xref: /linux/kernel/trace/bpf_trace.c (revision 6863aaa88516292b885fdce5dd91925a00c3a3de)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3  * Copyright (c) 2016 Facebook
4  */
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/slab.h>
8 #include <linux/bpf.h>
9 #include <linux/bpf_perf_event.h>
10 #include <linux/btf.h>
11 #include <linux/filter.h>
12 #include <linux/uaccess.h>
13 #include <linux/ctype.h>
14 #include <linux/kprobes.h>
15 #include <linux/spinlock.h>
16 #include <linux/syscalls.h>
17 #include <linux/error-injection.h>
18 #include <linux/btf_ids.h>
19 #include <linux/bpf_lsm.h>
20 #include <linux/fprobe.h>
21 #include <linux/bsearch.h>
22 #include <linux/sort.h>
23 #include <linux/key.h>
24 #include <linux/verification.h>
25 
26 #include <net/bpf_sk_storage.h>
27 
28 #include <uapi/linux/bpf.h>
29 #include <uapi/linux/btf.h>
30 
31 #include <asm/tlb.h>
32 
33 #include "trace_probe.h"
34 #include "trace.h"
35 
36 #define CREATE_TRACE_POINTS
37 #include "bpf_trace.h"
38 
39 #define bpf_event_rcu_dereference(p)					\
40 	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
41 
42 #ifdef CONFIG_MODULES
43 struct bpf_trace_module {
44 	struct module *module;
45 	struct list_head list;
46 };
47 
48 static LIST_HEAD(bpf_trace_modules);
49 static DEFINE_MUTEX(bpf_module_mutex);
50 
51 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
52 {
53 	struct bpf_raw_event_map *btp, *ret = NULL;
54 	struct bpf_trace_module *btm;
55 	unsigned int i;
56 
57 	mutex_lock(&bpf_module_mutex);
58 	list_for_each_entry(btm, &bpf_trace_modules, list) {
59 		for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
60 			btp = &btm->module->bpf_raw_events[i];
61 			if (!strcmp(btp->tp->name, name)) {
62 				if (try_module_get(btm->module))
63 					ret = btp;
64 				goto out;
65 			}
66 		}
67 	}
68 out:
69 	mutex_unlock(&bpf_module_mutex);
70 	return ret;
71 }
72 #else
73 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
74 {
75 	return NULL;
76 }
77 #endif /* CONFIG_MODULES */
78 
79 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
80 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
81 
82 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
83 				  u64 flags, const struct btf **btf,
84 				  s32 *btf_id);
85 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
86 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
87 
88 /**
89  * trace_call_bpf - invoke BPF program
90  * @call: tracepoint event
91  * @ctx: opaque context pointer
92  *
93  * kprobe handlers execute BPF programs via this helper.
94  * Can be used from static tracepoints in the future.
95  *
96  * Return: BPF programs always return an integer which is interpreted by
97  * kprobe handler as:
98  * 0 - return from kprobe (event is filtered out)
99  * 1 - store kprobe event into ring buffer
100  * Other values are reserved and currently alias to 1
101  */
102 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
103 {
104 	unsigned int ret;
105 
106 	cant_sleep();
107 
108 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
109 		/*
110 		 * since some bpf program is already running on this cpu,
111 		 * don't call into another bpf program (same or different)
112 		 * and don't send kprobe event into ring-buffer,
113 		 * so return zero here
114 		 */
115 		ret = 0;
116 		goto out;
117 	}
118 
119 	/*
120 	 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
121 	 * to all call sites, we did a bpf_prog_array_valid() there to check
122 	 * whether call->prog_array is empty or not, which is
123 	 * a heuristic to speed up execution.
124 	 *
125 	 * If bpf_prog_array_valid() fetched prog_array was
126 	 * non-NULL, we go into trace_call_bpf() and do the actual
127 	 * proper rcu_dereference() under RCU lock.
128 	 * If it turns out that prog_array is NULL then, we bail out.
129 	 * For the opposite, if the bpf_prog_array_valid() fetched pointer
130 	 * was NULL, you'll skip the prog_array with the risk of missing
131 	 * out of events when it was updated in between this and the
132 	 * rcu_dereference() which is accepted risk.
133 	 */
134 	rcu_read_lock();
135 	ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
136 				 ctx, bpf_prog_run);
137 	rcu_read_unlock();
138 
139  out:
140 	__this_cpu_dec(bpf_prog_active);
141 
142 	return ret;
143 }
144 
145 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
146 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
147 {
148 	regs_set_return_value(regs, rc);
149 	override_function_with_return(regs);
150 	return 0;
151 }
152 
153 static const struct bpf_func_proto bpf_override_return_proto = {
154 	.func		= bpf_override_return,
155 	.gpl_only	= true,
156 	.ret_type	= RET_INTEGER,
157 	.arg1_type	= ARG_PTR_TO_CTX,
158 	.arg2_type	= ARG_ANYTHING,
159 };
160 #endif
161 
162 static __always_inline int
163 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
164 {
165 	int ret;
166 
167 	ret = copy_from_user_nofault(dst, unsafe_ptr, size);
168 	if (unlikely(ret < 0))
169 		memset(dst, 0, size);
170 	return ret;
171 }
172 
173 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
174 	   const void __user *, unsafe_ptr)
175 {
176 	return bpf_probe_read_user_common(dst, size, unsafe_ptr);
177 }
178 
179 const struct bpf_func_proto bpf_probe_read_user_proto = {
180 	.func		= bpf_probe_read_user,
181 	.gpl_only	= true,
182 	.ret_type	= RET_INTEGER,
183 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
184 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
185 	.arg3_type	= ARG_ANYTHING,
186 };
187 
188 static __always_inline int
189 bpf_probe_read_user_str_common(void *dst, u32 size,
190 			       const void __user *unsafe_ptr)
191 {
192 	int ret;
193 
194 	/*
195 	 * NB: We rely on strncpy_from_user() not copying junk past the NUL
196 	 * terminator into `dst`.
197 	 *
198 	 * strncpy_from_user() does long-sized strides in the fast path. If the
199 	 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
200 	 * then there could be junk after the NUL in `dst`. If user takes `dst`
201 	 * and keys a hash map with it, then semantically identical strings can
202 	 * occupy multiple entries in the map.
203 	 */
204 	ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
205 	if (unlikely(ret < 0))
206 		memset(dst, 0, size);
207 	return ret;
208 }
209 
210 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
211 	   const void __user *, unsafe_ptr)
212 {
213 	return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
214 }
215 
216 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
217 	.func		= bpf_probe_read_user_str,
218 	.gpl_only	= true,
219 	.ret_type	= RET_INTEGER,
220 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
221 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
222 	.arg3_type	= ARG_ANYTHING,
223 };
224 
225 static __always_inline int
226 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
227 {
228 	int ret;
229 
230 	ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
231 	if (unlikely(ret < 0))
232 		memset(dst, 0, size);
233 	return ret;
234 }
235 
236 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
237 	   const void *, unsafe_ptr)
238 {
239 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
240 }
241 
242 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
243 	.func		= bpf_probe_read_kernel,
244 	.gpl_only	= true,
245 	.ret_type	= RET_INTEGER,
246 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
247 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
248 	.arg3_type	= ARG_ANYTHING,
249 };
250 
251 static __always_inline int
252 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
253 {
254 	int ret;
255 
256 	/*
257 	 * The strncpy_from_kernel_nofault() call will likely not fill the
258 	 * entire buffer, but that's okay in this circumstance as we're probing
259 	 * arbitrary memory anyway similar to bpf_probe_read_*() and might
260 	 * as well probe the stack. Thus, memory is explicitly cleared
261 	 * only in error case, so that improper users ignoring return
262 	 * code altogether don't copy garbage; otherwise length of string
263 	 * is returned that can be used for bpf_perf_event_output() et al.
264 	 */
265 	ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
266 	if (unlikely(ret < 0))
267 		memset(dst, 0, size);
268 	return ret;
269 }
270 
271 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
272 	   const void *, unsafe_ptr)
273 {
274 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
275 }
276 
277 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
278 	.func		= bpf_probe_read_kernel_str,
279 	.gpl_only	= true,
280 	.ret_type	= RET_INTEGER,
281 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
282 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
283 	.arg3_type	= ARG_ANYTHING,
284 };
285 
286 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
287 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
288 	   const void *, unsafe_ptr)
289 {
290 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
291 		return bpf_probe_read_user_common(dst, size,
292 				(__force void __user *)unsafe_ptr);
293 	}
294 	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
295 }
296 
297 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
298 	.func		= bpf_probe_read_compat,
299 	.gpl_only	= true,
300 	.ret_type	= RET_INTEGER,
301 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
302 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
303 	.arg3_type	= ARG_ANYTHING,
304 };
305 
306 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
307 	   const void *, unsafe_ptr)
308 {
309 	if ((unsigned long)unsafe_ptr < TASK_SIZE) {
310 		return bpf_probe_read_user_str_common(dst, size,
311 				(__force void __user *)unsafe_ptr);
312 	}
313 	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
314 }
315 
316 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
317 	.func		= bpf_probe_read_compat_str,
318 	.gpl_only	= true,
319 	.ret_type	= RET_INTEGER,
320 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
321 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
322 	.arg3_type	= ARG_ANYTHING,
323 };
324 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
325 
326 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
327 	   u32, size)
328 {
329 	/*
330 	 * Ensure we're in user context which is safe for the helper to
331 	 * run. This helper has no business in a kthread.
332 	 *
333 	 * access_ok() should prevent writing to non-user memory, but in
334 	 * some situations (nommu, temporary switch, etc) access_ok() does
335 	 * not provide enough validation, hence the check on KERNEL_DS.
336 	 *
337 	 * nmi_uaccess_okay() ensures the probe is not run in an interim
338 	 * state, when the task or mm are switched. This is specifically
339 	 * required to prevent the use of temporary mm.
340 	 */
341 
342 	if (unlikely(in_interrupt() ||
343 		     current->flags & (PF_KTHREAD | PF_EXITING)))
344 		return -EPERM;
345 	if (unlikely(!nmi_uaccess_okay()))
346 		return -EPERM;
347 
348 	return copy_to_user_nofault(unsafe_ptr, src, size);
349 }
350 
351 static const struct bpf_func_proto bpf_probe_write_user_proto = {
352 	.func		= bpf_probe_write_user,
353 	.gpl_only	= true,
354 	.ret_type	= RET_INTEGER,
355 	.arg1_type	= ARG_ANYTHING,
356 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
357 	.arg3_type	= ARG_CONST_SIZE,
358 };
359 
360 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
361 {
362 	if (!capable(CAP_SYS_ADMIN))
363 		return NULL;
364 
365 	pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
366 			    current->comm, task_pid_nr(current));
367 
368 	return &bpf_probe_write_user_proto;
369 }
370 
371 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
372 
373 #define MAX_TRACE_PRINTK_VARARGS	3
374 #define BPF_TRACE_PRINTK_SIZE		1024
375 
376 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
377 	   u64, arg2, u64, arg3)
378 {
379 	u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
380 	u32 *bin_args;
381 	static char buf[BPF_TRACE_PRINTK_SIZE];
382 	unsigned long flags;
383 	int ret;
384 
385 	ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
386 				  MAX_TRACE_PRINTK_VARARGS);
387 	if (ret < 0)
388 		return ret;
389 
390 	raw_spin_lock_irqsave(&trace_printk_lock, flags);
391 	ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
392 
393 	trace_bpf_trace_printk(buf);
394 	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
395 
396 	bpf_bprintf_cleanup();
397 
398 	return ret;
399 }
400 
401 static const struct bpf_func_proto bpf_trace_printk_proto = {
402 	.func		= bpf_trace_printk,
403 	.gpl_only	= true,
404 	.ret_type	= RET_INTEGER,
405 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
406 	.arg2_type	= ARG_CONST_SIZE,
407 };
408 
409 static void __set_printk_clr_event(void)
410 {
411 	/*
412 	 * This program might be calling bpf_trace_printk,
413 	 * so enable the associated bpf_trace/bpf_trace_printk event.
414 	 * Repeat this each time as it is possible a user has
415 	 * disabled bpf_trace_printk events.  By loading a program
416 	 * calling bpf_trace_printk() however the user has expressed
417 	 * the intent to see such events.
418 	 */
419 	if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
420 		pr_warn_ratelimited("could not enable bpf_trace_printk events");
421 }
422 
423 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
424 {
425 	__set_printk_clr_event();
426 	return &bpf_trace_printk_proto;
427 }
428 
429 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
430 	   u32, data_len)
431 {
432 	static char buf[BPF_TRACE_PRINTK_SIZE];
433 	unsigned long flags;
434 	int ret, num_args;
435 	u32 *bin_args;
436 
437 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
438 	    (data_len && !data))
439 		return -EINVAL;
440 	num_args = data_len / 8;
441 
442 	ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
443 	if (ret < 0)
444 		return ret;
445 
446 	raw_spin_lock_irqsave(&trace_printk_lock, flags);
447 	ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
448 
449 	trace_bpf_trace_printk(buf);
450 	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
451 
452 	bpf_bprintf_cleanup();
453 
454 	return ret;
455 }
456 
457 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
458 	.func		= bpf_trace_vprintk,
459 	.gpl_only	= true,
460 	.ret_type	= RET_INTEGER,
461 	.arg1_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
462 	.arg2_type	= ARG_CONST_SIZE,
463 	.arg3_type	= ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
464 	.arg4_type	= ARG_CONST_SIZE_OR_ZERO,
465 };
466 
467 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
468 {
469 	__set_printk_clr_event();
470 	return &bpf_trace_vprintk_proto;
471 }
472 
473 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
474 	   const void *, data, u32, data_len)
475 {
476 	int err, num_args;
477 	u32 *bin_args;
478 
479 	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
480 	    (data_len && !data))
481 		return -EINVAL;
482 	num_args = data_len / 8;
483 
484 	err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
485 	if (err < 0)
486 		return err;
487 
488 	seq_bprintf(m, fmt, bin_args);
489 
490 	bpf_bprintf_cleanup();
491 
492 	return seq_has_overflowed(m) ? -EOVERFLOW : 0;
493 }
494 
495 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
496 
497 static const struct bpf_func_proto bpf_seq_printf_proto = {
498 	.func		= bpf_seq_printf,
499 	.gpl_only	= true,
500 	.ret_type	= RET_INTEGER,
501 	.arg1_type	= ARG_PTR_TO_BTF_ID,
502 	.arg1_btf_id	= &btf_seq_file_ids[0],
503 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
504 	.arg3_type	= ARG_CONST_SIZE,
505 	.arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
506 	.arg5_type      = ARG_CONST_SIZE_OR_ZERO,
507 };
508 
509 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
510 {
511 	return seq_write(m, data, len) ? -EOVERFLOW : 0;
512 }
513 
514 static const struct bpf_func_proto bpf_seq_write_proto = {
515 	.func		= bpf_seq_write,
516 	.gpl_only	= true,
517 	.ret_type	= RET_INTEGER,
518 	.arg1_type	= ARG_PTR_TO_BTF_ID,
519 	.arg1_btf_id	= &btf_seq_file_ids[0],
520 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
521 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
522 };
523 
524 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
525 	   u32, btf_ptr_size, u64, flags)
526 {
527 	const struct btf *btf;
528 	s32 btf_id;
529 	int ret;
530 
531 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
532 	if (ret)
533 		return ret;
534 
535 	return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
536 }
537 
538 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
539 	.func		= bpf_seq_printf_btf,
540 	.gpl_only	= true,
541 	.ret_type	= RET_INTEGER,
542 	.arg1_type	= ARG_PTR_TO_BTF_ID,
543 	.arg1_btf_id	= &btf_seq_file_ids[0],
544 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
545 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
546 	.arg4_type	= ARG_ANYTHING,
547 };
548 
549 static __always_inline int
550 get_map_perf_counter(struct bpf_map *map, u64 flags,
551 		     u64 *value, u64 *enabled, u64 *running)
552 {
553 	struct bpf_array *array = container_of(map, struct bpf_array, map);
554 	unsigned int cpu = smp_processor_id();
555 	u64 index = flags & BPF_F_INDEX_MASK;
556 	struct bpf_event_entry *ee;
557 
558 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
559 		return -EINVAL;
560 	if (index == BPF_F_CURRENT_CPU)
561 		index = cpu;
562 	if (unlikely(index >= array->map.max_entries))
563 		return -E2BIG;
564 
565 	ee = READ_ONCE(array->ptrs[index]);
566 	if (!ee)
567 		return -ENOENT;
568 
569 	return perf_event_read_local(ee->event, value, enabled, running);
570 }
571 
572 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
573 {
574 	u64 value = 0;
575 	int err;
576 
577 	err = get_map_perf_counter(map, flags, &value, NULL, NULL);
578 	/*
579 	 * this api is ugly since we miss [-22..-2] range of valid
580 	 * counter values, but that's uapi
581 	 */
582 	if (err)
583 		return err;
584 	return value;
585 }
586 
587 static const struct bpf_func_proto bpf_perf_event_read_proto = {
588 	.func		= bpf_perf_event_read,
589 	.gpl_only	= true,
590 	.ret_type	= RET_INTEGER,
591 	.arg1_type	= ARG_CONST_MAP_PTR,
592 	.arg2_type	= ARG_ANYTHING,
593 };
594 
595 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
596 	   struct bpf_perf_event_value *, buf, u32, size)
597 {
598 	int err = -EINVAL;
599 
600 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
601 		goto clear;
602 	err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
603 				   &buf->running);
604 	if (unlikely(err))
605 		goto clear;
606 	return 0;
607 clear:
608 	memset(buf, 0, size);
609 	return err;
610 }
611 
612 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
613 	.func		= bpf_perf_event_read_value,
614 	.gpl_only	= true,
615 	.ret_type	= RET_INTEGER,
616 	.arg1_type	= ARG_CONST_MAP_PTR,
617 	.arg2_type	= ARG_ANYTHING,
618 	.arg3_type	= ARG_PTR_TO_UNINIT_MEM,
619 	.arg4_type	= ARG_CONST_SIZE,
620 };
621 
622 static __always_inline u64
623 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
624 			u64 flags, struct perf_sample_data *sd)
625 {
626 	struct bpf_array *array = container_of(map, struct bpf_array, map);
627 	unsigned int cpu = smp_processor_id();
628 	u64 index = flags & BPF_F_INDEX_MASK;
629 	struct bpf_event_entry *ee;
630 	struct perf_event *event;
631 
632 	if (index == BPF_F_CURRENT_CPU)
633 		index = cpu;
634 	if (unlikely(index >= array->map.max_entries))
635 		return -E2BIG;
636 
637 	ee = READ_ONCE(array->ptrs[index]);
638 	if (!ee)
639 		return -ENOENT;
640 
641 	event = ee->event;
642 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
643 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
644 		return -EINVAL;
645 
646 	if (unlikely(event->oncpu != cpu))
647 		return -EOPNOTSUPP;
648 
649 	return perf_event_output(event, sd, regs);
650 }
651 
652 /*
653  * Support executing tracepoints in normal, irq, and nmi context that each call
654  * bpf_perf_event_output
655  */
656 struct bpf_trace_sample_data {
657 	struct perf_sample_data sds[3];
658 };
659 
660 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
661 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
662 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
663 	   u64, flags, void *, data, u64, size)
664 {
665 	struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
666 	int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
667 	struct perf_raw_record raw = {
668 		.frag = {
669 			.size = size,
670 			.data = data,
671 		},
672 	};
673 	struct perf_sample_data *sd;
674 	int err;
675 
676 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
677 		err = -EBUSY;
678 		goto out;
679 	}
680 
681 	sd = &sds->sds[nest_level - 1];
682 
683 	if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
684 		err = -EINVAL;
685 		goto out;
686 	}
687 
688 	perf_sample_data_init(sd, 0, 0);
689 	sd->raw = &raw;
690 
691 	err = __bpf_perf_event_output(regs, map, flags, sd);
692 
693 out:
694 	this_cpu_dec(bpf_trace_nest_level);
695 	return err;
696 }
697 
698 static const struct bpf_func_proto bpf_perf_event_output_proto = {
699 	.func		= bpf_perf_event_output,
700 	.gpl_only	= true,
701 	.ret_type	= RET_INTEGER,
702 	.arg1_type	= ARG_PTR_TO_CTX,
703 	.arg2_type	= ARG_CONST_MAP_PTR,
704 	.arg3_type	= ARG_ANYTHING,
705 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
706 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
707 };
708 
709 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
710 struct bpf_nested_pt_regs {
711 	struct pt_regs regs[3];
712 };
713 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
714 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
715 
716 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
717 		     void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
718 {
719 	int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
720 	struct perf_raw_frag frag = {
721 		.copy		= ctx_copy,
722 		.size		= ctx_size,
723 		.data		= ctx,
724 	};
725 	struct perf_raw_record raw = {
726 		.frag = {
727 			{
728 				.next	= ctx_size ? &frag : NULL,
729 			},
730 			.size	= meta_size,
731 			.data	= meta,
732 		},
733 	};
734 	struct perf_sample_data *sd;
735 	struct pt_regs *regs;
736 	u64 ret;
737 
738 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
739 		ret = -EBUSY;
740 		goto out;
741 	}
742 	sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
743 	regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
744 
745 	perf_fetch_caller_regs(regs);
746 	perf_sample_data_init(sd, 0, 0);
747 	sd->raw = &raw;
748 
749 	ret = __bpf_perf_event_output(regs, map, flags, sd);
750 out:
751 	this_cpu_dec(bpf_event_output_nest_level);
752 	return ret;
753 }
754 
755 BPF_CALL_0(bpf_get_current_task)
756 {
757 	return (long) current;
758 }
759 
760 const struct bpf_func_proto bpf_get_current_task_proto = {
761 	.func		= bpf_get_current_task,
762 	.gpl_only	= true,
763 	.ret_type	= RET_INTEGER,
764 };
765 
766 BPF_CALL_0(bpf_get_current_task_btf)
767 {
768 	return (unsigned long) current;
769 }
770 
771 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
772 	.func		= bpf_get_current_task_btf,
773 	.gpl_only	= true,
774 	.ret_type	= RET_PTR_TO_BTF_ID,
775 	.ret_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
776 };
777 
778 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
779 {
780 	return (unsigned long) task_pt_regs(task);
781 }
782 
783 BTF_ID_LIST(bpf_task_pt_regs_ids)
784 BTF_ID(struct, pt_regs)
785 
786 const struct bpf_func_proto bpf_task_pt_regs_proto = {
787 	.func		= bpf_task_pt_regs,
788 	.gpl_only	= true,
789 	.arg1_type	= ARG_PTR_TO_BTF_ID,
790 	.arg1_btf_id	= &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
791 	.ret_type	= RET_PTR_TO_BTF_ID,
792 	.ret_btf_id	= &bpf_task_pt_regs_ids[0],
793 };
794 
795 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
796 {
797 	struct bpf_array *array = container_of(map, struct bpf_array, map);
798 	struct cgroup *cgrp;
799 
800 	if (unlikely(idx >= array->map.max_entries))
801 		return -E2BIG;
802 
803 	cgrp = READ_ONCE(array->ptrs[idx]);
804 	if (unlikely(!cgrp))
805 		return -EAGAIN;
806 
807 	return task_under_cgroup_hierarchy(current, cgrp);
808 }
809 
810 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
811 	.func           = bpf_current_task_under_cgroup,
812 	.gpl_only       = false,
813 	.ret_type       = RET_INTEGER,
814 	.arg1_type      = ARG_CONST_MAP_PTR,
815 	.arg2_type      = ARG_ANYTHING,
816 };
817 
818 struct send_signal_irq_work {
819 	struct irq_work irq_work;
820 	struct task_struct *task;
821 	u32 sig;
822 	enum pid_type type;
823 };
824 
825 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
826 
827 static void do_bpf_send_signal(struct irq_work *entry)
828 {
829 	struct send_signal_irq_work *work;
830 
831 	work = container_of(entry, struct send_signal_irq_work, irq_work);
832 	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
833 }
834 
835 static int bpf_send_signal_common(u32 sig, enum pid_type type)
836 {
837 	struct send_signal_irq_work *work = NULL;
838 
839 	/* Similar to bpf_probe_write_user, task needs to be
840 	 * in a sound condition and kernel memory access be
841 	 * permitted in order to send signal to the current
842 	 * task.
843 	 */
844 	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
845 		return -EPERM;
846 	if (unlikely(!nmi_uaccess_okay()))
847 		return -EPERM;
848 
849 	if (irqs_disabled()) {
850 		/* Do an early check on signal validity. Otherwise,
851 		 * the error is lost in deferred irq_work.
852 		 */
853 		if (unlikely(!valid_signal(sig)))
854 			return -EINVAL;
855 
856 		work = this_cpu_ptr(&send_signal_work);
857 		if (irq_work_is_busy(&work->irq_work))
858 			return -EBUSY;
859 
860 		/* Add the current task, which is the target of sending signal,
861 		 * to the irq_work. The current task may change when queued
862 		 * irq works get executed.
863 		 */
864 		work->task = current;
865 		work->sig = sig;
866 		work->type = type;
867 		irq_work_queue(&work->irq_work);
868 		return 0;
869 	}
870 
871 	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
872 }
873 
874 BPF_CALL_1(bpf_send_signal, u32, sig)
875 {
876 	return bpf_send_signal_common(sig, PIDTYPE_TGID);
877 }
878 
879 static const struct bpf_func_proto bpf_send_signal_proto = {
880 	.func		= bpf_send_signal,
881 	.gpl_only	= false,
882 	.ret_type	= RET_INTEGER,
883 	.arg1_type	= ARG_ANYTHING,
884 };
885 
886 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
887 {
888 	return bpf_send_signal_common(sig, PIDTYPE_PID);
889 }
890 
891 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
892 	.func		= bpf_send_signal_thread,
893 	.gpl_only	= false,
894 	.ret_type	= RET_INTEGER,
895 	.arg1_type	= ARG_ANYTHING,
896 };
897 
898 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
899 {
900 	long len;
901 	char *p;
902 
903 	if (!sz)
904 		return 0;
905 
906 	p = d_path(path, buf, sz);
907 	if (IS_ERR(p)) {
908 		len = PTR_ERR(p);
909 	} else {
910 		len = buf + sz - p;
911 		memmove(buf, p, len);
912 	}
913 
914 	return len;
915 }
916 
917 BTF_SET_START(btf_allowlist_d_path)
918 #ifdef CONFIG_SECURITY
919 BTF_ID(func, security_file_permission)
920 BTF_ID(func, security_inode_getattr)
921 BTF_ID(func, security_file_open)
922 #endif
923 #ifdef CONFIG_SECURITY_PATH
924 BTF_ID(func, security_path_truncate)
925 #endif
926 BTF_ID(func, vfs_truncate)
927 BTF_ID(func, vfs_fallocate)
928 BTF_ID(func, dentry_open)
929 BTF_ID(func, vfs_getattr)
930 BTF_ID(func, filp_close)
931 BTF_SET_END(btf_allowlist_d_path)
932 
933 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
934 {
935 	if (prog->type == BPF_PROG_TYPE_TRACING &&
936 	    prog->expected_attach_type == BPF_TRACE_ITER)
937 		return true;
938 
939 	if (prog->type == BPF_PROG_TYPE_LSM)
940 		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
941 
942 	return btf_id_set_contains(&btf_allowlist_d_path,
943 				   prog->aux->attach_btf_id);
944 }
945 
946 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
947 
948 static const struct bpf_func_proto bpf_d_path_proto = {
949 	.func		= bpf_d_path,
950 	.gpl_only	= false,
951 	.ret_type	= RET_INTEGER,
952 	.arg1_type	= ARG_PTR_TO_BTF_ID,
953 	.arg1_btf_id	= &bpf_d_path_btf_ids[0],
954 	.arg2_type	= ARG_PTR_TO_MEM,
955 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
956 	.allowed	= bpf_d_path_allowed,
957 };
958 
959 #define BTF_F_ALL	(BTF_F_COMPACT  | BTF_F_NONAME | \
960 			 BTF_F_PTR_RAW | BTF_F_ZERO)
961 
962 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
963 				  u64 flags, const struct btf **btf,
964 				  s32 *btf_id)
965 {
966 	const struct btf_type *t;
967 
968 	if (unlikely(flags & ~(BTF_F_ALL)))
969 		return -EINVAL;
970 
971 	if (btf_ptr_size != sizeof(struct btf_ptr))
972 		return -EINVAL;
973 
974 	*btf = bpf_get_btf_vmlinux();
975 
976 	if (IS_ERR_OR_NULL(*btf))
977 		return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
978 
979 	if (ptr->type_id > 0)
980 		*btf_id = ptr->type_id;
981 	else
982 		return -EINVAL;
983 
984 	if (*btf_id > 0)
985 		t = btf_type_by_id(*btf, *btf_id);
986 	if (*btf_id <= 0 || !t)
987 		return -ENOENT;
988 
989 	return 0;
990 }
991 
992 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
993 	   u32, btf_ptr_size, u64, flags)
994 {
995 	const struct btf *btf;
996 	s32 btf_id;
997 	int ret;
998 
999 	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1000 	if (ret)
1001 		return ret;
1002 
1003 	return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1004 				      flags);
1005 }
1006 
1007 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1008 	.func		= bpf_snprintf_btf,
1009 	.gpl_only	= false,
1010 	.ret_type	= RET_INTEGER,
1011 	.arg1_type	= ARG_PTR_TO_MEM,
1012 	.arg2_type	= ARG_CONST_SIZE,
1013 	.arg3_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1014 	.arg4_type	= ARG_CONST_SIZE,
1015 	.arg5_type	= ARG_ANYTHING,
1016 };
1017 
1018 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1019 {
1020 	/* This helper call is inlined by verifier. */
1021 	return ((u64 *)ctx)[-2];
1022 }
1023 
1024 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1025 	.func		= bpf_get_func_ip_tracing,
1026 	.gpl_only	= true,
1027 	.ret_type	= RET_INTEGER,
1028 	.arg1_type	= ARG_PTR_TO_CTX,
1029 };
1030 
1031 #ifdef CONFIG_X86_KERNEL_IBT
1032 static unsigned long get_entry_ip(unsigned long fentry_ip)
1033 {
1034 	u32 instr;
1035 
1036 	/* Being extra safe in here in case entry ip is on the page-edge. */
1037 	if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1038 		return fentry_ip;
1039 	if (is_endbr(instr))
1040 		fentry_ip -= ENDBR_INSN_SIZE;
1041 	return fentry_ip;
1042 }
1043 #else
1044 #define get_entry_ip(fentry_ip) fentry_ip
1045 #endif
1046 
1047 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1048 {
1049 	struct kprobe *kp = kprobe_running();
1050 
1051 	if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1052 		return 0;
1053 
1054 	return get_entry_ip((uintptr_t)kp->addr);
1055 }
1056 
1057 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1058 	.func		= bpf_get_func_ip_kprobe,
1059 	.gpl_only	= true,
1060 	.ret_type	= RET_INTEGER,
1061 	.arg1_type	= ARG_PTR_TO_CTX,
1062 };
1063 
1064 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1065 {
1066 	return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1067 }
1068 
1069 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1070 	.func		= bpf_get_func_ip_kprobe_multi,
1071 	.gpl_only	= false,
1072 	.ret_type	= RET_INTEGER,
1073 	.arg1_type	= ARG_PTR_TO_CTX,
1074 };
1075 
1076 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1077 {
1078 	return bpf_kprobe_multi_cookie(current->bpf_ctx);
1079 }
1080 
1081 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1082 	.func		= bpf_get_attach_cookie_kprobe_multi,
1083 	.gpl_only	= false,
1084 	.ret_type	= RET_INTEGER,
1085 	.arg1_type	= ARG_PTR_TO_CTX,
1086 };
1087 
1088 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1089 {
1090 	struct bpf_trace_run_ctx *run_ctx;
1091 
1092 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1093 	return run_ctx->bpf_cookie;
1094 }
1095 
1096 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1097 	.func		= bpf_get_attach_cookie_trace,
1098 	.gpl_only	= false,
1099 	.ret_type	= RET_INTEGER,
1100 	.arg1_type	= ARG_PTR_TO_CTX,
1101 };
1102 
1103 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1104 {
1105 	return ctx->event->bpf_cookie;
1106 }
1107 
1108 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1109 	.func		= bpf_get_attach_cookie_pe,
1110 	.gpl_only	= false,
1111 	.ret_type	= RET_INTEGER,
1112 	.arg1_type	= ARG_PTR_TO_CTX,
1113 };
1114 
1115 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1116 {
1117 	struct bpf_trace_run_ctx *run_ctx;
1118 
1119 	run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1120 	return run_ctx->bpf_cookie;
1121 }
1122 
1123 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1124 	.func		= bpf_get_attach_cookie_tracing,
1125 	.gpl_only	= false,
1126 	.ret_type	= RET_INTEGER,
1127 	.arg1_type	= ARG_PTR_TO_CTX,
1128 };
1129 
1130 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1131 {
1132 #ifndef CONFIG_X86
1133 	return -ENOENT;
1134 #else
1135 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1136 	u32 entry_cnt = size / br_entry_size;
1137 
1138 	entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1139 
1140 	if (unlikely(flags))
1141 		return -EINVAL;
1142 
1143 	if (!entry_cnt)
1144 		return -ENOENT;
1145 
1146 	return entry_cnt * br_entry_size;
1147 #endif
1148 }
1149 
1150 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1151 	.func		= bpf_get_branch_snapshot,
1152 	.gpl_only	= true,
1153 	.ret_type	= RET_INTEGER,
1154 	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
1155 	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
1156 };
1157 
1158 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1159 {
1160 	/* This helper call is inlined by verifier. */
1161 	u64 nr_args = ((u64 *)ctx)[-1];
1162 
1163 	if ((u64) n >= nr_args)
1164 		return -EINVAL;
1165 	*value = ((u64 *)ctx)[n];
1166 	return 0;
1167 }
1168 
1169 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1170 	.func		= get_func_arg,
1171 	.ret_type	= RET_INTEGER,
1172 	.arg1_type	= ARG_PTR_TO_CTX,
1173 	.arg2_type	= ARG_ANYTHING,
1174 	.arg3_type	= ARG_PTR_TO_LONG,
1175 };
1176 
1177 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1178 {
1179 	/* This helper call is inlined by verifier. */
1180 	u64 nr_args = ((u64 *)ctx)[-1];
1181 
1182 	*value = ((u64 *)ctx)[nr_args];
1183 	return 0;
1184 }
1185 
1186 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1187 	.func		= get_func_ret,
1188 	.ret_type	= RET_INTEGER,
1189 	.arg1_type	= ARG_PTR_TO_CTX,
1190 	.arg2_type	= ARG_PTR_TO_LONG,
1191 };
1192 
1193 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1194 {
1195 	/* This helper call is inlined by verifier. */
1196 	return ((u64 *)ctx)[-1];
1197 }
1198 
1199 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1200 	.func		= get_func_arg_cnt,
1201 	.ret_type	= RET_INTEGER,
1202 	.arg1_type	= ARG_PTR_TO_CTX,
1203 };
1204 
1205 #ifdef CONFIG_KEYS
1206 __diag_push();
1207 __diag_ignore_all("-Wmissing-prototypes",
1208 		  "kfuncs which will be used in BPF programs");
1209 
1210 /**
1211  * bpf_lookup_user_key - lookup a key by its serial
1212  * @serial: key handle serial number
1213  * @flags: lookup-specific flags
1214  *
1215  * Search a key with a given *serial* and the provided *flags*.
1216  * If found, increment the reference count of the key by one, and
1217  * return it in the bpf_key structure.
1218  *
1219  * The bpf_key structure must be passed to bpf_key_put() when done
1220  * with it, so that the key reference count is decremented and the
1221  * bpf_key structure is freed.
1222  *
1223  * Permission checks are deferred to the time the key is used by
1224  * one of the available key-specific kfuncs.
1225  *
1226  * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1227  * special keyring (e.g. session keyring), if it doesn't yet exist.
1228  * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1229  * for the key construction, and to retrieve uninstantiated keys (keys
1230  * without data attached to them).
1231  *
1232  * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1233  *         NULL pointer otherwise.
1234  */
1235 struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1236 {
1237 	key_ref_t key_ref;
1238 	struct bpf_key *bkey;
1239 
1240 	if (flags & ~KEY_LOOKUP_ALL)
1241 		return NULL;
1242 
1243 	/*
1244 	 * Permission check is deferred until the key is used, as the
1245 	 * intent of the caller is unknown here.
1246 	 */
1247 	key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1248 	if (IS_ERR(key_ref))
1249 		return NULL;
1250 
1251 	bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1252 	if (!bkey) {
1253 		key_put(key_ref_to_ptr(key_ref));
1254 		return NULL;
1255 	}
1256 
1257 	bkey->key = key_ref_to_ptr(key_ref);
1258 	bkey->has_ref = true;
1259 
1260 	return bkey;
1261 }
1262 
1263 /**
1264  * bpf_lookup_system_key - lookup a key by a system-defined ID
1265  * @id: key ID
1266  *
1267  * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1268  * The key pointer is marked as invalid, to prevent bpf_key_put() from
1269  * attempting to decrement the key reference count on that pointer. The key
1270  * pointer set in such way is currently understood only by
1271  * verify_pkcs7_signature().
1272  *
1273  * Set *id* to one of the values defined in include/linux/verification.h:
1274  * 0 for the primary keyring (immutable keyring of system keys);
1275  * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1276  * (where keys can be added only if they are vouched for by existing keys
1277  * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1278  * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1279  * kerned image and, possibly, the initramfs signature).
1280  *
1281  * Return: a bpf_key pointer with an invalid key pointer set from the
1282  *         pre-determined ID on success, a NULL pointer otherwise
1283  */
1284 struct bpf_key *bpf_lookup_system_key(u64 id)
1285 {
1286 	struct bpf_key *bkey;
1287 
1288 	if (system_keyring_id_check(id) < 0)
1289 		return NULL;
1290 
1291 	bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1292 	if (!bkey)
1293 		return NULL;
1294 
1295 	bkey->key = (struct key *)(unsigned long)id;
1296 	bkey->has_ref = false;
1297 
1298 	return bkey;
1299 }
1300 
1301 /**
1302  * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1303  * @bkey: bpf_key structure
1304  *
1305  * Decrement the reference count of the key inside *bkey*, if the pointer
1306  * is valid, and free *bkey*.
1307  */
1308 void bpf_key_put(struct bpf_key *bkey)
1309 {
1310 	if (bkey->has_ref)
1311 		key_put(bkey->key);
1312 
1313 	kfree(bkey);
1314 }
1315 
1316 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1317 /**
1318  * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1319  * @data_ptr: data to verify
1320  * @sig_ptr: signature of the data
1321  * @trusted_keyring: keyring with keys trusted for signature verification
1322  *
1323  * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1324  * with keys in a keyring referenced by *trusted_keyring*.
1325  *
1326  * Return: 0 on success, a negative value on error.
1327  */
1328 int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1329 			       struct bpf_dynptr_kern *sig_ptr,
1330 			       struct bpf_key *trusted_keyring)
1331 {
1332 	int ret;
1333 
1334 	if (trusted_keyring->has_ref) {
1335 		/*
1336 		 * Do the permission check deferred in bpf_lookup_user_key().
1337 		 * See bpf_lookup_user_key() for more details.
1338 		 *
1339 		 * A call to key_task_permission() here would be redundant, as
1340 		 * it is already done by keyring_search() called by
1341 		 * find_asymmetric_key().
1342 		 */
1343 		ret = key_validate(trusted_keyring->key);
1344 		if (ret < 0)
1345 			return ret;
1346 	}
1347 
1348 	return verify_pkcs7_signature(data_ptr->data,
1349 				      bpf_dynptr_get_size(data_ptr),
1350 				      sig_ptr->data,
1351 				      bpf_dynptr_get_size(sig_ptr),
1352 				      trusted_keyring->key,
1353 				      VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1354 				      NULL);
1355 }
1356 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1357 
1358 __diag_pop();
1359 
1360 BTF_SET8_START(key_sig_kfunc_set)
1361 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1362 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1363 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1364 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1365 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1366 #endif
1367 BTF_SET8_END(key_sig_kfunc_set)
1368 
1369 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1370 	.owner = THIS_MODULE,
1371 	.set = &key_sig_kfunc_set,
1372 };
1373 
1374 static int __init bpf_key_sig_kfuncs_init(void)
1375 {
1376 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1377 					 &bpf_key_sig_kfunc_set);
1378 }
1379 
1380 late_initcall(bpf_key_sig_kfuncs_init);
1381 #endif /* CONFIG_KEYS */
1382 
1383 static const struct bpf_func_proto *
1384 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1385 {
1386 	switch (func_id) {
1387 	case BPF_FUNC_map_lookup_elem:
1388 		return &bpf_map_lookup_elem_proto;
1389 	case BPF_FUNC_map_update_elem:
1390 		return &bpf_map_update_elem_proto;
1391 	case BPF_FUNC_map_delete_elem:
1392 		return &bpf_map_delete_elem_proto;
1393 	case BPF_FUNC_map_push_elem:
1394 		return &bpf_map_push_elem_proto;
1395 	case BPF_FUNC_map_pop_elem:
1396 		return &bpf_map_pop_elem_proto;
1397 	case BPF_FUNC_map_peek_elem:
1398 		return &bpf_map_peek_elem_proto;
1399 	case BPF_FUNC_map_lookup_percpu_elem:
1400 		return &bpf_map_lookup_percpu_elem_proto;
1401 	case BPF_FUNC_ktime_get_ns:
1402 		return &bpf_ktime_get_ns_proto;
1403 	case BPF_FUNC_ktime_get_boot_ns:
1404 		return &bpf_ktime_get_boot_ns_proto;
1405 	case BPF_FUNC_tail_call:
1406 		return &bpf_tail_call_proto;
1407 	case BPF_FUNC_get_current_pid_tgid:
1408 		return &bpf_get_current_pid_tgid_proto;
1409 	case BPF_FUNC_get_current_task:
1410 		return &bpf_get_current_task_proto;
1411 	case BPF_FUNC_get_current_task_btf:
1412 		return &bpf_get_current_task_btf_proto;
1413 	case BPF_FUNC_task_pt_regs:
1414 		return &bpf_task_pt_regs_proto;
1415 	case BPF_FUNC_get_current_uid_gid:
1416 		return &bpf_get_current_uid_gid_proto;
1417 	case BPF_FUNC_get_current_comm:
1418 		return &bpf_get_current_comm_proto;
1419 	case BPF_FUNC_trace_printk:
1420 		return bpf_get_trace_printk_proto();
1421 	case BPF_FUNC_get_smp_processor_id:
1422 		return &bpf_get_smp_processor_id_proto;
1423 	case BPF_FUNC_get_numa_node_id:
1424 		return &bpf_get_numa_node_id_proto;
1425 	case BPF_FUNC_perf_event_read:
1426 		return &bpf_perf_event_read_proto;
1427 	case BPF_FUNC_current_task_under_cgroup:
1428 		return &bpf_current_task_under_cgroup_proto;
1429 	case BPF_FUNC_get_prandom_u32:
1430 		return &bpf_get_prandom_u32_proto;
1431 	case BPF_FUNC_probe_write_user:
1432 		return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1433 		       NULL : bpf_get_probe_write_proto();
1434 	case BPF_FUNC_probe_read_user:
1435 		return &bpf_probe_read_user_proto;
1436 	case BPF_FUNC_probe_read_kernel:
1437 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1438 		       NULL : &bpf_probe_read_kernel_proto;
1439 	case BPF_FUNC_probe_read_user_str:
1440 		return &bpf_probe_read_user_str_proto;
1441 	case BPF_FUNC_probe_read_kernel_str:
1442 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1443 		       NULL : &bpf_probe_read_kernel_str_proto;
1444 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1445 	case BPF_FUNC_probe_read:
1446 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1447 		       NULL : &bpf_probe_read_compat_proto;
1448 	case BPF_FUNC_probe_read_str:
1449 		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1450 		       NULL : &bpf_probe_read_compat_str_proto;
1451 #endif
1452 #ifdef CONFIG_CGROUPS
1453 	case BPF_FUNC_get_current_cgroup_id:
1454 		return &bpf_get_current_cgroup_id_proto;
1455 	case BPF_FUNC_get_current_ancestor_cgroup_id:
1456 		return &bpf_get_current_ancestor_cgroup_id_proto;
1457 #endif
1458 	case BPF_FUNC_send_signal:
1459 		return &bpf_send_signal_proto;
1460 	case BPF_FUNC_send_signal_thread:
1461 		return &bpf_send_signal_thread_proto;
1462 	case BPF_FUNC_perf_event_read_value:
1463 		return &bpf_perf_event_read_value_proto;
1464 	case BPF_FUNC_get_ns_current_pid_tgid:
1465 		return &bpf_get_ns_current_pid_tgid_proto;
1466 	case BPF_FUNC_ringbuf_output:
1467 		return &bpf_ringbuf_output_proto;
1468 	case BPF_FUNC_ringbuf_reserve:
1469 		return &bpf_ringbuf_reserve_proto;
1470 	case BPF_FUNC_ringbuf_submit:
1471 		return &bpf_ringbuf_submit_proto;
1472 	case BPF_FUNC_ringbuf_discard:
1473 		return &bpf_ringbuf_discard_proto;
1474 	case BPF_FUNC_ringbuf_query:
1475 		return &bpf_ringbuf_query_proto;
1476 	case BPF_FUNC_jiffies64:
1477 		return &bpf_jiffies64_proto;
1478 	case BPF_FUNC_get_task_stack:
1479 		return &bpf_get_task_stack_proto;
1480 	case BPF_FUNC_copy_from_user:
1481 		return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1482 	case BPF_FUNC_copy_from_user_task:
1483 		return prog->aux->sleepable ? &bpf_copy_from_user_task_proto : NULL;
1484 	case BPF_FUNC_snprintf_btf:
1485 		return &bpf_snprintf_btf_proto;
1486 	case BPF_FUNC_per_cpu_ptr:
1487 		return &bpf_per_cpu_ptr_proto;
1488 	case BPF_FUNC_this_cpu_ptr:
1489 		return &bpf_this_cpu_ptr_proto;
1490 	case BPF_FUNC_task_storage_get:
1491 		return &bpf_task_storage_get_proto;
1492 	case BPF_FUNC_task_storage_delete:
1493 		return &bpf_task_storage_delete_proto;
1494 	case BPF_FUNC_for_each_map_elem:
1495 		return &bpf_for_each_map_elem_proto;
1496 	case BPF_FUNC_snprintf:
1497 		return &bpf_snprintf_proto;
1498 	case BPF_FUNC_get_func_ip:
1499 		return &bpf_get_func_ip_proto_tracing;
1500 	case BPF_FUNC_get_branch_snapshot:
1501 		return &bpf_get_branch_snapshot_proto;
1502 	case BPF_FUNC_find_vma:
1503 		return &bpf_find_vma_proto;
1504 	case BPF_FUNC_trace_vprintk:
1505 		return bpf_get_trace_vprintk_proto();
1506 	default:
1507 		return bpf_base_func_proto(func_id);
1508 	}
1509 }
1510 
1511 static const struct bpf_func_proto *
1512 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1513 {
1514 	switch (func_id) {
1515 	case BPF_FUNC_perf_event_output:
1516 		return &bpf_perf_event_output_proto;
1517 	case BPF_FUNC_get_stackid:
1518 		return &bpf_get_stackid_proto;
1519 	case BPF_FUNC_get_stack:
1520 		return &bpf_get_stack_proto;
1521 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1522 	case BPF_FUNC_override_return:
1523 		return &bpf_override_return_proto;
1524 #endif
1525 	case BPF_FUNC_get_func_ip:
1526 		return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1527 			&bpf_get_func_ip_proto_kprobe_multi :
1528 			&bpf_get_func_ip_proto_kprobe;
1529 	case BPF_FUNC_get_attach_cookie:
1530 		return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1531 			&bpf_get_attach_cookie_proto_kmulti :
1532 			&bpf_get_attach_cookie_proto_trace;
1533 	default:
1534 		return bpf_tracing_func_proto(func_id, prog);
1535 	}
1536 }
1537 
1538 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1539 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1540 					const struct bpf_prog *prog,
1541 					struct bpf_insn_access_aux *info)
1542 {
1543 	if (off < 0 || off >= sizeof(struct pt_regs))
1544 		return false;
1545 	if (type != BPF_READ)
1546 		return false;
1547 	if (off % size != 0)
1548 		return false;
1549 	/*
1550 	 * Assertion for 32 bit to make sure last 8 byte access
1551 	 * (BPF_DW) to the last 4 byte member is disallowed.
1552 	 */
1553 	if (off + size > sizeof(struct pt_regs))
1554 		return false;
1555 
1556 	return true;
1557 }
1558 
1559 const struct bpf_verifier_ops kprobe_verifier_ops = {
1560 	.get_func_proto  = kprobe_prog_func_proto,
1561 	.is_valid_access = kprobe_prog_is_valid_access,
1562 };
1563 
1564 const struct bpf_prog_ops kprobe_prog_ops = {
1565 };
1566 
1567 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1568 	   u64, flags, void *, data, u64, size)
1569 {
1570 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1571 
1572 	/*
1573 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1574 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1575 	 * from there and call the same bpf_perf_event_output() helper inline.
1576 	 */
1577 	return ____bpf_perf_event_output(regs, map, flags, data, size);
1578 }
1579 
1580 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1581 	.func		= bpf_perf_event_output_tp,
1582 	.gpl_only	= true,
1583 	.ret_type	= RET_INTEGER,
1584 	.arg1_type	= ARG_PTR_TO_CTX,
1585 	.arg2_type	= ARG_CONST_MAP_PTR,
1586 	.arg3_type	= ARG_ANYTHING,
1587 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1588 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1589 };
1590 
1591 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1592 	   u64, flags)
1593 {
1594 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1595 
1596 	/*
1597 	 * Same comment as in bpf_perf_event_output_tp(), only that this time
1598 	 * the other helper's function body cannot be inlined due to being
1599 	 * external, thus we need to call raw helper function.
1600 	 */
1601 	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1602 			       flags, 0, 0);
1603 }
1604 
1605 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1606 	.func		= bpf_get_stackid_tp,
1607 	.gpl_only	= true,
1608 	.ret_type	= RET_INTEGER,
1609 	.arg1_type	= ARG_PTR_TO_CTX,
1610 	.arg2_type	= ARG_CONST_MAP_PTR,
1611 	.arg3_type	= ARG_ANYTHING,
1612 };
1613 
1614 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1615 	   u64, flags)
1616 {
1617 	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1618 
1619 	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1620 			     (unsigned long) size, flags, 0);
1621 }
1622 
1623 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1624 	.func		= bpf_get_stack_tp,
1625 	.gpl_only	= true,
1626 	.ret_type	= RET_INTEGER,
1627 	.arg1_type	= ARG_PTR_TO_CTX,
1628 	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
1629 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1630 	.arg4_type	= ARG_ANYTHING,
1631 };
1632 
1633 static const struct bpf_func_proto *
1634 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1635 {
1636 	switch (func_id) {
1637 	case BPF_FUNC_perf_event_output:
1638 		return &bpf_perf_event_output_proto_tp;
1639 	case BPF_FUNC_get_stackid:
1640 		return &bpf_get_stackid_proto_tp;
1641 	case BPF_FUNC_get_stack:
1642 		return &bpf_get_stack_proto_tp;
1643 	case BPF_FUNC_get_attach_cookie:
1644 		return &bpf_get_attach_cookie_proto_trace;
1645 	default:
1646 		return bpf_tracing_func_proto(func_id, prog);
1647 	}
1648 }
1649 
1650 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1651 				    const struct bpf_prog *prog,
1652 				    struct bpf_insn_access_aux *info)
1653 {
1654 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1655 		return false;
1656 	if (type != BPF_READ)
1657 		return false;
1658 	if (off % size != 0)
1659 		return false;
1660 
1661 	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1662 	return true;
1663 }
1664 
1665 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1666 	.get_func_proto  = tp_prog_func_proto,
1667 	.is_valid_access = tp_prog_is_valid_access,
1668 };
1669 
1670 const struct bpf_prog_ops tracepoint_prog_ops = {
1671 };
1672 
1673 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1674 	   struct bpf_perf_event_value *, buf, u32, size)
1675 {
1676 	int err = -EINVAL;
1677 
1678 	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1679 		goto clear;
1680 	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1681 				    &buf->running);
1682 	if (unlikely(err))
1683 		goto clear;
1684 	return 0;
1685 clear:
1686 	memset(buf, 0, size);
1687 	return err;
1688 }
1689 
1690 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1691          .func           = bpf_perf_prog_read_value,
1692          .gpl_only       = true,
1693          .ret_type       = RET_INTEGER,
1694          .arg1_type      = ARG_PTR_TO_CTX,
1695          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1696          .arg3_type      = ARG_CONST_SIZE,
1697 };
1698 
1699 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1700 	   void *, buf, u32, size, u64, flags)
1701 {
1702 	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1703 	struct perf_branch_stack *br_stack = ctx->data->br_stack;
1704 	u32 to_copy;
1705 
1706 	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1707 		return -EINVAL;
1708 
1709 	if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1710 		return -ENOENT;
1711 
1712 	if (unlikely(!br_stack))
1713 		return -ENOENT;
1714 
1715 	if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1716 		return br_stack->nr * br_entry_size;
1717 
1718 	if (!buf || (size % br_entry_size != 0))
1719 		return -EINVAL;
1720 
1721 	to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1722 	memcpy(buf, br_stack->entries, to_copy);
1723 
1724 	return to_copy;
1725 }
1726 
1727 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1728 	.func           = bpf_read_branch_records,
1729 	.gpl_only       = true,
1730 	.ret_type       = RET_INTEGER,
1731 	.arg1_type      = ARG_PTR_TO_CTX,
1732 	.arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1733 	.arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1734 	.arg4_type      = ARG_ANYTHING,
1735 };
1736 
1737 static const struct bpf_func_proto *
1738 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1739 {
1740 	switch (func_id) {
1741 	case BPF_FUNC_perf_event_output:
1742 		return &bpf_perf_event_output_proto_tp;
1743 	case BPF_FUNC_get_stackid:
1744 		return &bpf_get_stackid_proto_pe;
1745 	case BPF_FUNC_get_stack:
1746 		return &bpf_get_stack_proto_pe;
1747 	case BPF_FUNC_perf_prog_read_value:
1748 		return &bpf_perf_prog_read_value_proto;
1749 	case BPF_FUNC_read_branch_records:
1750 		return &bpf_read_branch_records_proto;
1751 	case BPF_FUNC_get_attach_cookie:
1752 		return &bpf_get_attach_cookie_proto_pe;
1753 	default:
1754 		return bpf_tracing_func_proto(func_id, prog);
1755 	}
1756 }
1757 
1758 /*
1759  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1760  * to avoid potential recursive reuse issue when/if tracepoints are added
1761  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1762  *
1763  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1764  * in normal, irq, and nmi context.
1765  */
1766 struct bpf_raw_tp_regs {
1767 	struct pt_regs regs[3];
1768 };
1769 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1770 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1771 static struct pt_regs *get_bpf_raw_tp_regs(void)
1772 {
1773 	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1774 	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1775 
1776 	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1777 		this_cpu_dec(bpf_raw_tp_nest_level);
1778 		return ERR_PTR(-EBUSY);
1779 	}
1780 
1781 	return &tp_regs->regs[nest_level - 1];
1782 }
1783 
1784 static void put_bpf_raw_tp_regs(void)
1785 {
1786 	this_cpu_dec(bpf_raw_tp_nest_level);
1787 }
1788 
1789 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1790 	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1791 {
1792 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1793 	int ret;
1794 
1795 	if (IS_ERR(regs))
1796 		return PTR_ERR(regs);
1797 
1798 	perf_fetch_caller_regs(regs);
1799 	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1800 
1801 	put_bpf_raw_tp_regs();
1802 	return ret;
1803 }
1804 
1805 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1806 	.func		= bpf_perf_event_output_raw_tp,
1807 	.gpl_only	= true,
1808 	.ret_type	= RET_INTEGER,
1809 	.arg1_type	= ARG_PTR_TO_CTX,
1810 	.arg2_type	= ARG_CONST_MAP_PTR,
1811 	.arg3_type	= ARG_ANYTHING,
1812 	.arg4_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1813 	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1814 };
1815 
1816 extern const struct bpf_func_proto bpf_skb_output_proto;
1817 extern const struct bpf_func_proto bpf_xdp_output_proto;
1818 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1819 
1820 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1821 	   struct bpf_map *, map, u64, flags)
1822 {
1823 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1824 	int ret;
1825 
1826 	if (IS_ERR(regs))
1827 		return PTR_ERR(regs);
1828 
1829 	perf_fetch_caller_regs(regs);
1830 	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1831 	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1832 			      flags, 0, 0);
1833 	put_bpf_raw_tp_regs();
1834 	return ret;
1835 }
1836 
1837 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1838 	.func		= bpf_get_stackid_raw_tp,
1839 	.gpl_only	= true,
1840 	.ret_type	= RET_INTEGER,
1841 	.arg1_type	= ARG_PTR_TO_CTX,
1842 	.arg2_type	= ARG_CONST_MAP_PTR,
1843 	.arg3_type	= ARG_ANYTHING,
1844 };
1845 
1846 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1847 	   void *, buf, u32, size, u64, flags)
1848 {
1849 	struct pt_regs *regs = get_bpf_raw_tp_regs();
1850 	int ret;
1851 
1852 	if (IS_ERR(regs))
1853 		return PTR_ERR(regs);
1854 
1855 	perf_fetch_caller_regs(regs);
1856 	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1857 			    (unsigned long) size, flags, 0);
1858 	put_bpf_raw_tp_regs();
1859 	return ret;
1860 }
1861 
1862 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1863 	.func		= bpf_get_stack_raw_tp,
1864 	.gpl_only	= true,
1865 	.ret_type	= RET_INTEGER,
1866 	.arg1_type	= ARG_PTR_TO_CTX,
1867 	.arg2_type	= ARG_PTR_TO_MEM | MEM_RDONLY,
1868 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1869 	.arg4_type	= ARG_ANYTHING,
1870 };
1871 
1872 static const struct bpf_func_proto *
1873 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1874 {
1875 	switch (func_id) {
1876 	case BPF_FUNC_perf_event_output:
1877 		return &bpf_perf_event_output_proto_raw_tp;
1878 	case BPF_FUNC_get_stackid:
1879 		return &bpf_get_stackid_proto_raw_tp;
1880 	case BPF_FUNC_get_stack:
1881 		return &bpf_get_stack_proto_raw_tp;
1882 	default:
1883 		return bpf_tracing_func_proto(func_id, prog);
1884 	}
1885 }
1886 
1887 const struct bpf_func_proto *
1888 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1889 {
1890 	const struct bpf_func_proto *fn;
1891 
1892 	switch (func_id) {
1893 #ifdef CONFIG_NET
1894 	case BPF_FUNC_skb_output:
1895 		return &bpf_skb_output_proto;
1896 	case BPF_FUNC_xdp_output:
1897 		return &bpf_xdp_output_proto;
1898 	case BPF_FUNC_skc_to_tcp6_sock:
1899 		return &bpf_skc_to_tcp6_sock_proto;
1900 	case BPF_FUNC_skc_to_tcp_sock:
1901 		return &bpf_skc_to_tcp_sock_proto;
1902 	case BPF_FUNC_skc_to_tcp_timewait_sock:
1903 		return &bpf_skc_to_tcp_timewait_sock_proto;
1904 	case BPF_FUNC_skc_to_tcp_request_sock:
1905 		return &bpf_skc_to_tcp_request_sock_proto;
1906 	case BPF_FUNC_skc_to_udp6_sock:
1907 		return &bpf_skc_to_udp6_sock_proto;
1908 	case BPF_FUNC_skc_to_unix_sock:
1909 		return &bpf_skc_to_unix_sock_proto;
1910 	case BPF_FUNC_skc_to_mptcp_sock:
1911 		return &bpf_skc_to_mptcp_sock_proto;
1912 	case BPF_FUNC_sk_storage_get:
1913 		return &bpf_sk_storage_get_tracing_proto;
1914 	case BPF_FUNC_sk_storage_delete:
1915 		return &bpf_sk_storage_delete_tracing_proto;
1916 	case BPF_FUNC_sock_from_file:
1917 		return &bpf_sock_from_file_proto;
1918 	case BPF_FUNC_get_socket_cookie:
1919 		return &bpf_get_socket_ptr_cookie_proto;
1920 	case BPF_FUNC_xdp_get_buff_len:
1921 		return &bpf_xdp_get_buff_len_trace_proto;
1922 #endif
1923 	case BPF_FUNC_seq_printf:
1924 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1925 		       &bpf_seq_printf_proto :
1926 		       NULL;
1927 	case BPF_FUNC_seq_write:
1928 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1929 		       &bpf_seq_write_proto :
1930 		       NULL;
1931 	case BPF_FUNC_seq_printf_btf:
1932 		return prog->expected_attach_type == BPF_TRACE_ITER ?
1933 		       &bpf_seq_printf_btf_proto :
1934 		       NULL;
1935 	case BPF_FUNC_d_path:
1936 		return &bpf_d_path_proto;
1937 	case BPF_FUNC_get_func_arg:
1938 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1939 	case BPF_FUNC_get_func_ret:
1940 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1941 	case BPF_FUNC_get_func_arg_cnt:
1942 		return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1943 	case BPF_FUNC_get_attach_cookie:
1944 		return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1945 	default:
1946 		fn = raw_tp_prog_func_proto(func_id, prog);
1947 		if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1948 			fn = bpf_iter_get_func_proto(func_id, prog);
1949 		return fn;
1950 	}
1951 }
1952 
1953 static bool raw_tp_prog_is_valid_access(int off, int size,
1954 					enum bpf_access_type type,
1955 					const struct bpf_prog *prog,
1956 					struct bpf_insn_access_aux *info)
1957 {
1958 	return bpf_tracing_ctx_access(off, size, type);
1959 }
1960 
1961 static bool tracing_prog_is_valid_access(int off, int size,
1962 					 enum bpf_access_type type,
1963 					 const struct bpf_prog *prog,
1964 					 struct bpf_insn_access_aux *info)
1965 {
1966 	return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1967 }
1968 
1969 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1970 				     const union bpf_attr *kattr,
1971 				     union bpf_attr __user *uattr)
1972 {
1973 	return -ENOTSUPP;
1974 }
1975 
1976 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1977 	.get_func_proto  = raw_tp_prog_func_proto,
1978 	.is_valid_access = raw_tp_prog_is_valid_access,
1979 };
1980 
1981 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1982 #ifdef CONFIG_NET
1983 	.test_run = bpf_prog_test_run_raw_tp,
1984 #endif
1985 };
1986 
1987 const struct bpf_verifier_ops tracing_verifier_ops = {
1988 	.get_func_proto  = tracing_prog_func_proto,
1989 	.is_valid_access = tracing_prog_is_valid_access,
1990 };
1991 
1992 const struct bpf_prog_ops tracing_prog_ops = {
1993 	.test_run = bpf_prog_test_run_tracing,
1994 };
1995 
1996 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1997 						 enum bpf_access_type type,
1998 						 const struct bpf_prog *prog,
1999 						 struct bpf_insn_access_aux *info)
2000 {
2001 	if (off == 0) {
2002 		if (size != sizeof(u64) || type != BPF_READ)
2003 			return false;
2004 		info->reg_type = PTR_TO_TP_BUFFER;
2005 	}
2006 	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2007 }
2008 
2009 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2010 	.get_func_proto  = raw_tp_prog_func_proto,
2011 	.is_valid_access = raw_tp_writable_prog_is_valid_access,
2012 };
2013 
2014 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2015 };
2016 
2017 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2018 				    const struct bpf_prog *prog,
2019 				    struct bpf_insn_access_aux *info)
2020 {
2021 	const int size_u64 = sizeof(u64);
2022 
2023 	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2024 		return false;
2025 	if (type != BPF_READ)
2026 		return false;
2027 	if (off % size != 0) {
2028 		if (sizeof(unsigned long) != 4)
2029 			return false;
2030 		if (size != 8)
2031 			return false;
2032 		if (off % size != 4)
2033 			return false;
2034 	}
2035 
2036 	switch (off) {
2037 	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2038 		bpf_ctx_record_field_size(info, size_u64);
2039 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2040 			return false;
2041 		break;
2042 	case bpf_ctx_range(struct bpf_perf_event_data, addr):
2043 		bpf_ctx_record_field_size(info, size_u64);
2044 		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2045 			return false;
2046 		break;
2047 	default:
2048 		if (size != sizeof(long))
2049 			return false;
2050 	}
2051 
2052 	return true;
2053 }
2054 
2055 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2056 				      const struct bpf_insn *si,
2057 				      struct bpf_insn *insn_buf,
2058 				      struct bpf_prog *prog, u32 *target_size)
2059 {
2060 	struct bpf_insn *insn = insn_buf;
2061 
2062 	switch (si->off) {
2063 	case offsetof(struct bpf_perf_event_data, sample_period):
2064 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2065 						       data), si->dst_reg, si->src_reg,
2066 				      offsetof(struct bpf_perf_event_data_kern, data));
2067 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2068 				      bpf_target_off(struct perf_sample_data, period, 8,
2069 						     target_size));
2070 		break;
2071 	case offsetof(struct bpf_perf_event_data, addr):
2072 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2073 						       data), si->dst_reg, si->src_reg,
2074 				      offsetof(struct bpf_perf_event_data_kern, data));
2075 		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2076 				      bpf_target_off(struct perf_sample_data, addr, 8,
2077 						     target_size));
2078 		break;
2079 	default:
2080 		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2081 						       regs), si->dst_reg, si->src_reg,
2082 				      offsetof(struct bpf_perf_event_data_kern, regs));
2083 		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2084 				      si->off);
2085 		break;
2086 	}
2087 
2088 	return insn - insn_buf;
2089 }
2090 
2091 const struct bpf_verifier_ops perf_event_verifier_ops = {
2092 	.get_func_proto		= pe_prog_func_proto,
2093 	.is_valid_access	= pe_prog_is_valid_access,
2094 	.convert_ctx_access	= pe_prog_convert_ctx_access,
2095 };
2096 
2097 const struct bpf_prog_ops perf_event_prog_ops = {
2098 };
2099 
2100 static DEFINE_MUTEX(bpf_event_mutex);
2101 
2102 #define BPF_TRACE_MAX_PROGS 64
2103 
2104 int perf_event_attach_bpf_prog(struct perf_event *event,
2105 			       struct bpf_prog *prog,
2106 			       u64 bpf_cookie)
2107 {
2108 	struct bpf_prog_array *old_array;
2109 	struct bpf_prog_array *new_array;
2110 	int ret = -EEXIST;
2111 
2112 	/*
2113 	 * Kprobe override only works if they are on the function entry,
2114 	 * and only if they are on the opt-in list.
2115 	 */
2116 	if (prog->kprobe_override &&
2117 	    (!trace_kprobe_on_func_entry(event->tp_event) ||
2118 	     !trace_kprobe_error_injectable(event->tp_event)))
2119 		return -EINVAL;
2120 
2121 	mutex_lock(&bpf_event_mutex);
2122 
2123 	if (event->prog)
2124 		goto unlock;
2125 
2126 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2127 	if (old_array &&
2128 	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2129 		ret = -E2BIG;
2130 		goto unlock;
2131 	}
2132 
2133 	ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2134 	if (ret < 0)
2135 		goto unlock;
2136 
2137 	/* set the new array to event->tp_event and set event->prog */
2138 	event->prog = prog;
2139 	event->bpf_cookie = bpf_cookie;
2140 	rcu_assign_pointer(event->tp_event->prog_array, new_array);
2141 	bpf_prog_array_free_sleepable(old_array);
2142 
2143 unlock:
2144 	mutex_unlock(&bpf_event_mutex);
2145 	return ret;
2146 }
2147 
2148 void perf_event_detach_bpf_prog(struct perf_event *event)
2149 {
2150 	struct bpf_prog_array *old_array;
2151 	struct bpf_prog_array *new_array;
2152 	int ret;
2153 
2154 	mutex_lock(&bpf_event_mutex);
2155 
2156 	if (!event->prog)
2157 		goto unlock;
2158 
2159 	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2160 	ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2161 	if (ret == -ENOENT)
2162 		goto unlock;
2163 	if (ret < 0) {
2164 		bpf_prog_array_delete_safe(old_array, event->prog);
2165 	} else {
2166 		rcu_assign_pointer(event->tp_event->prog_array, new_array);
2167 		bpf_prog_array_free_sleepable(old_array);
2168 	}
2169 
2170 	bpf_prog_put(event->prog);
2171 	event->prog = NULL;
2172 
2173 unlock:
2174 	mutex_unlock(&bpf_event_mutex);
2175 }
2176 
2177 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2178 {
2179 	struct perf_event_query_bpf __user *uquery = info;
2180 	struct perf_event_query_bpf query = {};
2181 	struct bpf_prog_array *progs;
2182 	u32 *ids, prog_cnt, ids_len;
2183 	int ret;
2184 
2185 	if (!perfmon_capable())
2186 		return -EPERM;
2187 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
2188 		return -EINVAL;
2189 	if (copy_from_user(&query, uquery, sizeof(query)))
2190 		return -EFAULT;
2191 
2192 	ids_len = query.ids_len;
2193 	if (ids_len > BPF_TRACE_MAX_PROGS)
2194 		return -E2BIG;
2195 	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2196 	if (!ids)
2197 		return -ENOMEM;
2198 	/*
2199 	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2200 	 * is required when user only wants to check for uquery->prog_cnt.
2201 	 * There is no need to check for it since the case is handled
2202 	 * gracefully in bpf_prog_array_copy_info.
2203 	 */
2204 
2205 	mutex_lock(&bpf_event_mutex);
2206 	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2207 	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2208 	mutex_unlock(&bpf_event_mutex);
2209 
2210 	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2211 	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2212 		ret = -EFAULT;
2213 
2214 	kfree(ids);
2215 	return ret;
2216 }
2217 
2218 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2219 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2220 
2221 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2222 {
2223 	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2224 
2225 	for (; btp < __stop__bpf_raw_tp; btp++) {
2226 		if (!strcmp(btp->tp->name, name))
2227 			return btp;
2228 	}
2229 
2230 	return bpf_get_raw_tracepoint_module(name);
2231 }
2232 
2233 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2234 {
2235 	struct module *mod;
2236 
2237 	preempt_disable();
2238 	mod = __module_address((unsigned long)btp);
2239 	module_put(mod);
2240 	preempt_enable();
2241 }
2242 
2243 static __always_inline
2244 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2245 {
2246 	cant_sleep();
2247 	if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2248 		bpf_prog_inc_misses_counter(prog);
2249 		goto out;
2250 	}
2251 	rcu_read_lock();
2252 	(void) bpf_prog_run(prog, args);
2253 	rcu_read_unlock();
2254 out:
2255 	this_cpu_dec(*(prog->active));
2256 }
2257 
2258 #define UNPACK(...)			__VA_ARGS__
2259 #define REPEAT_1(FN, DL, X, ...)	FN(X)
2260 #define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2261 #define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2262 #define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2263 #define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2264 #define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2265 #define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2266 #define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2267 #define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2268 #define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2269 #define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2270 #define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2271 #define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
2272 
2273 #define SARG(X)		u64 arg##X
2274 #define COPY(X)		args[X] = arg##X
2275 
2276 #define __DL_COM	(,)
2277 #define __DL_SEM	(;)
2278 
2279 #define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2280 
2281 #define BPF_TRACE_DEFN_x(x)						\
2282 	void bpf_trace_run##x(struct bpf_prog *prog,			\
2283 			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
2284 	{								\
2285 		u64 args[x];						\
2286 		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
2287 		__bpf_trace_run(prog, args);				\
2288 	}								\
2289 	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2290 BPF_TRACE_DEFN_x(1);
2291 BPF_TRACE_DEFN_x(2);
2292 BPF_TRACE_DEFN_x(3);
2293 BPF_TRACE_DEFN_x(4);
2294 BPF_TRACE_DEFN_x(5);
2295 BPF_TRACE_DEFN_x(6);
2296 BPF_TRACE_DEFN_x(7);
2297 BPF_TRACE_DEFN_x(8);
2298 BPF_TRACE_DEFN_x(9);
2299 BPF_TRACE_DEFN_x(10);
2300 BPF_TRACE_DEFN_x(11);
2301 BPF_TRACE_DEFN_x(12);
2302 
2303 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2304 {
2305 	struct tracepoint *tp = btp->tp;
2306 
2307 	/*
2308 	 * check that program doesn't access arguments beyond what's
2309 	 * available in this tracepoint
2310 	 */
2311 	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2312 		return -EINVAL;
2313 
2314 	if (prog->aux->max_tp_access > btp->writable_size)
2315 		return -EINVAL;
2316 
2317 	return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2318 						   prog);
2319 }
2320 
2321 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2322 {
2323 	return __bpf_probe_register(btp, prog);
2324 }
2325 
2326 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2327 {
2328 	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2329 }
2330 
2331 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2332 			    u32 *fd_type, const char **buf,
2333 			    u64 *probe_offset, u64 *probe_addr)
2334 {
2335 	bool is_tracepoint, is_syscall_tp;
2336 	struct bpf_prog *prog;
2337 	int flags, err = 0;
2338 
2339 	prog = event->prog;
2340 	if (!prog)
2341 		return -ENOENT;
2342 
2343 	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2344 	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2345 		return -EOPNOTSUPP;
2346 
2347 	*prog_id = prog->aux->id;
2348 	flags = event->tp_event->flags;
2349 	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2350 	is_syscall_tp = is_syscall_trace_event(event->tp_event);
2351 
2352 	if (is_tracepoint || is_syscall_tp) {
2353 		*buf = is_tracepoint ? event->tp_event->tp->name
2354 				     : event->tp_event->name;
2355 		*fd_type = BPF_FD_TYPE_TRACEPOINT;
2356 		*probe_offset = 0x0;
2357 		*probe_addr = 0x0;
2358 	} else {
2359 		/* kprobe/uprobe */
2360 		err = -EOPNOTSUPP;
2361 #ifdef CONFIG_KPROBE_EVENTS
2362 		if (flags & TRACE_EVENT_FL_KPROBE)
2363 			err = bpf_get_kprobe_info(event, fd_type, buf,
2364 						  probe_offset, probe_addr,
2365 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2366 #endif
2367 #ifdef CONFIG_UPROBE_EVENTS
2368 		if (flags & TRACE_EVENT_FL_UPROBE)
2369 			err = bpf_get_uprobe_info(event, fd_type, buf,
2370 						  probe_offset,
2371 						  event->attr.type == PERF_TYPE_TRACEPOINT);
2372 #endif
2373 	}
2374 
2375 	return err;
2376 }
2377 
2378 static int __init send_signal_irq_work_init(void)
2379 {
2380 	int cpu;
2381 	struct send_signal_irq_work *work;
2382 
2383 	for_each_possible_cpu(cpu) {
2384 		work = per_cpu_ptr(&send_signal_work, cpu);
2385 		init_irq_work(&work->irq_work, do_bpf_send_signal);
2386 	}
2387 	return 0;
2388 }
2389 
2390 subsys_initcall(send_signal_irq_work_init);
2391 
2392 #ifdef CONFIG_MODULES
2393 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2394 			    void *module)
2395 {
2396 	struct bpf_trace_module *btm, *tmp;
2397 	struct module *mod = module;
2398 	int ret = 0;
2399 
2400 	if (mod->num_bpf_raw_events == 0 ||
2401 	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2402 		goto out;
2403 
2404 	mutex_lock(&bpf_module_mutex);
2405 
2406 	switch (op) {
2407 	case MODULE_STATE_COMING:
2408 		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2409 		if (btm) {
2410 			btm->module = module;
2411 			list_add(&btm->list, &bpf_trace_modules);
2412 		} else {
2413 			ret = -ENOMEM;
2414 		}
2415 		break;
2416 	case MODULE_STATE_GOING:
2417 		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2418 			if (btm->module == module) {
2419 				list_del(&btm->list);
2420 				kfree(btm);
2421 				break;
2422 			}
2423 		}
2424 		break;
2425 	}
2426 
2427 	mutex_unlock(&bpf_module_mutex);
2428 
2429 out:
2430 	return notifier_from_errno(ret);
2431 }
2432 
2433 static struct notifier_block bpf_module_nb = {
2434 	.notifier_call = bpf_event_notify,
2435 };
2436 
2437 static int __init bpf_event_init(void)
2438 {
2439 	register_module_notifier(&bpf_module_nb);
2440 	return 0;
2441 }
2442 
2443 fs_initcall(bpf_event_init);
2444 #endif /* CONFIG_MODULES */
2445 
2446 #ifdef CONFIG_FPROBE
2447 struct bpf_kprobe_multi_link {
2448 	struct bpf_link link;
2449 	struct fprobe fp;
2450 	unsigned long *addrs;
2451 	u64 *cookies;
2452 	u32 cnt;
2453 };
2454 
2455 struct bpf_kprobe_multi_run_ctx {
2456 	struct bpf_run_ctx run_ctx;
2457 	struct bpf_kprobe_multi_link *link;
2458 	unsigned long entry_ip;
2459 };
2460 
2461 struct user_syms {
2462 	const char **syms;
2463 	char *buf;
2464 };
2465 
2466 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2467 {
2468 	unsigned long __user usymbol;
2469 	const char **syms = NULL;
2470 	char *buf = NULL, *p;
2471 	int err = -ENOMEM;
2472 	unsigned int i;
2473 
2474 	syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2475 	if (!syms)
2476 		goto error;
2477 
2478 	buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2479 	if (!buf)
2480 		goto error;
2481 
2482 	for (p = buf, i = 0; i < cnt; i++) {
2483 		if (__get_user(usymbol, usyms + i)) {
2484 			err = -EFAULT;
2485 			goto error;
2486 		}
2487 		err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2488 		if (err == KSYM_NAME_LEN)
2489 			err = -E2BIG;
2490 		if (err < 0)
2491 			goto error;
2492 		syms[i] = p;
2493 		p += err + 1;
2494 	}
2495 
2496 	us->syms = syms;
2497 	us->buf = buf;
2498 	return 0;
2499 
2500 error:
2501 	if (err) {
2502 		kvfree(syms);
2503 		kvfree(buf);
2504 	}
2505 	return err;
2506 }
2507 
2508 static void free_user_syms(struct user_syms *us)
2509 {
2510 	kvfree(us->syms);
2511 	kvfree(us->buf);
2512 }
2513 
2514 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2515 {
2516 	struct bpf_kprobe_multi_link *kmulti_link;
2517 
2518 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2519 	unregister_fprobe(&kmulti_link->fp);
2520 }
2521 
2522 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2523 {
2524 	struct bpf_kprobe_multi_link *kmulti_link;
2525 
2526 	kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2527 	kvfree(kmulti_link->addrs);
2528 	kvfree(kmulti_link->cookies);
2529 	kfree(kmulti_link);
2530 }
2531 
2532 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2533 	.release = bpf_kprobe_multi_link_release,
2534 	.dealloc = bpf_kprobe_multi_link_dealloc,
2535 };
2536 
2537 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2538 {
2539 	const struct bpf_kprobe_multi_link *link = priv;
2540 	unsigned long *addr_a = a, *addr_b = b;
2541 	u64 *cookie_a, *cookie_b;
2542 
2543 	cookie_a = link->cookies + (addr_a - link->addrs);
2544 	cookie_b = link->cookies + (addr_b - link->addrs);
2545 
2546 	/* swap addr_a/addr_b and cookie_a/cookie_b values */
2547 	swap(*addr_a, *addr_b);
2548 	swap(*cookie_a, *cookie_b);
2549 }
2550 
2551 static int __bpf_kprobe_multi_cookie_cmp(const void *a, const void *b)
2552 {
2553 	const unsigned long *addr_a = a, *addr_b = b;
2554 
2555 	if (*addr_a == *addr_b)
2556 		return 0;
2557 	return *addr_a < *addr_b ? -1 : 1;
2558 }
2559 
2560 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2561 {
2562 	return __bpf_kprobe_multi_cookie_cmp(a, b);
2563 }
2564 
2565 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2566 {
2567 	struct bpf_kprobe_multi_run_ctx *run_ctx;
2568 	struct bpf_kprobe_multi_link *link;
2569 	u64 *cookie, entry_ip;
2570 	unsigned long *addr;
2571 
2572 	if (WARN_ON_ONCE(!ctx))
2573 		return 0;
2574 	run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2575 	link = run_ctx->link;
2576 	if (!link->cookies)
2577 		return 0;
2578 	entry_ip = run_ctx->entry_ip;
2579 	addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2580 		       __bpf_kprobe_multi_cookie_cmp);
2581 	if (!addr)
2582 		return 0;
2583 	cookie = link->cookies + (addr - link->addrs);
2584 	return *cookie;
2585 }
2586 
2587 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2588 {
2589 	struct bpf_kprobe_multi_run_ctx *run_ctx;
2590 
2591 	run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2592 	return run_ctx->entry_ip;
2593 }
2594 
2595 static int
2596 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2597 			   unsigned long entry_ip, struct pt_regs *regs)
2598 {
2599 	struct bpf_kprobe_multi_run_ctx run_ctx = {
2600 		.link = link,
2601 		.entry_ip = entry_ip,
2602 	};
2603 	struct bpf_run_ctx *old_run_ctx;
2604 	int err;
2605 
2606 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2607 		err = 0;
2608 		goto out;
2609 	}
2610 
2611 	migrate_disable();
2612 	rcu_read_lock();
2613 	old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2614 	err = bpf_prog_run(link->link.prog, regs);
2615 	bpf_reset_run_ctx(old_run_ctx);
2616 	rcu_read_unlock();
2617 	migrate_enable();
2618 
2619  out:
2620 	__this_cpu_dec(bpf_prog_active);
2621 	return err;
2622 }
2623 
2624 static void
2625 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2626 			  struct pt_regs *regs)
2627 {
2628 	struct bpf_kprobe_multi_link *link;
2629 
2630 	link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2631 	kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2632 }
2633 
2634 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2635 {
2636 	const char **str_a = (const char **) a;
2637 	const char **str_b = (const char **) b;
2638 
2639 	return strcmp(*str_a, *str_b);
2640 }
2641 
2642 struct multi_symbols_sort {
2643 	const char **funcs;
2644 	u64 *cookies;
2645 };
2646 
2647 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2648 {
2649 	const struct multi_symbols_sort *data = priv;
2650 	const char **name_a = a, **name_b = b;
2651 
2652 	swap(*name_a, *name_b);
2653 
2654 	/* If defined, swap also related cookies. */
2655 	if (data->cookies) {
2656 		u64 *cookie_a, *cookie_b;
2657 
2658 		cookie_a = data->cookies + (name_a - data->funcs);
2659 		cookie_b = data->cookies + (name_b - data->funcs);
2660 		swap(*cookie_a, *cookie_b);
2661 	}
2662 }
2663 
2664 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2665 {
2666 	struct bpf_kprobe_multi_link *link = NULL;
2667 	struct bpf_link_primer link_primer;
2668 	void __user *ucookies;
2669 	unsigned long *addrs;
2670 	u32 flags, cnt, size;
2671 	void __user *uaddrs;
2672 	u64 *cookies = NULL;
2673 	void __user *usyms;
2674 	int err;
2675 
2676 	/* no support for 32bit archs yet */
2677 	if (sizeof(u64) != sizeof(void *))
2678 		return -EOPNOTSUPP;
2679 
2680 	if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2681 		return -EINVAL;
2682 
2683 	flags = attr->link_create.kprobe_multi.flags;
2684 	if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2685 		return -EINVAL;
2686 
2687 	uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2688 	usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2689 	if (!!uaddrs == !!usyms)
2690 		return -EINVAL;
2691 
2692 	cnt = attr->link_create.kprobe_multi.cnt;
2693 	if (!cnt)
2694 		return -EINVAL;
2695 
2696 	size = cnt * sizeof(*addrs);
2697 	addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2698 	if (!addrs)
2699 		return -ENOMEM;
2700 
2701 	ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2702 	if (ucookies) {
2703 		cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2704 		if (!cookies) {
2705 			err = -ENOMEM;
2706 			goto error;
2707 		}
2708 		if (copy_from_user(cookies, ucookies, size)) {
2709 			err = -EFAULT;
2710 			goto error;
2711 		}
2712 	}
2713 
2714 	if (uaddrs) {
2715 		if (copy_from_user(addrs, uaddrs, size)) {
2716 			err = -EFAULT;
2717 			goto error;
2718 		}
2719 	} else {
2720 		struct multi_symbols_sort data = {
2721 			.cookies = cookies,
2722 		};
2723 		struct user_syms us;
2724 
2725 		err = copy_user_syms(&us, usyms, cnt);
2726 		if (err)
2727 			goto error;
2728 
2729 		if (cookies)
2730 			data.funcs = us.syms;
2731 
2732 		sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2733 		       symbols_swap_r, &data);
2734 
2735 		err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2736 		free_user_syms(&us);
2737 		if (err)
2738 			goto error;
2739 	}
2740 
2741 	link = kzalloc(sizeof(*link), GFP_KERNEL);
2742 	if (!link) {
2743 		err = -ENOMEM;
2744 		goto error;
2745 	}
2746 
2747 	bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2748 		      &bpf_kprobe_multi_link_lops, prog);
2749 
2750 	err = bpf_link_prime(&link->link, &link_primer);
2751 	if (err)
2752 		goto error;
2753 
2754 	if (flags & BPF_F_KPROBE_MULTI_RETURN)
2755 		link->fp.exit_handler = kprobe_multi_link_handler;
2756 	else
2757 		link->fp.entry_handler = kprobe_multi_link_handler;
2758 
2759 	link->addrs = addrs;
2760 	link->cookies = cookies;
2761 	link->cnt = cnt;
2762 
2763 	if (cookies) {
2764 		/*
2765 		 * Sorting addresses will trigger sorting cookies as well
2766 		 * (check bpf_kprobe_multi_cookie_swap). This way we can
2767 		 * find cookie based on the address in bpf_get_attach_cookie
2768 		 * helper.
2769 		 */
2770 		sort_r(addrs, cnt, sizeof(*addrs),
2771 		       bpf_kprobe_multi_cookie_cmp,
2772 		       bpf_kprobe_multi_cookie_swap,
2773 		       link);
2774 	}
2775 
2776 	err = register_fprobe_ips(&link->fp, addrs, cnt);
2777 	if (err) {
2778 		bpf_link_cleanup(&link_primer);
2779 		return err;
2780 	}
2781 
2782 	return bpf_link_settle(&link_primer);
2783 
2784 error:
2785 	kfree(link);
2786 	kvfree(addrs);
2787 	kvfree(cookies);
2788 	return err;
2789 }
2790 #else /* !CONFIG_FPROBE */
2791 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2792 {
2793 	return -EOPNOTSUPP;
2794 }
2795 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2796 {
2797 	return 0;
2798 }
2799 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2800 {
2801 	return 0;
2802 }
2803 #endif
2804