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"
23 struct bpf_trace_module {
24 struct module *module;
25 struct list_head list;
28 static LIST_HEAD(bpf_trace_modules);
29 static DEFINE_MUTEX(bpf_module_mutex);
31 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
33 struct bpf_raw_event_map *btp, *ret = NULL;
34 struct bpf_trace_module *btm;
37 mutex_lock(&bpf_module_mutex);
38 list_for_each_entry(btm, &bpf_trace_modules, list) {
39 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
40 btp = &btm->module->bpf_raw_events[i];
41 if (!strcmp(btp->tp->name, name)) {
42 if (try_module_get(btm->module))
49 mutex_unlock(&bpf_module_mutex);
53 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
57 #endif /* CONFIG_MODULES */
59 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
60 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
63 * trace_call_bpf - invoke BPF program
64 * @call: tracepoint event
65 * @ctx: opaque context pointer
67 * kprobe handlers execute BPF programs via this helper.
68 * Can be used from static tracepoints in the future.
70 * Return: BPF programs always return an integer which is interpreted by
72 * 0 - return from kprobe (event is filtered out)
73 * 1 - store kprobe event into ring buffer
74 * Other values are reserved and currently alias to 1
76 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
80 if (in_nmi()) /* not supported yet */
85 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
87 * since some bpf program is already running on this cpu,
88 * don't call into another bpf program (same or different)
89 * and don't send kprobe event into ring-buffer,
97 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
98 * to all call sites, we did a bpf_prog_array_valid() there to check
99 * whether call->prog_array is empty or not, which is
100 * a heurisitc to speed up execution.
102 * If bpf_prog_array_valid() fetched prog_array was
103 * non-NULL, we go into trace_call_bpf() and do the actual
104 * proper rcu_dereference() under RCU lock.
105 * If it turns out that prog_array is NULL then, we bail out.
106 * For the opposite, if the bpf_prog_array_valid() fetched pointer
107 * was NULL, you'll skip the prog_array with the risk of missing
108 * out of events when it was updated in between this and the
109 * rcu_dereference() which is accepted risk.
111 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
114 __this_cpu_dec(bpf_prog_active);
119 EXPORT_SYMBOL_GPL(trace_call_bpf);
121 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
122 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
124 regs_set_return_value(regs, rc);
125 override_function_with_return(regs);
129 static const struct bpf_func_proto bpf_override_return_proto = {
130 .func = bpf_override_return,
132 .ret_type = RET_INTEGER,
133 .arg1_type = ARG_PTR_TO_CTX,
134 .arg2_type = ARG_ANYTHING,
138 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
142 ret = probe_kernel_read(dst, unsafe_ptr, size);
143 if (unlikely(ret < 0))
144 memset(dst, 0, size);
149 static const struct bpf_func_proto bpf_probe_read_proto = {
150 .func = bpf_probe_read,
152 .ret_type = RET_INTEGER,
153 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
154 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
155 .arg3_type = ARG_ANYTHING,
158 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
162 * Ensure we're in user context which is safe for the helper to
163 * run. This helper has no business in a kthread.
165 * access_ok() should prevent writing to non-user memory, but in
166 * some situations (nommu, temporary switch, etc) access_ok() does
167 * not provide enough validation, hence the check on KERNEL_DS.
169 * nmi_uaccess_okay() ensures the probe is not run in an interim
170 * state, when the task or mm are switched. This is specifically
171 * required to prevent the use of temporary mm.
174 if (unlikely(in_interrupt() ||
175 current->flags & (PF_KTHREAD | PF_EXITING)))
177 if (unlikely(uaccess_kernel()))
179 if (unlikely(!nmi_uaccess_okay()))
181 if (!access_ok(unsafe_ptr, size))
184 return probe_kernel_write(unsafe_ptr, src, size);
187 static const struct bpf_func_proto bpf_probe_write_user_proto = {
188 .func = bpf_probe_write_user,
190 .ret_type = RET_INTEGER,
191 .arg1_type = ARG_ANYTHING,
192 .arg2_type = ARG_PTR_TO_MEM,
193 .arg3_type = ARG_CONST_SIZE,
196 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
198 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
199 current->comm, task_pid_nr(current));
201 return &bpf_probe_write_user_proto;
205 * Only limited trace_printk() conversion specifiers allowed:
206 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
208 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
209 u64, arg2, u64, arg3)
211 bool str_seen = false;
219 * bpf_check()->check_func_arg()->check_stack_boundary()
220 * guarantees that fmt points to bpf program stack,
221 * fmt_size bytes of it were initialized and fmt_size > 0
223 if (fmt[--fmt_size] != 0)
226 /* check format string for allowed specifiers */
227 for (i = 0; i < fmt_size; i++) {
228 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
237 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
242 } else if (fmt[i] == 'p' || fmt[i] == 's') {
244 /* disallow any further format extensions */
245 if (fmt[i + 1] != 0 &&
246 !isspace(fmt[i + 1]) &&
247 !ispunct(fmt[i + 1]))
252 /* allow only one '%s' per fmt string */
271 strncpy_from_unsafe(buf,
272 (void *) (long) unsafe_addr,
283 if (fmt[i] != 'i' && fmt[i] != 'd' &&
284 fmt[i] != 'u' && fmt[i] != 'x')
289 /* Horrid workaround for getting va_list handling working with different
290 * argument type combinations generically for 32 and 64 bit archs.
292 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
293 #define __BPF_TP(...) \
294 __trace_printk(0 /* Fake ip */, \
297 #define __BPF_ARG1_TP(...) \
298 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
299 ? __BPF_TP(arg1, ##__VA_ARGS__) \
300 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
301 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
302 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
304 #define __BPF_ARG2_TP(...) \
305 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
306 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
307 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
308 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
309 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
311 #define __BPF_ARG3_TP(...) \
312 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
313 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
314 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
315 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
316 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
318 return __BPF_TP_EMIT();
321 static const struct bpf_func_proto bpf_trace_printk_proto = {
322 .func = bpf_trace_printk,
324 .ret_type = RET_INTEGER,
325 .arg1_type = ARG_PTR_TO_MEM,
326 .arg2_type = ARG_CONST_SIZE,
329 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
332 * this program might be calling bpf_trace_printk,
333 * so allocate per-cpu printk buffers
335 trace_printk_init_buffers();
337 return &bpf_trace_printk_proto;
340 static __always_inline int
341 get_map_perf_counter(struct bpf_map *map, u64 flags,
342 u64 *value, u64 *enabled, u64 *running)
344 struct bpf_array *array = container_of(map, struct bpf_array, map);
345 unsigned int cpu = smp_processor_id();
346 u64 index = flags & BPF_F_INDEX_MASK;
347 struct bpf_event_entry *ee;
349 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
351 if (index == BPF_F_CURRENT_CPU)
353 if (unlikely(index >= array->map.max_entries))
356 ee = READ_ONCE(array->ptrs[index]);
360 return perf_event_read_local(ee->event, value, enabled, running);
363 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
368 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
370 * this api is ugly since we miss [-22..-2] range of valid
371 * counter values, but that's uapi
378 static const struct bpf_func_proto bpf_perf_event_read_proto = {
379 .func = bpf_perf_event_read,
381 .ret_type = RET_INTEGER,
382 .arg1_type = ARG_CONST_MAP_PTR,
383 .arg2_type = ARG_ANYTHING,
386 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
387 struct bpf_perf_event_value *, buf, u32, size)
391 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
393 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
399 memset(buf, 0, size);
403 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
404 .func = bpf_perf_event_read_value,
406 .ret_type = RET_INTEGER,
407 .arg1_type = ARG_CONST_MAP_PTR,
408 .arg2_type = ARG_ANYTHING,
409 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
410 .arg4_type = ARG_CONST_SIZE,
413 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
415 static __always_inline u64
416 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
417 u64 flags, struct perf_sample_data *sd)
419 struct bpf_array *array = container_of(map, struct bpf_array, map);
420 unsigned int cpu = smp_processor_id();
421 u64 index = flags & BPF_F_INDEX_MASK;
422 struct bpf_event_entry *ee;
423 struct perf_event *event;
425 if (index == BPF_F_CURRENT_CPU)
427 if (unlikely(index >= array->map.max_entries))
430 ee = READ_ONCE(array->ptrs[index]);
435 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
436 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
439 if (unlikely(event->oncpu != cpu))
442 return perf_event_output(event, sd, regs);
445 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
446 u64, flags, void *, data, u64, size)
448 struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
449 struct perf_raw_record raw = {
456 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
459 perf_sample_data_init(sd, 0, 0);
462 return __bpf_perf_event_output(regs, map, flags, sd);
465 static const struct bpf_func_proto bpf_perf_event_output_proto = {
466 .func = bpf_perf_event_output,
468 .ret_type = RET_INTEGER,
469 .arg1_type = ARG_PTR_TO_CTX,
470 .arg2_type = ARG_CONST_MAP_PTR,
471 .arg3_type = ARG_ANYTHING,
472 .arg4_type = ARG_PTR_TO_MEM,
473 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
476 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
477 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
479 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
480 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
482 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
483 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
484 struct perf_raw_frag frag = {
489 struct perf_raw_record raw = {
492 .next = ctx_size ? &frag : NULL,
499 perf_fetch_caller_regs(regs);
500 perf_sample_data_init(sd, 0, 0);
503 return __bpf_perf_event_output(regs, map, flags, sd);
506 BPF_CALL_0(bpf_get_current_task)
508 return (long) current;
511 static const struct bpf_func_proto bpf_get_current_task_proto = {
512 .func = bpf_get_current_task,
514 .ret_type = RET_INTEGER,
517 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
519 struct bpf_array *array = container_of(map, struct bpf_array, map);
522 if (unlikely(idx >= array->map.max_entries))
525 cgrp = READ_ONCE(array->ptrs[idx]);
529 return task_under_cgroup_hierarchy(current, cgrp);
532 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
533 .func = bpf_current_task_under_cgroup,
535 .ret_type = RET_INTEGER,
536 .arg1_type = ARG_CONST_MAP_PTR,
537 .arg2_type = ARG_ANYTHING,
540 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
541 const void *, unsafe_ptr)
546 * The strncpy_from_unsafe() call will likely not fill the entire
547 * buffer, but that's okay in this circumstance as we're probing
548 * arbitrary memory anyway similar to bpf_probe_read() and might
549 * as well probe the stack. Thus, memory is explicitly cleared
550 * only in error case, so that improper users ignoring return
551 * code altogether don't copy garbage; otherwise length of string
552 * is returned that can be used for bpf_perf_event_output() et al.
554 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
555 if (unlikely(ret < 0))
556 memset(dst, 0, size);
561 static const struct bpf_func_proto bpf_probe_read_str_proto = {
562 .func = bpf_probe_read_str,
564 .ret_type = RET_INTEGER,
565 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
566 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
567 .arg3_type = ARG_ANYTHING,
570 static const struct bpf_func_proto *
571 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
574 case BPF_FUNC_map_lookup_elem:
575 return &bpf_map_lookup_elem_proto;
576 case BPF_FUNC_map_update_elem:
577 return &bpf_map_update_elem_proto;
578 case BPF_FUNC_map_delete_elem:
579 return &bpf_map_delete_elem_proto;
580 case BPF_FUNC_map_push_elem:
581 return &bpf_map_push_elem_proto;
582 case BPF_FUNC_map_pop_elem:
583 return &bpf_map_pop_elem_proto;
584 case BPF_FUNC_map_peek_elem:
585 return &bpf_map_peek_elem_proto;
586 case BPF_FUNC_probe_read:
587 return &bpf_probe_read_proto;
588 case BPF_FUNC_ktime_get_ns:
589 return &bpf_ktime_get_ns_proto;
590 case BPF_FUNC_tail_call:
591 return &bpf_tail_call_proto;
592 case BPF_FUNC_get_current_pid_tgid:
593 return &bpf_get_current_pid_tgid_proto;
594 case BPF_FUNC_get_current_task:
595 return &bpf_get_current_task_proto;
596 case BPF_FUNC_get_current_uid_gid:
597 return &bpf_get_current_uid_gid_proto;
598 case BPF_FUNC_get_current_comm:
599 return &bpf_get_current_comm_proto;
600 case BPF_FUNC_trace_printk:
601 return bpf_get_trace_printk_proto();
602 case BPF_FUNC_get_smp_processor_id:
603 return &bpf_get_smp_processor_id_proto;
604 case BPF_FUNC_get_numa_node_id:
605 return &bpf_get_numa_node_id_proto;
606 case BPF_FUNC_perf_event_read:
607 return &bpf_perf_event_read_proto;
608 case BPF_FUNC_probe_write_user:
609 return bpf_get_probe_write_proto();
610 case BPF_FUNC_current_task_under_cgroup:
611 return &bpf_current_task_under_cgroup_proto;
612 case BPF_FUNC_get_prandom_u32:
613 return &bpf_get_prandom_u32_proto;
614 case BPF_FUNC_probe_read_str:
615 return &bpf_probe_read_str_proto;
616 #ifdef CONFIG_CGROUPS
617 case BPF_FUNC_get_current_cgroup_id:
618 return &bpf_get_current_cgroup_id_proto;
625 static const struct bpf_func_proto *
626 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
629 case BPF_FUNC_perf_event_output:
630 return &bpf_perf_event_output_proto;
631 case BPF_FUNC_get_stackid:
632 return &bpf_get_stackid_proto;
633 case BPF_FUNC_get_stack:
634 return &bpf_get_stack_proto;
635 case BPF_FUNC_perf_event_read_value:
636 return &bpf_perf_event_read_value_proto;
637 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
638 case BPF_FUNC_override_return:
639 return &bpf_override_return_proto;
642 return tracing_func_proto(func_id, prog);
646 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
647 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
648 const struct bpf_prog *prog,
649 struct bpf_insn_access_aux *info)
651 if (off < 0 || off >= sizeof(struct pt_regs))
653 if (type != BPF_READ)
658 * Assertion for 32 bit to make sure last 8 byte access
659 * (BPF_DW) to the last 4 byte member is disallowed.
661 if (off + size > sizeof(struct pt_regs))
667 const struct bpf_verifier_ops kprobe_verifier_ops = {
668 .get_func_proto = kprobe_prog_func_proto,
669 .is_valid_access = kprobe_prog_is_valid_access,
672 const struct bpf_prog_ops kprobe_prog_ops = {
675 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
676 u64, flags, void *, data, u64, size)
678 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
681 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
682 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
683 * from there and call the same bpf_perf_event_output() helper inline.
685 return ____bpf_perf_event_output(regs, map, flags, data, size);
688 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
689 .func = bpf_perf_event_output_tp,
691 .ret_type = RET_INTEGER,
692 .arg1_type = ARG_PTR_TO_CTX,
693 .arg2_type = ARG_CONST_MAP_PTR,
694 .arg3_type = ARG_ANYTHING,
695 .arg4_type = ARG_PTR_TO_MEM,
696 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
699 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
702 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
705 * Same comment as in bpf_perf_event_output_tp(), only that this time
706 * the other helper's function body cannot be inlined due to being
707 * external, thus we need to call raw helper function.
709 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
713 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
714 .func = bpf_get_stackid_tp,
716 .ret_type = RET_INTEGER,
717 .arg1_type = ARG_PTR_TO_CTX,
718 .arg2_type = ARG_CONST_MAP_PTR,
719 .arg3_type = ARG_ANYTHING,
722 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
725 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
727 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
728 (unsigned long) size, flags, 0);
731 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
732 .func = bpf_get_stack_tp,
734 .ret_type = RET_INTEGER,
735 .arg1_type = ARG_PTR_TO_CTX,
736 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
737 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
738 .arg4_type = ARG_ANYTHING,
741 static const struct bpf_func_proto *
742 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
745 case BPF_FUNC_perf_event_output:
746 return &bpf_perf_event_output_proto_tp;
747 case BPF_FUNC_get_stackid:
748 return &bpf_get_stackid_proto_tp;
749 case BPF_FUNC_get_stack:
750 return &bpf_get_stack_proto_tp;
752 return tracing_func_proto(func_id, prog);
756 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
757 const struct bpf_prog *prog,
758 struct bpf_insn_access_aux *info)
760 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
762 if (type != BPF_READ)
767 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
771 const struct bpf_verifier_ops tracepoint_verifier_ops = {
772 .get_func_proto = tp_prog_func_proto,
773 .is_valid_access = tp_prog_is_valid_access,
776 const struct bpf_prog_ops tracepoint_prog_ops = {
779 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
780 struct bpf_perf_event_value *, buf, u32, size)
784 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
786 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
792 memset(buf, 0, size);
796 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
797 .func = bpf_perf_prog_read_value,
799 .ret_type = RET_INTEGER,
800 .arg1_type = ARG_PTR_TO_CTX,
801 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
802 .arg3_type = ARG_CONST_SIZE,
805 static const struct bpf_func_proto *
806 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
809 case BPF_FUNC_perf_event_output:
810 return &bpf_perf_event_output_proto_tp;
811 case BPF_FUNC_get_stackid:
812 return &bpf_get_stackid_proto_tp;
813 case BPF_FUNC_get_stack:
814 return &bpf_get_stack_proto_tp;
815 case BPF_FUNC_perf_prog_read_value:
816 return &bpf_perf_prog_read_value_proto;
818 return tracing_func_proto(func_id, prog);
823 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
824 * to avoid potential recursive reuse issue when/if tracepoints are added
825 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
827 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
828 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
829 struct bpf_map *, map, u64, flags, void *, data, u64, size)
831 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
833 perf_fetch_caller_regs(regs);
834 return ____bpf_perf_event_output(regs, map, flags, data, size);
837 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
838 .func = bpf_perf_event_output_raw_tp,
840 .ret_type = RET_INTEGER,
841 .arg1_type = ARG_PTR_TO_CTX,
842 .arg2_type = ARG_CONST_MAP_PTR,
843 .arg3_type = ARG_ANYTHING,
844 .arg4_type = ARG_PTR_TO_MEM,
845 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
848 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
849 struct bpf_map *, map, u64, flags)
851 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
853 perf_fetch_caller_regs(regs);
854 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
855 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
859 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
860 .func = bpf_get_stackid_raw_tp,
862 .ret_type = RET_INTEGER,
863 .arg1_type = ARG_PTR_TO_CTX,
864 .arg2_type = ARG_CONST_MAP_PTR,
865 .arg3_type = ARG_ANYTHING,
868 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
869 void *, buf, u32, size, u64, flags)
871 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
873 perf_fetch_caller_regs(regs);
874 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
875 (unsigned long) size, flags, 0);
878 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
879 .func = bpf_get_stack_raw_tp,
881 .ret_type = RET_INTEGER,
882 .arg1_type = ARG_PTR_TO_CTX,
883 .arg2_type = ARG_PTR_TO_MEM,
884 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
885 .arg4_type = ARG_ANYTHING,
888 static const struct bpf_func_proto *
889 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
892 case BPF_FUNC_perf_event_output:
893 return &bpf_perf_event_output_proto_raw_tp;
894 case BPF_FUNC_get_stackid:
895 return &bpf_get_stackid_proto_raw_tp;
896 case BPF_FUNC_get_stack:
897 return &bpf_get_stack_proto_raw_tp;
899 return tracing_func_proto(func_id, prog);
903 static bool raw_tp_prog_is_valid_access(int off, int size,
904 enum bpf_access_type type,
905 const struct bpf_prog *prog,
906 struct bpf_insn_access_aux *info)
908 /* largest tracepoint in the kernel has 12 args */
909 if (off < 0 || off >= sizeof(__u64) * 12)
911 if (type != BPF_READ)
918 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
919 .get_func_proto = raw_tp_prog_func_proto,
920 .is_valid_access = raw_tp_prog_is_valid_access,
923 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
926 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
927 enum bpf_access_type type,
928 const struct bpf_prog *prog,
929 struct bpf_insn_access_aux *info)
932 if (size != sizeof(u64) || type != BPF_READ)
934 info->reg_type = PTR_TO_TP_BUFFER;
936 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
939 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
940 .get_func_proto = raw_tp_prog_func_proto,
941 .is_valid_access = raw_tp_writable_prog_is_valid_access,
944 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
947 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
948 const struct bpf_prog *prog,
949 struct bpf_insn_access_aux *info)
951 const int size_u64 = sizeof(u64);
953 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
955 if (type != BPF_READ)
957 if (off % size != 0) {
958 if (sizeof(unsigned long) != 4)
967 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
968 bpf_ctx_record_field_size(info, size_u64);
969 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
972 case bpf_ctx_range(struct bpf_perf_event_data, addr):
973 bpf_ctx_record_field_size(info, size_u64);
974 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
978 if (size != sizeof(long))
985 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
986 const struct bpf_insn *si,
987 struct bpf_insn *insn_buf,
988 struct bpf_prog *prog, u32 *target_size)
990 struct bpf_insn *insn = insn_buf;
993 case offsetof(struct bpf_perf_event_data, sample_period):
994 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
995 data), si->dst_reg, si->src_reg,
996 offsetof(struct bpf_perf_event_data_kern, data));
997 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
998 bpf_target_off(struct perf_sample_data, period, 8,
1001 case offsetof(struct bpf_perf_event_data, addr):
1002 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1003 data), si->dst_reg, si->src_reg,
1004 offsetof(struct bpf_perf_event_data_kern, data));
1005 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1006 bpf_target_off(struct perf_sample_data, addr, 8,
1010 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1011 regs), si->dst_reg, si->src_reg,
1012 offsetof(struct bpf_perf_event_data_kern, regs));
1013 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1018 return insn - insn_buf;
1021 const struct bpf_verifier_ops perf_event_verifier_ops = {
1022 .get_func_proto = pe_prog_func_proto,
1023 .is_valid_access = pe_prog_is_valid_access,
1024 .convert_ctx_access = pe_prog_convert_ctx_access,
1027 const struct bpf_prog_ops perf_event_prog_ops = {
1030 static DEFINE_MUTEX(bpf_event_mutex);
1032 #define BPF_TRACE_MAX_PROGS 64
1034 int perf_event_attach_bpf_prog(struct perf_event *event,
1035 struct bpf_prog *prog)
1037 struct bpf_prog_array __rcu *old_array;
1038 struct bpf_prog_array *new_array;
1042 * Kprobe override only works if they are on the function entry,
1043 * and only if they are on the opt-in list.
1045 if (prog->kprobe_override &&
1046 (!trace_kprobe_on_func_entry(event->tp_event) ||
1047 !trace_kprobe_error_injectable(event->tp_event)))
1050 mutex_lock(&bpf_event_mutex);
1055 old_array = event->tp_event->prog_array;
1057 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1062 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1066 /* set the new array to event->tp_event and set event->prog */
1068 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1069 bpf_prog_array_free(old_array);
1072 mutex_unlock(&bpf_event_mutex);
1076 void perf_event_detach_bpf_prog(struct perf_event *event)
1078 struct bpf_prog_array __rcu *old_array;
1079 struct bpf_prog_array *new_array;
1082 mutex_lock(&bpf_event_mutex);
1087 old_array = event->tp_event->prog_array;
1088 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1092 bpf_prog_array_delete_safe(old_array, event->prog);
1094 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1095 bpf_prog_array_free(old_array);
1098 bpf_prog_put(event->prog);
1102 mutex_unlock(&bpf_event_mutex);
1105 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1107 struct perf_event_query_bpf __user *uquery = info;
1108 struct perf_event_query_bpf query = {};
1109 u32 *ids, prog_cnt, ids_len;
1112 if (!capable(CAP_SYS_ADMIN))
1114 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1116 if (copy_from_user(&query, uquery, sizeof(query)))
1119 ids_len = query.ids_len;
1120 if (ids_len > BPF_TRACE_MAX_PROGS)
1122 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1126 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1127 * is required when user only wants to check for uquery->prog_cnt.
1128 * There is no need to check for it since the case is handled
1129 * gracefully in bpf_prog_array_copy_info.
1132 mutex_lock(&bpf_event_mutex);
1133 ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1137 mutex_unlock(&bpf_event_mutex);
1139 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1140 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1147 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1148 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1150 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1152 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1154 for (; btp < __stop__bpf_raw_tp; btp++) {
1155 if (!strcmp(btp->tp->name, name))
1159 return bpf_get_raw_tracepoint_module(name);
1162 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1164 struct module *mod = __module_address((unsigned long)btp);
1170 static __always_inline
1171 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1175 (void) BPF_PROG_RUN(prog, args);
1180 #define UNPACK(...) __VA_ARGS__
1181 #define REPEAT_1(FN, DL, X, ...) FN(X)
1182 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1183 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1184 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1185 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1186 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1187 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1188 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1189 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1190 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1191 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1192 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1193 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1195 #define SARG(X) u64 arg##X
1196 #define COPY(X) args[X] = arg##X
1198 #define __DL_COM (,)
1199 #define __DL_SEM (;)
1201 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1203 #define BPF_TRACE_DEFN_x(x) \
1204 void bpf_trace_run##x(struct bpf_prog *prog, \
1205 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1208 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1209 __bpf_trace_run(prog, args); \
1211 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1212 BPF_TRACE_DEFN_x(1);
1213 BPF_TRACE_DEFN_x(2);
1214 BPF_TRACE_DEFN_x(3);
1215 BPF_TRACE_DEFN_x(4);
1216 BPF_TRACE_DEFN_x(5);
1217 BPF_TRACE_DEFN_x(6);
1218 BPF_TRACE_DEFN_x(7);
1219 BPF_TRACE_DEFN_x(8);
1220 BPF_TRACE_DEFN_x(9);
1221 BPF_TRACE_DEFN_x(10);
1222 BPF_TRACE_DEFN_x(11);
1223 BPF_TRACE_DEFN_x(12);
1225 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1227 struct tracepoint *tp = btp->tp;
1230 * check that program doesn't access arguments beyond what's
1231 * available in this tracepoint
1233 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1236 if (prog->aux->max_tp_access > btp->writable_size)
1239 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1242 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1244 return __bpf_probe_register(btp, prog);
1247 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1249 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1252 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1253 u32 *fd_type, const char **buf,
1254 u64 *probe_offset, u64 *probe_addr)
1256 bool is_tracepoint, is_syscall_tp;
1257 struct bpf_prog *prog;
1264 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1265 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1268 *prog_id = prog->aux->id;
1269 flags = event->tp_event->flags;
1270 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1271 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1273 if (is_tracepoint || is_syscall_tp) {
1274 *buf = is_tracepoint ? event->tp_event->tp->name
1275 : event->tp_event->name;
1276 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1277 *probe_offset = 0x0;
1282 #ifdef CONFIG_KPROBE_EVENTS
1283 if (flags & TRACE_EVENT_FL_KPROBE)
1284 err = bpf_get_kprobe_info(event, fd_type, buf,
1285 probe_offset, probe_addr,
1286 event->attr.type == PERF_TYPE_TRACEPOINT);
1288 #ifdef CONFIG_UPROBE_EVENTS
1289 if (flags & TRACE_EVENT_FL_UPROBE)
1290 err = bpf_get_uprobe_info(event, fd_type, buf,
1292 event->attr.type == PERF_TYPE_TRACEPOINT);
1299 #ifdef CONFIG_MODULES
1300 int bpf_event_notify(struct notifier_block *nb, unsigned long op, void *module)
1302 struct bpf_trace_module *btm, *tmp;
1303 struct module *mod = module;
1305 if (mod->num_bpf_raw_events == 0 ||
1306 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1309 mutex_lock(&bpf_module_mutex);
1312 case MODULE_STATE_COMING:
1313 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1315 btm->module = module;
1316 list_add(&btm->list, &bpf_trace_modules);
1319 case MODULE_STATE_GOING:
1320 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1321 if (btm->module == module) {
1322 list_del(&btm->list);
1330 mutex_unlock(&bpf_module_mutex);
1335 static struct notifier_block bpf_module_nb = {
1336 .notifier_call = bpf_event_notify,
1339 int __init bpf_event_init(void)
1341 register_module_notifier(&bpf_module_nb);
1345 fs_initcall(bpf_event_init);
1346 #endif /* CONFIG_MODULES */