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/syscalls.h>
15 #include <linux/error-injection.h>
19 #include "trace_probe.h"
22 #define bpf_event_rcu_dereference(p) \
23 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
26 struct bpf_trace_module {
27 struct module *module;
28 struct list_head list;
31 static LIST_HEAD(bpf_trace_modules);
32 static DEFINE_MUTEX(bpf_module_mutex);
34 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
36 struct bpf_raw_event_map *btp, *ret = NULL;
37 struct bpf_trace_module *btm;
40 mutex_lock(&bpf_module_mutex);
41 list_for_each_entry(btm, &bpf_trace_modules, list) {
42 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
43 btp = &btm->module->bpf_raw_events[i];
44 if (!strcmp(btp->tp->name, name)) {
45 if (try_module_get(btm->module))
52 mutex_unlock(&bpf_module_mutex);
56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
60 #endif /* CONFIG_MODULES */
62 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
63 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
66 * trace_call_bpf - invoke BPF program
67 * @call: tracepoint event
68 * @ctx: opaque context pointer
70 * kprobe handlers execute BPF programs via this helper.
71 * Can be used from static tracepoints in the future.
73 * Return: BPF programs always return an integer which is interpreted by
75 * 0 - return from kprobe (event is filtered out)
76 * 1 - store kprobe event into ring buffer
77 * Other values are reserved and currently alias to 1
79 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
83 if (in_nmi()) /* not supported yet */
88 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
90 * since some bpf program is already running on this cpu,
91 * don't call into another bpf program (same or different)
92 * and don't send kprobe event into ring-buffer,
100 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
101 * to all call sites, we did a bpf_prog_array_valid() there to check
102 * whether call->prog_array is empty or not, which is
103 * a heurisitc to speed up execution.
105 * If bpf_prog_array_valid() fetched prog_array was
106 * non-NULL, we go into trace_call_bpf() and do the actual
107 * proper rcu_dereference() under RCU lock.
108 * If it turns out that prog_array is NULL then, we bail out.
109 * For the opposite, if the bpf_prog_array_valid() fetched pointer
110 * was NULL, you'll skip the prog_array with the risk of missing
111 * out of events when it was updated in between this and the
112 * rcu_dereference() which is accepted risk.
114 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
117 __this_cpu_dec(bpf_prog_active);
122 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
123 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
125 regs_set_return_value(regs, rc);
126 override_function_with_return(regs);
130 static const struct bpf_func_proto bpf_override_return_proto = {
131 .func = bpf_override_return,
133 .ret_type = RET_INTEGER,
134 .arg1_type = ARG_PTR_TO_CTX,
135 .arg2_type = ARG_ANYTHING,
139 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
140 const void __user *, unsafe_ptr)
142 int ret = probe_user_read(dst, unsafe_ptr, size);
144 if (unlikely(ret < 0))
145 memset(dst, 0, size);
150 const struct bpf_func_proto bpf_probe_read_user_proto = {
151 .func = bpf_probe_read_user,
153 .ret_type = RET_INTEGER,
154 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
155 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
156 .arg3_type = ARG_ANYTHING,
159 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
160 const void __user *, unsafe_ptr)
162 int ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
164 if (unlikely(ret < 0))
165 memset(dst, 0, size);
170 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
171 .func = bpf_probe_read_user_str,
173 .ret_type = RET_INTEGER,
174 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
175 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
176 .arg3_type = ARG_ANYTHING,
179 static __always_inline int
180 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr,
183 int ret = security_locked_down(LOCKDOWN_BPF_READ);
185 if (unlikely(ret < 0))
187 ret = compat ? probe_kernel_read(dst, unsafe_ptr, size) :
188 probe_kernel_read_strict(dst, unsafe_ptr, size);
189 if (unlikely(ret < 0))
191 memset(dst, 0, size);
195 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
196 const void *, unsafe_ptr)
198 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, false);
201 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
202 .func = bpf_probe_read_kernel,
204 .ret_type = RET_INTEGER,
205 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
206 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
207 .arg3_type = ARG_ANYTHING,
210 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
211 const void *, unsafe_ptr)
213 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, true);
216 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
217 .func = bpf_probe_read_compat,
219 .ret_type = RET_INTEGER,
220 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
221 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
222 .arg3_type = ARG_ANYTHING,
225 static __always_inline int
226 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr,
229 int ret = security_locked_down(LOCKDOWN_BPF_READ);
231 if (unlikely(ret < 0))
234 * The strncpy_from_unsafe_*() call will likely not fill the entire
235 * buffer, but that's okay in this circumstance as we're probing
236 * arbitrary memory anyway similar to bpf_probe_read_*() and might
237 * as well probe the stack. Thus, memory is explicitly cleared
238 * only in error case, so that improper users ignoring return
239 * code altogether don't copy garbage; otherwise length of string
240 * is returned that can be used for bpf_perf_event_output() et al.
242 ret = compat ? strncpy_from_unsafe(dst, unsafe_ptr, size) :
243 strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
244 if (unlikely(ret < 0))
246 memset(dst, 0, size);
250 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
251 const void *, unsafe_ptr)
253 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, false);
256 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
257 .func = bpf_probe_read_kernel_str,
259 .ret_type = RET_INTEGER,
260 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
261 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
262 .arg3_type = ARG_ANYTHING,
265 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
266 const void *, unsafe_ptr)
268 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, true);
271 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
272 .func = bpf_probe_read_compat_str,
274 .ret_type = RET_INTEGER,
275 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
276 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
277 .arg3_type = ARG_ANYTHING,
280 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
284 * Ensure we're in user context which is safe for the helper to
285 * run. This helper has no business in a kthread.
287 * access_ok() should prevent writing to non-user memory, but in
288 * some situations (nommu, temporary switch, etc) access_ok() does
289 * not provide enough validation, hence the check on KERNEL_DS.
291 * nmi_uaccess_okay() ensures the probe is not run in an interim
292 * state, when the task or mm are switched. This is specifically
293 * required to prevent the use of temporary mm.
296 if (unlikely(in_interrupt() ||
297 current->flags & (PF_KTHREAD | PF_EXITING)))
299 if (unlikely(uaccess_kernel()))
301 if (unlikely(!nmi_uaccess_okay()))
304 return probe_user_write(unsafe_ptr, src, size);
307 static const struct bpf_func_proto bpf_probe_write_user_proto = {
308 .func = bpf_probe_write_user,
310 .ret_type = RET_INTEGER,
311 .arg1_type = ARG_ANYTHING,
312 .arg2_type = ARG_PTR_TO_MEM,
313 .arg3_type = ARG_CONST_SIZE,
316 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
318 if (!capable(CAP_SYS_ADMIN))
321 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
322 current->comm, task_pid_nr(current));
324 return &bpf_probe_write_user_proto;
327 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
330 void __user *user_ptr = (__force void __user *)unsafe_ptr;
336 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
337 strncpy_from_unsafe(buf, unsafe_ptr, bufsz);
341 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
344 strncpy_from_user_nofault(buf, user_ptr, bufsz);
350 * Only limited trace_printk() conversion specifiers allowed:
351 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pks %pus %s
353 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
354 u64, arg2, u64, arg3)
356 int i, mod[3] = {}, fmt_cnt = 0;
357 char buf[64], fmt_ptype;
358 void *unsafe_ptr = NULL;
359 bool str_seen = false;
362 * bpf_check()->check_func_arg()->check_stack_boundary()
363 * guarantees that fmt points to bpf program stack,
364 * fmt_size bytes of it were initialized and fmt_size > 0
366 if (fmt[--fmt_size] != 0)
369 /* check format string for allowed specifiers */
370 for (i = 0; i < fmt_size; i++) {
371 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
380 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
385 } else if (fmt[i] == 'p') {
387 if ((fmt[i + 1] == 'k' ||
388 fmt[i + 1] == 'u') &&
390 fmt_ptype = fmt[i + 1];
395 /* disallow any further format extensions */
396 if (fmt[i + 1] != 0 &&
397 !isspace(fmt[i + 1]) &&
398 !ispunct(fmt[i + 1]))
402 } else if (fmt[i] == 's') {
407 /* allow only one '%s' per fmt string */
411 if (fmt[i + 1] != 0 &&
412 !isspace(fmt[i + 1]) &&
413 !ispunct(fmt[i + 1]))
418 unsafe_ptr = (void *)(long)arg1;
422 unsafe_ptr = (void *)(long)arg2;
426 unsafe_ptr = (void *)(long)arg3;
431 bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
441 if (fmt[i] != 'i' && fmt[i] != 'd' &&
442 fmt[i] != 'u' && fmt[i] != 'x')
448 /* Horrid workaround for getting va_list handling working with different
449 * argument type combinations generically for 32 and 64 bit archs.
451 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
452 #define __BPF_TP(...) \
453 __trace_printk(0 /* Fake ip */, \
456 #define __BPF_ARG1_TP(...) \
457 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
458 ? __BPF_TP(arg1, ##__VA_ARGS__) \
459 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
460 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
461 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
463 #define __BPF_ARG2_TP(...) \
464 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
465 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
466 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
467 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
468 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
470 #define __BPF_ARG3_TP(...) \
471 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
472 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
473 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
474 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
475 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
477 return __BPF_TP_EMIT();
480 static const struct bpf_func_proto bpf_trace_printk_proto = {
481 .func = bpf_trace_printk,
483 .ret_type = RET_INTEGER,
484 .arg1_type = ARG_PTR_TO_MEM,
485 .arg2_type = ARG_CONST_SIZE,
488 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
491 * this program might be calling bpf_trace_printk,
492 * so allocate per-cpu printk buffers
494 trace_printk_init_buffers();
496 return &bpf_trace_printk_proto;
499 #define MAX_SEQ_PRINTF_VARARGS 12
500 #define MAX_SEQ_PRINTF_MAX_MEMCPY 6
501 #define MAX_SEQ_PRINTF_STR_LEN 128
503 struct bpf_seq_printf_buf {
504 char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
506 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
507 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
509 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
510 const void *, data, u32, data_len)
512 int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
513 int i, buf_used, copy_size, num_args;
514 u64 params[MAX_SEQ_PRINTF_VARARGS];
515 struct bpf_seq_printf_buf *bufs;
516 const u64 *args = data;
518 buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
519 if (WARN_ON_ONCE(buf_used > 1)) {
524 bufs = this_cpu_ptr(&bpf_seq_printf_buf);
527 * bpf_check()->check_func_arg()->check_stack_boundary()
528 * guarantees that fmt points to bpf program stack,
529 * fmt_size bytes of it were initialized and fmt_size > 0
531 if (fmt[--fmt_size] != 0)
537 for (i = 0; i < fmt_size; i++) {
539 if (fmt[i + 1] == '%')
541 else if (!data || !data_len)
546 num_args = data_len / 8;
548 /* check format string for allowed specifiers */
549 for (i = 0; i < fmt_size; i++) {
550 /* only printable ascii for now. */
551 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
559 if (fmt[i + 1] == '%') {
564 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
569 if (fmt_cnt >= num_args) {
574 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
577 /* skip optional "[0 +-][num]" width formating field */
578 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
581 if (fmt[i] >= '1' && fmt[i] <= '9') {
583 while (fmt[i] >= '0' && fmt[i] <= '9')
588 /* try our best to copy */
589 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
594 err = strncpy_from_unsafe_strict(bufs->buf[memcpy_cnt],
595 (void *) (long) args[fmt_cnt],
596 MAX_SEQ_PRINTF_STR_LEN);
598 bufs->buf[memcpy_cnt][0] = '\0';
599 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
607 if (fmt[i + 1] == 0 ||
610 /* just kernel pointers */
611 params[fmt_cnt] = args[fmt_cnt];
616 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
617 if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
621 if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
626 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
632 copy_size = (fmt[i + 2] == '4') ? 4 : 16;
634 err = probe_kernel_read(bufs->buf[memcpy_cnt],
635 (void *) (long) args[fmt_cnt],
638 memset(bufs->buf[memcpy_cnt], 0, copy_size);
639 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
653 if (fmt[i] != 'i' && fmt[i] != 'd' &&
654 fmt[i] != 'u' && fmt[i] != 'x') {
659 params[fmt_cnt] = args[fmt_cnt];
663 /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
664 * all of them to seq_printf().
666 seq_printf(m, fmt, params[0], params[1], params[2], params[3],
667 params[4], params[5], params[6], params[7], params[8],
668 params[9], params[10], params[11]);
670 err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
672 this_cpu_dec(bpf_seq_printf_buf_used);
676 static int bpf_seq_printf_btf_ids[5];
677 static const struct bpf_func_proto bpf_seq_printf_proto = {
678 .func = bpf_seq_printf,
680 .ret_type = RET_INTEGER,
681 .arg1_type = ARG_PTR_TO_BTF_ID,
682 .arg2_type = ARG_PTR_TO_MEM,
683 .arg3_type = ARG_CONST_SIZE,
684 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
685 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
686 .btf_id = bpf_seq_printf_btf_ids,
689 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
691 return seq_write(m, data, len) ? -EOVERFLOW : 0;
694 static int bpf_seq_write_btf_ids[5];
695 static const struct bpf_func_proto bpf_seq_write_proto = {
696 .func = bpf_seq_write,
698 .ret_type = RET_INTEGER,
699 .arg1_type = ARG_PTR_TO_BTF_ID,
700 .arg2_type = ARG_PTR_TO_MEM,
701 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
702 .btf_id = bpf_seq_write_btf_ids,
705 static __always_inline int
706 get_map_perf_counter(struct bpf_map *map, u64 flags,
707 u64 *value, u64 *enabled, u64 *running)
709 struct bpf_array *array = container_of(map, struct bpf_array, map);
710 unsigned int cpu = smp_processor_id();
711 u64 index = flags & BPF_F_INDEX_MASK;
712 struct bpf_event_entry *ee;
714 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
716 if (index == BPF_F_CURRENT_CPU)
718 if (unlikely(index >= array->map.max_entries))
721 ee = READ_ONCE(array->ptrs[index]);
725 return perf_event_read_local(ee->event, value, enabled, running);
728 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
733 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
735 * this api is ugly since we miss [-22..-2] range of valid
736 * counter values, but that's uapi
743 static const struct bpf_func_proto bpf_perf_event_read_proto = {
744 .func = bpf_perf_event_read,
746 .ret_type = RET_INTEGER,
747 .arg1_type = ARG_CONST_MAP_PTR,
748 .arg2_type = ARG_ANYTHING,
751 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
752 struct bpf_perf_event_value *, buf, u32, size)
756 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
758 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
764 memset(buf, 0, size);
768 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
769 .func = bpf_perf_event_read_value,
771 .ret_type = RET_INTEGER,
772 .arg1_type = ARG_CONST_MAP_PTR,
773 .arg2_type = ARG_ANYTHING,
774 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
775 .arg4_type = ARG_CONST_SIZE,
778 static __always_inline u64
779 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
780 u64 flags, struct perf_sample_data *sd)
782 struct bpf_array *array = container_of(map, struct bpf_array, map);
783 unsigned int cpu = smp_processor_id();
784 u64 index = flags & BPF_F_INDEX_MASK;
785 struct bpf_event_entry *ee;
786 struct perf_event *event;
788 if (index == BPF_F_CURRENT_CPU)
790 if (unlikely(index >= array->map.max_entries))
793 ee = READ_ONCE(array->ptrs[index]);
798 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
799 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
802 if (unlikely(event->oncpu != cpu))
805 return perf_event_output(event, sd, regs);
809 * Support executing tracepoints in normal, irq, and nmi context that each call
810 * bpf_perf_event_output
812 struct bpf_trace_sample_data {
813 struct perf_sample_data sds[3];
816 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
817 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
818 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
819 u64, flags, void *, data, u64, size)
821 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
822 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
823 struct perf_raw_record raw = {
829 struct perf_sample_data *sd;
832 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
837 sd = &sds->sds[nest_level - 1];
839 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
844 perf_sample_data_init(sd, 0, 0);
847 err = __bpf_perf_event_output(regs, map, flags, sd);
850 this_cpu_dec(bpf_trace_nest_level);
854 static const struct bpf_func_proto bpf_perf_event_output_proto = {
855 .func = bpf_perf_event_output,
857 .ret_type = RET_INTEGER,
858 .arg1_type = ARG_PTR_TO_CTX,
859 .arg2_type = ARG_CONST_MAP_PTR,
860 .arg3_type = ARG_ANYTHING,
861 .arg4_type = ARG_PTR_TO_MEM,
862 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
865 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
866 struct bpf_nested_pt_regs {
867 struct pt_regs regs[3];
869 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
870 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
872 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
873 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
875 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
876 struct perf_raw_frag frag = {
881 struct perf_raw_record raw = {
884 .next = ctx_size ? &frag : NULL,
890 struct perf_sample_data *sd;
891 struct pt_regs *regs;
894 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
898 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
899 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
901 perf_fetch_caller_regs(regs);
902 perf_sample_data_init(sd, 0, 0);
905 ret = __bpf_perf_event_output(regs, map, flags, sd);
907 this_cpu_dec(bpf_event_output_nest_level);
911 BPF_CALL_0(bpf_get_current_task)
913 return (long) current;
916 const struct bpf_func_proto bpf_get_current_task_proto = {
917 .func = bpf_get_current_task,
919 .ret_type = RET_INTEGER,
922 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
924 struct bpf_array *array = container_of(map, struct bpf_array, map);
927 if (unlikely(idx >= array->map.max_entries))
930 cgrp = READ_ONCE(array->ptrs[idx]);
934 return task_under_cgroup_hierarchy(current, cgrp);
937 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
938 .func = bpf_current_task_under_cgroup,
940 .ret_type = RET_INTEGER,
941 .arg1_type = ARG_CONST_MAP_PTR,
942 .arg2_type = ARG_ANYTHING,
945 struct send_signal_irq_work {
946 struct irq_work irq_work;
947 struct task_struct *task;
952 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
954 static void do_bpf_send_signal(struct irq_work *entry)
956 struct send_signal_irq_work *work;
958 work = container_of(entry, struct send_signal_irq_work, irq_work);
959 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
962 static int bpf_send_signal_common(u32 sig, enum pid_type type)
964 struct send_signal_irq_work *work = NULL;
966 /* Similar to bpf_probe_write_user, task needs to be
967 * in a sound condition and kernel memory access be
968 * permitted in order to send signal to the current
971 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
973 if (unlikely(uaccess_kernel()))
975 if (unlikely(!nmi_uaccess_okay()))
978 if (irqs_disabled()) {
979 /* Do an early check on signal validity. Otherwise,
980 * the error is lost in deferred irq_work.
982 if (unlikely(!valid_signal(sig)))
985 work = this_cpu_ptr(&send_signal_work);
986 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
989 /* Add the current task, which is the target of sending signal,
990 * to the irq_work. The current task may change when queued
991 * irq works get executed.
993 work->task = current;
996 irq_work_queue(&work->irq_work);
1000 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1003 BPF_CALL_1(bpf_send_signal, u32, sig)
1005 return bpf_send_signal_common(sig, PIDTYPE_TGID);
1008 static const struct bpf_func_proto bpf_send_signal_proto = {
1009 .func = bpf_send_signal,
1011 .ret_type = RET_INTEGER,
1012 .arg1_type = ARG_ANYTHING,
1015 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1017 return bpf_send_signal_common(sig, PIDTYPE_PID);
1020 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1021 .func = bpf_send_signal_thread,
1023 .ret_type = RET_INTEGER,
1024 .arg1_type = ARG_ANYTHING,
1027 const struct bpf_func_proto *
1028 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1031 case BPF_FUNC_map_lookup_elem:
1032 return &bpf_map_lookup_elem_proto;
1033 case BPF_FUNC_map_update_elem:
1034 return &bpf_map_update_elem_proto;
1035 case BPF_FUNC_map_delete_elem:
1036 return &bpf_map_delete_elem_proto;
1037 case BPF_FUNC_map_push_elem:
1038 return &bpf_map_push_elem_proto;
1039 case BPF_FUNC_map_pop_elem:
1040 return &bpf_map_pop_elem_proto;
1041 case BPF_FUNC_map_peek_elem:
1042 return &bpf_map_peek_elem_proto;
1043 case BPF_FUNC_ktime_get_ns:
1044 return &bpf_ktime_get_ns_proto;
1045 case BPF_FUNC_ktime_get_boot_ns:
1046 return &bpf_ktime_get_boot_ns_proto;
1047 case BPF_FUNC_tail_call:
1048 return &bpf_tail_call_proto;
1049 case BPF_FUNC_get_current_pid_tgid:
1050 return &bpf_get_current_pid_tgid_proto;
1051 case BPF_FUNC_get_current_task:
1052 return &bpf_get_current_task_proto;
1053 case BPF_FUNC_get_current_uid_gid:
1054 return &bpf_get_current_uid_gid_proto;
1055 case BPF_FUNC_get_current_comm:
1056 return &bpf_get_current_comm_proto;
1057 case BPF_FUNC_trace_printk:
1058 return bpf_get_trace_printk_proto();
1059 case BPF_FUNC_get_smp_processor_id:
1060 return &bpf_get_smp_processor_id_proto;
1061 case BPF_FUNC_get_numa_node_id:
1062 return &bpf_get_numa_node_id_proto;
1063 case BPF_FUNC_perf_event_read:
1064 return &bpf_perf_event_read_proto;
1065 case BPF_FUNC_probe_write_user:
1066 return bpf_get_probe_write_proto();
1067 case BPF_FUNC_current_task_under_cgroup:
1068 return &bpf_current_task_under_cgroup_proto;
1069 case BPF_FUNC_get_prandom_u32:
1070 return &bpf_get_prandom_u32_proto;
1071 case BPF_FUNC_probe_read_user:
1072 return &bpf_probe_read_user_proto;
1073 case BPF_FUNC_probe_read_kernel:
1074 return &bpf_probe_read_kernel_proto;
1075 case BPF_FUNC_probe_read_user_str:
1076 return &bpf_probe_read_user_str_proto;
1077 case BPF_FUNC_probe_read_kernel_str:
1078 return &bpf_probe_read_kernel_str_proto;
1079 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1080 case BPF_FUNC_probe_read:
1081 return &bpf_probe_read_compat_proto;
1082 case BPF_FUNC_probe_read_str:
1083 return &bpf_probe_read_compat_str_proto;
1085 #ifdef CONFIG_CGROUPS
1086 case BPF_FUNC_get_current_cgroup_id:
1087 return &bpf_get_current_cgroup_id_proto;
1089 case BPF_FUNC_send_signal:
1090 return &bpf_send_signal_proto;
1091 case BPF_FUNC_send_signal_thread:
1092 return &bpf_send_signal_thread_proto;
1093 case BPF_FUNC_perf_event_read_value:
1094 return &bpf_perf_event_read_value_proto;
1095 case BPF_FUNC_get_ns_current_pid_tgid:
1096 return &bpf_get_ns_current_pid_tgid_proto;
1097 case BPF_FUNC_ringbuf_output:
1098 return &bpf_ringbuf_output_proto;
1099 case BPF_FUNC_ringbuf_reserve:
1100 return &bpf_ringbuf_reserve_proto;
1101 case BPF_FUNC_ringbuf_submit:
1102 return &bpf_ringbuf_submit_proto;
1103 case BPF_FUNC_ringbuf_discard:
1104 return &bpf_ringbuf_discard_proto;
1105 case BPF_FUNC_ringbuf_query:
1106 return &bpf_ringbuf_query_proto;
1112 static const struct bpf_func_proto *
1113 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1116 case BPF_FUNC_perf_event_output:
1117 return &bpf_perf_event_output_proto;
1118 case BPF_FUNC_get_stackid:
1119 return &bpf_get_stackid_proto;
1120 case BPF_FUNC_get_stack:
1121 return &bpf_get_stack_proto;
1122 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1123 case BPF_FUNC_override_return:
1124 return &bpf_override_return_proto;
1127 return bpf_tracing_func_proto(func_id, prog);
1131 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1132 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1133 const struct bpf_prog *prog,
1134 struct bpf_insn_access_aux *info)
1136 if (off < 0 || off >= sizeof(struct pt_regs))
1138 if (type != BPF_READ)
1140 if (off % size != 0)
1143 * Assertion for 32 bit to make sure last 8 byte access
1144 * (BPF_DW) to the last 4 byte member is disallowed.
1146 if (off + size > sizeof(struct pt_regs))
1152 const struct bpf_verifier_ops kprobe_verifier_ops = {
1153 .get_func_proto = kprobe_prog_func_proto,
1154 .is_valid_access = kprobe_prog_is_valid_access,
1157 const struct bpf_prog_ops kprobe_prog_ops = {
1160 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1161 u64, flags, void *, data, u64, size)
1163 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1166 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1167 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1168 * from there and call the same bpf_perf_event_output() helper inline.
1170 return ____bpf_perf_event_output(regs, map, flags, data, size);
1173 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1174 .func = bpf_perf_event_output_tp,
1176 .ret_type = RET_INTEGER,
1177 .arg1_type = ARG_PTR_TO_CTX,
1178 .arg2_type = ARG_CONST_MAP_PTR,
1179 .arg3_type = ARG_ANYTHING,
1180 .arg4_type = ARG_PTR_TO_MEM,
1181 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1184 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1187 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1190 * Same comment as in bpf_perf_event_output_tp(), only that this time
1191 * the other helper's function body cannot be inlined due to being
1192 * external, thus we need to call raw helper function.
1194 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1198 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1199 .func = bpf_get_stackid_tp,
1201 .ret_type = RET_INTEGER,
1202 .arg1_type = ARG_PTR_TO_CTX,
1203 .arg2_type = ARG_CONST_MAP_PTR,
1204 .arg3_type = ARG_ANYTHING,
1207 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1210 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1212 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1213 (unsigned long) size, flags, 0);
1216 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1217 .func = bpf_get_stack_tp,
1219 .ret_type = RET_INTEGER,
1220 .arg1_type = ARG_PTR_TO_CTX,
1221 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1222 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1223 .arg4_type = ARG_ANYTHING,
1226 static const struct bpf_func_proto *
1227 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1230 case BPF_FUNC_perf_event_output:
1231 return &bpf_perf_event_output_proto_tp;
1232 case BPF_FUNC_get_stackid:
1233 return &bpf_get_stackid_proto_tp;
1234 case BPF_FUNC_get_stack:
1235 return &bpf_get_stack_proto_tp;
1237 return bpf_tracing_func_proto(func_id, prog);
1241 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1242 const struct bpf_prog *prog,
1243 struct bpf_insn_access_aux *info)
1245 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1247 if (type != BPF_READ)
1249 if (off % size != 0)
1252 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1256 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1257 .get_func_proto = tp_prog_func_proto,
1258 .is_valid_access = tp_prog_is_valid_access,
1261 const struct bpf_prog_ops tracepoint_prog_ops = {
1264 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1265 struct bpf_perf_event_value *, buf, u32, size)
1269 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1271 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1277 memset(buf, 0, size);
1281 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1282 .func = bpf_perf_prog_read_value,
1284 .ret_type = RET_INTEGER,
1285 .arg1_type = ARG_PTR_TO_CTX,
1286 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1287 .arg3_type = ARG_CONST_SIZE,
1290 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1291 void *, buf, u32, size, u64, flags)
1296 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1297 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1300 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1303 if (unlikely(!br_stack))
1306 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1307 return br_stack->nr * br_entry_size;
1309 if (!buf || (size % br_entry_size != 0))
1312 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1313 memcpy(buf, br_stack->entries, to_copy);
1319 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1320 .func = bpf_read_branch_records,
1322 .ret_type = RET_INTEGER,
1323 .arg1_type = ARG_PTR_TO_CTX,
1324 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1325 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1326 .arg4_type = ARG_ANYTHING,
1329 static const struct bpf_func_proto *
1330 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1333 case BPF_FUNC_perf_event_output:
1334 return &bpf_perf_event_output_proto_tp;
1335 case BPF_FUNC_get_stackid:
1336 return &bpf_get_stackid_proto_tp;
1337 case BPF_FUNC_get_stack:
1338 return &bpf_get_stack_proto_tp;
1339 case BPF_FUNC_perf_prog_read_value:
1340 return &bpf_perf_prog_read_value_proto;
1341 case BPF_FUNC_read_branch_records:
1342 return &bpf_read_branch_records_proto;
1344 return bpf_tracing_func_proto(func_id, prog);
1349 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1350 * to avoid potential recursive reuse issue when/if tracepoints are added
1351 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1353 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1354 * in normal, irq, and nmi context.
1356 struct bpf_raw_tp_regs {
1357 struct pt_regs regs[3];
1359 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1360 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1361 static struct pt_regs *get_bpf_raw_tp_regs(void)
1363 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1364 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1366 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1367 this_cpu_dec(bpf_raw_tp_nest_level);
1368 return ERR_PTR(-EBUSY);
1371 return &tp_regs->regs[nest_level - 1];
1374 static void put_bpf_raw_tp_regs(void)
1376 this_cpu_dec(bpf_raw_tp_nest_level);
1379 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1380 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1382 struct pt_regs *regs = get_bpf_raw_tp_regs();
1386 return PTR_ERR(regs);
1388 perf_fetch_caller_regs(regs);
1389 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1391 put_bpf_raw_tp_regs();
1395 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1396 .func = bpf_perf_event_output_raw_tp,
1398 .ret_type = RET_INTEGER,
1399 .arg1_type = ARG_PTR_TO_CTX,
1400 .arg2_type = ARG_CONST_MAP_PTR,
1401 .arg3_type = ARG_ANYTHING,
1402 .arg4_type = ARG_PTR_TO_MEM,
1403 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1406 extern const struct bpf_func_proto bpf_skb_output_proto;
1407 extern const struct bpf_func_proto bpf_xdp_output_proto;
1409 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1410 struct bpf_map *, map, u64, flags)
1412 struct pt_regs *regs = get_bpf_raw_tp_regs();
1416 return PTR_ERR(regs);
1418 perf_fetch_caller_regs(regs);
1419 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1420 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1422 put_bpf_raw_tp_regs();
1426 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1427 .func = bpf_get_stackid_raw_tp,
1429 .ret_type = RET_INTEGER,
1430 .arg1_type = ARG_PTR_TO_CTX,
1431 .arg2_type = ARG_CONST_MAP_PTR,
1432 .arg3_type = ARG_ANYTHING,
1435 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1436 void *, buf, u32, size, u64, flags)
1438 struct pt_regs *regs = get_bpf_raw_tp_regs();
1442 return PTR_ERR(regs);
1444 perf_fetch_caller_regs(regs);
1445 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1446 (unsigned long) size, flags, 0);
1447 put_bpf_raw_tp_regs();
1451 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1452 .func = bpf_get_stack_raw_tp,
1454 .ret_type = RET_INTEGER,
1455 .arg1_type = ARG_PTR_TO_CTX,
1456 .arg2_type = ARG_PTR_TO_MEM,
1457 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1458 .arg4_type = ARG_ANYTHING,
1461 static const struct bpf_func_proto *
1462 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1465 case BPF_FUNC_perf_event_output:
1466 return &bpf_perf_event_output_proto_raw_tp;
1467 case BPF_FUNC_get_stackid:
1468 return &bpf_get_stackid_proto_raw_tp;
1469 case BPF_FUNC_get_stack:
1470 return &bpf_get_stack_proto_raw_tp;
1472 return bpf_tracing_func_proto(func_id, prog);
1476 const struct bpf_func_proto *
1477 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1481 case BPF_FUNC_skb_output:
1482 return &bpf_skb_output_proto;
1483 case BPF_FUNC_xdp_output:
1484 return &bpf_xdp_output_proto;
1486 case BPF_FUNC_seq_printf:
1487 return prog->expected_attach_type == BPF_TRACE_ITER ?
1488 &bpf_seq_printf_proto :
1490 case BPF_FUNC_seq_write:
1491 return prog->expected_attach_type == BPF_TRACE_ITER ?
1492 &bpf_seq_write_proto :
1495 return raw_tp_prog_func_proto(func_id, prog);
1499 static bool raw_tp_prog_is_valid_access(int off, int size,
1500 enum bpf_access_type type,
1501 const struct bpf_prog *prog,
1502 struct bpf_insn_access_aux *info)
1504 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1506 if (type != BPF_READ)
1508 if (off % size != 0)
1513 static bool tracing_prog_is_valid_access(int off, int size,
1514 enum bpf_access_type type,
1515 const struct bpf_prog *prog,
1516 struct bpf_insn_access_aux *info)
1518 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1520 if (type != BPF_READ)
1522 if (off % size != 0)
1524 return btf_ctx_access(off, size, type, prog, info);
1527 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1528 const union bpf_attr *kattr,
1529 union bpf_attr __user *uattr)
1534 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1535 .get_func_proto = raw_tp_prog_func_proto,
1536 .is_valid_access = raw_tp_prog_is_valid_access,
1539 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1542 const struct bpf_verifier_ops tracing_verifier_ops = {
1543 .get_func_proto = tracing_prog_func_proto,
1544 .is_valid_access = tracing_prog_is_valid_access,
1547 const struct bpf_prog_ops tracing_prog_ops = {
1548 .test_run = bpf_prog_test_run_tracing,
1551 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1552 enum bpf_access_type type,
1553 const struct bpf_prog *prog,
1554 struct bpf_insn_access_aux *info)
1557 if (size != sizeof(u64) || type != BPF_READ)
1559 info->reg_type = PTR_TO_TP_BUFFER;
1561 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1564 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1565 .get_func_proto = raw_tp_prog_func_proto,
1566 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1569 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1572 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1573 const struct bpf_prog *prog,
1574 struct bpf_insn_access_aux *info)
1576 const int size_u64 = sizeof(u64);
1578 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1580 if (type != BPF_READ)
1582 if (off % size != 0) {
1583 if (sizeof(unsigned long) != 4)
1587 if (off % size != 4)
1592 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1593 bpf_ctx_record_field_size(info, size_u64);
1594 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1597 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1598 bpf_ctx_record_field_size(info, size_u64);
1599 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1603 if (size != sizeof(long))
1610 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1611 const struct bpf_insn *si,
1612 struct bpf_insn *insn_buf,
1613 struct bpf_prog *prog, u32 *target_size)
1615 struct bpf_insn *insn = insn_buf;
1618 case offsetof(struct bpf_perf_event_data, sample_period):
1619 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1620 data), si->dst_reg, si->src_reg,
1621 offsetof(struct bpf_perf_event_data_kern, data));
1622 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1623 bpf_target_off(struct perf_sample_data, period, 8,
1626 case offsetof(struct bpf_perf_event_data, addr):
1627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1628 data), si->dst_reg, si->src_reg,
1629 offsetof(struct bpf_perf_event_data_kern, data));
1630 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1631 bpf_target_off(struct perf_sample_data, addr, 8,
1635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1636 regs), si->dst_reg, si->src_reg,
1637 offsetof(struct bpf_perf_event_data_kern, regs));
1638 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1643 return insn - insn_buf;
1646 const struct bpf_verifier_ops perf_event_verifier_ops = {
1647 .get_func_proto = pe_prog_func_proto,
1648 .is_valid_access = pe_prog_is_valid_access,
1649 .convert_ctx_access = pe_prog_convert_ctx_access,
1652 const struct bpf_prog_ops perf_event_prog_ops = {
1655 static DEFINE_MUTEX(bpf_event_mutex);
1657 #define BPF_TRACE_MAX_PROGS 64
1659 int perf_event_attach_bpf_prog(struct perf_event *event,
1660 struct bpf_prog *prog)
1662 struct bpf_prog_array *old_array;
1663 struct bpf_prog_array *new_array;
1667 * Kprobe override only works if they are on the function entry,
1668 * and only if they are on the opt-in list.
1670 if (prog->kprobe_override &&
1671 (!trace_kprobe_on_func_entry(event->tp_event) ||
1672 !trace_kprobe_error_injectable(event->tp_event)))
1675 mutex_lock(&bpf_event_mutex);
1680 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1682 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1687 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1691 /* set the new array to event->tp_event and set event->prog */
1693 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1694 bpf_prog_array_free(old_array);
1697 mutex_unlock(&bpf_event_mutex);
1701 void perf_event_detach_bpf_prog(struct perf_event *event)
1703 struct bpf_prog_array *old_array;
1704 struct bpf_prog_array *new_array;
1707 mutex_lock(&bpf_event_mutex);
1712 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1713 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1717 bpf_prog_array_delete_safe(old_array, event->prog);
1719 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1720 bpf_prog_array_free(old_array);
1723 bpf_prog_put(event->prog);
1727 mutex_unlock(&bpf_event_mutex);
1730 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1732 struct perf_event_query_bpf __user *uquery = info;
1733 struct perf_event_query_bpf query = {};
1734 struct bpf_prog_array *progs;
1735 u32 *ids, prog_cnt, ids_len;
1738 if (!perfmon_capable())
1740 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1742 if (copy_from_user(&query, uquery, sizeof(query)))
1745 ids_len = query.ids_len;
1746 if (ids_len > BPF_TRACE_MAX_PROGS)
1748 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1752 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1753 * is required when user only wants to check for uquery->prog_cnt.
1754 * There is no need to check for it since the case is handled
1755 * gracefully in bpf_prog_array_copy_info.
1758 mutex_lock(&bpf_event_mutex);
1759 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1760 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1761 mutex_unlock(&bpf_event_mutex);
1763 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1764 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1771 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1772 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1774 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1776 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1778 for (; btp < __stop__bpf_raw_tp; btp++) {
1779 if (!strcmp(btp->tp->name, name))
1783 return bpf_get_raw_tracepoint_module(name);
1786 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1788 struct module *mod = __module_address((unsigned long)btp);
1794 static __always_inline
1795 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1799 (void) BPF_PROG_RUN(prog, args);
1803 #define UNPACK(...) __VA_ARGS__
1804 #define REPEAT_1(FN, DL, X, ...) FN(X)
1805 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1806 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1807 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1808 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1809 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1810 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1811 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1812 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1813 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1814 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1815 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1816 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1818 #define SARG(X) u64 arg##X
1819 #define COPY(X) args[X] = arg##X
1821 #define __DL_COM (,)
1822 #define __DL_SEM (;)
1824 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1826 #define BPF_TRACE_DEFN_x(x) \
1827 void bpf_trace_run##x(struct bpf_prog *prog, \
1828 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1831 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1832 __bpf_trace_run(prog, args); \
1834 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1835 BPF_TRACE_DEFN_x(1);
1836 BPF_TRACE_DEFN_x(2);
1837 BPF_TRACE_DEFN_x(3);
1838 BPF_TRACE_DEFN_x(4);
1839 BPF_TRACE_DEFN_x(5);
1840 BPF_TRACE_DEFN_x(6);
1841 BPF_TRACE_DEFN_x(7);
1842 BPF_TRACE_DEFN_x(8);
1843 BPF_TRACE_DEFN_x(9);
1844 BPF_TRACE_DEFN_x(10);
1845 BPF_TRACE_DEFN_x(11);
1846 BPF_TRACE_DEFN_x(12);
1848 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1850 struct tracepoint *tp = btp->tp;
1853 * check that program doesn't access arguments beyond what's
1854 * available in this tracepoint
1856 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1859 if (prog->aux->max_tp_access > btp->writable_size)
1862 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1865 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1867 return __bpf_probe_register(btp, prog);
1870 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1872 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1875 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1876 u32 *fd_type, const char **buf,
1877 u64 *probe_offset, u64 *probe_addr)
1879 bool is_tracepoint, is_syscall_tp;
1880 struct bpf_prog *prog;
1887 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1888 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1891 *prog_id = prog->aux->id;
1892 flags = event->tp_event->flags;
1893 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1894 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1896 if (is_tracepoint || is_syscall_tp) {
1897 *buf = is_tracepoint ? event->tp_event->tp->name
1898 : event->tp_event->name;
1899 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1900 *probe_offset = 0x0;
1905 #ifdef CONFIG_KPROBE_EVENTS
1906 if (flags & TRACE_EVENT_FL_KPROBE)
1907 err = bpf_get_kprobe_info(event, fd_type, buf,
1908 probe_offset, probe_addr,
1909 event->attr.type == PERF_TYPE_TRACEPOINT);
1911 #ifdef CONFIG_UPROBE_EVENTS
1912 if (flags & TRACE_EVENT_FL_UPROBE)
1913 err = bpf_get_uprobe_info(event, fd_type, buf,
1915 event->attr.type == PERF_TYPE_TRACEPOINT);
1922 static int __init send_signal_irq_work_init(void)
1925 struct send_signal_irq_work *work;
1927 for_each_possible_cpu(cpu) {
1928 work = per_cpu_ptr(&send_signal_work, cpu);
1929 init_irq_work(&work->irq_work, do_bpf_send_signal);
1934 subsys_initcall(send_signal_irq_work_init);
1936 #ifdef CONFIG_MODULES
1937 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1940 struct bpf_trace_module *btm, *tmp;
1941 struct module *mod = module;
1943 if (mod->num_bpf_raw_events == 0 ||
1944 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1947 mutex_lock(&bpf_module_mutex);
1950 case MODULE_STATE_COMING:
1951 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1953 btm->module = module;
1954 list_add(&btm->list, &bpf_trace_modules);
1957 case MODULE_STATE_GOING:
1958 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1959 if (btm->module == module) {
1960 list_del(&btm->list);
1968 mutex_unlock(&bpf_module_mutex);
1973 static struct notifier_block bpf_module_nb = {
1974 .notifier_call = bpf_event_notify,
1977 static int __init bpf_event_init(void)
1979 register_module_notifier(&bpf_module_nb);
1983 fs_initcall(bpf_event_init);
1984 #endif /* CONFIG_MODULES */