xref: /linux/kernel/trace/bpf_trace.c (revision baaa68a9796ef2cadfe5caaf4c730412eda0f31c)
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
38 struct bpf_trace_module {
39 	struct module *module;
40 	struct list_head list;
41 };
42 
43 static LIST_HEAD(bpf_trace_modules);
44 static DEFINE_MUTEX(bpf_module_mutex);
45 
46 static 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 	}
63 out:
64 	mutex_unlock(&bpf_module_mutex);
65 	return ret;
66 }
67 #else
68 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
69 {
70 	return NULL;
71 }
72 #endif /* CONFIG_MODULES */
73 
74 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
75 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
76 
77 static 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  */
95 unsigned 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(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
136 BPF_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 
143 static 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 
152 static __always_inline int
153 bpf_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 
163 BPF_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 
169 const 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 
178 static __always_inline int
179 bpf_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 
200 BPF_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 
206 const 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 
215 static __always_inline int
216 bpf_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 
226 BPF_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 
232 const 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 
241 static __always_inline int
242 bpf_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 
261 BPF_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 
267 const 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
277 BPF_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 
287 static 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 
296 BPF_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 
306 static 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 
316 BPF_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(!nmi_uaccess_okay()))
336 		return -EPERM;
337 
338 	return copy_to_user_nofault(unsafe_ptr, src, size);
339 }
340 
341 static const struct bpf_func_proto bpf_probe_write_user_proto = {
342 	.func		= bpf_probe_write_user,
343 	.gpl_only	= true,
344 	.ret_type	= RET_INTEGER,
345 	.arg1_type	= ARG_ANYTHING,
346 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
347 	.arg3_type	= ARG_CONST_SIZE,
348 };
349 
350 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
351 {
352 	if (!capable(CAP_SYS_ADMIN))
353 		return NULL;
354 
355 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
356 			    current->comm, task_pid_nr(current));
357 
358 	return &bpf_probe_write_user_proto;
359 }
360 
361 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
362 
363 #define MAX_TRACE_PRINTK_VARARGS	3
364 #define BPF_TRACE_PRINTK_SIZE		1024
365 
366 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
367 	   u64, arg2, u64, arg3)
368 {
369 	u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
370 	u32 *bin_args;
371 	static char buf[BPF_TRACE_PRINTK_SIZE];
372 	unsigned long flags;
373 	int ret;
374 
375 	ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
376 				  MAX_TRACE_PRINTK_VARARGS);
377 	if (ret < 0)
378 		return ret;
379 
380 	raw_spin_lock_irqsave(&trace_printk_lock, flags);
381 	ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
382 
383 	trace_bpf_trace_printk(buf);
384 	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
385 
386 	bpf_bprintf_cleanup();
387 
388 	return ret;
389 }
390 
391 static const struct bpf_func_proto bpf_trace_printk_proto = {
392 	.func		= bpf_trace_printk,
393 	.gpl_only	= true,
394 	.ret_type	= RET_INTEGER,
395 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
396 	.arg2_type	= ARG_CONST_SIZE,
397 };
398 
399 static void __set_printk_clr_event(void)
400 {
401 	/*
402 	 * This program might be calling bpf_trace_printk,
403 	 * so enable the associated bpf_trace/bpf_trace_printk event.
404 	 * Repeat this each time as it is possible a user has
405 	 * disabled bpf_trace_printk events.  By loading a program
406 	 * calling bpf_trace_printk() however the user has expressed
407 	 * the intent to see such events.
408 	 */
409 	if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
410 		pr_warn_ratelimited("could not enable bpf_trace_printk events");
411 }
412 
413 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
414 {
415 	__set_printk_clr_event();
416 	return &bpf_trace_printk_proto;
417 }
418 
419 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
420 	   u32, data_len)
421 {
422 	static char buf[BPF_TRACE_PRINTK_SIZE];
423 	unsigned long flags;
424 	int ret, num_args;
425 	u32 *bin_args;
426 
427 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
428 	    (data_len && !data))
429 		return -EINVAL;
430 	num_args = data_len / 8;
431 
432 	ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
433 	if (ret < 0)
434 		return ret;
435 
436 	raw_spin_lock_irqsave(&trace_printk_lock, flags);
437 	ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
438 
439 	trace_bpf_trace_printk(buf);
440 	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
441 
442 	bpf_bprintf_cleanup();
443 
444 	return ret;
445 }
446 
447 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
448 	.func		= bpf_trace_vprintk,
449 	.gpl_only	= true,
450 	.ret_type	= RET_INTEGER,
451 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
452 	.arg2_type	= ARG_CONST_SIZE,
453 	.arg3_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
454 	.arg4_type	= ARG_CONST_SIZE_OR_ZERO,
455 };
456 
457 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
458 {
459 	__set_printk_clr_event();
460 	return &bpf_trace_vprintk_proto;
461 }
462 
463 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
464 	   const void *, data, u32, data_len)
465 {
466 	int err, num_args;
467 	u32 *bin_args;
468 
469 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
470 	    (data_len && !data))
471 		return -EINVAL;
472 	num_args = data_len / 8;
473 
474 	err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
475 	if (err < 0)
476 		return err;
477 
478 	seq_bprintf(m, fmt, bin_args);
479 
480 	bpf_bprintf_cleanup();
481 
482 	return seq_has_overflowed(m) ? -EOVERFLOW : 0;
483 }
484 
485 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
486 
487 static const struct bpf_func_proto bpf_seq_printf_proto = {
488 	.func		= bpf_seq_printf,
489 	.gpl_only	= true,
490 	.ret_type	= RET_INTEGER,
491 	.arg1_type	= ARG_PTR_TO_BTF_ID,
492 	.arg1_btf_id	= &btf_seq_file_ids[0],
493 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
494 	.arg3_type	= ARG_CONST_SIZE,
495 	.arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
496 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
497 };
498 
499 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
500 {
501 	return seq_write(m, data, len) ? -EOVERFLOW : 0;
502 }
503 
504 static const struct bpf_func_proto bpf_seq_write_proto = {
505 	.func		= bpf_seq_write,
506 	.gpl_only	= true,
507 	.ret_type	= RET_INTEGER,
508 	.arg1_type	= ARG_PTR_TO_BTF_ID,
509 	.arg1_btf_id	= &btf_seq_file_ids[0],
510 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
511 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
512 };
513 
514 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
515 	   u32, btf_ptr_size, u64, flags)
516 {
517 	const struct btf *btf;
518 	s32 btf_id;
519 	int ret;
520 
521 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
522 	if (ret)
523 		return ret;
524 
525 	return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
526 }
527 
528 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
529 	.func		= bpf_seq_printf_btf,
530 	.gpl_only	= true,
531 	.ret_type	= RET_INTEGER,
532 	.arg1_type	= ARG_PTR_TO_BTF_ID,
533 	.arg1_btf_id	= &btf_seq_file_ids[0],
534 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
535 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
536 	.arg4_type	= ARG_ANYTHING,
537 };
538 
539 static __always_inline int
540 get_map_perf_counter(struct bpf_map *map, u64 flags,
541 		     u64 *value, u64 *enabled, u64 *running)
542 {
543 	struct bpf_array *array = container_of(map, struct bpf_array, map);
544 	unsigned int cpu = smp_processor_id();
545 	u64 index = flags & BPF_F_INDEX_MASK;
546 	struct bpf_event_entry *ee;
547 
548 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
549 		return -EINVAL;
550 	if (index == BPF_F_CURRENT_CPU)
551 		index = cpu;
552 	if (unlikely(index >= array->map.max_entries))
553 		return -E2BIG;
554 
555 	ee = READ_ONCE(array->ptrs[index]);
556 	if (!ee)
557 		return -ENOENT;
558 
559 	return perf_event_read_local(ee->event, value, enabled, running);
560 }
561 
562 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
563 {
564 	u64 value = 0;
565 	int err;
566 
567 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
568 	/*
569 	 * this api is ugly since we miss [-22..-2] range of valid
570 	 * counter values, but that's uapi
571 	 */
572 	if (err)
573 		return err;
574 	return value;
575 }
576 
577 static const struct bpf_func_proto bpf_perf_event_read_proto = {
578 	.func		= bpf_perf_event_read,
579 	.gpl_only	= true,
580 	.ret_type	= RET_INTEGER,
581 	.arg1_type	= ARG_CONST_MAP_PTR,
582 	.arg2_type	= ARG_ANYTHING,
583 };
584 
585 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
586 	   struct bpf_perf_event_value *, buf, u32, size)
587 {
588 	int err = -EINVAL;
589 
590 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
591 		goto clear;
592 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
593 				   &buf->running);
594 	if (unlikely(err))
595 		goto clear;
596 	return 0;
597 clear:
598 	memset(buf, 0, size);
599 	return err;
600 }
601 
602 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
603 	.func		= bpf_perf_event_read_value,
604 	.gpl_only	= true,
605 	.ret_type	= RET_INTEGER,
606 	.arg1_type	= ARG_CONST_MAP_PTR,
607 	.arg2_type	= ARG_ANYTHING,
608 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
609 	.arg4_type	= ARG_CONST_SIZE,
610 };
611 
612 static __always_inline u64
613 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
614 			u64 flags, struct perf_sample_data *sd)
615 {
616 	struct bpf_array *array = container_of(map, struct bpf_array, map);
617 	unsigned int cpu = smp_processor_id();
618 	u64 index = flags & BPF_F_INDEX_MASK;
619 	struct bpf_event_entry *ee;
620 	struct perf_event *event;
621 
622 	if (index == BPF_F_CURRENT_CPU)
623 		index = cpu;
624 	if (unlikely(index >= array->map.max_entries))
625 		return -E2BIG;
626 
627 	ee = READ_ONCE(array->ptrs[index]);
628 	if (!ee)
629 		return -ENOENT;
630 
631 	event = ee->event;
632 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
633 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
634 		return -EINVAL;
635 
636 	if (unlikely(event->oncpu != cpu))
637 		return -EOPNOTSUPP;
638 
639 	return perf_event_output(event, sd, regs);
640 }
641 
642 /*
643  * Support executing tracepoints in normal, irq, and nmi context that each call
644  * bpf_perf_event_output
645  */
646 struct bpf_trace_sample_data {
647 	struct perf_sample_data sds[3];
648 };
649 
650 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
651 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
652 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
653 	   u64, flags, void *, data, u64, size)
654 {
655 	struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
656 	int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
657 	struct perf_raw_record raw = {
658 		.frag = {
659 			.size = size,
660 			.data = data,
661 		},
662 	};
663 	struct perf_sample_data *sd;
664 	int err;
665 
666 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
667 		err = -EBUSY;
668 		goto out;
669 	}
670 
671 	sd = &sds->sds[nest_level - 1];
672 
673 	if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
674 		err = -EINVAL;
675 		goto out;
676 	}
677 
678 	perf_sample_data_init(sd, 0, 0);
679 	sd->raw = &raw;
680 
681 	err = __bpf_perf_event_output(regs, map, flags, sd);
682 
683 out:
684 	this_cpu_dec(bpf_trace_nest_level);
685 	return err;
686 }
687 
688 static const struct bpf_func_proto bpf_perf_event_output_proto = {
689 	.func		= bpf_perf_event_output,
690 	.gpl_only	= true,
691 	.ret_type	= RET_INTEGER,
692 	.arg1_type	= ARG_PTR_TO_CTX,
693 	.arg2_type	= ARG_CONST_MAP_PTR,
694 	.arg3_type	= ARG_ANYTHING,
695 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
696 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
697 };
698 
699 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
700 struct bpf_nested_pt_regs {
701 	struct pt_regs regs[3];
702 };
703 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
704 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
705 
706 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
707 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
708 {
709 	int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
710 	struct perf_raw_frag frag = {
711 		.copy		= ctx_copy,
712 		.size		= ctx_size,
713 		.data		= ctx,
714 	};
715 	struct perf_raw_record raw = {
716 		.frag = {
717 			{
718 				.next	= ctx_size ? &frag : NULL,
719 			},
720 			.size	= meta_size,
721 			.data	= meta,
722 		},
723 	};
724 	struct perf_sample_data *sd;
725 	struct pt_regs *regs;
726 	u64 ret;
727 
728 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
729 		ret = -EBUSY;
730 		goto out;
731 	}
732 	sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
733 	regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
734 
735 	perf_fetch_caller_regs(regs);
736 	perf_sample_data_init(sd, 0, 0);
737 	sd->raw = &raw;
738 
739 	ret = __bpf_perf_event_output(regs, map, flags, sd);
740 out:
741 	this_cpu_dec(bpf_event_output_nest_level);
742 	return ret;
743 }
744 
745 BPF_CALL_0(bpf_get_current_task)
746 {
747 	return (long) current;
748 }
749 
750 const struct bpf_func_proto bpf_get_current_task_proto = {
751 	.func		= bpf_get_current_task,
752 	.gpl_only	= true,
753 	.ret_type	= RET_INTEGER,
754 };
755 
756 BPF_CALL_0(bpf_get_current_task_btf)
757 {
758 	return (unsigned long) current;
759 }
760 
761 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
762 	.func		= bpf_get_current_task_btf,
763 	.gpl_only	= true,
764 	.ret_type	= RET_PTR_TO_BTF_ID,
765 	.ret_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
766 };
767 
768 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
769 {
770 	return (unsigned long) task_pt_regs(task);
771 }
772 
773 BTF_ID_LIST(bpf_task_pt_regs_ids)
774 BTF_ID(struct, pt_regs)
775 
776 const struct bpf_func_proto bpf_task_pt_regs_proto = {
777 	.func		= bpf_task_pt_regs,
778 	.gpl_only	= true,
779 	.arg1_type	= ARG_PTR_TO_BTF_ID,
780 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
781 	.ret_type	= RET_PTR_TO_BTF_ID,
782 	.ret_btf_id	= &bpf_task_pt_regs_ids[0],
783 };
784 
785 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
786 {
787 	struct bpf_array *array = container_of(map, struct bpf_array, map);
788 	struct cgroup *cgrp;
789 
790 	if (unlikely(idx >= array->map.max_entries))
791 		return -E2BIG;
792 
793 	cgrp = READ_ONCE(array->ptrs[idx]);
794 	if (unlikely(!cgrp))
795 		return -EAGAIN;
796 
797 	return task_under_cgroup_hierarchy(current, cgrp);
798 }
799 
800 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
801 	.func           = bpf_current_task_under_cgroup,
802 	.gpl_only       = false,
803 	.ret_type       = RET_INTEGER,
804 	.arg1_type      = ARG_CONST_MAP_PTR,
805 	.arg2_type      = ARG_ANYTHING,
806 };
807 
808 struct send_signal_irq_work {
809 	struct irq_work irq_work;
810 	struct task_struct *task;
811 	u32 sig;
812 	enum pid_type type;
813 };
814 
815 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
816 
817 static void do_bpf_send_signal(struct irq_work *entry)
818 {
819 	struct send_signal_irq_work *work;
820 
821 	work = container_of(entry, struct send_signal_irq_work, irq_work);
822 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
823 }
824 
825 static int bpf_send_signal_common(u32 sig, enum pid_type type)
826 {
827 	struct send_signal_irq_work *work = NULL;
828 
829 	/* Similar to bpf_probe_write_user, task needs to be
830 	 * in a sound condition and kernel memory access be
831 	 * permitted in order to send signal to the current
832 	 * task.
833 	 */
834 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
835 		return -EPERM;
836 	if (unlikely(!nmi_uaccess_okay()))
837 		return -EPERM;
838 
839 	if (irqs_disabled()) {
840 		/* Do an early check on signal validity. Otherwise,
841 		 * the error is lost in deferred irq_work.
842 		 */
843 		if (unlikely(!valid_signal(sig)))
844 			return -EINVAL;
845 
846 		work = this_cpu_ptr(&send_signal_work);
847 		if (irq_work_is_busy(&work->irq_work))
848 			return -EBUSY;
849 
850 		/* Add the current task, which is the target of sending signal,
851 		 * to the irq_work. The current task may change when queued
852 		 * irq works get executed.
853 		 */
854 		work->task = current;
855 		work->sig = sig;
856 		work->type = type;
857 		irq_work_queue(&work->irq_work);
858 		return 0;
859 	}
860 
861 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
862 }
863 
864 BPF_CALL_1(bpf_send_signal, u32, sig)
865 {
866 	return bpf_send_signal_common(sig, PIDTYPE_TGID);
867 }
868 
869 static const struct bpf_func_proto bpf_send_signal_proto = {
870 	.func		= bpf_send_signal,
871 	.gpl_only	= false,
872 	.ret_type	= RET_INTEGER,
873 	.arg1_type	= ARG_ANYTHING,
874 };
875 
876 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
877 {
878 	return bpf_send_signal_common(sig, PIDTYPE_PID);
879 }
880 
881 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
882 	.func		= bpf_send_signal_thread,
883 	.gpl_only	= false,
884 	.ret_type	= RET_INTEGER,
885 	.arg1_type	= ARG_ANYTHING,
886 };
887 
888 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
889 {
890 	long len;
891 	char *p;
892 
893 	if (!sz)
894 		return 0;
895 
896 	p = d_path(path, buf, sz);
897 	if (IS_ERR(p)) {
898 		len = PTR_ERR(p);
899 	} else {
900 		len = buf + sz - p;
901 		memmove(buf, p, len);
902 	}
903 
904 	return len;
905 }
906 
907 BTF_SET_START(btf_allowlist_d_path)
908 #ifdef CONFIG_SECURITY
909 BTF_ID(func, security_file_permission)
910 BTF_ID(func, security_inode_getattr)
911 BTF_ID(func, security_file_open)
912 #endif
913 #ifdef CONFIG_SECURITY_PATH
914 BTF_ID(func, security_path_truncate)
915 #endif
916 BTF_ID(func, vfs_truncate)
917 BTF_ID(func, vfs_fallocate)
918 BTF_ID(func, dentry_open)
919 BTF_ID(func, vfs_getattr)
920 BTF_ID(func, filp_close)
921 BTF_SET_END(btf_allowlist_d_path)
922 
923 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
924 {
925 	if (prog->type == BPF_PROG_TYPE_TRACING &&
926 	    prog->expected_attach_type == BPF_TRACE_ITER)
927 		return true;
928 
929 	if (prog->type == BPF_PROG_TYPE_LSM)
930 		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
931 
932 	return btf_id_set_contains(&btf_allowlist_d_path,
933 				   prog->aux->attach_btf_id);
934 }
935 
936 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
937 
938 static const struct bpf_func_proto bpf_d_path_proto = {
939 	.func		= bpf_d_path,
940 	.gpl_only	= false,
941 	.ret_type	= RET_INTEGER,
942 	.arg1_type	= ARG_PTR_TO_BTF_ID,
943 	.arg1_btf_id	= &bpf_d_path_btf_ids[0],
944 	.arg2_type	= ARG_PTR_TO_MEM,
945 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
946 	.allowed	= bpf_d_path_allowed,
947 };
948 
949 #define BTF_F_ALL	(BTF_F_COMPACT  | BTF_F_NONAME | \
950 			 BTF_F_PTR_RAW | BTF_F_ZERO)
951 
952 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
953 				  u64 flags, const struct btf **btf,
954 				  s32 *btf_id)
955 {
956 	const struct btf_type *t;
957 
958 	if (unlikely(flags & ~(BTF_F_ALL)))
959 		return -EINVAL;
960 
961 	if (btf_ptr_size != sizeof(struct btf_ptr))
962 		return -EINVAL;
963 
964 	*btf = bpf_get_btf_vmlinux();
965 
966 	if (IS_ERR_OR_NULL(*btf))
967 		return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
968 
969 	if (ptr->type_id > 0)
970 		*btf_id = ptr->type_id;
971 	else
972 		return -EINVAL;
973 
974 	if (*btf_id > 0)
975 		t = btf_type_by_id(*btf, *btf_id);
976 	if (*btf_id <= 0 || !t)
977 		return -ENOENT;
978 
979 	return 0;
980 }
981 
982 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
983 	   u32, btf_ptr_size, u64, flags)
984 {
985 	const struct btf *btf;
986 	s32 btf_id;
987 	int ret;
988 
989 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
990 	if (ret)
991 		return ret;
992 
993 	return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
994 				      flags);
995 }
996 
997 const struct bpf_func_proto bpf_snprintf_btf_proto = {
998 	.func		= bpf_snprintf_btf,
999 	.gpl_only	= false,
1000 	.ret_type	= RET_INTEGER,
1001 	.arg1_type	= ARG_PTR_TO_MEM,
1002 	.arg2_type	= ARG_CONST_SIZE,
1003 	.arg3_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1004 	.arg4_type	= ARG_CONST_SIZE,
1005 	.arg5_type	= ARG_ANYTHING,
1006 };
1007 
1008 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1009 {
1010 	/* This helper call is inlined by verifier. */
1011 	return ((u64 *)ctx)[-2];
1012 }
1013 
1014 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1015 	.func		= bpf_get_func_ip_tracing,
1016 	.gpl_only	= true,
1017 	.ret_type	= RET_INTEGER,
1018 	.arg1_type	= ARG_PTR_TO_CTX,
1019 };
1020 
1021 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1022 {
1023 	struct kprobe *kp = kprobe_running();
1024 
1025 	return kp ? (uintptr_t)kp->addr : 0;
1026 }
1027 
1028 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1029 	.func		= bpf_get_func_ip_kprobe,
1030 	.gpl_only	= true,
1031 	.ret_type	= RET_INTEGER,
1032 	.arg1_type	= ARG_PTR_TO_CTX,
1033 };
1034 
1035 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1036 {
1037 	struct bpf_trace_run_ctx *run_ctx;
1038 
1039 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1040 	return run_ctx->bpf_cookie;
1041 }
1042 
1043 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1044 	.func		= bpf_get_attach_cookie_trace,
1045 	.gpl_only	= false,
1046 	.ret_type	= RET_INTEGER,
1047 	.arg1_type	= ARG_PTR_TO_CTX,
1048 };
1049 
1050 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1051 {
1052 	return ctx->event->bpf_cookie;
1053 }
1054 
1055 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1056 	.func		= bpf_get_attach_cookie_pe,
1057 	.gpl_only	= false,
1058 	.ret_type	= RET_INTEGER,
1059 	.arg1_type	= ARG_PTR_TO_CTX,
1060 };
1061 
1062 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1063 {
1064 #ifndef CONFIG_X86
1065 	return -ENOENT;
1066 #else
1067 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1068 	u32 entry_cnt = size / br_entry_size;
1069 
1070 	entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1071 
1072 	if (unlikely(flags))
1073 		return -EINVAL;
1074 
1075 	if (!entry_cnt)
1076 		return -ENOENT;
1077 
1078 	return entry_cnt * br_entry_size;
1079 #endif
1080 }
1081 
1082 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1083 	.func		= bpf_get_branch_snapshot,
1084 	.gpl_only	= true,
1085 	.ret_type	= RET_INTEGER,
1086 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1087 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1088 };
1089 
1090 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1091 {
1092 	/* This helper call is inlined by verifier. */
1093 	u64 nr_args = ((u64 *)ctx)[-1];
1094 
1095 	if ((u64) n >= nr_args)
1096 		return -EINVAL;
1097 	*value = ((u64 *)ctx)[n];
1098 	return 0;
1099 }
1100 
1101 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1102 	.func		= get_func_arg,
1103 	.ret_type	= RET_INTEGER,
1104 	.arg1_type	= ARG_PTR_TO_CTX,
1105 	.arg2_type	= ARG_ANYTHING,
1106 	.arg3_type	= ARG_PTR_TO_LONG,
1107 };
1108 
1109 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1110 {
1111 	/* This helper call is inlined by verifier. */
1112 	u64 nr_args = ((u64 *)ctx)[-1];
1113 
1114 	*value = ((u64 *)ctx)[nr_args];
1115 	return 0;
1116 }
1117 
1118 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1119 	.func		= get_func_ret,
1120 	.ret_type	= RET_INTEGER,
1121 	.arg1_type	= ARG_PTR_TO_CTX,
1122 	.arg2_type	= ARG_PTR_TO_LONG,
1123 };
1124 
1125 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1126 {
1127 	/* This helper call is inlined by verifier. */
1128 	return ((u64 *)ctx)[-1];
1129 }
1130 
1131 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1132 	.func		= get_func_arg_cnt,
1133 	.ret_type	= RET_INTEGER,
1134 	.arg1_type	= ARG_PTR_TO_CTX,
1135 };
1136 
1137 static const struct bpf_func_proto *
1138 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1139 {
1140 	switch (func_id) {
1141 	case BPF_FUNC_map_lookup_elem:
1142 		return &bpf_map_lookup_elem_proto;
1143 	case BPF_FUNC_map_update_elem:
1144 		return &bpf_map_update_elem_proto;
1145 	case BPF_FUNC_map_delete_elem:
1146 		return &bpf_map_delete_elem_proto;
1147 	case BPF_FUNC_map_push_elem:
1148 		return &bpf_map_push_elem_proto;
1149 	case BPF_FUNC_map_pop_elem:
1150 		return &bpf_map_pop_elem_proto;
1151 	case BPF_FUNC_map_peek_elem:
1152 		return &bpf_map_peek_elem_proto;
1153 	case BPF_FUNC_ktime_get_ns:
1154 		return &bpf_ktime_get_ns_proto;
1155 	case BPF_FUNC_ktime_get_boot_ns:
1156 		return &bpf_ktime_get_boot_ns_proto;
1157 	case BPF_FUNC_tail_call:
1158 		return &bpf_tail_call_proto;
1159 	case BPF_FUNC_get_current_pid_tgid:
1160 		return &bpf_get_current_pid_tgid_proto;
1161 	case BPF_FUNC_get_current_task:
1162 		return &bpf_get_current_task_proto;
1163 	case BPF_FUNC_get_current_task_btf:
1164 		return &bpf_get_current_task_btf_proto;
1165 	case BPF_FUNC_task_pt_regs:
1166 		return &bpf_task_pt_regs_proto;
1167 	case BPF_FUNC_get_current_uid_gid:
1168 		return &bpf_get_current_uid_gid_proto;
1169 	case BPF_FUNC_get_current_comm:
1170 		return &bpf_get_current_comm_proto;
1171 	case BPF_FUNC_trace_printk:
1172 		return bpf_get_trace_printk_proto();
1173 	case BPF_FUNC_get_smp_processor_id:
1174 		return &bpf_get_smp_processor_id_proto;
1175 	case BPF_FUNC_get_numa_node_id:
1176 		return &bpf_get_numa_node_id_proto;
1177 	case BPF_FUNC_perf_event_read:
1178 		return &bpf_perf_event_read_proto;
1179 	case BPF_FUNC_current_task_under_cgroup:
1180 		return &bpf_current_task_under_cgroup_proto;
1181 	case BPF_FUNC_get_prandom_u32:
1182 		return &bpf_get_prandom_u32_proto;
1183 	case BPF_FUNC_probe_write_user:
1184 		return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1185 		       NULL : bpf_get_probe_write_proto();
1186 	case BPF_FUNC_probe_read_user:
1187 		return &bpf_probe_read_user_proto;
1188 	case BPF_FUNC_probe_read_kernel:
1189 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1190 		       NULL : &bpf_probe_read_kernel_proto;
1191 	case BPF_FUNC_probe_read_user_str:
1192 		return &bpf_probe_read_user_str_proto;
1193 	case BPF_FUNC_probe_read_kernel_str:
1194 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1195 		       NULL : &bpf_probe_read_kernel_str_proto;
1196 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1197 	case BPF_FUNC_probe_read:
1198 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1199 		       NULL : &bpf_probe_read_compat_proto;
1200 	case BPF_FUNC_probe_read_str:
1201 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1202 		       NULL : &bpf_probe_read_compat_str_proto;
1203 #endif
1204 #ifdef CONFIG_CGROUPS
1205 	case BPF_FUNC_get_current_cgroup_id:
1206 		return &bpf_get_current_cgroup_id_proto;
1207 	case BPF_FUNC_get_current_ancestor_cgroup_id:
1208 		return &bpf_get_current_ancestor_cgroup_id_proto;
1209 #endif
1210 	case BPF_FUNC_send_signal:
1211 		return &bpf_send_signal_proto;
1212 	case BPF_FUNC_send_signal_thread:
1213 		return &bpf_send_signal_thread_proto;
1214 	case BPF_FUNC_perf_event_read_value:
1215 		return &bpf_perf_event_read_value_proto;
1216 	case BPF_FUNC_get_ns_current_pid_tgid:
1217 		return &bpf_get_ns_current_pid_tgid_proto;
1218 	case BPF_FUNC_ringbuf_output:
1219 		return &bpf_ringbuf_output_proto;
1220 	case BPF_FUNC_ringbuf_reserve:
1221 		return &bpf_ringbuf_reserve_proto;
1222 	case BPF_FUNC_ringbuf_submit:
1223 		return &bpf_ringbuf_submit_proto;
1224 	case BPF_FUNC_ringbuf_discard:
1225 		return &bpf_ringbuf_discard_proto;
1226 	case BPF_FUNC_ringbuf_query:
1227 		return &bpf_ringbuf_query_proto;
1228 	case BPF_FUNC_jiffies64:
1229 		return &bpf_jiffies64_proto;
1230 	case BPF_FUNC_get_task_stack:
1231 		return &bpf_get_task_stack_proto;
1232 	case BPF_FUNC_copy_from_user:
1233 		return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1234 	case BPF_FUNC_snprintf_btf:
1235 		return &bpf_snprintf_btf_proto;
1236 	case BPF_FUNC_per_cpu_ptr:
1237 		return &bpf_per_cpu_ptr_proto;
1238 	case BPF_FUNC_this_cpu_ptr:
1239 		return &bpf_this_cpu_ptr_proto;
1240 	case BPF_FUNC_task_storage_get:
1241 		return &bpf_task_storage_get_proto;
1242 	case BPF_FUNC_task_storage_delete:
1243 		return &bpf_task_storage_delete_proto;
1244 	case BPF_FUNC_for_each_map_elem:
1245 		return &bpf_for_each_map_elem_proto;
1246 	case BPF_FUNC_snprintf:
1247 		return &bpf_snprintf_proto;
1248 	case BPF_FUNC_get_func_ip:
1249 		return &bpf_get_func_ip_proto_tracing;
1250 	case BPF_FUNC_get_branch_snapshot:
1251 		return &bpf_get_branch_snapshot_proto;
1252 	case BPF_FUNC_find_vma:
1253 		return &bpf_find_vma_proto;
1254 	case BPF_FUNC_trace_vprintk:
1255 		return bpf_get_trace_vprintk_proto();
1256 	default:
1257 		return bpf_base_func_proto(func_id);
1258 	}
1259 }
1260 
1261 static const struct bpf_func_proto *
1262 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1263 {
1264 	switch (func_id) {
1265 	case BPF_FUNC_perf_event_output:
1266 		return &bpf_perf_event_output_proto;
1267 	case BPF_FUNC_get_stackid:
1268 		return &bpf_get_stackid_proto;
1269 	case BPF_FUNC_get_stack:
1270 		return &bpf_get_stack_proto;
1271 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1272 	case BPF_FUNC_override_return:
1273 		return &bpf_override_return_proto;
1274 #endif
1275 	case BPF_FUNC_get_func_ip:
1276 		return &bpf_get_func_ip_proto_kprobe;
1277 	case BPF_FUNC_get_attach_cookie:
1278 		return &bpf_get_attach_cookie_proto_trace;
1279 	default:
1280 		return bpf_tracing_func_proto(func_id, prog);
1281 	}
1282 }
1283 
1284 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1285 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1286 					const struct bpf_prog *prog,
1287 					struct bpf_insn_access_aux *info)
1288 {
1289 	if (off < 0 || off >= sizeof(struct pt_regs))
1290 		return false;
1291 	if (type != BPF_READ)
1292 		return false;
1293 	if (off % size != 0)
1294 		return false;
1295 	/*
1296 	 * Assertion for 32 bit to make sure last 8 byte access
1297 	 * (BPF_DW) to the last 4 byte member is disallowed.
1298 	 */
1299 	if (off + size > sizeof(struct pt_regs))
1300 		return false;
1301 
1302 	return true;
1303 }
1304 
1305 const struct bpf_verifier_ops kprobe_verifier_ops = {
1306 	.get_func_proto  = kprobe_prog_func_proto,
1307 	.is_valid_access = kprobe_prog_is_valid_access,
1308 };
1309 
1310 const struct bpf_prog_ops kprobe_prog_ops = {
1311 };
1312 
1313 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1314 	   u64, flags, void *, data, u64, size)
1315 {
1316 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1317 
1318 	/*
1319 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1320 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1321 	 * from there and call the same bpf_perf_event_output() helper inline.
1322 	 */
1323 	return ____bpf_perf_event_output(regs, map, flags, data, size);
1324 }
1325 
1326 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1327 	.func		= bpf_perf_event_output_tp,
1328 	.gpl_only	= true,
1329 	.ret_type	= RET_INTEGER,
1330 	.arg1_type	= ARG_PTR_TO_CTX,
1331 	.arg2_type	= ARG_CONST_MAP_PTR,
1332 	.arg3_type	= ARG_ANYTHING,
1333 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1334 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1335 };
1336 
1337 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1338 	   u64, flags)
1339 {
1340 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1341 
1342 	/*
1343 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
1344 	 * the other helper's function body cannot be inlined due to being
1345 	 * external, thus we need to call raw helper function.
1346 	 */
1347 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1348 			       flags, 0, 0);
1349 }
1350 
1351 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1352 	.func		= bpf_get_stackid_tp,
1353 	.gpl_only	= true,
1354 	.ret_type	= RET_INTEGER,
1355 	.arg1_type	= ARG_PTR_TO_CTX,
1356 	.arg2_type	= ARG_CONST_MAP_PTR,
1357 	.arg3_type	= ARG_ANYTHING,
1358 };
1359 
1360 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1361 	   u64, flags)
1362 {
1363 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1364 
1365 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1366 			     (unsigned long) size, flags, 0);
1367 }
1368 
1369 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1370 	.func		= bpf_get_stack_tp,
1371 	.gpl_only	= true,
1372 	.ret_type	= RET_INTEGER,
1373 	.arg1_type	= ARG_PTR_TO_CTX,
1374 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
1375 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1376 	.arg4_type	= ARG_ANYTHING,
1377 };
1378 
1379 static const struct bpf_func_proto *
1380 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1381 {
1382 	switch (func_id) {
1383 	case BPF_FUNC_perf_event_output:
1384 		return &bpf_perf_event_output_proto_tp;
1385 	case BPF_FUNC_get_stackid:
1386 		return &bpf_get_stackid_proto_tp;
1387 	case BPF_FUNC_get_stack:
1388 		return &bpf_get_stack_proto_tp;
1389 	case BPF_FUNC_get_attach_cookie:
1390 		return &bpf_get_attach_cookie_proto_trace;
1391 	default:
1392 		return bpf_tracing_func_proto(func_id, prog);
1393 	}
1394 }
1395 
1396 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1397 				    const struct bpf_prog *prog,
1398 				    struct bpf_insn_access_aux *info)
1399 {
1400 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1401 		return false;
1402 	if (type != BPF_READ)
1403 		return false;
1404 	if (off % size != 0)
1405 		return false;
1406 
1407 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1408 	return true;
1409 }
1410 
1411 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1412 	.get_func_proto  = tp_prog_func_proto,
1413 	.is_valid_access = tp_prog_is_valid_access,
1414 };
1415 
1416 const struct bpf_prog_ops tracepoint_prog_ops = {
1417 };
1418 
1419 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1420 	   struct bpf_perf_event_value *, buf, u32, size)
1421 {
1422 	int err = -EINVAL;
1423 
1424 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1425 		goto clear;
1426 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1427 				    &buf->running);
1428 	if (unlikely(err))
1429 		goto clear;
1430 	return 0;
1431 clear:
1432 	memset(buf, 0, size);
1433 	return err;
1434 }
1435 
1436 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1437          .func           = bpf_perf_prog_read_value,
1438          .gpl_only       = true,
1439          .ret_type       = RET_INTEGER,
1440          .arg1_type      = ARG_PTR_TO_CTX,
1441          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1442          .arg3_type      = ARG_CONST_SIZE,
1443 };
1444 
1445 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1446 	   void *, buf, u32, size, u64, flags)
1447 {
1448 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1449 	struct perf_branch_stack *br_stack = ctx->data->br_stack;
1450 	u32 to_copy;
1451 
1452 	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1453 		return -EINVAL;
1454 
1455 	if (unlikely(!br_stack))
1456 		return -ENOENT;
1457 
1458 	if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1459 		return br_stack->nr * br_entry_size;
1460 
1461 	if (!buf || (size % br_entry_size != 0))
1462 		return -EINVAL;
1463 
1464 	to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1465 	memcpy(buf, br_stack->entries, to_copy);
1466 
1467 	return to_copy;
1468 }
1469 
1470 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1471 	.func           = bpf_read_branch_records,
1472 	.gpl_only       = true,
1473 	.ret_type       = RET_INTEGER,
1474 	.arg1_type      = ARG_PTR_TO_CTX,
1475 	.arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1476 	.arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1477 	.arg4_type      = ARG_ANYTHING,
1478 };
1479 
1480 static const struct bpf_func_proto *
1481 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1482 {
1483 	switch (func_id) {
1484 	case BPF_FUNC_perf_event_output:
1485 		return &bpf_perf_event_output_proto_tp;
1486 	case BPF_FUNC_get_stackid:
1487 		return &bpf_get_stackid_proto_pe;
1488 	case BPF_FUNC_get_stack:
1489 		return &bpf_get_stack_proto_pe;
1490 	case BPF_FUNC_perf_prog_read_value:
1491 		return &bpf_perf_prog_read_value_proto;
1492 	case BPF_FUNC_read_branch_records:
1493 		return &bpf_read_branch_records_proto;
1494 	case BPF_FUNC_get_attach_cookie:
1495 		return &bpf_get_attach_cookie_proto_pe;
1496 	default:
1497 		return bpf_tracing_func_proto(func_id, prog);
1498 	}
1499 }
1500 
1501 /*
1502  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1503  * to avoid potential recursive reuse issue when/if tracepoints are added
1504  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1505  *
1506  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1507  * in normal, irq, and nmi context.
1508  */
1509 struct bpf_raw_tp_regs {
1510 	struct pt_regs regs[3];
1511 };
1512 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1513 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1514 static struct pt_regs *get_bpf_raw_tp_regs(void)
1515 {
1516 	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1517 	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1518 
1519 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1520 		this_cpu_dec(bpf_raw_tp_nest_level);
1521 		return ERR_PTR(-EBUSY);
1522 	}
1523 
1524 	return &tp_regs->regs[nest_level - 1];
1525 }
1526 
1527 static void put_bpf_raw_tp_regs(void)
1528 {
1529 	this_cpu_dec(bpf_raw_tp_nest_level);
1530 }
1531 
1532 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1533 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1534 {
1535 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1536 	int ret;
1537 
1538 	if (IS_ERR(regs))
1539 		return PTR_ERR(regs);
1540 
1541 	perf_fetch_caller_regs(regs);
1542 	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1543 
1544 	put_bpf_raw_tp_regs();
1545 	return ret;
1546 }
1547 
1548 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1549 	.func		= bpf_perf_event_output_raw_tp,
1550 	.gpl_only	= true,
1551 	.ret_type	= RET_INTEGER,
1552 	.arg1_type	= ARG_PTR_TO_CTX,
1553 	.arg2_type	= ARG_CONST_MAP_PTR,
1554 	.arg3_type	= ARG_ANYTHING,
1555 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1556 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1557 };
1558 
1559 extern const struct bpf_func_proto bpf_skb_output_proto;
1560 extern const struct bpf_func_proto bpf_xdp_output_proto;
1561 
1562 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1563 	   struct bpf_map *, map, u64, flags)
1564 {
1565 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1566 	int ret;
1567 
1568 	if (IS_ERR(regs))
1569 		return PTR_ERR(regs);
1570 
1571 	perf_fetch_caller_regs(regs);
1572 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1573 	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1574 			      flags, 0, 0);
1575 	put_bpf_raw_tp_regs();
1576 	return ret;
1577 }
1578 
1579 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1580 	.func		= bpf_get_stackid_raw_tp,
1581 	.gpl_only	= true,
1582 	.ret_type	= RET_INTEGER,
1583 	.arg1_type	= ARG_PTR_TO_CTX,
1584 	.arg2_type	= ARG_CONST_MAP_PTR,
1585 	.arg3_type	= ARG_ANYTHING,
1586 };
1587 
1588 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1589 	   void *, buf, u32, size, u64, flags)
1590 {
1591 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1592 	int ret;
1593 
1594 	if (IS_ERR(regs))
1595 		return PTR_ERR(regs);
1596 
1597 	perf_fetch_caller_regs(regs);
1598 	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1599 			    (unsigned long) size, flags, 0);
1600 	put_bpf_raw_tp_regs();
1601 	return ret;
1602 }
1603 
1604 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1605 	.func		= bpf_get_stack_raw_tp,
1606 	.gpl_only	= true,
1607 	.ret_type	= RET_INTEGER,
1608 	.arg1_type	= ARG_PTR_TO_CTX,
1609 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1610 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1611 	.arg4_type	= ARG_ANYTHING,
1612 };
1613 
1614 static const struct bpf_func_proto *
1615 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1616 {
1617 	switch (func_id) {
1618 	case BPF_FUNC_perf_event_output:
1619 		return &bpf_perf_event_output_proto_raw_tp;
1620 	case BPF_FUNC_get_stackid:
1621 		return &bpf_get_stackid_proto_raw_tp;
1622 	case BPF_FUNC_get_stack:
1623 		return &bpf_get_stack_proto_raw_tp;
1624 	default:
1625 		return bpf_tracing_func_proto(func_id, prog);
1626 	}
1627 }
1628 
1629 const struct bpf_func_proto *
1630 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1631 {
1632 	const struct bpf_func_proto *fn;
1633 
1634 	switch (func_id) {
1635 #ifdef CONFIG_NET
1636 	case BPF_FUNC_skb_output:
1637 		return &bpf_skb_output_proto;
1638 	case BPF_FUNC_xdp_output:
1639 		return &bpf_xdp_output_proto;
1640 	case BPF_FUNC_skc_to_tcp6_sock:
1641 		return &bpf_skc_to_tcp6_sock_proto;
1642 	case BPF_FUNC_skc_to_tcp_sock:
1643 		return &bpf_skc_to_tcp_sock_proto;
1644 	case BPF_FUNC_skc_to_tcp_timewait_sock:
1645 		return &bpf_skc_to_tcp_timewait_sock_proto;
1646 	case BPF_FUNC_skc_to_tcp_request_sock:
1647 		return &bpf_skc_to_tcp_request_sock_proto;
1648 	case BPF_FUNC_skc_to_udp6_sock:
1649 		return &bpf_skc_to_udp6_sock_proto;
1650 	case BPF_FUNC_skc_to_unix_sock:
1651 		return &bpf_skc_to_unix_sock_proto;
1652 	case BPF_FUNC_sk_storage_get:
1653 		return &bpf_sk_storage_get_tracing_proto;
1654 	case BPF_FUNC_sk_storage_delete:
1655 		return &bpf_sk_storage_delete_tracing_proto;
1656 	case BPF_FUNC_sock_from_file:
1657 		return &bpf_sock_from_file_proto;
1658 	case BPF_FUNC_get_socket_cookie:
1659 		return &bpf_get_socket_ptr_cookie_proto;
1660 #endif
1661 	case BPF_FUNC_seq_printf:
1662 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1663 		       &bpf_seq_printf_proto :
1664 		       NULL;
1665 	case BPF_FUNC_seq_write:
1666 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1667 		       &bpf_seq_write_proto :
1668 		       NULL;
1669 	case BPF_FUNC_seq_printf_btf:
1670 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1671 		       &bpf_seq_printf_btf_proto :
1672 		       NULL;
1673 	case BPF_FUNC_d_path:
1674 		return &bpf_d_path_proto;
1675 	case BPF_FUNC_get_func_arg:
1676 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1677 	case BPF_FUNC_get_func_ret:
1678 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1679 	case BPF_FUNC_get_func_arg_cnt:
1680 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1681 	default:
1682 		fn = raw_tp_prog_func_proto(func_id, prog);
1683 		if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1684 			fn = bpf_iter_get_func_proto(func_id, prog);
1685 		return fn;
1686 	}
1687 }
1688 
1689 static bool raw_tp_prog_is_valid_access(int off, int size,
1690 					enum bpf_access_type type,
1691 					const struct bpf_prog *prog,
1692 					struct bpf_insn_access_aux *info)
1693 {
1694 	return bpf_tracing_ctx_access(off, size, type);
1695 }
1696 
1697 static bool tracing_prog_is_valid_access(int off, int size,
1698 					 enum bpf_access_type type,
1699 					 const struct bpf_prog *prog,
1700 					 struct bpf_insn_access_aux *info)
1701 {
1702 	return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1703 }
1704 
1705 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1706 				     const union bpf_attr *kattr,
1707 				     union bpf_attr __user *uattr)
1708 {
1709 	return -ENOTSUPP;
1710 }
1711 
1712 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1713 	.get_func_proto  = raw_tp_prog_func_proto,
1714 	.is_valid_access = raw_tp_prog_is_valid_access,
1715 };
1716 
1717 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1718 #ifdef CONFIG_NET
1719 	.test_run = bpf_prog_test_run_raw_tp,
1720 #endif
1721 };
1722 
1723 const struct bpf_verifier_ops tracing_verifier_ops = {
1724 	.get_func_proto  = tracing_prog_func_proto,
1725 	.is_valid_access = tracing_prog_is_valid_access,
1726 };
1727 
1728 const struct bpf_prog_ops tracing_prog_ops = {
1729 	.test_run = bpf_prog_test_run_tracing,
1730 };
1731 
1732 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1733 						 enum bpf_access_type type,
1734 						 const struct bpf_prog *prog,
1735 						 struct bpf_insn_access_aux *info)
1736 {
1737 	if (off == 0) {
1738 		if (size != sizeof(u64) || type != BPF_READ)
1739 			return false;
1740 		info->reg_type = PTR_TO_TP_BUFFER;
1741 	}
1742 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1743 }
1744 
1745 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1746 	.get_func_proto  = raw_tp_prog_func_proto,
1747 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
1748 };
1749 
1750 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1751 };
1752 
1753 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1754 				    const struct bpf_prog *prog,
1755 				    struct bpf_insn_access_aux *info)
1756 {
1757 	const int size_u64 = sizeof(u64);
1758 
1759 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1760 		return false;
1761 	if (type != BPF_READ)
1762 		return false;
1763 	if (off % size != 0) {
1764 		if (sizeof(unsigned long) != 4)
1765 			return false;
1766 		if (size != 8)
1767 			return false;
1768 		if (off % size != 4)
1769 			return false;
1770 	}
1771 
1772 	switch (off) {
1773 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1774 		bpf_ctx_record_field_size(info, size_u64);
1775 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1776 			return false;
1777 		break;
1778 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
1779 		bpf_ctx_record_field_size(info, size_u64);
1780 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1781 			return false;
1782 		break;
1783 	default:
1784 		if (size != sizeof(long))
1785 			return false;
1786 	}
1787 
1788 	return true;
1789 }
1790 
1791 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1792 				      const struct bpf_insn *si,
1793 				      struct bpf_insn *insn_buf,
1794 				      struct bpf_prog *prog, u32 *target_size)
1795 {
1796 	struct bpf_insn *insn = insn_buf;
1797 
1798 	switch (si->off) {
1799 	case offsetof(struct bpf_perf_event_data, sample_period):
1800 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1801 						       data), si->dst_reg, si->src_reg,
1802 				      offsetof(struct bpf_perf_event_data_kern, data));
1803 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1804 				      bpf_target_off(struct perf_sample_data, period, 8,
1805 						     target_size));
1806 		break;
1807 	case offsetof(struct bpf_perf_event_data, addr):
1808 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1809 						       data), si->dst_reg, si->src_reg,
1810 				      offsetof(struct bpf_perf_event_data_kern, data));
1811 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1812 				      bpf_target_off(struct perf_sample_data, addr, 8,
1813 						     target_size));
1814 		break;
1815 	default:
1816 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1817 						       regs), si->dst_reg, si->src_reg,
1818 				      offsetof(struct bpf_perf_event_data_kern, regs));
1819 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1820 				      si->off);
1821 		break;
1822 	}
1823 
1824 	return insn - insn_buf;
1825 }
1826 
1827 const struct bpf_verifier_ops perf_event_verifier_ops = {
1828 	.get_func_proto		= pe_prog_func_proto,
1829 	.is_valid_access	= pe_prog_is_valid_access,
1830 	.convert_ctx_access	= pe_prog_convert_ctx_access,
1831 };
1832 
1833 const struct bpf_prog_ops perf_event_prog_ops = {
1834 };
1835 
1836 static DEFINE_MUTEX(bpf_event_mutex);
1837 
1838 #define BPF_TRACE_MAX_PROGS 64
1839 
1840 int perf_event_attach_bpf_prog(struct perf_event *event,
1841 			       struct bpf_prog *prog,
1842 			       u64 bpf_cookie)
1843 {
1844 	struct bpf_prog_array *old_array;
1845 	struct bpf_prog_array *new_array;
1846 	int ret = -EEXIST;
1847 
1848 	/*
1849 	 * Kprobe override only works if they are on the function entry,
1850 	 * and only if they are on the opt-in list.
1851 	 */
1852 	if (prog->kprobe_override &&
1853 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1854 	     !trace_kprobe_error_injectable(event->tp_event)))
1855 		return -EINVAL;
1856 
1857 	mutex_lock(&bpf_event_mutex);
1858 
1859 	if (event->prog)
1860 		goto unlock;
1861 
1862 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1863 	if (old_array &&
1864 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1865 		ret = -E2BIG;
1866 		goto unlock;
1867 	}
1868 
1869 	ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
1870 	if (ret < 0)
1871 		goto unlock;
1872 
1873 	/* set the new array to event->tp_event and set event->prog */
1874 	event->prog = prog;
1875 	event->bpf_cookie = bpf_cookie;
1876 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1877 	bpf_prog_array_free(old_array);
1878 
1879 unlock:
1880 	mutex_unlock(&bpf_event_mutex);
1881 	return ret;
1882 }
1883 
1884 void perf_event_detach_bpf_prog(struct perf_event *event)
1885 {
1886 	struct bpf_prog_array *old_array;
1887 	struct bpf_prog_array *new_array;
1888 	int ret;
1889 
1890 	mutex_lock(&bpf_event_mutex);
1891 
1892 	if (!event->prog)
1893 		goto unlock;
1894 
1895 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1896 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
1897 	if (ret == -ENOENT)
1898 		goto unlock;
1899 	if (ret < 0) {
1900 		bpf_prog_array_delete_safe(old_array, event->prog);
1901 	} else {
1902 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1903 		bpf_prog_array_free(old_array);
1904 	}
1905 
1906 	bpf_prog_put(event->prog);
1907 	event->prog = NULL;
1908 
1909 unlock:
1910 	mutex_unlock(&bpf_event_mutex);
1911 }
1912 
1913 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1914 {
1915 	struct perf_event_query_bpf __user *uquery = info;
1916 	struct perf_event_query_bpf query = {};
1917 	struct bpf_prog_array *progs;
1918 	u32 *ids, prog_cnt, ids_len;
1919 	int ret;
1920 
1921 	if (!perfmon_capable())
1922 		return -EPERM;
1923 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1924 		return -EINVAL;
1925 	if (copy_from_user(&query, uquery, sizeof(query)))
1926 		return -EFAULT;
1927 
1928 	ids_len = query.ids_len;
1929 	if (ids_len > BPF_TRACE_MAX_PROGS)
1930 		return -E2BIG;
1931 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1932 	if (!ids)
1933 		return -ENOMEM;
1934 	/*
1935 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1936 	 * is required when user only wants to check for uquery->prog_cnt.
1937 	 * There is no need to check for it since the case is handled
1938 	 * gracefully in bpf_prog_array_copy_info.
1939 	 */
1940 
1941 	mutex_lock(&bpf_event_mutex);
1942 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1943 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1944 	mutex_unlock(&bpf_event_mutex);
1945 
1946 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1947 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1948 		ret = -EFAULT;
1949 
1950 	kfree(ids);
1951 	return ret;
1952 }
1953 
1954 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1955 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1956 
1957 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1958 {
1959 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1960 
1961 	for (; btp < __stop__bpf_raw_tp; btp++) {
1962 		if (!strcmp(btp->tp->name, name))
1963 			return btp;
1964 	}
1965 
1966 	return bpf_get_raw_tracepoint_module(name);
1967 }
1968 
1969 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1970 {
1971 	struct module *mod;
1972 
1973 	preempt_disable();
1974 	mod = __module_address((unsigned long)btp);
1975 	module_put(mod);
1976 	preempt_enable();
1977 }
1978 
1979 static __always_inline
1980 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1981 {
1982 	cant_sleep();
1983 	rcu_read_lock();
1984 	(void) bpf_prog_run(prog, args);
1985 	rcu_read_unlock();
1986 }
1987 
1988 #define UNPACK(...)			__VA_ARGS__
1989 #define REPEAT_1(FN, DL, X, ...)	FN(X)
1990 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1991 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1992 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1993 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1994 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1995 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1996 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1997 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1998 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1999 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2000 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2001 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
2002 
2003 #define SARG(X)		u64 arg##X
2004 #define COPY(X)		args[X] = arg##X
2005 
2006 #define __DL_COM	(,)
2007 #define __DL_SEM	(;)
2008 
2009 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2010 
2011 #define BPF_TRACE_DEFN_x(x)						\
2012 	void bpf_trace_run##x(struct bpf_prog *prog,			\
2013 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
2014 	{								\
2015 		u64 args[x];						\
2016 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
2017 		__bpf_trace_run(prog, args);				\
2018 	}								\
2019 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2020 BPF_TRACE_DEFN_x(1);
2021 BPF_TRACE_DEFN_x(2);
2022 BPF_TRACE_DEFN_x(3);
2023 BPF_TRACE_DEFN_x(4);
2024 BPF_TRACE_DEFN_x(5);
2025 BPF_TRACE_DEFN_x(6);
2026 BPF_TRACE_DEFN_x(7);
2027 BPF_TRACE_DEFN_x(8);
2028 BPF_TRACE_DEFN_x(9);
2029 BPF_TRACE_DEFN_x(10);
2030 BPF_TRACE_DEFN_x(11);
2031 BPF_TRACE_DEFN_x(12);
2032 
2033 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2034 {
2035 	struct tracepoint *tp = btp->tp;
2036 
2037 	/*
2038 	 * check that program doesn't access arguments beyond what's
2039 	 * available in this tracepoint
2040 	 */
2041 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2042 		return -EINVAL;
2043 
2044 	if (prog->aux->max_tp_access > btp->writable_size)
2045 		return -EINVAL;
2046 
2047 	return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2048 						   prog);
2049 }
2050 
2051 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2052 {
2053 	return __bpf_probe_register(btp, prog);
2054 }
2055 
2056 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2057 {
2058 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2059 }
2060 
2061 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2062 			    u32 *fd_type, const char **buf,
2063 			    u64 *probe_offset, u64 *probe_addr)
2064 {
2065 	bool is_tracepoint, is_syscall_tp;
2066 	struct bpf_prog *prog;
2067 	int flags, err = 0;
2068 
2069 	prog = event->prog;
2070 	if (!prog)
2071 		return -ENOENT;
2072 
2073 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2074 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2075 		return -EOPNOTSUPP;
2076 
2077 	*prog_id = prog->aux->id;
2078 	flags = event->tp_event->flags;
2079 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2080 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
2081 
2082 	if (is_tracepoint || is_syscall_tp) {
2083 		*buf = is_tracepoint ? event->tp_event->tp->name
2084 				     : event->tp_event->name;
2085 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
2086 		*probe_offset = 0x0;
2087 		*probe_addr = 0x0;
2088 	} else {
2089 		/* kprobe/uprobe */
2090 		err = -EOPNOTSUPP;
2091 #ifdef CONFIG_KPROBE_EVENTS
2092 		if (flags & TRACE_EVENT_FL_KPROBE)
2093 			err = bpf_get_kprobe_info(event, fd_type, buf,
2094 						  probe_offset, probe_addr,
2095 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2096 #endif
2097 #ifdef CONFIG_UPROBE_EVENTS
2098 		if (flags & TRACE_EVENT_FL_UPROBE)
2099 			err = bpf_get_uprobe_info(event, fd_type, buf,
2100 						  probe_offset,
2101 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2102 #endif
2103 	}
2104 
2105 	return err;
2106 }
2107 
2108 static int __init send_signal_irq_work_init(void)
2109 {
2110 	int cpu;
2111 	struct send_signal_irq_work *work;
2112 
2113 	for_each_possible_cpu(cpu) {
2114 		work = per_cpu_ptr(&send_signal_work, cpu);
2115 		init_irq_work(&work->irq_work, do_bpf_send_signal);
2116 	}
2117 	return 0;
2118 }
2119 
2120 subsys_initcall(send_signal_irq_work_init);
2121 
2122 #ifdef CONFIG_MODULES
2123 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2124 			    void *module)
2125 {
2126 	struct bpf_trace_module *btm, *tmp;
2127 	struct module *mod = module;
2128 	int ret = 0;
2129 
2130 	if (mod->num_bpf_raw_events == 0 ||
2131 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2132 		goto out;
2133 
2134 	mutex_lock(&bpf_module_mutex);
2135 
2136 	switch (op) {
2137 	case MODULE_STATE_COMING:
2138 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2139 		if (btm) {
2140 			btm->module = module;
2141 			list_add(&btm->list, &bpf_trace_modules);
2142 		} else {
2143 			ret = -ENOMEM;
2144 		}
2145 		break;
2146 	case MODULE_STATE_GOING:
2147 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2148 			if (btm->module == module) {
2149 				list_del(&btm->list);
2150 				kfree(btm);
2151 				break;
2152 			}
2153 		}
2154 		break;
2155 	}
2156 
2157 	mutex_unlock(&bpf_module_mutex);
2158 
2159 out:
2160 	return notifier_from_errno(ret);
2161 }
2162 
2163 static struct notifier_block bpf_module_nb = {
2164 	.notifier_call = bpf_event_notify,
2165 };
2166 
2167 static int __init bpf_event_init(void)
2168 {
2169 	register_module_notifier(&bpf_module_nb);
2170 	return 0;
2171 }
2172 
2173 fs_initcall(bpf_event_init);
2174 #endif /* CONFIG_MODULES */
2175