Loading...
1/* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
3 *
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.
7 */
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>
16#include <linux/kprobes.h>
17#include <linux/error-injection.h>
18
19#include "trace_probe.h"
20#include "trace.h"
21
22u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
23
24/**
25 * trace_call_bpf - invoke BPF program
26 * @call: tracepoint event
27 * @ctx: opaque context pointer
28 *
29 * kprobe handlers execute BPF programs via this helper.
30 * Can be used from static tracepoints in the future.
31 *
32 * Return: BPF programs always return an integer which is interpreted by
33 * kprobe handler as:
34 * 0 - return from kprobe (event is filtered out)
35 * 1 - store kprobe event into ring buffer
36 * Other values are reserved and currently alias to 1
37 */
38unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
39{
40 unsigned int ret;
41
42 if (in_nmi()) /* not supported yet */
43 return 1;
44
45 preempt_disable();
46
47 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
48 /*
49 * since some bpf program is already running on this cpu,
50 * don't call into another bpf program (same or different)
51 * and don't send kprobe event into ring-buffer,
52 * so return zero here
53 */
54 ret = 0;
55 goto out;
56 }
57
58 /*
59 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
60 * to all call sites, we did a bpf_prog_array_valid() there to check
61 * whether call->prog_array is empty or not, which is
62 * a heurisitc to speed up execution.
63 *
64 * If bpf_prog_array_valid() fetched prog_array was
65 * non-NULL, we go into trace_call_bpf() and do the actual
66 * proper rcu_dereference() under RCU lock.
67 * If it turns out that prog_array is NULL then, we bail out.
68 * For the opposite, if the bpf_prog_array_valid() fetched pointer
69 * was NULL, you'll skip the prog_array with the risk of missing
70 * out of events when it was updated in between this and the
71 * rcu_dereference() which is accepted risk.
72 */
73 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
74
75 out:
76 __this_cpu_dec(bpf_prog_active);
77 preempt_enable();
78
79 return ret;
80}
81EXPORT_SYMBOL_GPL(trace_call_bpf);
82
83#ifdef CONFIG_BPF_KPROBE_OVERRIDE
84BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
85{
86 regs_set_return_value(regs, rc);
87 override_function_with_return(regs);
88 return 0;
89}
90
91static const struct bpf_func_proto bpf_override_return_proto = {
92 .func = bpf_override_return,
93 .gpl_only = true,
94 .ret_type = RET_INTEGER,
95 .arg1_type = ARG_PTR_TO_CTX,
96 .arg2_type = ARG_ANYTHING,
97};
98#endif
99
100BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
101{
102 int ret;
103
104 ret = probe_kernel_read(dst, unsafe_ptr, size);
105 if (unlikely(ret < 0))
106 memset(dst, 0, size);
107
108 return ret;
109}
110
111static const struct bpf_func_proto bpf_probe_read_proto = {
112 .func = bpf_probe_read,
113 .gpl_only = true,
114 .ret_type = RET_INTEGER,
115 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
116 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
117 .arg3_type = ARG_ANYTHING,
118};
119
120BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
121 u32, size)
122{
123 /*
124 * Ensure we're in user context which is safe for the helper to
125 * run. This helper has no business in a kthread.
126 *
127 * access_ok() should prevent writing to non-user memory, but in
128 * some situations (nommu, temporary switch, etc) access_ok() does
129 * not provide enough validation, hence the check on KERNEL_DS.
130 */
131
132 if (unlikely(in_interrupt() ||
133 current->flags & (PF_KTHREAD | PF_EXITING)))
134 return -EPERM;
135 if (unlikely(uaccess_kernel()))
136 return -EPERM;
137 if (!access_ok(VERIFY_WRITE, unsafe_ptr, size))
138 return -EPERM;
139
140 return probe_kernel_write(unsafe_ptr, src, size);
141}
142
143static const struct bpf_func_proto bpf_probe_write_user_proto = {
144 .func = bpf_probe_write_user,
145 .gpl_only = true,
146 .ret_type = RET_INTEGER,
147 .arg1_type = ARG_ANYTHING,
148 .arg2_type = ARG_PTR_TO_MEM,
149 .arg3_type = ARG_CONST_SIZE,
150};
151
152static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
153{
154 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
155 current->comm, task_pid_nr(current));
156
157 return &bpf_probe_write_user_proto;
158}
159
160/*
161 * Only limited trace_printk() conversion specifiers allowed:
162 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
163 */
164BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
165 u64, arg2, u64, arg3)
166{
167 bool str_seen = false;
168 int mod[3] = {};
169 int fmt_cnt = 0;
170 u64 unsafe_addr;
171 char buf[64];
172 int i;
173
174 /*
175 * bpf_check()->check_func_arg()->check_stack_boundary()
176 * guarantees that fmt points to bpf program stack,
177 * fmt_size bytes of it were initialized and fmt_size > 0
178 */
179 if (fmt[--fmt_size] != 0)
180 return -EINVAL;
181
182 /* check format string for allowed specifiers */
183 for (i = 0; i < fmt_size; i++) {
184 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
185 return -EINVAL;
186
187 if (fmt[i] != '%')
188 continue;
189
190 if (fmt_cnt >= 3)
191 return -EINVAL;
192
193 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
194 i++;
195 if (fmt[i] == 'l') {
196 mod[fmt_cnt]++;
197 i++;
198 } else if (fmt[i] == 'p' || fmt[i] == 's') {
199 mod[fmt_cnt]++;
200 i++;
201 if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
202 return -EINVAL;
203 fmt_cnt++;
204 if (fmt[i - 1] == 's') {
205 if (str_seen)
206 /* allow only one '%s' per fmt string */
207 return -EINVAL;
208 str_seen = true;
209
210 switch (fmt_cnt) {
211 case 1:
212 unsafe_addr = arg1;
213 arg1 = (long) buf;
214 break;
215 case 2:
216 unsafe_addr = arg2;
217 arg2 = (long) buf;
218 break;
219 case 3:
220 unsafe_addr = arg3;
221 arg3 = (long) buf;
222 break;
223 }
224 buf[0] = 0;
225 strncpy_from_unsafe(buf,
226 (void *) (long) unsafe_addr,
227 sizeof(buf));
228 }
229 continue;
230 }
231
232 if (fmt[i] == 'l') {
233 mod[fmt_cnt]++;
234 i++;
235 }
236
237 if (fmt[i] != 'i' && fmt[i] != 'd' &&
238 fmt[i] != 'u' && fmt[i] != 'x')
239 return -EINVAL;
240 fmt_cnt++;
241 }
242
243/* Horrid workaround for getting va_list handling working with different
244 * argument type combinations generically for 32 and 64 bit archs.
245 */
246#define __BPF_TP_EMIT() __BPF_ARG3_TP()
247#define __BPF_TP(...) \
248 __trace_printk(0 /* Fake ip */, \
249 fmt, ##__VA_ARGS__)
250
251#define __BPF_ARG1_TP(...) \
252 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
253 ? __BPF_TP(arg1, ##__VA_ARGS__) \
254 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
255 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
256 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
257
258#define __BPF_ARG2_TP(...) \
259 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
260 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
261 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
262 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
263 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
264
265#define __BPF_ARG3_TP(...) \
266 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
267 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
268 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
269 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
270 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
271
272 return __BPF_TP_EMIT();
273}
274
275static const struct bpf_func_proto bpf_trace_printk_proto = {
276 .func = bpf_trace_printk,
277 .gpl_only = true,
278 .ret_type = RET_INTEGER,
279 .arg1_type = ARG_PTR_TO_MEM,
280 .arg2_type = ARG_CONST_SIZE,
281};
282
283const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
284{
285 /*
286 * this program might be calling bpf_trace_printk,
287 * so allocate per-cpu printk buffers
288 */
289 trace_printk_init_buffers();
290
291 return &bpf_trace_printk_proto;
292}
293
294static __always_inline int
295get_map_perf_counter(struct bpf_map *map, u64 flags,
296 u64 *value, u64 *enabled, u64 *running)
297{
298 struct bpf_array *array = container_of(map, struct bpf_array, map);
299 unsigned int cpu = smp_processor_id();
300 u64 index = flags & BPF_F_INDEX_MASK;
301 struct bpf_event_entry *ee;
302
303 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
304 return -EINVAL;
305 if (index == BPF_F_CURRENT_CPU)
306 index = cpu;
307 if (unlikely(index >= array->map.max_entries))
308 return -E2BIG;
309
310 ee = READ_ONCE(array->ptrs[index]);
311 if (!ee)
312 return -ENOENT;
313
314 return perf_event_read_local(ee->event, value, enabled, running);
315}
316
317BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
318{
319 u64 value = 0;
320 int err;
321
322 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
323 /*
324 * this api is ugly since we miss [-22..-2] range of valid
325 * counter values, but that's uapi
326 */
327 if (err)
328 return err;
329 return value;
330}
331
332static const struct bpf_func_proto bpf_perf_event_read_proto = {
333 .func = bpf_perf_event_read,
334 .gpl_only = true,
335 .ret_type = RET_INTEGER,
336 .arg1_type = ARG_CONST_MAP_PTR,
337 .arg2_type = ARG_ANYTHING,
338};
339
340BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
341 struct bpf_perf_event_value *, buf, u32, size)
342{
343 int err = -EINVAL;
344
345 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
346 goto clear;
347 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
348 &buf->running);
349 if (unlikely(err))
350 goto clear;
351 return 0;
352clear:
353 memset(buf, 0, size);
354 return err;
355}
356
357static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
358 .func = bpf_perf_event_read_value,
359 .gpl_only = true,
360 .ret_type = RET_INTEGER,
361 .arg1_type = ARG_CONST_MAP_PTR,
362 .arg2_type = ARG_ANYTHING,
363 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
364 .arg4_type = ARG_CONST_SIZE,
365};
366
367static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
368
369static __always_inline u64
370__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
371 u64 flags, struct perf_sample_data *sd)
372{
373 struct bpf_array *array = container_of(map, struct bpf_array, map);
374 unsigned int cpu = smp_processor_id();
375 u64 index = flags & BPF_F_INDEX_MASK;
376 struct bpf_event_entry *ee;
377 struct perf_event *event;
378
379 if (index == BPF_F_CURRENT_CPU)
380 index = cpu;
381 if (unlikely(index >= array->map.max_entries))
382 return -E2BIG;
383
384 ee = READ_ONCE(array->ptrs[index]);
385 if (!ee)
386 return -ENOENT;
387
388 event = ee->event;
389 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
390 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
391 return -EINVAL;
392
393 if (unlikely(event->oncpu != cpu))
394 return -EOPNOTSUPP;
395
396 perf_event_output(event, sd, regs);
397 return 0;
398}
399
400BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
401 u64, flags, void *, data, u64, size)
402{
403 struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
404 struct perf_raw_record raw = {
405 .frag = {
406 .size = size,
407 .data = data,
408 },
409 };
410
411 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
412 return -EINVAL;
413
414 perf_sample_data_init(sd, 0, 0);
415 sd->raw = &raw;
416
417 return __bpf_perf_event_output(regs, map, flags, sd);
418}
419
420static const struct bpf_func_proto bpf_perf_event_output_proto = {
421 .func = bpf_perf_event_output,
422 .gpl_only = true,
423 .ret_type = RET_INTEGER,
424 .arg1_type = ARG_PTR_TO_CTX,
425 .arg2_type = ARG_CONST_MAP_PTR,
426 .arg3_type = ARG_ANYTHING,
427 .arg4_type = ARG_PTR_TO_MEM,
428 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
429};
430
431static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
432static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
433
434u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
435 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
436{
437 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
438 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
439 struct perf_raw_frag frag = {
440 .copy = ctx_copy,
441 .size = ctx_size,
442 .data = ctx,
443 };
444 struct perf_raw_record raw = {
445 .frag = {
446 {
447 .next = ctx_size ? &frag : NULL,
448 },
449 .size = meta_size,
450 .data = meta,
451 },
452 };
453
454 perf_fetch_caller_regs(regs);
455 perf_sample_data_init(sd, 0, 0);
456 sd->raw = &raw;
457
458 return __bpf_perf_event_output(regs, map, flags, sd);
459}
460
461BPF_CALL_0(bpf_get_current_task)
462{
463 return (long) current;
464}
465
466static const struct bpf_func_proto bpf_get_current_task_proto = {
467 .func = bpf_get_current_task,
468 .gpl_only = true,
469 .ret_type = RET_INTEGER,
470};
471
472BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
473{
474 struct bpf_array *array = container_of(map, struct bpf_array, map);
475 struct cgroup *cgrp;
476
477 if (unlikely(in_interrupt()))
478 return -EINVAL;
479 if (unlikely(idx >= array->map.max_entries))
480 return -E2BIG;
481
482 cgrp = READ_ONCE(array->ptrs[idx]);
483 if (unlikely(!cgrp))
484 return -EAGAIN;
485
486 return task_under_cgroup_hierarchy(current, cgrp);
487}
488
489static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
490 .func = bpf_current_task_under_cgroup,
491 .gpl_only = false,
492 .ret_type = RET_INTEGER,
493 .arg1_type = ARG_CONST_MAP_PTR,
494 .arg2_type = ARG_ANYTHING,
495};
496
497BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
498 const void *, unsafe_ptr)
499{
500 int ret;
501
502 /*
503 * The strncpy_from_unsafe() call will likely not fill the entire
504 * buffer, but that's okay in this circumstance as we're probing
505 * arbitrary memory anyway similar to bpf_probe_read() and might
506 * as well probe the stack. Thus, memory is explicitly cleared
507 * only in error case, so that improper users ignoring return
508 * code altogether don't copy garbage; otherwise length of string
509 * is returned that can be used for bpf_perf_event_output() et al.
510 */
511 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
512 if (unlikely(ret < 0))
513 memset(dst, 0, size);
514
515 return ret;
516}
517
518static const struct bpf_func_proto bpf_probe_read_str_proto = {
519 .func = bpf_probe_read_str,
520 .gpl_only = true,
521 .ret_type = RET_INTEGER,
522 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
523 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
524 .arg3_type = ARG_ANYTHING,
525};
526
527static const struct bpf_func_proto *
528tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
529{
530 switch (func_id) {
531 case BPF_FUNC_map_lookup_elem:
532 return &bpf_map_lookup_elem_proto;
533 case BPF_FUNC_map_update_elem:
534 return &bpf_map_update_elem_proto;
535 case BPF_FUNC_map_delete_elem:
536 return &bpf_map_delete_elem_proto;
537 case BPF_FUNC_probe_read:
538 return &bpf_probe_read_proto;
539 case BPF_FUNC_ktime_get_ns:
540 return &bpf_ktime_get_ns_proto;
541 case BPF_FUNC_tail_call:
542 return &bpf_tail_call_proto;
543 case BPF_FUNC_get_current_pid_tgid:
544 return &bpf_get_current_pid_tgid_proto;
545 case BPF_FUNC_get_current_task:
546 return &bpf_get_current_task_proto;
547 case BPF_FUNC_get_current_uid_gid:
548 return &bpf_get_current_uid_gid_proto;
549 case BPF_FUNC_get_current_comm:
550 return &bpf_get_current_comm_proto;
551 case BPF_FUNC_trace_printk:
552 return bpf_get_trace_printk_proto();
553 case BPF_FUNC_get_smp_processor_id:
554 return &bpf_get_smp_processor_id_proto;
555 case BPF_FUNC_get_numa_node_id:
556 return &bpf_get_numa_node_id_proto;
557 case BPF_FUNC_perf_event_read:
558 return &bpf_perf_event_read_proto;
559 case BPF_FUNC_probe_write_user:
560 return bpf_get_probe_write_proto();
561 case BPF_FUNC_current_task_under_cgroup:
562 return &bpf_current_task_under_cgroup_proto;
563 case BPF_FUNC_get_prandom_u32:
564 return &bpf_get_prandom_u32_proto;
565 case BPF_FUNC_probe_read_str:
566 return &bpf_probe_read_str_proto;
567 default:
568 return NULL;
569 }
570}
571
572static const struct bpf_func_proto *
573kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
574{
575 switch (func_id) {
576 case BPF_FUNC_perf_event_output:
577 return &bpf_perf_event_output_proto;
578 case BPF_FUNC_get_stackid:
579 return &bpf_get_stackid_proto;
580 case BPF_FUNC_perf_event_read_value:
581 return &bpf_perf_event_read_value_proto;
582#ifdef CONFIG_BPF_KPROBE_OVERRIDE
583 case BPF_FUNC_override_return:
584 return &bpf_override_return_proto;
585#endif
586 default:
587 return tracing_func_proto(func_id, prog);
588 }
589}
590
591/* bpf+kprobe programs can access fields of 'struct pt_regs' */
592static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
593 const struct bpf_prog *prog,
594 struct bpf_insn_access_aux *info)
595{
596 if (off < 0 || off >= sizeof(struct pt_regs))
597 return false;
598 if (type != BPF_READ)
599 return false;
600 if (off % size != 0)
601 return false;
602 /*
603 * Assertion for 32 bit to make sure last 8 byte access
604 * (BPF_DW) to the last 4 byte member is disallowed.
605 */
606 if (off + size > sizeof(struct pt_regs))
607 return false;
608
609 return true;
610}
611
612const struct bpf_verifier_ops kprobe_verifier_ops = {
613 .get_func_proto = kprobe_prog_func_proto,
614 .is_valid_access = kprobe_prog_is_valid_access,
615};
616
617const struct bpf_prog_ops kprobe_prog_ops = {
618};
619
620BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
621 u64, flags, void *, data, u64, size)
622{
623 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
624
625 /*
626 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
627 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
628 * from there and call the same bpf_perf_event_output() helper inline.
629 */
630 return ____bpf_perf_event_output(regs, map, flags, data, size);
631}
632
633static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
634 .func = bpf_perf_event_output_tp,
635 .gpl_only = true,
636 .ret_type = RET_INTEGER,
637 .arg1_type = ARG_PTR_TO_CTX,
638 .arg2_type = ARG_CONST_MAP_PTR,
639 .arg3_type = ARG_ANYTHING,
640 .arg4_type = ARG_PTR_TO_MEM,
641 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
642};
643
644BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
645 u64, flags)
646{
647 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
648
649 /*
650 * Same comment as in bpf_perf_event_output_tp(), only that this time
651 * the other helper's function body cannot be inlined due to being
652 * external, thus we need to call raw helper function.
653 */
654 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
655 flags, 0, 0);
656}
657
658static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
659 .func = bpf_get_stackid_tp,
660 .gpl_only = true,
661 .ret_type = RET_INTEGER,
662 .arg1_type = ARG_PTR_TO_CTX,
663 .arg2_type = ARG_CONST_MAP_PTR,
664 .arg3_type = ARG_ANYTHING,
665};
666
667static const struct bpf_func_proto *
668tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
669{
670 switch (func_id) {
671 case BPF_FUNC_perf_event_output:
672 return &bpf_perf_event_output_proto_tp;
673 case BPF_FUNC_get_stackid:
674 return &bpf_get_stackid_proto_tp;
675 default:
676 return tracing_func_proto(func_id, prog);
677 }
678}
679
680static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
681 const struct bpf_prog *prog,
682 struct bpf_insn_access_aux *info)
683{
684 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
685 return false;
686 if (type != BPF_READ)
687 return false;
688 if (off % size != 0)
689 return false;
690
691 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
692 return true;
693}
694
695const struct bpf_verifier_ops tracepoint_verifier_ops = {
696 .get_func_proto = tp_prog_func_proto,
697 .is_valid_access = tp_prog_is_valid_access,
698};
699
700const struct bpf_prog_ops tracepoint_prog_ops = {
701};
702
703BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
704 struct bpf_perf_event_value *, buf, u32, size)
705{
706 int err = -EINVAL;
707
708 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
709 goto clear;
710 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
711 &buf->running);
712 if (unlikely(err))
713 goto clear;
714 return 0;
715clear:
716 memset(buf, 0, size);
717 return err;
718}
719
720static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
721 .func = bpf_perf_prog_read_value,
722 .gpl_only = true,
723 .ret_type = RET_INTEGER,
724 .arg1_type = ARG_PTR_TO_CTX,
725 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
726 .arg3_type = ARG_CONST_SIZE,
727};
728
729static const struct bpf_func_proto *
730pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
731{
732 switch (func_id) {
733 case BPF_FUNC_perf_event_output:
734 return &bpf_perf_event_output_proto_tp;
735 case BPF_FUNC_get_stackid:
736 return &bpf_get_stackid_proto_tp;
737 case BPF_FUNC_perf_prog_read_value:
738 return &bpf_perf_prog_read_value_proto;
739 default:
740 return tracing_func_proto(func_id, prog);
741 }
742}
743
744/*
745 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
746 * to avoid potential recursive reuse issue when/if tracepoints are added
747 * inside bpf_*_event_output and/or bpf_get_stack_id
748 */
749static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
750BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
751 struct bpf_map *, map, u64, flags, void *, data, u64, size)
752{
753 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
754
755 perf_fetch_caller_regs(regs);
756 return ____bpf_perf_event_output(regs, map, flags, data, size);
757}
758
759static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
760 .func = bpf_perf_event_output_raw_tp,
761 .gpl_only = true,
762 .ret_type = RET_INTEGER,
763 .arg1_type = ARG_PTR_TO_CTX,
764 .arg2_type = ARG_CONST_MAP_PTR,
765 .arg3_type = ARG_ANYTHING,
766 .arg4_type = ARG_PTR_TO_MEM,
767 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
768};
769
770BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
771 struct bpf_map *, map, u64, flags)
772{
773 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
774
775 perf_fetch_caller_regs(regs);
776 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
777 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
778 flags, 0, 0);
779}
780
781static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
782 .func = bpf_get_stackid_raw_tp,
783 .gpl_only = true,
784 .ret_type = RET_INTEGER,
785 .arg1_type = ARG_PTR_TO_CTX,
786 .arg2_type = ARG_CONST_MAP_PTR,
787 .arg3_type = ARG_ANYTHING,
788};
789
790static const struct bpf_func_proto *
791raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
792{
793 switch (func_id) {
794 case BPF_FUNC_perf_event_output:
795 return &bpf_perf_event_output_proto_raw_tp;
796 case BPF_FUNC_get_stackid:
797 return &bpf_get_stackid_proto_raw_tp;
798 default:
799 return tracing_func_proto(func_id, prog);
800 }
801}
802
803static bool raw_tp_prog_is_valid_access(int off, int size,
804 enum bpf_access_type type,
805 const struct bpf_prog *prog,
806 struct bpf_insn_access_aux *info)
807{
808 /* largest tracepoint in the kernel has 12 args */
809 if (off < 0 || off >= sizeof(__u64) * 12)
810 return false;
811 if (type != BPF_READ)
812 return false;
813 if (off % size != 0)
814 return false;
815 return true;
816}
817
818const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
819 .get_func_proto = raw_tp_prog_func_proto,
820 .is_valid_access = raw_tp_prog_is_valid_access,
821};
822
823const struct bpf_prog_ops raw_tracepoint_prog_ops = {
824};
825
826static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
827 const struct bpf_prog *prog,
828 struct bpf_insn_access_aux *info)
829{
830 const int size_u64 = sizeof(u64);
831
832 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
833 return false;
834 if (type != BPF_READ)
835 return false;
836 if (off % size != 0)
837 return false;
838
839 switch (off) {
840 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
841 bpf_ctx_record_field_size(info, size_u64);
842 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
843 return false;
844 break;
845 case bpf_ctx_range(struct bpf_perf_event_data, addr):
846 bpf_ctx_record_field_size(info, size_u64);
847 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
848 return false;
849 break;
850 default:
851 if (size != sizeof(long))
852 return false;
853 }
854
855 return true;
856}
857
858static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
859 const struct bpf_insn *si,
860 struct bpf_insn *insn_buf,
861 struct bpf_prog *prog, u32 *target_size)
862{
863 struct bpf_insn *insn = insn_buf;
864
865 switch (si->off) {
866 case offsetof(struct bpf_perf_event_data, sample_period):
867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
868 data), si->dst_reg, si->src_reg,
869 offsetof(struct bpf_perf_event_data_kern, data));
870 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
871 bpf_target_off(struct perf_sample_data, period, 8,
872 target_size));
873 break;
874 case offsetof(struct bpf_perf_event_data, addr):
875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
876 data), si->dst_reg, si->src_reg,
877 offsetof(struct bpf_perf_event_data_kern, data));
878 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
879 bpf_target_off(struct perf_sample_data, addr, 8,
880 target_size));
881 break;
882 default:
883 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
884 regs), si->dst_reg, si->src_reg,
885 offsetof(struct bpf_perf_event_data_kern, regs));
886 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
887 si->off);
888 break;
889 }
890
891 return insn - insn_buf;
892}
893
894const struct bpf_verifier_ops perf_event_verifier_ops = {
895 .get_func_proto = pe_prog_func_proto,
896 .is_valid_access = pe_prog_is_valid_access,
897 .convert_ctx_access = pe_prog_convert_ctx_access,
898};
899
900const struct bpf_prog_ops perf_event_prog_ops = {
901};
902
903static DEFINE_MUTEX(bpf_event_mutex);
904
905#define BPF_TRACE_MAX_PROGS 64
906
907int perf_event_attach_bpf_prog(struct perf_event *event,
908 struct bpf_prog *prog)
909{
910 struct bpf_prog_array __rcu *old_array;
911 struct bpf_prog_array *new_array;
912 int ret = -EEXIST;
913
914 /*
915 * Kprobe override only works if they are on the function entry,
916 * and only if they are on the opt-in list.
917 */
918 if (prog->kprobe_override &&
919 (!trace_kprobe_on_func_entry(event->tp_event) ||
920 !trace_kprobe_error_injectable(event->tp_event)))
921 return -EINVAL;
922
923 mutex_lock(&bpf_event_mutex);
924
925 if (event->prog)
926 goto unlock;
927
928 old_array = event->tp_event->prog_array;
929 if (old_array &&
930 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
931 ret = -E2BIG;
932 goto unlock;
933 }
934
935 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
936 if (ret < 0)
937 goto unlock;
938
939 /* set the new array to event->tp_event and set event->prog */
940 event->prog = prog;
941 rcu_assign_pointer(event->tp_event->prog_array, new_array);
942 bpf_prog_array_free(old_array);
943
944unlock:
945 mutex_unlock(&bpf_event_mutex);
946 return ret;
947}
948
949void perf_event_detach_bpf_prog(struct perf_event *event)
950{
951 struct bpf_prog_array __rcu *old_array;
952 struct bpf_prog_array *new_array;
953 int ret;
954
955 mutex_lock(&bpf_event_mutex);
956
957 if (!event->prog)
958 goto unlock;
959
960 old_array = event->tp_event->prog_array;
961 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
962 if (ret < 0) {
963 bpf_prog_array_delete_safe(old_array, event->prog);
964 } else {
965 rcu_assign_pointer(event->tp_event->prog_array, new_array);
966 bpf_prog_array_free(old_array);
967 }
968
969 bpf_prog_put(event->prog);
970 event->prog = NULL;
971
972unlock:
973 mutex_unlock(&bpf_event_mutex);
974}
975
976int perf_event_query_prog_array(struct perf_event *event, void __user *info)
977{
978 struct perf_event_query_bpf __user *uquery = info;
979 struct perf_event_query_bpf query = {};
980 u32 *ids, prog_cnt, ids_len;
981 int ret;
982
983 if (!capable(CAP_SYS_ADMIN))
984 return -EPERM;
985 if (event->attr.type != PERF_TYPE_TRACEPOINT)
986 return -EINVAL;
987 if (copy_from_user(&query, uquery, sizeof(query)))
988 return -EFAULT;
989
990 ids_len = query.ids_len;
991 if (ids_len > BPF_TRACE_MAX_PROGS)
992 return -E2BIG;
993 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
994 if (!ids)
995 return -ENOMEM;
996 /*
997 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
998 * is required when user only wants to check for uquery->prog_cnt.
999 * There is no need to check for it since the case is handled
1000 * gracefully in bpf_prog_array_copy_info.
1001 */
1002
1003 mutex_lock(&bpf_event_mutex);
1004 ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1005 ids,
1006 ids_len,
1007 &prog_cnt);
1008 mutex_unlock(&bpf_event_mutex);
1009
1010 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1011 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1012 ret = -EFAULT;
1013
1014 kfree(ids);
1015 return ret;
1016}
1017
1018extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1019extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1020
1021struct bpf_raw_event_map *bpf_find_raw_tracepoint(const char *name)
1022{
1023 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1024
1025 for (; btp < __stop__bpf_raw_tp; btp++) {
1026 if (!strcmp(btp->tp->name, name))
1027 return btp;
1028 }
1029 return NULL;
1030}
1031
1032static __always_inline
1033void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1034{
1035 rcu_read_lock();
1036 preempt_disable();
1037 (void) BPF_PROG_RUN(prog, args);
1038 preempt_enable();
1039 rcu_read_unlock();
1040}
1041
1042#define UNPACK(...) __VA_ARGS__
1043#define REPEAT_1(FN, DL, X, ...) FN(X)
1044#define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1045#define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1046#define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1047#define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1048#define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1049#define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1050#define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1051#define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1052#define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1053#define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1054#define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1055#define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1056
1057#define SARG(X) u64 arg##X
1058#define COPY(X) args[X] = arg##X
1059
1060#define __DL_COM (,)
1061#define __DL_SEM (;)
1062
1063#define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1064
1065#define BPF_TRACE_DEFN_x(x) \
1066 void bpf_trace_run##x(struct bpf_prog *prog, \
1067 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1068 { \
1069 u64 args[x]; \
1070 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1071 __bpf_trace_run(prog, args); \
1072 } \
1073 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1074BPF_TRACE_DEFN_x(1);
1075BPF_TRACE_DEFN_x(2);
1076BPF_TRACE_DEFN_x(3);
1077BPF_TRACE_DEFN_x(4);
1078BPF_TRACE_DEFN_x(5);
1079BPF_TRACE_DEFN_x(6);
1080BPF_TRACE_DEFN_x(7);
1081BPF_TRACE_DEFN_x(8);
1082BPF_TRACE_DEFN_x(9);
1083BPF_TRACE_DEFN_x(10);
1084BPF_TRACE_DEFN_x(11);
1085BPF_TRACE_DEFN_x(12);
1086
1087static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1088{
1089 struct tracepoint *tp = btp->tp;
1090
1091 /*
1092 * check that program doesn't access arguments beyond what's
1093 * available in this tracepoint
1094 */
1095 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1096 return -EINVAL;
1097
1098 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1099}
1100
1101int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1102{
1103 int err;
1104
1105 mutex_lock(&bpf_event_mutex);
1106 err = __bpf_probe_register(btp, prog);
1107 mutex_unlock(&bpf_event_mutex);
1108 return err;
1109}
1110
1111int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1112{
1113 int err;
1114
1115 mutex_lock(&bpf_event_mutex);
1116 err = tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1117 mutex_unlock(&bpf_event_mutex);
1118 return err;
1119}
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
4 */
5#include <linux/kernel.h>
6#include <linux/types.h>
7#include <linux/slab.h>
8#include <linux/bpf.h>
9#include <linux/bpf_perf_event.h>
10#include <linux/btf.h>
11#include <linux/filter.h>
12#include <linux/uaccess.h>
13#include <linux/ctype.h>
14#include <linux/kprobes.h>
15#include <linux/spinlock.h>
16#include <linux/syscalls.h>
17#include <linux/error-injection.h>
18#include <linux/btf_ids.h>
19#include <linux/bpf_lsm.h>
20
21#include <net/bpf_sk_storage.h>
22
23#include <uapi/linux/bpf.h>
24#include <uapi/linux/btf.h>
25
26#include <asm/tlb.h>
27
28#include "trace_probe.h"
29#include "trace.h"
30
31#define CREATE_TRACE_POINTS
32#include "bpf_trace.h"
33
34#define bpf_event_rcu_dereference(p) \
35 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
36
37#ifdef CONFIG_MODULES
38struct bpf_trace_module {
39 struct module *module;
40 struct list_head list;
41};
42
43static LIST_HEAD(bpf_trace_modules);
44static DEFINE_MUTEX(bpf_module_mutex);
45
46static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
47{
48 struct bpf_raw_event_map *btp, *ret = NULL;
49 struct bpf_trace_module *btm;
50 unsigned int i;
51
52 mutex_lock(&bpf_module_mutex);
53 list_for_each_entry(btm, &bpf_trace_modules, list) {
54 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
55 btp = &btm->module->bpf_raw_events[i];
56 if (!strcmp(btp->tp->name, name)) {
57 if (try_module_get(btm->module))
58 ret = btp;
59 goto out;
60 }
61 }
62 }
63out:
64 mutex_unlock(&bpf_module_mutex);
65 return ret;
66}
67#else
68static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
69{
70 return NULL;
71}
72#endif /* CONFIG_MODULES */
73
74u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
75u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
76
77static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
78 u64 flags, const struct btf **btf,
79 s32 *btf_id);
80
81/**
82 * trace_call_bpf - invoke BPF program
83 * @call: tracepoint event
84 * @ctx: opaque context pointer
85 *
86 * kprobe handlers execute BPF programs via this helper.
87 * Can be used from static tracepoints in the future.
88 *
89 * Return: BPF programs always return an integer which is interpreted by
90 * kprobe handler as:
91 * 0 - return from kprobe (event is filtered out)
92 * 1 - store kprobe event into ring buffer
93 * Other values are reserved and currently alias to 1
94 */
95unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
96{
97 unsigned int ret;
98
99 cant_sleep();
100
101 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
102 /*
103 * since some bpf program is already running on this cpu,
104 * don't call into another bpf program (same or different)
105 * and don't send kprobe event into ring-buffer,
106 * so return zero here
107 */
108 ret = 0;
109 goto out;
110 }
111
112 /*
113 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
114 * to all call sites, we did a bpf_prog_array_valid() there to check
115 * whether call->prog_array is empty or not, which is
116 * a heuristic to speed up execution.
117 *
118 * If bpf_prog_array_valid() fetched prog_array was
119 * non-NULL, we go into trace_call_bpf() and do the actual
120 * proper rcu_dereference() under RCU lock.
121 * If it turns out that prog_array is NULL then, we bail out.
122 * For the opposite, if the bpf_prog_array_valid() fetched pointer
123 * was NULL, you'll skip the prog_array with the risk of missing
124 * out of events when it was updated in between this and the
125 * rcu_dereference() which is accepted risk.
126 */
127 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
128
129 out:
130 __this_cpu_dec(bpf_prog_active);
131
132 return ret;
133}
134
135#ifdef CONFIG_BPF_KPROBE_OVERRIDE
136BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
137{
138 regs_set_return_value(regs, rc);
139 override_function_with_return(regs);
140 return 0;
141}
142
143static const struct bpf_func_proto bpf_override_return_proto = {
144 .func = bpf_override_return,
145 .gpl_only = true,
146 .ret_type = RET_INTEGER,
147 .arg1_type = ARG_PTR_TO_CTX,
148 .arg2_type = ARG_ANYTHING,
149};
150#endif
151
152static __always_inline int
153bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
154{
155 int ret;
156
157 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
158 if (unlikely(ret < 0))
159 memset(dst, 0, size);
160 return ret;
161}
162
163BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
164 const void __user *, unsafe_ptr)
165{
166 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
167}
168
169const struct bpf_func_proto bpf_probe_read_user_proto = {
170 .func = bpf_probe_read_user,
171 .gpl_only = true,
172 .ret_type = RET_INTEGER,
173 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
174 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
175 .arg3_type = ARG_ANYTHING,
176};
177
178static __always_inline int
179bpf_probe_read_user_str_common(void *dst, u32 size,
180 const void __user *unsafe_ptr)
181{
182 int ret;
183
184 /*
185 * NB: We rely on strncpy_from_user() not copying junk past the NUL
186 * terminator into `dst`.
187 *
188 * strncpy_from_user() does long-sized strides in the fast path. If the
189 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
190 * then there could be junk after the NUL in `dst`. If user takes `dst`
191 * and keys a hash map with it, then semantically identical strings can
192 * occupy multiple entries in the map.
193 */
194 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
195 if (unlikely(ret < 0))
196 memset(dst, 0, size);
197 return ret;
198}
199
200BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
201 const void __user *, unsafe_ptr)
202{
203 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
204}
205
206const struct bpf_func_proto bpf_probe_read_user_str_proto = {
207 .func = bpf_probe_read_user_str,
208 .gpl_only = true,
209 .ret_type = RET_INTEGER,
210 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
211 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
212 .arg3_type = ARG_ANYTHING,
213};
214
215static __always_inline int
216bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
217{
218 int ret;
219
220 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
221 if (unlikely(ret < 0))
222 memset(dst, 0, size);
223 return ret;
224}
225
226BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
227 const void *, unsafe_ptr)
228{
229 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
230}
231
232const struct bpf_func_proto bpf_probe_read_kernel_proto = {
233 .func = bpf_probe_read_kernel,
234 .gpl_only = true,
235 .ret_type = RET_INTEGER,
236 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
237 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
238 .arg3_type = ARG_ANYTHING,
239};
240
241static __always_inline int
242bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
243{
244 int ret;
245
246 /*
247 * The strncpy_from_kernel_nofault() call will likely not fill the
248 * entire buffer, but that's okay in this circumstance as we're probing
249 * arbitrary memory anyway similar to bpf_probe_read_*() and might
250 * as well probe the stack. Thus, memory is explicitly cleared
251 * only in error case, so that improper users ignoring return
252 * code altogether don't copy garbage; otherwise length of string
253 * is returned that can be used for bpf_perf_event_output() et al.
254 */
255 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
256 if (unlikely(ret < 0))
257 memset(dst, 0, size);
258 return ret;
259}
260
261BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
262 const void *, unsafe_ptr)
263{
264 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
265}
266
267const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
268 .func = bpf_probe_read_kernel_str,
269 .gpl_only = true,
270 .ret_type = RET_INTEGER,
271 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
272 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
273 .arg3_type = ARG_ANYTHING,
274};
275
276#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
277BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
278 const void *, unsafe_ptr)
279{
280 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
281 return bpf_probe_read_user_common(dst, size,
282 (__force void __user *)unsafe_ptr);
283 }
284 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
285}
286
287static const struct bpf_func_proto bpf_probe_read_compat_proto = {
288 .func = bpf_probe_read_compat,
289 .gpl_only = true,
290 .ret_type = RET_INTEGER,
291 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
292 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
293 .arg3_type = ARG_ANYTHING,
294};
295
296BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
297 const void *, unsafe_ptr)
298{
299 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
300 return bpf_probe_read_user_str_common(dst, size,
301 (__force void __user *)unsafe_ptr);
302 }
303 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
304}
305
306static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
307 .func = bpf_probe_read_compat_str,
308 .gpl_only = true,
309 .ret_type = RET_INTEGER,
310 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
311 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
312 .arg3_type = ARG_ANYTHING,
313};
314#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
315
316BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
317 u32, size)
318{
319 /*
320 * Ensure we're in user context which is safe for the helper to
321 * run. This helper has no business in a kthread.
322 *
323 * access_ok() should prevent writing to non-user memory, but in
324 * some situations (nommu, temporary switch, etc) access_ok() does
325 * not provide enough validation, hence the check on KERNEL_DS.
326 *
327 * nmi_uaccess_okay() ensures the probe is not run in an interim
328 * state, when the task or mm are switched. This is specifically
329 * required to prevent the use of temporary mm.
330 */
331
332 if (unlikely(in_interrupt() ||
333 current->flags & (PF_KTHREAD | PF_EXITING)))
334 return -EPERM;
335 if (unlikely(uaccess_kernel()))
336 return -EPERM;
337 if (unlikely(!nmi_uaccess_okay()))
338 return -EPERM;
339
340 return copy_to_user_nofault(unsafe_ptr, src, size);
341}
342
343static const struct bpf_func_proto bpf_probe_write_user_proto = {
344 .func = bpf_probe_write_user,
345 .gpl_only = true,
346 .ret_type = RET_INTEGER,
347 .arg1_type = ARG_ANYTHING,
348 .arg2_type = ARG_PTR_TO_MEM,
349 .arg3_type = ARG_CONST_SIZE,
350};
351
352static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
353{
354 if (!capable(CAP_SYS_ADMIN))
355 return NULL;
356
357 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
358 current->comm, task_pid_nr(current));
359
360 return &bpf_probe_write_user_proto;
361}
362
363static DEFINE_RAW_SPINLOCK(trace_printk_lock);
364
365#define MAX_TRACE_PRINTK_VARARGS 3
366#define BPF_TRACE_PRINTK_SIZE 1024
367
368BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
369 u64, arg2, u64, arg3)
370{
371 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
372 u32 *bin_args;
373 static char buf[BPF_TRACE_PRINTK_SIZE];
374 unsigned long flags;
375 int ret;
376
377 ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
378 MAX_TRACE_PRINTK_VARARGS);
379 if (ret < 0)
380 return ret;
381
382 raw_spin_lock_irqsave(&trace_printk_lock, flags);
383 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
384
385 trace_bpf_trace_printk(buf);
386 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
387
388 bpf_bprintf_cleanup();
389
390 return ret;
391}
392
393static const struct bpf_func_proto bpf_trace_printk_proto = {
394 .func = bpf_trace_printk,
395 .gpl_only = true,
396 .ret_type = RET_INTEGER,
397 .arg1_type = ARG_PTR_TO_MEM,
398 .arg2_type = ARG_CONST_SIZE,
399};
400
401const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
402{
403 /*
404 * This program might be calling bpf_trace_printk,
405 * so enable the associated bpf_trace/bpf_trace_printk event.
406 * Repeat this each time as it is possible a user has
407 * disabled bpf_trace_printk events. By loading a program
408 * calling bpf_trace_printk() however the user has expressed
409 * the intent to see such events.
410 */
411 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
412 pr_warn_ratelimited("could not enable bpf_trace_printk events");
413
414 return &bpf_trace_printk_proto;
415}
416
417#define MAX_SEQ_PRINTF_VARARGS 12
418
419BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
420 const void *, data, u32, data_len)
421{
422 int err, num_args;
423 u32 *bin_args;
424
425 if (data_len & 7 || data_len > MAX_SEQ_PRINTF_VARARGS * 8 ||
426 (data_len && !data))
427 return -EINVAL;
428 num_args = data_len / 8;
429
430 err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
431 if (err < 0)
432 return err;
433
434 seq_bprintf(m, fmt, bin_args);
435
436 bpf_bprintf_cleanup();
437
438 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
439}
440
441BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
442
443static const struct bpf_func_proto bpf_seq_printf_proto = {
444 .func = bpf_seq_printf,
445 .gpl_only = true,
446 .ret_type = RET_INTEGER,
447 .arg1_type = ARG_PTR_TO_BTF_ID,
448 .arg1_btf_id = &btf_seq_file_ids[0],
449 .arg2_type = ARG_PTR_TO_MEM,
450 .arg3_type = ARG_CONST_SIZE,
451 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
452 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
453};
454
455BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
456{
457 return seq_write(m, data, len) ? -EOVERFLOW : 0;
458}
459
460static const struct bpf_func_proto bpf_seq_write_proto = {
461 .func = bpf_seq_write,
462 .gpl_only = true,
463 .ret_type = RET_INTEGER,
464 .arg1_type = ARG_PTR_TO_BTF_ID,
465 .arg1_btf_id = &btf_seq_file_ids[0],
466 .arg2_type = ARG_PTR_TO_MEM,
467 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
468};
469
470BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
471 u32, btf_ptr_size, u64, flags)
472{
473 const struct btf *btf;
474 s32 btf_id;
475 int ret;
476
477 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
478 if (ret)
479 return ret;
480
481 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
482}
483
484static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
485 .func = bpf_seq_printf_btf,
486 .gpl_only = true,
487 .ret_type = RET_INTEGER,
488 .arg1_type = ARG_PTR_TO_BTF_ID,
489 .arg1_btf_id = &btf_seq_file_ids[0],
490 .arg2_type = ARG_PTR_TO_MEM,
491 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
492 .arg4_type = ARG_ANYTHING,
493};
494
495static __always_inline int
496get_map_perf_counter(struct bpf_map *map, u64 flags,
497 u64 *value, u64 *enabled, u64 *running)
498{
499 struct bpf_array *array = container_of(map, struct bpf_array, map);
500 unsigned int cpu = smp_processor_id();
501 u64 index = flags & BPF_F_INDEX_MASK;
502 struct bpf_event_entry *ee;
503
504 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
505 return -EINVAL;
506 if (index == BPF_F_CURRENT_CPU)
507 index = cpu;
508 if (unlikely(index >= array->map.max_entries))
509 return -E2BIG;
510
511 ee = READ_ONCE(array->ptrs[index]);
512 if (!ee)
513 return -ENOENT;
514
515 return perf_event_read_local(ee->event, value, enabled, running);
516}
517
518BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
519{
520 u64 value = 0;
521 int err;
522
523 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
524 /*
525 * this api is ugly since we miss [-22..-2] range of valid
526 * counter values, but that's uapi
527 */
528 if (err)
529 return err;
530 return value;
531}
532
533static const struct bpf_func_proto bpf_perf_event_read_proto = {
534 .func = bpf_perf_event_read,
535 .gpl_only = true,
536 .ret_type = RET_INTEGER,
537 .arg1_type = ARG_CONST_MAP_PTR,
538 .arg2_type = ARG_ANYTHING,
539};
540
541BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
542 struct bpf_perf_event_value *, buf, u32, size)
543{
544 int err = -EINVAL;
545
546 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
547 goto clear;
548 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
549 &buf->running);
550 if (unlikely(err))
551 goto clear;
552 return 0;
553clear:
554 memset(buf, 0, size);
555 return err;
556}
557
558static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
559 .func = bpf_perf_event_read_value,
560 .gpl_only = true,
561 .ret_type = RET_INTEGER,
562 .arg1_type = ARG_CONST_MAP_PTR,
563 .arg2_type = ARG_ANYTHING,
564 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
565 .arg4_type = ARG_CONST_SIZE,
566};
567
568static __always_inline u64
569__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
570 u64 flags, struct perf_sample_data *sd)
571{
572 struct bpf_array *array = container_of(map, struct bpf_array, map);
573 unsigned int cpu = smp_processor_id();
574 u64 index = flags & BPF_F_INDEX_MASK;
575 struct bpf_event_entry *ee;
576 struct perf_event *event;
577
578 if (index == BPF_F_CURRENT_CPU)
579 index = cpu;
580 if (unlikely(index >= array->map.max_entries))
581 return -E2BIG;
582
583 ee = READ_ONCE(array->ptrs[index]);
584 if (!ee)
585 return -ENOENT;
586
587 event = ee->event;
588 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
589 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
590 return -EINVAL;
591
592 if (unlikely(event->oncpu != cpu))
593 return -EOPNOTSUPP;
594
595 return perf_event_output(event, sd, regs);
596}
597
598/*
599 * Support executing tracepoints in normal, irq, and nmi context that each call
600 * bpf_perf_event_output
601 */
602struct bpf_trace_sample_data {
603 struct perf_sample_data sds[3];
604};
605
606static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
607static DEFINE_PER_CPU(int, bpf_trace_nest_level);
608BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
609 u64, flags, void *, data, u64, size)
610{
611 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
612 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
613 struct perf_raw_record raw = {
614 .frag = {
615 .size = size,
616 .data = data,
617 },
618 };
619 struct perf_sample_data *sd;
620 int err;
621
622 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
623 err = -EBUSY;
624 goto out;
625 }
626
627 sd = &sds->sds[nest_level - 1];
628
629 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
630 err = -EINVAL;
631 goto out;
632 }
633
634 perf_sample_data_init(sd, 0, 0);
635 sd->raw = &raw;
636
637 err = __bpf_perf_event_output(regs, map, flags, sd);
638
639out:
640 this_cpu_dec(bpf_trace_nest_level);
641 return err;
642}
643
644static const struct bpf_func_proto bpf_perf_event_output_proto = {
645 .func = bpf_perf_event_output,
646 .gpl_only = true,
647 .ret_type = RET_INTEGER,
648 .arg1_type = ARG_PTR_TO_CTX,
649 .arg2_type = ARG_CONST_MAP_PTR,
650 .arg3_type = ARG_ANYTHING,
651 .arg4_type = ARG_PTR_TO_MEM,
652 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
653};
654
655static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
656struct bpf_nested_pt_regs {
657 struct pt_regs regs[3];
658};
659static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
660static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
661
662u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
663 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
664{
665 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
666 struct perf_raw_frag frag = {
667 .copy = ctx_copy,
668 .size = ctx_size,
669 .data = ctx,
670 };
671 struct perf_raw_record raw = {
672 .frag = {
673 {
674 .next = ctx_size ? &frag : NULL,
675 },
676 .size = meta_size,
677 .data = meta,
678 },
679 };
680 struct perf_sample_data *sd;
681 struct pt_regs *regs;
682 u64 ret;
683
684 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
685 ret = -EBUSY;
686 goto out;
687 }
688 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
689 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
690
691 perf_fetch_caller_regs(regs);
692 perf_sample_data_init(sd, 0, 0);
693 sd->raw = &raw;
694
695 ret = __bpf_perf_event_output(regs, map, flags, sd);
696out:
697 this_cpu_dec(bpf_event_output_nest_level);
698 return ret;
699}
700
701BPF_CALL_0(bpf_get_current_task)
702{
703 return (long) current;
704}
705
706const struct bpf_func_proto bpf_get_current_task_proto = {
707 .func = bpf_get_current_task,
708 .gpl_only = true,
709 .ret_type = RET_INTEGER,
710};
711
712BPF_CALL_0(bpf_get_current_task_btf)
713{
714 return (unsigned long) current;
715}
716
717BTF_ID_LIST_SINGLE(bpf_get_current_btf_ids, struct, task_struct)
718
719static const struct bpf_func_proto bpf_get_current_task_btf_proto = {
720 .func = bpf_get_current_task_btf,
721 .gpl_only = true,
722 .ret_type = RET_PTR_TO_BTF_ID,
723 .ret_btf_id = &bpf_get_current_btf_ids[0],
724};
725
726BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
727{
728 struct bpf_array *array = container_of(map, struct bpf_array, map);
729 struct cgroup *cgrp;
730
731 if (unlikely(idx >= array->map.max_entries))
732 return -E2BIG;
733
734 cgrp = READ_ONCE(array->ptrs[idx]);
735 if (unlikely(!cgrp))
736 return -EAGAIN;
737
738 return task_under_cgroup_hierarchy(current, cgrp);
739}
740
741static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
742 .func = bpf_current_task_under_cgroup,
743 .gpl_only = false,
744 .ret_type = RET_INTEGER,
745 .arg1_type = ARG_CONST_MAP_PTR,
746 .arg2_type = ARG_ANYTHING,
747};
748
749struct send_signal_irq_work {
750 struct irq_work irq_work;
751 struct task_struct *task;
752 u32 sig;
753 enum pid_type type;
754};
755
756static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
757
758static void do_bpf_send_signal(struct irq_work *entry)
759{
760 struct send_signal_irq_work *work;
761
762 work = container_of(entry, struct send_signal_irq_work, irq_work);
763 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
764}
765
766static int bpf_send_signal_common(u32 sig, enum pid_type type)
767{
768 struct send_signal_irq_work *work = NULL;
769
770 /* Similar to bpf_probe_write_user, task needs to be
771 * in a sound condition and kernel memory access be
772 * permitted in order to send signal to the current
773 * task.
774 */
775 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
776 return -EPERM;
777 if (unlikely(uaccess_kernel()))
778 return -EPERM;
779 if (unlikely(!nmi_uaccess_okay()))
780 return -EPERM;
781
782 if (irqs_disabled()) {
783 /* Do an early check on signal validity. Otherwise,
784 * the error is lost in deferred irq_work.
785 */
786 if (unlikely(!valid_signal(sig)))
787 return -EINVAL;
788
789 work = this_cpu_ptr(&send_signal_work);
790 if (irq_work_is_busy(&work->irq_work))
791 return -EBUSY;
792
793 /* Add the current task, which is the target of sending signal,
794 * to the irq_work. The current task may change when queued
795 * irq works get executed.
796 */
797 work->task = current;
798 work->sig = sig;
799 work->type = type;
800 irq_work_queue(&work->irq_work);
801 return 0;
802 }
803
804 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
805}
806
807BPF_CALL_1(bpf_send_signal, u32, sig)
808{
809 return bpf_send_signal_common(sig, PIDTYPE_TGID);
810}
811
812static const struct bpf_func_proto bpf_send_signal_proto = {
813 .func = bpf_send_signal,
814 .gpl_only = false,
815 .ret_type = RET_INTEGER,
816 .arg1_type = ARG_ANYTHING,
817};
818
819BPF_CALL_1(bpf_send_signal_thread, u32, sig)
820{
821 return bpf_send_signal_common(sig, PIDTYPE_PID);
822}
823
824static const struct bpf_func_proto bpf_send_signal_thread_proto = {
825 .func = bpf_send_signal_thread,
826 .gpl_only = false,
827 .ret_type = RET_INTEGER,
828 .arg1_type = ARG_ANYTHING,
829};
830
831BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
832{
833 long len;
834 char *p;
835
836 if (!sz)
837 return 0;
838
839 p = d_path(path, buf, sz);
840 if (IS_ERR(p)) {
841 len = PTR_ERR(p);
842 } else {
843 len = buf + sz - p;
844 memmove(buf, p, len);
845 }
846
847 return len;
848}
849
850BTF_SET_START(btf_allowlist_d_path)
851#ifdef CONFIG_SECURITY
852BTF_ID(func, security_file_permission)
853BTF_ID(func, security_inode_getattr)
854BTF_ID(func, security_file_open)
855#endif
856#ifdef CONFIG_SECURITY_PATH
857BTF_ID(func, security_path_truncate)
858#endif
859BTF_ID(func, vfs_truncate)
860BTF_ID(func, vfs_fallocate)
861BTF_ID(func, dentry_open)
862BTF_ID(func, vfs_getattr)
863BTF_ID(func, filp_close)
864BTF_SET_END(btf_allowlist_d_path)
865
866static bool bpf_d_path_allowed(const struct bpf_prog *prog)
867{
868 if (prog->type == BPF_PROG_TYPE_TRACING &&
869 prog->expected_attach_type == BPF_TRACE_ITER)
870 return true;
871
872 if (prog->type == BPF_PROG_TYPE_LSM)
873 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
874
875 return btf_id_set_contains(&btf_allowlist_d_path,
876 prog->aux->attach_btf_id);
877}
878
879BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
880
881static const struct bpf_func_proto bpf_d_path_proto = {
882 .func = bpf_d_path,
883 .gpl_only = false,
884 .ret_type = RET_INTEGER,
885 .arg1_type = ARG_PTR_TO_BTF_ID,
886 .arg1_btf_id = &bpf_d_path_btf_ids[0],
887 .arg2_type = ARG_PTR_TO_MEM,
888 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
889 .allowed = bpf_d_path_allowed,
890};
891
892#define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
893 BTF_F_PTR_RAW | BTF_F_ZERO)
894
895static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
896 u64 flags, const struct btf **btf,
897 s32 *btf_id)
898{
899 const struct btf_type *t;
900
901 if (unlikely(flags & ~(BTF_F_ALL)))
902 return -EINVAL;
903
904 if (btf_ptr_size != sizeof(struct btf_ptr))
905 return -EINVAL;
906
907 *btf = bpf_get_btf_vmlinux();
908
909 if (IS_ERR_OR_NULL(*btf))
910 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
911
912 if (ptr->type_id > 0)
913 *btf_id = ptr->type_id;
914 else
915 return -EINVAL;
916
917 if (*btf_id > 0)
918 t = btf_type_by_id(*btf, *btf_id);
919 if (*btf_id <= 0 || !t)
920 return -ENOENT;
921
922 return 0;
923}
924
925BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
926 u32, btf_ptr_size, u64, flags)
927{
928 const struct btf *btf;
929 s32 btf_id;
930 int ret;
931
932 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
933 if (ret)
934 return ret;
935
936 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
937 flags);
938}
939
940const struct bpf_func_proto bpf_snprintf_btf_proto = {
941 .func = bpf_snprintf_btf,
942 .gpl_only = false,
943 .ret_type = RET_INTEGER,
944 .arg1_type = ARG_PTR_TO_MEM,
945 .arg2_type = ARG_CONST_SIZE,
946 .arg3_type = ARG_PTR_TO_MEM,
947 .arg4_type = ARG_CONST_SIZE,
948 .arg5_type = ARG_ANYTHING,
949};
950
951const struct bpf_func_proto *
952bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
953{
954 switch (func_id) {
955 case BPF_FUNC_map_lookup_elem:
956 return &bpf_map_lookup_elem_proto;
957 case BPF_FUNC_map_update_elem:
958 return &bpf_map_update_elem_proto;
959 case BPF_FUNC_map_delete_elem:
960 return &bpf_map_delete_elem_proto;
961 case BPF_FUNC_map_push_elem:
962 return &bpf_map_push_elem_proto;
963 case BPF_FUNC_map_pop_elem:
964 return &bpf_map_pop_elem_proto;
965 case BPF_FUNC_map_peek_elem:
966 return &bpf_map_peek_elem_proto;
967 case BPF_FUNC_ktime_get_ns:
968 return &bpf_ktime_get_ns_proto;
969 case BPF_FUNC_ktime_get_boot_ns:
970 return &bpf_ktime_get_boot_ns_proto;
971 case BPF_FUNC_ktime_get_coarse_ns:
972 return &bpf_ktime_get_coarse_ns_proto;
973 case BPF_FUNC_tail_call:
974 return &bpf_tail_call_proto;
975 case BPF_FUNC_get_current_pid_tgid:
976 return &bpf_get_current_pid_tgid_proto;
977 case BPF_FUNC_get_current_task:
978 return &bpf_get_current_task_proto;
979 case BPF_FUNC_get_current_task_btf:
980 return &bpf_get_current_task_btf_proto;
981 case BPF_FUNC_get_current_uid_gid:
982 return &bpf_get_current_uid_gid_proto;
983 case BPF_FUNC_get_current_comm:
984 return &bpf_get_current_comm_proto;
985 case BPF_FUNC_trace_printk:
986 return bpf_get_trace_printk_proto();
987 case BPF_FUNC_get_smp_processor_id:
988 return &bpf_get_smp_processor_id_proto;
989 case BPF_FUNC_get_numa_node_id:
990 return &bpf_get_numa_node_id_proto;
991 case BPF_FUNC_perf_event_read:
992 return &bpf_perf_event_read_proto;
993 case BPF_FUNC_current_task_under_cgroup:
994 return &bpf_current_task_under_cgroup_proto;
995 case BPF_FUNC_get_prandom_u32:
996 return &bpf_get_prandom_u32_proto;
997 case BPF_FUNC_probe_write_user:
998 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
999 NULL : bpf_get_probe_write_proto();
1000 case BPF_FUNC_probe_read_user:
1001 return &bpf_probe_read_user_proto;
1002 case BPF_FUNC_probe_read_kernel:
1003 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1004 NULL : &bpf_probe_read_kernel_proto;
1005 case BPF_FUNC_probe_read_user_str:
1006 return &bpf_probe_read_user_str_proto;
1007 case BPF_FUNC_probe_read_kernel_str:
1008 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1009 NULL : &bpf_probe_read_kernel_str_proto;
1010#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1011 case BPF_FUNC_probe_read:
1012 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1013 NULL : &bpf_probe_read_compat_proto;
1014 case BPF_FUNC_probe_read_str:
1015 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1016 NULL : &bpf_probe_read_compat_str_proto;
1017#endif
1018#ifdef CONFIG_CGROUPS
1019 case BPF_FUNC_get_current_cgroup_id:
1020 return &bpf_get_current_cgroup_id_proto;
1021 case BPF_FUNC_get_current_ancestor_cgroup_id:
1022 return &bpf_get_current_ancestor_cgroup_id_proto;
1023#endif
1024 case BPF_FUNC_send_signal:
1025 return &bpf_send_signal_proto;
1026 case BPF_FUNC_send_signal_thread:
1027 return &bpf_send_signal_thread_proto;
1028 case BPF_FUNC_perf_event_read_value:
1029 return &bpf_perf_event_read_value_proto;
1030 case BPF_FUNC_get_ns_current_pid_tgid:
1031 return &bpf_get_ns_current_pid_tgid_proto;
1032 case BPF_FUNC_ringbuf_output:
1033 return &bpf_ringbuf_output_proto;
1034 case BPF_FUNC_ringbuf_reserve:
1035 return &bpf_ringbuf_reserve_proto;
1036 case BPF_FUNC_ringbuf_submit:
1037 return &bpf_ringbuf_submit_proto;
1038 case BPF_FUNC_ringbuf_discard:
1039 return &bpf_ringbuf_discard_proto;
1040 case BPF_FUNC_ringbuf_query:
1041 return &bpf_ringbuf_query_proto;
1042 case BPF_FUNC_jiffies64:
1043 return &bpf_jiffies64_proto;
1044 case BPF_FUNC_get_task_stack:
1045 return &bpf_get_task_stack_proto;
1046 case BPF_FUNC_copy_from_user:
1047 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1048 case BPF_FUNC_snprintf_btf:
1049 return &bpf_snprintf_btf_proto;
1050 case BPF_FUNC_per_cpu_ptr:
1051 return &bpf_per_cpu_ptr_proto;
1052 case BPF_FUNC_this_cpu_ptr:
1053 return &bpf_this_cpu_ptr_proto;
1054 case BPF_FUNC_task_storage_get:
1055 return &bpf_task_storage_get_proto;
1056 case BPF_FUNC_task_storage_delete:
1057 return &bpf_task_storage_delete_proto;
1058 case BPF_FUNC_for_each_map_elem:
1059 return &bpf_for_each_map_elem_proto;
1060 case BPF_FUNC_snprintf:
1061 return &bpf_snprintf_proto;
1062 default:
1063 return NULL;
1064 }
1065}
1066
1067static const struct bpf_func_proto *
1068kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1069{
1070 switch (func_id) {
1071 case BPF_FUNC_perf_event_output:
1072 return &bpf_perf_event_output_proto;
1073 case BPF_FUNC_get_stackid:
1074 return &bpf_get_stackid_proto;
1075 case BPF_FUNC_get_stack:
1076 return &bpf_get_stack_proto;
1077#ifdef CONFIG_BPF_KPROBE_OVERRIDE
1078 case BPF_FUNC_override_return:
1079 return &bpf_override_return_proto;
1080#endif
1081 default:
1082 return bpf_tracing_func_proto(func_id, prog);
1083 }
1084}
1085
1086/* bpf+kprobe programs can access fields of 'struct pt_regs' */
1087static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1088 const struct bpf_prog *prog,
1089 struct bpf_insn_access_aux *info)
1090{
1091 if (off < 0 || off >= sizeof(struct pt_regs))
1092 return false;
1093 if (type != BPF_READ)
1094 return false;
1095 if (off % size != 0)
1096 return false;
1097 /*
1098 * Assertion for 32 bit to make sure last 8 byte access
1099 * (BPF_DW) to the last 4 byte member is disallowed.
1100 */
1101 if (off + size > sizeof(struct pt_regs))
1102 return false;
1103
1104 return true;
1105}
1106
1107const struct bpf_verifier_ops kprobe_verifier_ops = {
1108 .get_func_proto = kprobe_prog_func_proto,
1109 .is_valid_access = kprobe_prog_is_valid_access,
1110};
1111
1112const struct bpf_prog_ops kprobe_prog_ops = {
1113};
1114
1115BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1116 u64, flags, void *, data, u64, size)
1117{
1118 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1119
1120 /*
1121 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1122 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1123 * from there and call the same bpf_perf_event_output() helper inline.
1124 */
1125 return ____bpf_perf_event_output(regs, map, flags, data, size);
1126}
1127
1128static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1129 .func = bpf_perf_event_output_tp,
1130 .gpl_only = true,
1131 .ret_type = RET_INTEGER,
1132 .arg1_type = ARG_PTR_TO_CTX,
1133 .arg2_type = ARG_CONST_MAP_PTR,
1134 .arg3_type = ARG_ANYTHING,
1135 .arg4_type = ARG_PTR_TO_MEM,
1136 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1137};
1138
1139BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1140 u64, flags)
1141{
1142 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1143
1144 /*
1145 * Same comment as in bpf_perf_event_output_tp(), only that this time
1146 * the other helper's function body cannot be inlined due to being
1147 * external, thus we need to call raw helper function.
1148 */
1149 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1150 flags, 0, 0);
1151}
1152
1153static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1154 .func = bpf_get_stackid_tp,
1155 .gpl_only = true,
1156 .ret_type = RET_INTEGER,
1157 .arg1_type = ARG_PTR_TO_CTX,
1158 .arg2_type = ARG_CONST_MAP_PTR,
1159 .arg3_type = ARG_ANYTHING,
1160};
1161
1162BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1163 u64, flags)
1164{
1165 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1166
1167 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1168 (unsigned long) size, flags, 0);
1169}
1170
1171static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1172 .func = bpf_get_stack_tp,
1173 .gpl_only = true,
1174 .ret_type = RET_INTEGER,
1175 .arg1_type = ARG_PTR_TO_CTX,
1176 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1177 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1178 .arg4_type = ARG_ANYTHING,
1179};
1180
1181static const struct bpf_func_proto *
1182tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1183{
1184 switch (func_id) {
1185 case BPF_FUNC_perf_event_output:
1186 return &bpf_perf_event_output_proto_tp;
1187 case BPF_FUNC_get_stackid:
1188 return &bpf_get_stackid_proto_tp;
1189 case BPF_FUNC_get_stack:
1190 return &bpf_get_stack_proto_tp;
1191 default:
1192 return bpf_tracing_func_proto(func_id, prog);
1193 }
1194}
1195
1196static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1197 const struct bpf_prog *prog,
1198 struct bpf_insn_access_aux *info)
1199{
1200 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1201 return false;
1202 if (type != BPF_READ)
1203 return false;
1204 if (off % size != 0)
1205 return false;
1206
1207 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1208 return true;
1209}
1210
1211const struct bpf_verifier_ops tracepoint_verifier_ops = {
1212 .get_func_proto = tp_prog_func_proto,
1213 .is_valid_access = tp_prog_is_valid_access,
1214};
1215
1216const struct bpf_prog_ops tracepoint_prog_ops = {
1217};
1218
1219BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1220 struct bpf_perf_event_value *, buf, u32, size)
1221{
1222 int err = -EINVAL;
1223
1224 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1225 goto clear;
1226 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1227 &buf->running);
1228 if (unlikely(err))
1229 goto clear;
1230 return 0;
1231clear:
1232 memset(buf, 0, size);
1233 return err;
1234}
1235
1236static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1237 .func = bpf_perf_prog_read_value,
1238 .gpl_only = true,
1239 .ret_type = RET_INTEGER,
1240 .arg1_type = ARG_PTR_TO_CTX,
1241 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1242 .arg3_type = ARG_CONST_SIZE,
1243};
1244
1245BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1246 void *, buf, u32, size, u64, flags)
1247{
1248#ifndef CONFIG_X86
1249 return -ENOENT;
1250#else
1251 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1252 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1253 u32 to_copy;
1254
1255 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1256 return -EINVAL;
1257
1258 if (unlikely(!br_stack))
1259 return -EINVAL;
1260
1261 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1262 return br_stack->nr * br_entry_size;
1263
1264 if (!buf || (size % br_entry_size != 0))
1265 return -EINVAL;
1266
1267 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1268 memcpy(buf, br_stack->entries, to_copy);
1269
1270 return to_copy;
1271#endif
1272}
1273
1274static const struct bpf_func_proto bpf_read_branch_records_proto = {
1275 .func = bpf_read_branch_records,
1276 .gpl_only = true,
1277 .ret_type = RET_INTEGER,
1278 .arg1_type = ARG_PTR_TO_CTX,
1279 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1280 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1281 .arg4_type = ARG_ANYTHING,
1282};
1283
1284static const struct bpf_func_proto *
1285pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1286{
1287 switch (func_id) {
1288 case BPF_FUNC_perf_event_output:
1289 return &bpf_perf_event_output_proto_tp;
1290 case BPF_FUNC_get_stackid:
1291 return &bpf_get_stackid_proto_pe;
1292 case BPF_FUNC_get_stack:
1293 return &bpf_get_stack_proto_pe;
1294 case BPF_FUNC_perf_prog_read_value:
1295 return &bpf_perf_prog_read_value_proto;
1296 case BPF_FUNC_read_branch_records:
1297 return &bpf_read_branch_records_proto;
1298 default:
1299 return bpf_tracing_func_proto(func_id, prog);
1300 }
1301}
1302
1303/*
1304 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1305 * to avoid potential recursive reuse issue when/if tracepoints are added
1306 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1307 *
1308 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1309 * in normal, irq, and nmi context.
1310 */
1311struct bpf_raw_tp_regs {
1312 struct pt_regs regs[3];
1313};
1314static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1315static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1316static struct pt_regs *get_bpf_raw_tp_regs(void)
1317{
1318 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1319 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1320
1321 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1322 this_cpu_dec(bpf_raw_tp_nest_level);
1323 return ERR_PTR(-EBUSY);
1324 }
1325
1326 return &tp_regs->regs[nest_level - 1];
1327}
1328
1329static void put_bpf_raw_tp_regs(void)
1330{
1331 this_cpu_dec(bpf_raw_tp_nest_level);
1332}
1333
1334BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1335 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1336{
1337 struct pt_regs *regs = get_bpf_raw_tp_regs();
1338 int ret;
1339
1340 if (IS_ERR(regs))
1341 return PTR_ERR(regs);
1342
1343 perf_fetch_caller_regs(regs);
1344 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1345
1346 put_bpf_raw_tp_regs();
1347 return ret;
1348}
1349
1350static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1351 .func = bpf_perf_event_output_raw_tp,
1352 .gpl_only = true,
1353 .ret_type = RET_INTEGER,
1354 .arg1_type = ARG_PTR_TO_CTX,
1355 .arg2_type = ARG_CONST_MAP_PTR,
1356 .arg3_type = ARG_ANYTHING,
1357 .arg4_type = ARG_PTR_TO_MEM,
1358 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1359};
1360
1361extern const struct bpf_func_proto bpf_skb_output_proto;
1362extern const struct bpf_func_proto bpf_xdp_output_proto;
1363
1364BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1365 struct bpf_map *, map, u64, flags)
1366{
1367 struct pt_regs *regs = get_bpf_raw_tp_regs();
1368 int ret;
1369
1370 if (IS_ERR(regs))
1371 return PTR_ERR(regs);
1372
1373 perf_fetch_caller_regs(regs);
1374 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1375 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1376 flags, 0, 0);
1377 put_bpf_raw_tp_regs();
1378 return ret;
1379}
1380
1381static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1382 .func = bpf_get_stackid_raw_tp,
1383 .gpl_only = true,
1384 .ret_type = RET_INTEGER,
1385 .arg1_type = ARG_PTR_TO_CTX,
1386 .arg2_type = ARG_CONST_MAP_PTR,
1387 .arg3_type = ARG_ANYTHING,
1388};
1389
1390BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1391 void *, buf, u32, size, u64, flags)
1392{
1393 struct pt_regs *regs = get_bpf_raw_tp_regs();
1394 int ret;
1395
1396 if (IS_ERR(regs))
1397 return PTR_ERR(regs);
1398
1399 perf_fetch_caller_regs(regs);
1400 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1401 (unsigned long) size, flags, 0);
1402 put_bpf_raw_tp_regs();
1403 return ret;
1404}
1405
1406static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1407 .func = bpf_get_stack_raw_tp,
1408 .gpl_only = true,
1409 .ret_type = RET_INTEGER,
1410 .arg1_type = ARG_PTR_TO_CTX,
1411 .arg2_type = ARG_PTR_TO_MEM,
1412 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1413 .arg4_type = ARG_ANYTHING,
1414};
1415
1416static const struct bpf_func_proto *
1417raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1418{
1419 switch (func_id) {
1420 case BPF_FUNC_perf_event_output:
1421 return &bpf_perf_event_output_proto_raw_tp;
1422 case BPF_FUNC_get_stackid:
1423 return &bpf_get_stackid_proto_raw_tp;
1424 case BPF_FUNC_get_stack:
1425 return &bpf_get_stack_proto_raw_tp;
1426 default:
1427 return bpf_tracing_func_proto(func_id, prog);
1428 }
1429}
1430
1431const struct bpf_func_proto *
1432tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1433{
1434 switch (func_id) {
1435#ifdef CONFIG_NET
1436 case BPF_FUNC_skb_output:
1437 return &bpf_skb_output_proto;
1438 case BPF_FUNC_xdp_output:
1439 return &bpf_xdp_output_proto;
1440 case BPF_FUNC_skc_to_tcp6_sock:
1441 return &bpf_skc_to_tcp6_sock_proto;
1442 case BPF_FUNC_skc_to_tcp_sock:
1443 return &bpf_skc_to_tcp_sock_proto;
1444 case BPF_FUNC_skc_to_tcp_timewait_sock:
1445 return &bpf_skc_to_tcp_timewait_sock_proto;
1446 case BPF_FUNC_skc_to_tcp_request_sock:
1447 return &bpf_skc_to_tcp_request_sock_proto;
1448 case BPF_FUNC_skc_to_udp6_sock:
1449 return &bpf_skc_to_udp6_sock_proto;
1450 case BPF_FUNC_sk_storage_get:
1451 return &bpf_sk_storage_get_tracing_proto;
1452 case BPF_FUNC_sk_storage_delete:
1453 return &bpf_sk_storage_delete_tracing_proto;
1454 case BPF_FUNC_sock_from_file:
1455 return &bpf_sock_from_file_proto;
1456 case BPF_FUNC_get_socket_cookie:
1457 return &bpf_get_socket_ptr_cookie_proto;
1458#endif
1459 case BPF_FUNC_seq_printf:
1460 return prog->expected_attach_type == BPF_TRACE_ITER ?
1461 &bpf_seq_printf_proto :
1462 NULL;
1463 case BPF_FUNC_seq_write:
1464 return prog->expected_attach_type == BPF_TRACE_ITER ?
1465 &bpf_seq_write_proto :
1466 NULL;
1467 case BPF_FUNC_seq_printf_btf:
1468 return prog->expected_attach_type == BPF_TRACE_ITER ?
1469 &bpf_seq_printf_btf_proto :
1470 NULL;
1471 case BPF_FUNC_d_path:
1472 return &bpf_d_path_proto;
1473 default:
1474 return raw_tp_prog_func_proto(func_id, prog);
1475 }
1476}
1477
1478static bool raw_tp_prog_is_valid_access(int off, int size,
1479 enum bpf_access_type type,
1480 const struct bpf_prog *prog,
1481 struct bpf_insn_access_aux *info)
1482{
1483 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1484 return false;
1485 if (type != BPF_READ)
1486 return false;
1487 if (off % size != 0)
1488 return false;
1489 return true;
1490}
1491
1492static bool tracing_prog_is_valid_access(int off, int size,
1493 enum bpf_access_type type,
1494 const struct bpf_prog *prog,
1495 struct bpf_insn_access_aux *info)
1496{
1497 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1498 return false;
1499 if (type != BPF_READ)
1500 return false;
1501 if (off % size != 0)
1502 return false;
1503 return btf_ctx_access(off, size, type, prog, info);
1504}
1505
1506int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1507 const union bpf_attr *kattr,
1508 union bpf_attr __user *uattr)
1509{
1510 return -ENOTSUPP;
1511}
1512
1513const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1514 .get_func_proto = raw_tp_prog_func_proto,
1515 .is_valid_access = raw_tp_prog_is_valid_access,
1516};
1517
1518const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1519#ifdef CONFIG_NET
1520 .test_run = bpf_prog_test_run_raw_tp,
1521#endif
1522};
1523
1524const struct bpf_verifier_ops tracing_verifier_ops = {
1525 .get_func_proto = tracing_prog_func_proto,
1526 .is_valid_access = tracing_prog_is_valid_access,
1527};
1528
1529const struct bpf_prog_ops tracing_prog_ops = {
1530 .test_run = bpf_prog_test_run_tracing,
1531};
1532
1533static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1534 enum bpf_access_type type,
1535 const struct bpf_prog *prog,
1536 struct bpf_insn_access_aux *info)
1537{
1538 if (off == 0) {
1539 if (size != sizeof(u64) || type != BPF_READ)
1540 return false;
1541 info->reg_type = PTR_TO_TP_BUFFER;
1542 }
1543 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1544}
1545
1546const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1547 .get_func_proto = raw_tp_prog_func_proto,
1548 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1549};
1550
1551const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1552};
1553
1554static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1555 const struct bpf_prog *prog,
1556 struct bpf_insn_access_aux *info)
1557{
1558 const int size_u64 = sizeof(u64);
1559
1560 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1561 return false;
1562 if (type != BPF_READ)
1563 return false;
1564 if (off % size != 0) {
1565 if (sizeof(unsigned long) != 4)
1566 return false;
1567 if (size != 8)
1568 return false;
1569 if (off % size != 4)
1570 return false;
1571 }
1572
1573 switch (off) {
1574 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1575 bpf_ctx_record_field_size(info, size_u64);
1576 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1577 return false;
1578 break;
1579 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1580 bpf_ctx_record_field_size(info, size_u64);
1581 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1582 return false;
1583 break;
1584 default:
1585 if (size != sizeof(long))
1586 return false;
1587 }
1588
1589 return true;
1590}
1591
1592static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1593 const struct bpf_insn *si,
1594 struct bpf_insn *insn_buf,
1595 struct bpf_prog *prog, u32 *target_size)
1596{
1597 struct bpf_insn *insn = insn_buf;
1598
1599 switch (si->off) {
1600 case offsetof(struct bpf_perf_event_data, sample_period):
1601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1602 data), si->dst_reg, si->src_reg,
1603 offsetof(struct bpf_perf_event_data_kern, data));
1604 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1605 bpf_target_off(struct perf_sample_data, period, 8,
1606 target_size));
1607 break;
1608 case offsetof(struct bpf_perf_event_data, addr):
1609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1610 data), si->dst_reg, si->src_reg,
1611 offsetof(struct bpf_perf_event_data_kern, data));
1612 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1613 bpf_target_off(struct perf_sample_data, addr, 8,
1614 target_size));
1615 break;
1616 default:
1617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1618 regs), si->dst_reg, si->src_reg,
1619 offsetof(struct bpf_perf_event_data_kern, regs));
1620 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1621 si->off);
1622 break;
1623 }
1624
1625 return insn - insn_buf;
1626}
1627
1628const struct bpf_verifier_ops perf_event_verifier_ops = {
1629 .get_func_proto = pe_prog_func_proto,
1630 .is_valid_access = pe_prog_is_valid_access,
1631 .convert_ctx_access = pe_prog_convert_ctx_access,
1632};
1633
1634const struct bpf_prog_ops perf_event_prog_ops = {
1635};
1636
1637static DEFINE_MUTEX(bpf_event_mutex);
1638
1639#define BPF_TRACE_MAX_PROGS 64
1640
1641int perf_event_attach_bpf_prog(struct perf_event *event,
1642 struct bpf_prog *prog)
1643{
1644 struct bpf_prog_array *old_array;
1645 struct bpf_prog_array *new_array;
1646 int ret = -EEXIST;
1647
1648 /*
1649 * Kprobe override only works if they are on the function entry,
1650 * and only if they are on the opt-in list.
1651 */
1652 if (prog->kprobe_override &&
1653 (!trace_kprobe_on_func_entry(event->tp_event) ||
1654 !trace_kprobe_error_injectable(event->tp_event)))
1655 return -EINVAL;
1656
1657 mutex_lock(&bpf_event_mutex);
1658
1659 if (event->prog)
1660 goto unlock;
1661
1662 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1663 if (old_array &&
1664 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1665 ret = -E2BIG;
1666 goto unlock;
1667 }
1668
1669 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1670 if (ret < 0)
1671 goto unlock;
1672
1673 /* set the new array to event->tp_event and set event->prog */
1674 event->prog = prog;
1675 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1676 bpf_prog_array_free(old_array);
1677
1678unlock:
1679 mutex_unlock(&bpf_event_mutex);
1680 return ret;
1681}
1682
1683void perf_event_detach_bpf_prog(struct perf_event *event)
1684{
1685 struct bpf_prog_array *old_array;
1686 struct bpf_prog_array *new_array;
1687 int ret;
1688
1689 mutex_lock(&bpf_event_mutex);
1690
1691 if (!event->prog)
1692 goto unlock;
1693
1694 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1695 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1696 if (ret == -ENOENT)
1697 goto unlock;
1698 if (ret < 0) {
1699 bpf_prog_array_delete_safe(old_array, event->prog);
1700 } else {
1701 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1702 bpf_prog_array_free(old_array);
1703 }
1704
1705 bpf_prog_put(event->prog);
1706 event->prog = NULL;
1707
1708unlock:
1709 mutex_unlock(&bpf_event_mutex);
1710}
1711
1712int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1713{
1714 struct perf_event_query_bpf __user *uquery = info;
1715 struct perf_event_query_bpf query = {};
1716 struct bpf_prog_array *progs;
1717 u32 *ids, prog_cnt, ids_len;
1718 int ret;
1719
1720 if (!perfmon_capable())
1721 return -EPERM;
1722 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1723 return -EINVAL;
1724 if (copy_from_user(&query, uquery, sizeof(query)))
1725 return -EFAULT;
1726
1727 ids_len = query.ids_len;
1728 if (ids_len > BPF_TRACE_MAX_PROGS)
1729 return -E2BIG;
1730 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1731 if (!ids)
1732 return -ENOMEM;
1733 /*
1734 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1735 * is required when user only wants to check for uquery->prog_cnt.
1736 * There is no need to check for it since the case is handled
1737 * gracefully in bpf_prog_array_copy_info.
1738 */
1739
1740 mutex_lock(&bpf_event_mutex);
1741 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1742 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1743 mutex_unlock(&bpf_event_mutex);
1744
1745 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1746 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1747 ret = -EFAULT;
1748
1749 kfree(ids);
1750 return ret;
1751}
1752
1753extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1754extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1755
1756struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1757{
1758 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1759
1760 for (; btp < __stop__bpf_raw_tp; btp++) {
1761 if (!strcmp(btp->tp->name, name))
1762 return btp;
1763 }
1764
1765 return bpf_get_raw_tracepoint_module(name);
1766}
1767
1768void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1769{
1770 struct module *mod;
1771
1772 preempt_disable();
1773 mod = __module_address((unsigned long)btp);
1774 module_put(mod);
1775 preempt_enable();
1776}
1777
1778static __always_inline
1779void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1780{
1781 cant_sleep();
1782 rcu_read_lock();
1783 (void) BPF_PROG_RUN(prog, args);
1784 rcu_read_unlock();
1785}
1786
1787#define UNPACK(...) __VA_ARGS__
1788#define REPEAT_1(FN, DL, X, ...) FN(X)
1789#define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1790#define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1791#define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1792#define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1793#define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1794#define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1795#define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1796#define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1797#define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1798#define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1799#define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1800#define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1801
1802#define SARG(X) u64 arg##X
1803#define COPY(X) args[X] = arg##X
1804
1805#define __DL_COM (,)
1806#define __DL_SEM (;)
1807
1808#define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1809
1810#define BPF_TRACE_DEFN_x(x) \
1811 void bpf_trace_run##x(struct bpf_prog *prog, \
1812 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1813 { \
1814 u64 args[x]; \
1815 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1816 __bpf_trace_run(prog, args); \
1817 } \
1818 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1819BPF_TRACE_DEFN_x(1);
1820BPF_TRACE_DEFN_x(2);
1821BPF_TRACE_DEFN_x(3);
1822BPF_TRACE_DEFN_x(4);
1823BPF_TRACE_DEFN_x(5);
1824BPF_TRACE_DEFN_x(6);
1825BPF_TRACE_DEFN_x(7);
1826BPF_TRACE_DEFN_x(8);
1827BPF_TRACE_DEFN_x(9);
1828BPF_TRACE_DEFN_x(10);
1829BPF_TRACE_DEFN_x(11);
1830BPF_TRACE_DEFN_x(12);
1831
1832static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1833{
1834 struct tracepoint *tp = btp->tp;
1835
1836 /*
1837 * check that program doesn't access arguments beyond what's
1838 * available in this tracepoint
1839 */
1840 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1841 return -EINVAL;
1842
1843 if (prog->aux->max_tp_access > btp->writable_size)
1844 return -EINVAL;
1845
1846 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
1847 prog);
1848}
1849
1850int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1851{
1852 return __bpf_probe_register(btp, prog);
1853}
1854
1855int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1856{
1857 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1858}
1859
1860int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1861 u32 *fd_type, const char **buf,
1862 u64 *probe_offset, u64 *probe_addr)
1863{
1864 bool is_tracepoint, is_syscall_tp;
1865 struct bpf_prog *prog;
1866 int flags, err = 0;
1867
1868 prog = event->prog;
1869 if (!prog)
1870 return -ENOENT;
1871
1872 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1873 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1874 return -EOPNOTSUPP;
1875
1876 *prog_id = prog->aux->id;
1877 flags = event->tp_event->flags;
1878 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1879 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1880
1881 if (is_tracepoint || is_syscall_tp) {
1882 *buf = is_tracepoint ? event->tp_event->tp->name
1883 : event->tp_event->name;
1884 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1885 *probe_offset = 0x0;
1886 *probe_addr = 0x0;
1887 } else {
1888 /* kprobe/uprobe */
1889 err = -EOPNOTSUPP;
1890#ifdef CONFIG_KPROBE_EVENTS
1891 if (flags & TRACE_EVENT_FL_KPROBE)
1892 err = bpf_get_kprobe_info(event, fd_type, buf,
1893 probe_offset, probe_addr,
1894 event->attr.type == PERF_TYPE_TRACEPOINT);
1895#endif
1896#ifdef CONFIG_UPROBE_EVENTS
1897 if (flags & TRACE_EVENT_FL_UPROBE)
1898 err = bpf_get_uprobe_info(event, fd_type, buf,
1899 probe_offset,
1900 event->attr.type == PERF_TYPE_TRACEPOINT);
1901#endif
1902 }
1903
1904 return err;
1905}
1906
1907static int __init send_signal_irq_work_init(void)
1908{
1909 int cpu;
1910 struct send_signal_irq_work *work;
1911
1912 for_each_possible_cpu(cpu) {
1913 work = per_cpu_ptr(&send_signal_work, cpu);
1914 init_irq_work(&work->irq_work, do_bpf_send_signal);
1915 }
1916 return 0;
1917}
1918
1919subsys_initcall(send_signal_irq_work_init);
1920
1921#ifdef CONFIG_MODULES
1922static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1923 void *module)
1924{
1925 struct bpf_trace_module *btm, *tmp;
1926 struct module *mod = module;
1927 int ret = 0;
1928
1929 if (mod->num_bpf_raw_events == 0 ||
1930 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1931 goto out;
1932
1933 mutex_lock(&bpf_module_mutex);
1934
1935 switch (op) {
1936 case MODULE_STATE_COMING:
1937 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1938 if (btm) {
1939 btm->module = module;
1940 list_add(&btm->list, &bpf_trace_modules);
1941 } else {
1942 ret = -ENOMEM;
1943 }
1944 break;
1945 case MODULE_STATE_GOING:
1946 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1947 if (btm->module == module) {
1948 list_del(&btm->list);
1949 kfree(btm);
1950 break;
1951 }
1952 }
1953 break;
1954 }
1955
1956 mutex_unlock(&bpf_module_mutex);
1957
1958out:
1959 return notifier_from_errno(ret);
1960}
1961
1962static struct notifier_block bpf_module_nb = {
1963 .notifier_call = bpf_event_notify,
1964};
1965
1966static int __init bpf_event_init(void)
1967{
1968 register_module_notifier(&bpf_module_nb);
1969 return 0;
1970}
1971
1972fs_initcall(bpf_event_init);
1973#endif /* CONFIG_MODULES */