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/filter.h>
11 #include <linux/uaccess.h>
12 #include <linux/ctype.h>
13 #include <linux/kprobes.h>
14 #include <linux/spinlock.h>
15 #include <linux/syscalls.h>
16 #include <linux/error-injection.h>
17 #include <linux/btf_ids.h>
21 #include "trace_probe.h"
24 #define CREATE_TRACE_POINTS
25 #include "bpf_trace.h"
27 #define bpf_event_rcu_dereference(p) \
28 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
31 struct bpf_trace_module {
32 struct module *module;
33 struct list_head list;
36 static LIST_HEAD(bpf_trace_modules);
37 static DEFINE_MUTEX(bpf_module_mutex);
39 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
41 struct bpf_raw_event_map *btp, *ret = NULL;
42 struct bpf_trace_module *btm;
45 mutex_lock(&bpf_module_mutex);
46 list_for_each_entry(btm, &bpf_trace_modules, list) {
47 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
48 btp = &btm->module->bpf_raw_events[i];
49 if (!strcmp(btp->tp->name, name)) {
50 if (try_module_get(btm->module))
57 mutex_unlock(&bpf_module_mutex);
61 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
65 #endif /* CONFIG_MODULES */
67 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
68 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
71 * trace_call_bpf - invoke BPF program
72 * @call: tracepoint event
73 * @ctx: opaque context pointer
75 * kprobe handlers execute BPF programs via this helper.
76 * Can be used from static tracepoints in the future.
78 * Return: BPF programs always return an integer which is interpreted by
80 * 0 - return from kprobe (event is filtered out)
81 * 1 - store kprobe event into ring buffer
82 * Other values are reserved and currently alias to 1
84 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
88 if (in_nmi()) /* not supported yet */
93 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
95 * since some bpf program is already running on this cpu,
96 * don't call into another bpf program (same or different)
97 * and don't send kprobe event into ring-buffer,
105 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
106 * to all call sites, we did a bpf_prog_array_valid() there to check
107 * whether call->prog_array is empty or not, which is
108 * a heurisitc to speed up execution.
110 * If bpf_prog_array_valid() fetched prog_array was
111 * non-NULL, we go into trace_call_bpf() and do the actual
112 * proper rcu_dereference() under RCU lock.
113 * If it turns out that prog_array is NULL then, we bail out.
114 * For the opposite, if the bpf_prog_array_valid() fetched pointer
115 * was NULL, you'll skip the prog_array with the risk of missing
116 * out of events when it was updated in between this and the
117 * rcu_dereference() which is accepted risk.
119 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
122 __this_cpu_dec(bpf_prog_active);
127 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
128 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
130 regs_set_return_value(regs, rc);
131 override_function_with_return(regs);
135 static const struct bpf_func_proto bpf_override_return_proto = {
136 .func = bpf_override_return,
138 .ret_type = RET_INTEGER,
139 .arg1_type = ARG_PTR_TO_CTX,
140 .arg2_type = ARG_ANYTHING,
144 static __always_inline int
145 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
149 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
150 if (unlikely(ret < 0))
151 memset(dst, 0, size);
155 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
156 const void __user *, unsafe_ptr)
158 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
161 const struct bpf_func_proto bpf_probe_read_user_proto = {
162 .func = bpf_probe_read_user,
164 .ret_type = RET_INTEGER,
165 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
166 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
167 .arg3_type = ARG_ANYTHING,
170 static __always_inline int
171 bpf_probe_read_user_str_common(void *dst, u32 size,
172 const void __user *unsafe_ptr)
176 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
177 if (unlikely(ret < 0))
178 memset(dst, 0, size);
182 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
183 const void __user *, unsafe_ptr)
185 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
188 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
189 .func = bpf_probe_read_user_str,
191 .ret_type = RET_INTEGER,
192 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
193 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
194 .arg3_type = ARG_ANYTHING,
197 static __always_inline int
198 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
200 int ret = security_locked_down(LOCKDOWN_BPF_READ);
202 if (unlikely(ret < 0))
204 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
205 if (unlikely(ret < 0))
209 memset(dst, 0, size);
213 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
214 const void *, unsafe_ptr)
216 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
219 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
220 .func = bpf_probe_read_kernel,
222 .ret_type = RET_INTEGER,
223 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
224 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
225 .arg3_type = ARG_ANYTHING,
228 static __always_inline int
229 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
231 int ret = security_locked_down(LOCKDOWN_BPF_READ);
233 if (unlikely(ret < 0))
237 * The strncpy_from_kernel_nofault() call will likely not fill the
238 * entire buffer, but that's okay in this circumstance as we're probing
239 * arbitrary memory anyway similar to bpf_probe_read_*() and might
240 * as well probe the stack. Thus, memory is explicitly cleared
241 * only in error case, so that improper users ignoring return
242 * code altogether don't copy garbage; otherwise length of string
243 * is returned that can be used for bpf_perf_event_output() et al.
245 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
246 if (unlikely(ret < 0))
251 memset(dst, 0, size);
255 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
256 const void *, unsafe_ptr)
258 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
261 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
262 .func = bpf_probe_read_kernel_str,
264 .ret_type = RET_INTEGER,
265 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
266 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
267 .arg3_type = ARG_ANYTHING,
270 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
271 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
272 const void *, unsafe_ptr)
274 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
275 return bpf_probe_read_user_common(dst, size,
276 (__force void __user *)unsafe_ptr);
278 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
281 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
282 .func = bpf_probe_read_compat,
284 .ret_type = RET_INTEGER,
285 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
286 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
287 .arg3_type = ARG_ANYTHING,
290 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
291 const void *, unsafe_ptr)
293 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
294 return bpf_probe_read_user_str_common(dst, size,
295 (__force void __user *)unsafe_ptr);
297 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
300 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
301 .func = bpf_probe_read_compat_str,
303 .ret_type = RET_INTEGER,
304 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
305 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
306 .arg3_type = ARG_ANYTHING,
308 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
310 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
314 * Ensure we're in user context which is safe for the helper to
315 * run. This helper has no business in a kthread.
317 * access_ok() should prevent writing to non-user memory, but in
318 * some situations (nommu, temporary switch, etc) access_ok() does
319 * not provide enough validation, hence the check on KERNEL_DS.
321 * nmi_uaccess_okay() ensures the probe is not run in an interim
322 * state, when the task or mm are switched. This is specifically
323 * required to prevent the use of temporary mm.
326 if (unlikely(in_interrupt() ||
327 current->flags & (PF_KTHREAD | PF_EXITING)))
329 if (unlikely(uaccess_kernel()))
331 if (unlikely(!nmi_uaccess_okay()))
334 return copy_to_user_nofault(unsafe_ptr, src, size);
337 static const struct bpf_func_proto bpf_probe_write_user_proto = {
338 .func = bpf_probe_write_user,
340 .ret_type = RET_INTEGER,
341 .arg1_type = ARG_ANYTHING,
342 .arg2_type = ARG_PTR_TO_MEM,
343 .arg3_type = ARG_CONST_SIZE,
346 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
348 if (!capable(CAP_SYS_ADMIN))
351 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
352 current->comm, task_pid_nr(current));
354 return &bpf_probe_write_user_proto;
357 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
360 void __user *user_ptr = (__force void __user *)unsafe_ptr;
366 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
367 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
368 strncpy_from_user_nofault(buf, user_ptr, bufsz);
374 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
377 strncpy_from_user_nofault(buf, user_ptr, bufsz);
382 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
384 #define BPF_TRACE_PRINTK_SIZE 1024
386 static __printf(1, 0) int bpf_do_trace_printk(const char *fmt, ...)
388 static char buf[BPF_TRACE_PRINTK_SIZE];
393 raw_spin_lock_irqsave(&trace_printk_lock, flags);
395 ret = vsnprintf(buf, sizeof(buf), fmt, ap);
397 /* vsnprintf() will not append null for zero-length strings */
400 trace_bpf_trace_printk(buf);
401 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
407 * Only limited trace_printk() conversion specifiers allowed:
408 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pB %pks %pus %s
410 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
411 u64, arg2, u64, arg3)
413 int i, mod[3] = {}, fmt_cnt = 0;
414 char buf[64], fmt_ptype;
415 void *unsafe_ptr = NULL;
416 bool str_seen = false;
419 * bpf_check()->check_func_arg()->check_stack_boundary()
420 * guarantees that fmt points to bpf program stack,
421 * fmt_size bytes of it were initialized and fmt_size > 0
423 if (fmt[--fmt_size] != 0)
426 /* check format string for allowed specifiers */
427 for (i = 0; i < fmt_size; i++) {
428 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
437 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
442 } else if (fmt[i] == 'p') {
444 if ((fmt[i + 1] == 'k' ||
445 fmt[i + 1] == 'u') &&
447 fmt_ptype = fmt[i + 1];
452 if (fmt[i + 1] == 'B') {
457 /* disallow any further format extensions */
458 if (fmt[i + 1] != 0 &&
459 !isspace(fmt[i + 1]) &&
460 !ispunct(fmt[i + 1]))
464 } else if (fmt[i] == 's') {
469 /* allow only one '%s' per fmt string */
473 if (fmt[i + 1] != 0 &&
474 !isspace(fmt[i + 1]) &&
475 !ispunct(fmt[i + 1]))
480 unsafe_ptr = (void *)(long)arg1;
484 unsafe_ptr = (void *)(long)arg2;
488 unsafe_ptr = (void *)(long)arg3;
493 bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
503 if (fmt[i] != 'i' && fmt[i] != 'd' &&
504 fmt[i] != 'u' && fmt[i] != 'x')
510 /* Horrid workaround for getting va_list handling working with different
511 * argument type combinations generically for 32 and 64 bit archs.
513 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
514 #define __BPF_TP(...) \
515 bpf_do_trace_printk(fmt, ##__VA_ARGS__)
517 #define __BPF_ARG1_TP(...) \
518 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
519 ? __BPF_TP(arg1, ##__VA_ARGS__) \
520 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
521 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
522 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
524 #define __BPF_ARG2_TP(...) \
525 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
526 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
527 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
528 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
529 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
531 #define __BPF_ARG3_TP(...) \
532 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
533 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
534 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
535 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
536 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
538 return __BPF_TP_EMIT();
541 static const struct bpf_func_proto bpf_trace_printk_proto = {
542 .func = bpf_trace_printk,
544 .ret_type = RET_INTEGER,
545 .arg1_type = ARG_PTR_TO_MEM,
546 .arg2_type = ARG_CONST_SIZE,
549 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
552 * This program might be calling bpf_trace_printk,
553 * so enable the associated bpf_trace/bpf_trace_printk event.
554 * Repeat this each time as it is possible a user has
555 * disabled bpf_trace_printk events. By loading a program
556 * calling bpf_trace_printk() however the user has expressed
557 * the intent to see such events.
559 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
560 pr_warn_ratelimited("could not enable bpf_trace_printk events");
562 return &bpf_trace_printk_proto;
565 #define MAX_SEQ_PRINTF_VARARGS 12
566 #define MAX_SEQ_PRINTF_MAX_MEMCPY 6
567 #define MAX_SEQ_PRINTF_STR_LEN 128
569 struct bpf_seq_printf_buf {
570 char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
572 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
573 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
575 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
576 const void *, data, u32, data_len)
578 int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
579 int i, buf_used, copy_size, num_args;
580 u64 params[MAX_SEQ_PRINTF_VARARGS];
581 struct bpf_seq_printf_buf *bufs;
582 const u64 *args = data;
584 buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
585 if (WARN_ON_ONCE(buf_used > 1)) {
590 bufs = this_cpu_ptr(&bpf_seq_printf_buf);
593 * bpf_check()->check_func_arg()->check_stack_boundary()
594 * guarantees that fmt points to bpf program stack,
595 * fmt_size bytes of it were initialized and fmt_size > 0
597 if (fmt[--fmt_size] != 0)
603 for (i = 0; i < fmt_size; i++) {
605 if (fmt[i + 1] == '%')
607 else if (!data || !data_len)
612 num_args = data_len / 8;
614 /* check format string for allowed specifiers */
615 for (i = 0; i < fmt_size; i++) {
616 /* only printable ascii for now. */
617 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
625 if (fmt[i + 1] == '%') {
630 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
635 if (fmt_cnt >= num_args) {
640 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
643 /* skip optional "[0 +-][num]" width formating field */
644 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
647 if (fmt[i] >= '1' && fmt[i] <= '9') {
649 while (fmt[i] >= '0' && fmt[i] <= '9')
656 /* try our best to copy */
657 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
662 unsafe_ptr = (void *)(long)args[fmt_cnt];
663 err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
664 unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
666 bufs->buf[memcpy_cnt][0] = '\0';
667 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
675 if (fmt[i + 1] == 0 ||
679 /* just kernel pointers */
680 params[fmt_cnt] = args[fmt_cnt];
685 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
686 if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
690 if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
695 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
701 copy_size = (fmt[i + 2] == '4') ? 4 : 16;
703 err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
704 (void *) (long) args[fmt_cnt],
707 memset(bufs->buf[memcpy_cnt], 0, copy_size);
708 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
722 if (fmt[i] != 'i' && fmt[i] != 'd' &&
723 fmt[i] != 'u' && fmt[i] != 'x' &&
729 params[fmt_cnt] = args[fmt_cnt];
733 /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
734 * all of them to seq_printf().
736 seq_printf(m, fmt, params[0], params[1], params[2], params[3],
737 params[4], params[5], params[6], params[7], params[8],
738 params[9], params[10], params[11]);
740 err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
742 this_cpu_dec(bpf_seq_printf_buf_used);
746 BTF_ID_LIST(bpf_seq_printf_btf_ids)
747 BTF_ID(struct, seq_file)
749 static const struct bpf_func_proto bpf_seq_printf_proto = {
750 .func = bpf_seq_printf,
752 .ret_type = RET_INTEGER,
753 .arg1_type = ARG_PTR_TO_BTF_ID,
754 .arg2_type = ARG_PTR_TO_MEM,
755 .arg3_type = ARG_CONST_SIZE,
756 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
757 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
758 .btf_id = bpf_seq_printf_btf_ids,
761 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
763 return seq_write(m, data, len) ? -EOVERFLOW : 0;
766 BTF_ID_LIST(bpf_seq_write_btf_ids)
767 BTF_ID(struct, seq_file)
769 static const struct bpf_func_proto bpf_seq_write_proto = {
770 .func = bpf_seq_write,
772 .ret_type = RET_INTEGER,
773 .arg1_type = ARG_PTR_TO_BTF_ID,
774 .arg2_type = ARG_PTR_TO_MEM,
775 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
776 .btf_id = bpf_seq_write_btf_ids,
779 static __always_inline int
780 get_map_perf_counter(struct bpf_map *map, u64 flags,
781 u64 *value, u64 *enabled, u64 *running)
783 struct bpf_array *array = container_of(map, struct bpf_array, map);
784 unsigned int cpu = smp_processor_id();
785 u64 index = flags & BPF_F_INDEX_MASK;
786 struct bpf_event_entry *ee;
788 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
790 if (index == BPF_F_CURRENT_CPU)
792 if (unlikely(index >= array->map.max_entries))
795 ee = READ_ONCE(array->ptrs[index]);
799 return perf_event_read_local(ee->event, value, enabled, running);
802 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
807 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
809 * this api is ugly since we miss [-22..-2] range of valid
810 * counter values, but that's uapi
817 static const struct bpf_func_proto bpf_perf_event_read_proto = {
818 .func = bpf_perf_event_read,
820 .ret_type = RET_INTEGER,
821 .arg1_type = ARG_CONST_MAP_PTR,
822 .arg2_type = ARG_ANYTHING,
825 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
826 struct bpf_perf_event_value *, buf, u32, size)
830 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
832 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
838 memset(buf, 0, size);
842 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
843 .func = bpf_perf_event_read_value,
845 .ret_type = RET_INTEGER,
846 .arg1_type = ARG_CONST_MAP_PTR,
847 .arg2_type = ARG_ANYTHING,
848 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
849 .arg4_type = ARG_CONST_SIZE,
852 static __always_inline u64
853 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
854 u64 flags, struct perf_sample_data *sd)
856 struct bpf_array *array = container_of(map, struct bpf_array, map);
857 unsigned int cpu = smp_processor_id();
858 u64 index = flags & BPF_F_INDEX_MASK;
859 struct bpf_event_entry *ee;
860 struct perf_event *event;
862 if (index == BPF_F_CURRENT_CPU)
864 if (unlikely(index >= array->map.max_entries))
867 ee = READ_ONCE(array->ptrs[index]);
872 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
873 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
876 if (unlikely(event->oncpu != cpu))
879 return perf_event_output(event, sd, regs);
883 * Support executing tracepoints in normal, irq, and nmi context that each call
884 * bpf_perf_event_output
886 struct bpf_trace_sample_data {
887 struct perf_sample_data sds[3];
890 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
891 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
892 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
893 u64, flags, void *, data, u64, size)
895 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
896 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
897 struct perf_raw_record raw = {
903 struct perf_sample_data *sd;
906 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
911 sd = &sds->sds[nest_level - 1];
913 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
918 perf_sample_data_init(sd, 0, 0);
921 err = __bpf_perf_event_output(regs, map, flags, sd);
924 this_cpu_dec(bpf_trace_nest_level);
928 static const struct bpf_func_proto bpf_perf_event_output_proto = {
929 .func = bpf_perf_event_output,
931 .ret_type = RET_INTEGER,
932 .arg1_type = ARG_PTR_TO_CTX,
933 .arg2_type = ARG_CONST_MAP_PTR,
934 .arg3_type = ARG_ANYTHING,
935 .arg4_type = ARG_PTR_TO_MEM,
936 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
939 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
940 struct bpf_nested_pt_regs {
941 struct pt_regs regs[3];
943 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
944 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
946 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
947 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
949 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
950 struct perf_raw_frag frag = {
955 struct perf_raw_record raw = {
958 .next = ctx_size ? &frag : NULL,
964 struct perf_sample_data *sd;
965 struct pt_regs *regs;
968 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
972 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
973 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
975 perf_fetch_caller_regs(regs);
976 perf_sample_data_init(sd, 0, 0);
979 ret = __bpf_perf_event_output(regs, map, flags, sd);
981 this_cpu_dec(bpf_event_output_nest_level);
985 BPF_CALL_0(bpf_get_current_task)
987 return (long) current;
990 const struct bpf_func_proto bpf_get_current_task_proto = {
991 .func = bpf_get_current_task,
993 .ret_type = RET_INTEGER,
996 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
998 struct bpf_array *array = container_of(map, struct bpf_array, map);
1001 if (unlikely(idx >= array->map.max_entries))
1004 cgrp = READ_ONCE(array->ptrs[idx]);
1005 if (unlikely(!cgrp))
1008 return task_under_cgroup_hierarchy(current, cgrp);
1011 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
1012 .func = bpf_current_task_under_cgroup,
1014 .ret_type = RET_INTEGER,
1015 .arg1_type = ARG_CONST_MAP_PTR,
1016 .arg2_type = ARG_ANYTHING,
1019 struct send_signal_irq_work {
1020 struct irq_work irq_work;
1021 struct task_struct *task;
1026 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
1028 static void do_bpf_send_signal(struct irq_work *entry)
1030 struct send_signal_irq_work *work;
1032 work = container_of(entry, struct send_signal_irq_work, irq_work);
1033 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
1036 static int bpf_send_signal_common(u32 sig, enum pid_type type)
1038 struct send_signal_irq_work *work = NULL;
1040 /* Similar to bpf_probe_write_user, task needs to be
1041 * in a sound condition and kernel memory access be
1042 * permitted in order to send signal to the current
1045 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1047 if (unlikely(uaccess_kernel()))
1049 if (unlikely(!nmi_uaccess_okay()))
1052 if (irqs_disabled()) {
1053 /* Do an early check on signal validity. Otherwise,
1054 * the error is lost in deferred irq_work.
1056 if (unlikely(!valid_signal(sig)))
1059 work = this_cpu_ptr(&send_signal_work);
1060 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
1063 /* Add the current task, which is the target of sending signal,
1064 * to the irq_work. The current task may change when queued
1065 * irq works get executed.
1067 work->task = current;
1070 irq_work_queue(&work->irq_work);
1074 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1077 BPF_CALL_1(bpf_send_signal, u32, sig)
1079 return bpf_send_signal_common(sig, PIDTYPE_TGID);
1082 static const struct bpf_func_proto bpf_send_signal_proto = {
1083 .func = bpf_send_signal,
1085 .ret_type = RET_INTEGER,
1086 .arg1_type = ARG_ANYTHING,
1089 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1091 return bpf_send_signal_common(sig, PIDTYPE_PID);
1094 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1095 .func = bpf_send_signal_thread,
1097 .ret_type = RET_INTEGER,
1098 .arg1_type = ARG_ANYTHING,
1101 const struct bpf_func_proto *
1102 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1105 case BPF_FUNC_map_lookup_elem:
1106 return &bpf_map_lookup_elem_proto;
1107 case BPF_FUNC_map_update_elem:
1108 return &bpf_map_update_elem_proto;
1109 case BPF_FUNC_map_delete_elem:
1110 return &bpf_map_delete_elem_proto;
1111 case BPF_FUNC_map_push_elem:
1112 return &bpf_map_push_elem_proto;
1113 case BPF_FUNC_map_pop_elem:
1114 return &bpf_map_pop_elem_proto;
1115 case BPF_FUNC_map_peek_elem:
1116 return &bpf_map_peek_elem_proto;
1117 case BPF_FUNC_ktime_get_ns:
1118 return &bpf_ktime_get_ns_proto;
1119 case BPF_FUNC_ktime_get_boot_ns:
1120 return &bpf_ktime_get_boot_ns_proto;
1121 case BPF_FUNC_tail_call:
1122 return &bpf_tail_call_proto;
1123 case BPF_FUNC_get_current_pid_tgid:
1124 return &bpf_get_current_pid_tgid_proto;
1125 case BPF_FUNC_get_current_task:
1126 return &bpf_get_current_task_proto;
1127 case BPF_FUNC_get_current_uid_gid:
1128 return &bpf_get_current_uid_gid_proto;
1129 case BPF_FUNC_get_current_comm:
1130 return &bpf_get_current_comm_proto;
1131 case BPF_FUNC_trace_printk:
1132 return bpf_get_trace_printk_proto();
1133 case BPF_FUNC_get_smp_processor_id:
1134 return &bpf_get_smp_processor_id_proto;
1135 case BPF_FUNC_get_numa_node_id:
1136 return &bpf_get_numa_node_id_proto;
1137 case BPF_FUNC_perf_event_read:
1138 return &bpf_perf_event_read_proto;
1139 case BPF_FUNC_probe_write_user:
1140 return bpf_get_probe_write_proto();
1141 case BPF_FUNC_current_task_under_cgroup:
1142 return &bpf_current_task_under_cgroup_proto;
1143 case BPF_FUNC_get_prandom_u32:
1144 return &bpf_get_prandom_u32_proto;
1145 case BPF_FUNC_probe_read_user:
1146 return &bpf_probe_read_user_proto;
1147 case BPF_FUNC_probe_read_kernel:
1148 return &bpf_probe_read_kernel_proto;
1149 case BPF_FUNC_probe_read_user_str:
1150 return &bpf_probe_read_user_str_proto;
1151 case BPF_FUNC_probe_read_kernel_str:
1152 return &bpf_probe_read_kernel_str_proto;
1153 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1154 case BPF_FUNC_probe_read:
1155 return &bpf_probe_read_compat_proto;
1156 case BPF_FUNC_probe_read_str:
1157 return &bpf_probe_read_compat_str_proto;
1159 #ifdef CONFIG_CGROUPS
1160 case BPF_FUNC_get_current_cgroup_id:
1161 return &bpf_get_current_cgroup_id_proto;
1163 case BPF_FUNC_send_signal:
1164 return &bpf_send_signal_proto;
1165 case BPF_FUNC_send_signal_thread:
1166 return &bpf_send_signal_thread_proto;
1167 case BPF_FUNC_perf_event_read_value:
1168 return &bpf_perf_event_read_value_proto;
1169 case BPF_FUNC_get_ns_current_pid_tgid:
1170 return &bpf_get_ns_current_pid_tgid_proto;
1171 case BPF_FUNC_ringbuf_output:
1172 return &bpf_ringbuf_output_proto;
1173 case BPF_FUNC_ringbuf_reserve:
1174 return &bpf_ringbuf_reserve_proto;
1175 case BPF_FUNC_ringbuf_submit:
1176 return &bpf_ringbuf_submit_proto;
1177 case BPF_FUNC_ringbuf_discard:
1178 return &bpf_ringbuf_discard_proto;
1179 case BPF_FUNC_ringbuf_query:
1180 return &bpf_ringbuf_query_proto;
1181 case BPF_FUNC_jiffies64:
1182 return &bpf_jiffies64_proto;
1183 case BPF_FUNC_get_task_stack:
1184 return &bpf_get_task_stack_proto;
1190 static const struct bpf_func_proto *
1191 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1194 case BPF_FUNC_perf_event_output:
1195 return &bpf_perf_event_output_proto;
1196 case BPF_FUNC_get_stackid:
1197 return &bpf_get_stackid_proto;
1198 case BPF_FUNC_get_stack:
1199 return &bpf_get_stack_proto;
1200 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1201 case BPF_FUNC_override_return:
1202 return &bpf_override_return_proto;
1205 return bpf_tracing_func_proto(func_id, prog);
1209 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1210 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1211 const struct bpf_prog *prog,
1212 struct bpf_insn_access_aux *info)
1214 if (off < 0 || off >= sizeof(struct pt_regs))
1216 if (type != BPF_READ)
1218 if (off % size != 0)
1221 * Assertion for 32 bit to make sure last 8 byte access
1222 * (BPF_DW) to the last 4 byte member is disallowed.
1224 if (off + size > sizeof(struct pt_regs))
1230 const struct bpf_verifier_ops kprobe_verifier_ops = {
1231 .get_func_proto = kprobe_prog_func_proto,
1232 .is_valid_access = kprobe_prog_is_valid_access,
1235 const struct bpf_prog_ops kprobe_prog_ops = {
1238 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1239 u64, flags, void *, data, u64, size)
1241 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1244 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1245 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1246 * from there and call the same bpf_perf_event_output() helper inline.
1248 return ____bpf_perf_event_output(regs, map, flags, data, size);
1251 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1252 .func = bpf_perf_event_output_tp,
1254 .ret_type = RET_INTEGER,
1255 .arg1_type = ARG_PTR_TO_CTX,
1256 .arg2_type = ARG_CONST_MAP_PTR,
1257 .arg3_type = ARG_ANYTHING,
1258 .arg4_type = ARG_PTR_TO_MEM,
1259 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1262 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1265 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1268 * Same comment as in bpf_perf_event_output_tp(), only that this time
1269 * the other helper's function body cannot be inlined due to being
1270 * external, thus we need to call raw helper function.
1272 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1276 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1277 .func = bpf_get_stackid_tp,
1279 .ret_type = RET_INTEGER,
1280 .arg1_type = ARG_PTR_TO_CTX,
1281 .arg2_type = ARG_CONST_MAP_PTR,
1282 .arg3_type = ARG_ANYTHING,
1285 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1288 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1290 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1291 (unsigned long) size, flags, 0);
1294 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1295 .func = bpf_get_stack_tp,
1297 .ret_type = RET_INTEGER,
1298 .arg1_type = ARG_PTR_TO_CTX,
1299 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1300 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1301 .arg4_type = ARG_ANYTHING,
1304 static const struct bpf_func_proto *
1305 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1308 case BPF_FUNC_perf_event_output:
1309 return &bpf_perf_event_output_proto_tp;
1310 case BPF_FUNC_get_stackid:
1311 return &bpf_get_stackid_proto_tp;
1312 case BPF_FUNC_get_stack:
1313 return &bpf_get_stack_proto_tp;
1315 return bpf_tracing_func_proto(func_id, prog);
1319 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1320 const struct bpf_prog *prog,
1321 struct bpf_insn_access_aux *info)
1323 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1325 if (type != BPF_READ)
1327 if (off % size != 0)
1330 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1334 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1335 .get_func_proto = tp_prog_func_proto,
1336 .is_valid_access = tp_prog_is_valid_access,
1339 const struct bpf_prog_ops tracepoint_prog_ops = {
1342 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1343 struct bpf_perf_event_value *, buf, u32, size)
1347 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1349 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1355 memset(buf, 0, size);
1359 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1360 .func = bpf_perf_prog_read_value,
1362 .ret_type = RET_INTEGER,
1363 .arg1_type = ARG_PTR_TO_CTX,
1364 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1365 .arg3_type = ARG_CONST_SIZE,
1368 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1369 void *, buf, u32, size, u64, flags)
1374 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1375 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1378 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1381 if (unlikely(!br_stack))
1384 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1385 return br_stack->nr * br_entry_size;
1387 if (!buf || (size % br_entry_size != 0))
1390 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1391 memcpy(buf, br_stack->entries, to_copy);
1397 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1398 .func = bpf_read_branch_records,
1400 .ret_type = RET_INTEGER,
1401 .arg1_type = ARG_PTR_TO_CTX,
1402 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1403 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1404 .arg4_type = ARG_ANYTHING,
1407 static const struct bpf_func_proto *
1408 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1411 case BPF_FUNC_perf_event_output:
1412 return &bpf_perf_event_output_proto_tp;
1413 case BPF_FUNC_get_stackid:
1414 return &bpf_get_stackid_proto_pe;
1415 case BPF_FUNC_get_stack:
1416 return &bpf_get_stack_proto_pe;
1417 case BPF_FUNC_perf_prog_read_value:
1418 return &bpf_perf_prog_read_value_proto;
1419 case BPF_FUNC_read_branch_records:
1420 return &bpf_read_branch_records_proto;
1422 return bpf_tracing_func_proto(func_id, prog);
1427 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1428 * to avoid potential recursive reuse issue when/if tracepoints are added
1429 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1431 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1432 * in normal, irq, and nmi context.
1434 struct bpf_raw_tp_regs {
1435 struct pt_regs regs[3];
1437 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1438 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1439 static struct pt_regs *get_bpf_raw_tp_regs(void)
1441 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1442 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1444 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1445 this_cpu_dec(bpf_raw_tp_nest_level);
1446 return ERR_PTR(-EBUSY);
1449 return &tp_regs->regs[nest_level - 1];
1452 static void put_bpf_raw_tp_regs(void)
1454 this_cpu_dec(bpf_raw_tp_nest_level);
1457 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1458 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1460 struct pt_regs *regs = get_bpf_raw_tp_regs();
1464 return PTR_ERR(regs);
1466 perf_fetch_caller_regs(regs);
1467 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1469 put_bpf_raw_tp_regs();
1473 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1474 .func = bpf_perf_event_output_raw_tp,
1476 .ret_type = RET_INTEGER,
1477 .arg1_type = ARG_PTR_TO_CTX,
1478 .arg2_type = ARG_CONST_MAP_PTR,
1479 .arg3_type = ARG_ANYTHING,
1480 .arg4_type = ARG_PTR_TO_MEM,
1481 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1484 extern const struct bpf_func_proto bpf_skb_output_proto;
1485 extern const struct bpf_func_proto bpf_xdp_output_proto;
1487 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1488 struct bpf_map *, map, u64, flags)
1490 struct pt_regs *regs = get_bpf_raw_tp_regs();
1494 return PTR_ERR(regs);
1496 perf_fetch_caller_regs(regs);
1497 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1498 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1500 put_bpf_raw_tp_regs();
1504 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1505 .func = bpf_get_stackid_raw_tp,
1507 .ret_type = RET_INTEGER,
1508 .arg1_type = ARG_PTR_TO_CTX,
1509 .arg2_type = ARG_CONST_MAP_PTR,
1510 .arg3_type = ARG_ANYTHING,
1513 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1514 void *, buf, u32, size, u64, flags)
1516 struct pt_regs *regs = get_bpf_raw_tp_regs();
1520 return PTR_ERR(regs);
1522 perf_fetch_caller_regs(regs);
1523 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1524 (unsigned long) size, flags, 0);
1525 put_bpf_raw_tp_regs();
1529 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1530 .func = bpf_get_stack_raw_tp,
1532 .ret_type = RET_INTEGER,
1533 .arg1_type = ARG_PTR_TO_CTX,
1534 .arg2_type = ARG_PTR_TO_MEM,
1535 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1536 .arg4_type = ARG_ANYTHING,
1539 static const struct bpf_func_proto *
1540 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1543 case BPF_FUNC_perf_event_output:
1544 return &bpf_perf_event_output_proto_raw_tp;
1545 case BPF_FUNC_get_stackid:
1546 return &bpf_get_stackid_proto_raw_tp;
1547 case BPF_FUNC_get_stack:
1548 return &bpf_get_stack_proto_raw_tp;
1550 return bpf_tracing_func_proto(func_id, prog);
1554 const struct bpf_func_proto *
1555 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1559 case BPF_FUNC_skb_output:
1560 return &bpf_skb_output_proto;
1561 case BPF_FUNC_xdp_output:
1562 return &bpf_xdp_output_proto;
1563 case BPF_FUNC_skc_to_tcp6_sock:
1564 return &bpf_skc_to_tcp6_sock_proto;
1565 case BPF_FUNC_skc_to_tcp_sock:
1566 return &bpf_skc_to_tcp_sock_proto;
1567 case BPF_FUNC_skc_to_tcp_timewait_sock:
1568 return &bpf_skc_to_tcp_timewait_sock_proto;
1569 case BPF_FUNC_skc_to_tcp_request_sock:
1570 return &bpf_skc_to_tcp_request_sock_proto;
1571 case BPF_FUNC_skc_to_udp6_sock:
1572 return &bpf_skc_to_udp6_sock_proto;
1574 case BPF_FUNC_seq_printf:
1575 return prog->expected_attach_type == BPF_TRACE_ITER ?
1576 &bpf_seq_printf_proto :
1578 case BPF_FUNC_seq_write:
1579 return prog->expected_attach_type == BPF_TRACE_ITER ?
1580 &bpf_seq_write_proto :
1583 return raw_tp_prog_func_proto(func_id, prog);
1587 static bool raw_tp_prog_is_valid_access(int off, int size,
1588 enum bpf_access_type type,
1589 const struct bpf_prog *prog,
1590 struct bpf_insn_access_aux *info)
1592 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1594 if (type != BPF_READ)
1596 if (off % size != 0)
1601 static bool tracing_prog_is_valid_access(int off, int size,
1602 enum bpf_access_type type,
1603 const struct bpf_prog *prog,
1604 struct bpf_insn_access_aux *info)
1606 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1608 if (type != BPF_READ)
1610 if (off % size != 0)
1612 return btf_ctx_access(off, size, type, prog, info);
1615 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1616 const union bpf_attr *kattr,
1617 union bpf_attr __user *uattr)
1622 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1623 .get_func_proto = raw_tp_prog_func_proto,
1624 .is_valid_access = raw_tp_prog_is_valid_access,
1627 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1630 const struct bpf_verifier_ops tracing_verifier_ops = {
1631 .get_func_proto = tracing_prog_func_proto,
1632 .is_valid_access = tracing_prog_is_valid_access,
1635 const struct bpf_prog_ops tracing_prog_ops = {
1636 .test_run = bpf_prog_test_run_tracing,
1639 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1640 enum bpf_access_type type,
1641 const struct bpf_prog *prog,
1642 struct bpf_insn_access_aux *info)
1645 if (size != sizeof(u64) || type != BPF_READ)
1647 info->reg_type = PTR_TO_TP_BUFFER;
1649 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1652 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1653 .get_func_proto = raw_tp_prog_func_proto,
1654 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1657 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1660 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1661 const struct bpf_prog *prog,
1662 struct bpf_insn_access_aux *info)
1664 const int size_u64 = sizeof(u64);
1666 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1668 if (type != BPF_READ)
1670 if (off % size != 0) {
1671 if (sizeof(unsigned long) != 4)
1675 if (off % size != 4)
1680 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1681 bpf_ctx_record_field_size(info, size_u64);
1682 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1685 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1686 bpf_ctx_record_field_size(info, size_u64);
1687 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1691 if (size != sizeof(long))
1698 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1699 const struct bpf_insn *si,
1700 struct bpf_insn *insn_buf,
1701 struct bpf_prog *prog, u32 *target_size)
1703 struct bpf_insn *insn = insn_buf;
1706 case offsetof(struct bpf_perf_event_data, sample_period):
1707 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1708 data), si->dst_reg, si->src_reg,
1709 offsetof(struct bpf_perf_event_data_kern, data));
1710 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1711 bpf_target_off(struct perf_sample_data, period, 8,
1714 case offsetof(struct bpf_perf_event_data, addr):
1715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1716 data), si->dst_reg, si->src_reg,
1717 offsetof(struct bpf_perf_event_data_kern, data));
1718 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1719 bpf_target_off(struct perf_sample_data, addr, 8,
1723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1724 regs), si->dst_reg, si->src_reg,
1725 offsetof(struct bpf_perf_event_data_kern, regs));
1726 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1731 return insn - insn_buf;
1734 const struct bpf_verifier_ops perf_event_verifier_ops = {
1735 .get_func_proto = pe_prog_func_proto,
1736 .is_valid_access = pe_prog_is_valid_access,
1737 .convert_ctx_access = pe_prog_convert_ctx_access,
1740 const struct bpf_prog_ops perf_event_prog_ops = {
1743 static DEFINE_MUTEX(bpf_event_mutex);
1745 #define BPF_TRACE_MAX_PROGS 64
1747 int perf_event_attach_bpf_prog(struct perf_event *event,
1748 struct bpf_prog *prog)
1750 struct bpf_prog_array *old_array;
1751 struct bpf_prog_array *new_array;
1755 * Kprobe override only works if they are on the function entry,
1756 * and only if they are on the opt-in list.
1758 if (prog->kprobe_override &&
1759 (!trace_kprobe_on_func_entry(event->tp_event) ||
1760 !trace_kprobe_error_injectable(event->tp_event)))
1763 mutex_lock(&bpf_event_mutex);
1768 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1770 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1775 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1779 /* set the new array to event->tp_event and set event->prog */
1781 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1782 bpf_prog_array_free(old_array);
1785 mutex_unlock(&bpf_event_mutex);
1789 void perf_event_detach_bpf_prog(struct perf_event *event)
1791 struct bpf_prog_array *old_array;
1792 struct bpf_prog_array *new_array;
1795 mutex_lock(&bpf_event_mutex);
1800 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1801 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1805 bpf_prog_array_delete_safe(old_array, event->prog);
1807 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1808 bpf_prog_array_free(old_array);
1811 bpf_prog_put(event->prog);
1815 mutex_unlock(&bpf_event_mutex);
1818 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1820 struct perf_event_query_bpf __user *uquery = info;
1821 struct perf_event_query_bpf query = {};
1822 struct bpf_prog_array *progs;
1823 u32 *ids, prog_cnt, ids_len;
1826 if (!perfmon_capable())
1828 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1830 if (copy_from_user(&query, uquery, sizeof(query)))
1833 ids_len = query.ids_len;
1834 if (ids_len > BPF_TRACE_MAX_PROGS)
1836 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1840 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1841 * is required when user only wants to check for uquery->prog_cnt.
1842 * There is no need to check for it since the case is handled
1843 * gracefully in bpf_prog_array_copy_info.
1846 mutex_lock(&bpf_event_mutex);
1847 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1848 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1849 mutex_unlock(&bpf_event_mutex);
1851 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1852 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1859 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1860 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1862 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1864 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1866 for (; btp < __stop__bpf_raw_tp; btp++) {
1867 if (!strcmp(btp->tp->name, name))
1871 return bpf_get_raw_tracepoint_module(name);
1874 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1876 struct module *mod = __module_address((unsigned long)btp);
1882 static __always_inline
1883 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1887 (void) BPF_PROG_RUN(prog, args);
1891 #define UNPACK(...) __VA_ARGS__
1892 #define REPEAT_1(FN, DL, X, ...) FN(X)
1893 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1894 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1895 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1896 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1897 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1898 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1899 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1900 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1901 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1902 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1903 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1904 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1906 #define SARG(X) u64 arg##X
1907 #define COPY(X) args[X] = arg##X
1909 #define __DL_COM (,)
1910 #define __DL_SEM (;)
1912 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1914 #define BPF_TRACE_DEFN_x(x) \
1915 void bpf_trace_run##x(struct bpf_prog *prog, \
1916 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1919 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1920 __bpf_trace_run(prog, args); \
1922 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1923 BPF_TRACE_DEFN_x(1);
1924 BPF_TRACE_DEFN_x(2);
1925 BPF_TRACE_DEFN_x(3);
1926 BPF_TRACE_DEFN_x(4);
1927 BPF_TRACE_DEFN_x(5);
1928 BPF_TRACE_DEFN_x(6);
1929 BPF_TRACE_DEFN_x(7);
1930 BPF_TRACE_DEFN_x(8);
1931 BPF_TRACE_DEFN_x(9);
1932 BPF_TRACE_DEFN_x(10);
1933 BPF_TRACE_DEFN_x(11);
1934 BPF_TRACE_DEFN_x(12);
1936 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1938 struct tracepoint *tp = btp->tp;
1941 * check that program doesn't access arguments beyond what's
1942 * available in this tracepoint
1944 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1947 if (prog->aux->max_tp_access > btp->writable_size)
1950 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1953 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1955 return __bpf_probe_register(btp, prog);
1958 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1960 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1963 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1964 u32 *fd_type, const char **buf,
1965 u64 *probe_offset, u64 *probe_addr)
1967 bool is_tracepoint, is_syscall_tp;
1968 struct bpf_prog *prog;
1975 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1976 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1979 *prog_id = prog->aux->id;
1980 flags = event->tp_event->flags;
1981 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1982 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1984 if (is_tracepoint || is_syscall_tp) {
1985 *buf = is_tracepoint ? event->tp_event->tp->name
1986 : event->tp_event->name;
1987 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1988 *probe_offset = 0x0;
1993 #ifdef CONFIG_KPROBE_EVENTS
1994 if (flags & TRACE_EVENT_FL_KPROBE)
1995 err = bpf_get_kprobe_info(event, fd_type, buf,
1996 probe_offset, probe_addr,
1997 event->attr.type == PERF_TYPE_TRACEPOINT);
1999 #ifdef CONFIG_UPROBE_EVENTS
2000 if (flags & TRACE_EVENT_FL_UPROBE)
2001 err = bpf_get_uprobe_info(event, fd_type, buf,
2003 event->attr.type == PERF_TYPE_TRACEPOINT);
2010 static int __init send_signal_irq_work_init(void)
2013 struct send_signal_irq_work *work;
2015 for_each_possible_cpu(cpu) {
2016 work = per_cpu_ptr(&send_signal_work, cpu);
2017 init_irq_work(&work->irq_work, do_bpf_send_signal);
2022 subsys_initcall(send_signal_irq_work_init);
2024 #ifdef CONFIG_MODULES
2025 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2028 struct bpf_trace_module *btm, *tmp;
2029 struct module *mod = module;
2031 if (mod->num_bpf_raw_events == 0 ||
2032 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2035 mutex_lock(&bpf_module_mutex);
2038 case MODULE_STATE_COMING:
2039 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2041 btm->module = module;
2042 list_add(&btm->list, &bpf_trace_modules);
2045 case MODULE_STATE_GOING:
2046 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2047 if (btm->module == module) {
2048 list_del(&btm->list);
2056 mutex_unlock(&bpf_module_mutex);
2061 static struct notifier_block bpf_module_nb = {
2062 .notifier_call = bpf_event_notify,
2065 static int __init bpf_event_init(void)
2067 register_module_notifier(&bpf_module_nb);
2071 fs_initcall(bpf_event_init);
2072 #endif /* CONFIG_MODULES */