?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hugetlb: introduce generic version of hugetlb_free_pgd_range
[linux/fpc-iii.git] / kernel / trace / bpf_trace.c
blob08fcfe440c6374e336b02c1dde33a478fc58615a
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/filter.h>
11 #include <linux/uaccess.h>
12 #include <linux/ctype.h>
13 #include <linux/kprobes.h>
14 #include <linux/syscalls.h>
15 #include <linux/error-injection.h>
16
17 #include "trace_probe.h"
18 #include "trace.h"
19
20 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
21 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
22
23 /**
24 * trace_call_bpf - invoke BPF program
25 * @call: tracepoint event
26 * @ctx: opaque context pointer
27 *
28 * kprobe handlers execute BPF programs via this helper.
29 * Can be used from static tracepoints in the future.
30 *
31 * Return: BPF programs always return an integer which is interpreted by
32 * kprobe handler as:
33 * 0 - return from kprobe (event is filtered out)
34 * 1 - store kprobe event into ring buffer
35 * Other values are reserved and currently alias to 1
36 */
37 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
38 {
39 unsigned int ret;
40
41 if (in_nmi()) /* not supported yet */
42 return 1;
43
44 preempt_disable();
45
46 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
47 /*
48 * since some bpf program is already running on this cpu,
49 * don't call into another bpf program (same or different)
50 * and don't send kprobe event into ring-buffer,
51 * so return zero here
52 */
53 ret = 0;
54 goto out;
55 }
56
57 /*
58 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
59 * to all call sites, we did a bpf_prog_array_valid() there to check
60 * whether call->prog_array is empty or not, which is
61 * a heurisitc to speed up execution.
62 *
63 * If bpf_prog_array_valid() fetched prog_array was
64 * non-NULL, we go into trace_call_bpf() and do the actual
65 * proper rcu_dereference() under RCU lock.
66 * If it turns out that prog_array is NULL then, we bail out.
67 * For the opposite, if the bpf_prog_array_valid() fetched pointer
68 * was NULL, you'll skip the prog_array with the risk of missing
69 * out of events when it was updated in between this and the
70 * rcu_dereference() which is accepted risk.
71 */
72 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
73
74 out:
75 __this_cpu_dec(bpf_prog_active);
76 preempt_enable();
77
78 return ret;
79 }
80 EXPORT_SYMBOL_GPL(trace_call_bpf);
81
82 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
83 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
84 {
85 regs_set_return_value(regs, rc);
86 override_function_with_return(regs);
87 return 0;
88 }
89
90 static const struct bpf_func_proto bpf_override_return_proto = {
91 .func = bpf_override_return,
92 .gpl_only = true,
93 .ret_type = RET_INTEGER,
94 .arg1_type = ARG_PTR_TO_CTX,
95 .arg2_type = ARG_ANYTHING,
96 };
97 #endif
98
99 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
100 {
101 int ret;
102
103 ret = probe_kernel_read(dst, unsafe_ptr, size);
104 if (unlikely(ret < 0))
105 memset(dst, 0, size);
106
107 return ret;
108 }
109
110 static const struct bpf_func_proto bpf_probe_read_proto = {
111 .func = bpf_probe_read,
112 .gpl_only = true,
113 .ret_type = RET_INTEGER,
114 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
115 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
116 .arg3_type = ARG_ANYTHING,
117 };
118
119 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
120 u32, size)
121 {
122 /*
123 * Ensure we're in user context which is safe for the helper to
124 * run. This helper has no business in a kthread.
125 *
126 * access_ok() should prevent writing to non-user memory, but in
127 * some situations (nommu, temporary switch, etc) access_ok() does
128 * not provide enough validation, hence the check on KERNEL_DS.
129 */
130
131 if (unlikely(in_interrupt() ||
132 current->flags & (PF_KTHREAD | PF_EXITING)))
133 return -EPERM;
134 if (unlikely(uaccess_kernel()))
135 return -EPERM;
136 if (!access_ok(VERIFY_WRITE, unsafe_ptr, size))
137 return -EPERM;
138
139 return probe_kernel_write(unsafe_ptr, src, size);
140 }
141
142 static const struct bpf_func_proto bpf_probe_write_user_proto = {
143 .func = bpf_probe_write_user,
144 .gpl_only = true,
145 .ret_type = RET_INTEGER,
146 .arg1_type = ARG_ANYTHING,
147 .arg2_type = ARG_PTR_TO_MEM,
148 .arg3_type = ARG_CONST_SIZE,
149 };
150
151 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
152 {
153 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
154 current->comm, task_pid_nr(current));
155
156 return &bpf_probe_write_user_proto;
157 }
158
159 /*
160 * Only limited trace_printk() conversion specifiers allowed:
161 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
162 */
163 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
164 u64, arg2, u64, arg3)
165 {
166 bool str_seen = false;
167 int mod[3] = {};
168 int fmt_cnt = 0;
169 u64 unsafe_addr;
170 char buf[64];
171 int i;
172
173 /*
174 * bpf_check()->check_func_arg()->check_stack_boundary()
175 * guarantees that fmt points to bpf program stack,
176 * fmt_size bytes of it were initialized and fmt_size > 0
177 */
178 if (fmt[--fmt_size] != 0)
179 return -EINVAL;
180
181 /* check format string for allowed specifiers */
182 for (i = 0; i < fmt_size; i++) {
183 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
184 return -EINVAL;
185
186 if (fmt[i] != '%')
187 continue;
188
189 if (fmt_cnt >= 3)
190 return -EINVAL;
191
192 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
193 i++;
194 if (fmt[i] == 'l') {
195 mod[fmt_cnt]++;
196 i++;
197 } else if (fmt[i] == 'p' || fmt[i] == 's') {
198 mod[fmt_cnt]++;
199 i++;
200 if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
201 return -EINVAL;
202 fmt_cnt++;
203 if (fmt[i - 1] == 's') {
204 if (str_seen)
205 /* allow only one '%s' per fmt string */
206 return -EINVAL;
207 str_seen = true;
208
209 switch (fmt_cnt) {
210 case 1:
211 unsafe_addr = arg1;
212 arg1 = (long) buf;
213 break;
214 case 2:
215 unsafe_addr = arg2;
216 arg2 = (long) buf;
217 break;
218 case 3:
219 unsafe_addr = arg3;
220 arg3 = (long) buf;
221 break;
222 }
223 buf[0] = 0;
224 strncpy_from_unsafe(buf,
225 (void *) (long) unsafe_addr,
226 sizeof(buf));
227 }
228 continue;
229 }
230
231 if (fmt[i] == 'l') {
232 mod[fmt_cnt]++;
233 i++;
234 }
235
236 if (fmt[i] != 'i' && fmt[i] != 'd' &&
237 fmt[i] != 'u' && fmt[i] != 'x')
238 return -EINVAL;
239 fmt_cnt++;
240 }
241
242 /* Horrid workaround for getting va_list handling working with different
243 * argument type combinations generically for 32 and 64 bit archs.
244 */
245 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
246 #define __BPF_TP(...) \
247 __trace_printk(0 /* Fake ip */, \
248 fmt, ##__VA_ARGS__)
249
250 #define __BPF_ARG1_TP(...) \
251 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
252 ? __BPF_TP(arg1, ##__VA_ARGS__) \
253 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
254 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
255 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
256
257 #define __BPF_ARG2_TP(...) \
258 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
259 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
260 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
261 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
262 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
263
264 #define __BPF_ARG3_TP(...) \
265 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
266 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
267 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
268 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
269 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
270
271 return __BPF_TP_EMIT();
272 }
273
274 static const struct bpf_func_proto bpf_trace_printk_proto = {
275 .func = bpf_trace_printk,
276 .gpl_only = true,
277 .ret_type = RET_INTEGER,
278 .arg1_type = ARG_PTR_TO_MEM,
279 .arg2_type = ARG_CONST_SIZE,
280 };
281
282 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
283 {
284 /*
285 * this program might be calling bpf_trace_printk,
286 * so allocate per-cpu printk buffers
287 */
288 trace_printk_init_buffers();
289
290 return &bpf_trace_printk_proto;
291 }
292
293 static __always_inline int
294 get_map_perf_counter(struct bpf_map *map, u64 flags,
295 u64 *value, u64 *enabled, u64 *running)
296 {
297 struct bpf_array *array = container_of(map, struct bpf_array, map);
298 unsigned int cpu = smp_processor_id();
299 u64 index = flags & BPF_F_INDEX_MASK;
300 struct bpf_event_entry *ee;
301
302 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
303 return -EINVAL;
304 if (index == BPF_F_CURRENT_CPU)
305 index = cpu;
306 if (unlikely(index >= array->map.max_entries))
307 return -E2BIG;
308
309 ee = READ_ONCE(array->ptrs[index]);
310 if (!ee)
311 return -ENOENT;
312
313 return perf_event_read_local(ee->event, value, enabled, running);
314 }
315
316 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
317 {
318 u64 value = 0;
319 int err;
320
321 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
322 /*
323 * this api is ugly since we miss [-22..-2] range of valid
324 * counter values, but that's uapi
325 */
326 if (err)
327 return err;
328 return value;
329 }
330
331 static const struct bpf_func_proto bpf_perf_event_read_proto = {
332 .func = bpf_perf_event_read,
333 .gpl_only = true,
334 .ret_type = RET_INTEGER,
335 .arg1_type = ARG_CONST_MAP_PTR,
336 .arg2_type = ARG_ANYTHING,
337 };
338
339 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
340 struct bpf_perf_event_value *, buf, u32, size)
341 {
342 int err = -EINVAL;
343
344 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
345 goto clear;
346 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
347 &buf->running);
348 if (unlikely(err))
349 goto clear;
350 return 0;
351 clear:
352 memset(buf, 0, size);
353 return err;
354 }
355
356 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
357 .func = bpf_perf_event_read_value,
358 .gpl_only = true,
359 .ret_type = RET_INTEGER,
360 .arg1_type = ARG_CONST_MAP_PTR,
361 .arg2_type = ARG_ANYTHING,
362 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
363 .arg4_type = ARG_CONST_SIZE,
364 };
365
366 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd);
367
368 static __always_inline u64
369 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
370 u64 flags, struct perf_sample_data *sd)
371 {
372 struct bpf_array *array = container_of(map, struct bpf_array, map);
373 unsigned int cpu = smp_processor_id();
374 u64 index = flags & BPF_F_INDEX_MASK;
375 struct bpf_event_entry *ee;
376 struct perf_event *event;
377
378 if (index == BPF_F_CURRENT_CPU)
379 index = cpu;
380 if (unlikely(index >= array->map.max_entries))
381 return -E2BIG;
382
383 ee = READ_ONCE(array->ptrs[index]);
384 if (!ee)
385 return -ENOENT;
386
387 event = ee->event;
388 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
389 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
390 return -EINVAL;
391
392 if (unlikely(event->oncpu != cpu))
393 return -EOPNOTSUPP;
394
395 perf_event_output(event, sd, regs);
396 return 0;
397 }
398
399 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
400 u64, flags, void *, data, u64, size)
401 {
402 struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd);
403 struct perf_raw_record raw = {
404 .frag = {
405 .size = size,
406 .data = data,
407 },
408 };
409
410 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
411 return -EINVAL;
412
413 perf_sample_data_init(sd, 0, 0);
414 sd->raw = &raw;
415
416 return __bpf_perf_event_output(regs, map, flags, sd);
417 }
418
419 static const struct bpf_func_proto bpf_perf_event_output_proto = {
420 .func = bpf_perf_event_output,
421 .gpl_only = true,
422 .ret_type = RET_INTEGER,
423 .arg1_type = ARG_PTR_TO_CTX,
424 .arg2_type = ARG_CONST_MAP_PTR,
425 .arg3_type = ARG_ANYTHING,
426 .arg4_type = ARG_PTR_TO_MEM,
427 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
428 };
429
430 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
431 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd);
432
433 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
434 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
435 {
436 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd);
437 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
438 struct perf_raw_frag frag = {
439 .copy = ctx_copy,
440 .size = ctx_size,
441 .data = ctx,
442 };
443 struct perf_raw_record raw = {
444 .frag = {
445 {
446 .next = ctx_size ? &frag : NULL,
447 },
448 .size = meta_size,
449 .data = meta,
450 },
451 };
452
453 perf_fetch_caller_regs(regs);
454 perf_sample_data_init(sd, 0, 0);
455 sd->raw = &raw;
456
457 return __bpf_perf_event_output(regs, map, flags, sd);
458 }
459
460 BPF_CALL_0(bpf_get_current_task)
461 {
462 return (long) current;
463 }
464
465 static const struct bpf_func_proto bpf_get_current_task_proto = {
466 .func = bpf_get_current_task,
467 .gpl_only = true,
468 .ret_type = RET_INTEGER,
469 };
470
471 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
472 {
473 struct bpf_array *array = container_of(map, struct bpf_array, map);
474 struct cgroup *cgrp;
475
476 if (unlikely(idx >= array->map.max_entries))
477 return -E2BIG;
478
479 cgrp = READ_ONCE(array->ptrs[idx]);
480 if (unlikely(!cgrp))
481 return -EAGAIN;
482
483 return task_under_cgroup_hierarchy(current, cgrp);
484 }
485
486 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
487 .func = bpf_current_task_under_cgroup,
488 .gpl_only = false,
489 .ret_type = RET_INTEGER,
490 .arg1_type = ARG_CONST_MAP_PTR,
491 .arg2_type = ARG_ANYTHING,
492 };
493
494 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
495 const void *, unsafe_ptr)
496 {
497 int ret;
498
499 /*
500 * The strncpy_from_unsafe() call will likely not fill the entire
501 * buffer, but that's okay in this circumstance as we're probing
502 * arbitrary memory anyway similar to bpf_probe_read() and might
503 * as well probe the stack. Thus, memory is explicitly cleared
504 * only in error case, so that improper users ignoring return
505 * code altogether don't copy garbage; otherwise length of string
506 * is returned that can be used for bpf_perf_event_output() et al.
507 */
508 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
509 if (unlikely(ret < 0))
510 memset(dst, 0, size);
511
512 return ret;
513 }
514
515 static const struct bpf_func_proto bpf_probe_read_str_proto = {
516 .func = bpf_probe_read_str,
517 .gpl_only = true,
518 .ret_type = RET_INTEGER,
519 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
520 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
521 .arg3_type = ARG_ANYTHING,
522 };
523
524 static const struct bpf_func_proto *
525 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
526 {
527 switch (func_id) {
528 case BPF_FUNC_map_lookup_elem:
529 return &bpf_map_lookup_elem_proto;
530 case BPF_FUNC_map_update_elem:
531 return &bpf_map_update_elem_proto;
532 case BPF_FUNC_map_delete_elem:
533 return &bpf_map_delete_elem_proto;
534 case BPF_FUNC_probe_read:
535 return &bpf_probe_read_proto;
536 case BPF_FUNC_ktime_get_ns:
537 return &bpf_ktime_get_ns_proto;
538 case BPF_FUNC_tail_call:
539 return &bpf_tail_call_proto;
540 case BPF_FUNC_get_current_pid_tgid:
541 return &bpf_get_current_pid_tgid_proto;
542 case BPF_FUNC_get_current_task:
543 return &bpf_get_current_task_proto;
544 case BPF_FUNC_get_current_uid_gid:
545 return &bpf_get_current_uid_gid_proto;
546 case BPF_FUNC_get_current_comm:
547 return &bpf_get_current_comm_proto;
548 case BPF_FUNC_trace_printk:
549 return bpf_get_trace_printk_proto();
550 case BPF_FUNC_get_smp_processor_id:
551 return &bpf_get_smp_processor_id_proto;
552 case BPF_FUNC_get_numa_node_id:
553 return &bpf_get_numa_node_id_proto;
554 case BPF_FUNC_perf_event_read:
555 return &bpf_perf_event_read_proto;
556 case BPF_FUNC_probe_write_user:
557 return bpf_get_probe_write_proto();
558 case BPF_FUNC_current_task_under_cgroup:
559 return &bpf_current_task_under_cgroup_proto;
560 case BPF_FUNC_get_prandom_u32:
561 return &bpf_get_prandom_u32_proto;
562 case BPF_FUNC_probe_read_str:
563 return &bpf_probe_read_str_proto;
564 #ifdef CONFIG_CGROUPS
565 case BPF_FUNC_get_current_cgroup_id:
566 return &bpf_get_current_cgroup_id_proto;
567 #endif
568 default:
569 return NULL;
570 }
571 }
572
573 static const struct bpf_func_proto *
574 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
575 {
576 switch (func_id) {
577 case BPF_FUNC_perf_event_output:
578 return &bpf_perf_event_output_proto;
579 case BPF_FUNC_get_stackid:
580 return &bpf_get_stackid_proto;
581 case BPF_FUNC_get_stack:
582 return &bpf_get_stack_proto;
583 case BPF_FUNC_perf_event_read_value:
584 return &bpf_perf_event_read_value_proto;
585 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
586 case BPF_FUNC_override_return:
587 return &bpf_override_return_proto;
588 #endif
589 default:
590 return tracing_func_proto(func_id, prog);
591 }
592 }
593
594 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
595 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
596 const struct bpf_prog *prog,
597 struct bpf_insn_access_aux *info)
598 {
599 if (off < 0 || off >= sizeof(struct pt_regs))
600 return false;
601 if (type != BPF_READ)
602 return false;
603 if (off % size != 0)
604 return false;
605 /*
606 * Assertion for 32 bit to make sure last 8 byte access
607 * (BPF_DW) to the last 4 byte member is disallowed.
608 */
609 if (off + size > sizeof(struct pt_regs))
610 return false;
611
612 return true;
613 }
614
615 const struct bpf_verifier_ops kprobe_verifier_ops = {
616 .get_func_proto = kprobe_prog_func_proto,
617 .is_valid_access = kprobe_prog_is_valid_access,
618 };
619
620 const struct bpf_prog_ops kprobe_prog_ops = {
621 };
622
623 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
624 u64, flags, void *, data, u64, size)
625 {
626 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
627
628 /*
629 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
630 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
631 * from there and call the same bpf_perf_event_output() helper inline.
632 */
633 return ____bpf_perf_event_output(regs, map, flags, data, size);
634 }
635
636 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
637 .func = bpf_perf_event_output_tp,
638 .gpl_only = true,
639 .ret_type = RET_INTEGER,
640 .arg1_type = ARG_PTR_TO_CTX,
641 .arg2_type = ARG_CONST_MAP_PTR,
642 .arg3_type = ARG_ANYTHING,
643 .arg4_type = ARG_PTR_TO_MEM,
644 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
645 };
646
647 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
648 u64, flags)
649 {
650 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
651
652 /*
653 * Same comment as in bpf_perf_event_output_tp(), only that this time
654 * the other helper's function body cannot be inlined due to being
655 * external, thus we need to call raw helper function.
656 */
657 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
658 flags, 0, 0);
659 }
660
661 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
662 .func = bpf_get_stackid_tp,
663 .gpl_only = true,
664 .ret_type = RET_INTEGER,
665 .arg1_type = ARG_PTR_TO_CTX,
666 .arg2_type = ARG_CONST_MAP_PTR,
667 .arg3_type = ARG_ANYTHING,
668 };
669
670 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
671 u64, flags)
672 {
673 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
674
675 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
676 (unsigned long) size, flags, 0);
677 }
678
679 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
680 .func = bpf_get_stack_tp,
681 .gpl_only = true,
682 .ret_type = RET_INTEGER,
683 .arg1_type = ARG_PTR_TO_CTX,
684 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
685 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
686 .arg4_type = ARG_ANYTHING,
687 };
688
689 static const struct bpf_func_proto *
690 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
691 {
692 switch (func_id) {
693 case BPF_FUNC_perf_event_output:
694 return &bpf_perf_event_output_proto_tp;
695 case BPF_FUNC_get_stackid:
696 return &bpf_get_stackid_proto_tp;
697 case BPF_FUNC_get_stack:
698 return &bpf_get_stack_proto_tp;
699 default:
700 return tracing_func_proto(func_id, prog);
701 }
702 }
703
704 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
705 const struct bpf_prog *prog,
706 struct bpf_insn_access_aux *info)
707 {
708 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
709 return false;
710 if (type != BPF_READ)
711 return false;
712 if (off % size != 0)
713 return false;
714
715 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
716 return true;
717 }
718
719 const struct bpf_verifier_ops tracepoint_verifier_ops = {
720 .get_func_proto = tp_prog_func_proto,
721 .is_valid_access = tp_prog_is_valid_access,
722 };
723
724 const struct bpf_prog_ops tracepoint_prog_ops = {
725 };
726
727 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
728 struct bpf_perf_event_value *, buf, u32, size)
729 {
730 int err = -EINVAL;
731
732 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
733 goto clear;
734 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
735 &buf->running);
736 if (unlikely(err))
737 goto clear;
738 return 0;
739 clear:
740 memset(buf, 0, size);
741 return err;
742 }
743
744 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
745 .func = bpf_perf_prog_read_value,
746 .gpl_only = true,
747 .ret_type = RET_INTEGER,
748 .arg1_type = ARG_PTR_TO_CTX,
749 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
750 .arg3_type = ARG_CONST_SIZE,
751 };
752
753 static const struct bpf_func_proto *
754 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
755 {
756 switch (func_id) {
757 case BPF_FUNC_perf_event_output:
758 return &bpf_perf_event_output_proto_tp;
759 case BPF_FUNC_get_stackid:
760 return &bpf_get_stackid_proto_tp;
761 case BPF_FUNC_get_stack:
762 return &bpf_get_stack_proto_tp;
763 case BPF_FUNC_perf_prog_read_value:
764 return &bpf_perf_prog_read_value_proto;
765 default:
766 return tracing_func_proto(func_id, prog);
767 }
768 }
769
770 /*
771 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
772 * to avoid potential recursive reuse issue when/if tracepoints are added
773 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack
774 */
775 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs);
776 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
777 struct bpf_map *, map, u64, flags, void *, data, u64, size)
778 {
779 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
780
781 perf_fetch_caller_regs(regs);
782 return ____bpf_perf_event_output(regs, map, flags, data, size);
783 }
784
785 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
786 .func = bpf_perf_event_output_raw_tp,
787 .gpl_only = true,
788 .ret_type = RET_INTEGER,
789 .arg1_type = ARG_PTR_TO_CTX,
790 .arg2_type = ARG_CONST_MAP_PTR,
791 .arg3_type = ARG_ANYTHING,
792 .arg4_type = ARG_PTR_TO_MEM,
793 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
794 };
795
796 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
797 struct bpf_map *, map, u64, flags)
798 {
799 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
800
801 perf_fetch_caller_regs(regs);
802 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
803 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
804 flags, 0, 0);
805 }
806
807 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
808 .func = bpf_get_stackid_raw_tp,
809 .gpl_only = true,
810 .ret_type = RET_INTEGER,
811 .arg1_type = ARG_PTR_TO_CTX,
812 .arg2_type = ARG_CONST_MAP_PTR,
813 .arg3_type = ARG_ANYTHING,
814 };
815
816 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
817 void *, buf, u32, size, u64, flags)
818 {
819 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs);
820
821 perf_fetch_caller_regs(regs);
822 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
823 (unsigned long) size, flags, 0);
824 }
825
826 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
827 .func = bpf_get_stack_raw_tp,
828 .gpl_only = true,
829 .ret_type = RET_INTEGER,
830 .arg1_type = ARG_PTR_TO_CTX,
831 .arg2_type = ARG_PTR_TO_MEM,
832 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
833 .arg4_type = ARG_ANYTHING,
834 };
835
836 static const struct bpf_func_proto *
837 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
838 {
839 switch (func_id) {
840 case BPF_FUNC_perf_event_output:
841 return &bpf_perf_event_output_proto_raw_tp;
842 case BPF_FUNC_get_stackid:
843 return &bpf_get_stackid_proto_raw_tp;
844 case BPF_FUNC_get_stack:
845 return &bpf_get_stack_proto_raw_tp;
846 default:
847 return tracing_func_proto(func_id, prog);
848 }
849 }
850
851 static bool raw_tp_prog_is_valid_access(int off, int size,
852 enum bpf_access_type type,
853 const struct bpf_prog *prog,
854 struct bpf_insn_access_aux *info)
855 {
856 /* largest tracepoint in the kernel has 12 args */
857 if (off < 0 || off >= sizeof(__u64) * 12)
858 return false;
859 if (type != BPF_READ)
860 return false;
861 if (off % size != 0)
862 return false;
863 return true;
864 }
865
866 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
867 .get_func_proto = raw_tp_prog_func_proto,
868 .is_valid_access = raw_tp_prog_is_valid_access,
869 };
870
871 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
872 };
873
874 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
875 const struct bpf_prog *prog,
876 struct bpf_insn_access_aux *info)
877 {
878 const int size_u64 = sizeof(u64);
879
880 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
881 return false;
882 if (type != BPF_READ)
883 return false;
884 if (off % size != 0) {
885 if (sizeof(unsigned long) != 4)
886 return false;
887 if (size != 8)
888 return false;
889 if (off % size != 4)
890 return false;
891 }
892
893 switch (off) {
894 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
895 bpf_ctx_record_field_size(info, size_u64);
896 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
897 return false;
898 break;
899 case bpf_ctx_range(struct bpf_perf_event_data, addr):
900 bpf_ctx_record_field_size(info, size_u64);
901 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
902 return false;
903 break;
904 default:
905 if (size != sizeof(long))
906 return false;
907 }
908
909 return true;
910 }
911
912 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
913 const struct bpf_insn *si,
914 struct bpf_insn *insn_buf,
915 struct bpf_prog *prog, u32 *target_size)
916 {
917 struct bpf_insn *insn = insn_buf;
918
919 switch (si->off) {
920 case offsetof(struct bpf_perf_event_data, sample_period):
921 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
922 data), si->dst_reg, si->src_reg,
923 offsetof(struct bpf_perf_event_data_kern, data));
924 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
925 bpf_target_off(struct perf_sample_data, period, 8,
926 target_size));
927 break;
928 case offsetof(struct bpf_perf_event_data, addr):
929 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
930 data), si->dst_reg, si->src_reg,
931 offsetof(struct bpf_perf_event_data_kern, data));
932 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
933 bpf_target_off(struct perf_sample_data, addr, 8,
934 target_size));
935 break;
936 default:
937 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
938 regs), si->dst_reg, si->src_reg,
939 offsetof(struct bpf_perf_event_data_kern, regs));
940 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
941 si->off);
942 break;
943 }
944
945 return insn - insn_buf;
946 }
947
948 const struct bpf_verifier_ops perf_event_verifier_ops = {
949 .get_func_proto = pe_prog_func_proto,
950 .is_valid_access = pe_prog_is_valid_access,
951 .convert_ctx_access = pe_prog_convert_ctx_access,
952 };
953
954 const struct bpf_prog_ops perf_event_prog_ops = {
955 };
956
957 static DEFINE_MUTEX(bpf_event_mutex);
958
959 #define BPF_TRACE_MAX_PROGS 64
960
961 int perf_event_attach_bpf_prog(struct perf_event *event,
962 struct bpf_prog *prog)
963 {
964 struct bpf_prog_array __rcu *old_array;
965 struct bpf_prog_array *new_array;
966 int ret = -EEXIST;
967
968 /*
969 * Kprobe override only works if they are on the function entry,
970 * and only if they are on the opt-in list.
971 */
972 if (prog->kprobe_override &&
973 (!trace_kprobe_on_func_entry(event->tp_event) ||
974 !trace_kprobe_error_injectable(event->tp_event)))
975 return -EINVAL;
976
977 mutex_lock(&bpf_event_mutex);
978
979 if (event->prog)
980 goto unlock;
981
982 old_array = event->tp_event->prog_array;
983 if (old_array &&
984 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
985 ret = -E2BIG;
986 goto unlock;
987 }
988
989 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
990 if (ret < 0)
991 goto unlock;
992
993 /* set the new array to event->tp_event and set event->prog */
994 event->prog = prog;
995 rcu_assign_pointer(event->tp_event->prog_array, new_array);
996 bpf_prog_array_free(old_array);
997
998 unlock:
999 mutex_unlock(&bpf_event_mutex);
1000 return ret;
1001 }
1002
1003 void perf_event_detach_bpf_prog(struct perf_event *event)
1004 {
1005 struct bpf_prog_array __rcu *old_array;
1006 struct bpf_prog_array *new_array;
1007 int ret;
1008
1009 mutex_lock(&bpf_event_mutex);
1010
1011 if (!event->prog)
1012 goto unlock;
1013
1014 old_array = event->tp_event->prog_array;
1015 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1016 if (ret == -ENOENT)
1017 goto unlock;
1018 if (ret < 0) {
1019 bpf_prog_array_delete_safe(old_array, event->prog);
1020 } else {
1021 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1022 bpf_prog_array_free(old_array);
1023 }
1024
1025 bpf_prog_put(event->prog);
1026 event->prog = NULL;
1027
1028 unlock:
1029 mutex_unlock(&bpf_event_mutex);
1030 }
1031
1032 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1033 {
1034 struct perf_event_query_bpf __user *uquery = info;
1035 struct perf_event_query_bpf query = {};
1036 u32 *ids, prog_cnt, ids_len;
1037 int ret;
1038
1039 if (!capable(CAP_SYS_ADMIN))
1040 return -EPERM;
1041 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1042 return -EINVAL;
1043 if (copy_from_user(&query, uquery, sizeof(query)))
1044 return -EFAULT;
1045
1046 ids_len = query.ids_len;
1047 if (ids_len > BPF_TRACE_MAX_PROGS)
1048 return -E2BIG;
1049 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1050 if (!ids)
1051 return -ENOMEM;
1052 /*
1053 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1054 * is required when user only wants to check for uquery->prog_cnt.
1055 * There is no need to check for it since the case is handled
1056 * gracefully in bpf_prog_array_copy_info.
1057 */
1058
1059 mutex_lock(&bpf_event_mutex);
1060 ret = bpf_prog_array_copy_info(event->tp_event->prog_array,
1061 ids,
1062 ids_len,
1063 &prog_cnt);
1064 mutex_unlock(&bpf_event_mutex);
1065
1066 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1067 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1068 ret = -EFAULT;
1069
1070 kfree(ids);
1071 return ret;
1072 }
1073
1074 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1075 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1076
1077 struct bpf_raw_event_map *bpf_find_raw_tracepoint(const char *name)
1078 {
1079 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1080
1081 for (; btp < __stop__bpf_raw_tp; btp++) {
1082 if (!strcmp(btp->tp->name, name))
1083 return btp;
1084 }
1085 return NULL;
1086 }
1087
1088 static __always_inline
1089 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1090 {
1091 rcu_read_lock();
1092 preempt_disable();
1093 (void) BPF_PROG_RUN(prog, args);
1094 preempt_enable();
1095 rcu_read_unlock();
1096 }
1097
1098 #define UNPACK(...) __VA_ARGS__
1099 #define REPEAT_1(FN, DL, X, ...) FN(X)
1100 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1101 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1102 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1103 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1104 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1105 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1106 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1107 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1108 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1109 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1110 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1111 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1112
1113 #define SARG(X) u64 arg##X
1114 #define COPY(X) args[X] = arg##X
1115
1116 #define __DL_COM (,)
1117 #define __DL_SEM (;)
1118
1119 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1120
1121 #define BPF_TRACE_DEFN_x(x) \
1122 void bpf_trace_run##x(struct bpf_prog *prog, \
1123 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1124 { \
1125 u64 args[x]; \
1126 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1127 __bpf_trace_run(prog, args); \
1128 } \
1129 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1130 BPF_TRACE_DEFN_x(1);
1131 BPF_TRACE_DEFN_x(2);
1132 BPF_TRACE_DEFN_x(3);
1133 BPF_TRACE_DEFN_x(4);
1134 BPF_TRACE_DEFN_x(5);
1135 BPF_TRACE_DEFN_x(6);
1136 BPF_TRACE_DEFN_x(7);
1137 BPF_TRACE_DEFN_x(8);
1138 BPF_TRACE_DEFN_x(9);
1139 BPF_TRACE_DEFN_x(10);
1140 BPF_TRACE_DEFN_x(11);
1141 BPF_TRACE_DEFN_x(12);
1142
1143 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1144 {
1145 struct tracepoint *tp = btp->tp;
1146
1147 /*
1148 * check that program doesn't access arguments beyond what's
1149 * available in this tracepoint
1150 */
1151 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1152 return -EINVAL;
1153
1154 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1155 }
1156
1157 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1158 {
1159 int err;
1160
1161 mutex_lock(&bpf_event_mutex);
1162 err = __bpf_probe_register(btp, prog);
1163 mutex_unlock(&bpf_event_mutex);
1164 return err;
1165 }
1166
1167 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1168 {
1169 int err;
1170
1171 mutex_lock(&bpf_event_mutex);
1172 err = tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1173 mutex_unlock(&bpf_event_mutex);
1174 return err;
1175 }
1176
1177 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1178 u32 *fd_type, const char **buf,
1179 u64 *probe_offset, u64 *probe_addr)
1180 {
1181 bool is_tracepoint, is_syscall_tp;
1182 struct bpf_prog *prog;
1183 int flags, err = 0;
1184
1185 prog = event->prog;
1186 if (!prog)
1187 return -ENOENT;
1188
1189 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1190 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1191 return -EOPNOTSUPP;
1192
1193 *prog_id = prog->aux->id;
1194 flags = event->tp_event->flags;
1195 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1196 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1197
1198 if (is_tracepoint || is_syscall_tp) {
1199 *buf = is_tracepoint ? event->tp_event->tp->name
1200 : event->tp_event->name;
1201 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1202 *probe_offset = 0x0;
1203 *probe_addr = 0x0;
1204 } else {
1205 /* kprobe/uprobe */
1206 err = -EOPNOTSUPP;
1207 #ifdef CONFIG_KPROBE_EVENTS
1208 if (flags & TRACE_EVENT_FL_KPROBE)
1209 err = bpf_get_kprobe_info(event, fd_type, buf,
1210 probe_offset, probe_addr,
1211 event->attr.type == PERF_TYPE_TRACEPOINT);
1212 #endif
1213 #ifdef CONFIG_UPROBE_EVENTS
1214 if (flags & TRACE_EVENT_FL_UPROBE)
1215 err = bpf_get_uprobe_info(event, fd_type, buf,
1216 probe_offset,
1217 event->attr.type == PERF_TYPE_TRACEPOINT);
1218 #endif
1219 }
1220
1221 return err;
1222 }
Linux with added FPC-III support