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 __always_inline u64
414 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
415 u64 flags, struct perf_sample_data *sd)
417 struct bpf_array *array = container_of(map, struct bpf_array, map);
418 unsigned int cpu = smp_processor_id();
419 u64 index = flags & BPF_F_INDEX_MASK;
420 struct bpf_event_entry *ee;
421 struct perf_event *event;
423 if (index == BPF_F_CURRENT_CPU)
425 if (unlikely(index >= array->map.max_entries))
428 ee = READ_ONCE(array->ptrs[index]);
433 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
434 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
437 if (unlikely(event->oncpu != cpu))
440 return perf_event_output(event, sd, regs);
444 * Support executing tracepoints in normal, irq, and nmi context that each call
445 * bpf_perf_event_output
447 struct bpf_trace_sample_data {
448 struct perf_sample_data sds[3];
451 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
452 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
453 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
454 u64, flags, void *, data, u64, size)
456 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
457 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
458 struct perf_raw_record raw = {
464 struct perf_sample_data *sd;
467 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
472 sd = &sds->sds[nest_level - 1];
474 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
479 perf_sample_data_init(sd, 0, 0);
482 err = __bpf_perf_event_output(regs, map, flags, sd);
485 this_cpu_dec(bpf_trace_nest_level);
489 static const struct bpf_func_proto bpf_perf_event_output_proto = {
490 .func = bpf_perf_event_output,
492 .ret_type = RET_INTEGER,
493 .arg1_type = ARG_PTR_TO_CTX,
494 .arg2_type = ARG_CONST_MAP_PTR,
495 .arg3_type = ARG_ANYTHING,
496 .arg4_type = ARG_PTR_TO_MEM,
497 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
500 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
501 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
503 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
504 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
506 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
507 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
508 struct perf_raw_frag frag = {
513 struct perf_raw_record raw = {
516 .next = ctx_size ? &frag : NULL,
523 perf_fetch_caller_regs(regs);
524 perf_sample_data_init(sd, 0, 0);
527 return __bpf_perf_event_output(regs, map, flags, sd);
530 BPF_CALL_0(bpf_get_current_task)
532 return (long) current;
535 static const struct bpf_func_proto bpf_get_current_task_proto = {
536 .func = bpf_get_current_task,
538 .ret_type = RET_INTEGER,
541 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
543 struct bpf_array *array = container_of(map, struct bpf_array, map);
546 if (unlikely(idx >= array->map.max_entries))
549 cgrp = READ_ONCE(array->ptrs[idx]);
553 return task_under_cgroup_hierarchy(current, cgrp);
556 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
557 .func = bpf_current_task_under_cgroup,
559 .ret_type = RET_INTEGER,
560 .arg1_type = ARG_CONST_MAP_PTR,
561 .arg2_type = ARG_ANYTHING,
564 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
565 const void *, unsafe_ptr)
570 * The strncpy_from_unsafe() call will likely not fill the entire
571 * buffer, but that's okay in this circumstance as we're probing
572 * arbitrary memory anyway similar to bpf_probe_read() and might
573 * as well probe the stack. Thus, memory is explicitly cleared
574 * only in error case, so that improper users ignoring return
575 * code altogether don't copy garbage; otherwise length of string
576 * is returned that can be used for bpf_perf_event_output() et al.
578 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
579 if (unlikely(ret < 0))
580 memset(dst, 0, size);
585 static const struct bpf_func_proto bpf_probe_read_str_proto = {
586 .func = bpf_probe_read_str,
588 .ret_type = RET_INTEGER,
589 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
590 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
591 .arg3_type = ARG_ANYTHING,
594 static const struct bpf_func_proto *
595 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
598 case BPF_FUNC_map_lookup_elem:
599 return &bpf_map_lookup_elem_proto;
600 case BPF_FUNC_map_update_elem:
601 return &bpf_map_update_elem_proto;
602 case BPF_FUNC_map_delete_elem:
603 return &bpf_map_delete_elem_proto;
604 case BPF_FUNC_map_push_elem:
605 return &bpf_map_push_elem_proto;
606 case BPF_FUNC_map_pop_elem:
607 return &bpf_map_pop_elem_proto;
608 case BPF_FUNC_map_peek_elem:
609 return &bpf_map_peek_elem_proto;
610 case BPF_FUNC_probe_read:
611 return &bpf_probe_read_proto;
612 case BPF_FUNC_ktime_get_ns:
613 return &bpf_ktime_get_ns_proto;
614 case BPF_FUNC_tail_call:
615 return &bpf_tail_call_proto;
616 case BPF_FUNC_get_current_pid_tgid:
617 return &bpf_get_current_pid_tgid_proto;
618 case BPF_FUNC_get_current_task:
619 return &bpf_get_current_task_proto;
620 case BPF_FUNC_get_current_uid_gid:
621 return &bpf_get_current_uid_gid_proto;
622 case BPF_FUNC_get_current_comm:
623 return &bpf_get_current_comm_proto;
624 case BPF_FUNC_trace_printk:
625 return bpf_get_trace_printk_proto();
626 case BPF_FUNC_get_smp_processor_id:
627 return &bpf_get_smp_processor_id_proto;
628 case BPF_FUNC_get_numa_node_id:
629 return &bpf_get_numa_node_id_proto;
630 case BPF_FUNC_perf_event_read:
631 return &bpf_perf_event_read_proto;
632 case BPF_FUNC_probe_write_user:
633 return bpf_get_probe_write_proto();
634 case BPF_FUNC_current_task_under_cgroup:
635 return &bpf_current_task_under_cgroup_proto;
636 case BPF_FUNC_get_prandom_u32:
637 return &bpf_get_prandom_u32_proto;
638 case BPF_FUNC_probe_read_str:
639 return &bpf_probe_read_str_proto;
640 #ifdef CONFIG_CGROUPS
641 case BPF_FUNC_get_current_cgroup_id:
642 return &bpf_get_current_cgroup_id_proto;
649 static const struct bpf_func_proto *
650 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
653 case BPF_FUNC_perf_event_output:
654 return &bpf_perf_event_output_proto;
655 case BPF_FUNC_get_stackid:
656 return &bpf_get_stackid_proto;
657 case BPF_FUNC_get_stack:
658 return &bpf_get_stack_proto;
659 case BPF_FUNC_perf_event_read_value:
660 return &bpf_perf_event_read_value_proto;
661 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
662 case BPF_FUNC_override_return:
663 return &bpf_override_return_proto;
666 return tracing_func_proto(func_id, prog);
670 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
671 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
672 const struct bpf_prog *prog,
673 struct bpf_insn_access_aux *info)
675 if (off < 0 || off >= sizeof(struct pt_regs))
677 if (type != BPF_READ)
682 * Assertion for 32 bit to make sure last 8 byte access
683 * (BPF_DW) to the last 4 byte member is disallowed.
685 if (off + size > sizeof(struct pt_regs))
691 const struct bpf_verifier_ops kprobe_verifier_ops = {
692 .get_func_proto = kprobe_prog_func_proto,
693 .is_valid_access = kprobe_prog_is_valid_access,
696 const struct bpf_prog_ops kprobe_prog_ops = {
699 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
700 u64, flags, void *, data, u64, size)
702 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
705 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
706 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
707 * from there and call the same bpf_perf_event_output() helper inline.
709 return ____bpf_perf_event_output(regs, map, flags, data, size);
712 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
713 .func = bpf_perf_event_output_tp,
715 .ret_type = RET_INTEGER,
716 .arg1_type = ARG_PTR_TO_CTX,
717 .arg2_type = ARG_CONST_MAP_PTR,
718 .arg3_type = ARG_ANYTHING,
719 .arg4_type = ARG_PTR_TO_MEM,
720 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
723 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
726 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
729 * Same comment as in bpf_perf_event_output_tp(), only that this time
730 * the other helper's function body cannot be inlined due to being
731 * external, thus we need to call raw helper function.
733 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
737 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
738 .func = bpf_get_stackid_tp,
740 .ret_type = RET_INTEGER,
741 .arg1_type = ARG_PTR_TO_CTX,
742 .arg2_type = ARG_CONST_MAP_PTR,
743 .arg3_type = ARG_ANYTHING,
746 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
749 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
751 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
752 (unsigned long) size, flags, 0);
755 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
756 .func = bpf_get_stack_tp,
758 .ret_type = RET_INTEGER,
759 .arg1_type = ARG_PTR_TO_CTX,
760 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
761 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
762 .arg4_type = ARG_ANYTHING,
765 static const struct bpf_func_proto *
766 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
769 case BPF_FUNC_perf_event_output:
770 return &bpf_perf_event_output_proto_tp;
771 case BPF_FUNC_get_stackid:
772 return &bpf_get_stackid_proto_tp;
773 case BPF_FUNC_get_stack:
774 return &bpf_get_stack_proto_tp;
776 return tracing_func_proto(func_id, prog);
780 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
781 const struct bpf_prog *prog,
782 struct bpf_insn_access_aux *info)
784 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
786 if (type != BPF_READ)
791 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
795 const struct bpf_verifier_ops tracepoint_verifier_ops = {
796 .get_func_proto = tp_prog_func_proto,
797 .is_valid_access = tp_prog_is_valid_access,
800 const struct bpf_prog_ops tracepoint_prog_ops = {
803 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
804 struct bpf_perf_event_value *, buf, u32, size)
808 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
810 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
816 memset(buf, 0, size);
820 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
821 .func = bpf_perf_prog_read_value,
823 .ret_type = RET_INTEGER,
824 .arg1_type = ARG_PTR_TO_CTX,
825 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
826 .arg3_type = ARG_CONST_SIZE,
829 static const struct bpf_func_proto *
830 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
833 case BPF_FUNC_perf_event_output:
834 return &bpf_perf_event_output_proto_tp;
835 case BPF_FUNC_get_stackid:
836 return &bpf_get_stackid_proto_tp;
837 case BPF_FUNC_get_stack:
838 return &bpf_get_stack_proto_tp;
839 case BPF_FUNC_perf_prog_read_value:
840 return &bpf_perf_prog_read_value_proto;
842 return tracing_func_proto(func_id, prog);
847 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
848 * to avoid potential recursive reuse issue when/if tracepoints are added
849 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
851 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
852 * in normal, irq, and nmi context.
854 struct bpf_raw_tp_regs {
855 struct pt_regs regs[3];
857 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
858 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
859 static struct pt_regs *get_bpf_raw_tp_regs(void)
861 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
862 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
864 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
865 this_cpu_dec(bpf_raw_tp_nest_level);
866 return ERR_PTR(-EBUSY);
869 return &tp_regs->regs[nest_level - 1];
872 static void put_bpf_raw_tp_regs(void)
874 this_cpu_dec(bpf_raw_tp_nest_level);
877 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
878 struct bpf_map *, map, u64, flags, void *, data, u64, size)
880 struct pt_regs *regs = get_bpf_raw_tp_regs();
884 return PTR_ERR(regs);
886 perf_fetch_caller_regs(regs);
887 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
889 put_bpf_raw_tp_regs();
893 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
894 .func = bpf_perf_event_output_raw_tp,
896 .ret_type = RET_INTEGER,
897 .arg1_type = ARG_PTR_TO_CTX,
898 .arg2_type = ARG_CONST_MAP_PTR,
899 .arg3_type = ARG_ANYTHING,
900 .arg4_type = ARG_PTR_TO_MEM,
901 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
904 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
905 struct bpf_map *, map, u64, flags)
907 struct pt_regs *regs = get_bpf_raw_tp_regs();
911 return PTR_ERR(regs);
913 perf_fetch_caller_regs(regs);
914 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
915 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
917 put_bpf_raw_tp_regs();
921 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
922 .func = bpf_get_stackid_raw_tp,
924 .ret_type = RET_INTEGER,
925 .arg1_type = ARG_PTR_TO_CTX,
926 .arg2_type = ARG_CONST_MAP_PTR,
927 .arg3_type = ARG_ANYTHING,
930 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
931 void *, buf, u32, size, u64, flags)
933 struct pt_regs *regs = get_bpf_raw_tp_regs();
937 return PTR_ERR(regs);
939 perf_fetch_caller_regs(regs);
940 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
941 (unsigned long) size, flags, 0);
942 put_bpf_raw_tp_regs();
946 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
947 .func = bpf_get_stack_raw_tp,
949 .ret_type = RET_INTEGER,
950 .arg1_type = ARG_PTR_TO_CTX,
951 .arg2_type = ARG_PTR_TO_MEM,
952 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
953 .arg4_type = ARG_ANYTHING,
956 static const struct bpf_func_proto *
957 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
960 case BPF_FUNC_perf_event_output:
961 return &bpf_perf_event_output_proto_raw_tp;
962 case BPF_FUNC_get_stackid:
963 return &bpf_get_stackid_proto_raw_tp;
964 case BPF_FUNC_get_stack:
965 return &bpf_get_stack_proto_raw_tp;
967 return tracing_func_proto(func_id, prog);
971 static bool raw_tp_prog_is_valid_access(int off, int size,
972 enum bpf_access_type type,
973 const struct bpf_prog *prog,
974 struct bpf_insn_access_aux *info)
976 /* largest tracepoint in the kernel has 12 args */
977 if (off < 0 || off >= sizeof(__u64) * 12)
979 if (type != BPF_READ)
986 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
987 .get_func_proto = raw_tp_prog_func_proto,
988 .is_valid_access = raw_tp_prog_is_valid_access,
991 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
994 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
995 enum bpf_access_type type,
996 const struct bpf_prog *prog,
997 struct bpf_insn_access_aux *info)
1000 if (size != sizeof(u64) || type != BPF_READ)
1002 info->reg_type = PTR_TO_TP_BUFFER;
1004 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1007 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1008 .get_func_proto = raw_tp_prog_func_proto,
1009 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1012 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1015 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1016 const struct bpf_prog *prog,
1017 struct bpf_insn_access_aux *info)
1019 const int size_u64 = sizeof(u64);
1021 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1023 if (type != BPF_READ)
1025 if (off % size != 0) {
1026 if (sizeof(unsigned long) != 4)
1030 if (off % size != 4)
1035 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1036 bpf_ctx_record_field_size(info, size_u64);
1037 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1040 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1041 bpf_ctx_record_field_size(info, size_u64);
1042 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1046 if (size != sizeof(long))
1053 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1054 const struct bpf_insn *si,
1055 struct bpf_insn *insn_buf,
1056 struct bpf_prog *prog, u32 *target_size)
1058 struct bpf_insn *insn = insn_buf;
1061 case offsetof(struct bpf_perf_event_data, sample_period):
1062 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1063 data), si->dst_reg, si->src_reg,
1064 offsetof(struct bpf_perf_event_data_kern, data));
1065 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1066 bpf_target_off(struct perf_sample_data, period, 8,
1069 case offsetof(struct bpf_perf_event_data, addr):
1070 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1071 data), si->dst_reg, si->src_reg,
1072 offsetof(struct bpf_perf_event_data_kern, data));
1073 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1074 bpf_target_off(struct perf_sample_data, addr, 8,
1078 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1079 regs), si->dst_reg, si->src_reg,
1080 offsetof(struct bpf_perf_event_data_kern, regs));
1081 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1086 return insn - insn_buf;
1089 const struct bpf_verifier_ops perf_event_verifier_ops = {
1090 .get_func_proto = pe_prog_func_proto,
1091 .is_valid_access = pe_prog_is_valid_access,
1092 .convert_ctx_access = pe_prog_convert_ctx_access,
1095 const struct bpf_prog_ops perf_event_prog_ops = {
1098 static DEFINE_MUTEX(bpf_event_mutex);
1100 #define BPF_TRACE_MAX_PROGS 64
1102 int perf_event_attach_bpf_prog(struct perf_event *event,
1103 struct bpf_prog *prog)
1105 struct bpf_prog_array __rcu *old_array;
1106 struct bpf_prog_array *new_array;
1110 * Kprobe override only works if they are on the function entry,
1111 * and only if they are on the opt-in list.
1113 if (prog->kprobe_override &&
1114 (!trace_kprobe_on_func_entry(event->tp_event) ||
1115 !trace_kprobe_error_injectable(event->tp_event)))
1118 mutex_lock(&bpf_event_mutex);
1123 old_array = event->tp_event->prog_array;
1125 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1130 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1134 /* set the new array to event->tp_event and set event->prog */
1136 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1137 bpf_prog_array_free(old_array);
1140 mutex_unlock(&bpf_event_mutex);
1144 void perf_event_detach_bpf_prog(struct perf_event *event)
1146 struct bpf_prog_array __rcu *old_array;
1147 struct bpf_prog_array *new_array;
1150 mutex_lock(&bpf_event_mutex);
1155 old_array = event->tp_event->prog_array;
1156 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1160 bpf_prog_array_delete_safe(old_array, event->prog);
1162 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1163 bpf_prog_array_free(old_array);
1166 bpf_prog_put(event->prog);
1170 mutex_unlock(&bpf_event_mutex);
1173 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1175 struct perf_event_query_bpf __user *uquery = info;
1176 struct perf_event_query_bpf query = {};
1177 u32 *ids, prog_cnt, ids_len;
1180 if (!capable(CAP_SYS_ADMIN))
1182 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1184 if (copy_from_user(&query, uquery, sizeof(query)))
1187 ids_len = query.ids_len;
1188 if (ids_len > BPF_TRACE_MAX_PROGS)
1190 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1194 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1195 * is required when user only wants to check for uquery->prog_cnt.
1196 * There is no need to check for it since the case is handled
1197 * gracefully in bpf_prog_array_copy_info.
1200 mutex_lock(&bpf_event_mutex);
1201 ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1205 mutex_unlock(&bpf_event_mutex);
1207 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1208 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1215 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1216 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1218 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1220 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1222 for (; btp < __stop__bpf_raw_tp; btp++) {
1223 if (!strcmp(btp->tp->name, name))
1227 return bpf_get_raw_tracepoint_module(name);
1230 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1232 struct module *mod = __module_address((unsigned long)btp);
1238 static __always_inline
1239 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1243 (void) BPF_PROG_RUN(prog, args);
1248 #define UNPACK(...) __VA_ARGS__
1249 #define REPEAT_1(FN, DL, X, ...) FN(X)
1250 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1251 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1252 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1253 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1254 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1255 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1256 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1257 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1258 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1259 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1260 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1261 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1263 #define SARG(X) u64 arg##X
1264 #define COPY(X) args[X] = arg##X
1266 #define __DL_COM (,)
1267 #define __DL_SEM (;)
1269 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1271 #define BPF_TRACE_DEFN_x(x) \
1272 void bpf_trace_run##x(struct bpf_prog *prog, \
1273 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1276 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1277 __bpf_trace_run(prog, args); \
1279 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1280 BPF_TRACE_DEFN_x(1);
1281 BPF_TRACE_DEFN_x(2);
1282 BPF_TRACE_DEFN_x(3);
1283 BPF_TRACE_DEFN_x(4);
1284 BPF_TRACE_DEFN_x(5);
1285 BPF_TRACE_DEFN_x(6);
1286 BPF_TRACE_DEFN_x(7);
1287 BPF_TRACE_DEFN_x(8);
1288 BPF_TRACE_DEFN_x(9);
1289 BPF_TRACE_DEFN_x(10);
1290 BPF_TRACE_DEFN_x(11);
1291 BPF_TRACE_DEFN_x(12);
1293 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1295 struct tracepoint *tp = btp->tp;
1298 * check that program doesn't access arguments beyond what's
1299 * available in this tracepoint
1301 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1304 if (prog->aux->max_tp_access > btp->writable_size)
1307 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1310 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1312 return __bpf_probe_register(btp, prog);
1315 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1317 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1320 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1321 u32 *fd_type, const char **buf,
1322 u64 *probe_offset, u64 *probe_addr)
1324 bool is_tracepoint, is_syscall_tp;
1325 struct bpf_prog *prog;
1332 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1333 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1336 *prog_id = prog->aux->id;
1337 flags = event->tp_event->flags;
1338 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1339 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1341 if (is_tracepoint || is_syscall_tp) {
1342 *buf = is_tracepoint ? event->tp_event->tp->name
1343 : event->tp_event->name;
1344 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1345 *probe_offset = 0x0;
1350 #ifdef CONFIG_KPROBE_EVENTS
1351 if (flags & TRACE_EVENT_FL_KPROBE)
1352 err = bpf_get_kprobe_info(event, fd_type, buf,
1353 probe_offset, probe_addr,
1354 event->attr.type == PERF_TYPE_TRACEPOINT);
1356 #ifdef CONFIG_UPROBE_EVENTS
1357 if (flags & TRACE_EVENT_FL_UPROBE)
1358 err = bpf_get_uprobe_info(event, fd_type, buf,
1360 event->attr.type == PERF_TYPE_TRACEPOINT);
1367 #ifdef CONFIG_MODULES
1368 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1371 struct bpf_trace_module *btm, *tmp;
1372 struct module *mod = module;
1374 if (mod->num_bpf_raw_events == 0 ||
1375 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1378 mutex_lock(&bpf_module_mutex);
1381 case MODULE_STATE_COMING:
1382 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1384 btm->module = module;
1385 list_add(&btm->list, &bpf_trace_modules);
1388 case MODULE_STATE_GOING:
1389 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1390 if (btm->module == module) {
1391 list_del(&btm->list);
1399 mutex_unlock(&bpf_module_mutex);
1404 static struct notifier_block bpf_module_nb = {
1405 .notifier_call = bpf_event_notify,
1408 static int __init bpf_event_init(void)
1410 register_module_notifier(&bpf_module_nb);
1414 fs_initcall(bpf_event_init);
1415 #endif /* CONFIG_MODULES */