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 static __always_inline int
140 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
144 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
145 if (unlikely(ret < 0))
146 memset(dst, 0, size);
150 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
151 const void __user *, unsafe_ptr)
153 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
156 const struct bpf_func_proto bpf_probe_read_user_proto = {
157 .func = bpf_probe_read_user,
159 .ret_type = RET_INTEGER,
160 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
161 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
162 .arg3_type = ARG_ANYTHING,
165 static __always_inline int
166 bpf_probe_read_user_str_common(void *dst, u32 size,
167 const void __user *unsafe_ptr)
171 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
172 if (unlikely(ret < 0))
173 memset(dst, 0, size);
177 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
178 const void __user *, unsafe_ptr)
180 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
183 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
184 .func = bpf_probe_read_user_str,
186 .ret_type = RET_INTEGER,
187 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
188 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
189 .arg3_type = ARG_ANYTHING,
192 static __always_inline int
193 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
195 int ret = security_locked_down(LOCKDOWN_BPF_READ);
197 if (unlikely(ret < 0))
199 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
200 if (unlikely(ret < 0))
204 memset(dst, 0, size);
208 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
209 const void *, unsafe_ptr)
211 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
214 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
215 .func = bpf_probe_read_kernel,
217 .ret_type = RET_INTEGER,
218 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
219 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
220 .arg3_type = ARG_ANYTHING,
223 static __always_inline int
224 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
226 int ret = security_locked_down(LOCKDOWN_BPF_READ);
228 if (unlikely(ret < 0))
232 * The strncpy_from_kernel_nofault() call will likely not fill the
233 * entire buffer, but that's okay in this circumstance as we're probing
234 * arbitrary memory anyway similar to bpf_probe_read_*() and might
235 * as well probe the stack. Thus, memory is explicitly cleared
236 * only in error case, so that improper users ignoring return
237 * code altogether don't copy garbage; otherwise length of string
238 * is returned that can be used for bpf_perf_event_output() et al.
240 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
241 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);
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 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
266 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
267 const void *, unsafe_ptr)
269 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
270 return bpf_probe_read_user_common(dst, size,
271 (__force void __user *)unsafe_ptr);
273 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
276 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
277 .func = bpf_probe_read_compat,
279 .ret_type = RET_INTEGER,
280 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
281 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
282 .arg3_type = ARG_ANYTHING,
285 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
286 const void *, unsafe_ptr)
288 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
289 return bpf_probe_read_user_str_common(dst, size,
290 (__force void __user *)unsafe_ptr);
292 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
295 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
296 .func = bpf_probe_read_compat_str,
298 .ret_type = RET_INTEGER,
299 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
300 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
301 .arg3_type = ARG_ANYTHING,
303 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
305 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
309 * Ensure we're in user context which is safe for the helper to
310 * run. This helper has no business in a kthread.
312 * access_ok() should prevent writing to non-user memory, but in
313 * some situations (nommu, temporary switch, etc) access_ok() does
314 * not provide enough validation, hence the check on KERNEL_DS.
316 * nmi_uaccess_okay() ensures the probe is not run in an interim
317 * state, when the task or mm are switched. This is specifically
318 * required to prevent the use of temporary mm.
321 if (unlikely(in_interrupt() ||
322 current->flags & (PF_KTHREAD | PF_EXITING)))
324 if (unlikely(uaccess_kernel()))
326 if (unlikely(!nmi_uaccess_okay()))
329 return copy_to_user_nofault(unsafe_ptr, src, size);
332 static const struct bpf_func_proto bpf_probe_write_user_proto = {
333 .func = bpf_probe_write_user,
335 .ret_type = RET_INTEGER,
336 .arg1_type = ARG_ANYTHING,
337 .arg2_type = ARG_PTR_TO_MEM,
338 .arg3_type = ARG_CONST_SIZE,
341 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
343 if (!capable(CAP_SYS_ADMIN))
346 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
347 current->comm, task_pid_nr(current));
349 return &bpf_probe_write_user_proto;
352 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
355 void __user *user_ptr = (__force void __user *)unsafe_ptr;
361 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
362 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
363 strncpy_from_user_nofault(buf, user_ptr, bufsz);
369 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
372 strncpy_from_user_nofault(buf, user_ptr, bufsz);
378 * Only limited trace_printk() conversion specifiers allowed:
379 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pks %pus %s
381 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
382 u64, arg2, u64, arg3)
384 int i, mod[3] = {}, fmt_cnt = 0;
385 char buf[64], fmt_ptype;
386 void *unsafe_ptr = NULL;
387 bool str_seen = false;
390 * bpf_check()->check_func_arg()->check_stack_boundary()
391 * guarantees that fmt points to bpf program stack,
392 * fmt_size bytes of it were initialized and fmt_size > 0
394 if (fmt[--fmt_size] != 0)
397 /* check format string for allowed specifiers */
398 for (i = 0; i < fmt_size; i++) {
399 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
408 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
413 } else if (fmt[i] == 'p') {
415 if ((fmt[i + 1] == 'k' ||
416 fmt[i + 1] == 'u') &&
418 fmt_ptype = fmt[i + 1];
423 /* disallow any further format extensions */
424 if (fmt[i + 1] != 0 &&
425 !isspace(fmt[i + 1]) &&
426 !ispunct(fmt[i + 1]))
430 } else if (fmt[i] == 's') {
435 /* allow only one '%s' per fmt string */
439 if (fmt[i + 1] != 0 &&
440 !isspace(fmt[i + 1]) &&
441 !ispunct(fmt[i + 1]))
446 unsafe_ptr = (void *)(long)arg1;
450 unsafe_ptr = (void *)(long)arg2;
454 unsafe_ptr = (void *)(long)arg3;
459 bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
469 if (fmt[i] != 'i' && fmt[i] != 'd' &&
470 fmt[i] != 'u' && fmt[i] != 'x')
476 /* Horrid workaround for getting va_list handling working with different
477 * argument type combinations generically for 32 and 64 bit archs.
479 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
480 #define __BPF_TP(...) \
481 __trace_printk(0 /* Fake ip */, \
484 #define __BPF_ARG1_TP(...) \
485 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
486 ? __BPF_TP(arg1, ##__VA_ARGS__) \
487 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
488 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
489 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
491 #define __BPF_ARG2_TP(...) \
492 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
493 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
494 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
495 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
496 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
498 #define __BPF_ARG3_TP(...) \
499 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
500 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
501 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
502 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
503 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
505 return __BPF_TP_EMIT();
508 static const struct bpf_func_proto bpf_trace_printk_proto = {
509 .func = bpf_trace_printk,
511 .ret_type = RET_INTEGER,
512 .arg1_type = ARG_PTR_TO_MEM,
513 .arg2_type = ARG_CONST_SIZE,
516 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
519 * this program might be calling bpf_trace_printk,
520 * so allocate per-cpu printk buffers
522 trace_printk_init_buffers();
524 return &bpf_trace_printk_proto;
527 #define MAX_SEQ_PRINTF_VARARGS 12
528 #define MAX_SEQ_PRINTF_MAX_MEMCPY 6
529 #define MAX_SEQ_PRINTF_STR_LEN 128
531 struct bpf_seq_printf_buf {
532 char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
534 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
535 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
537 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
538 const void *, data, u32, data_len)
540 int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
541 int i, buf_used, copy_size, num_args;
542 u64 params[MAX_SEQ_PRINTF_VARARGS];
543 struct bpf_seq_printf_buf *bufs;
544 const u64 *args = data;
546 buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
547 if (WARN_ON_ONCE(buf_used > 1)) {
552 bufs = this_cpu_ptr(&bpf_seq_printf_buf);
555 * bpf_check()->check_func_arg()->check_stack_boundary()
556 * guarantees that fmt points to bpf program stack,
557 * fmt_size bytes of it were initialized and fmt_size > 0
559 if (fmt[--fmt_size] != 0)
565 for (i = 0; i < fmt_size; i++) {
567 if (fmt[i + 1] == '%')
569 else if (!data || !data_len)
574 num_args = data_len / 8;
576 /* check format string for allowed specifiers */
577 for (i = 0; i < fmt_size; i++) {
578 /* only printable ascii for now. */
579 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
587 if (fmt[i + 1] == '%') {
592 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
597 if (fmt_cnt >= num_args) {
602 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
605 /* skip optional "[0 +-][num]" width formating field */
606 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
609 if (fmt[i] >= '1' && fmt[i] <= '9') {
611 while (fmt[i] >= '0' && fmt[i] <= '9')
618 /* try our best to copy */
619 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
624 unsafe_ptr = (void *)(long)args[fmt_cnt];
625 err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
626 unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
628 bufs->buf[memcpy_cnt][0] = '\0';
629 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
637 if (fmt[i + 1] == 0 ||
640 /* just kernel pointers */
641 params[fmt_cnt] = args[fmt_cnt];
646 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
647 if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
651 if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
656 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
662 copy_size = (fmt[i + 2] == '4') ? 4 : 16;
664 err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
665 (void *) (long) args[fmt_cnt],
668 memset(bufs->buf[memcpy_cnt], 0, copy_size);
669 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
683 if (fmt[i] != 'i' && fmt[i] != 'd' &&
684 fmt[i] != 'u' && fmt[i] != 'x') {
689 params[fmt_cnt] = args[fmt_cnt];
693 /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
694 * all of them to seq_printf().
696 seq_printf(m, fmt, params[0], params[1], params[2], params[3],
697 params[4], params[5], params[6], params[7], params[8],
698 params[9], params[10], params[11]);
700 err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
702 this_cpu_dec(bpf_seq_printf_buf_used);
706 static int bpf_seq_printf_btf_ids[5];
707 static const struct bpf_func_proto bpf_seq_printf_proto = {
708 .func = bpf_seq_printf,
710 .ret_type = RET_INTEGER,
711 .arg1_type = ARG_PTR_TO_BTF_ID,
712 .arg2_type = ARG_PTR_TO_MEM,
713 .arg3_type = ARG_CONST_SIZE,
714 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
715 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
716 .btf_id = bpf_seq_printf_btf_ids,
719 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
721 return seq_write(m, data, len) ? -EOVERFLOW : 0;
724 static int bpf_seq_write_btf_ids[5];
725 static const struct bpf_func_proto bpf_seq_write_proto = {
726 .func = bpf_seq_write,
728 .ret_type = RET_INTEGER,
729 .arg1_type = ARG_PTR_TO_BTF_ID,
730 .arg2_type = ARG_PTR_TO_MEM,
731 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
732 .btf_id = bpf_seq_write_btf_ids,
735 static __always_inline int
736 get_map_perf_counter(struct bpf_map *map, u64 flags,
737 u64 *value, u64 *enabled, u64 *running)
739 struct bpf_array *array = container_of(map, struct bpf_array, map);
740 unsigned int cpu = smp_processor_id();
741 u64 index = flags & BPF_F_INDEX_MASK;
742 struct bpf_event_entry *ee;
744 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
746 if (index == BPF_F_CURRENT_CPU)
748 if (unlikely(index >= array->map.max_entries))
751 ee = READ_ONCE(array->ptrs[index]);
755 return perf_event_read_local(ee->event, value, enabled, running);
758 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
763 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
765 * this api is ugly since we miss [-22..-2] range of valid
766 * counter values, but that's uapi
773 static const struct bpf_func_proto bpf_perf_event_read_proto = {
774 .func = bpf_perf_event_read,
776 .ret_type = RET_INTEGER,
777 .arg1_type = ARG_CONST_MAP_PTR,
778 .arg2_type = ARG_ANYTHING,
781 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
782 struct bpf_perf_event_value *, buf, u32, size)
786 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
788 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
794 memset(buf, 0, size);
798 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
799 .func = bpf_perf_event_read_value,
801 .ret_type = RET_INTEGER,
802 .arg1_type = ARG_CONST_MAP_PTR,
803 .arg2_type = ARG_ANYTHING,
804 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
805 .arg4_type = ARG_CONST_SIZE,
808 static __always_inline u64
809 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
810 u64 flags, struct perf_sample_data *sd)
812 struct bpf_array *array = container_of(map, struct bpf_array, map);
813 unsigned int cpu = smp_processor_id();
814 u64 index = flags & BPF_F_INDEX_MASK;
815 struct bpf_event_entry *ee;
816 struct perf_event *event;
818 if (index == BPF_F_CURRENT_CPU)
820 if (unlikely(index >= array->map.max_entries))
823 ee = READ_ONCE(array->ptrs[index]);
828 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
829 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
832 if (unlikely(event->oncpu != cpu))
835 return perf_event_output(event, sd, regs);
839 * Support executing tracepoints in normal, irq, and nmi context that each call
840 * bpf_perf_event_output
842 struct bpf_trace_sample_data {
843 struct perf_sample_data sds[3];
846 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
847 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
848 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
849 u64, flags, void *, data, u64, size)
851 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
852 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
853 struct perf_raw_record raw = {
859 struct perf_sample_data *sd;
862 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
867 sd = &sds->sds[nest_level - 1];
869 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
874 perf_sample_data_init(sd, 0, 0);
877 err = __bpf_perf_event_output(regs, map, flags, sd);
880 this_cpu_dec(bpf_trace_nest_level);
884 static const struct bpf_func_proto bpf_perf_event_output_proto = {
885 .func = bpf_perf_event_output,
887 .ret_type = RET_INTEGER,
888 .arg1_type = ARG_PTR_TO_CTX,
889 .arg2_type = ARG_CONST_MAP_PTR,
890 .arg3_type = ARG_ANYTHING,
891 .arg4_type = ARG_PTR_TO_MEM,
892 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
895 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
896 struct bpf_nested_pt_regs {
897 struct pt_regs regs[3];
899 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
900 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
902 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
903 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
905 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
906 struct perf_raw_frag frag = {
911 struct perf_raw_record raw = {
914 .next = ctx_size ? &frag : NULL,
920 struct perf_sample_data *sd;
921 struct pt_regs *regs;
924 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
928 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
929 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
931 perf_fetch_caller_regs(regs);
932 perf_sample_data_init(sd, 0, 0);
935 ret = __bpf_perf_event_output(regs, map, flags, sd);
937 this_cpu_dec(bpf_event_output_nest_level);
941 BPF_CALL_0(bpf_get_current_task)
943 return (long) current;
946 const struct bpf_func_proto bpf_get_current_task_proto = {
947 .func = bpf_get_current_task,
949 .ret_type = RET_INTEGER,
952 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
954 struct bpf_array *array = container_of(map, struct bpf_array, map);
957 if (unlikely(idx >= array->map.max_entries))
960 cgrp = READ_ONCE(array->ptrs[idx]);
964 return task_under_cgroup_hierarchy(current, cgrp);
967 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
968 .func = bpf_current_task_under_cgroup,
970 .ret_type = RET_INTEGER,
971 .arg1_type = ARG_CONST_MAP_PTR,
972 .arg2_type = ARG_ANYTHING,
975 struct send_signal_irq_work {
976 struct irq_work irq_work;
977 struct task_struct *task;
982 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
984 static void do_bpf_send_signal(struct irq_work *entry)
986 struct send_signal_irq_work *work;
988 work = container_of(entry, struct send_signal_irq_work, irq_work);
989 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
992 static int bpf_send_signal_common(u32 sig, enum pid_type type)
994 struct send_signal_irq_work *work = NULL;
996 /* Similar to bpf_probe_write_user, task needs to be
997 * in a sound condition and kernel memory access be
998 * permitted in order to send signal to the current
1001 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1003 if (unlikely(uaccess_kernel()))
1005 if (unlikely(!nmi_uaccess_okay()))
1008 if (irqs_disabled()) {
1009 /* Do an early check on signal validity. Otherwise,
1010 * the error is lost in deferred irq_work.
1012 if (unlikely(!valid_signal(sig)))
1015 work = this_cpu_ptr(&send_signal_work);
1016 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
1019 /* Add the current task, which is the target of sending signal,
1020 * to the irq_work. The current task may change when queued
1021 * irq works get executed.
1023 work->task = current;
1026 irq_work_queue(&work->irq_work);
1030 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1033 BPF_CALL_1(bpf_send_signal, u32, sig)
1035 return bpf_send_signal_common(sig, PIDTYPE_TGID);
1038 static const struct bpf_func_proto bpf_send_signal_proto = {
1039 .func = bpf_send_signal,
1041 .ret_type = RET_INTEGER,
1042 .arg1_type = ARG_ANYTHING,
1045 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1047 return bpf_send_signal_common(sig, PIDTYPE_PID);
1050 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1051 .func = bpf_send_signal_thread,
1053 .ret_type = RET_INTEGER,
1054 .arg1_type = ARG_ANYTHING,
1057 const struct bpf_func_proto *
1058 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1061 case BPF_FUNC_map_lookup_elem:
1062 return &bpf_map_lookup_elem_proto;
1063 case BPF_FUNC_map_update_elem:
1064 return &bpf_map_update_elem_proto;
1065 case BPF_FUNC_map_delete_elem:
1066 return &bpf_map_delete_elem_proto;
1067 case BPF_FUNC_map_push_elem:
1068 return &bpf_map_push_elem_proto;
1069 case BPF_FUNC_map_pop_elem:
1070 return &bpf_map_pop_elem_proto;
1071 case BPF_FUNC_map_peek_elem:
1072 return &bpf_map_peek_elem_proto;
1073 case BPF_FUNC_ktime_get_ns:
1074 return &bpf_ktime_get_ns_proto;
1075 case BPF_FUNC_ktime_get_boot_ns:
1076 return &bpf_ktime_get_boot_ns_proto;
1077 case BPF_FUNC_tail_call:
1078 return &bpf_tail_call_proto;
1079 case BPF_FUNC_get_current_pid_tgid:
1080 return &bpf_get_current_pid_tgid_proto;
1081 case BPF_FUNC_get_current_task:
1082 return &bpf_get_current_task_proto;
1083 case BPF_FUNC_get_current_uid_gid:
1084 return &bpf_get_current_uid_gid_proto;
1085 case BPF_FUNC_get_current_comm:
1086 return &bpf_get_current_comm_proto;
1087 case BPF_FUNC_trace_printk:
1088 return bpf_get_trace_printk_proto();
1089 case BPF_FUNC_get_smp_processor_id:
1090 return &bpf_get_smp_processor_id_proto;
1091 case BPF_FUNC_get_numa_node_id:
1092 return &bpf_get_numa_node_id_proto;
1093 case BPF_FUNC_perf_event_read:
1094 return &bpf_perf_event_read_proto;
1095 case BPF_FUNC_probe_write_user:
1096 return bpf_get_probe_write_proto();
1097 case BPF_FUNC_current_task_under_cgroup:
1098 return &bpf_current_task_under_cgroup_proto;
1099 case BPF_FUNC_get_prandom_u32:
1100 return &bpf_get_prandom_u32_proto;
1101 case BPF_FUNC_probe_read_user:
1102 return &bpf_probe_read_user_proto;
1103 case BPF_FUNC_probe_read_kernel:
1104 return &bpf_probe_read_kernel_proto;
1105 case BPF_FUNC_probe_read_user_str:
1106 return &bpf_probe_read_user_str_proto;
1107 case BPF_FUNC_probe_read_kernel_str:
1108 return &bpf_probe_read_kernel_str_proto;
1109 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1110 case BPF_FUNC_probe_read:
1111 return &bpf_probe_read_compat_proto;
1112 case BPF_FUNC_probe_read_str:
1113 return &bpf_probe_read_compat_str_proto;
1115 #ifdef CONFIG_CGROUPS
1116 case BPF_FUNC_get_current_cgroup_id:
1117 return &bpf_get_current_cgroup_id_proto;
1119 case BPF_FUNC_send_signal:
1120 return &bpf_send_signal_proto;
1121 case BPF_FUNC_send_signal_thread:
1122 return &bpf_send_signal_thread_proto;
1123 case BPF_FUNC_perf_event_read_value:
1124 return &bpf_perf_event_read_value_proto;
1125 case BPF_FUNC_get_ns_current_pid_tgid:
1126 return &bpf_get_ns_current_pid_tgid_proto;
1127 case BPF_FUNC_ringbuf_output:
1128 return &bpf_ringbuf_output_proto;
1129 case BPF_FUNC_ringbuf_reserve:
1130 return &bpf_ringbuf_reserve_proto;
1131 case BPF_FUNC_ringbuf_submit:
1132 return &bpf_ringbuf_submit_proto;
1133 case BPF_FUNC_ringbuf_discard:
1134 return &bpf_ringbuf_discard_proto;
1135 case BPF_FUNC_ringbuf_query:
1136 return &bpf_ringbuf_query_proto;
1142 static const struct bpf_func_proto *
1143 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1146 case BPF_FUNC_perf_event_output:
1147 return &bpf_perf_event_output_proto;
1148 case BPF_FUNC_get_stackid:
1149 return &bpf_get_stackid_proto;
1150 case BPF_FUNC_get_stack:
1151 return &bpf_get_stack_proto;
1152 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1153 case BPF_FUNC_override_return:
1154 return &bpf_override_return_proto;
1157 return bpf_tracing_func_proto(func_id, prog);
1161 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1162 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1163 const struct bpf_prog *prog,
1164 struct bpf_insn_access_aux *info)
1166 if (off < 0 || off >= sizeof(struct pt_regs))
1168 if (type != BPF_READ)
1170 if (off % size != 0)
1173 * Assertion for 32 bit to make sure last 8 byte access
1174 * (BPF_DW) to the last 4 byte member is disallowed.
1176 if (off + size > sizeof(struct pt_regs))
1182 const struct bpf_verifier_ops kprobe_verifier_ops = {
1183 .get_func_proto = kprobe_prog_func_proto,
1184 .is_valid_access = kprobe_prog_is_valid_access,
1187 const struct bpf_prog_ops kprobe_prog_ops = {
1190 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1191 u64, flags, void *, data, u64, size)
1193 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1196 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1197 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1198 * from there and call the same bpf_perf_event_output() helper inline.
1200 return ____bpf_perf_event_output(regs, map, flags, data, size);
1203 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1204 .func = bpf_perf_event_output_tp,
1206 .ret_type = RET_INTEGER,
1207 .arg1_type = ARG_PTR_TO_CTX,
1208 .arg2_type = ARG_CONST_MAP_PTR,
1209 .arg3_type = ARG_ANYTHING,
1210 .arg4_type = ARG_PTR_TO_MEM,
1211 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1214 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1217 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1220 * Same comment as in bpf_perf_event_output_tp(), only that this time
1221 * the other helper's function body cannot be inlined due to being
1222 * external, thus we need to call raw helper function.
1224 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1228 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1229 .func = bpf_get_stackid_tp,
1231 .ret_type = RET_INTEGER,
1232 .arg1_type = ARG_PTR_TO_CTX,
1233 .arg2_type = ARG_CONST_MAP_PTR,
1234 .arg3_type = ARG_ANYTHING,
1237 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1240 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1242 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1243 (unsigned long) size, flags, 0);
1246 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1247 .func = bpf_get_stack_tp,
1249 .ret_type = RET_INTEGER,
1250 .arg1_type = ARG_PTR_TO_CTX,
1251 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1252 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1253 .arg4_type = ARG_ANYTHING,
1256 static const struct bpf_func_proto *
1257 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1260 case BPF_FUNC_perf_event_output:
1261 return &bpf_perf_event_output_proto_tp;
1262 case BPF_FUNC_get_stackid:
1263 return &bpf_get_stackid_proto_tp;
1264 case BPF_FUNC_get_stack:
1265 return &bpf_get_stack_proto_tp;
1267 return bpf_tracing_func_proto(func_id, prog);
1271 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1272 const struct bpf_prog *prog,
1273 struct bpf_insn_access_aux *info)
1275 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1277 if (type != BPF_READ)
1279 if (off % size != 0)
1282 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1286 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1287 .get_func_proto = tp_prog_func_proto,
1288 .is_valid_access = tp_prog_is_valid_access,
1291 const struct bpf_prog_ops tracepoint_prog_ops = {
1294 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1295 struct bpf_perf_event_value *, buf, u32, size)
1299 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1301 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1307 memset(buf, 0, size);
1311 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1312 .func = bpf_perf_prog_read_value,
1314 .ret_type = RET_INTEGER,
1315 .arg1_type = ARG_PTR_TO_CTX,
1316 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1317 .arg3_type = ARG_CONST_SIZE,
1320 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1321 void *, buf, u32, size, u64, flags)
1326 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1327 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1330 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1333 if (unlikely(!br_stack))
1336 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1337 return br_stack->nr * br_entry_size;
1339 if (!buf || (size % br_entry_size != 0))
1342 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1343 memcpy(buf, br_stack->entries, to_copy);
1349 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1350 .func = bpf_read_branch_records,
1352 .ret_type = RET_INTEGER,
1353 .arg1_type = ARG_PTR_TO_CTX,
1354 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1355 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1356 .arg4_type = ARG_ANYTHING,
1359 static const struct bpf_func_proto *
1360 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1363 case BPF_FUNC_perf_event_output:
1364 return &bpf_perf_event_output_proto_tp;
1365 case BPF_FUNC_get_stackid:
1366 return &bpf_get_stackid_proto_tp;
1367 case BPF_FUNC_get_stack:
1368 return &bpf_get_stack_proto_tp;
1369 case BPF_FUNC_perf_prog_read_value:
1370 return &bpf_perf_prog_read_value_proto;
1371 case BPF_FUNC_read_branch_records:
1372 return &bpf_read_branch_records_proto;
1374 return bpf_tracing_func_proto(func_id, prog);
1379 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1380 * to avoid potential recursive reuse issue when/if tracepoints are added
1381 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1383 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1384 * in normal, irq, and nmi context.
1386 struct bpf_raw_tp_regs {
1387 struct pt_regs regs[3];
1389 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1390 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1391 static struct pt_regs *get_bpf_raw_tp_regs(void)
1393 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1394 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1396 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1397 this_cpu_dec(bpf_raw_tp_nest_level);
1398 return ERR_PTR(-EBUSY);
1401 return &tp_regs->regs[nest_level - 1];
1404 static void put_bpf_raw_tp_regs(void)
1406 this_cpu_dec(bpf_raw_tp_nest_level);
1409 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1410 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1412 struct pt_regs *regs = get_bpf_raw_tp_regs();
1416 return PTR_ERR(regs);
1418 perf_fetch_caller_regs(regs);
1419 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1421 put_bpf_raw_tp_regs();
1425 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1426 .func = bpf_perf_event_output_raw_tp,
1428 .ret_type = RET_INTEGER,
1429 .arg1_type = ARG_PTR_TO_CTX,
1430 .arg2_type = ARG_CONST_MAP_PTR,
1431 .arg3_type = ARG_ANYTHING,
1432 .arg4_type = ARG_PTR_TO_MEM,
1433 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1436 extern const struct bpf_func_proto bpf_skb_output_proto;
1437 extern const struct bpf_func_proto bpf_xdp_output_proto;
1439 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1440 struct bpf_map *, map, u64, flags)
1442 struct pt_regs *regs = get_bpf_raw_tp_regs();
1446 return PTR_ERR(regs);
1448 perf_fetch_caller_regs(regs);
1449 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1450 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1452 put_bpf_raw_tp_regs();
1456 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1457 .func = bpf_get_stackid_raw_tp,
1459 .ret_type = RET_INTEGER,
1460 .arg1_type = ARG_PTR_TO_CTX,
1461 .arg2_type = ARG_CONST_MAP_PTR,
1462 .arg3_type = ARG_ANYTHING,
1465 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1466 void *, buf, u32, size, u64, flags)
1468 struct pt_regs *regs = get_bpf_raw_tp_regs();
1472 return PTR_ERR(regs);
1474 perf_fetch_caller_regs(regs);
1475 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1476 (unsigned long) size, flags, 0);
1477 put_bpf_raw_tp_regs();
1481 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1482 .func = bpf_get_stack_raw_tp,
1484 .ret_type = RET_INTEGER,
1485 .arg1_type = ARG_PTR_TO_CTX,
1486 .arg2_type = ARG_PTR_TO_MEM,
1487 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1488 .arg4_type = ARG_ANYTHING,
1491 static const struct bpf_func_proto *
1492 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1495 case BPF_FUNC_perf_event_output:
1496 return &bpf_perf_event_output_proto_raw_tp;
1497 case BPF_FUNC_get_stackid:
1498 return &bpf_get_stackid_proto_raw_tp;
1499 case BPF_FUNC_get_stack:
1500 return &bpf_get_stack_proto_raw_tp;
1502 return bpf_tracing_func_proto(func_id, prog);
1506 const struct bpf_func_proto *
1507 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1511 case BPF_FUNC_skb_output:
1512 return &bpf_skb_output_proto;
1513 case BPF_FUNC_xdp_output:
1514 return &bpf_xdp_output_proto;
1516 case BPF_FUNC_seq_printf:
1517 return prog->expected_attach_type == BPF_TRACE_ITER ?
1518 &bpf_seq_printf_proto :
1520 case BPF_FUNC_seq_write:
1521 return prog->expected_attach_type == BPF_TRACE_ITER ?
1522 &bpf_seq_write_proto :
1525 return raw_tp_prog_func_proto(func_id, prog);
1529 static bool raw_tp_prog_is_valid_access(int off, int size,
1530 enum bpf_access_type type,
1531 const struct bpf_prog *prog,
1532 struct bpf_insn_access_aux *info)
1534 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1536 if (type != BPF_READ)
1538 if (off % size != 0)
1543 static bool tracing_prog_is_valid_access(int off, int size,
1544 enum bpf_access_type type,
1545 const struct bpf_prog *prog,
1546 struct bpf_insn_access_aux *info)
1548 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1550 if (type != BPF_READ)
1552 if (off % size != 0)
1554 return btf_ctx_access(off, size, type, prog, info);
1557 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1558 const union bpf_attr *kattr,
1559 union bpf_attr __user *uattr)
1564 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1565 .get_func_proto = raw_tp_prog_func_proto,
1566 .is_valid_access = raw_tp_prog_is_valid_access,
1569 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1572 const struct bpf_verifier_ops tracing_verifier_ops = {
1573 .get_func_proto = tracing_prog_func_proto,
1574 .is_valid_access = tracing_prog_is_valid_access,
1577 const struct bpf_prog_ops tracing_prog_ops = {
1578 .test_run = bpf_prog_test_run_tracing,
1581 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1582 enum bpf_access_type type,
1583 const struct bpf_prog *prog,
1584 struct bpf_insn_access_aux *info)
1587 if (size != sizeof(u64) || type != BPF_READ)
1589 info->reg_type = PTR_TO_TP_BUFFER;
1591 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1594 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1595 .get_func_proto = raw_tp_prog_func_proto,
1596 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1599 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1602 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1603 const struct bpf_prog *prog,
1604 struct bpf_insn_access_aux *info)
1606 const int size_u64 = sizeof(u64);
1608 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1610 if (type != BPF_READ)
1612 if (off % size != 0) {
1613 if (sizeof(unsigned long) != 4)
1617 if (off % size != 4)
1622 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1623 bpf_ctx_record_field_size(info, size_u64);
1624 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1627 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1628 bpf_ctx_record_field_size(info, size_u64);
1629 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1633 if (size != sizeof(long))
1640 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1641 const struct bpf_insn *si,
1642 struct bpf_insn *insn_buf,
1643 struct bpf_prog *prog, u32 *target_size)
1645 struct bpf_insn *insn = insn_buf;
1648 case offsetof(struct bpf_perf_event_data, sample_period):
1649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1650 data), si->dst_reg, si->src_reg,
1651 offsetof(struct bpf_perf_event_data_kern, data));
1652 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1653 bpf_target_off(struct perf_sample_data, period, 8,
1656 case offsetof(struct bpf_perf_event_data, addr):
1657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1658 data), si->dst_reg, si->src_reg,
1659 offsetof(struct bpf_perf_event_data_kern, data));
1660 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1661 bpf_target_off(struct perf_sample_data, addr, 8,
1665 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1666 regs), si->dst_reg, si->src_reg,
1667 offsetof(struct bpf_perf_event_data_kern, regs));
1668 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1673 return insn - insn_buf;
1676 const struct bpf_verifier_ops perf_event_verifier_ops = {
1677 .get_func_proto = pe_prog_func_proto,
1678 .is_valid_access = pe_prog_is_valid_access,
1679 .convert_ctx_access = pe_prog_convert_ctx_access,
1682 const struct bpf_prog_ops perf_event_prog_ops = {
1685 static DEFINE_MUTEX(bpf_event_mutex);
1687 #define BPF_TRACE_MAX_PROGS 64
1689 int perf_event_attach_bpf_prog(struct perf_event *event,
1690 struct bpf_prog *prog)
1692 struct bpf_prog_array *old_array;
1693 struct bpf_prog_array *new_array;
1697 * Kprobe override only works if they are on the function entry,
1698 * and only if they are on the opt-in list.
1700 if (prog->kprobe_override &&
1701 (!trace_kprobe_on_func_entry(event->tp_event) ||
1702 !trace_kprobe_error_injectable(event->tp_event)))
1705 mutex_lock(&bpf_event_mutex);
1710 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1712 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1717 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1721 /* set the new array to event->tp_event and set event->prog */
1723 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1724 bpf_prog_array_free(old_array);
1727 mutex_unlock(&bpf_event_mutex);
1731 void perf_event_detach_bpf_prog(struct perf_event *event)
1733 struct bpf_prog_array *old_array;
1734 struct bpf_prog_array *new_array;
1737 mutex_lock(&bpf_event_mutex);
1742 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1743 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1747 bpf_prog_array_delete_safe(old_array, event->prog);
1749 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1750 bpf_prog_array_free(old_array);
1753 bpf_prog_put(event->prog);
1757 mutex_unlock(&bpf_event_mutex);
1760 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1762 struct perf_event_query_bpf __user *uquery = info;
1763 struct perf_event_query_bpf query = {};
1764 struct bpf_prog_array *progs;
1765 u32 *ids, prog_cnt, ids_len;
1768 if (!perfmon_capable())
1770 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1772 if (copy_from_user(&query, uquery, sizeof(query)))
1775 ids_len = query.ids_len;
1776 if (ids_len > BPF_TRACE_MAX_PROGS)
1778 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1782 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1783 * is required when user only wants to check for uquery->prog_cnt.
1784 * There is no need to check for it since the case is handled
1785 * gracefully in bpf_prog_array_copy_info.
1788 mutex_lock(&bpf_event_mutex);
1789 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1790 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1791 mutex_unlock(&bpf_event_mutex);
1793 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1794 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1801 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1802 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1804 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1806 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1808 for (; btp < __stop__bpf_raw_tp; btp++) {
1809 if (!strcmp(btp->tp->name, name))
1813 return bpf_get_raw_tracepoint_module(name);
1816 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1818 struct module *mod = __module_address((unsigned long)btp);
1824 static __always_inline
1825 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1829 (void) BPF_PROG_RUN(prog, args);
1833 #define UNPACK(...) __VA_ARGS__
1834 #define REPEAT_1(FN, DL, X, ...) FN(X)
1835 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1836 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1837 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1838 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1839 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1840 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1841 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1842 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1843 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1844 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1845 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1846 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1848 #define SARG(X) u64 arg##X
1849 #define COPY(X) args[X] = arg##X
1851 #define __DL_COM (,)
1852 #define __DL_SEM (;)
1854 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1856 #define BPF_TRACE_DEFN_x(x) \
1857 void bpf_trace_run##x(struct bpf_prog *prog, \
1858 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1861 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1862 __bpf_trace_run(prog, args); \
1864 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1865 BPF_TRACE_DEFN_x(1);
1866 BPF_TRACE_DEFN_x(2);
1867 BPF_TRACE_DEFN_x(3);
1868 BPF_TRACE_DEFN_x(4);
1869 BPF_TRACE_DEFN_x(5);
1870 BPF_TRACE_DEFN_x(6);
1871 BPF_TRACE_DEFN_x(7);
1872 BPF_TRACE_DEFN_x(8);
1873 BPF_TRACE_DEFN_x(9);
1874 BPF_TRACE_DEFN_x(10);
1875 BPF_TRACE_DEFN_x(11);
1876 BPF_TRACE_DEFN_x(12);
1878 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1880 struct tracepoint *tp = btp->tp;
1883 * check that program doesn't access arguments beyond what's
1884 * available in this tracepoint
1886 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1889 if (prog->aux->max_tp_access > btp->writable_size)
1892 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1895 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1897 return __bpf_probe_register(btp, prog);
1900 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1902 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1905 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1906 u32 *fd_type, const char **buf,
1907 u64 *probe_offset, u64 *probe_addr)
1909 bool is_tracepoint, is_syscall_tp;
1910 struct bpf_prog *prog;
1917 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1918 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1921 *prog_id = prog->aux->id;
1922 flags = event->tp_event->flags;
1923 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1924 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1926 if (is_tracepoint || is_syscall_tp) {
1927 *buf = is_tracepoint ? event->tp_event->tp->name
1928 : event->tp_event->name;
1929 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1930 *probe_offset = 0x0;
1935 #ifdef CONFIG_KPROBE_EVENTS
1936 if (flags & TRACE_EVENT_FL_KPROBE)
1937 err = bpf_get_kprobe_info(event, fd_type, buf,
1938 probe_offset, probe_addr,
1939 event->attr.type == PERF_TYPE_TRACEPOINT);
1941 #ifdef CONFIG_UPROBE_EVENTS
1942 if (flags & TRACE_EVENT_FL_UPROBE)
1943 err = bpf_get_uprobe_info(event, fd_type, buf,
1945 event->attr.type == PERF_TYPE_TRACEPOINT);
1952 static int __init send_signal_irq_work_init(void)
1955 struct send_signal_irq_work *work;
1957 for_each_possible_cpu(cpu) {
1958 work = per_cpu_ptr(&send_signal_work, cpu);
1959 init_irq_work(&work->irq_work, do_bpf_send_signal);
1964 subsys_initcall(send_signal_irq_work_init);
1966 #ifdef CONFIG_MODULES
1967 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1970 struct bpf_trace_module *btm, *tmp;
1971 struct module *mod = module;
1973 if (mod->num_bpf_raw_events == 0 ||
1974 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1977 mutex_lock(&bpf_module_mutex);
1980 case MODULE_STATE_COMING:
1981 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1983 btm->module = module;
1984 list_add(&btm->list, &bpf_trace_modules);
1987 case MODULE_STATE_GOING:
1988 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1989 if (btm->module == module) {
1990 list_del(&btm->list);
1998 mutex_unlock(&bpf_module_mutex);
2003 static struct notifier_block bpf_module_nb = {
2004 .notifier_call = bpf_event_notify,
2007 static int __init bpf_event_init(void)
2009 register_module_notifier(&bpf_module_nb);
2013 fs_initcall(bpf_event_init);
2014 #endif /* CONFIG_MODULES */