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>
17 #include "trace_probe.h"
21 struct bpf_trace_module {
22 struct module *module;
23 struct list_head list;
26 static LIST_HEAD(bpf_trace_modules);
27 static DEFINE_MUTEX(bpf_module_mutex);
29 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
31 struct bpf_raw_event_map *btp, *ret = NULL;
32 struct bpf_trace_module *btm;
35 mutex_lock(&bpf_module_mutex);
36 list_for_each_entry(btm, &bpf_trace_modules, list) {
37 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
38 btp = &btm->module->bpf_raw_events[i];
39 if (!strcmp(btp->tp->name, name)) {
40 if (try_module_get(btm->module))
47 mutex_unlock(&bpf_module_mutex);
51 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
55 #endif /* CONFIG_MODULES */
57 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
58 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
61 * trace_call_bpf - invoke BPF program
62 * @call: tracepoint event
63 * @ctx: opaque context pointer
65 * kprobe handlers execute BPF programs via this helper.
66 * Can be used from static tracepoints in the future.
68 * Return: BPF programs always return an integer which is interpreted by
70 * 0 - return from kprobe (event is filtered out)
71 * 1 - store kprobe event into ring buffer
72 * Other values are reserved and currently alias to 1
74 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
78 if (in_nmi()) /* not supported yet */
83 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
85 * since some bpf program is already running on this cpu,
86 * don't call into another bpf program (same or different)
87 * and don't send kprobe event into ring-buffer,
95 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
96 * to all call sites, we did a bpf_prog_array_valid() there to check
97 * whether call->prog_array is empty or not, which is
98 * a heurisitc to speed up execution.
100 * If bpf_prog_array_valid() fetched prog_array was
101 * non-NULL, we go into trace_call_bpf() and do the actual
102 * proper rcu_dereference() under RCU lock.
103 * If it turns out that prog_array is NULL then, we bail out.
104 * For the opposite, if the bpf_prog_array_valid() fetched pointer
105 * was NULL, you'll skip the prog_array with the risk of missing
106 * out of events when it was updated in between this and the
107 * rcu_dereference() which is accepted risk.
109 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
112 __this_cpu_dec(bpf_prog_active);
117 EXPORT_SYMBOL_GPL(trace_call_bpf);
119 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
120 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
122 regs_set_return_value(regs, rc);
123 override_function_with_return(regs);
127 static const struct bpf_func_proto bpf_override_return_proto = {
128 .func = bpf_override_return,
130 .ret_type = RET_INTEGER,
131 .arg1_type = ARG_PTR_TO_CTX,
132 .arg2_type = ARG_ANYTHING,
136 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
140 ret = probe_kernel_read(dst, unsafe_ptr, size);
141 if (unlikely(ret < 0))
142 memset(dst, 0, size);
147 static const struct bpf_func_proto bpf_probe_read_proto = {
148 .func = bpf_probe_read,
150 .ret_type = RET_INTEGER,
151 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
152 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
153 .arg3_type = ARG_ANYTHING,
156 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
160 * Ensure we're in user context which is safe for the helper to
161 * run. This helper has no business in a kthread.
163 * access_ok() should prevent writing to non-user memory, but in
164 * some situations (nommu, temporary switch, etc) access_ok() does
165 * not provide enough validation, hence the check on KERNEL_DS.
168 if (unlikely(in_interrupt() ||
169 current->flags & (PF_KTHREAD | PF_EXITING)))
171 if (unlikely(uaccess_kernel()))
173 if (!access_ok(unsafe_ptr, size))
176 return probe_kernel_write(unsafe_ptr, src, size);
179 static const struct bpf_func_proto bpf_probe_write_user_proto = {
180 .func = bpf_probe_write_user,
182 .ret_type = RET_INTEGER,
183 .arg1_type = ARG_ANYTHING,
184 .arg2_type = ARG_PTR_TO_MEM,
185 .arg3_type = ARG_CONST_SIZE,
188 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
190 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
191 current->comm, task_pid_nr(current));
193 return &bpf_probe_write_user_proto;
197 * Only limited trace_printk() conversion specifiers allowed:
198 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
200 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
201 u64, arg2, u64, arg3)
203 bool str_seen = false;
211 * bpf_check()->check_func_arg()->check_stack_boundary()
212 * guarantees that fmt points to bpf program stack,
213 * fmt_size bytes of it were initialized and fmt_size > 0
215 if (fmt[--fmt_size] != 0)
218 /* check format string for allowed specifiers */
219 for (i = 0; i < fmt_size; i++) {
220 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
229 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
234 } else if (fmt[i] == 'p' || fmt[i] == 's') {
236 /* disallow any further format extensions */
237 if (fmt[i + 1] != 0 &&
238 !isspace(fmt[i + 1]) &&
239 !ispunct(fmt[i + 1]))
244 /* allow only one '%s' per fmt string */
263 strncpy_from_unsafe(buf,
264 (void *) (long) unsafe_addr,
275 if (fmt[i] != 'i' && fmt[i] != 'd' &&
276 fmt[i] != 'u' && fmt[i] != 'x')
281 /* Horrid workaround for getting va_list handling working with different
282 * argument type combinations generically for 32 and 64 bit archs.
284 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
285 #define __BPF_TP(...) \
286 __trace_printk(0 /* Fake ip */, \
289 #define __BPF_ARG1_TP(...) \
290 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
291 ? __BPF_TP(arg1, ##__VA_ARGS__) \
292 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
293 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
294 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
296 #define __BPF_ARG2_TP(...) \
297 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
298 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
299 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
300 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
301 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
303 #define __BPF_ARG3_TP(...) \
304 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
305 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
306 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
307 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
308 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
310 return __BPF_TP_EMIT();
313 static const struct bpf_func_proto bpf_trace_printk_proto = {
314 .func = bpf_trace_printk,
316 .ret_type = RET_INTEGER,
317 .arg1_type = ARG_PTR_TO_MEM,
318 .arg2_type = ARG_CONST_SIZE,
321 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
324 * this program might be calling bpf_trace_printk,
325 * so allocate per-cpu printk buffers
327 trace_printk_init_buffers();
329 return &bpf_trace_printk_proto;
332 static __always_inline int
333 get_map_perf_counter(struct bpf_map *map, u64 flags,
334 u64 *value, u64 *enabled, u64 *running)
336 struct bpf_array *array = container_of(map, struct bpf_array, map);
337 unsigned int cpu = smp_processor_id();
338 u64 index = flags & BPF_F_INDEX_MASK;
339 struct bpf_event_entry *ee;
341 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
343 if (index == BPF_F_CURRENT_CPU)
345 if (unlikely(index >= array->map.max_entries))
348 ee = READ_ONCE(array->ptrs[index]);
352 return perf_event_read_local(ee->event, value, enabled, running);
355 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
360 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
362 * this api is ugly since we miss [-22..-2] range of valid
363 * counter values, but that's uapi
370 static const struct bpf_func_proto bpf_perf_event_read_proto = {
371 .func = bpf_perf_event_read,
373 .ret_type = RET_INTEGER,
374 .arg1_type = ARG_CONST_MAP_PTR,
375 .arg2_type = ARG_ANYTHING,
378 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
379 struct bpf_perf_event_value *, buf, u32, size)
383 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
385 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
391 memset(buf, 0, size);
395 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
396 .func = bpf_perf_event_read_value,
398 .ret_type = RET_INTEGER,
399 .arg1_type = ARG_CONST_MAP_PTR,
400 .arg2_type = ARG_ANYTHING,
401 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
402 .arg4_type = ARG_CONST_SIZE,
405 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
407 static __always_inline u64
408 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
409 u64 flags, struct perf_sample_data *sd)
411 struct bpf_array *array = container_of(map, struct bpf_array, map);
412 unsigned int cpu = smp_processor_id();
413 u64 index = flags & BPF_F_INDEX_MASK;
414 struct bpf_event_entry *ee;
415 struct perf_event *event;
417 if (index == BPF_F_CURRENT_CPU)
419 if (unlikely(index >= array->map.max_entries))
422 ee = READ_ONCE(array->ptrs[index]);
427 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
428 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
431 if (unlikely(event->oncpu != cpu))
434 return perf_event_output(event, sd, regs);
437 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
438 u64, flags, void *, data, u64, size)
440 struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
441 struct perf_raw_record raw = {
448 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
451 perf_sample_data_init(sd, 0, 0);
454 return __bpf_perf_event_output(regs, map, flags, sd);
457 static const struct bpf_func_proto bpf_perf_event_output_proto = {
458 .func = bpf_perf_event_output,
460 .ret_type = RET_INTEGER,
461 .arg1_type = ARG_PTR_TO_CTX,
462 .arg2_type = ARG_CONST_MAP_PTR,
463 .arg3_type = ARG_ANYTHING,
464 .arg4_type = ARG_PTR_TO_MEM,
465 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
468 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
469 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
471 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
472 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
474 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
475 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
476 struct perf_raw_frag frag = {
481 struct perf_raw_record raw = {
484 .next = ctx_size ? &frag : NULL,
491 perf_fetch_caller_regs(regs);
492 perf_sample_data_init(sd, 0, 0);
495 return __bpf_perf_event_output(regs, map, flags, sd);
498 BPF_CALL_0(bpf_get_current_task)
500 return (long) current;
503 static const struct bpf_func_proto bpf_get_current_task_proto = {
504 .func = bpf_get_current_task,
506 .ret_type = RET_INTEGER,
509 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
511 struct bpf_array *array = container_of(map, struct bpf_array, map);
514 if (unlikely(idx >= array->map.max_entries))
517 cgrp = READ_ONCE(array->ptrs[idx]);
521 return task_under_cgroup_hierarchy(current, cgrp);
524 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
525 .func = bpf_current_task_under_cgroup,
527 .ret_type = RET_INTEGER,
528 .arg1_type = ARG_CONST_MAP_PTR,
529 .arg2_type = ARG_ANYTHING,
532 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
533 const void *, unsafe_ptr)
538 * The strncpy_from_unsafe() call will likely not fill the entire
539 * buffer, but that's okay in this circumstance as we're probing
540 * arbitrary memory anyway similar to bpf_probe_read() and might
541 * as well probe the stack. Thus, memory is explicitly cleared
542 * only in error case, so that improper users ignoring return
543 * code altogether don't copy garbage; otherwise length of string
544 * is returned that can be used for bpf_perf_event_output() et al.
546 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
547 if (unlikely(ret < 0))
548 memset(dst, 0, size);
553 static const struct bpf_func_proto bpf_probe_read_str_proto = {
554 .func = bpf_probe_read_str,
556 .ret_type = RET_INTEGER,
557 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
558 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
559 .arg3_type = ARG_ANYTHING,
562 static const struct bpf_func_proto *
563 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
566 case BPF_FUNC_map_lookup_elem:
567 return &bpf_map_lookup_elem_proto;
568 case BPF_FUNC_map_update_elem:
569 return &bpf_map_update_elem_proto;
570 case BPF_FUNC_map_delete_elem:
571 return &bpf_map_delete_elem_proto;
572 case BPF_FUNC_probe_read:
573 return &bpf_probe_read_proto;
574 case BPF_FUNC_ktime_get_ns:
575 return &bpf_ktime_get_ns_proto;
576 case BPF_FUNC_tail_call:
577 return &bpf_tail_call_proto;
578 case BPF_FUNC_get_current_pid_tgid:
579 return &bpf_get_current_pid_tgid_proto;
580 case BPF_FUNC_get_current_task:
581 return &bpf_get_current_task_proto;
582 case BPF_FUNC_get_current_uid_gid:
583 return &bpf_get_current_uid_gid_proto;
584 case BPF_FUNC_get_current_comm:
585 return &bpf_get_current_comm_proto;
586 case BPF_FUNC_trace_printk:
587 return bpf_get_trace_printk_proto();
588 case BPF_FUNC_get_smp_processor_id:
589 return &bpf_get_smp_processor_id_proto;
590 case BPF_FUNC_get_numa_node_id:
591 return &bpf_get_numa_node_id_proto;
592 case BPF_FUNC_perf_event_read:
593 return &bpf_perf_event_read_proto;
594 case BPF_FUNC_probe_write_user:
595 return bpf_get_probe_write_proto();
596 case BPF_FUNC_current_task_under_cgroup:
597 return &bpf_current_task_under_cgroup_proto;
598 case BPF_FUNC_get_prandom_u32:
599 return &bpf_get_prandom_u32_proto;
600 case BPF_FUNC_probe_read_str:
601 return &bpf_probe_read_str_proto;
602 #ifdef CONFIG_CGROUPS
603 case BPF_FUNC_get_current_cgroup_id:
604 return &bpf_get_current_cgroup_id_proto;
611 static const struct bpf_func_proto *
612 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
615 case BPF_FUNC_perf_event_output:
616 return &bpf_perf_event_output_proto;
617 case BPF_FUNC_get_stackid:
618 return &bpf_get_stackid_proto;
619 case BPF_FUNC_get_stack:
620 return &bpf_get_stack_proto;
621 case BPF_FUNC_perf_event_read_value:
622 return &bpf_perf_event_read_value_proto;
623 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
624 case BPF_FUNC_override_return:
625 return &bpf_override_return_proto;
628 return tracing_func_proto(func_id, prog);
632 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
633 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
634 const struct bpf_prog *prog,
635 struct bpf_insn_access_aux *info)
637 if (off < 0 || off >= sizeof(struct pt_regs))
639 if (type != BPF_READ)
644 * Assertion for 32 bit to make sure last 8 byte access
645 * (BPF_DW) to the last 4 byte member is disallowed.
647 if (off + size > sizeof(struct pt_regs))
653 const struct bpf_verifier_ops kprobe_verifier_ops = {
654 .get_func_proto = kprobe_prog_func_proto,
655 .is_valid_access = kprobe_prog_is_valid_access,
658 const struct bpf_prog_ops kprobe_prog_ops = {
661 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
662 u64, flags, void *, data, u64, size)
664 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
667 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
668 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
669 * from there and call the same bpf_perf_event_output() helper inline.
671 return ____bpf_perf_event_output(regs, map, flags, data, size);
674 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
675 .func = bpf_perf_event_output_tp,
677 .ret_type = RET_INTEGER,
678 .arg1_type = ARG_PTR_TO_CTX,
679 .arg2_type = ARG_CONST_MAP_PTR,
680 .arg3_type = ARG_ANYTHING,
681 .arg4_type = ARG_PTR_TO_MEM,
682 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
685 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
688 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
691 * Same comment as in bpf_perf_event_output_tp(), only that this time
692 * the other helper's function body cannot be inlined due to being
693 * external, thus we need to call raw helper function.
695 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
699 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
700 .func = bpf_get_stackid_tp,
702 .ret_type = RET_INTEGER,
703 .arg1_type = ARG_PTR_TO_CTX,
704 .arg2_type = ARG_CONST_MAP_PTR,
705 .arg3_type = ARG_ANYTHING,
708 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
711 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
713 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
714 (unsigned long) size, flags, 0);
717 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
718 .func = bpf_get_stack_tp,
720 .ret_type = RET_INTEGER,
721 .arg1_type = ARG_PTR_TO_CTX,
722 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
723 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
724 .arg4_type = ARG_ANYTHING,
727 static const struct bpf_func_proto *
728 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
731 case BPF_FUNC_perf_event_output:
732 return &bpf_perf_event_output_proto_tp;
733 case BPF_FUNC_get_stackid:
734 return &bpf_get_stackid_proto_tp;
735 case BPF_FUNC_get_stack:
736 return &bpf_get_stack_proto_tp;
738 return tracing_func_proto(func_id, prog);
742 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
743 const struct bpf_prog *prog,
744 struct bpf_insn_access_aux *info)
746 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
748 if (type != BPF_READ)
753 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
757 const struct bpf_verifier_ops tracepoint_verifier_ops = {
758 .get_func_proto = tp_prog_func_proto,
759 .is_valid_access = tp_prog_is_valid_access,
762 const struct bpf_prog_ops tracepoint_prog_ops = {
765 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
766 struct bpf_perf_event_value *, buf, u32, size)
770 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
772 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
778 memset(buf, 0, size);
782 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
783 .func = bpf_perf_prog_read_value,
785 .ret_type = RET_INTEGER,
786 .arg1_type = ARG_PTR_TO_CTX,
787 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
788 .arg3_type = ARG_CONST_SIZE,
791 static const struct bpf_func_proto *
792 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
795 case BPF_FUNC_perf_event_output:
796 return &bpf_perf_event_output_proto_tp;
797 case BPF_FUNC_get_stackid:
798 return &bpf_get_stackid_proto_tp;
799 case BPF_FUNC_get_stack:
800 return &bpf_get_stack_proto_tp;
801 case BPF_FUNC_perf_prog_read_value:
802 return &bpf_perf_prog_read_value_proto;
804 return tracing_func_proto(func_id, prog);
809 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
810 * to avoid potential recursive reuse issue when/if tracepoints are added
811 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
813 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
814 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
815 struct bpf_map *, map, u64, flags, void *, data, u64, size)
817 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
819 perf_fetch_caller_regs(regs);
820 return ____bpf_perf_event_output(regs, map, flags, data, size);
823 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
824 .func = bpf_perf_event_output_raw_tp,
826 .ret_type = RET_INTEGER,
827 .arg1_type = ARG_PTR_TO_CTX,
828 .arg2_type = ARG_CONST_MAP_PTR,
829 .arg3_type = ARG_ANYTHING,
830 .arg4_type = ARG_PTR_TO_MEM,
831 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
834 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
835 struct bpf_map *, map, u64, flags)
837 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
839 perf_fetch_caller_regs(regs);
840 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
841 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
845 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
846 .func = bpf_get_stackid_raw_tp,
848 .ret_type = RET_INTEGER,
849 .arg1_type = ARG_PTR_TO_CTX,
850 .arg2_type = ARG_CONST_MAP_PTR,
851 .arg3_type = ARG_ANYTHING,
854 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
855 void *, buf, u32, size, u64, flags)
857 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
859 perf_fetch_caller_regs(regs);
860 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
861 (unsigned long) size, flags, 0);
864 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
865 .func = bpf_get_stack_raw_tp,
867 .ret_type = RET_INTEGER,
868 .arg1_type = ARG_PTR_TO_CTX,
869 .arg2_type = ARG_PTR_TO_MEM,
870 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
871 .arg4_type = ARG_ANYTHING,
874 static const struct bpf_func_proto *
875 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
878 case BPF_FUNC_perf_event_output:
879 return &bpf_perf_event_output_proto_raw_tp;
880 case BPF_FUNC_get_stackid:
881 return &bpf_get_stackid_proto_raw_tp;
882 case BPF_FUNC_get_stack:
883 return &bpf_get_stack_proto_raw_tp;
885 return tracing_func_proto(func_id, prog);
889 static bool raw_tp_prog_is_valid_access(int off, int size,
890 enum bpf_access_type type,
891 const struct bpf_prog *prog,
892 struct bpf_insn_access_aux *info)
894 /* largest tracepoint in the kernel has 12 args */
895 if (off < 0 || off >= sizeof(__u64) * 12)
897 if (type != BPF_READ)
904 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
905 .get_func_proto = raw_tp_prog_func_proto,
906 .is_valid_access = raw_tp_prog_is_valid_access,
909 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
912 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
913 const struct bpf_prog *prog,
914 struct bpf_insn_access_aux *info)
916 const int size_u64 = sizeof(u64);
918 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
920 if (type != BPF_READ)
922 if (off % size != 0) {
923 if (sizeof(unsigned long) != 4)
932 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
933 bpf_ctx_record_field_size(info, size_u64);
934 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
937 case bpf_ctx_range(struct bpf_perf_event_data, addr):
938 bpf_ctx_record_field_size(info, size_u64);
939 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
943 if (size != sizeof(long))
950 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
951 const struct bpf_insn *si,
952 struct bpf_insn *insn_buf,
953 struct bpf_prog *prog, u32 *target_size)
955 struct bpf_insn *insn = insn_buf;
958 case offsetof(struct bpf_perf_event_data, sample_period):
959 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
960 data), si->dst_reg, si->src_reg,
961 offsetof(struct bpf_perf_event_data_kern, data));
962 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
963 bpf_target_off(struct perf_sample_data, period, 8,
966 case offsetof(struct bpf_perf_event_data, addr):
967 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
968 data), si->dst_reg, si->src_reg,
969 offsetof(struct bpf_perf_event_data_kern, data));
970 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
971 bpf_target_off(struct perf_sample_data, addr, 8,
975 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
976 regs), si->dst_reg, si->src_reg,
977 offsetof(struct bpf_perf_event_data_kern, regs));
978 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
983 return insn - insn_buf;
986 const struct bpf_verifier_ops perf_event_verifier_ops = {
987 .get_func_proto = pe_prog_func_proto,
988 .is_valid_access = pe_prog_is_valid_access,
989 .convert_ctx_access = pe_prog_convert_ctx_access,
992 const struct bpf_prog_ops perf_event_prog_ops = {
995 static DEFINE_MUTEX(bpf_event_mutex);
997 #define BPF_TRACE_MAX_PROGS 64
999 int perf_event_attach_bpf_prog(struct perf_event *event,
1000 struct bpf_prog *prog)
1002 struct bpf_prog_array __rcu *old_array;
1003 struct bpf_prog_array *new_array;
1007 * Kprobe override only works if they are on the function entry,
1008 * and only if they are on the opt-in list.
1010 if (prog->kprobe_override &&
1011 (!trace_kprobe_on_func_entry(event->tp_event) ||
1012 !trace_kprobe_error_injectable(event->tp_event)))
1015 mutex_lock(&bpf_event_mutex);
1020 old_array = event->tp_event->prog_array;
1022 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1027 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1031 /* set the new array to event->tp_event and set event->prog */
1033 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1034 bpf_prog_array_free(old_array);
1037 mutex_unlock(&bpf_event_mutex);
1041 void perf_event_detach_bpf_prog(struct perf_event *event)
1043 struct bpf_prog_array __rcu *old_array;
1044 struct bpf_prog_array *new_array;
1047 mutex_lock(&bpf_event_mutex);
1052 old_array = event->tp_event->prog_array;
1053 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1057 bpf_prog_array_delete_safe(old_array, event->prog);
1059 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1060 bpf_prog_array_free(old_array);
1063 bpf_prog_put(event->prog);
1067 mutex_unlock(&bpf_event_mutex);
1070 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1072 struct perf_event_query_bpf __user *uquery = info;
1073 struct perf_event_query_bpf query = {};
1074 u32 *ids, prog_cnt, ids_len;
1077 if (!capable(CAP_SYS_ADMIN))
1079 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1081 if (copy_from_user(&query, uquery, sizeof(query)))
1084 ids_len = query.ids_len;
1085 if (ids_len > BPF_TRACE_MAX_PROGS)
1087 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1091 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1092 * is required when user only wants to check for uquery->prog_cnt.
1093 * There is no need to check for it since the case is handled
1094 * gracefully in bpf_prog_array_copy_info.
1097 mutex_lock(&bpf_event_mutex);
1098 ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1102 mutex_unlock(&bpf_event_mutex);
1104 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1105 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1112 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1113 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1115 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1117 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1119 for (; btp < __stop__bpf_raw_tp; btp++) {
1120 if (!strcmp(btp->tp->name, name))
1124 return bpf_get_raw_tracepoint_module(name);
1127 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1129 struct module *mod = __module_address((unsigned long)btp);
1135 static __always_inline
1136 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1140 (void) BPF_PROG_RUN(prog, args);
1145 #define UNPACK(...) __VA_ARGS__
1146 #define REPEAT_1(FN, DL, X, ...) FN(X)
1147 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1148 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1149 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1150 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1151 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1152 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1153 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1154 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1155 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1156 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1157 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1158 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1160 #define SARG(X) u64 arg##X
1161 #define COPY(X) args[X] = arg##X
1163 #define __DL_COM (,)
1164 #define __DL_SEM (;)
1166 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1168 #define BPF_TRACE_DEFN_x(x) \
1169 void bpf_trace_run##x(struct bpf_prog *prog, \
1170 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1173 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1174 __bpf_trace_run(prog, args); \
1176 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1177 BPF_TRACE_DEFN_x(1);
1178 BPF_TRACE_DEFN_x(2);
1179 BPF_TRACE_DEFN_x(3);
1180 BPF_TRACE_DEFN_x(4);
1181 BPF_TRACE_DEFN_x(5);
1182 BPF_TRACE_DEFN_x(6);
1183 BPF_TRACE_DEFN_x(7);
1184 BPF_TRACE_DEFN_x(8);
1185 BPF_TRACE_DEFN_x(9);
1186 BPF_TRACE_DEFN_x(10);
1187 BPF_TRACE_DEFN_x(11);
1188 BPF_TRACE_DEFN_x(12);
1190 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1192 struct tracepoint *tp = btp->tp;
1195 * check that program doesn't access arguments beyond what's
1196 * available in this tracepoint
1198 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1201 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1204 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1206 return __bpf_probe_register(btp, prog);
1209 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1211 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1214 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1215 u32 *fd_type, const char **buf,
1216 u64 *probe_offset, u64 *probe_addr)
1218 bool is_tracepoint, is_syscall_tp;
1219 struct bpf_prog *prog;
1226 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1227 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1230 *prog_id = prog->aux->id;
1231 flags = event->tp_event->flags;
1232 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1233 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1235 if (is_tracepoint || is_syscall_tp) {
1236 *buf = is_tracepoint ? event->tp_event->tp->name
1237 : event->tp_event->name;
1238 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1239 *probe_offset = 0x0;
1244 #ifdef CONFIG_KPROBE_EVENTS
1245 if (flags & TRACE_EVENT_FL_KPROBE)
1246 err = bpf_get_kprobe_info(event, fd_type, buf,
1247 probe_offset, probe_addr,
1248 event->attr.type == PERF_TYPE_TRACEPOINT);
1250 #ifdef CONFIG_UPROBE_EVENTS
1251 if (flags & TRACE_EVENT_FL_UPROBE)
1252 err = bpf_get_uprobe_info(event, fd_type, buf,
1254 event->attr.type == PERF_TYPE_TRACEPOINT);
1261 #ifdef CONFIG_MODULES
1262 int bpf_event_notify(struct notifier_block *nb, unsigned long op, void *module)
1264 struct bpf_trace_module *btm, *tmp;
1265 struct module *mod = module;
1267 if (mod->num_bpf_raw_events == 0 ||
1268 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1271 mutex_lock(&bpf_module_mutex);
1274 case MODULE_STATE_COMING:
1275 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1277 btm->module = module;
1278 list_add(&btm->list, &bpf_trace_modules);
1281 case MODULE_STATE_GOING:
1282 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1283 if (btm->module == module) {
1284 list_del(&btm->list);
1292 mutex_unlock(&bpf_module_mutex);
1297 static struct notifier_block bpf_module_nb = {
1298 .notifier_call = bpf_event_notify,
1301 int __init bpf_event_init(void)
1303 register_module_notifier(&bpf_module_nb);
1307 fs_initcall(bpf_event_init);
1308 #endif /* CONFIG_MODULES */