1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
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>
27 #include "../../lib/kstrtox.h"
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
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.
38 BPF_CALL_2(bpf_map_lookup_elem, struct bpf_map *, map, void *, key)
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);
45 const struct bpf_func_proto bpf_map_lookup_elem_proto = {
46 .func = bpf_map_lookup_elem,
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,
54 BPF_CALL_4(bpf_map_update_elem, struct bpf_map *, map, void *, key,
55 void *, value, u64, flags)
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);
62 const struct bpf_func_proto bpf_map_update_elem_proto = {
63 .func = bpf_map_update_elem,
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,
73 BPF_CALL_2(bpf_map_delete_elem, struct bpf_map *, map, void *, key)
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);
80 const struct bpf_func_proto bpf_map_delete_elem_proto = {
81 .func = bpf_map_delete_elem,
84 .ret_type = RET_INTEGER,
85 .arg1_type = ARG_CONST_MAP_PTR,
86 .arg2_type = ARG_PTR_TO_MAP_KEY,
89 BPF_CALL_3(bpf_map_push_elem, struct bpf_map *, map, void *, value, u64, flags)
91 return map->ops->map_push_elem(map, value, flags);
94 const struct bpf_func_proto bpf_map_push_elem_proto = {
95 .func = bpf_map_push_elem,
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,
104 BPF_CALL_2(bpf_map_pop_elem, struct bpf_map *, map, void *, value)
106 return map->ops->map_pop_elem(map, value);
109 const struct bpf_func_proto bpf_map_pop_elem_proto = {
110 .func = bpf_map_pop_elem,
112 .ret_type = RET_INTEGER,
113 .arg1_type = ARG_CONST_MAP_PTR,
114 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
117 BPF_CALL_2(bpf_map_peek_elem, struct bpf_map *, map, void *, value)
119 return map->ops->map_peek_elem(map, value);
122 const struct bpf_func_proto bpf_map_peek_elem_proto = {
123 .func = bpf_map_peek_elem,
125 .ret_type = RET_INTEGER,
126 .arg1_type = ARG_CONST_MAP_PTR,
127 .arg2_type = ARG_PTR_TO_MAP_VALUE | MEM_UNINIT,
130 BPF_CALL_3(bpf_map_lookup_percpu_elem, struct bpf_map *, map, void *, key, u32, cpu)
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);
136 const struct bpf_func_proto bpf_map_lookup_percpu_elem_proto = {
137 .func = bpf_map_lookup_percpu_elem,
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,
146 const struct bpf_func_proto bpf_get_prandom_u32_proto = {
147 .func = bpf_user_rnd_u32,
149 .ret_type = RET_INTEGER,
152 BPF_CALL_0(bpf_get_smp_processor_id)
154 return smp_processor_id();
157 const struct bpf_func_proto bpf_get_smp_processor_id_proto = {
158 .func = bpf_get_smp_processor_id,
160 .ret_type = RET_INTEGER,
163 BPF_CALL_0(bpf_get_numa_node_id)
165 return numa_node_id();
168 const struct bpf_func_proto bpf_get_numa_node_id_proto = {
169 .func = bpf_get_numa_node_id,
171 .ret_type = RET_INTEGER,
174 BPF_CALL_0(bpf_ktime_get_ns)
176 /* NMI safe access to clock monotonic */
177 return ktime_get_mono_fast_ns();
180 const struct bpf_func_proto bpf_ktime_get_ns_proto = {
181 .func = bpf_ktime_get_ns,
183 .ret_type = RET_INTEGER,
186 BPF_CALL_0(bpf_ktime_get_boot_ns)
188 /* NMI safe access to clock boottime */
189 return ktime_get_boot_fast_ns();
192 const struct bpf_func_proto bpf_ktime_get_boot_ns_proto = {
193 .func = bpf_ktime_get_boot_ns,
195 .ret_type = RET_INTEGER,
198 BPF_CALL_0(bpf_ktime_get_coarse_ns)
200 return ktime_get_coarse_ns();
203 const struct bpf_func_proto bpf_ktime_get_coarse_ns_proto = {
204 .func = bpf_ktime_get_coarse_ns,
206 .ret_type = RET_INTEGER,
209 BPF_CALL_0(bpf_ktime_get_tai_ns)
211 /* NMI safe access to clock tai */
212 return ktime_get_tai_fast_ns();
215 const struct bpf_func_proto bpf_ktime_get_tai_ns_proto = {
216 .func = bpf_ktime_get_tai_ns,
218 .ret_type = RET_INTEGER,
221 BPF_CALL_0(bpf_get_current_pid_tgid)
223 struct task_struct *task = current;
228 return (u64) task->tgid << 32 | task->pid;
231 const struct bpf_func_proto bpf_get_current_pid_tgid_proto = {
232 .func = bpf_get_current_pid_tgid,
234 .ret_type = RET_INTEGER,
237 BPF_CALL_0(bpf_get_current_uid_gid)
239 struct task_struct *task = current;
246 current_uid_gid(&uid, &gid);
247 return (u64) from_kgid(&init_user_ns, gid) << 32 |
248 from_kuid(&init_user_ns, uid);
251 const struct bpf_func_proto bpf_get_current_uid_gid_proto = {
252 .func = bpf_get_current_uid_gid,
254 .ret_type = RET_INTEGER,
257 BPF_CALL_2(bpf_get_current_comm, char *, buf, u32, size)
259 struct task_struct *task = current;
264 /* Verifier guarantees that size > 0 */
265 strscpy_pad(buf, task->comm, size);
268 memset(buf, 0, size);
272 const struct bpf_func_proto bpf_get_current_comm_proto = {
273 .func = bpf_get_current_comm,
275 .ret_type = RET_INTEGER,
276 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
277 .arg2_type = ARG_CONST_SIZE,
280 #if defined(CONFIG_QUEUED_SPINLOCKS) || defined(CONFIG_BPF_ARCH_SPINLOCK)
282 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
284 arch_spinlock_t *l = (void *)lock;
287 arch_spinlock_t lock;
288 } u = { .lock = __ARCH_SPIN_LOCK_UNLOCKED };
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));
297 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
299 arch_spinlock_t *l = (void *)lock;
307 static inline void __bpf_spin_lock(struct bpf_spin_lock *lock)
309 atomic_t *l = (void *)lock;
311 BUILD_BUG_ON(sizeof(*l) != sizeof(*lock));
313 atomic_cond_read_relaxed(l, !VAL);
314 } while (atomic_xchg(l, 1));
317 static inline void __bpf_spin_unlock(struct bpf_spin_lock *lock)
319 atomic_t *l = (void *)lock;
321 atomic_set_release(l, 0);
326 static DEFINE_PER_CPU(unsigned long, irqsave_flags);
328 static inline void __bpf_spin_lock_irqsave(struct bpf_spin_lock *lock)
332 local_irq_save(flags);
333 __bpf_spin_lock(lock);
334 __this_cpu_write(irqsave_flags, flags);
337 NOTRACE_BPF_CALL_1(bpf_spin_lock, struct bpf_spin_lock *, lock)
339 __bpf_spin_lock_irqsave(lock);
343 const struct bpf_func_proto bpf_spin_lock_proto = {
344 .func = bpf_spin_lock,
346 .ret_type = RET_VOID,
347 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
348 .arg1_btf_id = BPF_PTR_POISON,
351 static inline void __bpf_spin_unlock_irqrestore(struct bpf_spin_lock *lock)
355 flags = __this_cpu_read(irqsave_flags);
356 __bpf_spin_unlock(lock);
357 local_irq_restore(flags);
360 NOTRACE_BPF_CALL_1(bpf_spin_unlock, struct bpf_spin_lock *, lock)
362 __bpf_spin_unlock_irqrestore(lock);
366 const struct bpf_func_proto bpf_spin_unlock_proto = {
367 .func = bpf_spin_unlock,
369 .ret_type = RET_VOID,
370 .arg1_type = ARG_PTR_TO_SPIN_LOCK,
371 .arg1_btf_id = BPF_PTR_POISON,
374 void copy_map_value_locked(struct bpf_map *map, void *dst, void *src,
377 struct bpf_spin_lock *lock;
380 lock = src + map->record->spin_lock_off;
382 lock = dst + map->record->spin_lock_off;
384 __bpf_spin_lock_irqsave(lock);
385 copy_map_value(map, dst, src);
386 __bpf_spin_unlock_irqrestore(lock);
390 BPF_CALL_0(bpf_jiffies64)
392 return get_jiffies_64();
395 const struct bpf_func_proto bpf_jiffies64_proto = {
396 .func = bpf_jiffies64,
398 .ret_type = RET_INTEGER,
401 #ifdef CONFIG_CGROUPS
402 BPF_CALL_0(bpf_get_current_cgroup_id)
408 cgrp = task_dfl_cgroup(current);
409 cgrp_id = cgroup_id(cgrp);
415 const struct bpf_func_proto bpf_get_current_cgroup_id_proto = {
416 .func = bpf_get_current_cgroup_id,
418 .ret_type = RET_INTEGER,
421 BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
424 struct cgroup *ancestor;
428 cgrp = task_dfl_cgroup(current);
429 ancestor = cgroup_ancestor(cgrp, ancestor_level);
430 cgrp_id = ancestor ? cgroup_id(ancestor) : 0;
436 const struct bpf_func_proto bpf_get_current_ancestor_cgroup_id_proto = {
437 .func = bpf_get_current_ancestor_cgroup_id,
439 .ret_type = RET_INTEGER,
440 .arg1_type = ARG_ANYTHING,
442 #endif /* CONFIG_CGROUPS */
444 #define BPF_STRTOX_BASE_MASK 0x1F
446 static int __bpf_strtoull(const char *buf, size_t buf_len, u64 flags,
447 unsigned long long *res, bool *is_negative)
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;
456 if (!buf || !buf_len || !res || !is_negative)
459 if (base != 0 && base != 8 && base != 10 && base != 16)
462 if (flags & ~BPF_STRTOX_BASE_MASK)
465 while (cur_buf < buf + buf_len && isspace(*cur_buf))
468 *is_negative = (cur_buf < buf + buf_len && *cur_buf == '-');
472 consumed = cur_buf - buf;
477 cur_len = min(cur_len, sizeof(str) - 1);
478 memcpy(str, cur_buf, cur_len);
482 cur_buf = _parse_integer_fixup_radix(cur_buf, &base);
483 val_len = _parse_integer(cur_buf, base, res);
485 if (val_len & KSTRTOX_OVERFLOW)
492 consumed += cur_buf - str;
497 static int __bpf_strtoll(const char *buf, size_t buf_len, u64 flags,
500 unsigned long long _res;
504 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
508 if ((long long)-_res > 0)
512 if ((long long)_res < 0)
519 BPF_CALL_4(bpf_strtol, const char *, buf, size_t, buf_len, u64, flags,
525 err = __bpf_strtoll(buf, buf_len, flags, &_res);
528 if (_res != (long)_res)
534 const struct bpf_func_proto bpf_strtol_proto = {
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,
544 BPF_CALL_4(bpf_strtoul, const char *, buf, size_t, buf_len, u64, flags,
545 unsigned long *, res)
547 unsigned long long _res;
551 err = __bpf_strtoull(buf, buf_len, flags, &_res, &is_negative);
556 if (_res != (unsigned long)_res)
562 const struct bpf_func_proto bpf_strtoul_proto = {
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,
572 BPF_CALL_3(bpf_strncmp, const char *, s1, u32, s1_sz, const char *, s2)
574 return strncmp(s1, s2, s1_sz);
577 static const struct bpf_func_proto bpf_strncmp_proto = {
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,
586 BPF_CALL_4(bpf_get_ns_current_pid_tgid, u64, dev, u64, ino,
587 struct bpf_pidns_info *, nsdata, u32, size)
589 struct task_struct *task = current;
590 struct pid_namespace *pidns;
593 if (unlikely(size != sizeof(struct bpf_pidns_info)))
596 if (unlikely((u64)(dev_t)dev != dev))
602 pidns = task_active_pid_ns(task);
603 if (unlikely(!pidns)) {
608 if (!ns_match(&pidns->ns, (dev_t)dev, ino))
611 nsdata->pid = task_pid_nr_ns(task, pidns);
612 nsdata->tgid = task_tgid_nr_ns(task, pidns);
615 memset((void *)nsdata, 0, (size_t) size);
619 const struct bpf_func_proto bpf_get_ns_current_pid_tgid_proto = {
620 .func = bpf_get_ns_current_pid_tgid,
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,
629 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
630 .func = bpf_get_raw_cpu_id,
632 .ret_type = RET_INTEGER,
635 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map,
636 u64, flags, void *, data, u64, size)
638 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
641 return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
644 const struct bpf_func_proto bpf_event_output_data_proto = {
645 .func = bpf_event_output_data,
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,
655 BPF_CALL_3(bpf_copy_from_user, void *, dst, u32, size,
656 const void __user *, user_ptr)
658 int ret = copy_from_user(dst, user_ptr, size);
661 memset(dst, 0, size);
668 const struct bpf_func_proto bpf_copy_from_user_proto = {
669 .func = bpf_copy_from_user,
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,
678 BPF_CALL_5(bpf_copy_from_user_task, void *, dst, u32, size,
679 const void __user *, user_ptr, struct task_struct *, tsk, u64, flags)
683 /* flags is not used yet */
690 ret = access_process_vm(tsk, (unsigned long)user_ptr, dst, size, 0);
694 memset(dst, 0, size);
695 /* Return -EFAULT for partial read */
696 return ret < 0 ? ret : -EFAULT;
699 const struct bpf_func_proto bpf_copy_from_user_task_proto = {
700 .func = bpf_copy_from_user_task,
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
712 BPF_CALL_2(bpf_per_cpu_ptr, const void *, ptr, u32, cpu)
714 if (cpu >= nr_cpu_ids)
715 return (unsigned long)NULL;
717 return (unsigned long)per_cpu_ptr((const void __percpu *)ptr, cpu);
720 const struct bpf_func_proto bpf_per_cpu_ptr_proto = {
721 .func = bpf_per_cpu_ptr,
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,
728 BPF_CALL_1(bpf_this_cpu_ptr, const void *, percpu_ptr)
730 return (unsigned long)this_cpu_ptr((const void __percpu *)percpu_ptr);
733 const struct bpf_func_proto bpf_this_cpu_ptr_proto = {
734 .func = bpf_this_cpu_ptr,
736 .ret_type = RET_PTR_TO_MEM_OR_BTF_ID | MEM_RDONLY,
737 .arg1_type = ARG_PTR_TO_PERCPU_BTF_ID,
740 static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
743 void __user *user_ptr = (__force void __user *)unsafe_ptr;
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);
755 return strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
757 return strncpy_from_user_nofault(buf, user_ptr, bufsz);
763 /* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
764 * arguments representation.
766 #define MAX_BPRINTF_BIN_ARGS 512
768 /* Support executing three nested bprintf helper calls on a given CPU */
769 #define MAX_BPRINTF_NEST_LEVEL 3
770 struct bpf_bprintf_buffers {
771 char bin_args[MAX_BPRINTF_BIN_ARGS];
772 char buf[MAX_BPRINTF_BUF];
775 static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
776 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
778 static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
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);
789 *bufs = this_cpu_ptr(&bpf_bprintf_bufs[nest_level - 1]);
794 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
796 if (!data->bin_args && !data->buf)
798 if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
800 this_cpu_dec(bpf_bprintf_nest_level);
805 * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
807 * Returns a negative value if fmt is an invalid format string or 0 otherwise.
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.
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.
818 int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
819 u32 num_args, struct bpf_bprintf_data *data)
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;
827 char fmt_ptype, cur_ip[16], ip_spec[] = "%pXX";
829 fmt_end = strnchr(fmt, fmt_size, 0);
832 fmt_size = fmt_end - fmt;
834 if (get_buffers && try_get_buffers(&buffers))
837 if (data->get_bin_args) {
839 tmp_buf = buffers->bin_args;
840 tmp_buf_end = tmp_buf + MAX_BPRINTF_BIN_ARGS;
841 data->bin_args = (u32 *)tmp_buf;
845 data->buf = buffers->buf;
847 for (i = 0; i < fmt_size; i++) {
848 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
856 if (fmt[i + 1] == '%') {
861 if (num_spec >= num_args) {
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
871 /* skip optional "[0 +-][num]" width formatting field */
872 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
875 if (fmt[i] >= '1' && fmt[i] <= '9') {
877 while (fmt[i] >= '0' && fmt[i] <= '9')
882 sizeof_cur_arg = sizeof(long);
884 if ((fmt[i + 1] == 'k' || fmt[i + 1] == 'u') &&
886 fmt_ptype = fmt[i + 1];
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' ||
895 /* just kernel pointers */
897 cur_arg = raw_args[num_spec];
902 if (fmt[i + 1] == 'B') {
904 err = snprintf(tmp_buf,
905 (tmp_buf_end - tmp_buf),
907 (void *)(long)raw_args[num_spec]);
908 tmp_buf += (err + 1);
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')) {
927 sizeof_cur_ip = (fmt[i] == '4') ? 4 : 16;
928 if (tmp_buf_end - tmp_buf < sizeof_cur_ip) {
933 unsafe_ptr = (char *)(long)raw_args[num_spec];
934 err = copy_from_kernel_nofault(cur_ip, unsafe_ptr,
937 memset(cur_ip, 0, sizeof_cur_ip);
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.
943 ip_spec[2] = fmt[i - 1];
945 err = snprintf(tmp_buf, tmp_buf_end - tmp_buf,
952 } else if (fmt[i] == 's') {
955 if (fmt[i + 1] != 0 &&
956 !isspace(fmt[i + 1]) &&
957 !ispunct(fmt[i + 1])) {
965 if (tmp_buf_end == tmp_buf) {
970 unsafe_ptr = (char *)(long)raw_args[num_spec];
971 err = bpf_trace_copy_string(tmp_buf, unsafe_ptr,
973 tmp_buf_end - tmp_buf);
983 } else if (fmt[i] == 'c') {
987 if (tmp_buf_end == tmp_buf) {
992 *tmp_buf = raw_args[num_spec];
999 sizeof_cur_arg = sizeof(int);
1001 if (fmt[i] == 'l') {
1002 sizeof_cur_arg = sizeof(long);
1005 if (fmt[i] == 'l') {
1006 sizeof_cur_arg = sizeof(long long);
1010 if (fmt[i] != 'i' && fmt[i] != 'd' && fmt[i] != 'u' &&
1011 fmt[i] != 'x' && fmt[i] != 'X') {
1017 cur_arg = raw_args[num_spec];
1020 tmp_buf = PTR_ALIGN(tmp_buf, sizeof(u32));
1021 if (tmp_buf_end - tmp_buf < sizeof_cur_arg) {
1026 if (sizeof_cur_arg == 8) {
1027 *(u32 *)tmp_buf = *(u32 *)&cur_arg;
1028 *(u32 *)(tmp_buf + 4) = *((u32 *)&cur_arg + 1);
1030 *(u32 *)tmp_buf = (u32)(long)cur_arg;
1032 tmp_buf += sizeof_cur_arg;
1040 bpf_bprintf_cleanup(data);
1044 BPF_CALL_5(bpf_snprintf, char *, str, u32, str_size, char *, fmt,
1045 const void *, args, u32, data_len)
1047 struct bpf_bprintf_data data = {
1048 .get_bin_args = true,
1052 if (data_len % 8 || data_len > MAX_BPRINTF_VARARGS * 8 ||
1053 (data_len && !args))
1055 num_args = data_len / 8;
1057 /* ARG_PTR_TO_CONST_STR guarantees that fmt is zero-terminated so we
1058 * can safely give an unbounded size.
1060 err = bpf_bprintf_prepare(fmt, UINT_MAX, args, num_args, &data);
1064 err = bstr_printf(str, str_size, fmt, data.bin_args);
1066 bpf_bprintf_cleanup(&data);
1071 const struct bpf_func_proto bpf_snprintf_proto = {
1072 .func = bpf_snprintf,
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,
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.
1098 struct bpf_hrtimer {
1099 struct hrtimer timer;
1100 struct bpf_map *map;
1101 struct bpf_prog *prog;
1102 void __rcu *callback_fn;
1106 /* the actual struct hidden inside uapi struct bpf_timer */
1107 struct bpf_timer_kern {
1108 struct bpf_hrtimer *timer;
1109 /* bpf_spin_lock is used here instead of spinlock_t to make
1110 * sure that it always fits into space reserved by struct bpf_timer
1111 * regardless of LOCKDEP and spinlock debug flags.
1113 struct bpf_spin_lock lock;
1114 } __attribute__((aligned(8)));
1116 static DEFINE_PER_CPU(struct bpf_hrtimer *, hrtimer_running);
1118 static enum hrtimer_restart bpf_timer_cb(struct hrtimer *hrtimer)
1120 struct bpf_hrtimer *t = container_of(hrtimer, struct bpf_hrtimer, timer);
1121 struct bpf_map *map = t->map;
1122 void *value = t->value;
1123 bpf_callback_t callback_fn;
1127 BTF_TYPE_EMIT(struct bpf_timer);
1128 callback_fn = rcu_dereference_check(t->callback_fn, rcu_read_lock_bh_held());
1132 /* bpf_timer_cb() runs in hrtimer_run_softirq. It doesn't migrate and
1133 * cannot be preempted by another bpf_timer_cb() on the same cpu.
1134 * Remember the timer this callback is servicing to prevent
1135 * deadlock if callback_fn() calls bpf_timer_cancel() or
1136 * bpf_map_delete_elem() on the same timer.
1138 this_cpu_write(hrtimer_running, t);
1139 if (map->map_type == BPF_MAP_TYPE_ARRAY) {
1140 struct bpf_array *array = container_of(map, struct bpf_array, map);
1142 /* compute the key */
1143 idx = ((char *)value - array->value) / array->elem_size;
1145 } else { /* hash or lru */
1146 key = value - round_up(map->key_size, 8);
1149 callback_fn((u64)(long)map, (u64)(long)key, (u64)(long)value, 0, 0);
1150 /* The verifier checked that return value is zero. */
1152 this_cpu_write(hrtimer_running, NULL);
1154 return HRTIMER_NORESTART;
1157 BPF_CALL_3(bpf_timer_init, struct bpf_timer_kern *, timer, struct bpf_map *, map,
1160 clockid_t clockid = flags & (MAX_CLOCKS - 1);
1161 struct bpf_hrtimer *t;
1164 BUILD_BUG_ON(MAX_CLOCKS != 16);
1165 BUILD_BUG_ON(sizeof(struct bpf_timer_kern) > sizeof(struct bpf_timer));
1166 BUILD_BUG_ON(__alignof__(struct bpf_timer_kern) != __alignof__(struct bpf_timer));
1171 if (flags >= MAX_CLOCKS ||
1172 /* similar to timerfd except _ALARM variants are not supported */
1173 (clockid != CLOCK_MONOTONIC &&
1174 clockid != CLOCK_REALTIME &&
1175 clockid != CLOCK_BOOTTIME))
1177 __bpf_spin_lock_irqsave(&timer->lock);
1183 /* allocate hrtimer via map_kmalloc to use memcg accounting */
1184 t = bpf_map_kmalloc_node(map, sizeof(*t), GFP_ATOMIC, map->numa_node);
1189 t->value = (void *)timer - map->record->timer_off;
1192 rcu_assign_pointer(t->callback_fn, NULL);
1193 hrtimer_init(&t->timer, clockid, HRTIMER_MODE_REL_SOFT);
1194 t->timer.function = bpf_timer_cb;
1195 WRITE_ONCE(timer->timer, t);
1196 /* Guarantee the order between timer->timer and map->usercnt. So
1197 * when there are concurrent uref release and bpf timer init, either
1198 * bpf_timer_cancel_and_free() called by uref release reads a no-NULL
1199 * timer or atomic64_read() below returns a zero usercnt.
1202 if (!atomic64_read(&map->usercnt)) {
1203 /* maps with timers must be either held by user space
1204 * or pinned in bpffs.
1206 WRITE_ONCE(timer->timer, NULL);
1211 __bpf_spin_unlock_irqrestore(&timer->lock);
1215 static const struct bpf_func_proto bpf_timer_init_proto = {
1216 .func = bpf_timer_init,
1218 .ret_type = RET_INTEGER,
1219 .arg1_type = ARG_PTR_TO_TIMER,
1220 .arg2_type = ARG_CONST_MAP_PTR,
1221 .arg3_type = ARG_ANYTHING,
1224 BPF_CALL_3(bpf_timer_set_callback, struct bpf_timer_kern *, timer, void *, callback_fn,
1225 struct bpf_prog_aux *, aux)
1227 struct bpf_prog *prev, *prog = aux->prog;
1228 struct bpf_hrtimer *t;
1233 __bpf_spin_lock_irqsave(&timer->lock);
1239 if (!atomic64_read(&t->map->usercnt)) {
1240 /* maps with timers must be either held by user space
1241 * or pinned in bpffs. Otherwise timer might still be
1242 * running even when bpf prog is detached and user space
1243 * is gone, since map_release_uref won't ever be called.
1250 /* Bump prog refcnt once. Every bpf_timer_set_callback()
1251 * can pick different callback_fn-s within the same prog.
1253 prog = bpf_prog_inc_not_zero(prog);
1255 ret = PTR_ERR(prog);
1259 /* Drop prev prog refcnt when swapping with new prog */
1263 rcu_assign_pointer(t->callback_fn, callback_fn);
1265 __bpf_spin_unlock_irqrestore(&timer->lock);
1269 static const struct bpf_func_proto bpf_timer_set_callback_proto = {
1270 .func = bpf_timer_set_callback,
1272 .ret_type = RET_INTEGER,
1273 .arg1_type = ARG_PTR_TO_TIMER,
1274 .arg2_type = ARG_PTR_TO_FUNC,
1277 BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, flags)
1279 struct bpf_hrtimer *t;
1281 enum hrtimer_mode mode;
1285 if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
1287 __bpf_spin_lock_irqsave(&timer->lock);
1289 if (!t || !t->prog) {
1294 if (flags & BPF_F_TIMER_ABS)
1295 mode = HRTIMER_MODE_ABS_SOFT;
1297 mode = HRTIMER_MODE_REL_SOFT;
1299 if (flags & BPF_F_TIMER_CPU_PIN)
1300 mode |= HRTIMER_MODE_PINNED;
1302 hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
1304 __bpf_spin_unlock_irqrestore(&timer->lock);
1308 static const struct bpf_func_proto bpf_timer_start_proto = {
1309 .func = bpf_timer_start,
1311 .ret_type = RET_INTEGER,
1312 .arg1_type = ARG_PTR_TO_TIMER,
1313 .arg2_type = ARG_ANYTHING,
1314 .arg3_type = ARG_ANYTHING,
1317 static void drop_prog_refcnt(struct bpf_hrtimer *t)
1319 struct bpf_prog *prog = t->prog;
1324 rcu_assign_pointer(t->callback_fn, NULL);
1328 BPF_CALL_1(bpf_timer_cancel, struct bpf_timer_kern *, timer)
1330 struct bpf_hrtimer *t;
1335 __bpf_spin_lock_irqsave(&timer->lock);
1341 if (this_cpu_read(hrtimer_running) == t) {
1342 /* If bpf callback_fn is trying to bpf_timer_cancel()
1343 * its own timer the hrtimer_cancel() will deadlock
1344 * since it waits for callback_fn to finish
1349 drop_prog_refcnt(t);
1351 __bpf_spin_unlock_irqrestore(&timer->lock);
1352 /* Cancel the timer and wait for associated callback to finish
1353 * if it was running.
1355 ret = ret ?: hrtimer_cancel(&t->timer);
1359 static const struct bpf_func_proto bpf_timer_cancel_proto = {
1360 .func = bpf_timer_cancel,
1362 .ret_type = RET_INTEGER,
1363 .arg1_type = ARG_PTR_TO_TIMER,
1366 /* This function is called by map_delete/update_elem for individual element and
1367 * by ops->map_release_uref when the user space reference to a map reaches zero.
1369 void bpf_timer_cancel_and_free(void *val)
1371 struct bpf_timer_kern *timer = val;
1372 struct bpf_hrtimer *t;
1374 /* Performance optimization: read timer->timer without lock first. */
1375 if (!READ_ONCE(timer->timer))
1378 __bpf_spin_lock_irqsave(&timer->lock);
1379 /* re-read it under lock */
1383 drop_prog_refcnt(t);
1384 /* The subsequent bpf_timer_start/cancel() helpers won't be able to use
1385 * this timer, since it won't be initialized.
1387 WRITE_ONCE(timer->timer, NULL);
1389 __bpf_spin_unlock_irqrestore(&timer->lock);
1392 /* Cancel the timer and wait for callback to complete if it was running.
1393 * If hrtimer_cancel() can be safely called it's safe to call kfree(t)
1394 * right after for both preallocated and non-preallocated maps.
1395 * The timer->timer = NULL was already done and no code path can
1396 * see address 't' anymore.
1398 * Check that bpf_map_delete/update_elem() wasn't called from timer
1399 * callback_fn. In such case don't call hrtimer_cancel() (since it will
1400 * deadlock) and don't call hrtimer_try_to_cancel() (since it will just
1401 * return -1). Though callback_fn is still running on this cpu it's
1402 * safe to do kfree(t) because bpf_timer_cb() read everything it needed
1403 * from 't'. The bpf subprog callback_fn won't be able to access 't',
1404 * since timer->timer = NULL was already done. The timer will be
1405 * effectively cancelled because bpf_timer_cb() will return
1406 * HRTIMER_NORESTART.
1408 if (this_cpu_read(hrtimer_running) != t)
1409 hrtimer_cancel(&t->timer);
1413 BPF_CALL_2(bpf_kptr_xchg, void *, map_value, void *, ptr)
1415 unsigned long *kptr = map_value;
1417 /* This helper may be inlined by verifier. */
1418 return xchg(kptr, (unsigned long)ptr);
1421 /* Unlike other PTR_TO_BTF_ID helpers the btf_id in bpf_kptr_xchg()
1422 * helper is determined dynamically by the verifier. Use BPF_PTR_POISON to
1423 * denote type that verifier will determine.
1425 static const struct bpf_func_proto bpf_kptr_xchg_proto = {
1426 .func = bpf_kptr_xchg,
1428 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
1429 .ret_btf_id = BPF_PTR_POISON,
1430 .arg1_type = ARG_PTR_TO_KPTR,
1431 .arg2_type = ARG_PTR_TO_BTF_ID_OR_NULL | OBJ_RELEASE,
1432 .arg2_btf_id = BPF_PTR_POISON,
1435 /* Since the upper 8 bits of dynptr->size is reserved, the
1436 * maximum supported size is 2^24 - 1.
1438 #define DYNPTR_MAX_SIZE ((1UL << 24) - 1)
1439 #define DYNPTR_TYPE_SHIFT 28
1440 #define DYNPTR_SIZE_MASK 0xFFFFFF
1441 #define DYNPTR_RDONLY_BIT BIT(31)
1443 static bool __bpf_dynptr_is_rdonly(const struct bpf_dynptr_kern *ptr)
1445 return ptr->size & DYNPTR_RDONLY_BIT;
1448 void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
1450 ptr->size |= DYNPTR_RDONLY_BIT;
1453 static void bpf_dynptr_set_type(struct bpf_dynptr_kern *ptr, enum bpf_dynptr_type type)
1455 ptr->size |= type << DYNPTR_TYPE_SHIFT;
1458 static enum bpf_dynptr_type bpf_dynptr_get_type(const struct bpf_dynptr_kern *ptr)
1460 return (ptr->size & ~(DYNPTR_RDONLY_BIT)) >> DYNPTR_TYPE_SHIFT;
1463 u32 __bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
1465 return ptr->size & DYNPTR_SIZE_MASK;
1468 static void bpf_dynptr_set_size(struct bpf_dynptr_kern *ptr, u32 new_size)
1470 u32 metadata = ptr->size & ~DYNPTR_SIZE_MASK;
1472 ptr->size = new_size | metadata;
1475 int bpf_dynptr_check_size(u32 size)
1477 return size > DYNPTR_MAX_SIZE ? -E2BIG : 0;
1480 void bpf_dynptr_init(struct bpf_dynptr_kern *ptr, void *data,
1481 enum bpf_dynptr_type type, u32 offset, u32 size)
1484 ptr->offset = offset;
1486 bpf_dynptr_set_type(ptr, type);
1489 void bpf_dynptr_set_null(struct bpf_dynptr_kern *ptr)
1491 memset(ptr, 0, sizeof(*ptr));
1494 static int bpf_dynptr_check_off_len(const struct bpf_dynptr_kern *ptr, u32 offset, u32 len)
1496 u32 size = __bpf_dynptr_size(ptr);
1498 if (len > size || offset > size - len)
1504 BPF_CALL_4(bpf_dynptr_from_mem, void *, data, u32, size, u64, flags, struct bpf_dynptr_kern *, ptr)
1508 BTF_TYPE_EMIT(struct bpf_dynptr);
1510 err = bpf_dynptr_check_size(size);
1514 /* flags is currently unsupported */
1520 bpf_dynptr_init(ptr, data, BPF_DYNPTR_TYPE_LOCAL, 0, size);
1525 bpf_dynptr_set_null(ptr);
1529 static const struct bpf_func_proto bpf_dynptr_from_mem_proto = {
1530 .func = bpf_dynptr_from_mem,
1532 .ret_type = RET_INTEGER,
1533 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1534 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1535 .arg3_type = ARG_ANYTHING,
1536 .arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL | MEM_UNINIT,
1539 BPF_CALL_5(bpf_dynptr_read, void *, dst, u32, len, const struct bpf_dynptr_kern *, src,
1540 u32, offset, u64, flags)
1542 enum bpf_dynptr_type type;
1545 if (!src->data || flags)
1548 err = bpf_dynptr_check_off_len(src, offset, len);
1552 type = bpf_dynptr_get_type(src);
1555 case BPF_DYNPTR_TYPE_LOCAL:
1556 case BPF_DYNPTR_TYPE_RINGBUF:
1557 /* Source and destination may possibly overlap, hence use memmove to
1558 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1559 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1561 memmove(dst, src->data + src->offset + offset, len);
1563 case BPF_DYNPTR_TYPE_SKB:
1564 return __bpf_skb_load_bytes(src->data, src->offset + offset, dst, len);
1565 case BPF_DYNPTR_TYPE_XDP:
1566 return __bpf_xdp_load_bytes(src->data, src->offset + offset, dst, len);
1568 WARN_ONCE(true, "bpf_dynptr_read: unknown dynptr type %d\n", type);
1573 static const struct bpf_func_proto bpf_dynptr_read_proto = {
1574 .func = bpf_dynptr_read,
1576 .ret_type = RET_INTEGER,
1577 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1578 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1579 .arg3_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1580 .arg4_type = ARG_ANYTHING,
1581 .arg5_type = ARG_ANYTHING,
1584 BPF_CALL_5(bpf_dynptr_write, const struct bpf_dynptr_kern *, dst, u32, offset, void *, src,
1585 u32, len, u64, flags)
1587 enum bpf_dynptr_type type;
1590 if (!dst->data || __bpf_dynptr_is_rdonly(dst))
1593 err = bpf_dynptr_check_off_len(dst, offset, len);
1597 type = bpf_dynptr_get_type(dst);
1600 case BPF_DYNPTR_TYPE_LOCAL:
1601 case BPF_DYNPTR_TYPE_RINGBUF:
1604 /* Source and destination may possibly overlap, hence use memmove to
1605 * copy the data. E.g. bpf_dynptr_from_mem may create two dynptr
1606 * pointing to overlapping PTR_TO_MAP_VALUE regions.
1608 memmove(dst->data + dst->offset + offset, src, len);
1610 case BPF_DYNPTR_TYPE_SKB:
1611 return __bpf_skb_store_bytes(dst->data, dst->offset + offset, src, len,
1613 case BPF_DYNPTR_TYPE_XDP:
1616 return __bpf_xdp_store_bytes(dst->data, dst->offset + offset, src, len);
1618 WARN_ONCE(true, "bpf_dynptr_write: unknown dynptr type %d\n", type);
1623 static const struct bpf_func_proto bpf_dynptr_write_proto = {
1624 .func = bpf_dynptr_write,
1626 .ret_type = RET_INTEGER,
1627 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1628 .arg2_type = ARG_ANYTHING,
1629 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1630 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
1631 .arg5_type = ARG_ANYTHING,
1634 BPF_CALL_3(bpf_dynptr_data, const struct bpf_dynptr_kern *, ptr, u32, offset, u32, len)
1636 enum bpf_dynptr_type type;
1642 err = bpf_dynptr_check_off_len(ptr, offset, len);
1646 if (__bpf_dynptr_is_rdonly(ptr))
1649 type = bpf_dynptr_get_type(ptr);
1652 case BPF_DYNPTR_TYPE_LOCAL:
1653 case BPF_DYNPTR_TYPE_RINGBUF:
1654 return (unsigned long)(ptr->data + ptr->offset + offset);
1655 case BPF_DYNPTR_TYPE_SKB:
1656 case BPF_DYNPTR_TYPE_XDP:
1657 /* skb and xdp dynptrs should use bpf_dynptr_slice / bpf_dynptr_slice_rdwr */
1660 WARN_ONCE(true, "bpf_dynptr_data: unknown dynptr type %d\n", type);
1665 static const struct bpf_func_proto bpf_dynptr_data_proto = {
1666 .func = bpf_dynptr_data,
1668 .ret_type = RET_PTR_TO_DYNPTR_MEM_OR_NULL,
1669 .arg1_type = ARG_PTR_TO_DYNPTR | MEM_RDONLY,
1670 .arg2_type = ARG_ANYTHING,
1671 .arg3_type = ARG_CONST_ALLOC_SIZE_OR_ZERO,
1674 const struct bpf_func_proto bpf_get_current_task_proto __weak;
1675 const struct bpf_func_proto bpf_get_current_task_btf_proto __weak;
1676 const struct bpf_func_proto bpf_probe_read_user_proto __weak;
1677 const struct bpf_func_proto bpf_probe_read_user_str_proto __weak;
1678 const struct bpf_func_proto bpf_probe_read_kernel_proto __weak;
1679 const struct bpf_func_proto bpf_probe_read_kernel_str_proto __weak;
1680 const struct bpf_func_proto bpf_task_pt_regs_proto __weak;
1682 const struct bpf_func_proto *
1683 bpf_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1686 case BPF_FUNC_map_lookup_elem:
1687 return &bpf_map_lookup_elem_proto;
1688 case BPF_FUNC_map_update_elem:
1689 return &bpf_map_update_elem_proto;
1690 case BPF_FUNC_map_delete_elem:
1691 return &bpf_map_delete_elem_proto;
1692 case BPF_FUNC_map_push_elem:
1693 return &bpf_map_push_elem_proto;
1694 case BPF_FUNC_map_pop_elem:
1695 return &bpf_map_pop_elem_proto;
1696 case BPF_FUNC_map_peek_elem:
1697 return &bpf_map_peek_elem_proto;
1698 case BPF_FUNC_map_lookup_percpu_elem:
1699 return &bpf_map_lookup_percpu_elem_proto;
1700 case BPF_FUNC_get_prandom_u32:
1701 return &bpf_get_prandom_u32_proto;
1702 case BPF_FUNC_get_smp_processor_id:
1703 return &bpf_get_raw_smp_processor_id_proto;
1704 case BPF_FUNC_get_numa_node_id:
1705 return &bpf_get_numa_node_id_proto;
1706 case BPF_FUNC_tail_call:
1707 return &bpf_tail_call_proto;
1708 case BPF_FUNC_ktime_get_ns:
1709 return &bpf_ktime_get_ns_proto;
1710 case BPF_FUNC_ktime_get_boot_ns:
1711 return &bpf_ktime_get_boot_ns_proto;
1712 case BPF_FUNC_ktime_get_tai_ns:
1713 return &bpf_ktime_get_tai_ns_proto;
1714 case BPF_FUNC_ringbuf_output:
1715 return &bpf_ringbuf_output_proto;
1716 case BPF_FUNC_ringbuf_reserve:
1717 return &bpf_ringbuf_reserve_proto;
1718 case BPF_FUNC_ringbuf_submit:
1719 return &bpf_ringbuf_submit_proto;
1720 case BPF_FUNC_ringbuf_discard:
1721 return &bpf_ringbuf_discard_proto;
1722 case BPF_FUNC_ringbuf_query:
1723 return &bpf_ringbuf_query_proto;
1724 case BPF_FUNC_strncmp:
1725 return &bpf_strncmp_proto;
1726 case BPF_FUNC_strtol:
1727 return &bpf_strtol_proto;
1728 case BPF_FUNC_strtoul:
1729 return &bpf_strtoul_proto;
1734 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
1738 case BPF_FUNC_spin_lock:
1739 return &bpf_spin_lock_proto;
1740 case BPF_FUNC_spin_unlock:
1741 return &bpf_spin_unlock_proto;
1742 case BPF_FUNC_jiffies64:
1743 return &bpf_jiffies64_proto;
1744 case BPF_FUNC_per_cpu_ptr:
1745 return &bpf_per_cpu_ptr_proto;
1746 case BPF_FUNC_this_cpu_ptr:
1747 return &bpf_this_cpu_ptr_proto;
1748 case BPF_FUNC_timer_init:
1749 return &bpf_timer_init_proto;
1750 case BPF_FUNC_timer_set_callback:
1751 return &bpf_timer_set_callback_proto;
1752 case BPF_FUNC_timer_start:
1753 return &bpf_timer_start_proto;
1754 case BPF_FUNC_timer_cancel:
1755 return &bpf_timer_cancel_proto;
1756 case BPF_FUNC_kptr_xchg:
1757 return &bpf_kptr_xchg_proto;
1758 case BPF_FUNC_for_each_map_elem:
1759 return &bpf_for_each_map_elem_proto;
1761 return &bpf_loop_proto;
1762 case BPF_FUNC_user_ringbuf_drain:
1763 return &bpf_user_ringbuf_drain_proto;
1764 case BPF_FUNC_ringbuf_reserve_dynptr:
1765 return &bpf_ringbuf_reserve_dynptr_proto;
1766 case BPF_FUNC_ringbuf_submit_dynptr:
1767 return &bpf_ringbuf_submit_dynptr_proto;
1768 case BPF_FUNC_ringbuf_discard_dynptr:
1769 return &bpf_ringbuf_discard_dynptr_proto;
1770 case BPF_FUNC_dynptr_from_mem:
1771 return &bpf_dynptr_from_mem_proto;
1772 case BPF_FUNC_dynptr_read:
1773 return &bpf_dynptr_read_proto;
1774 case BPF_FUNC_dynptr_write:
1775 return &bpf_dynptr_write_proto;
1776 case BPF_FUNC_dynptr_data:
1777 return &bpf_dynptr_data_proto;
1778 #ifdef CONFIG_CGROUPS
1779 case BPF_FUNC_cgrp_storage_get:
1780 return &bpf_cgrp_storage_get_proto;
1781 case BPF_FUNC_cgrp_storage_delete:
1782 return &bpf_cgrp_storage_delete_proto;
1783 case BPF_FUNC_get_current_cgroup_id:
1784 return &bpf_get_current_cgroup_id_proto;
1785 case BPF_FUNC_get_current_ancestor_cgroup_id:
1786 return &bpf_get_current_ancestor_cgroup_id_proto;
1792 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
1796 case BPF_FUNC_trace_printk:
1797 return bpf_get_trace_printk_proto();
1798 case BPF_FUNC_get_current_task:
1799 return &bpf_get_current_task_proto;
1800 case BPF_FUNC_get_current_task_btf:
1801 return &bpf_get_current_task_btf_proto;
1802 case BPF_FUNC_probe_read_user:
1803 return &bpf_probe_read_user_proto;
1804 case BPF_FUNC_probe_read_kernel:
1805 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1806 NULL : &bpf_probe_read_kernel_proto;
1807 case BPF_FUNC_probe_read_user_str:
1808 return &bpf_probe_read_user_str_proto;
1809 case BPF_FUNC_probe_read_kernel_str:
1810 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1811 NULL : &bpf_probe_read_kernel_str_proto;
1812 case BPF_FUNC_snprintf_btf:
1813 return &bpf_snprintf_btf_proto;
1814 case BPF_FUNC_snprintf:
1815 return &bpf_snprintf_proto;
1816 case BPF_FUNC_task_pt_regs:
1817 return &bpf_task_pt_regs_proto;
1818 case BPF_FUNC_trace_vprintk:
1819 return bpf_get_trace_vprintk_proto();
1825 void bpf_list_head_free(const struct btf_field *field, void *list_head,
1826 struct bpf_spin_lock *spin_lock)
1828 struct list_head *head = list_head, *orig_head = list_head;
1830 BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct bpf_list_head));
1831 BUILD_BUG_ON(__alignof__(struct list_head) > __alignof__(struct bpf_list_head));
1833 /* Do the actual list draining outside the lock to not hold the lock for
1834 * too long, and also prevent deadlocks if tracing programs end up
1835 * executing on entry/exit of functions called inside the critical
1836 * section, and end up doing map ops that call bpf_list_head_free for
1837 * the same map value again.
1839 __bpf_spin_lock_irqsave(spin_lock);
1840 if (!head->next || list_empty(head))
1844 INIT_LIST_HEAD(orig_head);
1845 __bpf_spin_unlock_irqrestore(spin_lock);
1847 while (head != orig_head) {
1850 obj -= field->graph_root.node_offset;
1852 /* The contained type can also have resources, including a
1853 * bpf_list_head which needs to be freed.
1856 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
1861 /* Like rbtree_postorder_for_each_entry_safe, but 'pos' and 'n' are
1862 * 'rb_node *', so field name of rb_node within containing struct is not
1865 * Since bpf_rb_tree's node type has a corresponding struct btf_field with
1866 * graph_root.node_offset, it's not necessary to know field name
1867 * or type of node struct
1869 #define bpf_rbtree_postorder_for_each_entry_safe(pos, n, root) \
1870 for (pos = rb_first_postorder(root); \
1871 pos && ({ n = rb_next_postorder(pos); 1; }); \
1874 void bpf_rb_root_free(const struct btf_field *field, void *rb_root,
1875 struct bpf_spin_lock *spin_lock)
1877 struct rb_root_cached orig_root, *root = rb_root;
1878 struct rb_node *pos, *n;
1881 BUILD_BUG_ON(sizeof(struct rb_root_cached) > sizeof(struct bpf_rb_root));
1882 BUILD_BUG_ON(__alignof__(struct rb_root_cached) > __alignof__(struct bpf_rb_root));
1884 __bpf_spin_lock_irqsave(spin_lock);
1886 *root = RB_ROOT_CACHED;
1887 __bpf_spin_unlock_irqrestore(spin_lock);
1889 bpf_rbtree_postorder_for_each_entry_safe(pos, n, &orig_root.rb_root) {
1891 obj -= field->graph_root.node_offset;
1895 __bpf_obj_drop_impl(obj, field->graph_root.value_rec, false);
1900 __bpf_kfunc_start_defs();
1902 __bpf_kfunc void *bpf_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1904 struct btf_struct_meta *meta = meta__ign;
1905 u64 size = local_type_id__k;
1908 p = bpf_mem_alloc(&bpf_global_ma, size);
1912 bpf_obj_init(meta->record, p);
1916 __bpf_kfunc void *bpf_percpu_obj_new_impl(u64 local_type_id__k, void *meta__ign)
1918 u64 size = local_type_id__k;
1920 /* The verifier has ensured that meta__ign must be NULL */
1921 return bpf_mem_alloc(&bpf_global_percpu_ma, size);
1924 /* Must be called under migrate_disable(), as required by bpf_mem_free */
1925 void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu)
1927 struct bpf_mem_alloc *ma;
1929 if (rec && rec->refcount_off >= 0 &&
1930 !refcount_dec_and_test((refcount_t *)(p + rec->refcount_off))) {
1931 /* Object is refcounted and refcount_dec didn't result in 0
1932 * refcount. Return without freeing the object
1938 bpf_obj_free_fields(rec, p);
1941 ma = &bpf_global_percpu_ma;
1943 ma = &bpf_global_ma;
1944 bpf_mem_free_rcu(ma, p);
1947 __bpf_kfunc void bpf_obj_drop_impl(void *p__alloc, void *meta__ign)
1949 struct btf_struct_meta *meta = meta__ign;
1952 __bpf_obj_drop_impl(p, meta ? meta->record : NULL, false);
1955 __bpf_kfunc void bpf_percpu_obj_drop_impl(void *p__alloc, void *meta__ign)
1957 /* The verifier has ensured that meta__ign must be NULL */
1958 bpf_mem_free_rcu(&bpf_global_percpu_ma, p__alloc);
1961 __bpf_kfunc void *bpf_refcount_acquire_impl(void *p__refcounted_kptr, void *meta__ign)
1963 struct btf_struct_meta *meta = meta__ign;
1964 struct bpf_refcount *ref;
1966 /* Could just cast directly to refcount_t *, but need some code using
1967 * bpf_refcount type so that it is emitted in vmlinux BTF
1969 ref = (struct bpf_refcount *)(p__refcounted_kptr + meta->record->refcount_off);
1970 if (!refcount_inc_not_zero((refcount_t *)ref))
1973 /* Verifier strips KF_RET_NULL if input is owned ref, see is_kfunc_ret_null
1976 return (void *)p__refcounted_kptr;
1979 static int __bpf_list_add(struct bpf_list_node_kern *node,
1980 struct bpf_list_head *head,
1981 bool tail, struct btf_record *rec, u64 off)
1983 struct list_head *n = &node->list_head, *h = (void *)head;
1985 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
1986 * called on its fields, so init here
1988 if (unlikely(!h->next))
1991 /* node->owner != NULL implies !list_empty(n), no need to separately
1994 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
1995 /* Only called from BPF prog, no need to migrate_disable */
1996 __bpf_obj_drop_impl((void *)n - off, rec, false);
2000 tail ? list_add_tail(n, h) : list_add(n, h);
2001 WRITE_ONCE(node->owner, head);
2006 __bpf_kfunc int bpf_list_push_front_impl(struct bpf_list_head *head,
2007 struct bpf_list_node *node,
2008 void *meta__ign, u64 off)
2010 struct bpf_list_node_kern *n = (void *)node;
2011 struct btf_struct_meta *meta = meta__ign;
2013 return __bpf_list_add(n, head, false, meta ? meta->record : NULL, off);
2016 __bpf_kfunc int bpf_list_push_back_impl(struct bpf_list_head *head,
2017 struct bpf_list_node *node,
2018 void *meta__ign, u64 off)
2020 struct bpf_list_node_kern *n = (void *)node;
2021 struct btf_struct_meta *meta = meta__ign;
2023 return __bpf_list_add(n, head, true, meta ? meta->record : NULL, off);
2026 static struct bpf_list_node *__bpf_list_del(struct bpf_list_head *head, bool tail)
2028 struct list_head *n, *h = (void *)head;
2029 struct bpf_list_node_kern *node;
2031 /* If list_head was 0-initialized by map, bpf_obj_init_field wasn't
2032 * called on its fields, so init here
2034 if (unlikely(!h->next))
2039 n = tail ? h->prev : h->next;
2040 node = container_of(n, struct bpf_list_node_kern, list_head);
2041 if (WARN_ON_ONCE(READ_ONCE(node->owner) != head))
2045 WRITE_ONCE(node->owner, NULL);
2046 return (struct bpf_list_node *)n;
2049 __bpf_kfunc struct bpf_list_node *bpf_list_pop_front(struct bpf_list_head *head)
2051 return __bpf_list_del(head, false);
2054 __bpf_kfunc struct bpf_list_node *bpf_list_pop_back(struct bpf_list_head *head)
2056 return __bpf_list_del(head, true);
2059 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_remove(struct bpf_rb_root *root,
2060 struct bpf_rb_node *node)
2062 struct bpf_rb_node_kern *node_internal = (struct bpf_rb_node_kern *)node;
2063 struct rb_root_cached *r = (struct rb_root_cached *)root;
2064 struct rb_node *n = &node_internal->rb_node;
2066 /* node_internal->owner != root implies either RB_EMPTY_NODE(n) or
2067 * n is owned by some other tree. No need to check RB_EMPTY_NODE(n)
2069 if (READ_ONCE(node_internal->owner) != root)
2072 rb_erase_cached(n, r);
2074 WRITE_ONCE(node_internal->owner, NULL);
2075 return (struct bpf_rb_node *)n;
2078 /* Need to copy rbtree_add_cached's logic here because our 'less' is a BPF
2081 static int __bpf_rbtree_add(struct bpf_rb_root *root,
2082 struct bpf_rb_node_kern *node,
2083 void *less, struct btf_record *rec, u64 off)
2085 struct rb_node **link = &((struct rb_root_cached *)root)->rb_root.rb_node;
2086 struct rb_node *parent = NULL, *n = &node->rb_node;
2087 bpf_callback_t cb = (bpf_callback_t)less;
2088 bool leftmost = true;
2090 /* node->owner != NULL implies !RB_EMPTY_NODE(n), no need to separately
2093 if (cmpxchg(&node->owner, NULL, BPF_PTR_POISON)) {
2094 /* Only called from BPF prog, no need to migrate_disable */
2095 __bpf_obj_drop_impl((void *)n - off, rec, false);
2101 if (cb((uintptr_t)node, (uintptr_t)parent, 0, 0, 0)) {
2102 link = &parent->rb_left;
2104 link = &parent->rb_right;
2109 rb_link_node(n, parent, link);
2110 rb_insert_color_cached(n, (struct rb_root_cached *)root, leftmost);
2111 WRITE_ONCE(node->owner, root);
2115 __bpf_kfunc int bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node,
2116 bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b),
2117 void *meta__ign, u64 off)
2119 struct btf_struct_meta *meta = meta__ign;
2120 struct bpf_rb_node_kern *n = (void *)node;
2122 return __bpf_rbtree_add(root, n, (void *)less, meta ? meta->record : NULL, off);
2125 __bpf_kfunc struct bpf_rb_node *bpf_rbtree_first(struct bpf_rb_root *root)
2127 struct rb_root_cached *r = (struct rb_root_cached *)root;
2129 return (struct bpf_rb_node *)rb_first_cached(r);
2133 * bpf_task_acquire - Acquire a reference to a task. A task acquired by this
2134 * kfunc which is not stored in a map as a kptr, must be released by calling
2135 * bpf_task_release().
2136 * @p: The task on which a reference is being acquired.
2138 __bpf_kfunc struct task_struct *bpf_task_acquire(struct task_struct *p)
2140 if (refcount_inc_not_zero(&p->rcu_users))
2146 * bpf_task_release - Release the reference acquired on a task.
2147 * @p: The task on which a reference is being released.
2149 __bpf_kfunc void bpf_task_release(struct task_struct *p)
2151 put_task_struct_rcu_user(p);
2154 __bpf_kfunc void bpf_task_release_dtor(void *p)
2156 put_task_struct_rcu_user(p);
2158 CFI_NOSEAL(bpf_task_release_dtor);
2160 #ifdef CONFIG_CGROUPS
2162 * bpf_cgroup_acquire - Acquire a reference to a cgroup. A cgroup acquired by
2163 * this kfunc which is not stored in a map as a kptr, must be released by
2164 * calling bpf_cgroup_release().
2165 * @cgrp: The cgroup on which a reference is being acquired.
2167 __bpf_kfunc struct cgroup *bpf_cgroup_acquire(struct cgroup *cgrp)
2169 return cgroup_tryget(cgrp) ? cgrp : NULL;
2173 * bpf_cgroup_release - Release the reference acquired on a cgroup.
2174 * If this kfunc is invoked in an RCU read region, the cgroup is guaranteed to
2175 * not be freed until the current grace period has ended, even if its refcount
2177 * @cgrp: The cgroup on which a reference is being released.
2179 __bpf_kfunc void bpf_cgroup_release(struct cgroup *cgrp)
2184 __bpf_kfunc void bpf_cgroup_release_dtor(void *cgrp)
2188 CFI_NOSEAL(bpf_cgroup_release_dtor);
2191 * bpf_cgroup_ancestor - Perform a lookup on an entry in a cgroup's ancestor
2192 * array. A cgroup returned by this kfunc which is not subsequently stored in a
2193 * map, must be released by calling bpf_cgroup_release().
2194 * @cgrp: The cgroup for which we're performing a lookup.
2195 * @level: The level of ancestor to look up.
2197 __bpf_kfunc struct cgroup *bpf_cgroup_ancestor(struct cgroup *cgrp, int level)
2199 struct cgroup *ancestor;
2201 if (level > cgrp->level || level < 0)
2204 /* cgrp's refcnt could be 0 here, but ancestors can still be accessed */
2205 ancestor = cgrp->ancestors[level];
2206 if (!cgroup_tryget(ancestor))
2212 * bpf_cgroup_from_id - Find a cgroup from its ID. A cgroup returned by this
2213 * kfunc which is not subsequently stored in a map, must be released by calling
2214 * bpf_cgroup_release().
2217 __bpf_kfunc struct cgroup *bpf_cgroup_from_id(u64 cgid)
2219 struct cgroup *cgrp;
2221 cgrp = cgroup_get_from_id(cgid);
2228 * bpf_task_under_cgroup - wrap task_under_cgroup_hierarchy() as a kfunc, test
2229 * task's membership of cgroup ancestry.
2230 * @task: the task to be tested
2231 * @ancestor: possible ancestor of @task's cgroup
2233 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor.
2234 * It follows all the same rules as cgroup_is_descendant, and only applies
2235 * to the default hierarchy.
2237 __bpf_kfunc long bpf_task_under_cgroup(struct task_struct *task,
2238 struct cgroup *ancestor)
2243 ret = task_under_cgroup_hierarchy(task, ancestor);
2249 * bpf_task_get_cgroup1 - Acquires the associated cgroup of a task within a
2250 * specific cgroup1 hierarchy. The cgroup1 hierarchy is identified by its
2252 * @task: The target task
2253 * @hierarchy_id: The ID of a cgroup1 hierarchy
2255 * On success, the cgroup is returen. On failure, NULL is returned.
2257 __bpf_kfunc struct cgroup *
2258 bpf_task_get_cgroup1(struct task_struct *task, int hierarchy_id)
2260 struct cgroup *cgrp = task_get_cgroup1(task, hierarchy_id);
2266 #endif /* CONFIG_CGROUPS */
2269 * bpf_task_from_pid - Find a struct task_struct from its pid by looking it up
2270 * in the root pid namespace idr. If a task is returned, it must either be
2271 * stored in a map, or released with bpf_task_release().
2272 * @pid: The pid of the task being looked up.
2274 __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
2276 struct task_struct *p;
2279 p = find_task_by_pid_ns(pid, &init_pid_ns);
2281 p = bpf_task_acquire(p);
2288 * bpf_dynptr_slice() - Obtain a read-only pointer to the dynptr data.
2289 * @ptr: The dynptr whose data slice to retrieve
2290 * @offset: Offset into the dynptr
2291 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2292 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2293 * length of the requested slice. This must be a constant.
2295 * For non-skb and non-xdp type dynptrs, there is no difference between
2296 * bpf_dynptr_slice and bpf_dynptr_data.
2298 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2300 * If the intention is to write to the data slice, please use
2301 * bpf_dynptr_slice_rdwr.
2303 * The user must check that the returned pointer is not null before using it.
2305 * Please note that in the case of skb and xdp dynptrs, bpf_dynptr_slice
2306 * does not change the underlying packet data pointers, so a call to
2307 * bpf_dynptr_slice will not invalidate any ctx->data/data_end pointers in
2310 * Return: NULL if the call failed (eg invalid dynptr), pointer to a read-only
2311 * data slice (can be either direct pointer to the data or a pointer to the user
2312 * provided buffer, with its contents containing the data, if unable to obtain
2315 __bpf_kfunc void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset,
2316 void *buffer__opt, u32 buffer__szk)
2318 enum bpf_dynptr_type type;
2319 u32 len = buffer__szk;
2325 err = bpf_dynptr_check_off_len(ptr, offset, len);
2329 type = bpf_dynptr_get_type(ptr);
2332 case BPF_DYNPTR_TYPE_LOCAL:
2333 case BPF_DYNPTR_TYPE_RINGBUF:
2334 return ptr->data + ptr->offset + offset;
2335 case BPF_DYNPTR_TYPE_SKB:
2337 return skb_header_pointer(ptr->data, ptr->offset + offset, len, buffer__opt);
2339 return skb_pointer_if_linear(ptr->data, ptr->offset + offset, len);
2340 case BPF_DYNPTR_TYPE_XDP:
2342 void *xdp_ptr = bpf_xdp_pointer(ptr->data, ptr->offset + offset, len);
2343 if (!IS_ERR_OR_NULL(xdp_ptr))
2348 bpf_xdp_copy_buf(ptr->data, ptr->offset + offset, buffer__opt, len, false);
2352 WARN_ONCE(true, "unknown dynptr type %d\n", type);
2358 * bpf_dynptr_slice_rdwr() - Obtain a writable pointer to the dynptr data.
2359 * @ptr: The dynptr whose data slice to retrieve
2360 * @offset: Offset into the dynptr
2361 * @buffer__opt: User-provided buffer to copy contents into. May be NULL
2362 * @buffer__szk: Size (in bytes) of the buffer if present. This is the
2363 * length of the requested slice. This must be a constant.
2365 * For non-skb and non-xdp type dynptrs, there is no difference between
2366 * bpf_dynptr_slice and bpf_dynptr_data.
2368 * If buffer__opt is NULL, the call will fail if buffer_opt was needed.
2370 * The returned pointer is writable and may point to either directly the dynptr
2371 * data at the requested offset or to the buffer if unable to obtain a direct
2372 * data pointer to (example: the requested slice is to the paged area of an skb
2373 * packet). In the case where the returned pointer is to the buffer, the user
2374 * is responsible for persisting writes through calling bpf_dynptr_write(). This
2375 * usually looks something like this pattern:
2377 * struct eth_hdr *eth = bpf_dynptr_slice_rdwr(&dynptr, 0, buffer, sizeof(buffer));
2379 * return TC_ACT_SHOT;
2381 * // mutate eth header //
2383 * if (eth == buffer)
2384 * bpf_dynptr_write(&ptr, 0, buffer, sizeof(buffer), 0);
2386 * Please note that, as in the example above, the user must check that the
2387 * returned pointer is not null before using it.
2389 * Please also note that in the case of skb and xdp dynptrs, bpf_dynptr_slice_rdwr
2390 * does not change the underlying packet data pointers, so a call to
2391 * bpf_dynptr_slice_rdwr will not invalidate any ctx->data/data_end pointers in
2394 * Return: NULL if the call failed (eg invalid dynptr), pointer to a
2395 * data slice (can be either direct pointer to the data or a pointer to the user
2396 * provided buffer, with its contents containing the data, if unable to obtain
2399 __bpf_kfunc void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset,
2400 void *buffer__opt, u32 buffer__szk)
2402 if (!ptr->data || __bpf_dynptr_is_rdonly(ptr))
2405 /* bpf_dynptr_slice_rdwr is the same logic as bpf_dynptr_slice.
2407 * For skb-type dynptrs, it is safe to write into the returned pointer
2408 * if the bpf program allows skb data writes. There are two possiblities
2409 * that may occur when calling bpf_dynptr_slice_rdwr:
2411 * 1) The requested slice is in the head of the skb. In this case, the
2412 * returned pointer is directly to skb data, and if the skb is cloned, the
2413 * verifier will have uncloned it (see bpf_unclone_prologue()) already.
2414 * The pointer can be directly written into.
2416 * 2) Some portion of the requested slice is in the paged buffer area.
2417 * In this case, the requested data will be copied out into the buffer
2418 * and the returned pointer will be a pointer to the buffer. The skb
2419 * will not be pulled. To persist the write, the user will need to call
2420 * bpf_dynptr_write(), which will pull the skb and commit the write.
2422 * Similarly for xdp programs, if the requested slice is not across xdp
2423 * fragments, then a direct pointer will be returned, otherwise the data
2424 * will be copied out into the buffer and the user will need to call
2425 * bpf_dynptr_write() to commit changes.
2427 return bpf_dynptr_slice(ptr, offset, buffer__opt, buffer__szk);
2430 __bpf_kfunc int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end)
2434 if (!ptr->data || start > end)
2437 size = __bpf_dynptr_size(ptr);
2439 if (start > size || end > size)
2442 ptr->offset += start;
2443 bpf_dynptr_set_size(ptr, end - start);
2448 __bpf_kfunc bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr)
2453 __bpf_kfunc bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr)
2458 return __bpf_dynptr_is_rdonly(ptr);
2461 __bpf_kfunc __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr)
2466 return __bpf_dynptr_size(ptr);
2469 __bpf_kfunc int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr,
2470 struct bpf_dynptr_kern *clone__uninit)
2473 bpf_dynptr_set_null(clone__uninit);
2477 *clone__uninit = *ptr;
2482 __bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
2487 __bpf_kfunc void *bpf_rdonly_cast(const void *obj__ign, u32 btf_id__k)
2489 return (void *)obj__ign;
2492 __bpf_kfunc void bpf_rcu_read_lock(void)
2497 __bpf_kfunc void bpf_rcu_read_unlock(void)
2502 struct bpf_throw_ctx {
2503 struct bpf_prog_aux *aux;
2509 static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
2511 struct bpf_throw_ctx *ctx = cookie;
2512 struct bpf_prog *prog;
2514 if (!is_bpf_text_address(ip))
2516 prog = bpf_prog_ksym_find(ip);
2518 if (bpf_is_subprog(prog))
2520 ctx->aux = prog->aux;
2526 __bpf_kfunc void bpf_throw(u64 cookie)
2528 struct bpf_throw_ctx ctx = {};
2530 arch_bpf_stack_walk(bpf_stack_walker, &ctx);
2531 WARN_ON_ONCE(!ctx.aux);
2533 WARN_ON_ONCE(!ctx.aux->exception_boundary);
2534 WARN_ON_ONCE(!ctx.bp);
2535 WARN_ON_ONCE(!ctx.cnt);
2536 /* Prevent KASAN false positives for CONFIG_KASAN_STACK by unpoisoning
2537 * deeper stack depths than ctx.sp as we do not return from bpf_throw,
2538 * which skips compiler generated instrumentation to do the same.
2540 kasan_unpoison_task_stack_below((void *)(long)ctx.sp);
2541 ctx.aux->bpf_exception_cb(cookie, ctx.sp, ctx.bp, 0, 0);
2542 WARN(1, "A call to BPF exception callback should never return\n");
2545 __bpf_kfunc_end_defs();
2547 BTF_KFUNCS_START(generic_btf_ids)
2548 #ifdef CONFIG_KEXEC_CORE
2549 BTF_ID_FLAGS(func, crash_kexec, KF_DESTRUCTIVE)
2551 BTF_ID_FLAGS(func, bpf_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2552 BTF_ID_FLAGS(func, bpf_percpu_obj_new_impl, KF_ACQUIRE | KF_RET_NULL)
2553 BTF_ID_FLAGS(func, bpf_obj_drop_impl, KF_RELEASE)
2554 BTF_ID_FLAGS(func, bpf_percpu_obj_drop_impl, KF_RELEASE)
2555 BTF_ID_FLAGS(func, bpf_refcount_acquire_impl, KF_ACQUIRE | KF_RET_NULL | KF_RCU)
2556 BTF_ID_FLAGS(func, bpf_list_push_front_impl)
2557 BTF_ID_FLAGS(func, bpf_list_push_back_impl)
2558 BTF_ID_FLAGS(func, bpf_list_pop_front, KF_ACQUIRE | KF_RET_NULL)
2559 BTF_ID_FLAGS(func, bpf_list_pop_back, KF_ACQUIRE | KF_RET_NULL)
2560 BTF_ID_FLAGS(func, bpf_task_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2561 BTF_ID_FLAGS(func, bpf_task_release, KF_RELEASE)
2562 BTF_ID_FLAGS(func, bpf_rbtree_remove, KF_ACQUIRE | KF_RET_NULL)
2563 BTF_ID_FLAGS(func, bpf_rbtree_add_impl)
2564 BTF_ID_FLAGS(func, bpf_rbtree_first, KF_RET_NULL)
2566 #ifdef CONFIG_CGROUPS
2567 BTF_ID_FLAGS(func, bpf_cgroup_acquire, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2568 BTF_ID_FLAGS(func, bpf_cgroup_release, KF_RELEASE)
2569 BTF_ID_FLAGS(func, bpf_cgroup_ancestor, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2570 BTF_ID_FLAGS(func, bpf_cgroup_from_id, KF_ACQUIRE | KF_RET_NULL)
2571 BTF_ID_FLAGS(func, bpf_task_under_cgroup, KF_RCU)
2572 BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL)
2574 BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL)
2575 BTF_ID_FLAGS(func, bpf_throw)
2576 BTF_KFUNCS_END(generic_btf_ids)
2578 static const struct btf_kfunc_id_set generic_kfunc_set = {
2579 .owner = THIS_MODULE,
2580 .set = &generic_btf_ids,
2584 BTF_ID_LIST(generic_dtor_ids)
2585 BTF_ID(struct, task_struct)
2586 BTF_ID(func, bpf_task_release_dtor)
2587 #ifdef CONFIG_CGROUPS
2588 BTF_ID(struct, cgroup)
2589 BTF_ID(func, bpf_cgroup_release_dtor)
2592 BTF_KFUNCS_START(common_btf_ids)
2593 BTF_ID_FLAGS(func, bpf_cast_to_kern_ctx)
2594 BTF_ID_FLAGS(func, bpf_rdonly_cast)
2595 BTF_ID_FLAGS(func, bpf_rcu_read_lock)
2596 BTF_ID_FLAGS(func, bpf_rcu_read_unlock)
2597 BTF_ID_FLAGS(func, bpf_dynptr_slice, KF_RET_NULL)
2598 BTF_ID_FLAGS(func, bpf_dynptr_slice_rdwr, KF_RET_NULL)
2599 BTF_ID_FLAGS(func, bpf_iter_num_new, KF_ITER_NEW)
2600 BTF_ID_FLAGS(func, bpf_iter_num_next, KF_ITER_NEXT | KF_RET_NULL)
2601 BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
2602 BTF_ID_FLAGS(func, bpf_iter_task_vma_new, KF_ITER_NEW | KF_RCU)
2603 BTF_ID_FLAGS(func, bpf_iter_task_vma_next, KF_ITER_NEXT | KF_RET_NULL)
2604 BTF_ID_FLAGS(func, bpf_iter_task_vma_destroy, KF_ITER_DESTROY)
2605 #ifdef CONFIG_CGROUPS
2606 BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS)
2607 BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
2608 BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
2609 BTF_ID_FLAGS(func, bpf_iter_css_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2610 BTF_ID_FLAGS(func, bpf_iter_css_next, KF_ITER_NEXT | KF_RET_NULL)
2611 BTF_ID_FLAGS(func, bpf_iter_css_destroy, KF_ITER_DESTROY)
2613 BTF_ID_FLAGS(func, bpf_iter_task_new, KF_ITER_NEW | KF_TRUSTED_ARGS | KF_RCU_PROTECTED)
2614 BTF_ID_FLAGS(func, bpf_iter_task_next, KF_ITER_NEXT | KF_RET_NULL)
2615 BTF_ID_FLAGS(func, bpf_iter_task_destroy, KF_ITER_DESTROY)
2616 BTF_ID_FLAGS(func, bpf_dynptr_adjust)
2617 BTF_ID_FLAGS(func, bpf_dynptr_is_null)
2618 BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
2619 BTF_ID_FLAGS(func, bpf_dynptr_size)
2620 BTF_ID_FLAGS(func, bpf_dynptr_clone)
2621 BTF_KFUNCS_END(common_btf_ids)
2623 static const struct btf_kfunc_id_set common_kfunc_set = {
2624 .owner = THIS_MODULE,
2625 .set = &common_btf_ids,
2628 static int __init kfunc_init(void)
2631 const struct btf_id_dtor_kfunc generic_dtors[] = {
2633 .btf_id = generic_dtor_ids[0],
2634 .kfunc_btf_id = generic_dtor_ids[1]
2636 #ifdef CONFIG_CGROUPS
2638 .btf_id = generic_dtor_ids[2],
2639 .kfunc_btf_id = generic_dtor_ids[3]
2644 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &generic_kfunc_set);
2645 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &generic_kfunc_set);
2646 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &generic_kfunc_set);
2647 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &generic_kfunc_set);
2648 ret = ret ?: register_btf_id_dtor_kfuncs(generic_dtors,
2649 ARRAY_SIZE(generic_dtors),
2651 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_UNSPEC, &common_kfunc_set);
2654 late_initcall(kfunc_init);
2656 /* Get a pointer to dynptr data up to len bytes for read only access. If
2657 * the dynptr doesn't have continuous data up to len bytes, return NULL.
2659 const void *__bpf_dynptr_data(const struct bpf_dynptr_kern *ptr, u32 len)
2661 return bpf_dynptr_slice(ptr, 0, NULL, len);
2664 /* Get a pointer to dynptr data up to len bytes for read write access. If
2665 * the dynptr doesn't have continuous data up to len bytes, or the dynptr
2666 * is read only, return NULL.
2668 void *__bpf_dynptr_data_rw(const struct bpf_dynptr_kern *ptr, u32 len)
2670 if (__bpf_dynptr_is_rdonly(ptr))
2672 return (void *)__bpf_dynptr_data(ptr, len);