Loading...
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
27#include <net/bpf_sk_storage.h>
28
29#include <uapi/linux/bpf.h>
30#include <uapi/linux/btf.h>
31
32#include <asm/tlb.h>
33
34#include "trace_probe.h"
35#include "trace.h"
36
37#define CREATE_TRACE_POINTS
38#include "bpf_trace.h"
39
40#define bpf_event_rcu_dereference(p) \
41 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
42
43#ifdef CONFIG_MODULES
44struct bpf_trace_module {
45 struct module *module;
46 struct list_head list;
47};
48
49static LIST_HEAD(bpf_trace_modules);
50static DEFINE_MUTEX(bpf_module_mutex);
51
52static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
53{
54 struct bpf_raw_event_map *btp, *ret = NULL;
55 struct bpf_trace_module *btm;
56 unsigned int i;
57
58 mutex_lock(&bpf_module_mutex);
59 list_for_each_entry(btm, &bpf_trace_modules, list) {
60 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
61 btp = &btm->module->bpf_raw_events[i];
62 if (!strcmp(btp->tp->name, name)) {
63 if (try_module_get(btm->module))
64 ret = btp;
65 goto out;
66 }
67 }
68 }
69out:
70 mutex_unlock(&bpf_module_mutex);
71 return ret;
72}
73#else
74static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
75{
76 return NULL;
77}
78#endif /* CONFIG_MODULES */
79
80u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
81u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
82
83static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
84 u64 flags, const struct btf **btf,
85 s32 *btf_id);
86static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
87static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
88
89/**
90 * trace_call_bpf - invoke BPF program
91 * @call: tracepoint event
92 * @ctx: opaque context pointer
93 *
94 * kprobe handlers execute BPF programs via this helper.
95 * Can be used from static tracepoints in the future.
96 *
97 * Return: BPF programs always return an integer which is interpreted by
98 * kprobe handler as:
99 * 0 - return from kprobe (event is filtered out)
100 * 1 - store kprobe event into ring buffer
101 * Other values are reserved and currently alias to 1
102 */
103unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
104{
105 unsigned int ret;
106
107 cant_sleep();
108
109 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
110 /*
111 * since some bpf program is already running on this cpu,
112 * don't call into another bpf program (same or different)
113 * and don't send kprobe event into ring-buffer,
114 * so return zero here
115 */
116 ret = 0;
117 goto out;
118 }
119
120 /*
121 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
122 * to all call sites, we did a bpf_prog_array_valid() there to check
123 * whether call->prog_array is empty or not, which is
124 * a heuristic to speed up execution.
125 *
126 * If bpf_prog_array_valid() fetched prog_array was
127 * non-NULL, we go into trace_call_bpf() and do the actual
128 * proper rcu_dereference() under RCU lock.
129 * If it turns out that prog_array is NULL then, we bail out.
130 * For the opposite, if the bpf_prog_array_valid() fetched pointer
131 * was NULL, you'll skip the prog_array with the risk of missing
132 * out of events when it was updated in between this and the
133 * rcu_dereference() which is accepted risk.
134 */
135 rcu_read_lock();
136 ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
137 ctx, bpf_prog_run);
138 rcu_read_unlock();
139
140 out:
141 __this_cpu_dec(bpf_prog_active);
142
143 return ret;
144}
145
146#ifdef CONFIG_BPF_KPROBE_OVERRIDE
147BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
148{
149 regs_set_return_value(regs, rc);
150 override_function_with_return(regs);
151 return 0;
152}
153
154static const struct bpf_func_proto bpf_override_return_proto = {
155 .func = bpf_override_return,
156 .gpl_only = true,
157 .ret_type = RET_INTEGER,
158 .arg1_type = ARG_PTR_TO_CTX,
159 .arg2_type = ARG_ANYTHING,
160};
161#endif
162
163static __always_inline int
164bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
165{
166 int ret;
167
168 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
169 if (unlikely(ret < 0))
170 memset(dst, 0, size);
171 return ret;
172}
173
174BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
175 const void __user *, unsafe_ptr)
176{
177 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
178}
179
180const struct bpf_func_proto bpf_probe_read_user_proto = {
181 .func = bpf_probe_read_user,
182 .gpl_only = true,
183 .ret_type = RET_INTEGER,
184 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
185 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
186 .arg3_type = ARG_ANYTHING,
187};
188
189static __always_inline int
190bpf_probe_read_user_str_common(void *dst, u32 size,
191 const void __user *unsafe_ptr)
192{
193 int ret;
194
195 /*
196 * NB: We rely on strncpy_from_user() not copying junk past the NUL
197 * terminator into `dst`.
198 *
199 * strncpy_from_user() does long-sized strides in the fast path. If the
200 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
201 * then there could be junk after the NUL in `dst`. If user takes `dst`
202 * and keys a hash map with it, then semantically identical strings can
203 * occupy multiple entries in the map.
204 */
205 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
206 if (unlikely(ret < 0))
207 memset(dst, 0, size);
208 return ret;
209}
210
211BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
212 const void __user *, unsafe_ptr)
213{
214 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
215}
216
217const struct bpf_func_proto bpf_probe_read_user_str_proto = {
218 .func = bpf_probe_read_user_str,
219 .gpl_only = true,
220 .ret_type = RET_INTEGER,
221 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
222 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
223 .arg3_type = ARG_ANYTHING,
224};
225
226static __always_inline int
227bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
228{
229 int ret;
230
231 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
232 if (unlikely(ret < 0))
233 memset(dst, 0, size);
234 return ret;
235}
236
237BPF_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
243const 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
252static __always_inline int
253bpf_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
272BPF_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
278const 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
288BPF_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
298static 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
307BPF_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
317static 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
327BPF_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
352static 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
361static 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
372static DEFINE_RAW_SPINLOCK(trace_printk_lock);
373
374#define MAX_TRACE_PRINTK_VARARGS 3
375#define BPF_TRACE_PRINTK_SIZE 1024
376
377BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
378 u64, arg2, u64, arg3)
379{
380 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
381 u32 *bin_args;
382 static char buf[BPF_TRACE_PRINTK_SIZE];
383 unsigned long flags;
384 int ret;
385
386 ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
387 MAX_TRACE_PRINTK_VARARGS);
388 if (ret < 0)
389 return ret;
390
391 raw_spin_lock_irqsave(&trace_printk_lock, flags);
392 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
393
394 trace_bpf_trace_printk(buf);
395 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
396
397 bpf_bprintf_cleanup();
398
399 return ret;
400}
401
402static const struct bpf_func_proto bpf_trace_printk_proto = {
403 .func = bpf_trace_printk,
404 .gpl_only = true,
405 .ret_type = RET_INTEGER,
406 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
407 .arg2_type = ARG_CONST_SIZE,
408};
409
410static void __set_printk_clr_event(void)
411{
412 /*
413 * This program might be calling bpf_trace_printk,
414 * so enable the associated bpf_trace/bpf_trace_printk event.
415 * Repeat this each time as it is possible a user has
416 * disabled bpf_trace_printk events. By loading a program
417 * calling bpf_trace_printk() however the user has expressed
418 * the intent to see such events.
419 */
420 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
421 pr_warn_ratelimited("could not enable bpf_trace_printk events");
422}
423
424const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
425{
426 __set_printk_clr_event();
427 return &bpf_trace_printk_proto;
428}
429
430BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
431 u32, data_len)
432{
433 static char buf[BPF_TRACE_PRINTK_SIZE];
434 unsigned long flags;
435 int ret, num_args;
436 u32 *bin_args;
437
438 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
439 (data_len && !data))
440 return -EINVAL;
441 num_args = data_len / 8;
442
443 ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
444 if (ret < 0)
445 return ret;
446
447 raw_spin_lock_irqsave(&trace_printk_lock, flags);
448 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
449
450 trace_bpf_trace_printk(buf);
451 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
452
453 bpf_bprintf_cleanup();
454
455 return ret;
456}
457
458static const struct bpf_func_proto bpf_trace_vprintk_proto = {
459 .func = bpf_trace_vprintk,
460 .gpl_only = true,
461 .ret_type = RET_INTEGER,
462 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
463 .arg2_type = ARG_CONST_SIZE,
464 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
465 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
466};
467
468const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
469{
470 __set_printk_clr_event();
471 return &bpf_trace_vprintk_proto;
472}
473
474BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
475 const void *, data, u32, data_len)
476{
477 int err, num_args;
478 u32 *bin_args;
479
480 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
481 (data_len && !data))
482 return -EINVAL;
483 num_args = data_len / 8;
484
485 err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
486 if (err < 0)
487 return err;
488
489 seq_bprintf(m, fmt, bin_args);
490
491 bpf_bprintf_cleanup();
492
493 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
494}
495
496BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
497
498static const struct bpf_func_proto bpf_seq_printf_proto = {
499 .func = bpf_seq_printf,
500 .gpl_only = true,
501 .ret_type = RET_INTEGER,
502 .arg1_type = ARG_PTR_TO_BTF_ID,
503 .arg1_btf_id = &btf_seq_file_ids[0],
504 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
505 .arg3_type = ARG_CONST_SIZE,
506 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
507 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
508};
509
510BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
511{
512 return seq_write(m, data, len) ? -EOVERFLOW : 0;
513}
514
515static const struct bpf_func_proto bpf_seq_write_proto = {
516 .func = bpf_seq_write,
517 .gpl_only = true,
518 .ret_type = RET_INTEGER,
519 .arg1_type = ARG_PTR_TO_BTF_ID,
520 .arg1_btf_id = &btf_seq_file_ids[0],
521 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
522 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
523};
524
525BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
526 u32, btf_ptr_size, u64, flags)
527{
528 const struct btf *btf;
529 s32 btf_id;
530 int ret;
531
532 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
533 if (ret)
534 return ret;
535
536 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
537}
538
539static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
540 .func = bpf_seq_printf_btf,
541 .gpl_only = true,
542 .ret_type = RET_INTEGER,
543 .arg1_type = ARG_PTR_TO_BTF_ID,
544 .arg1_btf_id = &btf_seq_file_ids[0],
545 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
546 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
547 .arg4_type = ARG_ANYTHING,
548};
549
550static __always_inline int
551get_map_perf_counter(struct bpf_map *map, u64 flags,
552 u64 *value, u64 *enabled, u64 *running)
553{
554 struct bpf_array *array = container_of(map, struct bpf_array, map);
555 unsigned int cpu = smp_processor_id();
556 u64 index = flags & BPF_F_INDEX_MASK;
557 struct bpf_event_entry *ee;
558
559 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
560 return -EINVAL;
561 if (index == BPF_F_CURRENT_CPU)
562 index = cpu;
563 if (unlikely(index >= array->map.max_entries))
564 return -E2BIG;
565
566 ee = READ_ONCE(array->ptrs[index]);
567 if (!ee)
568 return -ENOENT;
569
570 return perf_event_read_local(ee->event, value, enabled, running);
571}
572
573BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
574{
575 u64 value = 0;
576 int err;
577
578 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
579 /*
580 * this api is ugly since we miss [-22..-2] range of valid
581 * counter values, but that's uapi
582 */
583 if (err)
584 return err;
585 return value;
586}
587
588static const struct bpf_func_proto bpf_perf_event_read_proto = {
589 .func = bpf_perf_event_read,
590 .gpl_only = true,
591 .ret_type = RET_INTEGER,
592 .arg1_type = ARG_CONST_MAP_PTR,
593 .arg2_type = ARG_ANYTHING,
594};
595
596BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
597 struct bpf_perf_event_value *, buf, u32, size)
598{
599 int err = -EINVAL;
600
601 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
602 goto clear;
603 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
604 &buf->running);
605 if (unlikely(err))
606 goto clear;
607 return 0;
608clear:
609 memset(buf, 0, size);
610 return err;
611}
612
613static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
614 .func = bpf_perf_event_read_value,
615 .gpl_only = true,
616 .ret_type = RET_INTEGER,
617 .arg1_type = ARG_CONST_MAP_PTR,
618 .arg2_type = ARG_ANYTHING,
619 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
620 .arg4_type = ARG_CONST_SIZE,
621};
622
623static __always_inline u64
624__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
625 u64 flags, struct perf_sample_data *sd)
626{
627 struct bpf_array *array = container_of(map, struct bpf_array, map);
628 unsigned int cpu = smp_processor_id();
629 u64 index = flags & BPF_F_INDEX_MASK;
630 struct bpf_event_entry *ee;
631 struct perf_event *event;
632
633 if (index == BPF_F_CURRENT_CPU)
634 index = cpu;
635 if (unlikely(index >= array->map.max_entries))
636 return -E2BIG;
637
638 ee = READ_ONCE(array->ptrs[index]);
639 if (!ee)
640 return -ENOENT;
641
642 event = ee->event;
643 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
644 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
645 return -EINVAL;
646
647 if (unlikely(event->oncpu != cpu))
648 return -EOPNOTSUPP;
649
650 return perf_event_output(event, sd, regs);
651}
652
653/*
654 * Support executing tracepoints in normal, irq, and nmi context that each call
655 * bpf_perf_event_output
656 */
657struct bpf_trace_sample_data {
658 struct perf_sample_data sds[3];
659};
660
661static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
662static DEFINE_PER_CPU(int, bpf_trace_nest_level);
663BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
664 u64, flags, void *, data, u64, size)
665{
666 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
667 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
668 struct perf_raw_record raw = {
669 .frag = {
670 .size = size,
671 .data = data,
672 },
673 };
674 struct perf_sample_data *sd;
675 int err;
676
677 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
678 err = -EBUSY;
679 goto out;
680 }
681
682 sd = &sds->sds[nest_level - 1];
683
684 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
685 err = -EINVAL;
686 goto out;
687 }
688
689 perf_sample_data_init(sd, 0, 0);
690 sd->raw = &raw;
691 sd->sample_flags |= PERF_SAMPLE_RAW;
692
693 err = __bpf_perf_event_output(regs, map, flags, sd);
694
695out:
696 this_cpu_dec(bpf_trace_nest_level);
697 return err;
698}
699
700static 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
711static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
712struct bpf_nested_pt_regs {
713 struct pt_regs regs[3];
714};
715static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
716static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
717
718u64 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 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
722 struct perf_raw_frag frag = {
723 .copy = ctx_copy,
724 .size = ctx_size,
725 .data = ctx,
726 };
727 struct perf_raw_record raw = {
728 .frag = {
729 {
730 .next = ctx_size ? &frag : NULL,
731 },
732 .size = meta_size,
733 .data = meta,
734 },
735 };
736 struct perf_sample_data *sd;
737 struct pt_regs *regs;
738 u64 ret;
739
740 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
741 ret = -EBUSY;
742 goto out;
743 }
744 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
745 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
746
747 perf_fetch_caller_regs(regs);
748 perf_sample_data_init(sd, 0, 0);
749 sd->raw = &raw;
750 sd->sample_flags |= PERF_SAMPLE_RAW;
751
752 ret = __bpf_perf_event_output(regs, map, flags, sd);
753out:
754 this_cpu_dec(bpf_event_output_nest_level);
755 return ret;
756}
757
758BPF_CALL_0(bpf_get_current_task)
759{
760 return (long) current;
761}
762
763const struct bpf_func_proto bpf_get_current_task_proto = {
764 .func = bpf_get_current_task,
765 .gpl_only = true,
766 .ret_type = RET_INTEGER,
767};
768
769BPF_CALL_0(bpf_get_current_task_btf)
770{
771 return (unsigned long) current;
772}
773
774const struct bpf_func_proto bpf_get_current_task_btf_proto = {
775 .func = bpf_get_current_task_btf,
776 .gpl_only = true,
777 .ret_type = RET_PTR_TO_BTF_ID_TRUSTED,
778 .ret_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
779};
780
781BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
782{
783 return (unsigned long) task_pt_regs(task);
784}
785
786BTF_ID_LIST(bpf_task_pt_regs_ids)
787BTF_ID(struct, pt_regs)
788
789const struct bpf_func_proto bpf_task_pt_regs_proto = {
790 .func = bpf_task_pt_regs,
791 .gpl_only = true,
792 .arg1_type = ARG_PTR_TO_BTF_ID,
793 .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
794 .ret_type = RET_PTR_TO_BTF_ID,
795 .ret_btf_id = &bpf_task_pt_regs_ids[0],
796};
797
798BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
799{
800 struct bpf_array *array = container_of(map, struct bpf_array, map);
801 struct cgroup *cgrp;
802
803 if (unlikely(idx >= array->map.max_entries))
804 return -E2BIG;
805
806 cgrp = READ_ONCE(array->ptrs[idx]);
807 if (unlikely(!cgrp))
808 return -EAGAIN;
809
810 return task_under_cgroup_hierarchy(current, cgrp);
811}
812
813static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
814 .func = bpf_current_task_under_cgroup,
815 .gpl_only = false,
816 .ret_type = RET_INTEGER,
817 .arg1_type = ARG_CONST_MAP_PTR,
818 .arg2_type = ARG_ANYTHING,
819};
820
821struct send_signal_irq_work {
822 struct irq_work irq_work;
823 struct task_struct *task;
824 u32 sig;
825 enum pid_type type;
826};
827
828static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
829
830static void do_bpf_send_signal(struct irq_work *entry)
831{
832 struct send_signal_irq_work *work;
833
834 work = container_of(entry, struct send_signal_irq_work, irq_work);
835 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
836 put_task_struct(work->task);
837}
838
839static int bpf_send_signal_common(u32 sig, enum pid_type type)
840{
841 struct send_signal_irq_work *work = NULL;
842
843 /* Similar to bpf_probe_write_user, task needs to be
844 * in a sound condition and kernel memory access be
845 * permitted in order to send signal to the current
846 * task.
847 */
848 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
849 return -EPERM;
850 if (unlikely(!nmi_uaccess_okay()))
851 return -EPERM;
852 /* Task should not be pid=1 to avoid kernel panic. */
853 if (unlikely(is_global_init(current)))
854 return -EPERM;
855
856 if (irqs_disabled()) {
857 /* Do an early check on signal validity. Otherwise,
858 * the error is lost in deferred irq_work.
859 */
860 if (unlikely(!valid_signal(sig)))
861 return -EINVAL;
862
863 work = this_cpu_ptr(&send_signal_work);
864 if (irq_work_is_busy(&work->irq_work))
865 return -EBUSY;
866
867 /* Add the current task, which is the target of sending signal,
868 * to the irq_work. The current task may change when queued
869 * irq works get executed.
870 */
871 work->task = get_task_struct(current);
872 work->sig = sig;
873 work->type = type;
874 irq_work_queue(&work->irq_work);
875 return 0;
876 }
877
878 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
879}
880
881BPF_CALL_1(bpf_send_signal, u32, sig)
882{
883 return bpf_send_signal_common(sig, PIDTYPE_TGID);
884}
885
886static const struct bpf_func_proto bpf_send_signal_proto = {
887 .func = bpf_send_signal,
888 .gpl_only = false,
889 .ret_type = RET_INTEGER,
890 .arg1_type = ARG_ANYTHING,
891};
892
893BPF_CALL_1(bpf_send_signal_thread, u32, sig)
894{
895 return bpf_send_signal_common(sig, PIDTYPE_PID);
896}
897
898static const struct bpf_func_proto bpf_send_signal_thread_proto = {
899 .func = bpf_send_signal_thread,
900 .gpl_only = false,
901 .ret_type = RET_INTEGER,
902 .arg1_type = ARG_ANYTHING,
903};
904
905BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
906{
907 long len;
908 char *p;
909
910 if (!sz)
911 return 0;
912
913 p = d_path(path, buf, sz);
914 if (IS_ERR(p)) {
915 len = PTR_ERR(p);
916 } else {
917 len = buf + sz - p;
918 memmove(buf, p, len);
919 }
920
921 return len;
922}
923
924BTF_SET_START(btf_allowlist_d_path)
925#ifdef CONFIG_SECURITY
926BTF_ID(func, security_file_permission)
927BTF_ID(func, security_inode_getattr)
928BTF_ID(func, security_file_open)
929#endif
930#ifdef CONFIG_SECURITY_PATH
931BTF_ID(func, security_path_truncate)
932#endif
933BTF_ID(func, vfs_truncate)
934BTF_ID(func, vfs_fallocate)
935BTF_ID(func, dentry_open)
936BTF_ID(func, vfs_getattr)
937BTF_ID(func, filp_close)
938BTF_SET_END(btf_allowlist_d_path)
939
940static bool bpf_d_path_allowed(const struct bpf_prog *prog)
941{
942 if (prog->type == BPF_PROG_TYPE_TRACING &&
943 prog->expected_attach_type == BPF_TRACE_ITER)
944 return true;
945
946 if (prog->type == BPF_PROG_TYPE_LSM)
947 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
948
949 return btf_id_set_contains(&btf_allowlist_d_path,
950 prog->aux->attach_btf_id);
951}
952
953BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
954
955static const struct bpf_func_proto bpf_d_path_proto = {
956 .func = bpf_d_path,
957 .gpl_only = false,
958 .ret_type = RET_INTEGER,
959 .arg1_type = ARG_PTR_TO_BTF_ID,
960 .arg1_btf_id = &bpf_d_path_btf_ids[0],
961 .arg2_type = ARG_PTR_TO_MEM,
962 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
963 .allowed = bpf_d_path_allowed,
964};
965
966#define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
967 BTF_F_PTR_RAW | BTF_F_ZERO)
968
969static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
970 u64 flags, const struct btf **btf,
971 s32 *btf_id)
972{
973 const struct btf_type *t;
974
975 if (unlikely(flags & ~(BTF_F_ALL)))
976 return -EINVAL;
977
978 if (btf_ptr_size != sizeof(struct btf_ptr))
979 return -EINVAL;
980
981 *btf = bpf_get_btf_vmlinux();
982
983 if (IS_ERR_OR_NULL(*btf))
984 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
985
986 if (ptr->type_id > 0)
987 *btf_id = ptr->type_id;
988 else
989 return -EINVAL;
990
991 if (*btf_id > 0)
992 t = btf_type_by_id(*btf, *btf_id);
993 if (*btf_id <= 0 || !t)
994 return -ENOENT;
995
996 return 0;
997}
998
999BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1000 u32, btf_ptr_size, u64, flags)
1001{
1002 const struct btf *btf;
1003 s32 btf_id;
1004 int ret;
1005
1006 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1007 if (ret)
1008 return ret;
1009
1010 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1011 flags);
1012}
1013
1014const struct bpf_func_proto bpf_snprintf_btf_proto = {
1015 .func = bpf_snprintf_btf,
1016 .gpl_only = false,
1017 .ret_type = RET_INTEGER,
1018 .arg1_type = ARG_PTR_TO_MEM,
1019 .arg2_type = ARG_CONST_SIZE,
1020 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1021 .arg4_type = ARG_CONST_SIZE,
1022 .arg5_type = ARG_ANYTHING,
1023};
1024
1025BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1026{
1027 /* This helper call is inlined by verifier. */
1028 return ((u64 *)ctx)[-2];
1029}
1030
1031static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1032 .func = bpf_get_func_ip_tracing,
1033 .gpl_only = true,
1034 .ret_type = RET_INTEGER,
1035 .arg1_type = ARG_PTR_TO_CTX,
1036};
1037
1038#ifdef CONFIG_X86_KERNEL_IBT
1039static unsigned long get_entry_ip(unsigned long fentry_ip)
1040{
1041 u32 instr;
1042
1043 /* Being extra safe in here in case entry ip is on the page-edge. */
1044 if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1045 return fentry_ip;
1046 if (is_endbr(instr))
1047 fentry_ip -= ENDBR_INSN_SIZE;
1048 return fentry_ip;
1049}
1050#else
1051#define get_entry_ip(fentry_ip) fentry_ip
1052#endif
1053
1054BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1055{
1056 struct kprobe *kp = kprobe_running();
1057
1058 if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1059 return 0;
1060
1061 return get_entry_ip((uintptr_t)kp->addr);
1062}
1063
1064static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1065 .func = bpf_get_func_ip_kprobe,
1066 .gpl_only = true,
1067 .ret_type = RET_INTEGER,
1068 .arg1_type = ARG_PTR_TO_CTX,
1069};
1070
1071BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1072{
1073 return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1074}
1075
1076static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1077 .func = bpf_get_func_ip_kprobe_multi,
1078 .gpl_only = false,
1079 .ret_type = RET_INTEGER,
1080 .arg1_type = ARG_PTR_TO_CTX,
1081};
1082
1083BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1084{
1085 return bpf_kprobe_multi_cookie(current->bpf_ctx);
1086}
1087
1088static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1089 .func = bpf_get_attach_cookie_kprobe_multi,
1090 .gpl_only = false,
1091 .ret_type = RET_INTEGER,
1092 .arg1_type = ARG_PTR_TO_CTX,
1093};
1094
1095BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1096{
1097 struct bpf_trace_run_ctx *run_ctx;
1098
1099 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1100 return run_ctx->bpf_cookie;
1101}
1102
1103static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1104 .func = bpf_get_attach_cookie_trace,
1105 .gpl_only = false,
1106 .ret_type = RET_INTEGER,
1107 .arg1_type = ARG_PTR_TO_CTX,
1108};
1109
1110BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1111{
1112 return ctx->event->bpf_cookie;
1113}
1114
1115static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1116 .func = bpf_get_attach_cookie_pe,
1117 .gpl_only = false,
1118 .ret_type = RET_INTEGER,
1119 .arg1_type = ARG_PTR_TO_CTX,
1120};
1121
1122BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1123{
1124 struct bpf_trace_run_ctx *run_ctx;
1125
1126 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1127 return run_ctx->bpf_cookie;
1128}
1129
1130static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1131 .func = bpf_get_attach_cookie_tracing,
1132 .gpl_only = false,
1133 .ret_type = RET_INTEGER,
1134 .arg1_type = ARG_PTR_TO_CTX,
1135};
1136
1137BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1138{
1139#ifndef CONFIG_X86
1140 return -ENOENT;
1141#else
1142 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1143 u32 entry_cnt = size / br_entry_size;
1144
1145 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1146
1147 if (unlikely(flags))
1148 return -EINVAL;
1149
1150 if (!entry_cnt)
1151 return -ENOENT;
1152
1153 return entry_cnt * br_entry_size;
1154#endif
1155}
1156
1157static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1158 .func = bpf_get_branch_snapshot,
1159 .gpl_only = true,
1160 .ret_type = RET_INTEGER,
1161 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1162 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1163};
1164
1165BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1166{
1167 /* This helper call is inlined by verifier. */
1168 u64 nr_args = ((u64 *)ctx)[-1];
1169
1170 if ((u64) n >= nr_args)
1171 return -EINVAL;
1172 *value = ((u64 *)ctx)[n];
1173 return 0;
1174}
1175
1176static const struct bpf_func_proto bpf_get_func_arg_proto = {
1177 .func = get_func_arg,
1178 .ret_type = RET_INTEGER,
1179 .arg1_type = ARG_PTR_TO_CTX,
1180 .arg2_type = ARG_ANYTHING,
1181 .arg3_type = ARG_PTR_TO_LONG,
1182};
1183
1184BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1185{
1186 /* This helper call is inlined by verifier. */
1187 u64 nr_args = ((u64 *)ctx)[-1];
1188
1189 *value = ((u64 *)ctx)[nr_args];
1190 return 0;
1191}
1192
1193static const struct bpf_func_proto bpf_get_func_ret_proto = {
1194 .func = get_func_ret,
1195 .ret_type = RET_INTEGER,
1196 .arg1_type = ARG_PTR_TO_CTX,
1197 .arg2_type = ARG_PTR_TO_LONG,
1198};
1199
1200BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1201{
1202 /* This helper call is inlined by verifier. */
1203 return ((u64 *)ctx)[-1];
1204}
1205
1206static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1207 .func = get_func_arg_cnt,
1208 .ret_type = RET_INTEGER,
1209 .arg1_type = ARG_PTR_TO_CTX,
1210};
1211
1212#ifdef CONFIG_KEYS
1213__diag_push();
1214__diag_ignore_all("-Wmissing-prototypes",
1215 "kfuncs which will be used in BPF programs");
1216
1217/**
1218 * bpf_lookup_user_key - lookup a key by its serial
1219 * @serial: key handle serial number
1220 * @flags: lookup-specific flags
1221 *
1222 * Search a key with a given *serial* and the provided *flags*.
1223 * If found, increment the reference count of the key by one, and
1224 * return it in the bpf_key structure.
1225 *
1226 * The bpf_key structure must be passed to bpf_key_put() when done
1227 * with it, so that the key reference count is decremented and the
1228 * bpf_key structure is freed.
1229 *
1230 * Permission checks are deferred to the time the key is used by
1231 * one of the available key-specific kfuncs.
1232 *
1233 * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1234 * special keyring (e.g. session keyring), if it doesn't yet exist.
1235 * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1236 * for the key construction, and to retrieve uninstantiated keys (keys
1237 * without data attached to them).
1238 *
1239 * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1240 * NULL pointer otherwise.
1241 */
1242struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1243{
1244 key_ref_t key_ref;
1245 struct bpf_key *bkey;
1246
1247 if (flags & ~KEY_LOOKUP_ALL)
1248 return NULL;
1249
1250 /*
1251 * Permission check is deferred until the key is used, as the
1252 * intent of the caller is unknown here.
1253 */
1254 key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1255 if (IS_ERR(key_ref))
1256 return NULL;
1257
1258 bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1259 if (!bkey) {
1260 key_put(key_ref_to_ptr(key_ref));
1261 return NULL;
1262 }
1263
1264 bkey->key = key_ref_to_ptr(key_ref);
1265 bkey->has_ref = true;
1266
1267 return bkey;
1268}
1269
1270/**
1271 * bpf_lookup_system_key - lookup a key by a system-defined ID
1272 * @id: key ID
1273 *
1274 * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1275 * The key pointer is marked as invalid, to prevent bpf_key_put() from
1276 * attempting to decrement the key reference count on that pointer. The key
1277 * pointer set in such way is currently understood only by
1278 * verify_pkcs7_signature().
1279 *
1280 * Set *id* to one of the values defined in include/linux/verification.h:
1281 * 0 for the primary keyring (immutable keyring of system keys);
1282 * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1283 * (where keys can be added only if they are vouched for by existing keys
1284 * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1285 * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1286 * kerned image and, possibly, the initramfs signature).
1287 *
1288 * Return: a bpf_key pointer with an invalid key pointer set from the
1289 * pre-determined ID on success, a NULL pointer otherwise
1290 */
1291struct bpf_key *bpf_lookup_system_key(u64 id)
1292{
1293 struct bpf_key *bkey;
1294
1295 if (system_keyring_id_check(id) < 0)
1296 return NULL;
1297
1298 bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1299 if (!bkey)
1300 return NULL;
1301
1302 bkey->key = (struct key *)(unsigned long)id;
1303 bkey->has_ref = false;
1304
1305 return bkey;
1306}
1307
1308/**
1309 * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1310 * @bkey: bpf_key structure
1311 *
1312 * Decrement the reference count of the key inside *bkey*, if the pointer
1313 * is valid, and free *bkey*.
1314 */
1315void bpf_key_put(struct bpf_key *bkey)
1316{
1317 if (bkey->has_ref)
1318 key_put(bkey->key);
1319
1320 kfree(bkey);
1321}
1322
1323#ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1324/**
1325 * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1326 * @data_ptr: data to verify
1327 * @sig_ptr: signature of the data
1328 * @trusted_keyring: keyring with keys trusted for signature verification
1329 *
1330 * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1331 * with keys in a keyring referenced by *trusted_keyring*.
1332 *
1333 * Return: 0 on success, a negative value on error.
1334 */
1335int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1336 struct bpf_dynptr_kern *sig_ptr,
1337 struct bpf_key *trusted_keyring)
1338{
1339 int ret;
1340
1341 if (trusted_keyring->has_ref) {
1342 /*
1343 * Do the permission check deferred in bpf_lookup_user_key().
1344 * See bpf_lookup_user_key() for more details.
1345 *
1346 * A call to key_task_permission() here would be redundant, as
1347 * it is already done by keyring_search() called by
1348 * find_asymmetric_key().
1349 */
1350 ret = key_validate(trusted_keyring->key);
1351 if (ret < 0)
1352 return ret;
1353 }
1354
1355 return verify_pkcs7_signature(data_ptr->data,
1356 bpf_dynptr_get_size(data_ptr),
1357 sig_ptr->data,
1358 bpf_dynptr_get_size(sig_ptr),
1359 trusted_keyring->key,
1360 VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1361 NULL);
1362}
1363#endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1364
1365__diag_pop();
1366
1367BTF_SET8_START(key_sig_kfunc_set)
1368BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1369BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1370BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1371#ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1372BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1373#endif
1374BTF_SET8_END(key_sig_kfunc_set)
1375
1376static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1377 .owner = THIS_MODULE,
1378 .set = &key_sig_kfunc_set,
1379};
1380
1381static int __init bpf_key_sig_kfuncs_init(void)
1382{
1383 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1384 &bpf_key_sig_kfunc_set);
1385}
1386
1387late_initcall(bpf_key_sig_kfuncs_init);
1388#endif /* CONFIG_KEYS */
1389
1390static const struct bpf_func_proto *
1391bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1392{
1393 switch (func_id) {
1394 case BPF_FUNC_map_lookup_elem:
1395 return &bpf_map_lookup_elem_proto;
1396 case BPF_FUNC_map_update_elem:
1397 return &bpf_map_update_elem_proto;
1398 case BPF_FUNC_map_delete_elem:
1399 return &bpf_map_delete_elem_proto;
1400 case BPF_FUNC_map_push_elem:
1401 return &bpf_map_push_elem_proto;
1402 case BPF_FUNC_map_pop_elem:
1403 return &bpf_map_pop_elem_proto;
1404 case BPF_FUNC_map_peek_elem:
1405 return &bpf_map_peek_elem_proto;
1406 case BPF_FUNC_map_lookup_percpu_elem:
1407 return &bpf_map_lookup_percpu_elem_proto;
1408 case BPF_FUNC_ktime_get_ns:
1409 return &bpf_ktime_get_ns_proto;
1410 case BPF_FUNC_ktime_get_boot_ns:
1411 return &bpf_ktime_get_boot_ns_proto;
1412 case BPF_FUNC_tail_call:
1413 return &bpf_tail_call_proto;
1414 case BPF_FUNC_get_current_pid_tgid:
1415 return &bpf_get_current_pid_tgid_proto;
1416 case BPF_FUNC_get_current_task:
1417 return &bpf_get_current_task_proto;
1418 case BPF_FUNC_get_current_task_btf:
1419 return &bpf_get_current_task_btf_proto;
1420 case BPF_FUNC_task_pt_regs:
1421 return &bpf_task_pt_regs_proto;
1422 case BPF_FUNC_get_current_uid_gid:
1423 return &bpf_get_current_uid_gid_proto;
1424 case BPF_FUNC_get_current_comm:
1425 return &bpf_get_current_comm_proto;
1426 case BPF_FUNC_trace_printk:
1427 return bpf_get_trace_printk_proto();
1428 case BPF_FUNC_get_smp_processor_id:
1429 return &bpf_get_smp_processor_id_proto;
1430 case BPF_FUNC_get_numa_node_id:
1431 return &bpf_get_numa_node_id_proto;
1432 case BPF_FUNC_perf_event_read:
1433 return &bpf_perf_event_read_proto;
1434 case BPF_FUNC_current_task_under_cgroup:
1435 return &bpf_current_task_under_cgroup_proto;
1436 case BPF_FUNC_get_prandom_u32:
1437 return &bpf_get_prandom_u32_proto;
1438 case BPF_FUNC_probe_write_user:
1439 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1440 NULL : bpf_get_probe_write_proto();
1441 case BPF_FUNC_probe_read_user:
1442 return &bpf_probe_read_user_proto;
1443 case BPF_FUNC_probe_read_kernel:
1444 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1445 NULL : &bpf_probe_read_kernel_proto;
1446 case BPF_FUNC_probe_read_user_str:
1447 return &bpf_probe_read_user_str_proto;
1448 case BPF_FUNC_probe_read_kernel_str:
1449 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1450 NULL : &bpf_probe_read_kernel_str_proto;
1451#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1452 case BPF_FUNC_probe_read:
1453 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1454 NULL : &bpf_probe_read_compat_proto;
1455 case BPF_FUNC_probe_read_str:
1456 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1457 NULL : &bpf_probe_read_compat_str_proto;
1458#endif
1459#ifdef CONFIG_CGROUPS
1460 case BPF_FUNC_get_current_cgroup_id:
1461 return &bpf_get_current_cgroup_id_proto;
1462 case BPF_FUNC_get_current_ancestor_cgroup_id:
1463 return &bpf_get_current_ancestor_cgroup_id_proto;
1464 case BPF_FUNC_cgrp_storage_get:
1465 return &bpf_cgrp_storage_get_proto;
1466 case BPF_FUNC_cgrp_storage_delete:
1467 return &bpf_cgrp_storage_delete_proto;
1468#endif
1469 case BPF_FUNC_send_signal:
1470 return &bpf_send_signal_proto;
1471 case BPF_FUNC_send_signal_thread:
1472 return &bpf_send_signal_thread_proto;
1473 case BPF_FUNC_perf_event_read_value:
1474 return &bpf_perf_event_read_value_proto;
1475 case BPF_FUNC_get_ns_current_pid_tgid:
1476 return &bpf_get_ns_current_pid_tgid_proto;
1477 case BPF_FUNC_ringbuf_output:
1478 return &bpf_ringbuf_output_proto;
1479 case BPF_FUNC_ringbuf_reserve:
1480 return &bpf_ringbuf_reserve_proto;
1481 case BPF_FUNC_ringbuf_submit:
1482 return &bpf_ringbuf_submit_proto;
1483 case BPF_FUNC_ringbuf_discard:
1484 return &bpf_ringbuf_discard_proto;
1485 case BPF_FUNC_ringbuf_query:
1486 return &bpf_ringbuf_query_proto;
1487 case BPF_FUNC_jiffies64:
1488 return &bpf_jiffies64_proto;
1489 case BPF_FUNC_get_task_stack:
1490 return &bpf_get_task_stack_proto;
1491 case BPF_FUNC_copy_from_user:
1492 return &bpf_copy_from_user_proto;
1493 case BPF_FUNC_copy_from_user_task:
1494 return &bpf_copy_from_user_task_proto;
1495 case BPF_FUNC_snprintf_btf:
1496 return &bpf_snprintf_btf_proto;
1497 case BPF_FUNC_per_cpu_ptr:
1498 return &bpf_per_cpu_ptr_proto;
1499 case BPF_FUNC_this_cpu_ptr:
1500 return &bpf_this_cpu_ptr_proto;
1501 case BPF_FUNC_task_storage_get:
1502 if (bpf_prog_check_recur(prog))
1503 return &bpf_task_storage_get_recur_proto;
1504 return &bpf_task_storage_get_proto;
1505 case BPF_FUNC_task_storage_delete:
1506 if (bpf_prog_check_recur(prog))
1507 return &bpf_task_storage_delete_recur_proto;
1508 return &bpf_task_storage_delete_proto;
1509 case BPF_FUNC_for_each_map_elem:
1510 return &bpf_for_each_map_elem_proto;
1511 case BPF_FUNC_snprintf:
1512 return &bpf_snprintf_proto;
1513 case BPF_FUNC_get_func_ip:
1514 return &bpf_get_func_ip_proto_tracing;
1515 case BPF_FUNC_get_branch_snapshot:
1516 return &bpf_get_branch_snapshot_proto;
1517 case BPF_FUNC_find_vma:
1518 return &bpf_find_vma_proto;
1519 case BPF_FUNC_trace_vprintk:
1520 return bpf_get_trace_vprintk_proto();
1521 default:
1522 return bpf_base_func_proto(func_id);
1523 }
1524}
1525
1526static const struct bpf_func_proto *
1527kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1528{
1529 switch (func_id) {
1530 case BPF_FUNC_perf_event_output:
1531 return &bpf_perf_event_output_proto;
1532 case BPF_FUNC_get_stackid:
1533 return &bpf_get_stackid_proto;
1534 case BPF_FUNC_get_stack:
1535 return &bpf_get_stack_proto;
1536#ifdef CONFIG_BPF_KPROBE_OVERRIDE
1537 case BPF_FUNC_override_return:
1538 return &bpf_override_return_proto;
1539#endif
1540 case BPF_FUNC_get_func_ip:
1541 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1542 &bpf_get_func_ip_proto_kprobe_multi :
1543 &bpf_get_func_ip_proto_kprobe;
1544 case BPF_FUNC_get_attach_cookie:
1545 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1546 &bpf_get_attach_cookie_proto_kmulti :
1547 &bpf_get_attach_cookie_proto_trace;
1548 default:
1549 return bpf_tracing_func_proto(func_id, prog);
1550 }
1551}
1552
1553/* bpf+kprobe programs can access fields of 'struct pt_regs' */
1554static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1555 const struct bpf_prog *prog,
1556 struct bpf_insn_access_aux *info)
1557{
1558 if (off < 0 || off >= sizeof(struct pt_regs))
1559 return false;
1560 if (type != BPF_READ)
1561 return false;
1562 if (off % size != 0)
1563 return false;
1564 /*
1565 * Assertion for 32 bit to make sure last 8 byte access
1566 * (BPF_DW) to the last 4 byte member is disallowed.
1567 */
1568 if (off + size > sizeof(struct pt_regs))
1569 return false;
1570
1571 return true;
1572}
1573
1574const struct bpf_verifier_ops kprobe_verifier_ops = {
1575 .get_func_proto = kprobe_prog_func_proto,
1576 .is_valid_access = kprobe_prog_is_valid_access,
1577};
1578
1579const struct bpf_prog_ops kprobe_prog_ops = {
1580};
1581
1582BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1583 u64, flags, void *, data, u64, size)
1584{
1585 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1586
1587 /*
1588 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1589 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1590 * from there and call the same bpf_perf_event_output() helper inline.
1591 */
1592 return ____bpf_perf_event_output(regs, map, flags, data, size);
1593}
1594
1595static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1596 .func = bpf_perf_event_output_tp,
1597 .gpl_only = true,
1598 .ret_type = RET_INTEGER,
1599 .arg1_type = ARG_PTR_TO_CTX,
1600 .arg2_type = ARG_CONST_MAP_PTR,
1601 .arg3_type = ARG_ANYTHING,
1602 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1603 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1604};
1605
1606BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1607 u64, flags)
1608{
1609 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1610
1611 /*
1612 * Same comment as in bpf_perf_event_output_tp(), only that this time
1613 * the other helper's function body cannot be inlined due to being
1614 * external, thus we need to call raw helper function.
1615 */
1616 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1617 flags, 0, 0);
1618}
1619
1620static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1621 .func = bpf_get_stackid_tp,
1622 .gpl_only = true,
1623 .ret_type = RET_INTEGER,
1624 .arg1_type = ARG_PTR_TO_CTX,
1625 .arg2_type = ARG_CONST_MAP_PTR,
1626 .arg3_type = ARG_ANYTHING,
1627};
1628
1629BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1630 u64, flags)
1631{
1632 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1633
1634 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1635 (unsigned long) size, flags, 0);
1636}
1637
1638static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1639 .func = bpf_get_stack_tp,
1640 .gpl_only = true,
1641 .ret_type = RET_INTEGER,
1642 .arg1_type = ARG_PTR_TO_CTX,
1643 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1644 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1645 .arg4_type = ARG_ANYTHING,
1646};
1647
1648static const struct bpf_func_proto *
1649tp_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_tp;
1654 case BPF_FUNC_get_stackid:
1655 return &bpf_get_stackid_proto_tp;
1656 case BPF_FUNC_get_stack:
1657 return &bpf_get_stack_proto_tp;
1658 case BPF_FUNC_get_attach_cookie:
1659 return &bpf_get_attach_cookie_proto_trace;
1660 default:
1661 return bpf_tracing_func_proto(func_id, prog);
1662 }
1663}
1664
1665static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1666 const struct bpf_prog *prog,
1667 struct bpf_insn_access_aux *info)
1668{
1669 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1670 return false;
1671 if (type != BPF_READ)
1672 return false;
1673 if (off % size != 0)
1674 return false;
1675
1676 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1677 return true;
1678}
1679
1680const struct bpf_verifier_ops tracepoint_verifier_ops = {
1681 .get_func_proto = tp_prog_func_proto,
1682 .is_valid_access = tp_prog_is_valid_access,
1683};
1684
1685const struct bpf_prog_ops tracepoint_prog_ops = {
1686};
1687
1688BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1689 struct bpf_perf_event_value *, buf, u32, size)
1690{
1691 int err = -EINVAL;
1692
1693 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1694 goto clear;
1695 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1696 &buf->running);
1697 if (unlikely(err))
1698 goto clear;
1699 return 0;
1700clear:
1701 memset(buf, 0, size);
1702 return err;
1703}
1704
1705static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1706 .func = bpf_perf_prog_read_value,
1707 .gpl_only = true,
1708 .ret_type = RET_INTEGER,
1709 .arg1_type = ARG_PTR_TO_CTX,
1710 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1711 .arg3_type = ARG_CONST_SIZE,
1712};
1713
1714BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1715 void *, buf, u32, size, u64, flags)
1716{
1717 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1718 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1719 u32 to_copy;
1720
1721 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1722 return -EINVAL;
1723
1724 if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1725 return -ENOENT;
1726
1727 if (unlikely(!br_stack))
1728 return -ENOENT;
1729
1730 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1731 return br_stack->nr * br_entry_size;
1732
1733 if (!buf || (size % br_entry_size != 0))
1734 return -EINVAL;
1735
1736 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1737 memcpy(buf, br_stack->entries, to_copy);
1738
1739 return to_copy;
1740}
1741
1742static const struct bpf_func_proto bpf_read_branch_records_proto = {
1743 .func = bpf_read_branch_records,
1744 .gpl_only = true,
1745 .ret_type = RET_INTEGER,
1746 .arg1_type = ARG_PTR_TO_CTX,
1747 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1748 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1749 .arg4_type = ARG_ANYTHING,
1750};
1751
1752static const struct bpf_func_proto *
1753pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1754{
1755 switch (func_id) {
1756 case BPF_FUNC_perf_event_output:
1757 return &bpf_perf_event_output_proto_tp;
1758 case BPF_FUNC_get_stackid:
1759 return &bpf_get_stackid_proto_pe;
1760 case BPF_FUNC_get_stack:
1761 return &bpf_get_stack_proto_pe;
1762 case BPF_FUNC_perf_prog_read_value:
1763 return &bpf_perf_prog_read_value_proto;
1764 case BPF_FUNC_read_branch_records:
1765 return &bpf_read_branch_records_proto;
1766 case BPF_FUNC_get_attach_cookie:
1767 return &bpf_get_attach_cookie_proto_pe;
1768 default:
1769 return bpf_tracing_func_proto(func_id, prog);
1770 }
1771}
1772
1773/*
1774 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1775 * to avoid potential recursive reuse issue when/if tracepoints are added
1776 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1777 *
1778 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1779 * in normal, irq, and nmi context.
1780 */
1781struct bpf_raw_tp_regs {
1782 struct pt_regs regs[3];
1783};
1784static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1785static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1786static struct pt_regs *get_bpf_raw_tp_regs(void)
1787{
1788 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1789 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1790
1791 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1792 this_cpu_dec(bpf_raw_tp_nest_level);
1793 return ERR_PTR(-EBUSY);
1794 }
1795
1796 return &tp_regs->regs[nest_level - 1];
1797}
1798
1799static void put_bpf_raw_tp_regs(void)
1800{
1801 this_cpu_dec(bpf_raw_tp_nest_level);
1802}
1803
1804BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1805 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1806{
1807 struct pt_regs *regs = get_bpf_raw_tp_regs();
1808 int ret;
1809
1810 if (IS_ERR(regs))
1811 return PTR_ERR(regs);
1812
1813 perf_fetch_caller_regs(regs);
1814 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1815
1816 put_bpf_raw_tp_regs();
1817 return ret;
1818}
1819
1820static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1821 .func = bpf_perf_event_output_raw_tp,
1822 .gpl_only = true,
1823 .ret_type = RET_INTEGER,
1824 .arg1_type = ARG_PTR_TO_CTX,
1825 .arg2_type = ARG_CONST_MAP_PTR,
1826 .arg3_type = ARG_ANYTHING,
1827 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1828 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1829};
1830
1831extern const struct bpf_func_proto bpf_skb_output_proto;
1832extern const struct bpf_func_proto bpf_xdp_output_proto;
1833extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1834
1835BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1836 struct bpf_map *, map, u64, flags)
1837{
1838 struct pt_regs *regs = get_bpf_raw_tp_regs();
1839 int ret;
1840
1841 if (IS_ERR(regs))
1842 return PTR_ERR(regs);
1843
1844 perf_fetch_caller_regs(regs);
1845 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1846 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1847 flags, 0, 0);
1848 put_bpf_raw_tp_regs();
1849 return ret;
1850}
1851
1852static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1853 .func = bpf_get_stackid_raw_tp,
1854 .gpl_only = true,
1855 .ret_type = RET_INTEGER,
1856 .arg1_type = ARG_PTR_TO_CTX,
1857 .arg2_type = ARG_CONST_MAP_PTR,
1858 .arg3_type = ARG_ANYTHING,
1859};
1860
1861BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1862 void *, buf, u32, size, u64, flags)
1863{
1864 struct pt_regs *regs = get_bpf_raw_tp_regs();
1865 int ret;
1866
1867 if (IS_ERR(regs))
1868 return PTR_ERR(regs);
1869
1870 perf_fetch_caller_regs(regs);
1871 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1872 (unsigned long) size, flags, 0);
1873 put_bpf_raw_tp_regs();
1874 return ret;
1875}
1876
1877static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1878 .func = bpf_get_stack_raw_tp,
1879 .gpl_only = true,
1880 .ret_type = RET_INTEGER,
1881 .arg1_type = ARG_PTR_TO_CTX,
1882 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1883 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1884 .arg4_type = ARG_ANYTHING,
1885};
1886
1887static const struct bpf_func_proto *
1888raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1889{
1890 switch (func_id) {
1891 case BPF_FUNC_perf_event_output:
1892 return &bpf_perf_event_output_proto_raw_tp;
1893 case BPF_FUNC_get_stackid:
1894 return &bpf_get_stackid_proto_raw_tp;
1895 case BPF_FUNC_get_stack:
1896 return &bpf_get_stack_proto_raw_tp;
1897 default:
1898 return bpf_tracing_func_proto(func_id, prog);
1899 }
1900}
1901
1902const struct bpf_func_proto *
1903tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1904{
1905 const struct bpf_func_proto *fn;
1906
1907 switch (func_id) {
1908#ifdef CONFIG_NET
1909 case BPF_FUNC_skb_output:
1910 return &bpf_skb_output_proto;
1911 case BPF_FUNC_xdp_output:
1912 return &bpf_xdp_output_proto;
1913 case BPF_FUNC_skc_to_tcp6_sock:
1914 return &bpf_skc_to_tcp6_sock_proto;
1915 case BPF_FUNC_skc_to_tcp_sock:
1916 return &bpf_skc_to_tcp_sock_proto;
1917 case BPF_FUNC_skc_to_tcp_timewait_sock:
1918 return &bpf_skc_to_tcp_timewait_sock_proto;
1919 case BPF_FUNC_skc_to_tcp_request_sock:
1920 return &bpf_skc_to_tcp_request_sock_proto;
1921 case BPF_FUNC_skc_to_udp6_sock:
1922 return &bpf_skc_to_udp6_sock_proto;
1923 case BPF_FUNC_skc_to_unix_sock:
1924 return &bpf_skc_to_unix_sock_proto;
1925 case BPF_FUNC_skc_to_mptcp_sock:
1926 return &bpf_skc_to_mptcp_sock_proto;
1927 case BPF_FUNC_sk_storage_get:
1928 return &bpf_sk_storage_get_tracing_proto;
1929 case BPF_FUNC_sk_storage_delete:
1930 return &bpf_sk_storage_delete_tracing_proto;
1931 case BPF_FUNC_sock_from_file:
1932 return &bpf_sock_from_file_proto;
1933 case BPF_FUNC_get_socket_cookie:
1934 return &bpf_get_socket_ptr_cookie_proto;
1935 case BPF_FUNC_xdp_get_buff_len:
1936 return &bpf_xdp_get_buff_len_trace_proto;
1937#endif
1938 case BPF_FUNC_seq_printf:
1939 return prog->expected_attach_type == BPF_TRACE_ITER ?
1940 &bpf_seq_printf_proto :
1941 NULL;
1942 case BPF_FUNC_seq_write:
1943 return prog->expected_attach_type == BPF_TRACE_ITER ?
1944 &bpf_seq_write_proto :
1945 NULL;
1946 case BPF_FUNC_seq_printf_btf:
1947 return prog->expected_attach_type == BPF_TRACE_ITER ?
1948 &bpf_seq_printf_btf_proto :
1949 NULL;
1950 case BPF_FUNC_d_path:
1951 return &bpf_d_path_proto;
1952 case BPF_FUNC_get_func_arg:
1953 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1954 case BPF_FUNC_get_func_ret:
1955 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1956 case BPF_FUNC_get_func_arg_cnt:
1957 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1958 case BPF_FUNC_get_attach_cookie:
1959 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1960 default:
1961 fn = raw_tp_prog_func_proto(func_id, prog);
1962 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1963 fn = bpf_iter_get_func_proto(func_id, prog);
1964 return fn;
1965 }
1966}
1967
1968static bool raw_tp_prog_is_valid_access(int off, int size,
1969 enum bpf_access_type type,
1970 const struct bpf_prog *prog,
1971 struct bpf_insn_access_aux *info)
1972{
1973 return bpf_tracing_ctx_access(off, size, type);
1974}
1975
1976static bool tracing_prog_is_valid_access(int off, int size,
1977 enum bpf_access_type type,
1978 const struct bpf_prog *prog,
1979 struct bpf_insn_access_aux *info)
1980{
1981 return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1982}
1983
1984int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1985 const union bpf_attr *kattr,
1986 union bpf_attr __user *uattr)
1987{
1988 return -ENOTSUPP;
1989}
1990
1991const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1992 .get_func_proto = raw_tp_prog_func_proto,
1993 .is_valid_access = raw_tp_prog_is_valid_access,
1994};
1995
1996const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1997#ifdef CONFIG_NET
1998 .test_run = bpf_prog_test_run_raw_tp,
1999#endif
2000};
2001
2002const struct bpf_verifier_ops tracing_verifier_ops = {
2003 .get_func_proto = tracing_prog_func_proto,
2004 .is_valid_access = tracing_prog_is_valid_access,
2005};
2006
2007const struct bpf_prog_ops tracing_prog_ops = {
2008 .test_run = bpf_prog_test_run_tracing,
2009};
2010
2011static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2012 enum bpf_access_type type,
2013 const struct bpf_prog *prog,
2014 struct bpf_insn_access_aux *info)
2015{
2016 if (off == 0) {
2017 if (size != sizeof(u64) || type != BPF_READ)
2018 return false;
2019 info->reg_type = PTR_TO_TP_BUFFER;
2020 }
2021 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2022}
2023
2024const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2025 .get_func_proto = raw_tp_prog_func_proto,
2026 .is_valid_access = raw_tp_writable_prog_is_valid_access,
2027};
2028
2029const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2030};
2031
2032static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2033 const struct bpf_prog *prog,
2034 struct bpf_insn_access_aux *info)
2035{
2036 const int size_u64 = sizeof(u64);
2037
2038 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2039 return false;
2040 if (type != BPF_READ)
2041 return false;
2042 if (off % size != 0) {
2043 if (sizeof(unsigned long) != 4)
2044 return false;
2045 if (size != 8)
2046 return false;
2047 if (off % size != 4)
2048 return false;
2049 }
2050
2051 switch (off) {
2052 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2053 bpf_ctx_record_field_size(info, size_u64);
2054 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2055 return false;
2056 break;
2057 case bpf_ctx_range(struct bpf_perf_event_data, addr):
2058 bpf_ctx_record_field_size(info, size_u64);
2059 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2060 return false;
2061 break;
2062 default:
2063 if (size != sizeof(long))
2064 return false;
2065 }
2066
2067 return true;
2068}
2069
2070static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2071 const struct bpf_insn *si,
2072 struct bpf_insn *insn_buf,
2073 struct bpf_prog *prog, u32 *target_size)
2074{
2075 struct bpf_insn *insn = insn_buf;
2076
2077 switch (si->off) {
2078 case offsetof(struct bpf_perf_event_data, sample_period):
2079 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2080 data), si->dst_reg, si->src_reg,
2081 offsetof(struct bpf_perf_event_data_kern, data));
2082 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2083 bpf_target_off(struct perf_sample_data, period, 8,
2084 target_size));
2085 break;
2086 case offsetof(struct bpf_perf_event_data, addr):
2087 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2088 data), si->dst_reg, si->src_reg,
2089 offsetof(struct bpf_perf_event_data_kern, data));
2090 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2091 bpf_target_off(struct perf_sample_data, addr, 8,
2092 target_size));
2093 break;
2094 default:
2095 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2096 regs), si->dst_reg, si->src_reg,
2097 offsetof(struct bpf_perf_event_data_kern, regs));
2098 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2099 si->off);
2100 break;
2101 }
2102
2103 return insn - insn_buf;
2104}
2105
2106const struct bpf_verifier_ops perf_event_verifier_ops = {
2107 .get_func_proto = pe_prog_func_proto,
2108 .is_valid_access = pe_prog_is_valid_access,
2109 .convert_ctx_access = pe_prog_convert_ctx_access,
2110};
2111
2112const struct bpf_prog_ops perf_event_prog_ops = {
2113};
2114
2115static DEFINE_MUTEX(bpf_event_mutex);
2116
2117#define BPF_TRACE_MAX_PROGS 64
2118
2119int perf_event_attach_bpf_prog(struct perf_event *event,
2120 struct bpf_prog *prog,
2121 u64 bpf_cookie)
2122{
2123 struct bpf_prog_array *old_array;
2124 struct bpf_prog_array *new_array;
2125 int ret = -EEXIST;
2126
2127 /*
2128 * Kprobe override only works if they are on the function entry,
2129 * and only if they are on the opt-in list.
2130 */
2131 if (prog->kprobe_override &&
2132 (!trace_kprobe_on_func_entry(event->tp_event) ||
2133 !trace_kprobe_error_injectable(event->tp_event)))
2134 return -EINVAL;
2135
2136 mutex_lock(&bpf_event_mutex);
2137
2138 if (event->prog)
2139 goto unlock;
2140
2141 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2142 if (old_array &&
2143 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2144 ret = -E2BIG;
2145 goto unlock;
2146 }
2147
2148 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2149 if (ret < 0)
2150 goto unlock;
2151
2152 /* set the new array to event->tp_event and set event->prog */
2153 event->prog = prog;
2154 event->bpf_cookie = bpf_cookie;
2155 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2156 bpf_prog_array_free_sleepable(old_array);
2157
2158unlock:
2159 mutex_unlock(&bpf_event_mutex);
2160 return ret;
2161}
2162
2163void perf_event_detach_bpf_prog(struct perf_event *event)
2164{
2165 struct bpf_prog_array *old_array;
2166 struct bpf_prog_array *new_array;
2167 int ret;
2168
2169 mutex_lock(&bpf_event_mutex);
2170
2171 if (!event->prog)
2172 goto unlock;
2173
2174 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2175 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2176 if (ret == -ENOENT)
2177 goto unlock;
2178 if (ret < 0) {
2179 bpf_prog_array_delete_safe(old_array, event->prog);
2180 } else {
2181 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2182 bpf_prog_array_free_sleepable(old_array);
2183 }
2184
2185 bpf_prog_put(event->prog);
2186 event->prog = NULL;
2187
2188unlock:
2189 mutex_unlock(&bpf_event_mutex);
2190}
2191
2192int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2193{
2194 struct perf_event_query_bpf __user *uquery = info;
2195 struct perf_event_query_bpf query = {};
2196 struct bpf_prog_array *progs;
2197 u32 *ids, prog_cnt, ids_len;
2198 int ret;
2199
2200 if (!perfmon_capable())
2201 return -EPERM;
2202 if (event->attr.type != PERF_TYPE_TRACEPOINT)
2203 return -EINVAL;
2204 if (copy_from_user(&query, uquery, sizeof(query)))
2205 return -EFAULT;
2206
2207 ids_len = query.ids_len;
2208 if (ids_len > BPF_TRACE_MAX_PROGS)
2209 return -E2BIG;
2210 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2211 if (!ids)
2212 return -ENOMEM;
2213 /*
2214 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2215 * is required when user only wants to check for uquery->prog_cnt.
2216 * There is no need to check for it since the case is handled
2217 * gracefully in bpf_prog_array_copy_info.
2218 */
2219
2220 mutex_lock(&bpf_event_mutex);
2221 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2222 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2223 mutex_unlock(&bpf_event_mutex);
2224
2225 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2226 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2227 ret = -EFAULT;
2228
2229 kfree(ids);
2230 return ret;
2231}
2232
2233extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2234extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2235
2236struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2237{
2238 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2239
2240 for (; btp < __stop__bpf_raw_tp; btp++) {
2241 if (!strcmp(btp->tp->name, name))
2242 return btp;
2243 }
2244
2245 return bpf_get_raw_tracepoint_module(name);
2246}
2247
2248void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2249{
2250 struct module *mod;
2251
2252 preempt_disable();
2253 mod = __module_address((unsigned long)btp);
2254 module_put(mod);
2255 preempt_enable();
2256}
2257
2258static __always_inline
2259void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2260{
2261 cant_sleep();
2262 if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2263 bpf_prog_inc_misses_counter(prog);
2264 goto out;
2265 }
2266 rcu_read_lock();
2267 (void) bpf_prog_run(prog, args);
2268 rcu_read_unlock();
2269out:
2270 this_cpu_dec(*(prog->active));
2271}
2272
2273#define UNPACK(...) __VA_ARGS__
2274#define REPEAT_1(FN, DL, X, ...) FN(X)
2275#define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2276#define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2277#define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2278#define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2279#define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2280#define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2281#define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2282#define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2283#define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2284#define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2285#define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2286#define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2287
2288#define SARG(X) u64 arg##X
2289#define COPY(X) args[X] = arg##X
2290
2291#define __DL_COM (,)
2292#define __DL_SEM (;)
2293
2294#define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2295
2296#define BPF_TRACE_DEFN_x(x) \
2297 void bpf_trace_run##x(struct bpf_prog *prog, \
2298 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2299 { \
2300 u64 args[x]; \
2301 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2302 __bpf_trace_run(prog, args); \
2303 } \
2304 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2305BPF_TRACE_DEFN_x(1);
2306BPF_TRACE_DEFN_x(2);
2307BPF_TRACE_DEFN_x(3);
2308BPF_TRACE_DEFN_x(4);
2309BPF_TRACE_DEFN_x(5);
2310BPF_TRACE_DEFN_x(6);
2311BPF_TRACE_DEFN_x(7);
2312BPF_TRACE_DEFN_x(8);
2313BPF_TRACE_DEFN_x(9);
2314BPF_TRACE_DEFN_x(10);
2315BPF_TRACE_DEFN_x(11);
2316BPF_TRACE_DEFN_x(12);
2317
2318static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2319{
2320 struct tracepoint *tp = btp->tp;
2321
2322 /*
2323 * check that program doesn't access arguments beyond what's
2324 * available in this tracepoint
2325 */
2326 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2327 return -EINVAL;
2328
2329 if (prog->aux->max_tp_access > btp->writable_size)
2330 return -EINVAL;
2331
2332 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2333 prog);
2334}
2335
2336int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2337{
2338 return __bpf_probe_register(btp, prog);
2339}
2340
2341int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2342{
2343 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2344}
2345
2346int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2347 u32 *fd_type, const char **buf,
2348 u64 *probe_offset, u64 *probe_addr)
2349{
2350 bool is_tracepoint, is_syscall_tp;
2351 struct bpf_prog *prog;
2352 int flags, err = 0;
2353
2354 prog = event->prog;
2355 if (!prog)
2356 return -ENOENT;
2357
2358 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2359 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2360 return -EOPNOTSUPP;
2361
2362 *prog_id = prog->aux->id;
2363 flags = event->tp_event->flags;
2364 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2365 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2366
2367 if (is_tracepoint || is_syscall_tp) {
2368 *buf = is_tracepoint ? event->tp_event->tp->name
2369 : event->tp_event->name;
2370 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2371 *probe_offset = 0x0;
2372 *probe_addr = 0x0;
2373 } else {
2374 /* kprobe/uprobe */
2375 err = -EOPNOTSUPP;
2376#ifdef CONFIG_KPROBE_EVENTS
2377 if (flags & TRACE_EVENT_FL_KPROBE)
2378 err = bpf_get_kprobe_info(event, fd_type, buf,
2379 probe_offset, probe_addr,
2380 event->attr.type == PERF_TYPE_TRACEPOINT);
2381#endif
2382#ifdef CONFIG_UPROBE_EVENTS
2383 if (flags & TRACE_EVENT_FL_UPROBE)
2384 err = bpf_get_uprobe_info(event, fd_type, buf,
2385 probe_offset,
2386 event->attr.type == PERF_TYPE_TRACEPOINT);
2387#endif
2388 }
2389
2390 return err;
2391}
2392
2393static int __init send_signal_irq_work_init(void)
2394{
2395 int cpu;
2396 struct send_signal_irq_work *work;
2397
2398 for_each_possible_cpu(cpu) {
2399 work = per_cpu_ptr(&send_signal_work, cpu);
2400 init_irq_work(&work->irq_work, do_bpf_send_signal);
2401 }
2402 return 0;
2403}
2404
2405subsys_initcall(send_signal_irq_work_init);
2406
2407#ifdef CONFIG_MODULES
2408static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2409 void *module)
2410{
2411 struct bpf_trace_module *btm, *tmp;
2412 struct module *mod = module;
2413 int ret = 0;
2414
2415 if (mod->num_bpf_raw_events == 0 ||
2416 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2417 goto out;
2418
2419 mutex_lock(&bpf_module_mutex);
2420
2421 switch (op) {
2422 case MODULE_STATE_COMING:
2423 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2424 if (btm) {
2425 btm->module = module;
2426 list_add(&btm->list, &bpf_trace_modules);
2427 } else {
2428 ret = -ENOMEM;
2429 }
2430 break;
2431 case MODULE_STATE_GOING:
2432 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2433 if (btm->module == module) {
2434 list_del(&btm->list);
2435 kfree(btm);
2436 break;
2437 }
2438 }
2439 break;
2440 }
2441
2442 mutex_unlock(&bpf_module_mutex);
2443
2444out:
2445 return notifier_from_errno(ret);
2446}
2447
2448static struct notifier_block bpf_module_nb = {
2449 .notifier_call = bpf_event_notify,
2450};
2451
2452static int __init bpf_event_init(void)
2453{
2454 register_module_notifier(&bpf_module_nb);
2455 return 0;
2456}
2457
2458fs_initcall(bpf_event_init);
2459#endif /* CONFIG_MODULES */
2460
2461#ifdef CONFIG_FPROBE
2462struct bpf_kprobe_multi_link {
2463 struct bpf_link link;
2464 struct fprobe fp;
2465 unsigned long *addrs;
2466 u64 *cookies;
2467 u32 cnt;
2468 u32 mods_cnt;
2469 struct module **mods;
2470};
2471
2472struct bpf_kprobe_multi_run_ctx {
2473 struct bpf_run_ctx run_ctx;
2474 struct bpf_kprobe_multi_link *link;
2475 unsigned long entry_ip;
2476};
2477
2478struct user_syms {
2479 const char **syms;
2480 char *buf;
2481};
2482
2483static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2484{
2485 unsigned long __user usymbol;
2486 const char **syms = NULL;
2487 char *buf = NULL, *p;
2488 int err = -ENOMEM;
2489 unsigned int i;
2490
2491 syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2492 if (!syms)
2493 goto error;
2494
2495 buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2496 if (!buf)
2497 goto error;
2498
2499 for (p = buf, i = 0; i < cnt; i++) {
2500 if (__get_user(usymbol, usyms + i)) {
2501 err = -EFAULT;
2502 goto error;
2503 }
2504 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2505 if (err == KSYM_NAME_LEN)
2506 err = -E2BIG;
2507 if (err < 0)
2508 goto error;
2509 syms[i] = p;
2510 p += err + 1;
2511 }
2512
2513 us->syms = syms;
2514 us->buf = buf;
2515 return 0;
2516
2517error:
2518 if (err) {
2519 kvfree(syms);
2520 kvfree(buf);
2521 }
2522 return err;
2523}
2524
2525static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2526{
2527 u32 i;
2528
2529 for (i = 0; i < cnt; i++)
2530 module_put(mods[i]);
2531}
2532
2533static void free_user_syms(struct user_syms *us)
2534{
2535 kvfree(us->syms);
2536 kvfree(us->buf);
2537}
2538
2539static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2540{
2541 struct bpf_kprobe_multi_link *kmulti_link;
2542
2543 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2544 unregister_fprobe(&kmulti_link->fp);
2545 kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2546}
2547
2548static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2549{
2550 struct bpf_kprobe_multi_link *kmulti_link;
2551
2552 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2553 kvfree(kmulti_link->addrs);
2554 kvfree(kmulti_link->cookies);
2555 kfree(kmulti_link->mods);
2556 kfree(kmulti_link);
2557}
2558
2559static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2560 .release = bpf_kprobe_multi_link_release,
2561 .dealloc = bpf_kprobe_multi_link_dealloc,
2562};
2563
2564static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2565{
2566 const struct bpf_kprobe_multi_link *link = priv;
2567 unsigned long *addr_a = a, *addr_b = b;
2568 u64 *cookie_a, *cookie_b;
2569
2570 cookie_a = link->cookies + (addr_a - link->addrs);
2571 cookie_b = link->cookies + (addr_b - link->addrs);
2572
2573 /* swap addr_a/addr_b and cookie_a/cookie_b values */
2574 swap(*addr_a, *addr_b);
2575 swap(*cookie_a, *cookie_b);
2576}
2577
2578static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2579{
2580 const unsigned long *addr_a = a, *addr_b = b;
2581
2582 if (*addr_a == *addr_b)
2583 return 0;
2584 return *addr_a < *addr_b ? -1 : 1;
2585}
2586
2587static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2588{
2589 return bpf_kprobe_multi_addrs_cmp(a, b);
2590}
2591
2592static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2593{
2594 struct bpf_kprobe_multi_run_ctx *run_ctx;
2595 struct bpf_kprobe_multi_link *link;
2596 u64 *cookie, entry_ip;
2597 unsigned long *addr;
2598
2599 if (WARN_ON_ONCE(!ctx))
2600 return 0;
2601 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2602 link = run_ctx->link;
2603 if (!link->cookies)
2604 return 0;
2605 entry_ip = run_ctx->entry_ip;
2606 addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2607 bpf_kprobe_multi_addrs_cmp);
2608 if (!addr)
2609 return 0;
2610 cookie = link->cookies + (addr - link->addrs);
2611 return *cookie;
2612}
2613
2614static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2615{
2616 struct bpf_kprobe_multi_run_ctx *run_ctx;
2617
2618 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2619 return run_ctx->entry_ip;
2620}
2621
2622static int
2623kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2624 unsigned long entry_ip, struct pt_regs *regs)
2625{
2626 struct bpf_kprobe_multi_run_ctx run_ctx = {
2627 .link = link,
2628 .entry_ip = entry_ip,
2629 };
2630 struct bpf_run_ctx *old_run_ctx;
2631 int err;
2632
2633 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2634 err = 0;
2635 goto out;
2636 }
2637
2638 migrate_disable();
2639 rcu_read_lock();
2640 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2641 err = bpf_prog_run(link->link.prog, regs);
2642 bpf_reset_run_ctx(old_run_ctx);
2643 rcu_read_unlock();
2644 migrate_enable();
2645
2646 out:
2647 __this_cpu_dec(bpf_prog_active);
2648 return err;
2649}
2650
2651static void
2652kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2653 struct pt_regs *regs)
2654{
2655 struct bpf_kprobe_multi_link *link;
2656
2657 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2658 kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2659}
2660
2661static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2662{
2663 const char **str_a = (const char **) a;
2664 const char **str_b = (const char **) b;
2665
2666 return strcmp(*str_a, *str_b);
2667}
2668
2669struct multi_symbols_sort {
2670 const char **funcs;
2671 u64 *cookies;
2672};
2673
2674static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2675{
2676 const struct multi_symbols_sort *data = priv;
2677 const char **name_a = a, **name_b = b;
2678
2679 swap(*name_a, *name_b);
2680
2681 /* If defined, swap also related cookies. */
2682 if (data->cookies) {
2683 u64 *cookie_a, *cookie_b;
2684
2685 cookie_a = data->cookies + (name_a - data->funcs);
2686 cookie_b = data->cookies + (name_b - data->funcs);
2687 swap(*cookie_a, *cookie_b);
2688 }
2689}
2690
2691struct module_addr_args {
2692 unsigned long *addrs;
2693 u32 addrs_cnt;
2694 struct module **mods;
2695 int mods_cnt;
2696 int mods_cap;
2697};
2698
2699static int module_callback(void *data, const char *name,
2700 struct module *mod, unsigned long addr)
2701{
2702 struct module_addr_args *args = data;
2703 struct module **mods;
2704
2705 /* We iterate all modules symbols and for each we:
2706 * - search for it in provided addresses array
2707 * - if found we check if we already have the module pointer stored
2708 * (we iterate modules sequentially, so we can check just the last
2709 * module pointer)
2710 * - take module reference and store it
2711 */
2712 if (!bsearch(&addr, args->addrs, args->addrs_cnt, sizeof(addr),
2713 bpf_kprobe_multi_addrs_cmp))
2714 return 0;
2715
2716 if (args->mods && args->mods[args->mods_cnt - 1] == mod)
2717 return 0;
2718
2719 if (args->mods_cnt == args->mods_cap) {
2720 args->mods_cap = max(16, args->mods_cap * 3 / 2);
2721 mods = krealloc_array(args->mods, args->mods_cap, sizeof(*mods), GFP_KERNEL);
2722 if (!mods)
2723 return -ENOMEM;
2724 args->mods = mods;
2725 }
2726
2727 if (!try_module_get(mod))
2728 return -EINVAL;
2729
2730 args->mods[args->mods_cnt] = mod;
2731 args->mods_cnt++;
2732 return 0;
2733}
2734
2735static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2736{
2737 struct module_addr_args args = {
2738 .addrs = addrs,
2739 .addrs_cnt = addrs_cnt,
2740 };
2741 int err;
2742
2743 /* We return either err < 0 in case of error, ... */
2744 err = module_kallsyms_on_each_symbol(module_callback, &args);
2745 if (err) {
2746 kprobe_multi_put_modules(args.mods, args.mods_cnt);
2747 kfree(args.mods);
2748 return err;
2749 }
2750
2751 /* or number of modules found if everything is ok. */
2752 *mods = args.mods;
2753 return args.mods_cnt;
2754}
2755
2756int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2757{
2758 struct bpf_kprobe_multi_link *link = NULL;
2759 struct bpf_link_primer link_primer;
2760 void __user *ucookies;
2761 unsigned long *addrs;
2762 u32 flags, cnt, size;
2763 void __user *uaddrs;
2764 u64 *cookies = NULL;
2765 void __user *usyms;
2766 int err;
2767
2768 /* no support for 32bit archs yet */
2769 if (sizeof(u64) != sizeof(void *))
2770 return -EOPNOTSUPP;
2771
2772 if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2773 return -EINVAL;
2774
2775 flags = attr->link_create.kprobe_multi.flags;
2776 if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2777 return -EINVAL;
2778
2779 uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2780 usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2781 if (!!uaddrs == !!usyms)
2782 return -EINVAL;
2783
2784 cnt = attr->link_create.kprobe_multi.cnt;
2785 if (!cnt)
2786 return -EINVAL;
2787
2788 size = cnt * sizeof(*addrs);
2789 addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2790 if (!addrs)
2791 return -ENOMEM;
2792
2793 ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2794 if (ucookies) {
2795 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2796 if (!cookies) {
2797 err = -ENOMEM;
2798 goto error;
2799 }
2800 if (copy_from_user(cookies, ucookies, size)) {
2801 err = -EFAULT;
2802 goto error;
2803 }
2804 }
2805
2806 if (uaddrs) {
2807 if (copy_from_user(addrs, uaddrs, size)) {
2808 err = -EFAULT;
2809 goto error;
2810 }
2811 } else {
2812 struct multi_symbols_sort data = {
2813 .cookies = cookies,
2814 };
2815 struct user_syms us;
2816
2817 err = copy_user_syms(&us, usyms, cnt);
2818 if (err)
2819 goto error;
2820
2821 if (cookies)
2822 data.funcs = us.syms;
2823
2824 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2825 symbols_swap_r, &data);
2826
2827 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2828 free_user_syms(&us);
2829 if (err)
2830 goto error;
2831 }
2832
2833 link = kzalloc(sizeof(*link), GFP_KERNEL);
2834 if (!link) {
2835 err = -ENOMEM;
2836 goto error;
2837 }
2838
2839 bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2840 &bpf_kprobe_multi_link_lops, prog);
2841
2842 err = bpf_link_prime(&link->link, &link_primer);
2843 if (err)
2844 goto error;
2845
2846 if (flags & BPF_F_KPROBE_MULTI_RETURN)
2847 link->fp.exit_handler = kprobe_multi_link_handler;
2848 else
2849 link->fp.entry_handler = kprobe_multi_link_handler;
2850
2851 link->addrs = addrs;
2852 link->cookies = cookies;
2853 link->cnt = cnt;
2854
2855 if (cookies) {
2856 /*
2857 * Sorting addresses will trigger sorting cookies as well
2858 * (check bpf_kprobe_multi_cookie_swap). This way we can
2859 * find cookie based on the address in bpf_get_attach_cookie
2860 * helper.
2861 */
2862 sort_r(addrs, cnt, sizeof(*addrs),
2863 bpf_kprobe_multi_cookie_cmp,
2864 bpf_kprobe_multi_cookie_swap,
2865 link);
2866 } else {
2867 /*
2868 * We need to sort addrs array even if there are no cookies
2869 * provided, to allow bsearch in get_modules_for_addrs.
2870 */
2871 sort(addrs, cnt, sizeof(*addrs),
2872 bpf_kprobe_multi_addrs_cmp, NULL);
2873 }
2874
2875 err = get_modules_for_addrs(&link->mods, addrs, cnt);
2876 if (err < 0) {
2877 bpf_link_cleanup(&link_primer);
2878 return err;
2879 }
2880 link->mods_cnt = err;
2881
2882 err = register_fprobe_ips(&link->fp, addrs, cnt);
2883 if (err) {
2884 kprobe_multi_put_modules(link->mods, link->mods_cnt);
2885 bpf_link_cleanup(&link_primer);
2886 return err;
2887 }
2888
2889 return bpf_link_settle(&link_primer);
2890
2891error:
2892 kfree(link);
2893 kvfree(addrs);
2894 kvfree(cookies);
2895 return err;
2896}
2897#else /* !CONFIG_FPROBE */
2898int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2899{
2900 return -EOPNOTSUPP;
2901}
2902static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2903{
2904 return 0;
2905}
2906static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2907{
2908 return 0;
2909}
2910#endif
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
4 */
5#include <linux/kernel.h>
6#include <linux/types.h>
7#include <linux/slab.h>
8#include <linux/bpf.h>
9#include <linux/bpf_perf_event.h>
10#include <linux/btf.h>
11#include <linux/filter.h>
12#include <linux/uaccess.h>
13#include <linux/ctype.h>
14#include <linux/kprobes.h>
15#include <linux/spinlock.h>
16#include <linux/syscalls.h>
17#include <linux/error-injection.h>
18#include <linux/btf_ids.h>
19#include <linux/bpf_lsm.h>
20
21#include <net/bpf_sk_storage.h>
22
23#include <uapi/linux/bpf.h>
24#include <uapi/linux/btf.h>
25
26#include <asm/tlb.h>
27
28#include "trace_probe.h"
29#include "trace.h"
30
31#define CREATE_TRACE_POINTS
32#include "bpf_trace.h"
33
34#define bpf_event_rcu_dereference(p) \
35 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
36
37#ifdef CONFIG_MODULES
38struct bpf_trace_module {
39 struct module *module;
40 struct list_head list;
41};
42
43static LIST_HEAD(bpf_trace_modules);
44static DEFINE_MUTEX(bpf_module_mutex);
45
46static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
47{
48 struct bpf_raw_event_map *btp, *ret = NULL;
49 struct bpf_trace_module *btm;
50 unsigned int i;
51
52 mutex_lock(&bpf_module_mutex);
53 list_for_each_entry(btm, &bpf_trace_modules, list) {
54 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
55 btp = &btm->module->bpf_raw_events[i];
56 if (!strcmp(btp->tp->name, name)) {
57 if (try_module_get(btm->module))
58 ret = btp;
59 goto out;
60 }
61 }
62 }
63out:
64 mutex_unlock(&bpf_module_mutex);
65 return ret;
66}
67#else
68static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
69{
70 return NULL;
71}
72#endif /* CONFIG_MODULES */
73
74u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
75u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
76
77static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
78 u64 flags, const struct btf **btf,
79 s32 *btf_id);
80
81/**
82 * trace_call_bpf - invoke BPF program
83 * @call: tracepoint event
84 * @ctx: opaque context pointer
85 *
86 * kprobe handlers execute BPF programs via this helper.
87 * Can be used from static tracepoints in the future.
88 *
89 * Return: BPF programs always return an integer which is interpreted by
90 * kprobe handler as:
91 * 0 - return from kprobe (event is filtered out)
92 * 1 - store kprobe event into ring buffer
93 * Other values are reserved and currently alias to 1
94 */
95unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
96{
97 unsigned int ret;
98
99 cant_sleep();
100
101 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
102 /*
103 * since some bpf program is already running on this cpu,
104 * don't call into another bpf program (same or different)
105 * and don't send kprobe event into ring-buffer,
106 * so return zero here
107 */
108 ret = 0;
109 goto out;
110 }
111
112 /*
113 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
114 * to all call sites, we did a bpf_prog_array_valid() there to check
115 * whether call->prog_array is empty or not, which is
116 * a heuristic to speed up execution.
117 *
118 * If bpf_prog_array_valid() fetched prog_array was
119 * non-NULL, we go into trace_call_bpf() and do the actual
120 * proper rcu_dereference() under RCU lock.
121 * If it turns out that prog_array is NULL then, we bail out.
122 * For the opposite, if the bpf_prog_array_valid() fetched pointer
123 * was NULL, you'll skip the prog_array with the risk of missing
124 * out of events when it was updated in between this and the
125 * rcu_dereference() which is accepted risk.
126 */
127 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
128
129 out:
130 __this_cpu_dec(bpf_prog_active);
131
132 return ret;
133}
134
135#ifdef CONFIG_BPF_KPROBE_OVERRIDE
136BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
137{
138 regs_set_return_value(regs, rc);
139 override_function_with_return(regs);
140 return 0;
141}
142
143static const struct bpf_func_proto bpf_override_return_proto = {
144 .func = bpf_override_return,
145 .gpl_only = true,
146 .ret_type = RET_INTEGER,
147 .arg1_type = ARG_PTR_TO_CTX,
148 .arg2_type = ARG_ANYTHING,
149};
150#endif
151
152static __always_inline int
153bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
154{
155 int ret;
156
157 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
158 if (unlikely(ret < 0))
159 memset(dst, 0, size);
160 return ret;
161}
162
163BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
164 const void __user *, unsafe_ptr)
165{
166 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
167}
168
169const struct bpf_func_proto bpf_probe_read_user_proto = {
170 .func = bpf_probe_read_user,
171 .gpl_only = true,
172 .ret_type = RET_INTEGER,
173 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
174 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
175 .arg3_type = ARG_ANYTHING,
176};
177
178static __always_inline int
179bpf_probe_read_user_str_common(void *dst, u32 size,
180 const void __user *unsafe_ptr)
181{
182 int ret;
183
184 /*
185 * NB: We rely on strncpy_from_user() not copying junk past the NUL
186 * terminator into `dst`.
187 *
188 * strncpy_from_user() does long-sized strides in the fast path. If the
189 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
190 * then there could be junk after the NUL in `dst`. If user takes `dst`
191 * and keys a hash map with it, then semantically identical strings can
192 * occupy multiple entries in the map.
193 */
194 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
195 if (unlikely(ret < 0))
196 memset(dst, 0, size);
197 return ret;
198}
199
200BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
201 const void __user *, unsafe_ptr)
202{
203 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
204}
205
206const struct bpf_func_proto bpf_probe_read_user_str_proto = {
207 .func = bpf_probe_read_user_str,
208 .gpl_only = true,
209 .ret_type = RET_INTEGER,
210 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
211 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
212 .arg3_type = ARG_ANYTHING,
213};
214
215static __always_inline int
216bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
217{
218 int ret;
219
220 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
221 if (unlikely(ret < 0))
222 memset(dst, 0, size);
223 return ret;
224}
225
226BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
227 const void *, unsafe_ptr)
228{
229 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
230}
231
232const struct bpf_func_proto bpf_probe_read_kernel_proto = {
233 .func = bpf_probe_read_kernel,
234 .gpl_only = true,
235 .ret_type = RET_INTEGER,
236 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
237 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
238 .arg3_type = ARG_ANYTHING,
239};
240
241static __always_inline int
242bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
243{
244 int ret;
245
246 /*
247 * The strncpy_from_kernel_nofault() call will likely not fill the
248 * entire buffer, but that's okay in this circumstance as we're probing
249 * arbitrary memory anyway similar to bpf_probe_read_*() and might
250 * as well probe the stack. Thus, memory is explicitly cleared
251 * only in error case, so that improper users ignoring return
252 * code altogether don't copy garbage; otherwise length of string
253 * is returned that can be used for bpf_perf_event_output() et al.
254 */
255 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
256 if (unlikely(ret < 0))
257 memset(dst, 0, size);
258 return ret;
259}
260
261BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
262 const void *, unsafe_ptr)
263{
264 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
265}
266
267const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
268 .func = bpf_probe_read_kernel_str,
269 .gpl_only = true,
270 .ret_type = RET_INTEGER,
271 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
272 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
273 .arg3_type = ARG_ANYTHING,
274};
275
276#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
277BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
278 const void *, unsafe_ptr)
279{
280 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
281 return bpf_probe_read_user_common(dst, size,
282 (__force void __user *)unsafe_ptr);
283 }
284 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
285}
286
287static const struct bpf_func_proto bpf_probe_read_compat_proto = {
288 .func = bpf_probe_read_compat,
289 .gpl_only = true,
290 .ret_type = RET_INTEGER,
291 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
292 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
293 .arg3_type = ARG_ANYTHING,
294};
295
296BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
297 const void *, unsafe_ptr)
298{
299 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
300 return bpf_probe_read_user_str_common(dst, size,
301 (__force void __user *)unsafe_ptr);
302 }
303 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
304}
305
306static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
307 .func = bpf_probe_read_compat_str,
308 .gpl_only = true,
309 .ret_type = RET_INTEGER,
310 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
311 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
312 .arg3_type = ARG_ANYTHING,
313};
314#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
315
316BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
317 u32, size)
318{
319 /*
320 * Ensure we're in user context which is safe for the helper to
321 * run. This helper has no business in a kthread.
322 *
323 * access_ok() should prevent writing to non-user memory, but in
324 * some situations (nommu, temporary switch, etc) access_ok() does
325 * not provide enough validation, hence the check on KERNEL_DS.
326 *
327 * nmi_uaccess_okay() ensures the probe is not run in an interim
328 * state, when the task or mm are switched. This is specifically
329 * required to prevent the use of temporary mm.
330 */
331
332 if (unlikely(in_interrupt() ||
333 current->flags & (PF_KTHREAD | PF_EXITING)))
334 return -EPERM;
335 if (unlikely(uaccess_kernel()))
336 return -EPERM;
337 if (unlikely(!nmi_uaccess_okay()))
338 return -EPERM;
339
340 return copy_to_user_nofault(unsafe_ptr, src, size);
341}
342
343static const struct bpf_func_proto bpf_probe_write_user_proto = {
344 .func = bpf_probe_write_user,
345 .gpl_only = true,
346 .ret_type = RET_INTEGER,
347 .arg1_type = ARG_ANYTHING,
348 .arg2_type = ARG_PTR_TO_MEM,
349 .arg3_type = ARG_CONST_SIZE,
350};
351
352static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
353{
354 if (!capable(CAP_SYS_ADMIN))
355 return NULL;
356
357 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
358 current->comm, task_pid_nr(current));
359
360 return &bpf_probe_write_user_proto;
361}
362
363static DEFINE_RAW_SPINLOCK(trace_printk_lock);
364
365#define MAX_TRACE_PRINTK_VARARGS 3
366#define BPF_TRACE_PRINTK_SIZE 1024
367
368BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
369 u64, arg2, u64, arg3)
370{
371 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
372 u32 *bin_args;
373 static char buf[BPF_TRACE_PRINTK_SIZE];
374 unsigned long flags;
375 int ret;
376
377 ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
378 MAX_TRACE_PRINTK_VARARGS);
379 if (ret < 0)
380 return ret;
381
382 raw_spin_lock_irqsave(&trace_printk_lock, flags);
383 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
384
385 trace_bpf_trace_printk(buf);
386 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
387
388 bpf_bprintf_cleanup();
389
390 return ret;
391}
392
393static const struct bpf_func_proto bpf_trace_printk_proto = {
394 .func = bpf_trace_printk,
395 .gpl_only = true,
396 .ret_type = RET_INTEGER,
397 .arg1_type = ARG_PTR_TO_MEM,
398 .arg2_type = ARG_CONST_SIZE,
399};
400
401const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
402{
403 /*
404 * This program might be calling bpf_trace_printk,
405 * so enable the associated bpf_trace/bpf_trace_printk event.
406 * Repeat this each time as it is possible a user has
407 * disabled bpf_trace_printk events. By loading a program
408 * calling bpf_trace_printk() however the user has expressed
409 * the intent to see such events.
410 */
411 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
412 pr_warn_ratelimited("could not enable bpf_trace_printk events");
413
414 return &bpf_trace_printk_proto;
415}
416
417#define MAX_SEQ_PRINTF_VARARGS 12
418
419BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
420 const void *, data, u32, data_len)
421{
422 int err, num_args;
423 u32 *bin_args;
424
425 if (data_len & 7 || data_len > MAX_SEQ_PRINTF_VARARGS * 8 ||
426 (data_len && !data))
427 return -EINVAL;
428 num_args = data_len / 8;
429
430 err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
431 if (err < 0)
432 return err;
433
434 seq_bprintf(m, fmt, bin_args);
435
436 bpf_bprintf_cleanup();
437
438 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
439}
440
441BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
442
443static const struct bpf_func_proto bpf_seq_printf_proto = {
444 .func = bpf_seq_printf,
445 .gpl_only = true,
446 .ret_type = RET_INTEGER,
447 .arg1_type = ARG_PTR_TO_BTF_ID,
448 .arg1_btf_id = &btf_seq_file_ids[0],
449 .arg2_type = ARG_PTR_TO_MEM,
450 .arg3_type = ARG_CONST_SIZE,
451 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
452 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
453};
454
455BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
456{
457 return seq_write(m, data, len) ? -EOVERFLOW : 0;
458}
459
460static const struct bpf_func_proto bpf_seq_write_proto = {
461 .func = bpf_seq_write,
462 .gpl_only = true,
463 .ret_type = RET_INTEGER,
464 .arg1_type = ARG_PTR_TO_BTF_ID,
465 .arg1_btf_id = &btf_seq_file_ids[0],
466 .arg2_type = ARG_PTR_TO_MEM,
467 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
468};
469
470BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
471 u32, btf_ptr_size, u64, flags)
472{
473 const struct btf *btf;
474 s32 btf_id;
475 int ret;
476
477 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
478 if (ret)
479 return ret;
480
481 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
482}
483
484static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
485 .func = bpf_seq_printf_btf,
486 .gpl_only = true,
487 .ret_type = RET_INTEGER,
488 .arg1_type = ARG_PTR_TO_BTF_ID,
489 .arg1_btf_id = &btf_seq_file_ids[0],
490 .arg2_type = ARG_PTR_TO_MEM,
491 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
492 .arg4_type = ARG_ANYTHING,
493};
494
495static __always_inline int
496get_map_perf_counter(struct bpf_map *map, u64 flags,
497 u64 *value, u64 *enabled, u64 *running)
498{
499 struct bpf_array *array = container_of(map, struct bpf_array, map);
500 unsigned int cpu = smp_processor_id();
501 u64 index = flags & BPF_F_INDEX_MASK;
502 struct bpf_event_entry *ee;
503
504 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
505 return -EINVAL;
506 if (index == BPF_F_CURRENT_CPU)
507 index = cpu;
508 if (unlikely(index >= array->map.max_entries))
509 return -E2BIG;
510
511 ee = READ_ONCE(array->ptrs[index]);
512 if (!ee)
513 return -ENOENT;
514
515 return perf_event_read_local(ee->event, value, enabled, running);
516}
517
518BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
519{
520 u64 value = 0;
521 int err;
522
523 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
524 /*
525 * this api is ugly since we miss [-22..-2] range of valid
526 * counter values, but that's uapi
527 */
528 if (err)
529 return err;
530 return value;
531}
532
533static const struct bpf_func_proto bpf_perf_event_read_proto = {
534 .func = bpf_perf_event_read,
535 .gpl_only = true,
536 .ret_type = RET_INTEGER,
537 .arg1_type = ARG_CONST_MAP_PTR,
538 .arg2_type = ARG_ANYTHING,
539};
540
541BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
542 struct bpf_perf_event_value *, buf, u32, size)
543{
544 int err = -EINVAL;
545
546 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
547 goto clear;
548 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
549 &buf->running);
550 if (unlikely(err))
551 goto clear;
552 return 0;
553clear:
554 memset(buf, 0, size);
555 return err;
556}
557
558static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
559 .func = bpf_perf_event_read_value,
560 .gpl_only = true,
561 .ret_type = RET_INTEGER,
562 .arg1_type = ARG_CONST_MAP_PTR,
563 .arg2_type = ARG_ANYTHING,
564 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
565 .arg4_type = ARG_CONST_SIZE,
566};
567
568static __always_inline u64
569__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
570 u64 flags, struct perf_sample_data *sd)
571{
572 struct bpf_array *array = container_of(map, struct bpf_array, map);
573 unsigned int cpu = smp_processor_id();
574 u64 index = flags & BPF_F_INDEX_MASK;
575 struct bpf_event_entry *ee;
576 struct perf_event *event;
577
578 if (index == BPF_F_CURRENT_CPU)
579 index = cpu;
580 if (unlikely(index >= array->map.max_entries))
581 return -E2BIG;
582
583 ee = READ_ONCE(array->ptrs[index]);
584 if (!ee)
585 return -ENOENT;
586
587 event = ee->event;
588 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
589 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
590 return -EINVAL;
591
592 if (unlikely(event->oncpu != cpu))
593 return -EOPNOTSUPP;
594
595 return perf_event_output(event, sd, regs);
596}
597
598/*
599 * Support executing tracepoints in normal, irq, and nmi context that each call
600 * bpf_perf_event_output
601 */
602struct bpf_trace_sample_data {
603 struct perf_sample_data sds[3];
604};
605
606static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
607static DEFINE_PER_CPU(int, bpf_trace_nest_level);
608BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
609 u64, flags, void *, data, u64, size)
610{
611 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
612 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
613 struct perf_raw_record raw = {
614 .frag = {
615 .size = size,
616 .data = data,
617 },
618 };
619 struct perf_sample_data *sd;
620 int err;
621
622 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
623 err = -EBUSY;
624 goto out;
625 }
626
627 sd = &sds->sds[nest_level - 1];
628
629 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
630 err = -EINVAL;
631 goto out;
632 }
633
634 perf_sample_data_init(sd, 0, 0);
635 sd->raw = &raw;
636
637 err = __bpf_perf_event_output(regs, map, flags, sd);
638
639out:
640 this_cpu_dec(bpf_trace_nest_level);
641 return err;
642}
643
644static const struct bpf_func_proto bpf_perf_event_output_proto = {
645 .func = bpf_perf_event_output,
646 .gpl_only = true,
647 .ret_type = RET_INTEGER,
648 .arg1_type = ARG_PTR_TO_CTX,
649 .arg2_type = ARG_CONST_MAP_PTR,
650 .arg3_type = ARG_ANYTHING,
651 .arg4_type = ARG_PTR_TO_MEM,
652 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
653};
654
655static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
656struct bpf_nested_pt_regs {
657 struct pt_regs regs[3];
658};
659static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
660static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
661
662u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
663 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
664{
665 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
666 struct perf_raw_frag frag = {
667 .copy = ctx_copy,
668 .size = ctx_size,
669 .data = ctx,
670 };
671 struct perf_raw_record raw = {
672 .frag = {
673 {
674 .next = ctx_size ? &frag : NULL,
675 },
676 .size = meta_size,
677 .data = meta,
678 },
679 };
680 struct perf_sample_data *sd;
681 struct pt_regs *regs;
682 u64 ret;
683
684 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
685 ret = -EBUSY;
686 goto out;
687 }
688 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
689 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
690
691 perf_fetch_caller_regs(regs);
692 perf_sample_data_init(sd, 0, 0);
693 sd->raw = &raw;
694
695 ret = __bpf_perf_event_output(regs, map, flags, sd);
696out:
697 this_cpu_dec(bpf_event_output_nest_level);
698 return ret;
699}
700
701BPF_CALL_0(bpf_get_current_task)
702{
703 return (long) current;
704}
705
706const struct bpf_func_proto bpf_get_current_task_proto = {
707 .func = bpf_get_current_task,
708 .gpl_only = true,
709 .ret_type = RET_INTEGER,
710};
711
712BPF_CALL_0(bpf_get_current_task_btf)
713{
714 return (unsigned long) current;
715}
716
717BTF_ID_LIST_SINGLE(bpf_get_current_btf_ids, struct, task_struct)
718
719static const struct bpf_func_proto bpf_get_current_task_btf_proto = {
720 .func = bpf_get_current_task_btf,
721 .gpl_only = true,
722 .ret_type = RET_PTR_TO_BTF_ID,
723 .ret_btf_id = &bpf_get_current_btf_ids[0],
724};
725
726BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
727{
728 struct bpf_array *array = container_of(map, struct bpf_array, map);
729 struct cgroup *cgrp;
730
731 if (unlikely(idx >= array->map.max_entries))
732 return -E2BIG;
733
734 cgrp = READ_ONCE(array->ptrs[idx]);
735 if (unlikely(!cgrp))
736 return -EAGAIN;
737
738 return task_under_cgroup_hierarchy(current, cgrp);
739}
740
741static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
742 .func = bpf_current_task_under_cgroup,
743 .gpl_only = false,
744 .ret_type = RET_INTEGER,
745 .arg1_type = ARG_CONST_MAP_PTR,
746 .arg2_type = ARG_ANYTHING,
747};
748
749struct send_signal_irq_work {
750 struct irq_work irq_work;
751 struct task_struct *task;
752 u32 sig;
753 enum pid_type type;
754};
755
756static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
757
758static void do_bpf_send_signal(struct irq_work *entry)
759{
760 struct send_signal_irq_work *work;
761
762 work = container_of(entry, struct send_signal_irq_work, irq_work);
763 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
764}
765
766static int bpf_send_signal_common(u32 sig, enum pid_type type)
767{
768 struct send_signal_irq_work *work = NULL;
769
770 /* Similar to bpf_probe_write_user, task needs to be
771 * in a sound condition and kernel memory access be
772 * permitted in order to send signal to the current
773 * task.
774 */
775 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
776 return -EPERM;
777 if (unlikely(uaccess_kernel()))
778 return -EPERM;
779 if (unlikely(!nmi_uaccess_okay()))
780 return -EPERM;
781
782 if (irqs_disabled()) {
783 /* Do an early check on signal validity. Otherwise,
784 * the error is lost in deferred irq_work.
785 */
786 if (unlikely(!valid_signal(sig)))
787 return -EINVAL;
788
789 work = this_cpu_ptr(&send_signal_work);
790 if (irq_work_is_busy(&work->irq_work))
791 return -EBUSY;
792
793 /* Add the current task, which is the target of sending signal,
794 * to the irq_work. The current task may change when queued
795 * irq works get executed.
796 */
797 work->task = current;
798 work->sig = sig;
799 work->type = type;
800 irq_work_queue(&work->irq_work);
801 return 0;
802 }
803
804 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
805}
806
807BPF_CALL_1(bpf_send_signal, u32, sig)
808{
809 return bpf_send_signal_common(sig, PIDTYPE_TGID);
810}
811
812static const struct bpf_func_proto bpf_send_signal_proto = {
813 .func = bpf_send_signal,
814 .gpl_only = false,
815 .ret_type = RET_INTEGER,
816 .arg1_type = ARG_ANYTHING,
817};
818
819BPF_CALL_1(bpf_send_signal_thread, u32, sig)
820{
821 return bpf_send_signal_common(sig, PIDTYPE_PID);
822}
823
824static const struct bpf_func_proto bpf_send_signal_thread_proto = {
825 .func = bpf_send_signal_thread,
826 .gpl_only = false,
827 .ret_type = RET_INTEGER,
828 .arg1_type = ARG_ANYTHING,
829};
830
831BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
832{
833 long len;
834 char *p;
835
836 if (!sz)
837 return 0;
838
839 p = d_path(path, buf, sz);
840 if (IS_ERR(p)) {
841 len = PTR_ERR(p);
842 } else {
843 len = buf + sz - p;
844 memmove(buf, p, len);
845 }
846
847 return len;
848}
849
850BTF_SET_START(btf_allowlist_d_path)
851#ifdef CONFIG_SECURITY
852BTF_ID(func, security_file_permission)
853BTF_ID(func, security_inode_getattr)
854BTF_ID(func, security_file_open)
855#endif
856#ifdef CONFIG_SECURITY_PATH
857BTF_ID(func, security_path_truncate)
858#endif
859BTF_ID(func, vfs_truncate)
860BTF_ID(func, vfs_fallocate)
861BTF_ID(func, dentry_open)
862BTF_ID(func, vfs_getattr)
863BTF_ID(func, filp_close)
864BTF_SET_END(btf_allowlist_d_path)
865
866static bool bpf_d_path_allowed(const struct bpf_prog *prog)
867{
868 if (prog->type == BPF_PROG_TYPE_TRACING &&
869 prog->expected_attach_type == BPF_TRACE_ITER)
870 return true;
871
872 if (prog->type == BPF_PROG_TYPE_LSM)
873 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
874
875 return btf_id_set_contains(&btf_allowlist_d_path,
876 prog->aux->attach_btf_id);
877}
878
879BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
880
881static const struct bpf_func_proto bpf_d_path_proto = {
882 .func = bpf_d_path,
883 .gpl_only = false,
884 .ret_type = RET_INTEGER,
885 .arg1_type = ARG_PTR_TO_BTF_ID,
886 .arg1_btf_id = &bpf_d_path_btf_ids[0],
887 .arg2_type = ARG_PTR_TO_MEM,
888 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
889 .allowed = bpf_d_path_allowed,
890};
891
892#define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
893 BTF_F_PTR_RAW | BTF_F_ZERO)
894
895static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
896 u64 flags, const struct btf **btf,
897 s32 *btf_id)
898{
899 const struct btf_type *t;
900
901 if (unlikely(flags & ~(BTF_F_ALL)))
902 return -EINVAL;
903
904 if (btf_ptr_size != sizeof(struct btf_ptr))
905 return -EINVAL;
906
907 *btf = bpf_get_btf_vmlinux();
908
909 if (IS_ERR_OR_NULL(*btf))
910 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
911
912 if (ptr->type_id > 0)
913 *btf_id = ptr->type_id;
914 else
915 return -EINVAL;
916
917 if (*btf_id > 0)
918 t = btf_type_by_id(*btf, *btf_id);
919 if (*btf_id <= 0 || !t)
920 return -ENOENT;
921
922 return 0;
923}
924
925BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
926 u32, btf_ptr_size, u64, flags)
927{
928 const struct btf *btf;
929 s32 btf_id;
930 int ret;
931
932 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
933 if (ret)
934 return ret;
935
936 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
937 flags);
938}
939
940const struct bpf_func_proto bpf_snprintf_btf_proto = {
941 .func = bpf_snprintf_btf,
942 .gpl_only = false,
943 .ret_type = RET_INTEGER,
944 .arg1_type = ARG_PTR_TO_MEM,
945 .arg2_type = ARG_CONST_SIZE,
946 .arg3_type = ARG_PTR_TO_MEM,
947 .arg4_type = ARG_CONST_SIZE,
948 .arg5_type = ARG_ANYTHING,
949};
950
951const struct bpf_func_proto *
952bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
953{
954 switch (func_id) {
955 case BPF_FUNC_map_lookup_elem:
956 return &bpf_map_lookup_elem_proto;
957 case BPF_FUNC_map_update_elem:
958 return &bpf_map_update_elem_proto;
959 case BPF_FUNC_map_delete_elem:
960 return &bpf_map_delete_elem_proto;
961 case BPF_FUNC_map_push_elem:
962 return &bpf_map_push_elem_proto;
963 case BPF_FUNC_map_pop_elem:
964 return &bpf_map_pop_elem_proto;
965 case BPF_FUNC_map_peek_elem:
966 return &bpf_map_peek_elem_proto;
967 case BPF_FUNC_ktime_get_ns:
968 return &bpf_ktime_get_ns_proto;
969 case BPF_FUNC_ktime_get_boot_ns:
970 return &bpf_ktime_get_boot_ns_proto;
971 case BPF_FUNC_ktime_get_coarse_ns:
972 return &bpf_ktime_get_coarse_ns_proto;
973 case BPF_FUNC_tail_call:
974 return &bpf_tail_call_proto;
975 case BPF_FUNC_get_current_pid_tgid:
976 return &bpf_get_current_pid_tgid_proto;
977 case BPF_FUNC_get_current_task:
978 return &bpf_get_current_task_proto;
979 case BPF_FUNC_get_current_task_btf:
980 return &bpf_get_current_task_btf_proto;
981 case BPF_FUNC_get_current_uid_gid:
982 return &bpf_get_current_uid_gid_proto;
983 case BPF_FUNC_get_current_comm:
984 return &bpf_get_current_comm_proto;
985 case BPF_FUNC_trace_printk:
986 return bpf_get_trace_printk_proto();
987 case BPF_FUNC_get_smp_processor_id:
988 return &bpf_get_smp_processor_id_proto;
989 case BPF_FUNC_get_numa_node_id:
990 return &bpf_get_numa_node_id_proto;
991 case BPF_FUNC_perf_event_read:
992 return &bpf_perf_event_read_proto;
993 case BPF_FUNC_current_task_under_cgroup:
994 return &bpf_current_task_under_cgroup_proto;
995 case BPF_FUNC_get_prandom_u32:
996 return &bpf_get_prandom_u32_proto;
997 case BPF_FUNC_probe_write_user:
998 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
999 NULL : bpf_get_probe_write_proto();
1000 case BPF_FUNC_probe_read_user:
1001 return &bpf_probe_read_user_proto;
1002 case BPF_FUNC_probe_read_kernel:
1003 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1004 NULL : &bpf_probe_read_kernel_proto;
1005 case BPF_FUNC_probe_read_user_str:
1006 return &bpf_probe_read_user_str_proto;
1007 case BPF_FUNC_probe_read_kernel_str:
1008 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1009 NULL : &bpf_probe_read_kernel_str_proto;
1010#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1011 case BPF_FUNC_probe_read:
1012 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1013 NULL : &bpf_probe_read_compat_proto;
1014 case BPF_FUNC_probe_read_str:
1015 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1016 NULL : &bpf_probe_read_compat_str_proto;
1017#endif
1018#ifdef CONFIG_CGROUPS
1019 case BPF_FUNC_get_current_cgroup_id:
1020 return &bpf_get_current_cgroup_id_proto;
1021 case BPF_FUNC_get_current_ancestor_cgroup_id:
1022 return &bpf_get_current_ancestor_cgroup_id_proto;
1023#endif
1024 case BPF_FUNC_send_signal:
1025 return &bpf_send_signal_proto;
1026 case BPF_FUNC_send_signal_thread:
1027 return &bpf_send_signal_thread_proto;
1028 case BPF_FUNC_perf_event_read_value:
1029 return &bpf_perf_event_read_value_proto;
1030 case BPF_FUNC_get_ns_current_pid_tgid:
1031 return &bpf_get_ns_current_pid_tgid_proto;
1032 case BPF_FUNC_ringbuf_output:
1033 return &bpf_ringbuf_output_proto;
1034 case BPF_FUNC_ringbuf_reserve:
1035 return &bpf_ringbuf_reserve_proto;
1036 case BPF_FUNC_ringbuf_submit:
1037 return &bpf_ringbuf_submit_proto;
1038 case BPF_FUNC_ringbuf_discard:
1039 return &bpf_ringbuf_discard_proto;
1040 case BPF_FUNC_ringbuf_query:
1041 return &bpf_ringbuf_query_proto;
1042 case BPF_FUNC_jiffies64:
1043 return &bpf_jiffies64_proto;
1044 case BPF_FUNC_get_task_stack:
1045 return &bpf_get_task_stack_proto;
1046 case BPF_FUNC_copy_from_user:
1047 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1048 case BPF_FUNC_snprintf_btf:
1049 return &bpf_snprintf_btf_proto;
1050 case BPF_FUNC_per_cpu_ptr:
1051 return &bpf_per_cpu_ptr_proto;
1052 case BPF_FUNC_this_cpu_ptr:
1053 return &bpf_this_cpu_ptr_proto;
1054 case BPF_FUNC_task_storage_get:
1055 return &bpf_task_storage_get_proto;
1056 case BPF_FUNC_task_storage_delete:
1057 return &bpf_task_storage_delete_proto;
1058 case BPF_FUNC_for_each_map_elem:
1059 return &bpf_for_each_map_elem_proto;
1060 case BPF_FUNC_snprintf:
1061 return &bpf_snprintf_proto;
1062 default:
1063 return NULL;
1064 }
1065}
1066
1067static const struct bpf_func_proto *
1068kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1069{
1070 switch (func_id) {
1071 case BPF_FUNC_perf_event_output:
1072 return &bpf_perf_event_output_proto;
1073 case BPF_FUNC_get_stackid:
1074 return &bpf_get_stackid_proto;
1075 case BPF_FUNC_get_stack:
1076 return &bpf_get_stack_proto;
1077#ifdef CONFIG_BPF_KPROBE_OVERRIDE
1078 case BPF_FUNC_override_return:
1079 return &bpf_override_return_proto;
1080#endif
1081 default:
1082 return bpf_tracing_func_proto(func_id, prog);
1083 }
1084}
1085
1086/* bpf+kprobe programs can access fields of 'struct pt_regs' */
1087static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1088 const struct bpf_prog *prog,
1089 struct bpf_insn_access_aux *info)
1090{
1091 if (off < 0 || off >= sizeof(struct pt_regs))
1092 return false;
1093 if (type != BPF_READ)
1094 return false;
1095 if (off % size != 0)
1096 return false;
1097 /*
1098 * Assertion for 32 bit to make sure last 8 byte access
1099 * (BPF_DW) to the last 4 byte member is disallowed.
1100 */
1101 if (off + size > sizeof(struct pt_regs))
1102 return false;
1103
1104 return true;
1105}
1106
1107const struct bpf_verifier_ops kprobe_verifier_ops = {
1108 .get_func_proto = kprobe_prog_func_proto,
1109 .is_valid_access = kprobe_prog_is_valid_access,
1110};
1111
1112const struct bpf_prog_ops kprobe_prog_ops = {
1113};
1114
1115BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1116 u64, flags, void *, data, u64, size)
1117{
1118 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1119
1120 /*
1121 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1122 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1123 * from there and call the same bpf_perf_event_output() helper inline.
1124 */
1125 return ____bpf_perf_event_output(regs, map, flags, data, size);
1126}
1127
1128static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1129 .func = bpf_perf_event_output_tp,
1130 .gpl_only = true,
1131 .ret_type = RET_INTEGER,
1132 .arg1_type = ARG_PTR_TO_CTX,
1133 .arg2_type = ARG_CONST_MAP_PTR,
1134 .arg3_type = ARG_ANYTHING,
1135 .arg4_type = ARG_PTR_TO_MEM,
1136 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1137};
1138
1139BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1140 u64, flags)
1141{
1142 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1143
1144 /*
1145 * Same comment as in bpf_perf_event_output_tp(), only that this time
1146 * the other helper's function body cannot be inlined due to being
1147 * external, thus we need to call raw helper function.
1148 */
1149 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1150 flags, 0, 0);
1151}
1152
1153static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1154 .func = bpf_get_stackid_tp,
1155 .gpl_only = true,
1156 .ret_type = RET_INTEGER,
1157 .arg1_type = ARG_PTR_TO_CTX,
1158 .arg2_type = ARG_CONST_MAP_PTR,
1159 .arg3_type = ARG_ANYTHING,
1160};
1161
1162BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1163 u64, flags)
1164{
1165 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1166
1167 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1168 (unsigned long) size, flags, 0);
1169}
1170
1171static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1172 .func = bpf_get_stack_tp,
1173 .gpl_only = true,
1174 .ret_type = RET_INTEGER,
1175 .arg1_type = ARG_PTR_TO_CTX,
1176 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1177 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1178 .arg4_type = ARG_ANYTHING,
1179};
1180
1181static const struct bpf_func_proto *
1182tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1183{
1184 switch (func_id) {
1185 case BPF_FUNC_perf_event_output:
1186 return &bpf_perf_event_output_proto_tp;
1187 case BPF_FUNC_get_stackid:
1188 return &bpf_get_stackid_proto_tp;
1189 case BPF_FUNC_get_stack:
1190 return &bpf_get_stack_proto_tp;
1191 default:
1192 return bpf_tracing_func_proto(func_id, prog);
1193 }
1194}
1195
1196static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1197 const struct bpf_prog *prog,
1198 struct bpf_insn_access_aux *info)
1199{
1200 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1201 return false;
1202 if (type != BPF_READ)
1203 return false;
1204 if (off % size != 0)
1205 return false;
1206
1207 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1208 return true;
1209}
1210
1211const struct bpf_verifier_ops tracepoint_verifier_ops = {
1212 .get_func_proto = tp_prog_func_proto,
1213 .is_valid_access = tp_prog_is_valid_access,
1214};
1215
1216const struct bpf_prog_ops tracepoint_prog_ops = {
1217};
1218
1219BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1220 struct bpf_perf_event_value *, buf, u32, size)
1221{
1222 int err = -EINVAL;
1223
1224 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1225 goto clear;
1226 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1227 &buf->running);
1228 if (unlikely(err))
1229 goto clear;
1230 return 0;
1231clear:
1232 memset(buf, 0, size);
1233 return err;
1234}
1235
1236static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1237 .func = bpf_perf_prog_read_value,
1238 .gpl_only = true,
1239 .ret_type = RET_INTEGER,
1240 .arg1_type = ARG_PTR_TO_CTX,
1241 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1242 .arg3_type = ARG_CONST_SIZE,
1243};
1244
1245BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1246 void *, buf, u32, size, u64, flags)
1247{
1248#ifndef CONFIG_X86
1249 return -ENOENT;
1250#else
1251 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1252 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1253 u32 to_copy;
1254
1255 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1256 return -EINVAL;
1257
1258 if (unlikely(!br_stack))
1259 return -EINVAL;
1260
1261 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1262 return br_stack->nr * br_entry_size;
1263
1264 if (!buf || (size % br_entry_size != 0))
1265 return -EINVAL;
1266
1267 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1268 memcpy(buf, br_stack->entries, to_copy);
1269
1270 return to_copy;
1271#endif
1272}
1273
1274static const struct bpf_func_proto bpf_read_branch_records_proto = {
1275 .func = bpf_read_branch_records,
1276 .gpl_only = true,
1277 .ret_type = RET_INTEGER,
1278 .arg1_type = ARG_PTR_TO_CTX,
1279 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1280 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1281 .arg4_type = ARG_ANYTHING,
1282};
1283
1284static const struct bpf_func_proto *
1285pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1286{
1287 switch (func_id) {
1288 case BPF_FUNC_perf_event_output:
1289 return &bpf_perf_event_output_proto_tp;
1290 case BPF_FUNC_get_stackid:
1291 return &bpf_get_stackid_proto_pe;
1292 case BPF_FUNC_get_stack:
1293 return &bpf_get_stack_proto_pe;
1294 case BPF_FUNC_perf_prog_read_value:
1295 return &bpf_perf_prog_read_value_proto;
1296 case BPF_FUNC_read_branch_records:
1297 return &bpf_read_branch_records_proto;
1298 default:
1299 return bpf_tracing_func_proto(func_id, prog);
1300 }
1301}
1302
1303/*
1304 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1305 * to avoid potential recursive reuse issue when/if tracepoints are added
1306 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1307 *
1308 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1309 * in normal, irq, and nmi context.
1310 */
1311struct bpf_raw_tp_regs {
1312 struct pt_regs regs[3];
1313};
1314static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1315static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1316static struct pt_regs *get_bpf_raw_tp_regs(void)
1317{
1318 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1319 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1320
1321 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1322 this_cpu_dec(bpf_raw_tp_nest_level);
1323 return ERR_PTR(-EBUSY);
1324 }
1325
1326 return &tp_regs->regs[nest_level - 1];
1327}
1328
1329static void put_bpf_raw_tp_regs(void)
1330{
1331 this_cpu_dec(bpf_raw_tp_nest_level);
1332}
1333
1334BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1335 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1336{
1337 struct pt_regs *regs = get_bpf_raw_tp_regs();
1338 int ret;
1339
1340 if (IS_ERR(regs))
1341 return PTR_ERR(regs);
1342
1343 perf_fetch_caller_regs(regs);
1344 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1345
1346 put_bpf_raw_tp_regs();
1347 return ret;
1348}
1349
1350static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1351 .func = bpf_perf_event_output_raw_tp,
1352 .gpl_only = true,
1353 .ret_type = RET_INTEGER,
1354 .arg1_type = ARG_PTR_TO_CTX,
1355 .arg2_type = ARG_CONST_MAP_PTR,
1356 .arg3_type = ARG_ANYTHING,
1357 .arg4_type = ARG_PTR_TO_MEM,
1358 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1359};
1360
1361extern const struct bpf_func_proto bpf_skb_output_proto;
1362extern const struct bpf_func_proto bpf_xdp_output_proto;
1363
1364BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1365 struct bpf_map *, map, u64, flags)
1366{
1367 struct pt_regs *regs = get_bpf_raw_tp_regs();
1368 int ret;
1369
1370 if (IS_ERR(regs))
1371 return PTR_ERR(regs);
1372
1373 perf_fetch_caller_regs(regs);
1374 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1375 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1376 flags, 0, 0);
1377 put_bpf_raw_tp_regs();
1378 return ret;
1379}
1380
1381static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1382 .func = bpf_get_stackid_raw_tp,
1383 .gpl_only = true,
1384 .ret_type = RET_INTEGER,
1385 .arg1_type = ARG_PTR_TO_CTX,
1386 .arg2_type = ARG_CONST_MAP_PTR,
1387 .arg3_type = ARG_ANYTHING,
1388};
1389
1390BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1391 void *, buf, u32, size, u64, flags)
1392{
1393 struct pt_regs *regs = get_bpf_raw_tp_regs();
1394 int ret;
1395
1396 if (IS_ERR(regs))
1397 return PTR_ERR(regs);
1398
1399 perf_fetch_caller_regs(regs);
1400 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1401 (unsigned long) size, flags, 0);
1402 put_bpf_raw_tp_regs();
1403 return ret;
1404}
1405
1406static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1407 .func = bpf_get_stack_raw_tp,
1408 .gpl_only = true,
1409 .ret_type = RET_INTEGER,
1410 .arg1_type = ARG_PTR_TO_CTX,
1411 .arg2_type = ARG_PTR_TO_MEM,
1412 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1413 .arg4_type = ARG_ANYTHING,
1414};
1415
1416static const struct bpf_func_proto *
1417raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1418{
1419 switch (func_id) {
1420 case BPF_FUNC_perf_event_output:
1421 return &bpf_perf_event_output_proto_raw_tp;
1422 case BPF_FUNC_get_stackid:
1423 return &bpf_get_stackid_proto_raw_tp;
1424 case BPF_FUNC_get_stack:
1425 return &bpf_get_stack_proto_raw_tp;
1426 default:
1427 return bpf_tracing_func_proto(func_id, prog);
1428 }
1429}
1430
1431const struct bpf_func_proto *
1432tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1433{
1434 switch (func_id) {
1435#ifdef CONFIG_NET
1436 case BPF_FUNC_skb_output:
1437 return &bpf_skb_output_proto;
1438 case BPF_FUNC_xdp_output:
1439 return &bpf_xdp_output_proto;
1440 case BPF_FUNC_skc_to_tcp6_sock:
1441 return &bpf_skc_to_tcp6_sock_proto;
1442 case BPF_FUNC_skc_to_tcp_sock:
1443 return &bpf_skc_to_tcp_sock_proto;
1444 case BPF_FUNC_skc_to_tcp_timewait_sock:
1445 return &bpf_skc_to_tcp_timewait_sock_proto;
1446 case BPF_FUNC_skc_to_tcp_request_sock:
1447 return &bpf_skc_to_tcp_request_sock_proto;
1448 case BPF_FUNC_skc_to_udp6_sock:
1449 return &bpf_skc_to_udp6_sock_proto;
1450 case BPF_FUNC_sk_storage_get:
1451 return &bpf_sk_storage_get_tracing_proto;
1452 case BPF_FUNC_sk_storage_delete:
1453 return &bpf_sk_storage_delete_tracing_proto;
1454 case BPF_FUNC_sock_from_file:
1455 return &bpf_sock_from_file_proto;
1456 case BPF_FUNC_get_socket_cookie:
1457 return &bpf_get_socket_ptr_cookie_proto;
1458#endif
1459 case BPF_FUNC_seq_printf:
1460 return prog->expected_attach_type == BPF_TRACE_ITER ?
1461 &bpf_seq_printf_proto :
1462 NULL;
1463 case BPF_FUNC_seq_write:
1464 return prog->expected_attach_type == BPF_TRACE_ITER ?
1465 &bpf_seq_write_proto :
1466 NULL;
1467 case BPF_FUNC_seq_printf_btf:
1468 return prog->expected_attach_type == BPF_TRACE_ITER ?
1469 &bpf_seq_printf_btf_proto :
1470 NULL;
1471 case BPF_FUNC_d_path:
1472 return &bpf_d_path_proto;
1473 default:
1474 return raw_tp_prog_func_proto(func_id, prog);
1475 }
1476}
1477
1478static bool raw_tp_prog_is_valid_access(int off, int size,
1479 enum bpf_access_type type,
1480 const struct bpf_prog *prog,
1481 struct bpf_insn_access_aux *info)
1482{
1483 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1484 return false;
1485 if (type != BPF_READ)
1486 return false;
1487 if (off % size != 0)
1488 return false;
1489 return true;
1490}
1491
1492static bool tracing_prog_is_valid_access(int off, int size,
1493 enum bpf_access_type type,
1494 const struct bpf_prog *prog,
1495 struct bpf_insn_access_aux *info)
1496{
1497 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1498 return false;
1499 if (type != BPF_READ)
1500 return false;
1501 if (off % size != 0)
1502 return false;
1503 return btf_ctx_access(off, size, type, prog, info);
1504}
1505
1506int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1507 const union bpf_attr *kattr,
1508 union bpf_attr __user *uattr)
1509{
1510 return -ENOTSUPP;
1511}
1512
1513const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1514 .get_func_proto = raw_tp_prog_func_proto,
1515 .is_valid_access = raw_tp_prog_is_valid_access,
1516};
1517
1518const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1519#ifdef CONFIG_NET
1520 .test_run = bpf_prog_test_run_raw_tp,
1521#endif
1522};
1523
1524const struct bpf_verifier_ops tracing_verifier_ops = {
1525 .get_func_proto = tracing_prog_func_proto,
1526 .is_valid_access = tracing_prog_is_valid_access,
1527};
1528
1529const struct bpf_prog_ops tracing_prog_ops = {
1530 .test_run = bpf_prog_test_run_tracing,
1531};
1532
1533static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1534 enum bpf_access_type type,
1535 const struct bpf_prog *prog,
1536 struct bpf_insn_access_aux *info)
1537{
1538 if (off == 0) {
1539 if (size != sizeof(u64) || type != BPF_READ)
1540 return false;
1541 info->reg_type = PTR_TO_TP_BUFFER;
1542 }
1543 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1544}
1545
1546const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1547 .get_func_proto = raw_tp_prog_func_proto,
1548 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1549};
1550
1551const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1552};
1553
1554static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1555 const struct bpf_prog *prog,
1556 struct bpf_insn_access_aux *info)
1557{
1558 const int size_u64 = sizeof(u64);
1559
1560 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1561 return false;
1562 if (type != BPF_READ)
1563 return false;
1564 if (off % size != 0) {
1565 if (sizeof(unsigned long) != 4)
1566 return false;
1567 if (size != 8)
1568 return false;
1569 if (off % size != 4)
1570 return false;
1571 }
1572
1573 switch (off) {
1574 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1575 bpf_ctx_record_field_size(info, size_u64);
1576 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1577 return false;
1578 break;
1579 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1580 bpf_ctx_record_field_size(info, size_u64);
1581 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1582 return false;
1583 break;
1584 default:
1585 if (size != sizeof(long))
1586 return false;
1587 }
1588
1589 return true;
1590}
1591
1592static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1593 const struct bpf_insn *si,
1594 struct bpf_insn *insn_buf,
1595 struct bpf_prog *prog, u32 *target_size)
1596{
1597 struct bpf_insn *insn = insn_buf;
1598
1599 switch (si->off) {
1600 case offsetof(struct bpf_perf_event_data, sample_period):
1601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1602 data), si->dst_reg, si->src_reg,
1603 offsetof(struct bpf_perf_event_data_kern, data));
1604 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1605 bpf_target_off(struct perf_sample_data, period, 8,
1606 target_size));
1607 break;
1608 case offsetof(struct bpf_perf_event_data, addr):
1609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1610 data), si->dst_reg, si->src_reg,
1611 offsetof(struct bpf_perf_event_data_kern, data));
1612 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1613 bpf_target_off(struct perf_sample_data, addr, 8,
1614 target_size));
1615 break;
1616 default:
1617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1618 regs), si->dst_reg, si->src_reg,
1619 offsetof(struct bpf_perf_event_data_kern, regs));
1620 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1621 si->off);
1622 break;
1623 }
1624
1625 return insn - insn_buf;
1626}
1627
1628const struct bpf_verifier_ops perf_event_verifier_ops = {
1629 .get_func_proto = pe_prog_func_proto,
1630 .is_valid_access = pe_prog_is_valid_access,
1631 .convert_ctx_access = pe_prog_convert_ctx_access,
1632};
1633
1634const struct bpf_prog_ops perf_event_prog_ops = {
1635};
1636
1637static DEFINE_MUTEX(bpf_event_mutex);
1638
1639#define BPF_TRACE_MAX_PROGS 64
1640
1641int perf_event_attach_bpf_prog(struct perf_event *event,
1642 struct bpf_prog *prog)
1643{
1644 struct bpf_prog_array *old_array;
1645 struct bpf_prog_array *new_array;
1646 int ret = -EEXIST;
1647
1648 /*
1649 * Kprobe override only works if they are on the function entry,
1650 * and only if they are on the opt-in list.
1651 */
1652 if (prog->kprobe_override &&
1653 (!trace_kprobe_on_func_entry(event->tp_event) ||
1654 !trace_kprobe_error_injectable(event->tp_event)))
1655 return -EINVAL;
1656
1657 mutex_lock(&bpf_event_mutex);
1658
1659 if (event->prog)
1660 goto unlock;
1661
1662 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1663 if (old_array &&
1664 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1665 ret = -E2BIG;
1666 goto unlock;
1667 }
1668
1669 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1670 if (ret < 0)
1671 goto unlock;
1672
1673 /* set the new array to event->tp_event and set event->prog */
1674 event->prog = prog;
1675 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1676 bpf_prog_array_free(old_array);
1677
1678unlock:
1679 mutex_unlock(&bpf_event_mutex);
1680 return ret;
1681}
1682
1683void perf_event_detach_bpf_prog(struct perf_event *event)
1684{
1685 struct bpf_prog_array *old_array;
1686 struct bpf_prog_array *new_array;
1687 int ret;
1688
1689 mutex_lock(&bpf_event_mutex);
1690
1691 if (!event->prog)
1692 goto unlock;
1693
1694 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1695 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1696 if (ret == -ENOENT)
1697 goto unlock;
1698 if (ret < 0) {
1699 bpf_prog_array_delete_safe(old_array, event->prog);
1700 } else {
1701 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1702 bpf_prog_array_free(old_array);
1703 }
1704
1705 bpf_prog_put(event->prog);
1706 event->prog = NULL;
1707
1708unlock:
1709 mutex_unlock(&bpf_event_mutex);
1710}
1711
1712int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1713{
1714 struct perf_event_query_bpf __user *uquery = info;
1715 struct perf_event_query_bpf query = {};
1716 struct bpf_prog_array *progs;
1717 u32 *ids, prog_cnt, ids_len;
1718 int ret;
1719
1720 if (!perfmon_capable())
1721 return -EPERM;
1722 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1723 return -EINVAL;
1724 if (copy_from_user(&query, uquery, sizeof(query)))
1725 return -EFAULT;
1726
1727 ids_len = query.ids_len;
1728 if (ids_len > BPF_TRACE_MAX_PROGS)
1729 return -E2BIG;
1730 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1731 if (!ids)
1732 return -ENOMEM;
1733 /*
1734 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1735 * is required when user only wants to check for uquery->prog_cnt.
1736 * There is no need to check for it since the case is handled
1737 * gracefully in bpf_prog_array_copy_info.
1738 */
1739
1740 mutex_lock(&bpf_event_mutex);
1741 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1742 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1743 mutex_unlock(&bpf_event_mutex);
1744
1745 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1746 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1747 ret = -EFAULT;
1748
1749 kfree(ids);
1750 return ret;
1751}
1752
1753extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1754extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1755
1756struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1757{
1758 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1759
1760 for (; btp < __stop__bpf_raw_tp; btp++) {
1761 if (!strcmp(btp->tp->name, name))
1762 return btp;
1763 }
1764
1765 return bpf_get_raw_tracepoint_module(name);
1766}
1767
1768void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1769{
1770 struct module *mod;
1771
1772 preempt_disable();
1773 mod = __module_address((unsigned long)btp);
1774 module_put(mod);
1775 preempt_enable();
1776}
1777
1778static __always_inline
1779void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1780{
1781 cant_sleep();
1782 rcu_read_lock();
1783 (void) BPF_PROG_RUN(prog, args);
1784 rcu_read_unlock();
1785}
1786
1787#define UNPACK(...) __VA_ARGS__
1788#define REPEAT_1(FN, DL, X, ...) FN(X)
1789#define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1790#define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1791#define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1792#define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1793#define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1794#define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1795#define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1796#define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1797#define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1798#define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1799#define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1800#define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1801
1802#define SARG(X) u64 arg##X
1803#define COPY(X) args[X] = arg##X
1804
1805#define __DL_COM (,)
1806#define __DL_SEM (;)
1807
1808#define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1809
1810#define BPF_TRACE_DEFN_x(x) \
1811 void bpf_trace_run##x(struct bpf_prog *prog, \
1812 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1813 { \
1814 u64 args[x]; \
1815 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1816 __bpf_trace_run(prog, args); \
1817 } \
1818 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1819BPF_TRACE_DEFN_x(1);
1820BPF_TRACE_DEFN_x(2);
1821BPF_TRACE_DEFN_x(3);
1822BPF_TRACE_DEFN_x(4);
1823BPF_TRACE_DEFN_x(5);
1824BPF_TRACE_DEFN_x(6);
1825BPF_TRACE_DEFN_x(7);
1826BPF_TRACE_DEFN_x(8);
1827BPF_TRACE_DEFN_x(9);
1828BPF_TRACE_DEFN_x(10);
1829BPF_TRACE_DEFN_x(11);
1830BPF_TRACE_DEFN_x(12);
1831
1832static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1833{
1834 struct tracepoint *tp = btp->tp;
1835
1836 /*
1837 * check that program doesn't access arguments beyond what's
1838 * available in this tracepoint
1839 */
1840 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1841 return -EINVAL;
1842
1843 if (prog->aux->max_tp_access > btp->writable_size)
1844 return -EINVAL;
1845
1846 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
1847 prog);
1848}
1849
1850int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1851{
1852 return __bpf_probe_register(btp, prog);
1853}
1854
1855int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1856{
1857 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1858}
1859
1860int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1861 u32 *fd_type, const char **buf,
1862 u64 *probe_offset, u64 *probe_addr)
1863{
1864 bool is_tracepoint, is_syscall_tp;
1865 struct bpf_prog *prog;
1866 int flags, err = 0;
1867
1868 prog = event->prog;
1869 if (!prog)
1870 return -ENOENT;
1871
1872 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1873 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1874 return -EOPNOTSUPP;
1875
1876 *prog_id = prog->aux->id;
1877 flags = event->tp_event->flags;
1878 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1879 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1880
1881 if (is_tracepoint || is_syscall_tp) {
1882 *buf = is_tracepoint ? event->tp_event->tp->name
1883 : event->tp_event->name;
1884 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1885 *probe_offset = 0x0;
1886 *probe_addr = 0x0;
1887 } else {
1888 /* kprobe/uprobe */
1889 err = -EOPNOTSUPP;
1890#ifdef CONFIG_KPROBE_EVENTS
1891 if (flags & TRACE_EVENT_FL_KPROBE)
1892 err = bpf_get_kprobe_info(event, fd_type, buf,
1893 probe_offset, probe_addr,
1894 event->attr.type == PERF_TYPE_TRACEPOINT);
1895#endif
1896#ifdef CONFIG_UPROBE_EVENTS
1897 if (flags & TRACE_EVENT_FL_UPROBE)
1898 err = bpf_get_uprobe_info(event, fd_type, buf,
1899 probe_offset,
1900 event->attr.type == PERF_TYPE_TRACEPOINT);
1901#endif
1902 }
1903
1904 return err;
1905}
1906
1907static int __init send_signal_irq_work_init(void)
1908{
1909 int cpu;
1910 struct send_signal_irq_work *work;
1911
1912 for_each_possible_cpu(cpu) {
1913 work = per_cpu_ptr(&send_signal_work, cpu);
1914 init_irq_work(&work->irq_work, do_bpf_send_signal);
1915 }
1916 return 0;
1917}
1918
1919subsys_initcall(send_signal_irq_work_init);
1920
1921#ifdef CONFIG_MODULES
1922static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1923 void *module)
1924{
1925 struct bpf_trace_module *btm, *tmp;
1926 struct module *mod = module;
1927 int ret = 0;
1928
1929 if (mod->num_bpf_raw_events == 0 ||
1930 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1931 goto out;
1932
1933 mutex_lock(&bpf_module_mutex);
1934
1935 switch (op) {
1936 case MODULE_STATE_COMING:
1937 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1938 if (btm) {
1939 btm->module = module;
1940 list_add(&btm->list, &bpf_trace_modules);
1941 } else {
1942 ret = -ENOMEM;
1943 }
1944 break;
1945 case MODULE_STATE_GOING:
1946 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1947 if (btm->module == module) {
1948 list_del(&btm->list);
1949 kfree(btm);
1950 break;
1951 }
1952 }
1953 break;
1954 }
1955
1956 mutex_unlock(&bpf_module_mutex);
1957
1958out:
1959 return notifier_from_errno(ret);
1960}
1961
1962static struct notifier_block bpf_module_nb = {
1963 .notifier_call = bpf_event_notify,
1964};
1965
1966static int __init bpf_event_init(void)
1967{
1968 register_module_notifier(&bpf_module_nb);
1969 return 0;
1970}
1971
1972fs_initcall(bpf_event_init);
1973#endif /* CONFIG_MODULES */