, cur_len);
444 cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
445 val_len = _parse_integer(cur_buf, base, res);
447 if (val_len & KSTRTOX_OVERFLOW)
454 consumed += cur_buf - str;
459 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
462 unsigned long long _res;
466 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
470 if ((long long)-_res > 0)
474 if ((long long)_res < 0)
481 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
487 err = __bpf_strtoll(buf, buf_len, flags, &_res);
490 if (_res != (long)_res)
496 const struct bpf_func_proto bpf_strtol_proto = {
499 .ret_type = RET_INTEGER,
500 .arg1_type = ARG_PTR_TO_MEM,
501 .arg2_type = ARG_CONST_SIZE,
502 .arg3_type = ARG_ANYTHING,
503 .arg4_type = ARG_PTR_TO_LONG,
506 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
507 unsigned long *, res)
509 unsigned long long _res;
513 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
518 if (_res != (unsigned long)_res)
524 const struct bpf_func_proto bpf_strtoul_proto = {
527 .ret_type = RET_INTEGER,
528 .arg1_type = ARG_PTR_TO_MEM,
529 .arg2_type = ARG_CONST_SIZE,
530 .arg3_type = ARG_ANYTHING,
531 .arg4_type = ARG_PTR_TO_LONG,
535 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
536 struct bpf_pidns_info *, nsdata, u32, size)
538 struct task_struct *task = current;
539 struct pid_namespace *pidns;
542 if (unlikely(size != sizeof(struct bpf_pidns_info)))
545 if (unlikely((u64)(dev_t)dev != dev))
551 pidns = task_active_pid_ns(task);
552 if (unlikely(!pidns)) {
557 if (!ns_match(&pidns->ns, (dev_t)dev, ino))
560 nsdata->pid = task_pid_nr_ns(task, pidns);
561 nsdata->tgid = task_tgid_nr_ns(task, pidns);
564 memset((void *)nsdata, 0, (size_t) size);
568 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
569 .func = bpf_get_ns_current_pid_tgid,
571 .ret_type = RET_INTEGER,
572 .arg1_type = ARG_ANYTHING,
573 .arg2_type = ARG_ANYTHING,
574 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
575 .arg4_type = ARG_CONST_SIZE,
578 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
579 .func = bpf_get_raw_cpu_id,
581 .ret_type = RET_INTEGER,
584 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
585 u64, flags, void *, data, u64, size)
587 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
590 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
593 const struct bpf_func_proto bpf_event_output_data_proto = {
594 .func = bpf_event_output_data,
596 .ret_type = RET_INTEGER,
597 .arg1_type = ARG_PTR_TO_CTX,
598 .arg2_type = ARG_CONST_MAP_PTR,
599 .arg3_type = ARG_ANYTHING,
600 .arg4_type = ARG_PTR_TO_MEM,
601 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
604 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
605 const void __user *, user_ptr)
607 int ret = copy_from_user(dst, user_ptr, size);
610 memset(dst, 0, size);
617 const struct bpf_func_proto bpf_copy_from_user_proto = {
618 .func = bpf_copy_from_user,
620 .ret_type = RET_INTEGER,
621 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
622 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
623 .arg3_type = ARG_ANYTHING,
626 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
628 if (cpu >= nr_cpu_ids)
629 return (unsigned long)NULL;
631 return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
634 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
635 .func = bpf_per_cpu_ptr,
637 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL,
638 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
639 .arg2_type = ARG_ANYTHING,
642 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
644 return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
647 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
648 .func = bpf_this_cpu_ptr,
650 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID,
651 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
654 const struct bpf_func_proto bpf_get_current_task_proto __weak;
655 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
656 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
657 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
658 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
660 const struct bpf_func_proto *
661 bpf_base_func_proto(enum bpf_func_id func_id)
664 case BPF_FUNC_map_lookup_elem:
665 return &bpf_map_lookup_elem_proto;
666 case BPF_FUNC_map_update_elem:
667 return &bpf_map_update_elem_proto;
668 case BPF_FUNC_map_delete_elem:
669 return &bpf_map_delete_elem_proto;
670 case BPF_FUNC_map_push_elem:
671 return &bpf_map_push_elem_proto;
672 case BPF_FUNC_map_pop_elem:
673 return &bpf_map_pop_elem_proto;
674 case BPF_FUNC_map_peek_elem:
675 return &bpf_map_peek_elem_proto;
676 case BPF_FUNC_get_prandom_u32:
677 return &bpf_get_prandom_u32_proto;
678 case BPF_FUNC_get_smp_processor_id:
679 return &bpf_get_raw_smp_processor_id_proto;
680 case BPF_FUNC_get_numa_node_id:
681 return &bpf_get_numa_node_id_proto;
682 case BPF_FUNC_tail_call:
683 return &bpf_tail_call_proto;
684 case BPF_FUNC_ktime_get_ns:
685 return &bpf_ktime_get_ns_proto;
686 case BPF_FUNC_ktime_get_boot_ns:
687 return &bpf_ktime_get_boot_ns_proto;
688 case BPF_FUNC_ringbuf_output:
689 return &bpf_ringbuf_output_proto;
690 case BPF_FUNC_ringbuf_reserve:
691 return &bpf_ringbuf_reserve_proto;
692 case BPF_FUNC_ringbuf_submit:
693 return &bpf_ringbuf_submit_proto;
694 case BPF_FUNC_ringbuf_discard:
695 return &bpf_ringbuf_discard_proto;
696 case BPF_FUNC_ringbuf_query:
697 return &bpf_ringbuf_query_proto;
706 case BPF_FUNC_spin_lock:
707 return &bpf_spin_lock_proto;
708 case BPF_FUNC_spin_unlock:
709 return &bpf_spin_unlock_proto;
710 case BPF_FUNC_trace_printk:
711 if (!perfmon_capable())
713 return bpf_get_trace_printk_proto();
714 case BPF_FUNC_snprintf_btf:
715 if (!perfmon_capable())
717 return &bpf_snprintf_btf_proto;
718 case BPF_FUNC_jiffies64:
719 return &bpf_jiffies64_proto;
720 case BPF_FUNC_bpf_per_cpu_ptr:
721 return &bpf_per_cpu_ptr_proto;
722 case BPF_FUNC_bpf_this_cpu_ptr:
723 return &bpf_this_cpu_ptr_proto;
728 if (!perfmon_capable())
732 case BPF_FUNC_get_current_task:
733 return &bpf_get_current_task_proto;
734 case BPF_FUNC_probe_read_user:
735 return &bpf_probe_read_user_proto;
736 case BPF_FUNC_probe_read_kernel:
737 return &bpf_probe_read_kernel_proto;
738 case BPF_FUNC_probe_read_user_str:
739 return &bpf_probe_read_user_str_proto;
740 case BPF_FUNC_probe_read_kernel_str:
741 return &bpf_probe_read_kernel_str_proto;