xref: /linux/kernel/trace/bpf_trace.c (revision ed7171ff9fabc49ae6ed42fbd082a576473836fc)
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_verifier.h>
10 #include <linux/bpf_perf_event.h>
11 #include <linux/btf.h>
12 #include <linux/filter.h>
13 #include <linux/uaccess.h>
14 #include <linux/ctype.h>
15 #include <linux/kprobes.h>
16 #include <linux/spinlock.h>
17 #include <linux/syscalls.h>
18 #include <linux/error-injection.h>
19 #include <linux/btf_ids.h>
20 #include <linux/bpf_lsm.h>
21 #include <linux/fprobe.h>
22 #include <linux/bsearch.h>
23 #include <linux/sort.h>
24 #include <linux/key.h>
25 #include <linux/verification.h>
26 #include <linux/namei.h>
27 #include <linux/fileattr.h>
28 
29 #include <net/bpf_sk_storage.h>
30 
31 #include <uapi/linux/bpf.h>
32 #include <uapi/linux/btf.h>
33 
34 #include <asm/tlb.h>
35 
36 #include "trace_probe.h"
37 #include "trace.h"
38 
39 #define CREATE_TRACE_POINTS
40 #include "bpf_trace.h"
41 
42 #define bpf_event_rcu_dereference(p)					\
43 	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
44 
45 #define MAX_UPROBE_MULTI_CNT (1U << 20)
46 #define MAX_KPROBE_MULTI_CNT (1U << 20)
47 
48 #ifdef CONFIG_MODULES
49 struct bpf_trace_module {
50 	struct module *module;
51 	struct list_head list;
52 };
53 
54 static LIST_HEAD(bpf_trace_modules);
55 static DEFINE_MUTEX(bpf_module_mutex);
56 
57 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
58 {
59 	struct bpf_raw_event_map *btp, *ret = NULL;
60 	struct bpf_trace_module *btm;
61 	unsigned int i;
62 
63 	mutex_lock(&bpf_module_mutex);
64 	list_for_each_entry(btm, &bpf_trace_modules, list) {
65 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
66 			btp = &btm->module->bpf_raw_events[i];
67 			if (!strcmp(btp->tp->name, name)) {
68 				if (try_module_get(btm->module))
69 					ret = btp;
70 				goto out;
71 			}
72 		}
73 	}
74 out:
75 	mutex_unlock(&bpf_module_mutex);
76 	return ret;
77 }
78 #else
79 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
80 {
81 	return NULL;
82 }
83 #endif /* CONFIG_MODULES */
84 
85 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
86 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
87 
88 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
89 				  u64 flags, const struct btf **btf,
90 				  s32 *btf_id);
91 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
92 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
93 
94 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx);
95 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
96 
97 /**
98  * trace_call_bpf - invoke BPF program
99  * @call: tracepoint event
100  * @ctx: opaque context pointer
101  *
102  * kprobe handlers execute BPF programs via this helper.
103  * Can be used from static tracepoints in the future.
104  *
105  * Return: BPF programs always return an integer which is interpreted by
106  * kprobe handler as:
107  * 0 - return from kprobe (event is filtered out)
108  * 1 - store kprobe event into ring buffer
109  * Other values are reserved and currently alias to 1
110  */
111 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
112 {
113 	unsigned int ret;
114 
115 	cant_sleep();
116 
117 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
118 		/*
119 		 * since some bpf program is already running on this cpu,
120 		 * don't call into another bpf program (same or different)
121 		 * and don't send kprobe event into ring-buffer,
122 		 * so return zero here
123 		 */
124 		rcu_read_lock();
125 		bpf_prog_inc_misses_counters(rcu_dereference(call->prog_array));
126 		rcu_read_unlock();
127 		ret = 0;
128 		goto out;
129 	}
130 
131 	/*
132 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
133 	 * to all call sites, we did a bpf_prog_array_valid() there to check
134 	 * whether call->prog_array is empty or not, which is
135 	 * a heuristic to speed up execution.
136 	 *
137 	 * If bpf_prog_array_valid() fetched prog_array was
138 	 * non-NULL, we go into trace_call_bpf() and do the actual
139 	 * proper rcu_dereference() under RCU lock.
140 	 * If it turns out that prog_array is NULL then, we bail out.
141 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
142 	 * was NULL, you'll skip the prog_array with the risk of missing
143 	 * out of events when it was updated in between this and the
144 	 * rcu_dereference() which is accepted risk.
145 	 */
146 	rcu_read_lock();
147 	ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
148 				 ctx, bpf_prog_run);
149 	rcu_read_unlock();
150 
151  out:
152 	__this_cpu_dec(bpf_prog_active);
153 
154 	return ret;
155 }
156 
157 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
158 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
159 {
160 	regs_set_return_value(regs, rc);
161 	override_function_with_return(regs);
162 	return 0;
163 }
164 
165 static const struct bpf_func_proto bpf_override_return_proto = {
166 	.func		= bpf_override_return,
167 	.gpl_only	= true,
168 	.ret_type	= RET_INTEGER,
169 	.arg1_type	= ARG_PTR_TO_CTX,
170 	.arg2_type	= ARG_ANYTHING,
171 };
172 #endif
173 
174 static __always_inline int
175 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
176 {
177 	int ret;
178 
179 	ret = copy_from_user_nofault(dst, unsafe_ptr, size);
180 	if (unlikely(ret < 0))
181 		memset(dst, 0, size);
182 	return ret;
183 }
184 
185 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
186 	   const void __user *, unsafe_ptr)
187 {
188 	return bpf_probe_read_user_common(dst, size, unsafe_ptr);
189 }
190 
191 const struct bpf_func_proto bpf_probe_read_user_proto = {
192 	.func		= bpf_probe_read_user,
193 	.gpl_only	= true,
194 	.ret_type	= RET_INTEGER,
195 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
196 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
197 	.arg3_type	= ARG_ANYTHING,
198 };
199 
200 static __always_inline int
201 bpf_probe_read_user_str_common(void *dst, u32 size,
202 			       const void __user *unsafe_ptr)
203 {
204 	int ret;
205 
206 	/*
207 	 * NB: We rely on strncpy_from_user() not copying junk past the NUL
208 	 * terminator into `dst`.
209 	 *
210 	 * strncpy_from_user() does long-sized strides in the fast path. If the
211 	 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
212 	 * then there could be junk after the NUL in `dst`. If user takes `dst`
213 	 * and keys a hash map with it, then semantically identical strings can
214 	 * occupy multiple entries in the map.
215 	 */
216 	ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
217 	if (unlikely(ret < 0))
218 		memset(dst, 0, size);
219 	return ret;
220 }
221 
222 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
223 	   const void __user *, unsafe_ptr)
224 {
225 	return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
226 }
227 
228 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
229 	.func		= bpf_probe_read_user_str,
230 	.gpl_only	= true,
231 	.ret_type	= RET_INTEGER,
232 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
233 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
234 	.arg3_type	= ARG_ANYTHING,
235 };
236 
237 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
238 	   const void *, unsafe_ptr)
239 {
240 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
241 }
242 
243 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
244 	.func		= bpf_probe_read_kernel,
245 	.gpl_only	= true,
246 	.ret_type	= RET_INTEGER,
247 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
248 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
249 	.arg3_type	= ARG_ANYTHING,
250 };
251 
252 static __always_inline int
253 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
254 {
255 	int ret;
256 
257 	/*
258 	 * The strncpy_from_kernel_nofault() call will likely not fill the
259 	 * entire buffer, but that's okay in this circumstance as we're probing
260 	 * arbitrary memory anyway similar to bpf_probe_read_*() and might
261 	 * as well probe the stack. Thus, memory is explicitly cleared
262 	 * only in error case, so that improper users ignoring return
263 	 * code altogether don't copy garbage; otherwise length of string
264 	 * is returned that can be used for bpf_perf_event_output() et al.
265 	 */
266 	ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
267 	if (unlikely(ret < 0))
268 		memset(dst, 0, size);
269 	return ret;
270 }
271 
272 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
273 	   const void *, unsafe_ptr)
274 {
275 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
276 }
277 
278 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
279 	.func		= bpf_probe_read_kernel_str,
280 	.gpl_only	= true,
281 	.ret_type	= RET_INTEGER,
282 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
283 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
284 	.arg3_type	= ARG_ANYTHING,
285 };
286 
287 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
288 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
289 	   const void *, unsafe_ptr)
290 {
291 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
292 		return bpf_probe_read_user_common(dst, size,
293 				(__force void __user *)unsafe_ptr);
294 	}
295 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
296 }
297 
298 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
299 	.func		= bpf_probe_read_compat,
300 	.gpl_only	= true,
301 	.ret_type	= RET_INTEGER,
302 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
303 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
304 	.arg3_type	= ARG_ANYTHING,
305 };
306 
307 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
308 	   const void *, unsafe_ptr)
309 {
310 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
311 		return bpf_probe_read_user_str_common(dst, size,
312 				(__force void __user *)unsafe_ptr);
313 	}
314 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
315 }
316 
317 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
318 	.func		= bpf_probe_read_compat_str,
319 	.gpl_only	= true,
320 	.ret_type	= RET_INTEGER,
321 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
322 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
323 	.arg3_type	= ARG_ANYTHING,
324 };
325 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
326 
327 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
328 	   u32, size)
329 {
330 	/*
331 	 * Ensure we're in user context which is safe for the helper to
332 	 * run. This helper has no business in a kthread.
333 	 *
334 	 * access_ok() should prevent writing to non-user memory, but in
335 	 * some situations (nommu, temporary switch, etc) access_ok() does
336 	 * not provide enough validation, hence the check on KERNEL_DS.
337 	 *
338 	 * nmi_uaccess_okay() ensures the probe is not run in an interim
339 	 * state, when the task or mm are switched. This is specifically
340 	 * required to prevent the use of temporary mm.
341 	 */
342 
343 	if (unlikely(in_interrupt() ||
344 		     current->flags & (PF_KTHREAD | PF_EXITING)))
345 		return -EPERM;
346 	if (unlikely(!nmi_uaccess_okay()))
347 		return -EPERM;
348 
349 	return copy_to_user_nofault(unsafe_ptr, src, size);
350 }
351 
352 static const struct bpf_func_proto bpf_probe_write_user_proto = {
353 	.func		= bpf_probe_write_user,
354 	.gpl_only	= true,
355 	.ret_type	= RET_INTEGER,
356 	.arg1_type	= ARG_ANYTHING,
357 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
358 	.arg3_type	= ARG_CONST_SIZE,
359 };
360 
361 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
362 {
363 	if (!capable(CAP_SYS_ADMIN))
364 		return NULL;
365 
366 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
367 			    current->comm, task_pid_nr(current));
368 
369 	return &bpf_probe_write_user_proto;
370 }
371 
372 #define MAX_TRACE_PRINTK_VARARGS	3
373 #define BPF_TRACE_PRINTK_SIZE		1024
374 
375 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
376 	   u64, arg2, u64, arg3)
377 {
378 	u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
379 	struct bpf_bprintf_data data = {
380 		.get_bin_args	= true,
381 		.get_buf	= true,
382 	};
383 	int ret;
384 
385 	ret = bpf_bprintf_prepare(fmt, fmt_size, args,
386 				  MAX_TRACE_PRINTK_VARARGS, &data);
387 	if (ret < 0)
388 		return ret;
389 
390 	ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
391 
392 	trace_bpf_trace_printk(data.buf);
393 
394 	bpf_bprintf_cleanup(&data);
395 
396 	return ret;
397 }
398 
399 static const struct bpf_func_proto bpf_trace_printk_proto = {
400 	.func		= bpf_trace_printk,
401 	.gpl_only	= true,
402 	.ret_type	= RET_INTEGER,
403 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
404 	.arg2_type	= ARG_CONST_SIZE,
405 };
406 
407 static void __set_printk_clr_event(void)
408 {
409 	/*
410 	 * This program might be calling bpf_trace_printk,
411 	 * so enable the associated bpf_trace/bpf_trace_printk event.
412 	 * Repeat this each time as it is possible a user has
413 	 * disabled bpf_trace_printk events.  By loading a program
414 	 * calling bpf_trace_printk() however the user has expressed
415 	 * the intent to see such events.
416 	 */
417 	if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
418 		pr_warn_ratelimited("could not enable bpf_trace_printk events");
419 }
420 
421 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
422 {
423 	__set_printk_clr_event();
424 	return &bpf_trace_printk_proto;
425 }
426 
427 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args,
428 	   u32, data_len)
429 {
430 	struct bpf_bprintf_data data = {
431 		.get_bin_args	= true,
432 		.get_buf	= true,
433 	};
434 	int ret, num_args;
435 
436 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
437 	    (data_len && !args))
438 		return -EINVAL;
439 	num_args = data_len / 8;
440 
441 	ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
442 	if (ret < 0)
443 		return ret;
444 
445 	ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
446 
447 	trace_bpf_trace_printk(data.buf);
448 
449 	bpf_bprintf_cleanup(&data);
450 
451 	return ret;
452 }
453 
454 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
455 	.func		= bpf_trace_vprintk,
456 	.gpl_only	= true,
457 	.ret_type	= RET_INTEGER,
458 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
459 	.arg2_type	= ARG_CONST_SIZE,
460 	.arg3_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
461 	.arg4_type	= ARG_CONST_SIZE_OR_ZERO,
462 };
463 
464 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
465 {
466 	__set_printk_clr_event();
467 	return &bpf_trace_vprintk_proto;
468 }
469 
470 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
471 	   const void *, args, u32, data_len)
472 {
473 	struct bpf_bprintf_data data = {
474 		.get_bin_args	= true,
475 	};
476 	int err, num_args;
477 
478 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
479 	    (data_len && !args))
480 		return -EINVAL;
481 	num_args = data_len / 8;
482 
483 	err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
484 	if (err < 0)
485 		return err;
486 
487 	seq_bprintf(m, fmt, data.bin_args);
488 
489 	bpf_bprintf_cleanup(&data);
490 
491 	return seq_has_overflowed(m) ? -EOVERFLOW : 0;
492 }
493 
494 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
495 
496 static const struct bpf_func_proto bpf_seq_printf_proto = {
497 	.func		= bpf_seq_printf,
498 	.gpl_only	= true,
499 	.ret_type	= RET_INTEGER,
500 	.arg1_type	= ARG_PTR_TO_BTF_ID,
501 	.arg1_btf_id	= &btf_seq_file_ids[0],
502 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
503 	.arg3_type	= ARG_CONST_SIZE,
504 	.arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
505 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
506 };
507 
508 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
509 {
510 	return seq_write(m, data, len) ? -EOVERFLOW : 0;
511 }
512 
513 static const struct bpf_func_proto bpf_seq_write_proto = {
514 	.func		= bpf_seq_write,
515 	.gpl_only	= true,
516 	.ret_type	= RET_INTEGER,
517 	.arg1_type	= ARG_PTR_TO_BTF_ID,
518 	.arg1_btf_id	= &btf_seq_file_ids[0],
519 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
520 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
521 };
522 
523 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
524 	   u32, btf_ptr_size, u64, flags)
525 {
526 	const struct btf *btf;
527 	s32 btf_id;
528 	int ret;
529 
530 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
531 	if (ret)
532 		return ret;
533 
534 	return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
535 }
536 
537 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
538 	.func		= bpf_seq_printf_btf,
539 	.gpl_only	= true,
540 	.ret_type	= RET_INTEGER,
541 	.arg1_type	= ARG_PTR_TO_BTF_ID,
542 	.arg1_btf_id	= &btf_seq_file_ids[0],
543 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
544 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
545 	.arg4_type	= ARG_ANYTHING,
546 };
547 
548 static __always_inline int
549 get_map_perf_counter(struct bpf_map *map, u64 flags,
550 		     u64 *value, u64 *enabled, u64 *running)
551 {
552 	struct bpf_array *array = container_of(map, struct bpf_array, map);
553 	unsigned int cpu = smp_processor_id();
554 	u64 index = flags & BPF_F_INDEX_MASK;
555 	struct bpf_event_entry *ee;
556 
557 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
558 		return -EINVAL;
559 	if (index == BPF_F_CURRENT_CPU)
560 		index = cpu;
561 	if (unlikely(index >= array->map.max_entries))
562 		return -E2BIG;
563 
564 	ee = READ_ONCE(array->ptrs[index]);
565 	if (!ee)
566 		return -ENOENT;
567 
568 	return perf_event_read_local(ee->event, value, enabled, running);
569 }
570 
571 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
572 {
573 	u64 value = 0;
574 	int err;
575 
576 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
577 	/*
578 	 * this api is ugly since we miss [-22..-2] range of valid
579 	 * counter values, but that's uapi
580 	 */
581 	if (err)
582 		return err;
583 	return value;
584 }
585 
586 static const struct bpf_func_proto bpf_perf_event_read_proto = {
587 	.func		= bpf_perf_event_read,
588 	.gpl_only	= true,
589 	.ret_type	= RET_INTEGER,
590 	.arg1_type	= ARG_CONST_MAP_PTR,
591 	.arg2_type	= ARG_ANYTHING,
592 };
593 
594 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
595 	   struct bpf_perf_event_value *, buf, u32, size)
596 {
597 	int err = -EINVAL;
598 
599 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
600 		goto clear;
601 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
602 				   &buf->running);
603 	if (unlikely(err))
604 		goto clear;
605 	return 0;
606 clear:
607 	memset(buf, 0, size);
608 	return err;
609 }
610 
611 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
612 	.func		= bpf_perf_event_read_value,
613 	.gpl_only	= true,
614 	.ret_type	= RET_INTEGER,
615 	.arg1_type	= ARG_CONST_MAP_PTR,
616 	.arg2_type	= ARG_ANYTHING,
617 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
618 	.arg4_type	= ARG_CONST_SIZE,
619 };
620 
621 static __always_inline u64
622 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
623 			u64 flags, struct perf_sample_data *sd)
624 {
625 	struct bpf_array *array = container_of(map, struct bpf_array, map);
626 	unsigned int cpu = smp_processor_id();
627 	u64 index = flags & BPF_F_INDEX_MASK;
628 	struct bpf_event_entry *ee;
629 	struct perf_event *event;
630 
631 	if (index == BPF_F_CURRENT_CPU)
632 		index = cpu;
633 	if (unlikely(index >= array->map.max_entries))
634 		return -E2BIG;
635 
636 	ee = READ_ONCE(array->ptrs[index]);
637 	if (!ee)
638 		return -ENOENT;
639 
640 	event = ee->event;
641 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
642 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
643 		return -EINVAL;
644 
645 	if (unlikely(event->oncpu != cpu))
646 		return -EOPNOTSUPP;
647 
648 	return perf_event_output(event, sd, regs);
649 }
650 
651 /*
652  * Support executing tracepoints in normal, irq, and nmi context that each call
653  * bpf_perf_event_output
654  */
655 struct bpf_trace_sample_data {
656 	struct perf_sample_data sds[3];
657 };
658 
659 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
660 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
661 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
662 	   u64, flags, void *, data, u64, size)
663 {
664 	struct bpf_trace_sample_data *sds;
665 	struct perf_raw_record raw = {
666 		.frag = {
667 			.size = size,
668 			.data = data,
669 		},
670 	};
671 	struct perf_sample_data *sd;
672 	int nest_level, err;
673 
674 	preempt_disable();
675 	sds = this_cpu_ptr(&bpf_trace_sds);
676 	nest_level = this_cpu_inc_return(bpf_trace_nest_level);
677 
678 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
679 		err = -EBUSY;
680 		goto out;
681 	}
682 
683 	sd = &sds->sds[nest_level - 1];
684 
685 	if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
686 		err = -EINVAL;
687 		goto out;
688 	}
689 
690 	perf_sample_data_init(sd, 0, 0);
691 	perf_sample_save_raw_data(sd, &raw);
692 
693 	err = __bpf_perf_event_output(regs, map, flags, sd);
694 out:
695 	this_cpu_dec(bpf_trace_nest_level);
696 	preempt_enable();
697 	return err;
698 }
699 
700 static const struct bpf_func_proto bpf_perf_event_output_proto = {
701 	.func		= bpf_perf_event_output,
702 	.gpl_only	= true,
703 	.ret_type	= RET_INTEGER,
704 	.arg1_type	= ARG_PTR_TO_CTX,
705 	.arg2_type	= ARG_CONST_MAP_PTR,
706 	.arg3_type	= ARG_ANYTHING,
707 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
708 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
709 };
710 
711 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
712 struct bpf_nested_pt_regs {
713 	struct pt_regs regs[3];
714 };
715 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
716 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
717 
718 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
719 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
720 {
721 	struct perf_raw_frag frag = {
722 		.copy		= ctx_copy,
723 		.size		= ctx_size,
724 		.data		= ctx,
725 	};
726 	struct perf_raw_record raw = {
727 		.frag = {
728 			{
729 				.next	= ctx_size ? &frag : NULL,
730 			},
731 			.size	= meta_size,
732 			.data	= meta,
733 		},
734 	};
735 	struct perf_sample_data *sd;
736 	struct pt_regs *regs;
737 	int nest_level;
738 	u64 ret;
739 
740 	preempt_disable();
741 	nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
742 
743 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
744 		ret = -EBUSY;
745 		goto out;
746 	}
747 	sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
748 	regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
749 
750 	perf_fetch_caller_regs(regs);
751 	perf_sample_data_init(sd, 0, 0);
752 	perf_sample_save_raw_data(sd, &raw);
753 
754 	ret = __bpf_perf_event_output(regs, map, flags, sd);
755 out:
756 	this_cpu_dec(bpf_event_output_nest_level);
757 	preempt_enable();
758 	return ret;
759 }
760 
761 BPF_CALL_0(bpf_get_current_task)
762 {
763 	return (long) current;
764 }
765 
766 const struct bpf_func_proto bpf_get_current_task_proto = {
767 	.func		= bpf_get_current_task,
768 	.gpl_only	= true,
769 	.ret_type	= RET_INTEGER,
770 };
771 
772 BPF_CALL_0(bpf_get_current_task_btf)
773 {
774 	return (unsigned long) current;
775 }
776 
777 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
778 	.func		= bpf_get_current_task_btf,
779 	.gpl_only	= true,
780 	.ret_type	= RET_PTR_TO_BTF_ID_TRUSTED,
781 	.ret_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
782 };
783 
784 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
785 {
786 	return (unsigned long) task_pt_regs(task);
787 }
788 
789 BTF_ID_LIST(bpf_task_pt_regs_ids)
790 BTF_ID(struct, pt_regs)
791 
792 const struct bpf_func_proto bpf_task_pt_regs_proto = {
793 	.func		= bpf_task_pt_regs,
794 	.gpl_only	= true,
795 	.arg1_type	= ARG_PTR_TO_BTF_ID,
796 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
797 	.ret_type	= RET_PTR_TO_BTF_ID,
798 	.ret_btf_id	= &bpf_task_pt_regs_ids[0],
799 };
800 
801 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
802 {
803 	struct bpf_array *array = container_of(map, struct bpf_array, map);
804 	struct cgroup *cgrp;
805 
806 	if (unlikely(idx >= array->map.max_entries))
807 		return -E2BIG;
808 
809 	cgrp = READ_ONCE(array->ptrs[idx]);
810 	if (unlikely(!cgrp))
811 		return -EAGAIN;
812 
813 	return task_under_cgroup_hierarchy(current, cgrp);
814 }
815 
816 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
817 	.func           = bpf_current_task_under_cgroup,
818 	.gpl_only       = false,
819 	.ret_type       = RET_INTEGER,
820 	.arg1_type      = ARG_CONST_MAP_PTR,
821 	.arg2_type      = ARG_ANYTHING,
822 };
823 
824 struct send_signal_irq_work {
825 	struct irq_work irq_work;
826 	struct task_struct *task;
827 	u32 sig;
828 	enum pid_type type;
829 };
830 
831 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
832 
833 static void do_bpf_send_signal(struct irq_work *entry)
834 {
835 	struct send_signal_irq_work *work;
836 
837 	work = container_of(entry, struct send_signal_irq_work, irq_work);
838 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
839 	put_task_struct(work->task);
840 }
841 
842 static int bpf_send_signal_common(u32 sig, enum pid_type type)
843 {
844 	struct send_signal_irq_work *work = NULL;
845 
846 	/* Similar to bpf_probe_write_user, task needs to be
847 	 * in a sound condition and kernel memory access be
848 	 * permitted in order to send signal to the current
849 	 * task.
850 	 */
851 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
852 		return -EPERM;
853 	if (unlikely(!nmi_uaccess_okay()))
854 		return -EPERM;
855 	/* Task should not be pid=1 to avoid kernel panic. */
856 	if (unlikely(is_global_init(current)))
857 		return -EPERM;
858 
859 	if (irqs_disabled()) {
860 		/* Do an early check on signal validity. Otherwise,
861 		 * the error is lost in deferred irq_work.
862 		 */
863 		if (unlikely(!valid_signal(sig)))
864 			return -EINVAL;
865 
866 		work = this_cpu_ptr(&send_signal_work);
867 		if (irq_work_is_busy(&work->irq_work))
868 			return -EBUSY;
869 
870 		/* Add the current task, which is the target of sending signal,
871 		 * to the irq_work. The current task may change when queued
872 		 * irq works get executed.
873 		 */
874 		work->task = get_task_struct(current);
875 		work->sig = sig;
876 		work->type = type;
877 		irq_work_queue(&work->irq_work);
878 		return 0;
879 	}
880 
881 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
882 }
883 
884 BPF_CALL_1(bpf_send_signal, u32, sig)
885 {
886 	return bpf_send_signal_common(sig, PIDTYPE_TGID);
887 }
888 
889 static const struct bpf_func_proto bpf_send_signal_proto = {
890 	.func		= bpf_send_signal,
891 	.gpl_only	= false,
892 	.ret_type	= RET_INTEGER,
893 	.arg1_type	= ARG_ANYTHING,
894 };
895 
896 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
897 {
898 	return bpf_send_signal_common(sig, PIDTYPE_PID);
899 }
900 
901 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
902 	.func		= bpf_send_signal_thread,
903 	.gpl_only	= false,
904 	.ret_type	= RET_INTEGER,
905 	.arg1_type	= ARG_ANYTHING,
906 };
907 
908 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
909 {
910 	struct path copy;
911 	long len;
912 	char *p;
913 
914 	if (!sz)
915 		return 0;
916 
917 	/*
918 	 * The path pointer is verified as trusted and safe to use,
919 	 * but let's double check it's valid anyway to workaround
920 	 * potentially broken verifier.
921 	 */
922 	len = copy_from_kernel_nofault(&copy, path, sizeof(*path));
923 	if (len < 0)
924 		return len;
925 
926 	p = d_path(&copy, buf, sz);
927 	if (IS_ERR(p)) {
928 		len = PTR_ERR(p);
929 	} else {
930 		len = buf + sz - p;
931 		memmove(buf, p, len);
932 	}
933 
934 	return len;
935 }
936 
937 BTF_SET_START(btf_allowlist_d_path)
938 #ifdef CONFIG_SECURITY
939 BTF_ID(func, security_file_permission)
940 BTF_ID(func, security_inode_getattr)
941 BTF_ID(func, security_file_open)
942 #endif
943 #ifdef CONFIG_SECURITY_PATH
944 BTF_ID(func, security_path_truncate)
945 #endif
946 BTF_ID(func, vfs_truncate)
947 BTF_ID(func, vfs_fallocate)
948 BTF_ID(func, dentry_open)
949 BTF_ID(func, vfs_getattr)
950 BTF_ID(func, filp_close)
951 BTF_SET_END(btf_allowlist_d_path)
952 
953 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
954 {
955 	if (prog->type == BPF_PROG_TYPE_TRACING &&
956 	    prog->expected_attach_type == BPF_TRACE_ITER)
957 		return true;
958 
959 	if (prog->type == BPF_PROG_TYPE_LSM)
960 		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
961 
962 	return btf_id_set_contains(&btf_allowlist_d_path,
963 				   prog->aux->attach_btf_id);
964 }
965 
966 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
967 
968 static const struct bpf_func_proto bpf_d_path_proto = {
969 	.func		= bpf_d_path,
970 	.gpl_only	= false,
971 	.ret_type	= RET_INTEGER,
972 	.arg1_type	= ARG_PTR_TO_BTF_ID,
973 	.arg1_btf_id	= &bpf_d_path_btf_ids[0],
974 	.arg2_type	= ARG_PTR_TO_MEM,
975 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
976 	.allowed	= bpf_d_path_allowed,
977 };
978 
979 #define BTF_F_ALL	(BTF_F_COMPACT  | BTF_F_NONAME | \
980 			 BTF_F_PTR_RAW | BTF_F_ZERO)
981 
982 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
983 				  u64 flags, const struct btf **btf,
984 				  s32 *btf_id)
985 {
986 	const struct btf_type *t;
987 
988 	if (unlikely(flags & ~(BTF_F_ALL)))
989 		return -EINVAL;
990 
991 	if (btf_ptr_size != sizeof(struct btf_ptr))
992 		return -EINVAL;
993 
994 	*btf = bpf_get_btf_vmlinux();
995 
996 	if (IS_ERR_OR_NULL(*btf))
997 		return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
998 
999 	if (ptr->type_id > 0)
1000 		*btf_id = ptr->type_id;
1001 	else
1002 		return -EINVAL;
1003 
1004 	if (*btf_id > 0)
1005 		t = btf_type_by_id(*btf, *btf_id);
1006 	if (*btf_id <= 0 || !t)
1007 		return -ENOENT;
1008 
1009 	return 0;
1010 }
1011 
1012 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1013 	   u32, btf_ptr_size, u64, flags)
1014 {
1015 	const struct btf *btf;
1016 	s32 btf_id;
1017 	int ret;
1018 
1019 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1020 	if (ret)
1021 		return ret;
1022 
1023 	return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1024 				      flags);
1025 }
1026 
1027 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1028 	.func		= bpf_snprintf_btf,
1029 	.gpl_only	= false,
1030 	.ret_type	= RET_INTEGER,
1031 	.arg1_type	= ARG_PTR_TO_MEM,
1032 	.arg2_type	= ARG_CONST_SIZE,
1033 	.arg3_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1034 	.arg4_type	= ARG_CONST_SIZE,
1035 	.arg5_type	= ARG_ANYTHING,
1036 };
1037 
1038 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1039 {
1040 	/* This helper call is inlined by verifier. */
1041 	return ((u64 *)ctx)[-2];
1042 }
1043 
1044 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1045 	.func		= bpf_get_func_ip_tracing,
1046 	.gpl_only	= true,
1047 	.ret_type	= RET_INTEGER,
1048 	.arg1_type	= ARG_PTR_TO_CTX,
1049 };
1050 
1051 #ifdef CONFIG_X86_KERNEL_IBT
1052 static unsigned long get_entry_ip(unsigned long fentry_ip)
1053 {
1054 	u32 instr;
1055 
1056 	/* We want to be extra safe in case entry ip is on the page edge,
1057 	 * but otherwise we need to avoid get_kernel_nofault()'s overhead.
1058 	 */
1059 	if ((fentry_ip & ~PAGE_MASK) < ENDBR_INSN_SIZE) {
1060 		if (get_kernel_nofault(instr, (u32 *)(fentry_ip - ENDBR_INSN_SIZE)))
1061 			return fentry_ip;
1062 	} else {
1063 		instr = *(u32 *)(fentry_ip - ENDBR_INSN_SIZE);
1064 	}
1065 	if (is_endbr(instr))
1066 		fentry_ip -= ENDBR_INSN_SIZE;
1067 	return fentry_ip;
1068 }
1069 #else
1070 #define get_entry_ip(fentry_ip) fentry_ip
1071 #endif
1072 
1073 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1074 {
1075 	struct bpf_trace_run_ctx *run_ctx __maybe_unused;
1076 	struct kprobe *kp;
1077 
1078 #ifdef CONFIG_UPROBES
1079 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1080 	if (run_ctx->is_uprobe)
1081 		return ((struct uprobe_dispatch_data *)current->utask->vaddr)->bp_addr;
1082 #endif
1083 
1084 	kp = kprobe_running();
1085 
1086 	if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1087 		return 0;
1088 
1089 	return get_entry_ip((uintptr_t)kp->addr);
1090 }
1091 
1092 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1093 	.func		= bpf_get_func_ip_kprobe,
1094 	.gpl_only	= true,
1095 	.ret_type	= RET_INTEGER,
1096 	.arg1_type	= ARG_PTR_TO_CTX,
1097 };
1098 
1099 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1100 {
1101 	return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1102 }
1103 
1104 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1105 	.func		= bpf_get_func_ip_kprobe_multi,
1106 	.gpl_only	= false,
1107 	.ret_type	= RET_INTEGER,
1108 	.arg1_type	= ARG_PTR_TO_CTX,
1109 };
1110 
1111 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1112 {
1113 	return bpf_kprobe_multi_cookie(current->bpf_ctx);
1114 }
1115 
1116 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1117 	.func		= bpf_get_attach_cookie_kprobe_multi,
1118 	.gpl_only	= false,
1119 	.ret_type	= RET_INTEGER,
1120 	.arg1_type	= ARG_PTR_TO_CTX,
1121 };
1122 
1123 BPF_CALL_1(bpf_get_func_ip_uprobe_multi, struct pt_regs *, regs)
1124 {
1125 	return bpf_uprobe_multi_entry_ip(current->bpf_ctx);
1126 }
1127 
1128 static const struct bpf_func_proto bpf_get_func_ip_proto_uprobe_multi = {
1129 	.func		= bpf_get_func_ip_uprobe_multi,
1130 	.gpl_only	= false,
1131 	.ret_type	= RET_INTEGER,
1132 	.arg1_type	= ARG_PTR_TO_CTX,
1133 };
1134 
1135 BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi, struct pt_regs *, regs)
1136 {
1137 	return bpf_uprobe_multi_cookie(current->bpf_ctx);
1138 }
1139 
1140 static const struct bpf_func_proto bpf_get_attach_cookie_proto_umulti = {
1141 	.func		= bpf_get_attach_cookie_uprobe_multi,
1142 	.gpl_only	= false,
1143 	.ret_type	= RET_INTEGER,
1144 	.arg1_type	= ARG_PTR_TO_CTX,
1145 };
1146 
1147 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1148 {
1149 	struct bpf_trace_run_ctx *run_ctx;
1150 
1151 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1152 	return run_ctx->bpf_cookie;
1153 }
1154 
1155 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1156 	.func		= bpf_get_attach_cookie_trace,
1157 	.gpl_only	= false,
1158 	.ret_type	= RET_INTEGER,
1159 	.arg1_type	= ARG_PTR_TO_CTX,
1160 };
1161 
1162 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1163 {
1164 	return ctx->event->bpf_cookie;
1165 }
1166 
1167 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1168 	.func		= bpf_get_attach_cookie_pe,
1169 	.gpl_only	= false,
1170 	.ret_type	= RET_INTEGER,
1171 	.arg1_type	= ARG_PTR_TO_CTX,
1172 };
1173 
1174 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1175 {
1176 	struct bpf_trace_run_ctx *run_ctx;
1177 
1178 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1179 	return run_ctx->bpf_cookie;
1180 }
1181 
1182 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1183 	.func		= bpf_get_attach_cookie_tracing,
1184 	.gpl_only	= false,
1185 	.ret_type	= RET_INTEGER,
1186 	.arg1_type	= ARG_PTR_TO_CTX,
1187 };
1188 
1189 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1190 {
1191 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1192 	u32 entry_cnt = size / br_entry_size;
1193 
1194 	entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1195 
1196 	if (unlikely(flags))
1197 		return -EINVAL;
1198 
1199 	if (!entry_cnt)
1200 		return -ENOENT;
1201 
1202 	return entry_cnt * br_entry_size;
1203 }
1204 
1205 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1206 	.func		= bpf_get_branch_snapshot,
1207 	.gpl_only	= true,
1208 	.ret_type	= RET_INTEGER,
1209 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1210 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1211 };
1212 
1213 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1214 {
1215 	/* This helper call is inlined by verifier. */
1216 	u64 nr_args = ((u64 *)ctx)[-1];
1217 
1218 	if ((u64) n >= nr_args)
1219 		return -EINVAL;
1220 	*value = ((u64 *)ctx)[n];
1221 	return 0;
1222 }
1223 
1224 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1225 	.func		= get_func_arg,
1226 	.ret_type	= RET_INTEGER,
1227 	.arg1_type	= ARG_PTR_TO_CTX,
1228 	.arg2_type	= ARG_ANYTHING,
1229 	.arg3_type	= ARG_PTR_TO_LONG,
1230 };
1231 
1232 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1233 {
1234 	/* This helper call is inlined by verifier. */
1235 	u64 nr_args = ((u64 *)ctx)[-1];
1236 
1237 	*value = ((u64 *)ctx)[nr_args];
1238 	return 0;
1239 }
1240 
1241 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1242 	.func		= get_func_ret,
1243 	.ret_type	= RET_INTEGER,
1244 	.arg1_type	= ARG_PTR_TO_CTX,
1245 	.arg2_type	= ARG_PTR_TO_LONG,
1246 };
1247 
1248 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1249 {
1250 	/* This helper call is inlined by verifier. */
1251 	return ((u64 *)ctx)[-1];
1252 }
1253 
1254 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1255 	.func		= get_func_arg_cnt,
1256 	.ret_type	= RET_INTEGER,
1257 	.arg1_type	= ARG_PTR_TO_CTX,
1258 };
1259 
1260 #ifdef CONFIG_KEYS
1261 __bpf_kfunc_start_defs();
1262 
1263 /**
1264  * bpf_lookup_user_key - lookup a key by its serial
1265  * @serial: key handle serial number
1266  * @flags: lookup-specific flags
1267  *
1268  * Search a key with a given *serial* and the provided *flags*.
1269  * If found, increment the reference count of the key by one, and
1270  * return it in the bpf_key structure.
1271  *
1272  * The bpf_key structure must be passed to bpf_key_put() when done
1273  * with it, so that the key reference count is decremented and the
1274  * bpf_key structure is freed.
1275  *
1276  * Permission checks are deferred to the time the key is used by
1277  * one of the available key-specific kfuncs.
1278  *
1279  * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1280  * special keyring (e.g. session keyring), if it doesn't yet exist.
1281  * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1282  * for the key construction, and to retrieve uninstantiated keys (keys
1283  * without data attached to them).
1284  *
1285  * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1286  *         NULL pointer otherwise.
1287  */
1288 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1289 {
1290 	key_ref_t key_ref;
1291 	struct bpf_key *bkey;
1292 
1293 	if (flags & ~KEY_LOOKUP_ALL)
1294 		return NULL;
1295 
1296 	/*
1297 	 * Permission check is deferred until the key is used, as the
1298 	 * intent of the caller is unknown here.
1299 	 */
1300 	key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1301 	if (IS_ERR(key_ref))
1302 		return NULL;
1303 
1304 	bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1305 	if (!bkey) {
1306 		key_put(key_ref_to_ptr(key_ref));
1307 		return NULL;
1308 	}
1309 
1310 	bkey->key = key_ref_to_ptr(key_ref);
1311 	bkey->has_ref = true;
1312 
1313 	return bkey;
1314 }
1315 
1316 /**
1317  * bpf_lookup_system_key - lookup a key by a system-defined ID
1318  * @id: key ID
1319  *
1320  * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1321  * The key pointer is marked as invalid, to prevent bpf_key_put() from
1322  * attempting to decrement the key reference count on that pointer. The key
1323  * pointer set in such way is currently understood only by
1324  * verify_pkcs7_signature().
1325  *
1326  * Set *id* to one of the values defined in include/linux/verification.h:
1327  * 0 for the primary keyring (immutable keyring of system keys);
1328  * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1329  * (where keys can be added only if they are vouched for by existing keys
1330  * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1331  * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1332  * kerned image and, possibly, the initramfs signature).
1333  *
1334  * Return: a bpf_key pointer with an invalid key pointer set from the
1335  *         pre-determined ID on success, a NULL pointer otherwise
1336  */
1337 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id)
1338 {
1339 	struct bpf_key *bkey;
1340 
1341 	if (system_keyring_id_check(id) < 0)
1342 		return NULL;
1343 
1344 	bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1345 	if (!bkey)
1346 		return NULL;
1347 
1348 	bkey->key = (struct key *)(unsigned long)id;
1349 	bkey->has_ref = false;
1350 
1351 	return bkey;
1352 }
1353 
1354 /**
1355  * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1356  * @bkey: bpf_key structure
1357  *
1358  * Decrement the reference count of the key inside *bkey*, if the pointer
1359  * is valid, and free *bkey*.
1360  */
1361 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey)
1362 {
1363 	if (bkey->has_ref)
1364 		key_put(bkey->key);
1365 
1366 	kfree(bkey);
1367 }
1368 
1369 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1370 /**
1371  * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1372  * @data_p: data to verify
1373  * @sig_p: signature of the data
1374  * @trusted_keyring: keyring with keys trusted for signature verification
1375  *
1376  * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1377  * with keys in a keyring referenced by *trusted_keyring*.
1378  *
1379  * Return: 0 on success, a negative value on error.
1380  */
1381 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_p,
1382 			       struct bpf_dynptr *sig_p,
1383 			       struct bpf_key *trusted_keyring)
1384 {
1385 	struct bpf_dynptr_kern *data_ptr = (struct bpf_dynptr_kern *)data_p;
1386 	struct bpf_dynptr_kern *sig_ptr = (struct bpf_dynptr_kern *)sig_p;
1387 	const void *data, *sig;
1388 	u32 data_len, sig_len;
1389 	int ret;
1390 
1391 	if (trusted_keyring->has_ref) {
1392 		/*
1393 		 * Do the permission check deferred in bpf_lookup_user_key().
1394 		 * See bpf_lookup_user_key() for more details.
1395 		 *
1396 		 * A call to key_task_permission() here would be redundant, as
1397 		 * it is already done by keyring_search() called by
1398 		 * find_asymmetric_key().
1399 		 */
1400 		ret = key_validate(trusted_keyring->key);
1401 		if (ret < 0)
1402 			return ret;
1403 	}
1404 
1405 	data_len = __bpf_dynptr_size(data_ptr);
1406 	data = __bpf_dynptr_data(data_ptr, data_len);
1407 	sig_len = __bpf_dynptr_size(sig_ptr);
1408 	sig = __bpf_dynptr_data(sig_ptr, sig_len);
1409 
1410 	return verify_pkcs7_signature(data, data_len, sig, sig_len,
1411 				      trusted_keyring->key,
1412 				      VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1413 				      NULL);
1414 }
1415 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1416 
1417 __bpf_kfunc_end_defs();
1418 
1419 BTF_KFUNCS_START(key_sig_kfunc_set)
1420 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1421 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1422 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1423 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1424 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1425 #endif
1426 BTF_KFUNCS_END(key_sig_kfunc_set)
1427 
1428 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1429 	.owner = THIS_MODULE,
1430 	.set = &key_sig_kfunc_set,
1431 };
1432 
1433 static int __init bpf_key_sig_kfuncs_init(void)
1434 {
1435 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1436 					 &bpf_key_sig_kfunc_set);
1437 }
1438 
1439 late_initcall(bpf_key_sig_kfuncs_init);
1440 #endif /* CONFIG_KEYS */
1441 
1442 /* filesystem kfuncs */
1443 __bpf_kfunc_start_defs();
1444 
1445 /**
1446  * bpf_get_file_xattr - get xattr of a file
1447  * @file: file to get xattr from
1448  * @name__str: name of the xattr
1449  * @value_p: output buffer of the xattr value
1450  *
1451  * Get xattr *name__str* of *file* and store the output in *value_ptr*.
1452  *
1453  * For security reasons, only *name__str* with prefix "user." is allowed.
1454  *
1455  * Return: 0 on success, a negative value on error.
1456  */
1457 __bpf_kfunc int bpf_get_file_xattr(struct file *file, const char *name__str,
1458 				   struct bpf_dynptr *value_p)
1459 {
1460 	struct bpf_dynptr_kern *value_ptr = (struct bpf_dynptr_kern *)value_p;
1461 	struct dentry *dentry;
1462 	u32 value_len;
1463 	void *value;
1464 	int ret;
1465 
1466 	if (strncmp(name__str, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
1467 		return -EPERM;
1468 
1469 	value_len = __bpf_dynptr_size(value_ptr);
1470 	value = __bpf_dynptr_data_rw(value_ptr, value_len);
1471 	if (!value)
1472 		return -EINVAL;
1473 
1474 	dentry = file_dentry(file);
1475 	ret = inode_permission(&nop_mnt_idmap, dentry->d_inode, MAY_READ);
1476 	if (ret)
1477 		return ret;
1478 	return __vfs_getxattr(dentry, dentry->d_inode, name__str, value, value_len);
1479 }
1480 
1481 __bpf_kfunc_end_defs();
1482 
1483 BTF_KFUNCS_START(fs_kfunc_set_ids)
1484 BTF_ID_FLAGS(func, bpf_get_file_xattr, KF_SLEEPABLE | KF_TRUSTED_ARGS)
1485 BTF_KFUNCS_END(fs_kfunc_set_ids)
1486 
1487 static int bpf_get_file_xattr_filter(const struct bpf_prog *prog, u32 kfunc_id)
1488 {
1489 	if (!btf_id_set8_contains(&fs_kfunc_set_ids, kfunc_id))
1490 		return 0;
1491 
1492 	/* Only allow to attach from LSM hooks, to avoid recursion */
1493 	return prog->type != BPF_PROG_TYPE_LSM ? -EACCES : 0;
1494 }
1495 
1496 static const struct btf_kfunc_id_set bpf_fs_kfunc_set = {
1497 	.owner = THIS_MODULE,
1498 	.set = &fs_kfunc_set_ids,
1499 	.filter = bpf_get_file_xattr_filter,
1500 };
1501 
1502 static int __init bpf_fs_kfuncs_init(void)
1503 {
1504 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_LSM, &bpf_fs_kfunc_set);
1505 }
1506 
1507 late_initcall(bpf_fs_kfuncs_init);
1508 
1509 static const struct bpf_func_proto *
1510 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1511 {
1512 	switch (func_id) {
1513 	case BPF_FUNC_map_lookup_elem:
1514 		return &bpf_map_lookup_elem_proto;
1515 	case BPF_FUNC_map_update_elem:
1516 		return &bpf_map_update_elem_proto;
1517 	case BPF_FUNC_map_delete_elem:
1518 		return &bpf_map_delete_elem_proto;
1519 	case BPF_FUNC_map_push_elem:
1520 		return &bpf_map_push_elem_proto;
1521 	case BPF_FUNC_map_pop_elem:
1522 		return &bpf_map_pop_elem_proto;
1523 	case BPF_FUNC_map_peek_elem:
1524 		return &bpf_map_peek_elem_proto;
1525 	case BPF_FUNC_map_lookup_percpu_elem:
1526 		return &bpf_map_lookup_percpu_elem_proto;
1527 	case BPF_FUNC_ktime_get_ns:
1528 		return &bpf_ktime_get_ns_proto;
1529 	case BPF_FUNC_ktime_get_boot_ns:
1530 		return &bpf_ktime_get_boot_ns_proto;
1531 	case BPF_FUNC_tail_call:
1532 		return &bpf_tail_call_proto;
1533 	case BPF_FUNC_get_current_task:
1534 		return &bpf_get_current_task_proto;
1535 	case BPF_FUNC_get_current_task_btf:
1536 		return &bpf_get_current_task_btf_proto;
1537 	case BPF_FUNC_task_pt_regs:
1538 		return &bpf_task_pt_regs_proto;
1539 	case BPF_FUNC_get_current_uid_gid:
1540 		return &bpf_get_current_uid_gid_proto;
1541 	case BPF_FUNC_get_current_comm:
1542 		return &bpf_get_current_comm_proto;
1543 	case BPF_FUNC_trace_printk:
1544 		return bpf_get_trace_printk_proto();
1545 	case BPF_FUNC_get_smp_processor_id:
1546 		return &bpf_get_smp_processor_id_proto;
1547 	case BPF_FUNC_get_numa_node_id:
1548 		return &bpf_get_numa_node_id_proto;
1549 	case BPF_FUNC_perf_event_read:
1550 		return &bpf_perf_event_read_proto;
1551 	case BPF_FUNC_current_task_under_cgroup:
1552 		return &bpf_current_task_under_cgroup_proto;
1553 	case BPF_FUNC_get_prandom_u32:
1554 		return &bpf_get_prandom_u32_proto;
1555 	case BPF_FUNC_probe_write_user:
1556 		return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1557 		       NULL : bpf_get_probe_write_proto();
1558 	case BPF_FUNC_probe_read_user:
1559 		return &bpf_probe_read_user_proto;
1560 	case BPF_FUNC_probe_read_kernel:
1561 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1562 		       NULL : &bpf_probe_read_kernel_proto;
1563 	case BPF_FUNC_probe_read_user_str:
1564 		return &bpf_probe_read_user_str_proto;
1565 	case BPF_FUNC_probe_read_kernel_str:
1566 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1567 		       NULL : &bpf_probe_read_kernel_str_proto;
1568 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1569 	case BPF_FUNC_probe_read:
1570 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1571 		       NULL : &bpf_probe_read_compat_proto;
1572 	case BPF_FUNC_probe_read_str:
1573 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1574 		       NULL : &bpf_probe_read_compat_str_proto;
1575 #endif
1576 #ifdef CONFIG_CGROUPS
1577 	case BPF_FUNC_cgrp_storage_get:
1578 		return &bpf_cgrp_storage_get_proto;
1579 	case BPF_FUNC_cgrp_storage_delete:
1580 		return &bpf_cgrp_storage_delete_proto;
1581 #endif
1582 	case BPF_FUNC_send_signal:
1583 		return &bpf_send_signal_proto;
1584 	case BPF_FUNC_send_signal_thread:
1585 		return &bpf_send_signal_thread_proto;
1586 	case BPF_FUNC_perf_event_read_value:
1587 		return &bpf_perf_event_read_value_proto;
1588 	case BPF_FUNC_ringbuf_output:
1589 		return &bpf_ringbuf_output_proto;
1590 	case BPF_FUNC_ringbuf_reserve:
1591 		return &bpf_ringbuf_reserve_proto;
1592 	case BPF_FUNC_ringbuf_submit:
1593 		return &bpf_ringbuf_submit_proto;
1594 	case BPF_FUNC_ringbuf_discard:
1595 		return &bpf_ringbuf_discard_proto;
1596 	case BPF_FUNC_ringbuf_query:
1597 		return &bpf_ringbuf_query_proto;
1598 	case BPF_FUNC_jiffies64:
1599 		return &bpf_jiffies64_proto;
1600 	case BPF_FUNC_get_task_stack:
1601 		return &bpf_get_task_stack_proto;
1602 	case BPF_FUNC_copy_from_user:
1603 		return &bpf_copy_from_user_proto;
1604 	case BPF_FUNC_copy_from_user_task:
1605 		return &bpf_copy_from_user_task_proto;
1606 	case BPF_FUNC_snprintf_btf:
1607 		return &bpf_snprintf_btf_proto;
1608 	case BPF_FUNC_per_cpu_ptr:
1609 		return &bpf_per_cpu_ptr_proto;
1610 	case BPF_FUNC_this_cpu_ptr:
1611 		return &bpf_this_cpu_ptr_proto;
1612 	case BPF_FUNC_task_storage_get:
1613 		if (bpf_prog_check_recur(prog))
1614 			return &bpf_task_storage_get_recur_proto;
1615 		return &bpf_task_storage_get_proto;
1616 	case BPF_FUNC_task_storage_delete:
1617 		if (bpf_prog_check_recur(prog))
1618 			return &bpf_task_storage_delete_recur_proto;
1619 		return &bpf_task_storage_delete_proto;
1620 	case BPF_FUNC_for_each_map_elem:
1621 		return &bpf_for_each_map_elem_proto;
1622 	case BPF_FUNC_snprintf:
1623 		return &bpf_snprintf_proto;
1624 	case BPF_FUNC_get_func_ip:
1625 		return &bpf_get_func_ip_proto_tracing;
1626 	case BPF_FUNC_get_branch_snapshot:
1627 		return &bpf_get_branch_snapshot_proto;
1628 	case BPF_FUNC_find_vma:
1629 		return &bpf_find_vma_proto;
1630 	case BPF_FUNC_trace_vprintk:
1631 		return bpf_get_trace_vprintk_proto();
1632 	default:
1633 		return bpf_base_func_proto(func_id, prog);
1634 	}
1635 }
1636 
1637 static bool is_kprobe_multi(const struct bpf_prog *prog)
1638 {
1639 	return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ||
1640 	       prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1641 }
1642 
1643 static inline bool is_kprobe_session(const struct bpf_prog *prog)
1644 {
1645 	return prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1646 }
1647 
1648 static const struct bpf_func_proto *
1649 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1650 {
1651 	switch (func_id) {
1652 	case BPF_FUNC_perf_event_output:
1653 		return &bpf_perf_event_output_proto;
1654 	case BPF_FUNC_get_stackid:
1655 		return &bpf_get_stackid_proto;
1656 	case BPF_FUNC_get_stack:
1657 		return &bpf_get_stack_proto;
1658 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1659 	case BPF_FUNC_override_return:
1660 		return &bpf_override_return_proto;
1661 #endif
1662 	case BPF_FUNC_get_func_ip:
1663 		if (is_kprobe_multi(prog))
1664 			return &bpf_get_func_ip_proto_kprobe_multi;
1665 		if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1666 			return &bpf_get_func_ip_proto_uprobe_multi;
1667 		return &bpf_get_func_ip_proto_kprobe;
1668 	case BPF_FUNC_get_attach_cookie:
1669 		if (is_kprobe_multi(prog))
1670 			return &bpf_get_attach_cookie_proto_kmulti;
1671 		if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1672 			return &bpf_get_attach_cookie_proto_umulti;
1673 		return &bpf_get_attach_cookie_proto_trace;
1674 	default:
1675 		return bpf_tracing_func_proto(func_id, prog);
1676 	}
1677 }
1678 
1679 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1680 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1681 					const struct bpf_prog *prog,
1682 					struct bpf_insn_access_aux *info)
1683 {
1684 	if (off < 0 || off >= sizeof(struct pt_regs))
1685 		return false;
1686 	if (type != BPF_READ)
1687 		return false;
1688 	if (off % size != 0)
1689 		return false;
1690 	/*
1691 	 * Assertion for 32 bit to make sure last 8 byte access
1692 	 * (BPF_DW) to the last 4 byte member is disallowed.
1693 	 */
1694 	if (off + size > sizeof(struct pt_regs))
1695 		return false;
1696 
1697 	return true;
1698 }
1699 
1700 const struct bpf_verifier_ops kprobe_verifier_ops = {
1701 	.get_func_proto  = kprobe_prog_func_proto,
1702 	.is_valid_access = kprobe_prog_is_valid_access,
1703 };
1704 
1705 const struct bpf_prog_ops kprobe_prog_ops = {
1706 };
1707 
1708 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1709 	   u64, flags, void *, data, u64, size)
1710 {
1711 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1712 
1713 	/*
1714 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1715 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1716 	 * from there and call the same bpf_perf_event_output() helper inline.
1717 	 */
1718 	return ____bpf_perf_event_output(regs, map, flags, data, size);
1719 }
1720 
1721 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1722 	.func		= bpf_perf_event_output_tp,
1723 	.gpl_only	= true,
1724 	.ret_type	= RET_INTEGER,
1725 	.arg1_type	= ARG_PTR_TO_CTX,
1726 	.arg2_type	= ARG_CONST_MAP_PTR,
1727 	.arg3_type	= ARG_ANYTHING,
1728 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1729 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1730 };
1731 
1732 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1733 	   u64, flags)
1734 {
1735 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1736 
1737 	/*
1738 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
1739 	 * the other helper's function body cannot be inlined due to being
1740 	 * external, thus we need to call raw helper function.
1741 	 */
1742 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1743 			       flags, 0, 0);
1744 }
1745 
1746 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1747 	.func		= bpf_get_stackid_tp,
1748 	.gpl_only	= true,
1749 	.ret_type	= RET_INTEGER,
1750 	.arg1_type	= ARG_PTR_TO_CTX,
1751 	.arg2_type	= ARG_CONST_MAP_PTR,
1752 	.arg3_type	= ARG_ANYTHING,
1753 };
1754 
1755 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1756 	   u64, flags)
1757 {
1758 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1759 
1760 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1761 			     (unsigned long) size, flags, 0);
1762 }
1763 
1764 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1765 	.func		= bpf_get_stack_tp,
1766 	.gpl_only	= true,
1767 	.ret_type	= RET_INTEGER,
1768 	.arg1_type	= ARG_PTR_TO_CTX,
1769 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
1770 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1771 	.arg4_type	= ARG_ANYTHING,
1772 };
1773 
1774 static const struct bpf_func_proto *
1775 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1776 {
1777 	switch (func_id) {
1778 	case BPF_FUNC_perf_event_output:
1779 		return &bpf_perf_event_output_proto_tp;
1780 	case BPF_FUNC_get_stackid:
1781 		return &bpf_get_stackid_proto_tp;
1782 	case BPF_FUNC_get_stack:
1783 		return &bpf_get_stack_proto_tp;
1784 	case BPF_FUNC_get_attach_cookie:
1785 		return &bpf_get_attach_cookie_proto_trace;
1786 	default:
1787 		return bpf_tracing_func_proto(func_id, prog);
1788 	}
1789 }
1790 
1791 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1792 				    const struct bpf_prog *prog,
1793 				    struct bpf_insn_access_aux *info)
1794 {
1795 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1796 		return false;
1797 	if (type != BPF_READ)
1798 		return false;
1799 	if (off % size != 0)
1800 		return false;
1801 
1802 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1803 	return true;
1804 }
1805 
1806 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1807 	.get_func_proto  = tp_prog_func_proto,
1808 	.is_valid_access = tp_prog_is_valid_access,
1809 };
1810 
1811 const struct bpf_prog_ops tracepoint_prog_ops = {
1812 };
1813 
1814 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1815 	   struct bpf_perf_event_value *, buf, u32, size)
1816 {
1817 	int err = -EINVAL;
1818 
1819 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1820 		goto clear;
1821 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1822 				    &buf->running);
1823 	if (unlikely(err))
1824 		goto clear;
1825 	return 0;
1826 clear:
1827 	memset(buf, 0, size);
1828 	return err;
1829 }
1830 
1831 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1832          .func           = bpf_perf_prog_read_value,
1833          .gpl_only       = true,
1834          .ret_type       = RET_INTEGER,
1835          .arg1_type      = ARG_PTR_TO_CTX,
1836          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1837          .arg3_type      = ARG_CONST_SIZE,
1838 };
1839 
1840 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1841 	   void *, buf, u32, size, u64, flags)
1842 {
1843 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1844 	struct perf_branch_stack *br_stack = ctx->data->br_stack;
1845 	u32 to_copy;
1846 
1847 	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1848 		return -EINVAL;
1849 
1850 	if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1851 		return -ENOENT;
1852 
1853 	if (unlikely(!br_stack))
1854 		return -ENOENT;
1855 
1856 	if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1857 		return br_stack->nr * br_entry_size;
1858 
1859 	if (!buf || (size % br_entry_size != 0))
1860 		return -EINVAL;
1861 
1862 	to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1863 	memcpy(buf, br_stack->entries, to_copy);
1864 
1865 	return to_copy;
1866 }
1867 
1868 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1869 	.func           = bpf_read_branch_records,
1870 	.gpl_only       = true,
1871 	.ret_type       = RET_INTEGER,
1872 	.arg1_type      = ARG_PTR_TO_CTX,
1873 	.arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1874 	.arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1875 	.arg4_type      = ARG_ANYTHING,
1876 };
1877 
1878 static const struct bpf_func_proto *
1879 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1880 {
1881 	switch (func_id) {
1882 	case BPF_FUNC_perf_event_output:
1883 		return &bpf_perf_event_output_proto_tp;
1884 	case BPF_FUNC_get_stackid:
1885 		return &bpf_get_stackid_proto_pe;
1886 	case BPF_FUNC_get_stack:
1887 		return &bpf_get_stack_proto_pe;
1888 	case BPF_FUNC_perf_prog_read_value:
1889 		return &bpf_perf_prog_read_value_proto;
1890 	case BPF_FUNC_read_branch_records:
1891 		return &bpf_read_branch_records_proto;
1892 	case BPF_FUNC_get_attach_cookie:
1893 		return &bpf_get_attach_cookie_proto_pe;
1894 	default:
1895 		return bpf_tracing_func_proto(func_id, prog);
1896 	}
1897 }
1898 
1899 /*
1900  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1901  * to avoid potential recursive reuse issue when/if tracepoints are added
1902  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1903  *
1904  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1905  * in normal, irq, and nmi context.
1906  */
1907 struct bpf_raw_tp_regs {
1908 	struct pt_regs regs[3];
1909 };
1910 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1911 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1912 static struct pt_regs *get_bpf_raw_tp_regs(void)
1913 {
1914 	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1915 	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1916 
1917 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1918 		this_cpu_dec(bpf_raw_tp_nest_level);
1919 		return ERR_PTR(-EBUSY);
1920 	}
1921 
1922 	return &tp_regs->regs[nest_level - 1];
1923 }
1924 
1925 static void put_bpf_raw_tp_regs(void)
1926 {
1927 	this_cpu_dec(bpf_raw_tp_nest_level);
1928 }
1929 
1930 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1931 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1932 {
1933 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1934 	int ret;
1935 
1936 	if (IS_ERR(regs))
1937 		return PTR_ERR(regs);
1938 
1939 	perf_fetch_caller_regs(regs);
1940 	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1941 
1942 	put_bpf_raw_tp_regs();
1943 	return ret;
1944 }
1945 
1946 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1947 	.func		= bpf_perf_event_output_raw_tp,
1948 	.gpl_only	= true,
1949 	.ret_type	= RET_INTEGER,
1950 	.arg1_type	= ARG_PTR_TO_CTX,
1951 	.arg2_type	= ARG_CONST_MAP_PTR,
1952 	.arg3_type	= ARG_ANYTHING,
1953 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1954 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1955 };
1956 
1957 extern const struct bpf_func_proto bpf_skb_output_proto;
1958 extern const struct bpf_func_proto bpf_xdp_output_proto;
1959 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1960 
1961 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1962 	   struct bpf_map *, map, u64, flags)
1963 {
1964 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1965 	int ret;
1966 
1967 	if (IS_ERR(regs))
1968 		return PTR_ERR(regs);
1969 
1970 	perf_fetch_caller_regs(regs);
1971 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1972 	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1973 			      flags, 0, 0);
1974 	put_bpf_raw_tp_regs();
1975 	return ret;
1976 }
1977 
1978 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1979 	.func		= bpf_get_stackid_raw_tp,
1980 	.gpl_only	= true,
1981 	.ret_type	= RET_INTEGER,
1982 	.arg1_type	= ARG_PTR_TO_CTX,
1983 	.arg2_type	= ARG_CONST_MAP_PTR,
1984 	.arg3_type	= ARG_ANYTHING,
1985 };
1986 
1987 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1988 	   void *, buf, u32, size, u64, flags)
1989 {
1990 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1991 	int ret;
1992 
1993 	if (IS_ERR(regs))
1994 		return PTR_ERR(regs);
1995 
1996 	perf_fetch_caller_regs(regs);
1997 	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1998 			    (unsigned long) size, flags, 0);
1999 	put_bpf_raw_tp_regs();
2000 	return ret;
2001 }
2002 
2003 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
2004 	.func		= bpf_get_stack_raw_tp,
2005 	.gpl_only	= true,
2006 	.ret_type	= RET_INTEGER,
2007 	.arg1_type	= ARG_PTR_TO_CTX,
2008 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
2009 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
2010 	.arg4_type	= ARG_ANYTHING,
2011 };
2012 
2013 static const struct bpf_func_proto *
2014 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2015 {
2016 	switch (func_id) {
2017 	case BPF_FUNC_perf_event_output:
2018 		return &bpf_perf_event_output_proto_raw_tp;
2019 	case BPF_FUNC_get_stackid:
2020 		return &bpf_get_stackid_proto_raw_tp;
2021 	case BPF_FUNC_get_stack:
2022 		return &bpf_get_stack_proto_raw_tp;
2023 	case BPF_FUNC_get_attach_cookie:
2024 		return &bpf_get_attach_cookie_proto_tracing;
2025 	default:
2026 		return bpf_tracing_func_proto(func_id, prog);
2027 	}
2028 }
2029 
2030 const struct bpf_func_proto *
2031 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
2032 {
2033 	const struct bpf_func_proto *fn;
2034 
2035 	switch (func_id) {
2036 #ifdef CONFIG_NET
2037 	case BPF_FUNC_skb_output:
2038 		return &bpf_skb_output_proto;
2039 	case BPF_FUNC_xdp_output:
2040 		return &bpf_xdp_output_proto;
2041 	case BPF_FUNC_skc_to_tcp6_sock:
2042 		return &bpf_skc_to_tcp6_sock_proto;
2043 	case BPF_FUNC_skc_to_tcp_sock:
2044 		return &bpf_skc_to_tcp_sock_proto;
2045 	case BPF_FUNC_skc_to_tcp_timewait_sock:
2046 		return &bpf_skc_to_tcp_timewait_sock_proto;
2047 	case BPF_FUNC_skc_to_tcp_request_sock:
2048 		return &bpf_skc_to_tcp_request_sock_proto;
2049 	case BPF_FUNC_skc_to_udp6_sock:
2050 		return &bpf_skc_to_udp6_sock_proto;
2051 	case BPF_FUNC_skc_to_unix_sock:
2052 		return &bpf_skc_to_unix_sock_proto;
2053 	case BPF_FUNC_skc_to_mptcp_sock:
2054 		return &bpf_skc_to_mptcp_sock_proto;
2055 	case BPF_FUNC_sk_storage_get:
2056 		return &bpf_sk_storage_get_tracing_proto;
2057 	case BPF_FUNC_sk_storage_delete:
2058 		return &bpf_sk_storage_delete_tracing_proto;
2059 	case BPF_FUNC_sock_from_file:
2060 		return &bpf_sock_from_file_proto;
2061 	case BPF_FUNC_get_socket_cookie:
2062 		return &bpf_get_socket_ptr_cookie_proto;
2063 	case BPF_FUNC_xdp_get_buff_len:
2064 		return &bpf_xdp_get_buff_len_trace_proto;
2065 #endif
2066 	case BPF_FUNC_seq_printf:
2067 		return prog->expected_attach_type == BPF_TRACE_ITER ?
2068 		       &bpf_seq_printf_proto :
2069 		       NULL;
2070 	case BPF_FUNC_seq_write:
2071 		return prog->expected_attach_type == BPF_TRACE_ITER ?
2072 		       &bpf_seq_write_proto :
2073 		       NULL;
2074 	case BPF_FUNC_seq_printf_btf:
2075 		return prog->expected_attach_type == BPF_TRACE_ITER ?
2076 		       &bpf_seq_printf_btf_proto :
2077 		       NULL;
2078 	case BPF_FUNC_d_path:
2079 		return &bpf_d_path_proto;
2080 	case BPF_FUNC_get_func_arg:
2081 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
2082 	case BPF_FUNC_get_func_ret:
2083 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
2084 	case BPF_FUNC_get_func_arg_cnt:
2085 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
2086 	case BPF_FUNC_get_attach_cookie:
2087 		if (prog->type == BPF_PROG_TYPE_TRACING &&
2088 		    prog->expected_attach_type == BPF_TRACE_RAW_TP)
2089 			return &bpf_get_attach_cookie_proto_tracing;
2090 		return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
2091 	default:
2092 		fn = raw_tp_prog_func_proto(func_id, prog);
2093 		if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
2094 			fn = bpf_iter_get_func_proto(func_id, prog);
2095 		return fn;
2096 	}
2097 }
2098 
2099 static bool raw_tp_prog_is_valid_access(int off, int size,
2100 					enum bpf_access_type type,
2101 					const struct bpf_prog *prog,
2102 					struct bpf_insn_access_aux *info)
2103 {
2104 	return bpf_tracing_ctx_access(off, size, type);
2105 }
2106 
2107 static bool tracing_prog_is_valid_access(int off, int size,
2108 					 enum bpf_access_type type,
2109 					 const struct bpf_prog *prog,
2110 					 struct bpf_insn_access_aux *info)
2111 {
2112 	return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
2113 }
2114 
2115 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
2116 				     const union bpf_attr *kattr,
2117 				     union bpf_attr __user *uattr)
2118 {
2119 	return -ENOTSUPP;
2120 }
2121 
2122 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
2123 	.get_func_proto  = raw_tp_prog_func_proto,
2124 	.is_valid_access = raw_tp_prog_is_valid_access,
2125 };
2126 
2127 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
2128 #ifdef CONFIG_NET
2129 	.test_run = bpf_prog_test_run_raw_tp,
2130 #endif
2131 };
2132 
2133 const struct bpf_verifier_ops tracing_verifier_ops = {
2134 	.get_func_proto  = tracing_prog_func_proto,
2135 	.is_valid_access = tracing_prog_is_valid_access,
2136 };
2137 
2138 const struct bpf_prog_ops tracing_prog_ops = {
2139 	.test_run = bpf_prog_test_run_tracing,
2140 };
2141 
2142 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2143 						 enum bpf_access_type type,
2144 						 const struct bpf_prog *prog,
2145 						 struct bpf_insn_access_aux *info)
2146 {
2147 	if (off == 0) {
2148 		if (size != sizeof(u64) || type != BPF_READ)
2149 			return false;
2150 		info->reg_type = PTR_TO_TP_BUFFER;
2151 	}
2152 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2153 }
2154 
2155 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2156 	.get_func_proto  = raw_tp_prog_func_proto,
2157 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
2158 };
2159 
2160 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2161 };
2162 
2163 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2164 				    const struct bpf_prog *prog,
2165 				    struct bpf_insn_access_aux *info)
2166 {
2167 	const int size_u64 = sizeof(u64);
2168 
2169 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2170 		return false;
2171 	if (type != BPF_READ)
2172 		return false;
2173 	if (off % size != 0) {
2174 		if (sizeof(unsigned long) != 4)
2175 			return false;
2176 		if (size != 8)
2177 			return false;
2178 		if (off % size != 4)
2179 			return false;
2180 	}
2181 
2182 	switch (off) {
2183 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2184 		bpf_ctx_record_field_size(info, size_u64);
2185 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2186 			return false;
2187 		break;
2188 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
2189 		bpf_ctx_record_field_size(info, size_u64);
2190 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2191 			return false;
2192 		break;
2193 	default:
2194 		if (size != sizeof(long))
2195 			return false;
2196 	}
2197 
2198 	return true;
2199 }
2200 
2201 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2202 				      const struct bpf_insn *si,
2203 				      struct bpf_insn *insn_buf,
2204 				      struct bpf_prog *prog, u32 *target_size)
2205 {
2206 	struct bpf_insn *insn = insn_buf;
2207 
2208 	switch (si->off) {
2209 	case offsetof(struct bpf_perf_event_data, sample_period):
2210 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2211 						       data), si->dst_reg, si->src_reg,
2212 				      offsetof(struct bpf_perf_event_data_kern, data));
2213 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2214 				      bpf_target_off(struct perf_sample_data, period, 8,
2215 						     target_size));
2216 		break;
2217 	case offsetof(struct bpf_perf_event_data, addr):
2218 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2219 						       data), si->dst_reg, si->src_reg,
2220 				      offsetof(struct bpf_perf_event_data_kern, data));
2221 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2222 				      bpf_target_off(struct perf_sample_data, addr, 8,
2223 						     target_size));
2224 		break;
2225 	default:
2226 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2227 						       regs), si->dst_reg, si->src_reg,
2228 				      offsetof(struct bpf_perf_event_data_kern, regs));
2229 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2230 				      si->off);
2231 		break;
2232 	}
2233 
2234 	return insn - insn_buf;
2235 }
2236 
2237 const struct bpf_verifier_ops perf_event_verifier_ops = {
2238 	.get_func_proto		= pe_prog_func_proto,
2239 	.is_valid_access	= pe_prog_is_valid_access,
2240 	.convert_ctx_access	= pe_prog_convert_ctx_access,
2241 };
2242 
2243 const struct bpf_prog_ops perf_event_prog_ops = {
2244 };
2245 
2246 static DEFINE_MUTEX(bpf_event_mutex);
2247 
2248 #define BPF_TRACE_MAX_PROGS 64
2249 
2250 int perf_event_attach_bpf_prog(struct perf_event *event,
2251 			       struct bpf_prog *prog,
2252 			       u64 bpf_cookie)
2253 {
2254 	struct bpf_prog_array *old_array;
2255 	struct bpf_prog_array *new_array;
2256 	int ret = -EEXIST;
2257 
2258 	/*
2259 	 * Kprobe override only works if they are on the function entry,
2260 	 * and only if they are on the opt-in list.
2261 	 */
2262 	if (prog->kprobe_override &&
2263 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
2264 	     !trace_kprobe_error_injectable(event->tp_event)))
2265 		return -EINVAL;
2266 
2267 	mutex_lock(&bpf_event_mutex);
2268 
2269 	if (event->prog)
2270 		goto unlock;
2271 
2272 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2273 	if (old_array &&
2274 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2275 		ret = -E2BIG;
2276 		goto unlock;
2277 	}
2278 
2279 	ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2280 	if (ret < 0)
2281 		goto unlock;
2282 
2283 	/* set the new array to event->tp_event and set event->prog */
2284 	event->prog = prog;
2285 	event->bpf_cookie = bpf_cookie;
2286 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
2287 	bpf_prog_array_free_sleepable(old_array);
2288 
2289 unlock:
2290 	mutex_unlock(&bpf_event_mutex);
2291 	return ret;
2292 }
2293 
2294 void perf_event_detach_bpf_prog(struct perf_event *event)
2295 {
2296 	struct bpf_prog_array *old_array;
2297 	struct bpf_prog_array *new_array;
2298 	int ret;
2299 
2300 	mutex_lock(&bpf_event_mutex);
2301 
2302 	if (!event->prog)
2303 		goto unlock;
2304 
2305 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2306 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2307 	if (ret == -ENOENT)
2308 		goto unlock;
2309 	if (ret < 0) {
2310 		bpf_prog_array_delete_safe(old_array, event->prog);
2311 	} else {
2312 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
2313 		bpf_prog_array_free_sleepable(old_array);
2314 	}
2315 
2316 	bpf_prog_put(event->prog);
2317 	event->prog = NULL;
2318 
2319 unlock:
2320 	mutex_unlock(&bpf_event_mutex);
2321 }
2322 
2323 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2324 {
2325 	struct perf_event_query_bpf __user *uquery = info;
2326 	struct perf_event_query_bpf query = {};
2327 	struct bpf_prog_array *progs;
2328 	u32 *ids, prog_cnt, ids_len;
2329 	int ret;
2330 
2331 	if (!perfmon_capable())
2332 		return -EPERM;
2333 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
2334 		return -EINVAL;
2335 	if (copy_from_user(&query, uquery, sizeof(query)))
2336 		return -EFAULT;
2337 
2338 	ids_len = query.ids_len;
2339 	if (ids_len > BPF_TRACE_MAX_PROGS)
2340 		return -E2BIG;
2341 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2342 	if (!ids)
2343 		return -ENOMEM;
2344 	/*
2345 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2346 	 * is required when user only wants to check for uquery->prog_cnt.
2347 	 * There is no need to check for it since the case is handled
2348 	 * gracefully in bpf_prog_array_copy_info.
2349 	 */
2350 
2351 	mutex_lock(&bpf_event_mutex);
2352 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2353 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2354 	mutex_unlock(&bpf_event_mutex);
2355 
2356 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2357 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2358 		ret = -EFAULT;
2359 
2360 	kfree(ids);
2361 	return ret;
2362 }
2363 
2364 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2365 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2366 
2367 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2368 {
2369 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2370 
2371 	for (; btp < __stop__bpf_raw_tp; btp++) {
2372 		if (!strcmp(btp->tp->name, name))
2373 			return btp;
2374 	}
2375 
2376 	return bpf_get_raw_tracepoint_module(name);
2377 }
2378 
2379 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2380 {
2381 	struct module *mod;
2382 
2383 	preempt_disable();
2384 	mod = __module_address((unsigned long)btp);
2385 	module_put(mod);
2386 	preempt_enable();
2387 }
2388 
2389 static __always_inline
2390 void __bpf_trace_run(struct bpf_raw_tp_link *link, u64 *args)
2391 {
2392 	struct bpf_prog *prog = link->link.prog;
2393 	struct bpf_run_ctx *old_run_ctx;
2394 	struct bpf_trace_run_ctx run_ctx;
2395 
2396 	cant_sleep();
2397 	if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2398 		bpf_prog_inc_misses_counter(prog);
2399 		goto out;
2400 	}
2401 
2402 	run_ctx.bpf_cookie = link->cookie;
2403 	old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2404 
2405 	rcu_read_lock();
2406 	(void) bpf_prog_run(prog, args);
2407 	rcu_read_unlock();
2408 
2409 	bpf_reset_run_ctx(old_run_ctx);
2410 out:
2411 	this_cpu_dec(*(prog->active));
2412 }
2413 
2414 #define UNPACK(...)			__VA_ARGS__
2415 #define REPEAT_1(FN, DL, X, ...)	FN(X)
2416 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2417 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2418 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2419 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2420 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2421 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2422 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2423 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2424 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2425 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2426 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2427 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
2428 
2429 #define SARG(X)		u64 arg##X
2430 #define COPY(X)		args[X] = arg##X
2431 
2432 #define __DL_COM	(,)
2433 #define __DL_SEM	(;)
2434 
2435 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2436 
2437 #define BPF_TRACE_DEFN_x(x)						\
2438 	void bpf_trace_run##x(struct bpf_raw_tp_link *link,		\
2439 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
2440 	{								\
2441 		u64 args[x];						\
2442 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
2443 		__bpf_trace_run(link, args);				\
2444 	}								\
2445 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2446 BPF_TRACE_DEFN_x(1);
2447 BPF_TRACE_DEFN_x(2);
2448 BPF_TRACE_DEFN_x(3);
2449 BPF_TRACE_DEFN_x(4);
2450 BPF_TRACE_DEFN_x(5);
2451 BPF_TRACE_DEFN_x(6);
2452 BPF_TRACE_DEFN_x(7);
2453 BPF_TRACE_DEFN_x(8);
2454 BPF_TRACE_DEFN_x(9);
2455 BPF_TRACE_DEFN_x(10);
2456 BPF_TRACE_DEFN_x(11);
2457 BPF_TRACE_DEFN_x(12);
2458 
2459 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2460 {
2461 	struct tracepoint *tp = btp->tp;
2462 	struct bpf_prog *prog = link->link.prog;
2463 
2464 	/*
2465 	 * check that program doesn't access arguments beyond what's
2466 	 * available in this tracepoint
2467 	 */
2468 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2469 		return -EINVAL;
2470 
2471 	if (prog->aux->max_tp_access > btp->writable_size)
2472 		return -EINVAL;
2473 
2474 	return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func, link);
2475 }
2476 
2477 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2478 {
2479 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, link);
2480 }
2481 
2482 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2483 			    u32 *fd_type, const char **buf,
2484 			    u64 *probe_offset, u64 *probe_addr,
2485 			    unsigned long *missed)
2486 {
2487 	bool is_tracepoint, is_syscall_tp;
2488 	struct bpf_prog *prog;
2489 	int flags, err = 0;
2490 
2491 	prog = event->prog;
2492 	if (!prog)
2493 		return -ENOENT;
2494 
2495 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2496 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2497 		return -EOPNOTSUPP;
2498 
2499 	*prog_id = prog->aux->id;
2500 	flags = event->tp_event->flags;
2501 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2502 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
2503 
2504 	if (is_tracepoint || is_syscall_tp) {
2505 		*buf = is_tracepoint ? event->tp_event->tp->name
2506 				     : event->tp_event->name;
2507 		/* We allow NULL pointer for tracepoint */
2508 		if (fd_type)
2509 			*fd_type = BPF_FD_TYPE_TRACEPOINT;
2510 		if (probe_offset)
2511 			*probe_offset = 0x0;
2512 		if (probe_addr)
2513 			*probe_addr = 0x0;
2514 	} else {
2515 		/* kprobe/uprobe */
2516 		err = -EOPNOTSUPP;
2517 #ifdef CONFIG_KPROBE_EVENTS
2518 		if (flags & TRACE_EVENT_FL_KPROBE)
2519 			err = bpf_get_kprobe_info(event, fd_type, buf,
2520 						  probe_offset, probe_addr, missed,
2521 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2522 #endif
2523 #ifdef CONFIG_UPROBE_EVENTS
2524 		if (flags & TRACE_EVENT_FL_UPROBE)
2525 			err = bpf_get_uprobe_info(event, fd_type, buf,
2526 						  probe_offset, probe_addr,
2527 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2528 #endif
2529 	}
2530 
2531 	return err;
2532 }
2533 
2534 static int __init send_signal_irq_work_init(void)
2535 {
2536 	int cpu;
2537 	struct send_signal_irq_work *work;
2538 
2539 	for_each_possible_cpu(cpu) {
2540 		work = per_cpu_ptr(&send_signal_work, cpu);
2541 		init_irq_work(&work->irq_work, do_bpf_send_signal);
2542 	}
2543 	return 0;
2544 }
2545 
2546 subsys_initcall(send_signal_irq_work_init);
2547 
2548 #ifdef CONFIG_MODULES
2549 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2550 			    void *module)
2551 {
2552 	struct bpf_trace_module *btm, *tmp;
2553 	struct module *mod = module;
2554 	int ret = 0;
2555 
2556 	if (mod->num_bpf_raw_events == 0 ||
2557 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2558 		goto out;
2559 
2560 	mutex_lock(&bpf_module_mutex);
2561 
2562 	switch (op) {
2563 	case MODULE_STATE_COMING:
2564 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2565 		if (btm) {
2566 			btm->module = module;
2567 			list_add(&btm->list, &bpf_trace_modules);
2568 		} else {
2569 			ret = -ENOMEM;
2570 		}
2571 		break;
2572 	case MODULE_STATE_GOING:
2573 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2574 			if (btm->module == module) {
2575 				list_del(&btm->list);
2576 				kfree(btm);
2577 				break;
2578 			}
2579 		}
2580 		break;
2581 	}
2582 
2583 	mutex_unlock(&bpf_module_mutex);
2584 
2585 out:
2586 	return notifier_from_errno(ret);
2587 }
2588 
2589 static struct notifier_block bpf_module_nb = {
2590 	.notifier_call = bpf_event_notify,
2591 };
2592 
2593 static int __init bpf_event_init(void)
2594 {
2595 	register_module_notifier(&bpf_module_nb);
2596 	return 0;
2597 }
2598 
2599 fs_initcall(bpf_event_init);
2600 #endif /* CONFIG_MODULES */
2601 
2602 struct bpf_session_run_ctx {
2603 	struct bpf_run_ctx run_ctx;
2604 	bool is_return;
2605 	void *data;
2606 };
2607 
2608 #ifdef CONFIG_FPROBE
2609 struct bpf_kprobe_multi_link {
2610 	struct bpf_link link;
2611 	struct fprobe fp;
2612 	unsigned long *addrs;
2613 	u64 *cookies;
2614 	u32 cnt;
2615 	u32 mods_cnt;
2616 	struct module **mods;
2617 	u32 flags;
2618 };
2619 
2620 struct bpf_kprobe_multi_run_ctx {
2621 	struct bpf_session_run_ctx session_ctx;
2622 	struct bpf_kprobe_multi_link *link;
2623 	unsigned long entry_ip;
2624 };
2625 
2626 struct user_syms {
2627 	const char **syms;
2628 	char *buf;
2629 };
2630 
2631 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2632 {
2633 	unsigned long __user usymbol;
2634 	const char **syms = NULL;
2635 	char *buf = NULL, *p;
2636 	int err = -ENOMEM;
2637 	unsigned int i;
2638 
2639 	syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2640 	if (!syms)
2641 		goto error;
2642 
2643 	buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2644 	if (!buf)
2645 		goto error;
2646 
2647 	for (p = buf, i = 0; i < cnt; i++) {
2648 		if (__get_user(usymbol, usyms + i)) {
2649 			err = -EFAULT;
2650 			goto error;
2651 		}
2652 		err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2653 		if (err == KSYM_NAME_LEN)
2654 			err = -E2BIG;
2655 		if (err < 0)
2656 			goto error;
2657 		syms[i] = p;
2658 		p += err + 1;
2659 	}
2660 
2661 	us->syms = syms;
2662 	us->buf = buf;
2663 	return 0;
2664 
2665 error:
2666 	if (err) {
2667 		kvfree(syms);
2668 		kvfree(buf);
2669 	}
2670 	return err;
2671 }
2672 
2673 static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2674 {
2675 	u32 i;
2676 
2677 	for (i = 0; i < cnt; i++)
2678 		module_put(mods[i]);
2679 }
2680 
2681 static void free_user_syms(struct user_syms *us)
2682 {
2683 	kvfree(us->syms);
2684 	kvfree(us->buf);
2685 }
2686 
2687 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2688 {
2689 	struct bpf_kprobe_multi_link *kmulti_link;
2690 
2691 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2692 	unregister_fprobe(&kmulti_link->fp);
2693 	kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2694 }
2695 
2696 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2697 {
2698 	struct bpf_kprobe_multi_link *kmulti_link;
2699 
2700 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2701 	kvfree(kmulti_link->addrs);
2702 	kvfree(kmulti_link->cookies);
2703 	kfree(kmulti_link->mods);
2704 	kfree(kmulti_link);
2705 }
2706 
2707 static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
2708 						struct bpf_link_info *info)
2709 {
2710 	u64 __user *ucookies = u64_to_user_ptr(info->kprobe_multi.cookies);
2711 	u64 __user *uaddrs = u64_to_user_ptr(info->kprobe_multi.addrs);
2712 	struct bpf_kprobe_multi_link *kmulti_link;
2713 	u32 ucount = info->kprobe_multi.count;
2714 	int err = 0, i;
2715 
2716 	if (!uaddrs ^ !ucount)
2717 		return -EINVAL;
2718 	if (ucookies && !ucount)
2719 		return -EINVAL;
2720 
2721 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2722 	info->kprobe_multi.count = kmulti_link->cnt;
2723 	info->kprobe_multi.flags = kmulti_link->flags;
2724 	info->kprobe_multi.missed = kmulti_link->fp.nmissed;
2725 
2726 	if (!uaddrs)
2727 		return 0;
2728 	if (ucount < kmulti_link->cnt)
2729 		err = -ENOSPC;
2730 	else
2731 		ucount = kmulti_link->cnt;
2732 
2733 	if (ucookies) {
2734 		if (kmulti_link->cookies) {
2735 			if (copy_to_user(ucookies, kmulti_link->cookies, ucount * sizeof(u64)))
2736 				return -EFAULT;
2737 		} else {
2738 			for (i = 0; i < ucount; i++) {
2739 				if (put_user(0, ucookies + i))
2740 					return -EFAULT;
2741 			}
2742 		}
2743 	}
2744 
2745 	if (kallsyms_show_value(current_cred())) {
2746 		if (copy_to_user(uaddrs, kmulti_link->addrs, ucount * sizeof(u64)))
2747 			return -EFAULT;
2748 	} else {
2749 		for (i = 0; i < ucount; i++) {
2750 			if (put_user(0, uaddrs + i))
2751 				return -EFAULT;
2752 		}
2753 	}
2754 	return err;
2755 }
2756 
2757 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2758 	.release = bpf_kprobe_multi_link_release,
2759 	.dealloc_deferred = bpf_kprobe_multi_link_dealloc,
2760 	.fill_link_info = bpf_kprobe_multi_link_fill_link_info,
2761 };
2762 
2763 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2764 {
2765 	const struct bpf_kprobe_multi_link *link = priv;
2766 	unsigned long *addr_a = a, *addr_b = b;
2767 	u64 *cookie_a, *cookie_b;
2768 
2769 	cookie_a = link->cookies + (addr_a - link->addrs);
2770 	cookie_b = link->cookies + (addr_b - link->addrs);
2771 
2772 	/* swap addr_a/addr_b and cookie_a/cookie_b values */
2773 	swap(*addr_a, *addr_b);
2774 	swap(*cookie_a, *cookie_b);
2775 }
2776 
2777 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2778 {
2779 	const unsigned long *addr_a = a, *addr_b = b;
2780 
2781 	if (*addr_a == *addr_b)
2782 		return 0;
2783 	return *addr_a < *addr_b ? -1 : 1;
2784 }
2785 
2786 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2787 {
2788 	return bpf_kprobe_multi_addrs_cmp(a, b);
2789 }
2790 
2791 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2792 {
2793 	struct bpf_kprobe_multi_run_ctx *run_ctx;
2794 	struct bpf_kprobe_multi_link *link;
2795 	u64 *cookie, entry_ip;
2796 	unsigned long *addr;
2797 
2798 	if (WARN_ON_ONCE(!ctx))
2799 		return 0;
2800 	run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2801 			       session_ctx.run_ctx);
2802 	link = run_ctx->link;
2803 	if (!link->cookies)
2804 		return 0;
2805 	entry_ip = run_ctx->entry_ip;
2806 	addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2807 		       bpf_kprobe_multi_addrs_cmp);
2808 	if (!addr)
2809 		return 0;
2810 	cookie = link->cookies + (addr - link->addrs);
2811 	return *cookie;
2812 }
2813 
2814 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2815 {
2816 	struct bpf_kprobe_multi_run_ctx *run_ctx;
2817 
2818 	run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2819 			       session_ctx.run_ctx);
2820 	return run_ctx->entry_ip;
2821 }
2822 
2823 static int
2824 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2825 			   unsigned long entry_ip, struct pt_regs *regs,
2826 			   bool is_return, void *data)
2827 {
2828 	struct bpf_kprobe_multi_run_ctx run_ctx = {
2829 		.session_ctx = {
2830 			.is_return = is_return,
2831 			.data = data,
2832 		},
2833 		.link = link,
2834 		.entry_ip = entry_ip,
2835 	};
2836 	struct bpf_run_ctx *old_run_ctx;
2837 	int err;
2838 
2839 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2840 		bpf_prog_inc_misses_counter(link->link.prog);
2841 		err = 0;
2842 		goto out;
2843 	}
2844 
2845 	migrate_disable();
2846 	rcu_read_lock();
2847 	old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
2848 	err = bpf_prog_run(link->link.prog, regs);
2849 	bpf_reset_run_ctx(old_run_ctx);
2850 	rcu_read_unlock();
2851 	migrate_enable();
2852 
2853  out:
2854 	__this_cpu_dec(bpf_prog_active);
2855 	return err;
2856 }
2857 
2858 static int
2859 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2860 			  unsigned long ret_ip, struct pt_regs *regs,
2861 			  void *data)
2862 {
2863 	struct bpf_kprobe_multi_link *link;
2864 	int err;
2865 
2866 	link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2867 	err = kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs, false, data);
2868 	return is_kprobe_session(link->link.prog) ? err : 0;
2869 }
2870 
2871 static void
2872 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip,
2873 			       unsigned long ret_ip, struct pt_regs *regs,
2874 			       void *data)
2875 {
2876 	struct bpf_kprobe_multi_link *link;
2877 
2878 	link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2879 	kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs, true, data);
2880 }
2881 
2882 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2883 {
2884 	const char **str_a = (const char **) a;
2885 	const char **str_b = (const char **) b;
2886 
2887 	return strcmp(*str_a, *str_b);
2888 }
2889 
2890 struct multi_symbols_sort {
2891 	const char **funcs;
2892 	u64 *cookies;
2893 };
2894 
2895 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2896 {
2897 	const struct multi_symbols_sort *data = priv;
2898 	const char **name_a = a, **name_b = b;
2899 
2900 	swap(*name_a, *name_b);
2901 
2902 	/* If defined, swap also related cookies. */
2903 	if (data->cookies) {
2904 		u64 *cookie_a, *cookie_b;
2905 
2906 		cookie_a = data->cookies + (name_a - data->funcs);
2907 		cookie_b = data->cookies + (name_b - data->funcs);
2908 		swap(*cookie_a, *cookie_b);
2909 	}
2910 }
2911 
2912 struct modules_array {
2913 	struct module **mods;
2914 	int mods_cnt;
2915 	int mods_cap;
2916 };
2917 
2918 static int add_module(struct modules_array *arr, struct module *mod)
2919 {
2920 	struct module **mods;
2921 
2922 	if (arr->mods_cnt == arr->mods_cap) {
2923 		arr->mods_cap = max(16, arr->mods_cap * 3 / 2);
2924 		mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL);
2925 		if (!mods)
2926 			return -ENOMEM;
2927 		arr->mods = mods;
2928 	}
2929 
2930 	arr->mods[arr->mods_cnt] = mod;
2931 	arr->mods_cnt++;
2932 	return 0;
2933 }
2934 
2935 static bool has_module(struct modules_array *arr, struct module *mod)
2936 {
2937 	int i;
2938 
2939 	for (i = arr->mods_cnt - 1; i >= 0; i--) {
2940 		if (arr->mods[i] == mod)
2941 			return true;
2942 	}
2943 	return false;
2944 }
2945 
2946 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2947 {
2948 	struct modules_array arr = {};
2949 	u32 i, err = 0;
2950 
2951 	for (i = 0; i < addrs_cnt; i++) {
2952 		struct module *mod;
2953 
2954 		preempt_disable();
2955 		mod = __module_address(addrs[i]);
2956 		/* Either no module or we it's already stored  */
2957 		if (!mod || has_module(&arr, mod)) {
2958 			preempt_enable();
2959 			continue;
2960 		}
2961 		if (!try_module_get(mod))
2962 			err = -EINVAL;
2963 		preempt_enable();
2964 		if (err)
2965 			break;
2966 		err = add_module(&arr, mod);
2967 		if (err) {
2968 			module_put(mod);
2969 			break;
2970 		}
2971 	}
2972 
2973 	/* We return either err < 0 in case of error, ... */
2974 	if (err) {
2975 		kprobe_multi_put_modules(arr.mods, arr.mods_cnt);
2976 		kfree(arr.mods);
2977 		return err;
2978 	}
2979 
2980 	/* or number of modules found if everything is ok. */
2981 	*mods = arr.mods;
2982 	return arr.mods_cnt;
2983 }
2984 
2985 static int addrs_check_error_injection_list(unsigned long *addrs, u32 cnt)
2986 {
2987 	u32 i;
2988 
2989 	for (i = 0; i < cnt; i++) {
2990 		if (!within_error_injection_list(addrs[i]))
2991 			return -EINVAL;
2992 	}
2993 	return 0;
2994 }
2995 
2996 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2997 {
2998 	struct bpf_kprobe_multi_link *link = NULL;
2999 	struct bpf_link_primer link_primer;
3000 	void __user *ucookies;
3001 	unsigned long *addrs;
3002 	u32 flags, cnt, size;
3003 	void __user *uaddrs;
3004 	u64 *cookies = NULL;
3005 	void __user *usyms;
3006 	int err;
3007 
3008 	/* no support for 32bit archs yet */
3009 	if (sizeof(u64) != sizeof(void *))
3010 		return -EOPNOTSUPP;
3011 
3012 	if (!is_kprobe_multi(prog))
3013 		return -EINVAL;
3014 
3015 	flags = attr->link_create.kprobe_multi.flags;
3016 	if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
3017 		return -EINVAL;
3018 
3019 	uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
3020 	usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
3021 	if (!!uaddrs == !!usyms)
3022 		return -EINVAL;
3023 
3024 	cnt = attr->link_create.kprobe_multi.cnt;
3025 	if (!cnt)
3026 		return -EINVAL;
3027 	if (cnt > MAX_KPROBE_MULTI_CNT)
3028 		return -E2BIG;
3029 
3030 	size = cnt * sizeof(*addrs);
3031 	addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
3032 	if (!addrs)
3033 		return -ENOMEM;
3034 
3035 	ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
3036 	if (ucookies) {
3037 		cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
3038 		if (!cookies) {
3039 			err = -ENOMEM;
3040 			goto error;
3041 		}
3042 		if (copy_from_user(cookies, ucookies, size)) {
3043 			err = -EFAULT;
3044 			goto error;
3045 		}
3046 	}
3047 
3048 	if (uaddrs) {
3049 		if (copy_from_user(addrs, uaddrs, size)) {
3050 			err = -EFAULT;
3051 			goto error;
3052 		}
3053 	} else {
3054 		struct multi_symbols_sort data = {
3055 			.cookies = cookies,
3056 		};
3057 		struct user_syms us;
3058 
3059 		err = copy_user_syms(&us, usyms, cnt);
3060 		if (err)
3061 			goto error;
3062 
3063 		if (cookies)
3064 			data.funcs = us.syms;
3065 
3066 		sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
3067 		       symbols_swap_r, &data);
3068 
3069 		err = ftrace_lookup_symbols(us.syms, cnt, addrs);
3070 		free_user_syms(&us);
3071 		if (err)
3072 			goto error;
3073 	}
3074 
3075 	if (prog->kprobe_override && addrs_check_error_injection_list(addrs, cnt)) {
3076 		err = -EINVAL;
3077 		goto error;
3078 	}
3079 
3080 	link = kzalloc(sizeof(*link), GFP_KERNEL);
3081 	if (!link) {
3082 		err = -ENOMEM;
3083 		goto error;
3084 	}
3085 
3086 	bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
3087 		      &bpf_kprobe_multi_link_lops, prog);
3088 
3089 	err = bpf_link_prime(&link->link, &link_primer);
3090 	if (err)
3091 		goto error;
3092 
3093 	if (!(flags & BPF_F_KPROBE_MULTI_RETURN))
3094 		link->fp.entry_handler = kprobe_multi_link_handler;
3095 	if ((flags & BPF_F_KPROBE_MULTI_RETURN) || is_kprobe_session(prog))
3096 		link->fp.exit_handler = kprobe_multi_link_exit_handler;
3097 	if (is_kprobe_session(prog))
3098 		link->fp.entry_data_size = sizeof(u64);
3099 
3100 	link->addrs = addrs;
3101 	link->cookies = cookies;
3102 	link->cnt = cnt;
3103 	link->flags = flags;
3104 
3105 	if (cookies) {
3106 		/*
3107 		 * Sorting addresses will trigger sorting cookies as well
3108 		 * (check bpf_kprobe_multi_cookie_swap). This way we can
3109 		 * find cookie based on the address in bpf_get_attach_cookie
3110 		 * helper.
3111 		 */
3112 		sort_r(addrs, cnt, sizeof(*addrs),
3113 		       bpf_kprobe_multi_cookie_cmp,
3114 		       bpf_kprobe_multi_cookie_swap,
3115 		       link);
3116 	}
3117 
3118 	err = get_modules_for_addrs(&link->mods, addrs, cnt);
3119 	if (err < 0) {
3120 		bpf_link_cleanup(&link_primer);
3121 		return err;
3122 	}
3123 	link->mods_cnt = err;
3124 
3125 	err = register_fprobe_ips(&link->fp, addrs, cnt);
3126 	if (err) {
3127 		kprobe_multi_put_modules(link->mods, link->mods_cnt);
3128 		bpf_link_cleanup(&link_primer);
3129 		return err;
3130 	}
3131 
3132 	return bpf_link_settle(&link_primer);
3133 
3134 error:
3135 	kfree(link);
3136 	kvfree(addrs);
3137 	kvfree(cookies);
3138 	return err;
3139 }
3140 #else /* !CONFIG_FPROBE */
3141 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3142 {
3143 	return -EOPNOTSUPP;
3144 }
3145 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
3146 {
3147 	return 0;
3148 }
3149 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3150 {
3151 	return 0;
3152 }
3153 #endif
3154 
3155 #ifdef CONFIG_UPROBES
3156 struct bpf_uprobe_multi_link;
3157 
3158 struct bpf_uprobe {
3159 	struct bpf_uprobe_multi_link *link;
3160 	loff_t offset;
3161 	unsigned long ref_ctr_offset;
3162 	u64 cookie;
3163 	struct uprobe_consumer consumer;
3164 };
3165 
3166 struct bpf_uprobe_multi_link {
3167 	struct path path;
3168 	struct bpf_link link;
3169 	u32 cnt;
3170 	u32 flags;
3171 	struct bpf_uprobe *uprobes;
3172 	struct task_struct *task;
3173 };
3174 
3175 struct bpf_uprobe_multi_run_ctx {
3176 	struct bpf_run_ctx run_ctx;
3177 	unsigned long entry_ip;
3178 	struct bpf_uprobe *uprobe;
3179 };
3180 
3181 static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes,
3182 				  u32 cnt)
3183 {
3184 	u32 i;
3185 
3186 	for (i = 0; i < cnt; i++) {
3187 		uprobe_unregister(d_real_inode(path->dentry), uprobes[i].offset,
3188 				  &uprobes[i].consumer);
3189 	}
3190 }
3191 
3192 static void bpf_uprobe_multi_link_release(struct bpf_link *link)
3193 {
3194 	struct bpf_uprobe_multi_link *umulti_link;
3195 
3196 	umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3197 	bpf_uprobe_unregister(&umulti_link->path, umulti_link->uprobes, umulti_link->cnt);
3198 	if (umulti_link->task)
3199 		put_task_struct(umulti_link->task);
3200 	path_put(&umulti_link->path);
3201 }
3202 
3203 static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)
3204 {
3205 	struct bpf_uprobe_multi_link *umulti_link;
3206 
3207 	umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3208 	kvfree(umulti_link->uprobes);
3209 	kfree(umulti_link);
3210 }
3211 
3212 static int bpf_uprobe_multi_link_fill_link_info(const struct bpf_link *link,
3213 						struct bpf_link_info *info)
3214 {
3215 	u64 __user *uref_ctr_offsets = u64_to_user_ptr(info->uprobe_multi.ref_ctr_offsets);
3216 	u64 __user *ucookies = u64_to_user_ptr(info->uprobe_multi.cookies);
3217 	u64 __user *uoffsets = u64_to_user_ptr(info->uprobe_multi.offsets);
3218 	u64 __user *upath = u64_to_user_ptr(info->uprobe_multi.path);
3219 	u32 upath_size = info->uprobe_multi.path_size;
3220 	struct bpf_uprobe_multi_link *umulti_link;
3221 	u32 ucount = info->uprobe_multi.count;
3222 	int err = 0, i;
3223 	long left;
3224 
3225 	if (!upath ^ !upath_size)
3226 		return -EINVAL;
3227 
3228 	if ((uoffsets || uref_ctr_offsets || ucookies) && !ucount)
3229 		return -EINVAL;
3230 
3231 	umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3232 	info->uprobe_multi.count = umulti_link->cnt;
3233 	info->uprobe_multi.flags = umulti_link->flags;
3234 	info->uprobe_multi.pid = umulti_link->task ?
3235 				 task_pid_nr_ns(umulti_link->task, task_active_pid_ns(current)) : 0;
3236 
3237 	if (upath) {
3238 		char *p, *buf;
3239 
3240 		upath_size = min_t(u32, upath_size, PATH_MAX);
3241 
3242 		buf = kmalloc(upath_size, GFP_KERNEL);
3243 		if (!buf)
3244 			return -ENOMEM;
3245 		p = d_path(&umulti_link->path, buf, upath_size);
3246 		if (IS_ERR(p)) {
3247 			kfree(buf);
3248 			return PTR_ERR(p);
3249 		}
3250 		upath_size = buf + upath_size - p;
3251 		left = copy_to_user(upath, p, upath_size);
3252 		kfree(buf);
3253 		if (left)
3254 			return -EFAULT;
3255 		info->uprobe_multi.path_size = upath_size;
3256 	}
3257 
3258 	if (!uoffsets && !ucookies && !uref_ctr_offsets)
3259 		return 0;
3260 
3261 	if (ucount < umulti_link->cnt)
3262 		err = -ENOSPC;
3263 	else
3264 		ucount = umulti_link->cnt;
3265 
3266 	for (i = 0; i < ucount; i++) {
3267 		if (uoffsets &&
3268 		    put_user(umulti_link->uprobes[i].offset, uoffsets + i))
3269 			return -EFAULT;
3270 		if (uref_ctr_offsets &&
3271 		    put_user(umulti_link->uprobes[i].ref_ctr_offset, uref_ctr_offsets + i))
3272 			return -EFAULT;
3273 		if (ucookies &&
3274 		    put_user(umulti_link->uprobes[i].cookie, ucookies + i))
3275 			return -EFAULT;
3276 	}
3277 
3278 	return err;
3279 }
3280 
3281 static const struct bpf_link_ops bpf_uprobe_multi_link_lops = {
3282 	.release = bpf_uprobe_multi_link_release,
3283 	.dealloc_deferred = bpf_uprobe_multi_link_dealloc,
3284 	.fill_link_info = bpf_uprobe_multi_link_fill_link_info,
3285 };
3286 
3287 static int uprobe_prog_run(struct bpf_uprobe *uprobe,
3288 			   unsigned long entry_ip,
3289 			   struct pt_regs *regs)
3290 {
3291 	struct bpf_uprobe_multi_link *link = uprobe->link;
3292 	struct bpf_uprobe_multi_run_ctx run_ctx = {
3293 		.entry_ip = entry_ip,
3294 		.uprobe = uprobe,
3295 	};
3296 	struct bpf_prog *prog = link->link.prog;
3297 	bool sleepable = prog->sleepable;
3298 	struct bpf_run_ctx *old_run_ctx;
3299 	int err = 0;
3300 
3301 	if (link->task && current->mm != link->task->mm)
3302 		return 0;
3303 
3304 	if (sleepable)
3305 		rcu_read_lock_trace();
3306 	else
3307 		rcu_read_lock();
3308 
3309 	migrate_disable();
3310 
3311 	old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
3312 	err = bpf_prog_run(link->link.prog, regs);
3313 	bpf_reset_run_ctx(old_run_ctx);
3314 
3315 	migrate_enable();
3316 
3317 	if (sleepable)
3318 		rcu_read_unlock_trace();
3319 	else
3320 		rcu_read_unlock();
3321 	return err;
3322 }
3323 
3324 static bool
3325 uprobe_multi_link_filter(struct uprobe_consumer *con, enum uprobe_filter_ctx ctx,
3326 			 struct mm_struct *mm)
3327 {
3328 	struct bpf_uprobe *uprobe;
3329 
3330 	uprobe = container_of(con, struct bpf_uprobe, consumer);
3331 	return uprobe->link->task->mm == mm;
3332 }
3333 
3334 static int
3335 uprobe_multi_link_handler(struct uprobe_consumer *con, struct pt_regs *regs)
3336 {
3337 	struct bpf_uprobe *uprobe;
3338 
3339 	uprobe = container_of(con, struct bpf_uprobe, consumer);
3340 	return uprobe_prog_run(uprobe, instruction_pointer(regs), regs);
3341 }
3342 
3343 static int
3344 uprobe_multi_link_ret_handler(struct uprobe_consumer *con, unsigned long func, struct pt_regs *regs)
3345 {
3346 	struct bpf_uprobe *uprobe;
3347 
3348 	uprobe = container_of(con, struct bpf_uprobe, consumer);
3349 	return uprobe_prog_run(uprobe, func, regs);
3350 }
3351 
3352 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3353 {
3354 	struct bpf_uprobe_multi_run_ctx *run_ctx;
3355 
3356 	run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3357 	return run_ctx->entry_ip;
3358 }
3359 
3360 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3361 {
3362 	struct bpf_uprobe_multi_run_ctx *run_ctx;
3363 
3364 	run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3365 	return run_ctx->uprobe->cookie;
3366 }
3367 
3368 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3369 {
3370 	struct bpf_uprobe_multi_link *link = NULL;
3371 	unsigned long __user *uref_ctr_offsets;
3372 	struct bpf_link_primer link_primer;
3373 	struct bpf_uprobe *uprobes = NULL;
3374 	struct task_struct *task = NULL;
3375 	unsigned long __user *uoffsets;
3376 	u64 __user *ucookies;
3377 	void __user *upath;
3378 	u32 flags, cnt, i;
3379 	struct path path;
3380 	char *name;
3381 	pid_t pid;
3382 	int err;
3383 
3384 	/* no support for 32bit archs yet */
3385 	if (sizeof(u64) != sizeof(void *))
3386 		return -EOPNOTSUPP;
3387 
3388 	if (prog->expected_attach_type != BPF_TRACE_UPROBE_MULTI)
3389 		return -EINVAL;
3390 
3391 	flags = attr->link_create.uprobe_multi.flags;
3392 	if (flags & ~BPF_F_UPROBE_MULTI_RETURN)
3393 		return -EINVAL;
3394 
3395 	/*
3396 	 * path, offsets and cnt are mandatory,
3397 	 * ref_ctr_offsets and cookies are optional
3398 	 */
3399 	upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
3400 	uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
3401 	cnt = attr->link_create.uprobe_multi.cnt;
3402 	pid = attr->link_create.uprobe_multi.pid;
3403 
3404 	if (!upath || !uoffsets || !cnt || pid < 0)
3405 		return -EINVAL;
3406 	if (cnt > MAX_UPROBE_MULTI_CNT)
3407 		return -E2BIG;
3408 
3409 	uref_ctr_offsets = u64_to_user_ptr(attr->link_create.uprobe_multi.ref_ctr_offsets);
3410 	ucookies = u64_to_user_ptr(attr->link_create.uprobe_multi.cookies);
3411 
3412 	name = strndup_user(upath, PATH_MAX);
3413 	if (IS_ERR(name)) {
3414 		err = PTR_ERR(name);
3415 		return err;
3416 	}
3417 
3418 	err = kern_path(name, LOOKUP_FOLLOW, &path);
3419 	kfree(name);
3420 	if (err)
3421 		return err;
3422 
3423 	if (!d_is_reg(path.dentry)) {
3424 		err = -EBADF;
3425 		goto error_path_put;
3426 	}
3427 
3428 	if (pid) {
3429 		task = get_pid_task(find_vpid(pid), PIDTYPE_TGID);
3430 		if (!task) {
3431 			err = -ESRCH;
3432 			goto error_path_put;
3433 		}
3434 	}
3435 
3436 	err = -ENOMEM;
3437 
3438 	link = kzalloc(sizeof(*link), GFP_KERNEL);
3439 	uprobes = kvcalloc(cnt, sizeof(*uprobes), GFP_KERNEL);
3440 
3441 	if (!uprobes || !link)
3442 		goto error_free;
3443 
3444 	for (i = 0; i < cnt; i++) {
3445 		if (__get_user(uprobes[i].offset, uoffsets + i)) {
3446 			err = -EFAULT;
3447 			goto error_free;
3448 		}
3449 		if (uprobes[i].offset < 0) {
3450 			err = -EINVAL;
3451 			goto error_free;
3452 		}
3453 		if (uref_ctr_offsets && __get_user(uprobes[i].ref_ctr_offset, uref_ctr_offsets + i)) {
3454 			err = -EFAULT;
3455 			goto error_free;
3456 		}
3457 		if (ucookies && __get_user(uprobes[i].cookie, ucookies + i)) {
3458 			err = -EFAULT;
3459 			goto error_free;
3460 		}
3461 
3462 		uprobes[i].link = link;
3463 
3464 		if (flags & BPF_F_UPROBE_MULTI_RETURN)
3465 			uprobes[i].consumer.ret_handler = uprobe_multi_link_ret_handler;
3466 		else
3467 			uprobes[i].consumer.handler = uprobe_multi_link_handler;
3468 
3469 		if (pid)
3470 			uprobes[i].consumer.filter = uprobe_multi_link_filter;
3471 	}
3472 
3473 	link->cnt = cnt;
3474 	link->uprobes = uprobes;
3475 	link->path = path;
3476 	link->task = task;
3477 	link->flags = flags;
3478 
3479 	bpf_link_init(&link->link, BPF_LINK_TYPE_UPROBE_MULTI,
3480 		      &bpf_uprobe_multi_link_lops, prog);
3481 
3482 	for (i = 0; i < cnt; i++) {
3483 		err = uprobe_register_refctr(d_real_inode(link->path.dentry),
3484 					     uprobes[i].offset,
3485 					     uprobes[i].ref_ctr_offset,
3486 					     &uprobes[i].consumer);
3487 		if (err) {
3488 			bpf_uprobe_unregister(&path, uprobes, i);
3489 			goto error_free;
3490 		}
3491 	}
3492 
3493 	err = bpf_link_prime(&link->link, &link_primer);
3494 	if (err)
3495 		goto error_free;
3496 
3497 	return bpf_link_settle(&link_primer);
3498 
3499 error_free:
3500 	kvfree(uprobes);
3501 	kfree(link);
3502 	if (task)
3503 		put_task_struct(task);
3504 error_path_put:
3505 	path_put(&path);
3506 	return err;
3507 }
3508 #else /* !CONFIG_UPROBES */
3509 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3510 {
3511 	return -EOPNOTSUPP;
3512 }
3513 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3514 {
3515 	return 0;
3516 }
3517 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3518 {
3519 	return 0;
3520 }
3521 #endif /* CONFIG_UPROBES */
3522 
3523 __bpf_kfunc_start_defs();
3524 
3525 __bpf_kfunc bool bpf_session_is_return(void)
3526 {
3527 	struct bpf_session_run_ctx *session_ctx;
3528 
3529 	session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3530 	return session_ctx->is_return;
3531 }
3532 
3533 __bpf_kfunc __u64 *bpf_session_cookie(void)
3534 {
3535 	struct bpf_session_run_ctx *session_ctx;
3536 
3537 	session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3538 	return session_ctx->data;
3539 }
3540 
3541 __bpf_kfunc_end_defs();
3542 
3543 BTF_KFUNCS_START(kprobe_multi_kfunc_set_ids)
3544 BTF_ID_FLAGS(func, bpf_session_is_return)
3545 BTF_ID_FLAGS(func, bpf_session_cookie)
3546 BTF_KFUNCS_END(kprobe_multi_kfunc_set_ids)
3547 
3548 static int bpf_kprobe_multi_filter(const struct bpf_prog *prog, u32 kfunc_id)
3549 {
3550 	if (!btf_id_set8_contains(&kprobe_multi_kfunc_set_ids, kfunc_id))
3551 		return 0;
3552 
3553 	if (!is_kprobe_session(prog))
3554 		return -EACCES;
3555 
3556 	return 0;
3557 }
3558 
3559 static const struct btf_kfunc_id_set bpf_kprobe_multi_kfunc_set = {
3560 	.owner = THIS_MODULE,
3561 	.set = &kprobe_multi_kfunc_set_ids,
3562 	.filter = bpf_kprobe_multi_filter,
3563 };
3564 
3565 static int __init bpf_kprobe_multi_kfuncs_init(void)
3566 {
3567 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_KPROBE, &bpf_kprobe_multi_kfunc_set);
3568 }
3569 
3570 late_initcall(bpf_kprobe_multi_kfuncs_init);
3571