1 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/slab.h>
11 #include <linux/bpf.h>
12 #include <linux/bpf_perf_event.h>
13 #include <linux/filter.h>
14 #include <linux/uaccess.h>
15 #include <linux/ctype.h>
19 * trace_call_bpf - invoke BPF program
21 * @ctx: opaque context pointer
23 * kprobe handlers execute BPF programs via this helper.
24 * Can be used from static tracepoints in the future.
26 * Return: BPF programs always return an integer which is interpreted by
28 * 0 - return from kprobe (event is filtered out)
29 * 1 - store kprobe event into ring buffer
30 * Other values are reserved and currently alias to 1
32 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
36 if (in_nmi()) /* not supported yet */
41 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
43 * since some bpf program is already running on this cpu,
44 * don't call into another bpf program (same or different)
45 * and don't send kprobe event into ring-buffer,
53 ret = BPF_PROG_RUN(prog, ctx);
57 __this_cpu_dec(bpf_prog_active);
62 EXPORT_SYMBOL_GPL(trace_call_bpf);
64 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
68 ret = probe_kernel_read(dst, unsafe_ptr, size);
69 if (unlikely(ret < 0))
75 static const struct bpf_func_proto bpf_probe_read_proto = {
76 .func = bpf_probe_read,
78 .ret_type = RET_INTEGER,
79 .arg1_type = ARG_PTR_TO_RAW_STACK,
80 .arg2_type = ARG_CONST_STACK_SIZE,
81 .arg3_type = ARG_ANYTHING,
84 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
88 * Ensure we're in user context which is safe for the helper to
89 * run. This helper has no business in a kthread.
91 * access_ok() should prevent writing to non-user memory, but in
92 * some situations (nommu, temporary switch, etc) access_ok() does
93 * not provide enough validation, hence the check on KERNEL_DS.
96 if (unlikely(in_interrupt() ||
97 current->flags & (PF_KTHREAD | PF_EXITING)))
99 if (unlikely(segment_eq(get_fs(), KERNEL_DS)))
101 if (!access_ok(VERIFY_WRITE, unsafe_ptr, size))
104 return probe_kernel_write(unsafe_ptr, src, size);
107 static const struct bpf_func_proto bpf_probe_write_user_proto = {
108 .func = bpf_probe_write_user,
110 .ret_type = RET_INTEGER,
111 .arg1_type = ARG_ANYTHING,
112 .arg2_type = ARG_PTR_TO_STACK,
113 .arg3_type = ARG_CONST_STACK_SIZE,
116 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
118 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
119 current->comm, task_pid_nr(current));
121 return &bpf_probe_write_user_proto;
125 * limited trace_printk()
126 * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
128 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
129 u64, arg2, u64, arg3)
131 bool str_seen = false;
139 * bpf_check()->check_func_arg()->check_stack_boundary()
140 * guarantees that fmt points to bpf program stack,
141 * fmt_size bytes of it were initialized and fmt_size > 0
143 if (fmt[--fmt_size] != 0)
146 /* check format string for allowed specifiers */
147 for (i = 0; i < fmt_size; i++) {
148 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
157 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
162 } else if (fmt[i] == 'p' || fmt[i] == 's') {
165 if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
168 if (fmt[i - 1] == 's') {
170 /* allow only one '%s' per fmt string */
189 strncpy_from_unsafe(buf,
190 (void *) (long) unsafe_addr,
201 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
206 return __trace_printk(1/* fake ip will not be printed */, fmt,
207 mod[0] == 2 ? arg1 : mod[0] == 1 ? (long) arg1 : (u32) arg1,
208 mod[1] == 2 ? arg2 : mod[1] == 1 ? (long) arg2 : (u32) arg2,
209 mod[2] == 2 ? arg3 : mod[2] == 1 ? (long) arg3 : (u32) arg3);
212 static const struct bpf_func_proto bpf_trace_printk_proto = {
213 .func = bpf_trace_printk,
215 .ret_type = RET_INTEGER,
216 .arg1_type = ARG_PTR_TO_STACK,
217 .arg2_type = ARG_CONST_STACK_SIZE,
220 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
223 * this program might be calling bpf_trace_printk,
224 * so allocate per-cpu printk buffers
226 trace_printk_init_buffers();
228 return &bpf_trace_printk_proto;
231 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
233 struct bpf_array *array = container_of(map, struct bpf_array, map);
234 unsigned int cpu = smp_processor_id();
235 u64 index = flags & BPF_F_INDEX_MASK;
236 struct bpf_event_entry *ee;
237 struct perf_event *event;
239 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
241 if (index == BPF_F_CURRENT_CPU)
243 if (unlikely(index >= array->map.max_entries))
246 ee = READ_ONCE(array->ptrs[index]);
251 if (unlikely(event->attr.type != PERF_TYPE_HARDWARE &&
252 event->attr.type != PERF_TYPE_RAW))
255 /* make sure event is local and doesn't have pmu::count */
256 if (unlikely(event->oncpu != cpu || event->pmu->count))
260 * we don't know if the function is run successfully by the
261 * return value. It can be judged in other places, such as
264 return perf_event_read_local(event);
267 static const struct bpf_func_proto bpf_perf_event_read_proto = {
268 .func = bpf_perf_event_read,
270 .ret_type = RET_INTEGER,
271 .arg1_type = ARG_CONST_MAP_PTR,
272 .arg2_type = ARG_ANYTHING,
275 static __always_inline u64
276 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
277 u64 flags, struct perf_raw_record *raw)
279 struct bpf_array *array = container_of(map, struct bpf_array, map);
280 unsigned int cpu = smp_processor_id();
281 u64 index = flags & BPF_F_INDEX_MASK;
282 struct perf_sample_data sample_data;
283 struct bpf_event_entry *ee;
284 struct perf_event *event;
286 if (index == BPF_F_CURRENT_CPU)
288 if (unlikely(index >= array->map.max_entries))
291 ee = READ_ONCE(array->ptrs[index]);
296 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
297 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
300 if (unlikely(event->oncpu != cpu))
303 perf_sample_data_init(&sample_data, 0, 0);
304 sample_data.raw = raw;
305 perf_event_output(event, &sample_data, regs);
309 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
310 u64, flags, void *, data, u64, size)
312 struct perf_raw_record raw = {
319 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
322 return __bpf_perf_event_output(regs, map, flags, &raw);
325 static const struct bpf_func_proto bpf_perf_event_output_proto = {
326 .func = bpf_perf_event_output,
328 .ret_type = RET_INTEGER,
329 .arg1_type = ARG_PTR_TO_CTX,
330 .arg2_type = ARG_CONST_MAP_PTR,
331 .arg3_type = ARG_ANYTHING,
332 .arg4_type = ARG_PTR_TO_STACK,
333 .arg5_type = ARG_CONST_STACK_SIZE,
336 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
338 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
339 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
341 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
342 struct perf_raw_frag frag = {
347 struct perf_raw_record raw = {
350 .next = ctx_size ? &frag : NULL,
357 perf_fetch_caller_regs(regs);
359 return __bpf_perf_event_output(regs, map, flags, &raw);
362 BPF_CALL_0(bpf_get_current_task)
364 return (long) current;
367 static const struct bpf_func_proto bpf_get_current_task_proto = {
368 .func = bpf_get_current_task,
370 .ret_type = RET_INTEGER,
373 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
375 struct bpf_array *array = container_of(map, struct bpf_array, map);
378 if (unlikely(in_interrupt()))
380 if (unlikely(idx >= array->map.max_entries))
383 cgrp = READ_ONCE(array->ptrs[idx]);
387 return task_under_cgroup_hierarchy(current, cgrp);
390 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
391 .func = bpf_current_task_under_cgroup,
393 .ret_type = RET_INTEGER,
394 .arg1_type = ARG_CONST_MAP_PTR,
395 .arg2_type = ARG_ANYTHING,
398 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
401 case BPF_FUNC_map_lookup_elem:
402 return &bpf_map_lookup_elem_proto;
403 case BPF_FUNC_map_update_elem:
404 return &bpf_map_update_elem_proto;
405 case BPF_FUNC_map_delete_elem:
406 return &bpf_map_delete_elem_proto;
407 case BPF_FUNC_probe_read:
408 return &bpf_probe_read_proto;
409 case BPF_FUNC_ktime_get_ns:
410 return &bpf_ktime_get_ns_proto;
411 case BPF_FUNC_tail_call:
412 return &bpf_tail_call_proto;
413 case BPF_FUNC_get_current_pid_tgid:
414 return &bpf_get_current_pid_tgid_proto;
415 case BPF_FUNC_get_current_task:
416 return &bpf_get_current_task_proto;
417 case BPF_FUNC_get_current_uid_gid:
418 return &bpf_get_current_uid_gid_proto;
419 case BPF_FUNC_get_current_comm:
420 return &bpf_get_current_comm_proto;
421 case BPF_FUNC_trace_printk:
422 return bpf_get_trace_printk_proto();
423 case BPF_FUNC_get_smp_processor_id:
424 return &bpf_get_smp_processor_id_proto;
425 case BPF_FUNC_perf_event_read:
426 return &bpf_perf_event_read_proto;
427 case BPF_FUNC_probe_write_user:
428 return bpf_get_probe_write_proto();
429 case BPF_FUNC_current_task_under_cgroup:
430 return &bpf_current_task_under_cgroup_proto;
431 case BPF_FUNC_get_prandom_u32:
432 return &bpf_get_prandom_u32_proto;
438 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
441 case BPF_FUNC_perf_event_output:
442 return &bpf_perf_event_output_proto;
443 case BPF_FUNC_get_stackid:
444 return &bpf_get_stackid_proto;
446 return tracing_func_proto(func_id);
450 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
451 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
452 enum bpf_reg_type *reg_type)
454 if (off < 0 || off >= sizeof(struct pt_regs))
456 if (type != BPF_READ)
463 static const struct bpf_verifier_ops kprobe_prog_ops = {
464 .get_func_proto = kprobe_prog_func_proto,
465 .is_valid_access = kprobe_prog_is_valid_access,
468 static struct bpf_prog_type_list kprobe_tl = {
469 .ops = &kprobe_prog_ops,
470 .type = BPF_PROG_TYPE_KPROBE,
473 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
474 u64, flags, void *, data, u64, size)
476 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
479 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
480 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
481 * from there and call the same bpf_perf_event_output() helper inline.
483 return ____bpf_perf_event_output(regs, map, flags, data, size);
486 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
487 .func = bpf_perf_event_output_tp,
489 .ret_type = RET_INTEGER,
490 .arg1_type = ARG_PTR_TO_CTX,
491 .arg2_type = ARG_CONST_MAP_PTR,
492 .arg3_type = ARG_ANYTHING,
493 .arg4_type = ARG_PTR_TO_STACK,
494 .arg5_type = ARG_CONST_STACK_SIZE,
497 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
500 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
503 * Same comment as in bpf_perf_event_output_tp(), only that this time
504 * the other helper's function body cannot be inlined due to being
505 * external, thus we need to call raw helper function.
507 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
511 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
512 .func = bpf_get_stackid_tp,
514 .ret_type = RET_INTEGER,
515 .arg1_type = ARG_PTR_TO_CTX,
516 .arg2_type = ARG_CONST_MAP_PTR,
517 .arg3_type = ARG_ANYTHING,
520 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
523 case BPF_FUNC_perf_event_output:
524 return &bpf_perf_event_output_proto_tp;
525 case BPF_FUNC_get_stackid:
526 return &bpf_get_stackid_proto_tp;
528 return tracing_func_proto(func_id);
532 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
533 enum bpf_reg_type *reg_type)
535 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
537 if (type != BPF_READ)
544 static const struct bpf_verifier_ops tracepoint_prog_ops = {
545 .get_func_proto = tp_prog_func_proto,
546 .is_valid_access = tp_prog_is_valid_access,
549 static struct bpf_prog_type_list tracepoint_tl = {
550 .ops = &tracepoint_prog_ops,
551 .type = BPF_PROG_TYPE_TRACEPOINT,
554 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
555 enum bpf_reg_type *reg_type)
557 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
559 if (type != BPF_READ)
563 if (off == offsetof(struct bpf_perf_event_data, sample_period)) {
564 if (size != sizeof(u64))
567 if (size != sizeof(long))
573 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, int dst_reg,
574 int src_reg, int ctx_off,
575 struct bpf_insn *insn_buf,
576 struct bpf_prog *prog)
578 struct bpf_insn *insn = insn_buf;
581 case offsetof(struct bpf_perf_event_data, sample_period):
582 BUILD_BUG_ON(FIELD_SIZEOF(struct perf_sample_data, period) != sizeof(u64));
584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
585 data), dst_reg, src_reg,
586 offsetof(struct bpf_perf_event_data_kern, data));
587 *insn++ = BPF_LDX_MEM(BPF_DW, dst_reg, dst_reg,
588 offsetof(struct perf_sample_data, period));
591 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
592 regs), dst_reg, src_reg,
593 offsetof(struct bpf_perf_event_data_kern, regs));
594 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), dst_reg, dst_reg, ctx_off);
598 return insn - insn_buf;
601 static const struct bpf_verifier_ops perf_event_prog_ops = {
602 .get_func_proto = tp_prog_func_proto,
603 .is_valid_access = pe_prog_is_valid_access,
604 .convert_ctx_access = pe_prog_convert_ctx_access,
607 static struct bpf_prog_type_list perf_event_tl = {
608 .ops = &perf_event_prog_ops,
609 .type = BPF_PROG_TYPE_PERF_EVENT,
612 static int __init register_kprobe_prog_ops(void)
614 bpf_register_prog_type(&kprobe_tl);
615 bpf_register_prog_type(&tracepoint_tl);
616 bpf_register_prog_type(&perf_event_tl);
619 late_initcall(register_kprobe_prog_ops);