xref: /openbmc/linux/kernel/trace/bpf_trace.c (revision 3e30a927)
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 <asm/tlb.h>
18 
19 #include "trace_probe.h"
20 #include "trace.h"
21 
22 #define bpf_event_rcu_dereference(p)					\
23 	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
24 
25 #ifdef CONFIG_MODULES
26 struct bpf_trace_module {
27 	struct module *module;
28 	struct list_head list;
29 };
30 
31 static LIST_HEAD(bpf_trace_modules);
32 static DEFINE_MUTEX(bpf_module_mutex);
33 
34 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
35 {
36 	struct bpf_raw_event_map *btp, *ret = NULL;
37 	struct bpf_trace_module *btm;
38 	unsigned int i;
39 
40 	mutex_lock(&bpf_module_mutex);
41 	list_for_each_entry(btm, &bpf_trace_modules, list) {
42 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
43 			btp = &btm->module->bpf_raw_events[i];
44 			if (!strcmp(btp->tp->name, name)) {
45 				if (try_module_get(btm->module))
46 					ret = btp;
47 				goto out;
48 			}
49 		}
50 	}
51 out:
52 	mutex_unlock(&bpf_module_mutex);
53 	return ret;
54 }
55 #else
56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
57 {
58 	return NULL;
59 }
60 #endif /* CONFIG_MODULES */
61 
62 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
63 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
64 
65 /**
66  * trace_call_bpf - invoke BPF program
67  * @call: tracepoint event
68  * @ctx: opaque context pointer
69  *
70  * kprobe handlers execute BPF programs via this helper.
71  * Can be used from static tracepoints in the future.
72  *
73  * Return: BPF programs always return an integer which is interpreted by
74  * kprobe handler as:
75  * 0 - return from kprobe (event is filtered out)
76  * 1 - store kprobe event into ring buffer
77  * Other values are reserved and currently alias to 1
78  */
79 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
80 {
81 	unsigned int ret;
82 
83 	if (in_nmi()) /* not supported yet */
84 		return 1;
85 
86 	preempt_disable();
87 
88 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
89 		/*
90 		 * since some bpf program is already running on this cpu,
91 		 * don't call into another bpf program (same or different)
92 		 * and don't send kprobe event into ring-buffer,
93 		 * so return zero here
94 		 */
95 		ret = 0;
96 		goto out;
97 	}
98 
99 	/*
100 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
101 	 * to all call sites, we did a bpf_prog_array_valid() there to check
102 	 * whether call->prog_array is empty or not, which is
103 	 * a heurisitc to speed up execution.
104 	 *
105 	 * If bpf_prog_array_valid() fetched prog_array was
106 	 * non-NULL, we go into trace_call_bpf() and do the actual
107 	 * proper rcu_dereference() under RCU lock.
108 	 * If it turns out that prog_array is NULL then, we bail out.
109 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
110 	 * was NULL, you'll skip the prog_array with the risk of missing
111 	 * out of events when it was updated in between this and the
112 	 * rcu_dereference() which is accepted risk.
113 	 */
114 	ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
115 
116  out:
117 	__this_cpu_dec(bpf_prog_active);
118 	preempt_enable();
119 
120 	return ret;
121 }
122 EXPORT_SYMBOL_GPL(trace_call_bpf);
123 
124 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
125 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
126 {
127 	regs_set_return_value(regs, rc);
128 	override_function_with_return(regs);
129 	return 0;
130 }
131 
132 static const struct bpf_func_proto bpf_override_return_proto = {
133 	.func		= bpf_override_return,
134 	.gpl_only	= true,
135 	.ret_type	= RET_INTEGER,
136 	.arg1_type	= ARG_PTR_TO_CTX,
137 	.arg2_type	= ARG_ANYTHING,
138 };
139 #endif
140 
141 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
142 	   const void __user *, unsafe_ptr)
143 {
144 	int ret = probe_user_read(dst, unsafe_ptr, size);
145 
146 	if (unlikely(ret < 0))
147 		memset(dst, 0, size);
148 
149 	return ret;
150 }
151 
152 static const struct bpf_func_proto bpf_probe_read_user_proto = {
153 	.func		= bpf_probe_read_user,
154 	.gpl_only	= true,
155 	.ret_type	= RET_INTEGER,
156 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
157 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
158 	.arg3_type	= ARG_ANYTHING,
159 };
160 
161 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
162 	   const void __user *, unsafe_ptr)
163 {
164 	int ret = strncpy_from_unsafe_user(dst, unsafe_ptr, size);
165 
166 	if (unlikely(ret < 0))
167 		memset(dst, 0, size);
168 
169 	return ret;
170 }
171 
172 static const struct bpf_func_proto bpf_probe_read_user_str_proto = {
173 	.func		= bpf_probe_read_user_str,
174 	.gpl_only	= true,
175 	.ret_type	= RET_INTEGER,
176 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
177 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
178 	.arg3_type	= ARG_ANYTHING,
179 };
180 
181 static __always_inline int
182 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr,
183 			     const bool compat)
184 {
185 	int ret = security_locked_down(LOCKDOWN_BPF_READ);
186 
187 	if (unlikely(ret < 0))
188 		goto out;
189 	ret = compat ? probe_kernel_read(dst, unsafe_ptr, size) :
190 	      probe_kernel_read_strict(dst, unsafe_ptr, size);
191 	if (unlikely(ret < 0))
192 out:
193 		memset(dst, 0, size);
194 	return ret;
195 }
196 
197 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
198 	   const void *, unsafe_ptr)
199 {
200 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, false);
201 }
202 
203 static const struct bpf_func_proto bpf_probe_read_kernel_proto = {
204 	.func		= bpf_probe_read_kernel,
205 	.gpl_only	= true,
206 	.ret_type	= RET_INTEGER,
207 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
208 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
209 	.arg3_type	= ARG_ANYTHING,
210 };
211 
212 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
213 	   const void *, unsafe_ptr)
214 {
215 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, true);
216 }
217 
218 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
219 	.func		= bpf_probe_read_compat,
220 	.gpl_only	= true,
221 	.ret_type	= RET_INTEGER,
222 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
223 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
224 	.arg3_type	= ARG_ANYTHING,
225 };
226 
227 static __always_inline int
228 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr,
229 				 const bool compat)
230 {
231 	int ret = security_locked_down(LOCKDOWN_BPF_READ);
232 
233 	if (unlikely(ret < 0))
234 		goto out;
235 	/*
236 	 * The strncpy_from_unsafe_*() call will likely not fill the entire
237 	 * buffer, but that's okay in this circumstance as we're probing
238 	 * arbitrary memory anyway similar to bpf_probe_read_*() and might
239 	 * as well probe the stack. Thus, memory is explicitly cleared
240 	 * only in error case, so that improper users ignoring return
241 	 * code altogether don't copy garbage; otherwise length of string
242 	 * is returned that can be used for bpf_perf_event_output() et al.
243 	 */
244 	ret = compat ? strncpy_from_unsafe(dst, unsafe_ptr, size) :
245 	      strncpy_from_unsafe_strict(dst, unsafe_ptr, size);
246 	if (unlikely(ret < 0))
247 out:
248 		memset(dst, 0, size);
249 	return ret;
250 }
251 
252 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
253 	   const void *, unsafe_ptr)
254 {
255 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, false);
256 }
257 
258 static const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
259 	.func		= bpf_probe_read_kernel_str,
260 	.gpl_only	= true,
261 	.ret_type	= RET_INTEGER,
262 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
263 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
264 	.arg3_type	= ARG_ANYTHING,
265 };
266 
267 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
268 	   const void *, unsafe_ptr)
269 {
270 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, true);
271 }
272 
273 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
274 	.func		= bpf_probe_read_compat_str,
275 	.gpl_only	= true,
276 	.ret_type	= RET_INTEGER,
277 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
278 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
279 	.arg3_type	= ARG_ANYTHING,
280 };
281 
282 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
283 	   u32, size)
284 {
285 	/*
286 	 * Ensure we're in user context which is safe for the helper to
287 	 * run. This helper has no business in a kthread.
288 	 *
289 	 * access_ok() should prevent writing to non-user memory, but in
290 	 * some situations (nommu, temporary switch, etc) access_ok() does
291 	 * not provide enough validation, hence the check on KERNEL_DS.
292 	 *
293 	 * nmi_uaccess_okay() ensures the probe is not run in an interim
294 	 * state, when the task or mm are switched. This is specifically
295 	 * required to prevent the use of temporary mm.
296 	 */
297 
298 	if (unlikely(in_interrupt() ||
299 		     current->flags & (PF_KTHREAD | PF_EXITING)))
300 		return -EPERM;
301 	if (unlikely(uaccess_kernel()))
302 		return -EPERM;
303 	if (unlikely(!nmi_uaccess_okay()))
304 		return -EPERM;
305 
306 	return probe_user_write(unsafe_ptr, src, size);
307 }
308 
309 static const struct bpf_func_proto bpf_probe_write_user_proto = {
310 	.func		= bpf_probe_write_user,
311 	.gpl_only	= true,
312 	.ret_type	= RET_INTEGER,
313 	.arg1_type	= ARG_ANYTHING,
314 	.arg2_type	= ARG_PTR_TO_MEM,
315 	.arg3_type	= ARG_CONST_SIZE,
316 };
317 
318 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
319 {
320 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
321 			    current->comm, task_pid_nr(current));
322 
323 	return &bpf_probe_write_user_proto;
324 }
325 
326 /*
327  * Only limited trace_printk() conversion specifiers allowed:
328  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
329  */
330 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
331 	   u64, arg2, u64, arg3)
332 {
333 	bool str_seen = false;
334 	int mod[3] = {};
335 	int fmt_cnt = 0;
336 	u64 unsafe_addr;
337 	char buf[64];
338 	int i;
339 
340 	/*
341 	 * bpf_check()->check_func_arg()->check_stack_boundary()
342 	 * guarantees that fmt points to bpf program stack,
343 	 * fmt_size bytes of it were initialized and fmt_size > 0
344 	 */
345 	if (fmt[--fmt_size] != 0)
346 		return -EINVAL;
347 
348 	/* check format string for allowed specifiers */
349 	for (i = 0; i < fmt_size; i++) {
350 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
351 			return -EINVAL;
352 
353 		if (fmt[i] != '%')
354 			continue;
355 
356 		if (fmt_cnt >= 3)
357 			return -EINVAL;
358 
359 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
360 		i++;
361 		if (fmt[i] == 'l') {
362 			mod[fmt_cnt]++;
363 			i++;
364 		} else if (fmt[i] == 'p' || fmt[i] == 's') {
365 			mod[fmt_cnt]++;
366 			/* disallow any further format extensions */
367 			if (fmt[i + 1] != 0 &&
368 			    !isspace(fmt[i + 1]) &&
369 			    !ispunct(fmt[i + 1]))
370 				return -EINVAL;
371 			fmt_cnt++;
372 			if (fmt[i] == 's') {
373 				if (str_seen)
374 					/* allow only one '%s' per fmt string */
375 					return -EINVAL;
376 				str_seen = true;
377 
378 				switch (fmt_cnt) {
379 				case 1:
380 					unsafe_addr = arg1;
381 					arg1 = (long) buf;
382 					break;
383 				case 2:
384 					unsafe_addr = arg2;
385 					arg2 = (long) buf;
386 					break;
387 				case 3:
388 					unsafe_addr = arg3;
389 					arg3 = (long) buf;
390 					break;
391 				}
392 				buf[0] = 0;
393 				strncpy_from_unsafe(buf,
394 						    (void *) (long) unsafe_addr,
395 						    sizeof(buf));
396 			}
397 			continue;
398 		}
399 
400 		if (fmt[i] == 'l') {
401 			mod[fmt_cnt]++;
402 			i++;
403 		}
404 
405 		if (fmt[i] != 'i' && fmt[i] != 'd' &&
406 		    fmt[i] != 'u' && fmt[i] != 'x')
407 			return -EINVAL;
408 		fmt_cnt++;
409 	}
410 
411 /* Horrid workaround for getting va_list handling working with different
412  * argument type combinations generically for 32 and 64 bit archs.
413  */
414 #define __BPF_TP_EMIT()	__BPF_ARG3_TP()
415 #define __BPF_TP(...)							\
416 	__trace_printk(0 /* Fake ip */,					\
417 		       fmt, ##__VA_ARGS__)
418 
419 #define __BPF_ARG1_TP(...)						\
420 	((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))	\
421 	  ? __BPF_TP(arg1, ##__VA_ARGS__)				\
422 	  : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))	\
423 	      ? __BPF_TP((long)arg1, ##__VA_ARGS__)			\
424 	      : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
425 
426 #define __BPF_ARG2_TP(...)						\
427 	((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))	\
428 	  ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)				\
429 	  : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))	\
430 	      ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)		\
431 	      : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
432 
433 #define __BPF_ARG3_TP(...)						\
434 	((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))	\
435 	  ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)				\
436 	  : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))	\
437 	      ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)		\
438 	      : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
439 
440 	return __BPF_TP_EMIT();
441 }
442 
443 static const struct bpf_func_proto bpf_trace_printk_proto = {
444 	.func		= bpf_trace_printk,
445 	.gpl_only	= true,
446 	.ret_type	= RET_INTEGER,
447 	.arg1_type	= ARG_PTR_TO_MEM,
448 	.arg2_type	= ARG_CONST_SIZE,
449 };
450 
451 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
452 {
453 	/*
454 	 * this program might be calling bpf_trace_printk,
455 	 * so allocate per-cpu printk buffers
456 	 */
457 	trace_printk_init_buffers();
458 
459 	return &bpf_trace_printk_proto;
460 }
461 
462 static __always_inline int
463 get_map_perf_counter(struct bpf_map *map, u64 flags,
464 		     u64 *value, u64 *enabled, u64 *running)
465 {
466 	struct bpf_array *array = container_of(map, struct bpf_array, map);
467 	unsigned int cpu = smp_processor_id();
468 	u64 index = flags & BPF_F_INDEX_MASK;
469 	struct bpf_event_entry *ee;
470 
471 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
472 		return -EINVAL;
473 	if (index == BPF_F_CURRENT_CPU)
474 		index = cpu;
475 	if (unlikely(index >= array->map.max_entries))
476 		return -E2BIG;
477 
478 	ee = READ_ONCE(array->ptrs[index]);
479 	if (!ee)
480 		return -ENOENT;
481 
482 	return perf_event_read_local(ee->event, value, enabled, running);
483 }
484 
485 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
486 {
487 	u64 value = 0;
488 	int err;
489 
490 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
491 	/*
492 	 * this api is ugly since we miss [-22..-2] range of valid
493 	 * counter values, but that's uapi
494 	 */
495 	if (err)
496 		return err;
497 	return value;
498 }
499 
500 static const struct bpf_func_proto bpf_perf_event_read_proto = {
501 	.func		= bpf_perf_event_read,
502 	.gpl_only	= true,
503 	.ret_type	= RET_INTEGER,
504 	.arg1_type	= ARG_CONST_MAP_PTR,
505 	.arg2_type	= ARG_ANYTHING,
506 };
507 
508 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
509 	   struct bpf_perf_event_value *, buf, u32, size)
510 {
511 	int err = -EINVAL;
512 
513 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
514 		goto clear;
515 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
516 				   &buf->running);
517 	if (unlikely(err))
518 		goto clear;
519 	return 0;
520 clear:
521 	memset(buf, 0, size);
522 	return err;
523 }
524 
525 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
526 	.func		= bpf_perf_event_read_value,
527 	.gpl_only	= true,
528 	.ret_type	= RET_INTEGER,
529 	.arg1_type	= ARG_CONST_MAP_PTR,
530 	.arg2_type	= ARG_ANYTHING,
531 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
532 	.arg4_type	= ARG_CONST_SIZE,
533 };
534 
535 static __always_inline u64
536 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
537 			u64 flags, struct perf_sample_data *sd)
538 {
539 	struct bpf_array *array = container_of(map, struct bpf_array, map);
540 	unsigned int cpu = smp_processor_id();
541 	u64 index = flags & BPF_F_INDEX_MASK;
542 	struct bpf_event_entry *ee;
543 	struct perf_event *event;
544 
545 	if (index == BPF_F_CURRENT_CPU)
546 		index = cpu;
547 	if (unlikely(index >= array->map.max_entries))
548 		return -E2BIG;
549 
550 	ee = READ_ONCE(array->ptrs[index]);
551 	if (!ee)
552 		return -ENOENT;
553 
554 	event = ee->event;
555 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
556 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
557 		return -EINVAL;
558 
559 	if (unlikely(event->oncpu != cpu))
560 		return -EOPNOTSUPP;
561 
562 	return perf_event_output(event, sd, regs);
563 }
564 
565 /*
566  * Support executing tracepoints in normal, irq, and nmi context that each call
567  * bpf_perf_event_output
568  */
569 struct bpf_trace_sample_data {
570 	struct perf_sample_data sds[3];
571 };
572 
573 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
574 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
575 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
576 	   u64, flags, void *, data, u64, size)
577 {
578 	struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
579 	int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
580 	struct perf_raw_record raw = {
581 		.frag = {
582 			.size = size,
583 			.data = data,
584 		},
585 	};
586 	struct perf_sample_data *sd;
587 	int err;
588 
589 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
590 		err = -EBUSY;
591 		goto out;
592 	}
593 
594 	sd = &sds->sds[nest_level - 1];
595 
596 	if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
597 		err = -EINVAL;
598 		goto out;
599 	}
600 
601 	perf_sample_data_init(sd, 0, 0);
602 	sd->raw = &raw;
603 
604 	err = __bpf_perf_event_output(regs, map, flags, sd);
605 
606 out:
607 	this_cpu_dec(bpf_trace_nest_level);
608 	return err;
609 }
610 
611 static const struct bpf_func_proto bpf_perf_event_output_proto = {
612 	.func		= bpf_perf_event_output,
613 	.gpl_only	= true,
614 	.ret_type	= RET_INTEGER,
615 	.arg1_type	= ARG_PTR_TO_CTX,
616 	.arg2_type	= ARG_CONST_MAP_PTR,
617 	.arg3_type	= ARG_ANYTHING,
618 	.arg4_type	= ARG_PTR_TO_MEM,
619 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
620 };
621 
622 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
623 struct bpf_nested_pt_regs {
624 	struct pt_regs regs[3];
625 };
626 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
627 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
628 
629 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
630 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
631 {
632 	int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
633 	struct perf_raw_frag frag = {
634 		.copy		= ctx_copy,
635 		.size		= ctx_size,
636 		.data		= ctx,
637 	};
638 	struct perf_raw_record raw = {
639 		.frag = {
640 			{
641 				.next	= ctx_size ? &frag : NULL,
642 			},
643 			.size	= meta_size,
644 			.data	= meta,
645 		},
646 	};
647 	struct perf_sample_data *sd;
648 	struct pt_regs *regs;
649 	u64 ret;
650 
651 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
652 		ret = -EBUSY;
653 		goto out;
654 	}
655 	sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
656 	regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
657 
658 	perf_fetch_caller_regs(regs);
659 	perf_sample_data_init(sd, 0, 0);
660 	sd->raw = &raw;
661 
662 	ret = __bpf_perf_event_output(regs, map, flags, sd);
663 out:
664 	this_cpu_dec(bpf_event_output_nest_level);
665 	return ret;
666 }
667 
668 BPF_CALL_0(bpf_get_current_task)
669 {
670 	return (long) current;
671 }
672 
673 static const struct bpf_func_proto bpf_get_current_task_proto = {
674 	.func		= bpf_get_current_task,
675 	.gpl_only	= true,
676 	.ret_type	= RET_INTEGER,
677 };
678 
679 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
680 {
681 	struct bpf_array *array = container_of(map, struct bpf_array, map);
682 	struct cgroup *cgrp;
683 
684 	if (unlikely(idx >= array->map.max_entries))
685 		return -E2BIG;
686 
687 	cgrp = READ_ONCE(array->ptrs[idx]);
688 	if (unlikely(!cgrp))
689 		return -EAGAIN;
690 
691 	return task_under_cgroup_hierarchy(current, cgrp);
692 }
693 
694 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
695 	.func           = bpf_current_task_under_cgroup,
696 	.gpl_only       = false,
697 	.ret_type       = RET_INTEGER,
698 	.arg1_type      = ARG_CONST_MAP_PTR,
699 	.arg2_type      = ARG_ANYTHING,
700 };
701 
702 struct send_signal_irq_work {
703 	struct irq_work irq_work;
704 	struct task_struct *task;
705 	u32 sig;
706 	enum pid_type type;
707 };
708 
709 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
710 
711 static void do_bpf_send_signal(struct irq_work *entry)
712 {
713 	struct send_signal_irq_work *work;
714 
715 	work = container_of(entry, struct send_signal_irq_work, irq_work);
716 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
717 }
718 
719 static int bpf_send_signal_common(u32 sig, enum pid_type type)
720 {
721 	struct send_signal_irq_work *work = NULL;
722 
723 	/* Similar to bpf_probe_write_user, task needs to be
724 	 * in a sound condition and kernel memory access be
725 	 * permitted in order to send signal to the current
726 	 * task.
727 	 */
728 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
729 		return -EPERM;
730 	if (unlikely(uaccess_kernel()))
731 		return -EPERM;
732 	if (unlikely(!nmi_uaccess_okay()))
733 		return -EPERM;
734 
735 	if (in_nmi()) {
736 		/* Do an early check on signal validity. Otherwise,
737 		 * the error is lost in deferred irq_work.
738 		 */
739 		if (unlikely(!valid_signal(sig)))
740 			return -EINVAL;
741 
742 		work = this_cpu_ptr(&send_signal_work);
743 		if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
744 			return -EBUSY;
745 
746 		/* Add the current task, which is the target of sending signal,
747 		 * to the irq_work. The current task may change when queued
748 		 * irq works get executed.
749 		 */
750 		work->task = current;
751 		work->sig = sig;
752 		work->type = type;
753 		irq_work_queue(&work->irq_work);
754 		return 0;
755 	}
756 
757 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
758 }
759 
760 BPF_CALL_1(bpf_send_signal, u32, sig)
761 {
762 	return bpf_send_signal_common(sig, PIDTYPE_TGID);
763 }
764 
765 static const struct bpf_func_proto bpf_send_signal_proto = {
766 	.func		= bpf_send_signal,
767 	.gpl_only	= false,
768 	.ret_type	= RET_INTEGER,
769 	.arg1_type	= ARG_ANYTHING,
770 };
771 
772 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
773 {
774 	return bpf_send_signal_common(sig, PIDTYPE_PID);
775 }
776 
777 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
778 	.func		= bpf_send_signal_thread,
779 	.gpl_only	= false,
780 	.ret_type	= RET_INTEGER,
781 	.arg1_type	= ARG_ANYTHING,
782 };
783 
784 static const struct bpf_func_proto *
785 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
786 {
787 	switch (func_id) {
788 	case BPF_FUNC_map_lookup_elem:
789 		return &bpf_map_lookup_elem_proto;
790 	case BPF_FUNC_map_update_elem:
791 		return &bpf_map_update_elem_proto;
792 	case BPF_FUNC_map_delete_elem:
793 		return &bpf_map_delete_elem_proto;
794 	case BPF_FUNC_map_push_elem:
795 		return &bpf_map_push_elem_proto;
796 	case BPF_FUNC_map_pop_elem:
797 		return &bpf_map_pop_elem_proto;
798 	case BPF_FUNC_map_peek_elem:
799 		return &bpf_map_peek_elem_proto;
800 	case BPF_FUNC_ktime_get_ns:
801 		return &bpf_ktime_get_ns_proto;
802 	case BPF_FUNC_tail_call:
803 		return &bpf_tail_call_proto;
804 	case BPF_FUNC_get_current_pid_tgid:
805 		return &bpf_get_current_pid_tgid_proto;
806 	case BPF_FUNC_get_current_task:
807 		return &bpf_get_current_task_proto;
808 	case BPF_FUNC_get_current_uid_gid:
809 		return &bpf_get_current_uid_gid_proto;
810 	case BPF_FUNC_get_current_comm:
811 		return &bpf_get_current_comm_proto;
812 	case BPF_FUNC_trace_printk:
813 		return bpf_get_trace_printk_proto();
814 	case BPF_FUNC_get_smp_processor_id:
815 		return &bpf_get_smp_processor_id_proto;
816 	case BPF_FUNC_get_numa_node_id:
817 		return &bpf_get_numa_node_id_proto;
818 	case BPF_FUNC_perf_event_read:
819 		return &bpf_perf_event_read_proto;
820 	case BPF_FUNC_probe_write_user:
821 		return bpf_get_probe_write_proto();
822 	case BPF_FUNC_current_task_under_cgroup:
823 		return &bpf_current_task_under_cgroup_proto;
824 	case BPF_FUNC_get_prandom_u32:
825 		return &bpf_get_prandom_u32_proto;
826 	case BPF_FUNC_probe_read_user:
827 		return &bpf_probe_read_user_proto;
828 	case BPF_FUNC_probe_read_kernel:
829 		return &bpf_probe_read_kernel_proto;
830 	case BPF_FUNC_probe_read:
831 		return &bpf_probe_read_compat_proto;
832 	case BPF_FUNC_probe_read_user_str:
833 		return &bpf_probe_read_user_str_proto;
834 	case BPF_FUNC_probe_read_kernel_str:
835 		return &bpf_probe_read_kernel_str_proto;
836 	case BPF_FUNC_probe_read_str:
837 		return &bpf_probe_read_compat_str_proto;
838 #ifdef CONFIG_CGROUPS
839 	case BPF_FUNC_get_current_cgroup_id:
840 		return &bpf_get_current_cgroup_id_proto;
841 #endif
842 	case BPF_FUNC_send_signal:
843 		return &bpf_send_signal_proto;
844 	case BPF_FUNC_send_signal_thread:
845 		return &bpf_send_signal_thread_proto;
846 	default:
847 		return NULL;
848 	}
849 }
850 
851 static const struct bpf_func_proto *
852 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
853 {
854 	switch (func_id) {
855 	case BPF_FUNC_perf_event_output:
856 		return &bpf_perf_event_output_proto;
857 	case BPF_FUNC_get_stackid:
858 		return &bpf_get_stackid_proto;
859 	case BPF_FUNC_get_stack:
860 		return &bpf_get_stack_proto;
861 	case BPF_FUNC_perf_event_read_value:
862 		return &bpf_perf_event_read_value_proto;
863 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
864 	case BPF_FUNC_override_return:
865 		return &bpf_override_return_proto;
866 #endif
867 	default:
868 		return tracing_func_proto(func_id, prog);
869 	}
870 }
871 
872 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
873 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
874 					const struct bpf_prog *prog,
875 					struct bpf_insn_access_aux *info)
876 {
877 	if (off < 0 || off >= sizeof(struct pt_regs))
878 		return false;
879 	if (type != BPF_READ)
880 		return false;
881 	if (off % size != 0)
882 		return false;
883 	/*
884 	 * Assertion for 32 bit to make sure last 8 byte access
885 	 * (BPF_DW) to the last 4 byte member is disallowed.
886 	 */
887 	if (off + size > sizeof(struct pt_regs))
888 		return false;
889 
890 	return true;
891 }
892 
893 const struct bpf_verifier_ops kprobe_verifier_ops = {
894 	.get_func_proto  = kprobe_prog_func_proto,
895 	.is_valid_access = kprobe_prog_is_valid_access,
896 };
897 
898 const struct bpf_prog_ops kprobe_prog_ops = {
899 };
900 
901 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
902 	   u64, flags, void *, data, u64, size)
903 {
904 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
905 
906 	/*
907 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
908 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
909 	 * from there and call the same bpf_perf_event_output() helper inline.
910 	 */
911 	return ____bpf_perf_event_output(regs, map, flags, data, size);
912 }
913 
914 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
915 	.func		= bpf_perf_event_output_tp,
916 	.gpl_only	= true,
917 	.ret_type	= RET_INTEGER,
918 	.arg1_type	= ARG_PTR_TO_CTX,
919 	.arg2_type	= ARG_CONST_MAP_PTR,
920 	.arg3_type	= ARG_ANYTHING,
921 	.arg4_type	= ARG_PTR_TO_MEM,
922 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
923 };
924 
925 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
926 	   u64, flags)
927 {
928 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
929 
930 	/*
931 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
932 	 * the other helper's function body cannot be inlined due to being
933 	 * external, thus we need to call raw helper function.
934 	 */
935 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
936 			       flags, 0, 0);
937 }
938 
939 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
940 	.func		= bpf_get_stackid_tp,
941 	.gpl_only	= true,
942 	.ret_type	= RET_INTEGER,
943 	.arg1_type	= ARG_PTR_TO_CTX,
944 	.arg2_type	= ARG_CONST_MAP_PTR,
945 	.arg3_type	= ARG_ANYTHING,
946 };
947 
948 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
949 	   u64, flags)
950 {
951 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
952 
953 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
954 			     (unsigned long) size, flags, 0);
955 }
956 
957 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
958 	.func		= bpf_get_stack_tp,
959 	.gpl_only	= true,
960 	.ret_type	= RET_INTEGER,
961 	.arg1_type	= ARG_PTR_TO_CTX,
962 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
963 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
964 	.arg4_type	= ARG_ANYTHING,
965 };
966 
967 static const struct bpf_func_proto *
968 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
969 {
970 	switch (func_id) {
971 	case BPF_FUNC_perf_event_output:
972 		return &bpf_perf_event_output_proto_tp;
973 	case BPF_FUNC_get_stackid:
974 		return &bpf_get_stackid_proto_tp;
975 	case BPF_FUNC_get_stack:
976 		return &bpf_get_stack_proto_tp;
977 	default:
978 		return tracing_func_proto(func_id, prog);
979 	}
980 }
981 
982 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
983 				    const struct bpf_prog *prog,
984 				    struct bpf_insn_access_aux *info)
985 {
986 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
987 		return false;
988 	if (type != BPF_READ)
989 		return false;
990 	if (off % size != 0)
991 		return false;
992 
993 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
994 	return true;
995 }
996 
997 const struct bpf_verifier_ops tracepoint_verifier_ops = {
998 	.get_func_proto  = tp_prog_func_proto,
999 	.is_valid_access = tp_prog_is_valid_access,
1000 };
1001 
1002 const struct bpf_prog_ops tracepoint_prog_ops = {
1003 };
1004 
1005 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1006 	   struct bpf_perf_event_value *, buf, u32, size)
1007 {
1008 	int err = -EINVAL;
1009 
1010 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1011 		goto clear;
1012 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1013 				    &buf->running);
1014 	if (unlikely(err))
1015 		goto clear;
1016 	return 0;
1017 clear:
1018 	memset(buf, 0, size);
1019 	return err;
1020 }
1021 
1022 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1023          .func           = bpf_perf_prog_read_value,
1024          .gpl_only       = true,
1025          .ret_type       = RET_INTEGER,
1026          .arg1_type      = ARG_PTR_TO_CTX,
1027          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1028          .arg3_type      = ARG_CONST_SIZE,
1029 };
1030 
1031 static const struct bpf_func_proto *
1032 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1033 {
1034 	switch (func_id) {
1035 	case BPF_FUNC_perf_event_output:
1036 		return &bpf_perf_event_output_proto_tp;
1037 	case BPF_FUNC_get_stackid:
1038 		return &bpf_get_stackid_proto_tp;
1039 	case BPF_FUNC_get_stack:
1040 		return &bpf_get_stack_proto_tp;
1041 	case BPF_FUNC_perf_prog_read_value:
1042 		return &bpf_perf_prog_read_value_proto;
1043 	default:
1044 		return tracing_func_proto(func_id, prog);
1045 	}
1046 }
1047 
1048 /*
1049  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1050  * to avoid potential recursive reuse issue when/if tracepoints are added
1051  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1052  *
1053  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1054  * in normal, irq, and nmi context.
1055  */
1056 struct bpf_raw_tp_regs {
1057 	struct pt_regs regs[3];
1058 };
1059 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1060 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1061 static struct pt_regs *get_bpf_raw_tp_regs(void)
1062 {
1063 	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1064 	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1065 
1066 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1067 		this_cpu_dec(bpf_raw_tp_nest_level);
1068 		return ERR_PTR(-EBUSY);
1069 	}
1070 
1071 	return &tp_regs->regs[nest_level - 1];
1072 }
1073 
1074 static void put_bpf_raw_tp_regs(void)
1075 {
1076 	this_cpu_dec(bpf_raw_tp_nest_level);
1077 }
1078 
1079 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1080 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1081 {
1082 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1083 	int ret;
1084 
1085 	if (IS_ERR(regs))
1086 		return PTR_ERR(regs);
1087 
1088 	perf_fetch_caller_regs(regs);
1089 	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1090 
1091 	put_bpf_raw_tp_regs();
1092 	return ret;
1093 }
1094 
1095 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1096 	.func		= bpf_perf_event_output_raw_tp,
1097 	.gpl_only	= true,
1098 	.ret_type	= RET_INTEGER,
1099 	.arg1_type	= ARG_PTR_TO_CTX,
1100 	.arg2_type	= ARG_CONST_MAP_PTR,
1101 	.arg3_type	= ARG_ANYTHING,
1102 	.arg4_type	= ARG_PTR_TO_MEM,
1103 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1104 };
1105 
1106 extern const struct bpf_func_proto bpf_skb_output_proto;
1107 
1108 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1109 	   struct bpf_map *, map, u64, flags)
1110 {
1111 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1112 	int ret;
1113 
1114 	if (IS_ERR(regs))
1115 		return PTR_ERR(regs);
1116 
1117 	perf_fetch_caller_regs(regs);
1118 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1119 	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1120 			      flags, 0, 0);
1121 	put_bpf_raw_tp_regs();
1122 	return ret;
1123 }
1124 
1125 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1126 	.func		= bpf_get_stackid_raw_tp,
1127 	.gpl_only	= true,
1128 	.ret_type	= RET_INTEGER,
1129 	.arg1_type	= ARG_PTR_TO_CTX,
1130 	.arg2_type	= ARG_CONST_MAP_PTR,
1131 	.arg3_type	= ARG_ANYTHING,
1132 };
1133 
1134 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1135 	   void *, buf, u32, size, u64, flags)
1136 {
1137 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1138 	int ret;
1139 
1140 	if (IS_ERR(regs))
1141 		return PTR_ERR(regs);
1142 
1143 	perf_fetch_caller_regs(regs);
1144 	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1145 			    (unsigned long) size, flags, 0);
1146 	put_bpf_raw_tp_regs();
1147 	return ret;
1148 }
1149 
1150 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1151 	.func		= bpf_get_stack_raw_tp,
1152 	.gpl_only	= true,
1153 	.ret_type	= RET_INTEGER,
1154 	.arg1_type	= ARG_PTR_TO_CTX,
1155 	.arg2_type	= ARG_PTR_TO_MEM,
1156 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1157 	.arg4_type	= ARG_ANYTHING,
1158 };
1159 
1160 static const struct bpf_func_proto *
1161 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1162 {
1163 	switch (func_id) {
1164 	case BPF_FUNC_perf_event_output:
1165 		return &bpf_perf_event_output_proto_raw_tp;
1166 	case BPF_FUNC_get_stackid:
1167 		return &bpf_get_stackid_proto_raw_tp;
1168 	case BPF_FUNC_get_stack:
1169 		return &bpf_get_stack_proto_raw_tp;
1170 	default:
1171 		return tracing_func_proto(func_id, prog);
1172 	}
1173 }
1174 
1175 static const struct bpf_func_proto *
1176 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1177 {
1178 	switch (func_id) {
1179 #ifdef CONFIG_NET
1180 	case BPF_FUNC_skb_output:
1181 		return &bpf_skb_output_proto;
1182 #endif
1183 	default:
1184 		return raw_tp_prog_func_proto(func_id, prog);
1185 	}
1186 }
1187 
1188 static bool raw_tp_prog_is_valid_access(int off, int size,
1189 					enum bpf_access_type type,
1190 					const struct bpf_prog *prog,
1191 					struct bpf_insn_access_aux *info)
1192 {
1193 	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1194 		return false;
1195 	if (type != BPF_READ)
1196 		return false;
1197 	if (off % size != 0)
1198 		return false;
1199 	return true;
1200 }
1201 
1202 static bool tracing_prog_is_valid_access(int off, int size,
1203 					 enum bpf_access_type type,
1204 					 const struct bpf_prog *prog,
1205 					 struct bpf_insn_access_aux *info)
1206 {
1207 	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1208 		return false;
1209 	if (type != BPF_READ)
1210 		return false;
1211 	if (off % size != 0)
1212 		return false;
1213 	return btf_ctx_access(off, size, type, prog, info);
1214 }
1215 
1216 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1217 	.get_func_proto  = raw_tp_prog_func_proto,
1218 	.is_valid_access = raw_tp_prog_is_valid_access,
1219 };
1220 
1221 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1222 };
1223 
1224 const struct bpf_verifier_ops tracing_verifier_ops = {
1225 	.get_func_proto  = tracing_prog_func_proto,
1226 	.is_valid_access = tracing_prog_is_valid_access,
1227 };
1228 
1229 const struct bpf_prog_ops tracing_prog_ops = {
1230 };
1231 
1232 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1233 						 enum bpf_access_type type,
1234 						 const struct bpf_prog *prog,
1235 						 struct bpf_insn_access_aux *info)
1236 {
1237 	if (off == 0) {
1238 		if (size != sizeof(u64) || type != BPF_READ)
1239 			return false;
1240 		info->reg_type = PTR_TO_TP_BUFFER;
1241 	}
1242 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1243 }
1244 
1245 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1246 	.get_func_proto  = raw_tp_prog_func_proto,
1247 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
1248 };
1249 
1250 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1251 };
1252 
1253 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1254 				    const struct bpf_prog *prog,
1255 				    struct bpf_insn_access_aux *info)
1256 {
1257 	const int size_u64 = sizeof(u64);
1258 
1259 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1260 		return false;
1261 	if (type != BPF_READ)
1262 		return false;
1263 	if (off % size != 0) {
1264 		if (sizeof(unsigned long) != 4)
1265 			return false;
1266 		if (size != 8)
1267 			return false;
1268 		if (off % size != 4)
1269 			return false;
1270 	}
1271 
1272 	switch (off) {
1273 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1274 		bpf_ctx_record_field_size(info, size_u64);
1275 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1276 			return false;
1277 		break;
1278 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
1279 		bpf_ctx_record_field_size(info, size_u64);
1280 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1281 			return false;
1282 		break;
1283 	default:
1284 		if (size != sizeof(long))
1285 			return false;
1286 	}
1287 
1288 	return true;
1289 }
1290 
1291 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1292 				      const struct bpf_insn *si,
1293 				      struct bpf_insn *insn_buf,
1294 				      struct bpf_prog *prog, u32 *target_size)
1295 {
1296 	struct bpf_insn *insn = insn_buf;
1297 
1298 	switch (si->off) {
1299 	case offsetof(struct bpf_perf_event_data, sample_period):
1300 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1301 						       data), si->dst_reg, si->src_reg,
1302 				      offsetof(struct bpf_perf_event_data_kern, data));
1303 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1304 				      bpf_target_off(struct perf_sample_data, period, 8,
1305 						     target_size));
1306 		break;
1307 	case offsetof(struct bpf_perf_event_data, addr):
1308 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1309 						       data), si->dst_reg, si->src_reg,
1310 				      offsetof(struct bpf_perf_event_data_kern, data));
1311 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1312 				      bpf_target_off(struct perf_sample_data, addr, 8,
1313 						     target_size));
1314 		break;
1315 	default:
1316 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1317 						       regs), si->dst_reg, si->src_reg,
1318 				      offsetof(struct bpf_perf_event_data_kern, regs));
1319 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1320 				      si->off);
1321 		break;
1322 	}
1323 
1324 	return insn - insn_buf;
1325 }
1326 
1327 const struct bpf_verifier_ops perf_event_verifier_ops = {
1328 	.get_func_proto		= pe_prog_func_proto,
1329 	.is_valid_access	= pe_prog_is_valid_access,
1330 	.convert_ctx_access	= pe_prog_convert_ctx_access,
1331 };
1332 
1333 const struct bpf_prog_ops perf_event_prog_ops = {
1334 };
1335 
1336 static DEFINE_MUTEX(bpf_event_mutex);
1337 
1338 #define BPF_TRACE_MAX_PROGS 64
1339 
1340 int perf_event_attach_bpf_prog(struct perf_event *event,
1341 			       struct bpf_prog *prog)
1342 {
1343 	struct bpf_prog_array *old_array;
1344 	struct bpf_prog_array *new_array;
1345 	int ret = -EEXIST;
1346 
1347 	/*
1348 	 * Kprobe override only works if they are on the function entry,
1349 	 * and only if they are on the opt-in list.
1350 	 */
1351 	if (prog->kprobe_override &&
1352 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1353 	     !trace_kprobe_error_injectable(event->tp_event)))
1354 		return -EINVAL;
1355 
1356 	mutex_lock(&bpf_event_mutex);
1357 
1358 	if (event->prog)
1359 		goto unlock;
1360 
1361 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1362 	if (old_array &&
1363 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1364 		ret = -E2BIG;
1365 		goto unlock;
1366 	}
1367 
1368 	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1369 	if (ret < 0)
1370 		goto unlock;
1371 
1372 	/* set the new array to event->tp_event and set event->prog */
1373 	event->prog = prog;
1374 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1375 	bpf_prog_array_free(old_array);
1376 
1377 unlock:
1378 	mutex_unlock(&bpf_event_mutex);
1379 	return ret;
1380 }
1381 
1382 void perf_event_detach_bpf_prog(struct perf_event *event)
1383 {
1384 	struct bpf_prog_array *old_array;
1385 	struct bpf_prog_array *new_array;
1386 	int ret;
1387 
1388 	mutex_lock(&bpf_event_mutex);
1389 
1390 	if (!event->prog)
1391 		goto unlock;
1392 
1393 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1394 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1395 	if (ret == -ENOENT)
1396 		goto unlock;
1397 	if (ret < 0) {
1398 		bpf_prog_array_delete_safe(old_array, event->prog);
1399 	} else {
1400 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1401 		bpf_prog_array_free(old_array);
1402 	}
1403 
1404 	bpf_prog_put(event->prog);
1405 	event->prog = NULL;
1406 
1407 unlock:
1408 	mutex_unlock(&bpf_event_mutex);
1409 }
1410 
1411 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1412 {
1413 	struct perf_event_query_bpf __user *uquery = info;
1414 	struct perf_event_query_bpf query = {};
1415 	struct bpf_prog_array *progs;
1416 	u32 *ids, prog_cnt, ids_len;
1417 	int ret;
1418 
1419 	if (!capable(CAP_SYS_ADMIN))
1420 		return -EPERM;
1421 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1422 		return -EINVAL;
1423 	if (copy_from_user(&query, uquery, sizeof(query)))
1424 		return -EFAULT;
1425 
1426 	ids_len = query.ids_len;
1427 	if (ids_len > BPF_TRACE_MAX_PROGS)
1428 		return -E2BIG;
1429 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1430 	if (!ids)
1431 		return -ENOMEM;
1432 	/*
1433 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1434 	 * is required when user only wants to check for uquery->prog_cnt.
1435 	 * There is no need to check for it since the case is handled
1436 	 * gracefully in bpf_prog_array_copy_info.
1437 	 */
1438 
1439 	mutex_lock(&bpf_event_mutex);
1440 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1441 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1442 	mutex_unlock(&bpf_event_mutex);
1443 
1444 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1445 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1446 		ret = -EFAULT;
1447 
1448 	kfree(ids);
1449 	return ret;
1450 }
1451 
1452 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1453 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1454 
1455 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1456 {
1457 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1458 
1459 	for (; btp < __stop__bpf_raw_tp; btp++) {
1460 		if (!strcmp(btp->tp->name, name))
1461 			return btp;
1462 	}
1463 
1464 	return bpf_get_raw_tracepoint_module(name);
1465 }
1466 
1467 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1468 {
1469 	struct module *mod = __module_address((unsigned long)btp);
1470 
1471 	if (mod)
1472 		module_put(mod);
1473 }
1474 
1475 static __always_inline
1476 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1477 {
1478 	rcu_read_lock();
1479 	preempt_disable();
1480 	(void) BPF_PROG_RUN(prog, args);
1481 	preempt_enable();
1482 	rcu_read_unlock();
1483 }
1484 
1485 #define UNPACK(...)			__VA_ARGS__
1486 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1487 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1488 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1489 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1490 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1491 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1492 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1493 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1494 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1495 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1496 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1497 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1498 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1499 
1500 #define SARG(X)		u64 arg##X
1501 #define COPY(X)		args[X] = arg##X
1502 
1503 #define __DL_COM	(,)
1504 #define __DL_SEM	(;)
1505 
1506 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1507 
1508 #define BPF_TRACE_DEFN_x(x)						\
1509 	void bpf_trace_run##x(struct bpf_prog *prog,			\
1510 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1511 	{								\
1512 		u64 args[x];						\
1513 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1514 		__bpf_trace_run(prog, args);				\
1515 	}								\
1516 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1517 BPF_TRACE_DEFN_x(1);
1518 BPF_TRACE_DEFN_x(2);
1519 BPF_TRACE_DEFN_x(3);
1520 BPF_TRACE_DEFN_x(4);
1521 BPF_TRACE_DEFN_x(5);
1522 BPF_TRACE_DEFN_x(6);
1523 BPF_TRACE_DEFN_x(7);
1524 BPF_TRACE_DEFN_x(8);
1525 BPF_TRACE_DEFN_x(9);
1526 BPF_TRACE_DEFN_x(10);
1527 BPF_TRACE_DEFN_x(11);
1528 BPF_TRACE_DEFN_x(12);
1529 
1530 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1531 {
1532 	struct tracepoint *tp = btp->tp;
1533 
1534 	/*
1535 	 * check that program doesn't access arguments beyond what's
1536 	 * available in this tracepoint
1537 	 */
1538 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1539 		return -EINVAL;
1540 
1541 	if (prog->aux->max_tp_access > btp->writable_size)
1542 		return -EINVAL;
1543 
1544 	return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1545 }
1546 
1547 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1548 {
1549 	return __bpf_probe_register(btp, prog);
1550 }
1551 
1552 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1553 {
1554 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1555 }
1556 
1557 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1558 			    u32 *fd_type, const char **buf,
1559 			    u64 *probe_offset, u64 *probe_addr)
1560 {
1561 	bool is_tracepoint, is_syscall_tp;
1562 	struct bpf_prog *prog;
1563 	int flags, err = 0;
1564 
1565 	prog = event->prog;
1566 	if (!prog)
1567 		return -ENOENT;
1568 
1569 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1570 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1571 		return -EOPNOTSUPP;
1572 
1573 	*prog_id = prog->aux->id;
1574 	flags = event->tp_event->flags;
1575 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1576 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
1577 
1578 	if (is_tracepoint || is_syscall_tp) {
1579 		*buf = is_tracepoint ? event->tp_event->tp->name
1580 				     : event->tp_event->name;
1581 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
1582 		*probe_offset = 0x0;
1583 		*probe_addr = 0x0;
1584 	} else {
1585 		/* kprobe/uprobe */
1586 		err = -EOPNOTSUPP;
1587 #ifdef CONFIG_KPROBE_EVENTS
1588 		if (flags & TRACE_EVENT_FL_KPROBE)
1589 			err = bpf_get_kprobe_info(event, fd_type, buf,
1590 						  probe_offset, probe_addr,
1591 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1592 #endif
1593 #ifdef CONFIG_UPROBE_EVENTS
1594 		if (flags & TRACE_EVENT_FL_UPROBE)
1595 			err = bpf_get_uprobe_info(event, fd_type, buf,
1596 						  probe_offset,
1597 						  event->attr.type == PERF_TYPE_TRACEPOINT);
1598 #endif
1599 	}
1600 
1601 	return err;
1602 }
1603 
1604 static int __init send_signal_irq_work_init(void)
1605 {
1606 	int cpu;
1607 	struct send_signal_irq_work *work;
1608 
1609 	for_each_possible_cpu(cpu) {
1610 		work = per_cpu_ptr(&send_signal_work, cpu);
1611 		init_irq_work(&work->irq_work, do_bpf_send_signal);
1612 	}
1613 	return 0;
1614 }
1615 
1616 subsys_initcall(send_signal_irq_work_init);
1617 
1618 #ifdef CONFIG_MODULES
1619 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1620 			    void *module)
1621 {
1622 	struct bpf_trace_module *btm, *tmp;
1623 	struct module *mod = module;
1624 
1625 	if (mod->num_bpf_raw_events == 0 ||
1626 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1627 		return 0;
1628 
1629 	mutex_lock(&bpf_module_mutex);
1630 
1631 	switch (op) {
1632 	case MODULE_STATE_COMING:
1633 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1634 		if (btm) {
1635 			btm->module = module;
1636 			list_add(&btm->list, &bpf_trace_modules);
1637 		}
1638 		break;
1639 	case MODULE_STATE_GOING:
1640 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1641 			if (btm->module == module) {
1642 				list_del(&btm->list);
1643 				kfree(btm);
1644 				break;
1645 			}
1646 		}
1647 		break;
1648 	}
1649 
1650 	mutex_unlock(&bpf_module_mutex);
1651 
1652 	return 0;
1653 }
1654 
1655 static struct notifier_block bpf_module_nb = {
1656 	.notifier_call = bpf_event_notify,
1657 };
1658 
1659 static int __init bpf_event_init(void)
1660 {
1661 	register_module_notifier(&bpf_module_nb);
1662 	return 0;
1663 }
1664 
1665 fs_initcall(bpf_event_init);
1666 #endif /* CONFIG_MODULES */
1667