1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/slab.h>
9 #include <linux/bpf_perf_event.h>
10 #include <linux/btf.h>
11 #include <linux/filter.h>
12 #include <linux/uaccess.h>
13 #include <linux/ctype.h>
14 #include <linux/kprobes.h>
15 #include <linux/spinlock.h>
16 #include <linux/syscalls.h>
17 #include <linux/error-injection.h>
18 #include <linux/btf_ids.h>
20 #include <uapi/linux/bpf.h>
21 #include <uapi/linux/btf.h>
25 #include "trace_probe.h"
28 #define CREATE_TRACE_POINTS
29 #include "bpf_trace.h"
31 #define bpf_event_rcu_dereference(p) \
32 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
35 struct bpf_trace_module {
36 struct module *module;
37 struct list_head list;
40 static LIST_HEAD(bpf_trace_modules);
41 static DEFINE_MUTEX(bpf_module_mutex);
43 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
45 struct bpf_raw_event_map *btp, *ret = NULL;
46 struct bpf_trace_module *btm;
49 mutex_lock(&bpf_module_mutex);
50 list_for_each_entry(btm, &bpf_trace_modules, list) {
51 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
52 btp = &btm->module->bpf_raw_events[i];
53 if (!strcmp(btp->tp->name, name)) {
54 if (try_module_get(btm->module))
61 mutex_unlock(&bpf_module_mutex);
65 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
69 #endif /* CONFIG_MODULES */
71 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
72 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
74 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
75 u64 flags, const struct btf **btf,
79 * trace_call_bpf - invoke BPF program
80 * @call: tracepoint event
81 * @ctx: opaque context pointer
83 * kprobe handlers execute BPF programs via this helper.
84 * Can be used from static tracepoints in the future.
86 * Return: BPF programs always return an integer which is interpreted by
88 * 0 - return from kprobe (event is filtered out)
89 * 1 - store kprobe event into ring buffer
90 * Other values are reserved and currently alias to 1
92 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
96 if (in_nmi()) /* not supported yet */
101 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
103 * since some bpf program is already running on this cpu,
104 * don't call into another bpf program (same or different)
105 * and don't send kprobe event into ring-buffer,
106 * so return zero here
113 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
114 * to all call sites, we did a bpf_prog_array_valid() there to check
115 * whether call->prog_array is empty or not, which is
116 * a heurisitc to speed up execution.
118 * If bpf_prog_array_valid() fetched prog_array was
119 * non-NULL, we go into trace_call_bpf() and do the actual
120 * proper rcu_dereference() under RCU lock.
121 * If it turns out that prog_array is NULL then, we bail out.
122 * For the opposite, if the bpf_prog_array_valid() fetched pointer
123 * was NULL, you'll skip the prog_array with the risk of missing
124 * out of events when it was updated in between this and the
125 * rcu_dereference() which is accepted risk.
127 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
130 __this_cpu_dec(bpf_prog_active);
135 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
136 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
138 regs_set_return_value(regs, rc);
139 override_function_with_return(regs);
143 static const struct bpf_func_proto bpf_override_return_proto = {
144 .func = bpf_override_return,
146 .ret_type = RET_INTEGER,
147 .arg1_type = ARG_PTR_TO_CTX,
148 .arg2_type = ARG_ANYTHING,
152 static __always_inline int
153 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
157 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
158 if (unlikely(ret < 0))
159 memset(dst, 0, size);
163 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
164 const void __user *, unsafe_ptr)
166 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
169 const struct bpf_func_proto bpf_probe_read_user_proto = {
170 .func = bpf_probe_read_user,
172 .ret_type = RET_INTEGER,
173 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
174 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
175 .arg3_type = ARG_ANYTHING,
178 static __always_inline int
179 bpf_probe_read_user_str_common(void *dst, u32 size,
180 const void __user *unsafe_ptr)
185 * NB: We rely on strncpy_from_user() not copying junk past the NUL
186 * terminator into `dst`.
188 * strncpy_from_user() does long-sized strides in the fast path. If the
189 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
190 * then there could be junk after the NUL in `dst`. If user takes `dst`
191 * and keys a hash map with it, then semantically identical strings can
192 * occupy multiple entries in the map.
194 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
195 if (unlikely(ret < 0))
196 memset(dst, 0, size);
200 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
201 const void __user *, unsafe_ptr)
203 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
206 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
207 .func = bpf_probe_read_user_str,
209 .ret_type = RET_INTEGER,
210 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
211 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
212 .arg3_type = ARG_ANYTHING,
215 static __always_inline int
216 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
218 int ret = security_locked_down(LOCKDOWN_BPF_READ);
220 if (unlikely(ret < 0))
222 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
223 if (unlikely(ret < 0))
227 memset(dst, 0, size);
231 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
232 const void *, unsafe_ptr)
234 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
237 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
238 .func = bpf_probe_read_kernel,
240 .ret_type = RET_INTEGER,
241 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
242 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
243 .arg3_type = ARG_ANYTHING,
246 static __always_inline int
247 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
249 int ret = security_locked_down(LOCKDOWN_BPF_READ);
251 if (unlikely(ret < 0))
255 * The strncpy_from_kernel_nofault() call will likely not fill the
256 * entire buffer, but that's okay in this circumstance as we're probing
257 * arbitrary memory anyway similar to bpf_probe_read_*() and might
258 * as well probe the stack. Thus, memory is explicitly cleared
259 * only in error case, so that improper users ignoring return
260 * code altogether don't copy garbage; otherwise length of string
261 * is returned that can be used for bpf_perf_event_output() et al.
263 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
264 if (unlikely(ret < 0))
269 memset(dst, 0, size);
273 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
274 const void *, unsafe_ptr)
276 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
279 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
280 .func = bpf_probe_read_kernel_str,
282 .ret_type = RET_INTEGER,
283 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
284 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
285 .arg3_type = ARG_ANYTHING,
288 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
289 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
290 const void *, unsafe_ptr)
292 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
293 return bpf_probe_read_user_common(dst, size,
294 (__force void __user *)unsafe_ptr);
296 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
299 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
300 .func = bpf_probe_read_compat,
302 .ret_type = RET_INTEGER,
303 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
304 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
305 .arg3_type = ARG_ANYTHING,
308 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
309 const void *, unsafe_ptr)
311 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
312 return bpf_probe_read_user_str_common(dst, size,
313 (__force void __user *)unsafe_ptr);
315 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
318 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
319 .func = bpf_probe_read_compat_str,
321 .ret_type = RET_INTEGER,
322 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
323 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
324 .arg3_type = ARG_ANYTHING,
326 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
328 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
332 * Ensure we're in user context which is safe for the helper to
333 * run. This helper has no business in a kthread.
335 * access_ok() should prevent writing to non-user memory, but in
336 * some situations (nommu, temporary switch, etc) access_ok() does
337 * not provide enough validation, hence the check on KERNEL_DS.
339 * nmi_uaccess_okay() ensures the probe is not run in an interim
340 * state, when the task or mm are switched. This is specifically
341 * required to prevent the use of temporary mm.
344 if (unlikely(in_interrupt() ||
345 current->flags & (PF_KTHREAD | PF_EXITING)))
347 if (unlikely(uaccess_kernel()))
349 if (unlikely(!nmi_uaccess_okay()))
352 return copy_to_user_nofault(unsafe_ptr, src, size);
355 static const struct bpf_func_proto bpf_probe_write_user_proto = {
356 .func = bpf_probe_write_user,
358 .ret_type = RET_INTEGER,
359 .arg1_type = ARG_ANYTHING,
360 .arg2_type = ARG_PTR_TO_MEM,
361 .arg3_type = ARG_CONST_SIZE,
364 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
366 if (!capable(CAP_SYS_ADMIN))
369 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
370 current->comm, task_pid_nr(current));
372 return &bpf_probe_write_user_proto;
375 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
378 void __user *user_ptr = (__force void __user *)unsafe_ptr;
384 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
385 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
386 strncpy_from_user_nofault(buf, user_ptr, bufsz);
392 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
395 strncpy_from_user_nofault(buf, user_ptr, bufsz);
400 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
402 #define BPF_TRACE_PRINTK_SIZE 1024
404 static __printf(1, 0) int bpf_do_trace_printk(const char *fmt, ...)
406 static char buf[BPF_TRACE_PRINTK_SIZE];
411 raw_spin_lock_irqsave(&trace_printk_lock, flags);
413 ret = vsnprintf(buf, sizeof(buf), fmt, ap);
415 /* vsnprintf() will not append null for zero-length strings */
418 trace_bpf_trace_printk(buf);
419 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
425 * Only limited trace_printk() conversion specifiers allowed:
426 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pB %pks %pus %s
428 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
429 u64, arg2, u64, arg3)
431 int i, mod[3] = {}, fmt_cnt = 0;
432 char buf[64], fmt_ptype;
433 void *unsafe_ptr = NULL;
434 bool str_seen = false;
437 * bpf_check()->check_func_arg()->check_stack_boundary()
438 * guarantees that fmt points to bpf program stack,
439 * fmt_size bytes of it were initialized and fmt_size > 0
441 if (fmt[--fmt_size] != 0)
444 /* check format string for allowed specifiers */
445 for (i = 0; i < fmt_size; i++) {
446 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
455 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
460 } else if (fmt[i] == 'p') {
462 if ((fmt[i + 1] == 'k' ||
463 fmt[i + 1] == 'u') &&
465 fmt_ptype = fmt[i + 1];
470 if (fmt[i + 1] == 'B') {
475 /* disallow any further format extensions */
476 if (fmt[i + 1] != 0 &&
477 !isspace(fmt[i + 1]) &&
478 !ispunct(fmt[i + 1]))
482 } else if (fmt[i] == 's') {
487 /* allow only one '%s' per fmt string */
491 if (fmt[i + 1] != 0 &&
492 !isspace(fmt[i + 1]) &&
493 !ispunct(fmt[i + 1]))
498 unsafe_ptr = (void *)(long)arg1;
502 unsafe_ptr = (void *)(long)arg2;
506 unsafe_ptr = (void *)(long)arg3;
511 bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
521 if (fmt[i] != 'i' && fmt[i] != 'd' &&
522 fmt[i] != 'u' && fmt[i] != 'x')
528 /* Horrid workaround for getting va_list handling working with different
529 * argument type combinations generically for 32 and 64 bit archs.
531 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
532 #define __BPF_TP(...) \
533 bpf_do_trace_printk(fmt, ##__VA_ARGS__)
535 #define __BPF_ARG1_TP(...) \
536 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
537 ? __BPF_TP(arg1, ##__VA_ARGS__) \
538 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
539 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
540 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
542 #define __BPF_ARG2_TP(...) \
543 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
544 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
545 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
546 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
547 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
549 #define __BPF_ARG3_TP(...) \
550 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
551 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
552 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
553 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
554 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
556 return __BPF_TP_EMIT();
559 static const struct bpf_func_proto bpf_trace_printk_proto = {
560 .func = bpf_trace_printk,
562 .ret_type = RET_INTEGER,
563 .arg1_type = ARG_PTR_TO_MEM,
564 .arg2_type = ARG_CONST_SIZE,
567 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
570 * This program might be calling bpf_trace_printk,
571 * so enable the associated bpf_trace/bpf_trace_printk event.
572 * Repeat this each time as it is possible a user has
573 * disabled bpf_trace_printk events. By loading a program
574 * calling bpf_trace_printk() however the user has expressed
575 * the intent to see such events.
577 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
578 pr_warn_ratelimited("could not enable bpf_trace_printk events");
580 return &bpf_trace_printk_proto;
583 #define MAX_SEQ_PRINTF_VARARGS 12
584 #define MAX_SEQ_PRINTF_MAX_MEMCPY 6
585 #define MAX_SEQ_PRINTF_STR_LEN 128
587 struct bpf_seq_printf_buf {
588 char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
590 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
591 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
593 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
594 const void *, data, u32, data_len)
596 int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
597 int i, buf_used, copy_size, num_args;
598 u64 params[MAX_SEQ_PRINTF_VARARGS];
599 struct bpf_seq_printf_buf *bufs;
600 const u64 *args = data;
602 buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
603 if (WARN_ON_ONCE(buf_used > 1)) {
608 bufs = this_cpu_ptr(&bpf_seq_printf_buf);
611 * bpf_check()->check_func_arg()->check_stack_boundary()
612 * guarantees that fmt points to bpf program stack,
613 * fmt_size bytes of it were initialized and fmt_size > 0
615 if (fmt[--fmt_size] != 0)
621 for (i = 0; i < fmt_size; i++) {
623 if (fmt[i + 1] == '%')
625 else if (!data || !data_len)
630 num_args = data_len / 8;
632 /* check format string for allowed specifiers */
633 for (i = 0; i < fmt_size; i++) {
634 /* only printable ascii for now. */
635 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
643 if (fmt[i + 1] == '%') {
648 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
653 if (fmt_cnt >= num_args) {
658 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
661 /* skip optional "[0 +-][num]" width formating field */
662 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
665 if (fmt[i] >= '1' && fmt[i] <= '9') {
667 while (fmt[i] >= '0' && fmt[i] <= '9')
674 /* try our best to copy */
675 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
680 unsafe_ptr = (void *)(long)args[fmt_cnt];
681 err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
682 unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
684 bufs->buf[memcpy_cnt][0] = '\0';
685 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
693 if (fmt[i + 1] == 0 ||
697 /* just kernel pointers */
698 params[fmt_cnt] = args[fmt_cnt];
703 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
704 if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
708 if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
713 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
719 copy_size = (fmt[i + 2] == '4') ? 4 : 16;
721 err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
722 (void *) (long) args[fmt_cnt],
725 memset(bufs->buf[memcpy_cnt], 0, copy_size);
726 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
740 if (fmt[i] != 'i' && fmt[i] != 'd' &&
741 fmt[i] != 'u' && fmt[i] != 'x' &&
747 params[fmt_cnt] = args[fmt_cnt];
751 /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
752 * all of them to seq_printf().
754 seq_printf(m, fmt, params[0], params[1], params[2], params[3],
755 params[4], params[5], params[6], params[7], params[8],
756 params[9], params[10], params[11]);
758 err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
760 this_cpu_dec(bpf_seq_printf_buf_used);
764 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
766 static const struct bpf_func_proto bpf_seq_printf_proto = {
767 .func = bpf_seq_printf,
769 .ret_type = RET_INTEGER,
770 .arg1_type = ARG_PTR_TO_BTF_ID,
771 .arg1_btf_id = &btf_seq_file_ids[0],
772 .arg2_type = ARG_PTR_TO_MEM,
773 .arg3_type = ARG_CONST_SIZE,
774 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
775 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
778 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
780 return seq_write(m, data, len) ? -EOVERFLOW : 0;
783 static const struct bpf_func_proto bpf_seq_write_proto = {
784 .func = bpf_seq_write,
786 .ret_type = RET_INTEGER,
787 .arg1_type = ARG_PTR_TO_BTF_ID,
788 .arg1_btf_id = &btf_seq_file_ids[0],
789 .arg2_type = ARG_PTR_TO_MEM,
790 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
793 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
794 u32, btf_ptr_size, u64, flags)
796 const struct btf *btf;
800 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
804 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
807 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
808 .func = bpf_seq_printf_btf,
810 .ret_type = RET_INTEGER,
811 .arg1_type = ARG_PTR_TO_BTF_ID,
812 .arg1_btf_id = &btf_seq_file_ids[0],
813 .arg2_type = ARG_PTR_TO_MEM,
814 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
815 .arg4_type = ARG_ANYTHING,
818 static __always_inline int
819 get_map_perf_counter(struct bpf_map *map, u64 flags,
820 u64 *value, u64 *enabled, u64 *running)
822 struct bpf_array *array = container_of(map, struct bpf_array, map);
823 unsigned int cpu = smp_processor_id();
824 u64 index = flags & BPF_F_INDEX_MASK;
825 struct bpf_event_entry *ee;
827 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
829 if (index == BPF_F_CURRENT_CPU)
831 if (unlikely(index >= array->map.max_entries))
834 ee = READ_ONCE(array->ptrs[index]);
838 return perf_event_read_local(ee->event, value, enabled, running);
841 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
846 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
848 * this api is ugly since we miss [-22..-2] range of valid
849 * counter values, but that's uapi
856 static const struct bpf_func_proto bpf_perf_event_read_proto = {
857 .func = bpf_perf_event_read,
859 .ret_type = RET_INTEGER,
860 .arg1_type = ARG_CONST_MAP_PTR,
861 .arg2_type = ARG_ANYTHING,
864 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
865 struct bpf_perf_event_value *, buf, u32, size)
869 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
871 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
877 memset(buf, 0, size);
881 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
882 .func = bpf_perf_event_read_value,
884 .ret_type = RET_INTEGER,
885 .arg1_type = ARG_CONST_MAP_PTR,
886 .arg2_type = ARG_ANYTHING,
887 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
888 .arg4_type = ARG_CONST_SIZE,
891 static __always_inline u64
892 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
893 u64 flags, struct perf_sample_data *sd)
895 struct bpf_array *array = container_of(map, struct bpf_array, map);
896 unsigned int cpu = smp_processor_id();
897 u64 index = flags & BPF_F_INDEX_MASK;
898 struct bpf_event_entry *ee;
899 struct perf_event *event;
901 if (index == BPF_F_CURRENT_CPU)
903 if (unlikely(index >= array->map.max_entries))
906 ee = READ_ONCE(array->ptrs[index]);
911 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
912 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
915 if (unlikely(event->oncpu != cpu))
918 return perf_event_output(event, sd, regs);
922 * Support executing tracepoints in normal, irq, and nmi context that each call
923 * bpf_perf_event_output
925 struct bpf_trace_sample_data {
926 struct perf_sample_data sds[3];
929 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
930 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
931 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
932 u64, flags, void *, data, u64, size)
934 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
935 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
936 struct perf_raw_record raw = {
942 struct perf_sample_data *sd;
945 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
950 sd = &sds->sds[nest_level - 1];
952 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
957 perf_sample_data_init(sd, 0, 0);
960 err = __bpf_perf_event_output(regs, map, flags, sd);
963 this_cpu_dec(bpf_trace_nest_level);
967 static const struct bpf_func_proto bpf_perf_event_output_proto = {
968 .func = bpf_perf_event_output,
970 .ret_type = RET_INTEGER,
971 .arg1_type = ARG_PTR_TO_CTX,
972 .arg2_type = ARG_CONST_MAP_PTR,
973 .arg3_type = ARG_ANYTHING,
974 .arg4_type = ARG_PTR_TO_MEM,
975 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
978 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
979 struct bpf_nested_pt_regs {
980 struct pt_regs regs[3];
982 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
983 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
985 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
986 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
988 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
989 struct perf_raw_frag frag = {
994 struct perf_raw_record raw = {
997 .next = ctx_size ? &frag : NULL,
1003 struct perf_sample_data *sd;
1004 struct pt_regs *regs;
1007 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
1011 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
1012 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
1014 perf_fetch_caller_regs(regs);
1015 perf_sample_data_init(sd, 0, 0);
1018 ret = __bpf_perf_event_output(regs, map, flags, sd);
1020 this_cpu_dec(bpf_event_output_nest_level);
1024 BPF_CALL_0(bpf_get_current_task)
1026 return (long) current;
1029 const struct bpf_func_proto bpf_get_current_task_proto = {
1030 .func = bpf_get_current_task,
1032 .ret_type = RET_INTEGER,
1035 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
1037 struct bpf_array *array = container_of(map, struct bpf_array, map);
1038 struct cgroup *cgrp;
1040 if (unlikely(idx >= array->map.max_entries))
1043 cgrp = READ_ONCE(array->ptrs[idx]);
1044 if (unlikely(!cgrp))
1047 return task_under_cgroup_hierarchy(current, cgrp);
1050 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
1051 .func = bpf_current_task_under_cgroup,
1053 .ret_type = RET_INTEGER,
1054 .arg1_type = ARG_CONST_MAP_PTR,
1055 .arg2_type = ARG_ANYTHING,
1058 struct send_signal_irq_work {
1059 struct irq_work irq_work;
1060 struct task_struct *task;
1065 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
1067 static void do_bpf_send_signal(struct irq_work *entry)
1069 struct send_signal_irq_work *work;
1071 work = container_of(entry, struct send_signal_irq_work, irq_work);
1072 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
1075 static int bpf_send_signal_common(u32 sig, enum pid_type type)
1077 struct send_signal_irq_work *work = NULL;
1079 /* Similar to bpf_probe_write_user, task needs to be
1080 * in a sound condition and kernel memory access be
1081 * permitted in order to send signal to the current
1084 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1086 if (unlikely(uaccess_kernel()))
1088 if (unlikely(!nmi_uaccess_okay()))
1091 if (irqs_disabled()) {
1092 /* Do an early check on signal validity. Otherwise,
1093 * the error is lost in deferred irq_work.
1095 if (unlikely(!valid_signal(sig)))
1098 work = this_cpu_ptr(&send_signal_work);
1099 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
1102 /* Add the current task, which is the target of sending signal,
1103 * to the irq_work. The current task may change when queued
1104 * irq works get executed.
1106 work->task = current;
1109 irq_work_queue(&work->irq_work);
1113 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1116 BPF_CALL_1(bpf_send_signal, u32, sig)
1118 return bpf_send_signal_common(sig, PIDTYPE_TGID);
1121 static const struct bpf_func_proto bpf_send_signal_proto = {
1122 .func = bpf_send_signal,
1124 .ret_type = RET_INTEGER,
1125 .arg1_type = ARG_ANYTHING,
1128 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1130 return bpf_send_signal_common(sig, PIDTYPE_PID);
1133 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1134 .func = bpf_send_signal_thread,
1136 .ret_type = RET_INTEGER,
1137 .arg1_type = ARG_ANYTHING,
1140 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
1148 p = d_path(path, buf, sz);
1153 memmove(buf, p, len);
1159 BTF_SET_START(btf_allowlist_d_path)
1160 #ifdef CONFIG_SECURITY
1161 BTF_ID(func, security_file_permission)
1162 BTF_ID(func, security_inode_getattr)
1163 BTF_ID(func, security_file_open)
1165 #ifdef CONFIG_SECURITY_PATH
1166 BTF_ID(func, security_path_truncate)
1168 BTF_ID(func, vfs_truncate)
1169 BTF_ID(func, vfs_fallocate)
1170 BTF_ID(func, dentry_open)
1171 BTF_ID(func, vfs_getattr)
1172 BTF_ID(func, filp_close)
1173 BTF_SET_END(btf_allowlist_d_path)
1175 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
1177 return btf_id_set_contains(&btf_allowlist_d_path, prog->aux->attach_btf_id);
1180 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
1182 static const struct bpf_func_proto bpf_d_path_proto = {
1185 .ret_type = RET_INTEGER,
1186 .arg1_type = ARG_PTR_TO_BTF_ID,
1187 .arg1_btf_id = &bpf_d_path_btf_ids[0],
1188 .arg2_type = ARG_PTR_TO_MEM,
1189 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1190 .allowed = bpf_d_path_allowed,
1193 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
1194 BTF_F_PTR_RAW | BTF_F_ZERO)
1196 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
1197 u64 flags, const struct btf **btf,
1200 const struct btf_type *t;
1202 if (unlikely(flags & ~(BTF_F_ALL)))
1205 if (btf_ptr_size != sizeof(struct btf_ptr))
1208 *btf = bpf_get_btf_vmlinux();
1210 if (IS_ERR_OR_NULL(*btf))
1211 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
1213 if (ptr->type_id > 0)
1214 *btf_id = ptr->type_id;
1219 t = btf_type_by_id(*btf, *btf_id);
1220 if (*btf_id <= 0 || !t)
1226 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1227 u32, btf_ptr_size, u64, flags)
1229 const struct btf *btf;
1233 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1237 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1241 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1242 .func = bpf_snprintf_btf,
1244 .ret_type = RET_INTEGER,
1245 .arg1_type = ARG_PTR_TO_MEM,
1246 .arg2_type = ARG_CONST_SIZE,
1247 .arg3_type = ARG_PTR_TO_MEM,
1248 .arg4_type = ARG_CONST_SIZE,
1249 .arg5_type = ARG_ANYTHING,
1252 const struct bpf_func_proto *
1253 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1256 case BPF_FUNC_map_lookup_elem:
1257 return &bpf_map_lookup_elem_proto;
1258 case BPF_FUNC_map_update_elem:
1259 return &bpf_map_update_elem_proto;
1260 case BPF_FUNC_map_delete_elem:
1261 return &bpf_map_delete_elem_proto;
1262 case BPF_FUNC_map_push_elem:
1263 return &bpf_map_push_elem_proto;
1264 case BPF_FUNC_map_pop_elem:
1265 return &bpf_map_pop_elem_proto;
1266 case BPF_FUNC_map_peek_elem:
1267 return &bpf_map_peek_elem_proto;
1268 case BPF_FUNC_ktime_get_ns:
1269 return &bpf_ktime_get_ns_proto;
1270 case BPF_FUNC_ktime_get_boot_ns:
1271 return &bpf_ktime_get_boot_ns_proto;
1272 case BPF_FUNC_tail_call:
1273 return &bpf_tail_call_proto;
1274 case BPF_FUNC_get_current_pid_tgid:
1275 return &bpf_get_current_pid_tgid_proto;
1276 case BPF_FUNC_get_current_task:
1277 return &bpf_get_current_task_proto;
1278 case BPF_FUNC_get_current_uid_gid:
1279 return &bpf_get_current_uid_gid_proto;
1280 case BPF_FUNC_get_current_comm:
1281 return &bpf_get_current_comm_proto;
1282 case BPF_FUNC_trace_printk:
1283 return bpf_get_trace_printk_proto();
1284 case BPF_FUNC_get_smp_processor_id:
1285 return &bpf_get_smp_processor_id_proto;
1286 case BPF_FUNC_get_numa_node_id:
1287 return &bpf_get_numa_node_id_proto;
1288 case BPF_FUNC_perf_event_read:
1289 return &bpf_perf_event_read_proto;
1290 case BPF_FUNC_probe_write_user:
1291 return bpf_get_probe_write_proto();
1292 case BPF_FUNC_current_task_under_cgroup:
1293 return &bpf_current_task_under_cgroup_proto;
1294 case BPF_FUNC_get_prandom_u32:
1295 return &bpf_get_prandom_u32_proto;
1296 case BPF_FUNC_probe_read_user:
1297 return &bpf_probe_read_user_proto;
1298 case BPF_FUNC_probe_read_kernel:
1299 return &bpf_probe_read_kernel_proto;
1300 case BPF_FUNC_probe_read_user_str:
1301 return &bpf_probe_read_user_str_proto;
1302 case BPF_FUNC_probe_read_kernel_str:
1303 return &bpf_probe_read_kernel_str_proto;
1304 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1305 case BPF_FUNC_probe_read:
1306 return &bpf_probe_read_compat_proto;
1307 case BPF_FUNC_probe_read_str:
1308 return &bpf_probe_read_compat_str_proto;
1310 #ifdef CONFIG_CGROUPS
1311 case BPF_FUNC_get_current_cgroup_id:
1312 return &bpf_get_current_cgroup_id_proto;
1314 case BPF_FUNC_send_signal:
1315 return &bpf_send_signal_proto;
1316 case BPF_FUNC_send_signal_thread:
1317 return &bpf_send_signal_thread_proto;
1318 case BPF_FUNC_perf_event_read_value:
1319 return &bpf_perf_event_read_value_proto;
1320 case BPF_FUNC_get_ns_current_pid_tgid:
1321 return &bpf_get_ns_current_pid_tgid_proto;
1322 case BPF_FUNC_ringbuf_output:
1323 return &bpf_ringbuf_output_proto;
1324 case BPF_FUNC_ringbuf_reserve:
1325 return &bpf_ringbuf_reserve_proto;
1326 case BPF_FUNC_ringbuf_submit:
1327 return &bpf_ringbuf_submit_proto;
1328 case BPF_FUNC_ringbuf_discard:
1329 return &bpf_ringbuf_discard_proto;
1330 case BPF_FUNC_ringbuf_query:
1331 return &bpf_ringbuf_query_proto;
1332 case BPF_FUNC_jiffies64:
1333 return &bpf_jiffies64_proto;
1334 case BPF_FUNC_get_task_stack:
1335 return &bpf_get_task_stack_proto;
1336 case BPF_FUNC_copy_from_user:
1337 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1338 case BPF_FUNC_snprintf_btf:
1339 return &bpf_snprintf_btf_proto;
1340 case BPF_FUNC_bpf_per_cpu_ptr:
1341 return &bpf_per_cpu_ptr_proto;
1342 case BPF_FUNC_bpf_this_cpu_ptr:
1343 return &bpf_this_cpu_ptr_proto;
1349 static const struct bpf_func_proto *
1350 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1353 case BPF_FUNC_perf_event_output:
1354 return &bpf_perf_event_output_proto;
1355 case BPF_FUNC_get_stackid:
1356 return &bpf_get_stackid_proto;
1357 case BPF_FUNC_get_stack:
1358 return &bpf_get_stack_proto;
1359 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1360 case BPF_FUNC_override_return:
1361 return &bpf_override_return_proto;
1364 return bpf_tracing_func_proto(func_id, prog);
1368 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1369 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1370 const struct bpf_prog *prog,
1371 struct bpf_insn_access_aux *info)
1373 if (off < 0 || off >= sizeof(struct pt_regs))
1375 if (type != BPF_READ)
1377 if (off % size != 0)
1380 * Assertion for 32 bit to make sure last 8 byte access
1381 * (BPF_DW) to the last 4 byte member is disallowed.
1383 if (off + size > sizeof(struct pt_regs))
1389 const struct bpf_verifier_ops kprobe_verifier_ops = {
1390 .get_func_proto = kprobe_prog_func_proto,
1391 .is_valid_access = kprobe_prog_is_valid_access,
1394 const struct bpf_prog_ops kprobe_prog_ops = {
1397 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1398 u64, flags, void *, data, u64, size)
1400 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1403 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1404 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1405 * from there and call the same bpf_perf_event_output() helper inline.
1407 return ____bpf_perf_event_output(regs, map, flags, data, size);
1410 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1411 .func = bpf_perf_event_output_tp,
1413 .ret_type = RET_INTEGER,
1414 .arg1_type = ARG_PTR_TO_CTX,
1415 .arg2_type = ARG_CONST_MAP_PTR,
1416 .arg3_type = ARG_ANYTHING,
1417 .arg4_type = ARG_PTR_TO_MEM,
1418 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1421 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1424 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1427 * Same comment as in bpf_perf_event_output_tp(), only that this time
1428 * the other helper's function body cannot be inlined due to being
1429 * external, thus we need to call raw helper function.
1431 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1435 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1436 .func = bpf_get_stackid_tp,
1438 .ret_type = RET_INTEGER,
1439 .arg1_type = ARG_PTR_TO_CTX,
1440 .arg2_type = ARG_CONST_MAP_PTR,
1441 .arg3_type = ARG_ANYTHING,
1444 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1447 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1449 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1450 (unsigned long) size, flags, 0);
1453 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1454 .func = bpf_get_stack_tp,
1456 .ret_type = RET_INTEGER,
1457 .arg1_type = ARG_PTR_TO_CTX,
1458 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1459 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1460 .arg4_type = ARG_ANYTHING,
1463 static const struct bpf_func_proto *
1464 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1467 case BPF_FUNC_perf_event_output:
1468 return &bpf_perf_event_output_proto_tp;
1469 case BPF_FUNC_get_stackid:
1470 return &bpf_get_stackid_proto_tp;
1471 case BPF_FUNC_get_stack:
1472 return &bpf_get_stack_proto_tp;
1474 return bpf_tracing_func_proto(func_id, prog);
1478 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1479 const struct bpf_prog *prog,
1480 struct bpf_insn_access_aux *info)
1482 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1484 if (type != BPF_READ)
1486 if (off % size != 0)
1489 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1493 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1494 .get_func_proto = tp_prog_func_proto,
1495 .is_valid_access = tp_prog_is_valid_access,
1498 const struct bpf_prog_ops tracepoint_prog_ops = {
1501 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1502 struct bpf_perf_event_value *, buf, u32, size)
1506 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1508 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1514 memset(buf, 0, size);
1518 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1519 .func = bpf_perf_prog_read_value,
1521 .ret_type = RET_INTEGER,
1522 .arg1_type = ARG_PTR_TO_CTX,
1523 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1524 .arg3_type = ARG_CONST_SIZE,
1527 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1528 void *, buf, u32, size, u64, flags)
1533 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1534 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1537 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1540 if (unlikely(!br_stack))
1543 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1544 return br_stack->nr * br_entry_size;
1546 if (!buf || (size % br_entry_size != 0))
1549 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1550 memcpy(buf, br_stack->entries, to_copy);
1556 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1557 .func = bpf_read_branch_records,
1559 .ret_type = RET_INTEGER,
1560 .arg1_type = ARG_PTR_TO_CTX,
1561 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1562 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1563 .arg4_type = ARG_ANYTHING,
1566 static const struct bpf_func_proto *
1567 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1570 case BPF_FUNC_perf_event_output:
1571 return &bpf_perf_event_output_proto_tp;
1572 case BPF_FUNC_get_stackid:
1573 return &bpf_get_stackid_proto_pe;
1574 case BPF_FUNC_get_stack:
1575 return &bpf_get_stack_proto_pe;
1576 case BPF_FUNC_perf_prog_read_value:
1577 return &bpf_perf_prog_read_value_proto;
1578 case BPF_FUNC_read_branch_records:
1579 return &bpf_read_branch_records_proto;
1581 return bpf_tracing_func_proto(func_id, prog);
1586 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1587 * to avoid potential recursive reuse issue when/if tracepoints are added
1588 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1590 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1591 * in normal, irq, and nmi context.
1593 struct bpf_raw_tp_regs {
1594 struct pt_regs regs[3];
1596 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1597 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1598 static struct pt_regs *get_bpf_raw_tp_regs(void)
1600 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1601 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1603 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1604 this_cpu_dec(bpf_raw_tp_nest_level);
1605 return ERR_PTR(-EBUSY);
1608 return &tp_regs->regs[nest_level - 1];
1611 static void put_bpf_raw_tp_regs(void)
1613 this_cpu_dec(bpf_raw_tp_nest_level);
1616 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1617 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1619 struct pt_regs *regs = get_bpf_raw_tp_regs();
1623 return PTR_ERR(regs);
1625 perf_fetch_caller_regs(regs);
1626 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1628 put_bpf_raw_tp_regs();
1632 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1633 .func = bpf_perf_event_output_raw_tp,
1635 .ret_type = RET_INTEGER,
1636 .arg1_type = ARG_PTR_TO_CTX,
1637 .arg2_type = ARG_CONST_MAP_PTR,
1638 .arg3_type = ARG_ANYTHING,
1639 .arg4_type = ARG_PTR_TO_MEM,
1640 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1643 extern const struct bpf_func_proto bpf_skb_output_proto;
1644 extern const struct bpf_func_proto bpf_xdp_output_proto;
1646 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1647 struct bpf_map *, map, u64, flags)
1649 struct pt_regs *regs = get_bpf_raw_tp_regs();
1653 return PTR_ERR(regs);
1655 perf_fetch_caller_regs(regs);
1656 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1657 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1659 put_bpf_raw_tp_regs();
1663 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1664 .func = bpf_get_stackid_raw_tp,
1666 .ret_type = RET_INTEGER,
1667 .arg1_type = ARG_PTR_TO_CTX,
1668 .arg2_type = ARG_CONST_MAP_PTR,
1669 .arg3_type = ARG_ANYTHING,
1672 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1673 void *, buf, u32, size, u64, flags)
1675 struct pt_regs *regs = get_bpf_raw_tp_regs();
1679 return PTR_ERR(regs);
1681 perf_fetch_caller_regs(regs);
1682 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1683 (unsigned long) size, flags, 0);
1684 put_bpf_raw_tp_regs();
1688 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1689 .func = bpf_get_stack_raw_tp,
1691 .ret_type = RET_INTEGER,
1692 .arg1_type = ARG_PTR_TO_CTX,
1693 .arg2_type = ARG_PTR_TO_MEM,
1694 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1695 .arg4_type = ARG_ANYTHING,
1698 static const struct bpf_func_proto *
1699 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1702 case BPF_FUNC_perf_event_output:
1703 return &bpf_perf_event_output_proto_raw_tp;
1704 case BPF_FUNC_get_stackid:
1705 return &bpf_get_stackid_proto_raw_tp;
1706 case BPF_FUNC_get_stack:
1707 return &bpf_get_stack_proto_raw_tp;
1709 return bpf_tracing_func_proto(func_id, prog);
1713 const struct bpf_func_proto *
1714 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1718 case BPF_FUNC_skb_output:
1719 return &bpf_skb_output_proto;
1720 case BPF_FUNC_xdp_output:
1721 return &bpf_xdp_output_proto;
1722 case BPF_FUNC_skc_to_tcp6_sock:
1723 return &bpf_skc_to_tcp6_sock_proto;
1724 case BPF_FUNC_skc_to_tcp_sock:
1725 return &bpf_skc_to_tcp_sock_proto;
1726 case BPF_FUNC_skc_to_tcp_timewait_sock:
1727 return &bpf_skc_to_tcp_timewait_sock_proto;
1728 case BPF_FUNC_skc_to_tcp_request_sock:
1729 return &bpf_skc_to_tcp_request_sock_proto;
1730 case BPF_FUNC_skc_to_udp6_sock:
1731 return &bpf_skc_to_udp6_sock_proto;
1733 case BPF_FUNC_seq_printf:
1734 return prog->expected_attach_type == BPF_TRACE_ITER ?
1735 &bpf_seq_printf_proto :
1737 case BPF_FUNC_seq_write:
1738 return prog->expected_attach_type == BPF_TRACE_ITER ?
1739 &bpf_seq_write_proto :
1741 case BPF_FUNC_seq_printf_btf:
1742 return prog->expected_attach_type == BPF_TRACE_ITER ?
1743 &bpf_seq_printf_btf_proto :
1745 case BPF_FUNC_d_path:
1746 return &bpf_d_path_proto;
1748 return raw_tp_prog_func_proto(func_id, prog);
1752 static bool raw_tp_prog_is_valid_access(int off, int size,
1753 enum bpf_access_type type,
1754 const struct bpf_prog *prog,
1755 struct bpf_insn_access_aux *info)
1757 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1759 if (type != BPF_READ)
1761 if (off % size != 0)
1766 static bool tracing_prog_is_valid_access(int off, int size,
1767 enum bpf_access_type type,
1768 const struct bpf_prog *prog,
1769 struct bpf_insn_access_aux *info)
1771 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1773 if (type != BPF_READ)
1775 if (off % size != 0)
1777 return btf_ctx_access(off, size, type, prog, info);
1780 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1781 const union bpf_attr *kattr,
1782 union bpf_attr __user *uattr)
1787 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1788 .get_func_proto = raw_tp_prog_func_proto,
1789 .is_valid_access = raw_tp_prog_is_valid_access,
1792 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1794 .test_run = bpf_prog_test_run_raw_tp,
1798 const struct bpf_verifier_ops tracing_verifier_ops = {
1799 .get_func_proto = tracing_prog_func_proto,
1800 .is_valid_access = tracing_prog_is_valid_access,
1803 const struct bpf_prog_ops tracing_prog_ops = {
1804 .test_run = bpf_prog_test_run_tracing,
1807 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1808 enum bpf_access_type type,
1809 const struct bpf_prog *prog,
1810 struct bpf_insn_access_aux *info)
1813 if (size != sizeof(u64) || type != BPF_READ)
1815 info->reg_type = PTR_TO_TP_BUFFER;
1817 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1820 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1821 .get_func_proto = raw_tp_prog_func_proto,
1822 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1825 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1828 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1829 const struct bpf_prog *prog,
1830 struct bpf_insn_access_aux *info)
1832 const int size_u64 = sizeof(u64);
1834 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1836 if (type != BPF_READ)
1838 if (off % size != 0) {
1839 if (sizeof(unsigned long) != 4)
1843 if (off % size != 4)
1848 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1849 bpf_ctx_record_field_size(info, size_u64);
1850 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1853 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1854 bpf_ctx_record_field_size(info, size_u64);
1855 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1859 if (size != sizeof(long))
1866 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1867 const struct bpf_insn *si,
1868 struct bpf_insn *insn_buf,
1869 struct bpf_prog *prog, u32 *target_size)
1871 struct bpf_insn *insn = insn_buf;
1874 case offsetof(struct bpf_perf_event_data, sample_period):
1875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1876 data), si->dst_reg, si->src_reg,
1877 offsetof(struct bpf_perf_event_data_kern, data));
1878 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1879 bpf_target_off(struct perf_sample_data, period, 8,
1882 case offsetof(struct bpf_perf_event_data, addr):
1883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1884 data), si->dst_reg, si->src_reg,
1885 offsetof(struct bpf_perf_event_data_kern, data));
1886 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1887 bpf_target_off(struct perf_sample_data, addr, 8,
1891 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1892 regs), si->dst_reg, si->src_reg,
1893 offsetof(struct bpf_perf_event_data_kern, regs));
1894 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1899 return insn - insn_buf;
1902 const struct bpf_verifier_ops perf_event_verifier_ops = {
1903 .get_func_proto = pe_prog_func_proto,
1904 .is_valid_access = pe_prog_is_valid_access,
1905 .convert_ctx_access = pe_prog_convert_ctx_access,
1908 const struct bpf_prog_ops perf_event_prog_ops = {
1911 static DEFINE_MUTEX(bpf_event_mutex);
1913 #define BPF_TRACE_MAX_PROGS 64
1915 int perf_event_attach_bpf_prog(struct perf_event *event,
1916 struct bpf_prog *prog)
1918 struct bpf_prog_array *old_array;
1919 struct bpf_prog_array *new_array;
1923 * Kprobe override only works if they are on the function entry,
1924 * and only if they are on the opt-in list.
1926 if (prog->kprobe_override &&
1927 (!trace_kprobe_on_func_entry(event->tp_event) ||
1928 !trace_kprobe_error_injectable(event->tp_event)))
1931 mutex_lock(&bpf_event_mutex);
1936 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1938 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1943 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1947 /* set the new array to event->tp_event and set event->prog */
1949 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1950 bpf_prog_array_free(old_array);
1953 mutex_unlock(&bpf_event_mutex);
1957 void perf_event_detach_bpf_prog(struct perf_event *event)
1959 struct bpf_prog_array *old_array;
1960 struct bpf_prog_array *new_array;
1963 mutex_lock(&bpf_event_mutex);
1968 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1969 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1973 bpf_prog_array_delete_safe(old_array, event->prog);
1975 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1976 bpf_prog_array_free(old_array);
1979 bpf_prog_put(event->prog);
1983 mutex_unlock(&bpf_event_mutex);
1986 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1988 struct perf_event_query_bpf __user *uquery = info;
1989 struct perf_event_query_bpf query = {};
1990 struct bpf_prog_array *progs;
1991 u32 *ids, prog_cnt, ids_len;
1994 if (!perfmon_capable())
1996 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1998 if (copy_from_user(&query, uquery, sizeof(query)))
2001 ids_len = query.ids_len;
2002 if (ids_len > BPF_TRACE_MAX_PROGS)
2004 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2008 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2009 * is required when user only wants to check for uquery->prog_cnt.
2010 * There is no need to check for it since the case is handled
2011 * gracefully in bpf_prog_array_copy_info.
2014 mutex_lock(&bpf_event_mutex);
2015 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2016 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2017 mutex_unlock(&bpf_event_mutex);
2019 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2020 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2027 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2028 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2030 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2032 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2034 for (; btp < __stop__bpf_raw_tp; btp++) {
2035 if (!strcmp(btp->tp->name, name))
2039 return bpf_get_raw_tracepoint_module(name);
2042 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2044 struct module *mod = __module_address((unsigned long)btp);
2050 static __always_inline
2051 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2055 (void) BPF_PROG_RUN(prog, args);
2059 #define UNPACK(...) __VA_ARGS__
2060 #define REPEAT_1(FN, DL, X, ...) FN(X)
2061 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2062 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2063 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2064 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2065 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2066 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2067 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2068 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2069 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2070 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2071 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2072 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2074 #define SARG(X) u64 arg##X
2075 #define COPY(X) args[X] = arg##X
2077 #define __DL_COM (,)
2078 #define __DL_SEM (;)
2080 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2082 #define BPF_TRACE_DEFN_x(x) \
2083 void bpf_trace_run##x(struct bpf_prog *prog, \
2084 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2087 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2088 __bpf_trace_run(prog, args); \
2090 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2091 BPF_TRACE_DEFN_x(1);
2092 BPF_TRACE_DEFN_x(2);
2093 BPF_TRACE_DEFN_x(3);
2094 BPF_TRACE_DEFN_x(4);
2095 BPF_TRACE_DEFN_x(5);
2096 BPF_TRACE_DEFN_x(6);
2097 BPF_TRACE_DEFN_x(7);
2098 BPF_TRACE_DEFN_x(8);
2099 BPF_TRACE_DEFN_x(9);
2100 BPF_TRACE_DEFN_x(10);
2101 BPF_TRACE_DEFN_x(11);
2102 BPF_TRACE_DEFN_x(12);
2104 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2106 struct tracepoint *tp = btp->tp;
2109 * check that program doesn't access arguments beyond what's
2110 * available in this tracepoint
2112 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2115 if (prog->aux->max_tp_access > btp->writable_size)
2118 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
2121 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2123 return __bpf_probe_register(btp, prog);
2126 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2128 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2131 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2132 u32 *fd_type, const char **buf,
2133 u64 *probe_offset, u64 *probe_addr)
2135 bool is_tracepoint, is_syscall_tp;
2136 struct bpf_prog *prog;
2143 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2144 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2147 *prog_id = prog->aux->id;
2148 flags = event->tp_event->flags;
2149 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2150 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2152 if (is_tracepoint || is_syscall_tp) {
2153 *buf = is_tracepoint ? event->tp_event->tp->name
2154 : event->tp_event->name;
2155 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2156 *probe_offset = 0x0;
2161 #ifdef CONFIG_KPROBE_EVENTS
2162 if (flags & TRACE_EVENT_FL_KPROBE)
2163 err = bpf_get_kprobe_info(event, fd_type, buf,
2164 probe_offset, probe_addr,
2165 event->attr.type == PERF_TYPE_TRACEPOINT);
2167 #ifdef CONFIG_UPROBE_EVENTS
2168 if (flags & TRACE_EVENT_FL_UPROBE)
2169 err = bpf_get_uprobe_info(event, fd_type, buf,
2171 event->attr.type == PERF_TYPE_TRACEPOINT);
2178 static int __init send_signal_irq_work_init(void)
2181 struct send_signal_irq_work *work;
2183 for_each_possible_cpu(cpu) {
2184 work = per_cpu_ptr(&send_signal_work, cpu);
2185 init_irq_work(&work->irq_work, do_bpf_send_signal);
2190 subsys_initcall(send_signal_irq_work_init);
2192 #ifdef CONFIG_MODULES
2193 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2196 struct bpf_trace_module *btm, *tmp;
2197 struct module *mod = module;
2200 if (mod->num_bpf_raw_events == 0 ||
2201 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2204 mutex_lock(&bpf_module_mutex);
2207 case MODULE_STATE_COMING:
2208 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2210 btm->module = module;
2211 list_add(&btm->list, &bpf_trace_modules);
2216 case MODULE_STATE_GOING:
2217 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2218 if (btm->module == module) {
2219 list_del(&btm->list);
2227 mutex_unlock(&bpf_module_mutex);
2230 return notifier_from_errno(ret);
2233 static struct notifier_block bpf_module_nb = {
2234 .notifier_call = bpf_event_notify,
2237 static int __init bpf_event_init(void)
2239 register_module_notifier(&bpf_module_nb);
2243 fs_initcall(bpf_event_init);
2244 #endif /* CONFIG_MODULES */