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_UNINIT_MEM,
80 .arg2_type = ARG_CONST_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(uaccess_kernel()))
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_MEM,
113 .arg3_type = ARG_CONST_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 * Only limited trace_printk() conversion specifiers allowed:
126 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
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] != 'i' && fmt[i] != 'd' &&
202 fmt[i] != 'u' && fmt[i] != 'x')
207 /* Horrid workaround for getting va_list handling working with different
208 * argument type combinations generically for 32 and 64 bit archs.
210 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
211 #define __BPF_TP(...) \
212 __trace_printk(1 /* Fake ip will not be printed. */, \
215 #define __BPF_ARG1_TP(...) \
216 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
217 ? __BPF_TP(arg1, ##__VA_ARGS__) \
218 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
219 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
220 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
222 #define __BPF_ARG2_TP(...) \
223 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
224 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
225 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
226 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
227 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
229 #define __BPF_ARG3_TP(...) \
230 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
231 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
232 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
233 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
234 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
236 return __BPF_TP_EMIT();
239 static const struct bpf_func_proto bpf_trace_printk_proto = {
240 .func = bpf_trace_printk,
242 .ret_type = RET_INTEGER,
243 .arg1_type = ARG_PTR_TO_MEM,
244 .arg2_type = ARG_CONST_SIZE,
247 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
250 * this program might be calling bpf_trace_printk,
251 * so allocate per-cpu printk buffers
253 trace_printk_init_buffers();
255 return &bpf_trace_printk_proto;
258 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
260 struct bpf_array *array = container_of(map, struct bpf_array, map);
261 unsigned int cpu = smp_processor_id();
262 u64 index = flags & BPF_F_INDEX_MASK;
263 struct bpf_event_entry *ee;
267 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
269 if (index == BPF_F_CURRENT_CPU)
271 if (unlikely(index >= array->map.max_entries))
274 ee = READ_ONCE(array->ptrs[index]);
278 err = perf_event_read_local(ee->event, &value);
280 * this api is ugly since we miss [-22..-2] range of valid
281 * counter values, but that's uapi
288 static const struct bpf_func_proto bpf_perf_event_read_proto = {
289 .func = bpf_perf_event_read,
291 .ret_type = RET_INTEGER,
292 .arg1_type = ARG_CONST_MAP_PTR,
293 .arg2_type = ARG_ANYTHING,
296 static DEFINE_PER_CPU(struct perf_sample_data, bpf_sd);
298 static __always_inline u64
299 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
300 u64 flags, struct perf_raw_record *raw)
302 struct bpf_array *array = container_of(map, struct bpf_array, map);
303 struct perf_sample_data *sd = this_cpu_ptr(&bpf_sd);
304 unsigned int cpu = smp_processor_id();
305 u64 index = flags & BPF_F_INDEX_MASK;
306 struct bpf_event_entry *ee;
307 struct perf_event *event;
309 if (index == BPF_F_CURRENT_CPU)
311 if (unlikely(index >= array->map.max_entries))
314 ee = READ_ONCE(array->ptrs[index]);
319 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
320 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
323 if (unlikely(event->oncpu != cpu))
326 perf_sample_data_init(sd, 0, 0);
328 perf_event_output(event, sd, regs);
332 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
333 u64, flags, void *, data, u64, size)
335 struct perf_raw_record raw = {
342 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
345 return __bpf_perf_event_output(regs, map, flags, &raw);
348 static const struct bpf_func_proto bpf_perf_event_output_proto = {
349 .func = bpf_perf_event_output,
351 .ret_type = RET_INTEGER,
352 .arg1_type = ARG_PTR_TO_CTX,
353 .arg2_type = ARG_CONST_MAP_PTR,
354 .arg3_type = ARG_ANYTHING,
355 .arg4_type = ARG_PTR_TO_MEM,
356 .arg5_type = ARG_CONST_SIZE,
359 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
361 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
362 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
364 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
365 struct perf_raw_frag frag = {
370 struct perf_raw_record raw = {
373 .next = ctx_size ? &frag : NULL,
380 perf_fetch_caller_regs(regs);
382 return __bpf_perf_event_output(regs, map, flags, &raw);
385 BPF_CALL_0(bpf_get_current_task)
387 return (long) current;
390 static const struct bpf_func_proto bpf_get_current_task_proto = {
391 .func = bpf_get_current_task,
393 .ret_type = RET_INTEGER,
396 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
398 struct bpf_array *array = container_of(map, struct bpf_array, map);
401 if (unlikely(in_interrupt()))
403 if (unlikely(idx >= array->map.max_entries))
406 cgrp = READ_ONCE(array->ptrs[idx]);
410 return task_under_cgroup_hierarchy(current, cgrp);
413 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
414 .func = bpf_current_task_under_cgroup,
416 .ret_type = RET_INTEGER,
417 .arg1_type = ARG_CONST_MAP_PTR,
418 .arg2_type = ARG_ANYTHING,
421 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
422 const void *, unsafe_ptr)
427 * The strncpy_from_unsafe() call will likely not fill the entire
428 * buffer, but that's okay in this circumstance as we're probing
429 * arbitrary memory anyway similar to bpf_probe_read() and might
430 * as well probe the stack. Thus, memory is explicitly cleared
431 * only in error case, so that improper users ignoring return
432 * code altogether don't copy garbage; otherwise length of string
433 * is returned that can be used for bpf_perf_event_output() et al.
435 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
436 if (unlikely(ret < 0))
437 memset(dst, 0, size);
442 static const struct bpf_func_proto bpf_probe_read_str_proto = {
443 .func = bpf_probe_read_str,
445 .ret_type = RET_INTEGER,
446 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
447 .arg2_type = ARG_CONST_SIZE,
448 .arg3_type = ARG_ANYTHING,
451 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
454 case BPF_FUNC_map_lookup_elem:
455 return &bpf_map_lookup_elem_proto;
456 case BPF_FUNC_map_update_elem:
457 return &bpf_map_update_elem_proto;
458 case BPF_FUNC_map_delete_elem:
459 return &bpf_map_delete_elem_proto;
460 case BPF_FUNC_probe_read:
461 return &bpf_probe_read_proto;
462 case BPF_FUNC_ktime_get_ns:
463 return &bpf_ktime_get_ns_proto;
464 case BPF_FUNC_tail_call:
465 return &bpf_tail_call_proto;
466 case BPF_FUNC_get_current_pid_tgid:
467 return &bpf_get_current_pid_tgid_proto;
468 case BPF_FUNC_get_current_task:
469 return &bpf_get_current_task_proto;
470 case BPF_FUNC_get_current_uid_gid:
471 return &bpf_get_current_uid_gid_proto;
472 case BPF_FUNC_get_current_comm:
473 return &bpf_get_current_comm_proto;
474 case BPF_FUNC_trace_printk:
475 return bpf_get_trace_printk_proto();
476 case BPF_FUNC_get_smp_processor_id:
477 return &bpf_get_smp_processor_id_proto;
478 case BPF_FUNC_get_numa_node_id:
479 return &bpf_get_numa_node_id_proto;
480 case BPF_FUNC_perf_event_read:
481 return &bpf_perf_event_read_proto;
482 case BPF_FUNC_probe_write_user:
483 return bpf_get_probe_write_proto();
484 case BPF_FUNC_current_task_under_cgroup:
485 return &bpf_current_task_under_cgroup_proto;
486 case BPF_FUNC_get_prandom_u32:
487 return &bpf_get_prandom_u32_proto;
488 case BPF_FUNC_probe_read_str:
489 return &bpf_probe_read_str_proto;
495 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
498 case BPF_FUNC_perf_event_output:
499 return &bpf_perf_event_output_proto;
500 case BPF_FUNC_get_stackid:
501 return &bpf_get_stackid_proto;
503 return tracing_func_proto(func_id);
507 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
508 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
509 struct bpf_insn_access_aux *info)
511 if (off < 0 || off >= sizeof(struct pt_regs))
513 if (type != BPF_READ)
518 * Assertion for 32 bit to make sure last 8 byte access
519 * (BPF_DW) to the last 4 byte member is disallowed.
521 if (off + size > sizeof(struct pt_regs))
527 const struct bpf_verifier_ops kprobe_prog_ops = {
528 .get_func_proto = kprobe_prog_func_proto,
529 .is_valid_access = kprobe_prog_is_valid_access,
532 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
533 u64, flags, void *, data, u64, size)
535 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
538 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
539 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
540 * from there and call the same bpf_perf_event_output() helper inline.
542 return ____bpf_perf_event_output(regs, map, flags, data, size);
545 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
546 .func = bpf_perf_event_output_tp,
548 .ret_type = RET_INTEGER,
549 .arg1_type = ARG_PTR_TO_CTX,
550 .arg2_type = ARG_CONST_MAP_PTR,
551 .arg3_type = ARG_ANYTHING,
552 .arg4_type = ARG_PTR_TO_MEM,
553 .arg5_type = ARG_CONST_SIZE,
556 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
559 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
562 * Same comment as in bpf_perf_event_output_tp(), only that this time
563 * the other helper's function body cannot be inlined due to being
564 * external, thus we need to call raw helper function.
566 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
570 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
571 .func = bpf_get_stackid_tp,
573 .ret_type = RET_INTEGER,
574 .arg1_type = ARG_PTR_TO_CTX,
575 .arg2_type = ARG_CONST_MAP_PTR,
576 .arg3_type = ARG_ANYTHING,
579 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
582 case BPF_FUNC_perf_event_output:
583 return &bpf_perf_event_output_proto_tp;
584 case BPF_FUNC_get_stackid:
585 return &bpf_get_stackid_proto_tp;
587 return tracing_func_proto(func_id);
591 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
592 struct bpf_insn_access_aux *info)
594 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
596 if (type != BPF_READ)
601 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
605 const struct bpf_verifier_ops tracepoint_prog_ops = {
606 .get_func_proto = tp_prog_func_proto,
607 .is_valid_access = tp_prog_is_valid_access,
610 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
611 struct bpf_insn_access_aux *info)
613 const int size_sp = FIELD_SIZEOF(struct bpf_perf_event_data,
616 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
618 if (type != BPF_READ)
624 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
625 bpf_ctx_record_field_size(info, size_sp);
626 if (!bpf_ctx_narrow_access_ok(off, size, size_sp))
630 if (size != sizeof(long))
637 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
638 const struct bpf_insn *si,
639 struct bpf_insn *insn_buf,
640 struct bpf_prog *prog, u32 *target_size)
642 struct bpf_insn *insn = insn_buf;
645 case offsetof(struct bpf_perf_event_data, sample_period):
646 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
647 data), si->dst_reg, si->src_reg,
648 offsetof(struct bpf_perf_event_data_kern, data));
649 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
650 bpf_target_off(struct perf_sample_data, period, 8,
654 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
655 regs), si->dst_reg, si->src_reg,
656 offsetof(struct bpf_perf_event_data_kern, regs));
657 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
662 return insn - insn_buf;
665 const struct bpf_verifier_ops perf_event_prog_ops = {
666 .get_func_proto = tp_prog_func_proto,
667 .is_valid_access = pe_prog_is_valid_access,
668 .convert_ctx_access = pe_prog_convert_ctx_access,