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
20 #include <uapi/linux/bpf.h>
21 #include <uapi/linux/btf.h>
22
23 #include <asm/tlb.h>
24
25 #include "trace_probe.h"
26 #include "trace.h"
27
28 #define CREATE_TRACE_POINTS
29 #include "bpf_trace.h"
30
31 #define bpf_event_rcu_dereference(p) \
32 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
33
34 #ifdef CONFIG_MODULES
35 struct bpf_trace_module {
36 struct module *module;
37 struct list_head list;
38 };
39
40 static LIST_HEAD(bpf_trace_modules);
41 static DEFINE_MUTEX(bpf_module_mutex);
42
bpf_get_raw_tracepoint_module(const char * name)43 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
44 {
45 struct bpf_raw_event_map *btp, *ret = NULL;
46 struct bpf_trace_module *btm;
47 unsigned int i;
48
49 mutex_lock(&bpf_module_mutex);
50 list_for_each_entry(btm, &bpf_trace_modules, list) {
51 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
52 btp = &btm->module->bpf_raw_events[i];
53 if (!strcmp(btp->tp->name, name)) {
54 if (try_module_get(btm->module))
55 ret = btp;
56 goto out;
57 }
58 }
59 }
60 out:
61 mutex_unlock(&bpf_module_mutex);
62 return ret;
63 }
64 #else
bpf_get_raw_tracepoint_module(const char * name)65 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
66 {
67 return NULL;
68 }
69 #endif /* CONFIG_MODULES */
70
71 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
72 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
73
74 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
75 u64 flags, const struct btf **btf,
76 s32 *btf_id);
77
78 /**
79 * trace_call_bpf - invoke BPF program
80 * @call: tracepoint event
81 * @ctx: opaque context pointer
82 *
83 * kprobe handlers execute BPF programs via this helper.
84 * Can be used from static tracepoints in the future.
85 *
86 * Return: BPF programs always return an integer which is interpreted by
87 * kprobe handler as:
88 * 0 - return from kprobe (event is filtered out)
89 * 1 - store kprobe event into ring buffer
90 * Other values are reserved and currently alias to 1
91 */
trace_call_bpf(struct trace_event_call * call,void * ctx)92 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
93 {
94 unsigned int ret;
95
96 cant_sleep();
97
98 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
99 /*
100 * since some bpf program is already running on this cpu,
101 * don't call into another bpf program (same or different)
102 * and don't send kprobe event into ring-buffer,
103 * so return zero here
104 */
105 ret = 0;
106 goto out;
107 }
108
109 /*
110 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
111 * to all call sites, we did a bpf_prog_array_valid() there to check
112 * whether call->prog_array is empty or not, which is
113 * a heurisitc to speed up execution.
114 *
115 * If bpf_prog_array_valid() fetched prog_array was
116 * non-NULL, we go into trace_call_bpf() and do the actual
117 * proper rcu_dereference() under RCU lock.
118 * If it turns out that prog_array is NULL then, we bail out.
119 * For the opposite, if the bpf_prog_array_valid() fetched pointer
120 * was NULL, you'll skip the prog_array with the risk of missing
121 * out of events when it was updated in between this and the
122 * rcu_dereference() which is accepted risk.
123 */
124 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
125
126 out:
127 __this_cpu_dec(bpf_prog_active);
128
129 return ret;
130 }
131
132 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
BPF_CALL_2(bpf_override_return,struct pt_regs *,regs,unsigned long,rc)133 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
134 {
135 regs_set_return_value(regs, rc);
136 override_function_with_return(regs);
137 return 0;
138 }
139
140 static const struct bpf_func_proto bpf_override_return_proto = {
141 .func = bpf_override_return,
142 .gpl_only = true,
143 .ret_type = RET_INTEGER,
144 .arg1_type = ARG_PTR_TO_CTX,
145 .arg2_type = ARG_ANYTHING,
146 };
147 #endif
148
149 static __always_inline int
bpf_probe_read_user_common(void * dst,u32 size,const void __user * unsafe_ptr)150 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
151 {
152 int ret;
153
154 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
155 if (unlikely(ret < 0))
156 memset(dst, 0, size);
157 return ret;
158 }
159
BPF_CALL_3(bpf_probe_read_user,void *,dst,u32,size,const void __user *,unsafe_ptr)160 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
161 const void __user *, unsafe_ptr)
162 {
163 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
164 }
165
166 const struct bpf_func_proto bpf_probe_read_user_proto = {
167 .func = bpf_probe_read_user,
168 .gpl_only = true,
169 .ret_type = RET_INTEGER,
170 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
171 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
172 .arg3_type = ARG_ANYTHING,
173 };
174
175 static __always_inline int
bpf_probe_read_user_str_common(void * dst,u32 size,const void __user * unsafe_ptr)176 bpf_probe_read_user_str_common(void *dst, u32 size,
177 const void __user *unsafe_ptr)
178 {
179 int ret;
180
181 /*
182 * NB: We rely on strncpy_from_user() not copying junk past the NUL
183 * terminator into `dst`.
184 *
185 * strncpy_from_user() does long-sized strides in the fast path. If the
186 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
187 * then there could be junk after the NUL in `dst`. If user takes `dst`
188 * and keys a hash map with it, then semantically identical strings can
189 * occupy multiple entries in the map.
190 */
191 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
192 if (unlikely(ret < 0))
193 memset(dst, 0, size);
194 return ret;
195 }
196
BPF_CALL_3(bpf_probe_read_user_str,void *,dst,u32,size,const void __user *,unsafe_ptr)197 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
198 const void __user *, unsafe_ptr)
199 {
200 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
201 }
202
203 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
204 .func = bpf_probe_read_user_str,
205 .gpl_only = true,
206 .ret_type = RET_INTEGER,
207 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
208 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
209 .arg3_type = ARG_ANYTHING,
210 };
211
212 static __always_inline int
bpf_probe_read_kernel_common(void * dst,u32 size,const void * unsafe_ptr)213 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
214 {
215 int ret;
216
217 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
218 if (unlikely(ret < 0))
219 memset(dst, 0, size);
220 return ret;
221 }
222
BPF_CALL_3(bpf_probe_read_kernel,void *,dst,u32,size,const void *,unsafe_ptr)223 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
224 const void *, unsafe_ptr)
225 {
226 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
227 }
228
229 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
230 .func = bpf_probe_read_kernel,
231 .gpl_only = true,
232 .ret_type = RET_INTEGER,
233 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
234 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
235 .arg3_type = ARG_ANYTHING,
236 };
237
238 static __always_inline int
bpf_probe_read_kernel_str_common(void * dst,u32 size,const void * unsafe_ptr)239 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
240 {
241 int ret;
242
243 /*
244 * The strncpy_from_kernel_nofault() call will likely not fill the
245 * entire buffer, but that's okay in this circumstance as we're probing
246 * arbitrary memory anyway similar to bpf_probe_read_*() and might
247 * as well probe the stack. Thus, memory is explicitly cleared
248 * only in error case, so that improper users ignoring return
249 * code altogether don't copy garbage; otherwise length of string
250 * is returned that can be used for bpf_perf_event_output() et al.
251 */
252 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
253 if (unlikely(ret < 0))
254 memset(dst, 0, size);
255 return ret;
256 }
257
BPF_CALL_3(bpf_probe_read_kernel_str,void *,dst,u32,size,const void *,unsafe_ptr)258 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
259 const void *, unsafe_ptr)
260 {
261 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
262 }
263
264 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
265 .func = bpf_probe_read_kernel_str,
266 .gpl_only = true,
267 .ret_type = RET_INTEGER,
268 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
269 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
270 .arg3_type = ARG_ANYTHING,
271 };
272
273 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
BPF_CALL_3(bpf_probe_read_compat,void *,dst,u32,size,const void *,unsafe_ptr)274 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
275 const void *, unsafe_ptr)
276 {
277 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
278 return bpf_probe_read_user_common(dst, size,
279 (__force void __user *)unsafe_ptr);
280 }
281 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
282 }
283
284 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
285 .func = bpf_probe_read_compat,
286 .gpl_only = true,
287 .ret_type = RET_INTEGER,
288 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
289 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
290 .arg3_type = ARG_ANYTHING,
291 };
292
BPF_CALL_3(bpf_probe_read_compat_str,void *,dst,u32,size,const void *,unsafe_ptr)293 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
294 const void *, unsafe_ptr)
295 {
296 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
297 return bpf_probe_read_user_str_common(dst, size,
298 (__force void __user *)unsafe_ptr);
299 }
300 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
301 }
302
303 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
304 .func = bpf_probe_read_compat_str,
305 .gpl_only = true,
306 .ret_type = RET_INTEGER,
307 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
308 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
309 .arg3_type = ARG_ANYTHING,
310 };
311 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
312
BPF_CALL_3(bpf_probe_write_user,void __user *,unsafe_ptr,const void *,src,u32,size)313 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
314 u32, size)
315 {
316 /*
317 * Ensure we're in user context which is safe for the helper to
318 * run. This helper has no business in a kthread.
319 *
320 * access_ok() should prevent writing to non-user memory, but in
321 * some situations (nommu, temporary switch, etc) access_ok() does
322 * not provide enough validation, hence the check on KERNEL_DS.
323 *
324 * nmi_uaccess_okay() ensures the probe is not run in an interim
325 * state, when the task or mm are switched. This is specifically
326 * required to prevent the use of temporary mm.
327 */
328
329 if (unlikely(in_interrupt() ||
330 current->flags & (PF_KTHREAD | PF_EXITING)))
331 return -EPERM;
332 if (unlikely(uaccess_kernel()))
333 return -EPERM;
334 if (unlikely(!nmi_uaccess_okay()))
335 return -EPERM;
336
337 return copy_to_user_nofault(unsafe_ptr, src, size);
338 }
339
340 static const struct bpf_func_proto bpf_probe_write_user_proto = {
341 .func = bpf_probe_write_user,
342 .gpl_only = true,
343 .ret_type = RET_INTEGER,
344 .arg1_type = ARG_ANYTHING,
345 .arg2_type = ARG_PTR_TO_MEM,
346 .arg3_type = ARG_CONST_SIZE,
347 };
348
bpf_get_probe_write_proto(void)349 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
350 {
351 if (!capable(CAP_SYS_ADMIN))
352 return NULL;
353
354 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
355 current->comm, task_pid_nr(current));
356
357 return &bpf_probe_write_user_proto;
358 }
359
bpf_trace_copy_string(char * buf,void * unsafe_ptr,char fmt_ptype,size_t bufsz)360 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
361 size_t bufsz)
362 {
363 void __user *user_ptr = (__force void __user *)unsafe_ptr;
364
365 buf[0] = 0;
366
367 switch (fmt_ptype) {
368 case 's':
369 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
370 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
371 strncpy_from_user_nofault(buf, user_ptr, bufsz);
372 break;
373 }
374 fallthrough;
375 #endif
376 case 'k':
377 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
378 break;
379 case 'u':
380 strncpy_from_user_nofault(buf, user_ptr, bufsz);
381 break;
382 }
383 }
384
385 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
386
387 #define BPF_TRACE_PRINTK_SIZE 1024
388
bpf_do_trace_printk(const char * fmt,...)389 static __printf(1, 0) int bpf_do_trace_printk(const char *fmt, ...)
390 {
391 static char buf[BPF_TRACE_PRINTK_SIZE];
392 unsigned long flags;
393 va_list ap;
394 int ret;
395
396 raw_spin_lock_irqsave(&trace_printk_lock, flags);
397 va_start(ap, fmt);
398 ret = vsnprintf(buf, sizeof(buf), fmt, ap);
399 va_end(ap);
400 /* vsnprintf() will not append null for zero-length strings */
401 if (ret == 0)
402 buf[0] = '\0';
403 trace_bpf_trace_printk(buf);
404 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
405
406 return ret;
407 }
408
409 /*
410 * Only limited trace_printk() conversion specifiers allowed:
411 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pB %pks %pus %s
412 */
BPF_CALL_5(bpf_trace_printk,char *,fmt,u32,fmt_size,u64,arg1,u64,arg2,u64,arg3)413 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
414 u64, arg2, u64, arg3)
415 {
416 int i, mod[3] = {}, fmt_cnt = 0;
417 char buf[64], fmt_ptype;
418 void *unsafe_ptr = NULL;
419 bool str_seen = false;
420
421 /*
422 * bpf_check()->check_func_arg()->check_stack_boundary()
423 * guarantees that fmt points to bpf program stack,
424 * fmt_size bytes of it were initialized and fmt_size > 0
425 */
426 if (fmt[--fmt_size] != 0)
427 return -EINVAL;
428
429 /* check format string for allowed specifiers */
430 for (i = 0; i < fmt_size; i++) {
431 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
432 return -EINVAL;
433
434 if (fmt[i] != '%')
435 continue;
436
437 if (fmt_cnt >= 3)
438 return -EINVAL;
439
440 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
441 i++;
442 if (fmt[i] == 'l') {
443 mod[fmt_cnt]++;
444 i++;
445 } else if (fmt[i] == 'p') {
446 mod[fmt_cnt]++;
447 if ((fmt[i + 1] == 'k' ||
448 fmt[i + 1] == 'u') &&
449 fmt[i + 2] == 's') {
450 fmt_ptype = fmt[i + 1];
451 i += 2;
452 goto fmt_str;
453 }
454
455 if (fmt[i + 1] == 'B') {
456 i++;
457 goto fmt_next;
458 }
459
460 /* disallow any further format extensions */
461 if (fmt[i + 1] != 0 &&
462 !isspace(fmt[i + 1]) &&
463 !ispunct(fmt[i + 1]))
464 return -EINVAL;
465
466 goto fmt_next;
467 } else if (fmt[i] == 's') {
468 mod[fmt_cnt]++;
469 fmt_ptype = fmt[i];
470 fmt_str:
471 if (str_seen)
472 /* allow only one '%s' per fmt string */
473 return -EINVAL;
474 str_seen = true;
475
476 if (fmt[i + 1] != 0 &&
477 !isspace(fmt[i + 1]) &&
478 !ispunct(fmt[i + 1]))
479 return -EINVAL;
480
481 switch (fmt_cnt) {
482 case 0:
483 unsafe_ptr = (void *)(long)arg1;
484 arg1 = (long)buf;
485 break;
486 case 1:
487 unsafe_ptr = (void *)(long)arg2;
488 arg2 = (long)buf;
489 break;
490 case 2:
491 unsafe_ptr = (void *)(long)arg3;
492 arg3 = (long)buf;
493 break;
494 }
495
496 bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
497 sizeof(buf));
498 goto fmt_next;
499 }
500
501 if (fmt[i] == 'l') {
502 mod[fmt_cnt]++;
503 i++;
504 }
505
506 if (fmt[i] != 'i' && fmt[i] != 'd' &&
507 fmt[i] != 'u' && fmt[i] != 'x')
508 return -EINVAL;
509 fmt_next:
510 fmt_cnt++;
511 }
512
513 /* Horrid workaround for getting va_list handling working with different
514 * argument type combinations generically for 32 and 64 bit archs.
515 */
516 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
517 #define __BPF_TP(...) \
518 bpf_do_trace_printk(fmt, ##__VA_ARGS__)
519
520 #define __BPF_ARG1_TP(...) \
521 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
522 ? __BPF_TP(arg1, ##__VA_ARGS__) \
523 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
524 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
525 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
526
527 #define __BPF_ARG2_TP(...) \
528 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
529 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
530 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
531 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
532 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
533
534 #define __BPF_ARG3_TP(...) \
535 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
536 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
537 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
538 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
539 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
540
541 return __BPF_TP_EMIT();
542 }
543
544 static const struct bpf_func_proto bpf_trace_printk_proto = {
545 .func = bpf_trace_printk,
546 .gpl_only = true,
547 .ret_type = RET_INTEGER,
548 .arg1_type = ARG_PTR_TO_MEM,
549 .arg2_type = ARG_CONST_SIZE,
550 };
551
bpf_get_trace_printk_proto(void)552 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
553 {
554 /*
555 * This program might be calling bpf_trace_printk,
556 * so enable the associated bpf_trace/bpf_trace_printk event.
557 * Repeat this each time as it is possible a user has
558 * disabled bpf_trace_printk events. By loading a program
559 * calling bpf_trace_printk() however the user has expressed
560 * the intent to see such events.
561 */
562 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
563 pr_warn_ratelimited("could not enable bpf_trace_printk events");
564
565 return &bpf_trace_printk_proto;
566 }
567
568 #define MAX_SEQ_PRINTF_VARARGS 12
569 #define MAX_SEQ_PRINTF_MAX_MEMCPY 6
570 #define MAX_SEQ_PRINTF_STR_LEN 128
571
572 struct bpf_seq_printf_buf {
573 char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
574 };
575 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
576 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
577
BPF_CALL_5(bpf_seq_printf,struct seq_file *,m,char *,fmt,u32,fmt_size,const void *,data,u32,data_len)578 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
579 const void *, data, u32, data_len)
580 {
581 int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
582 int i, buf_used, copy_size, num_args;
583 u64 params[MAX_SEQ_PRINTF_VARARGS];
584 struct bpf_seq_printf_buf *bufs;
585 const u64 *args = data;
586
587 buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
588 if (WARN_ON_ONCE(buf_used > 1)) {
589 err = -EBUSY;
590 goto out;
591 }
592
593 bufs = this_cpu_ptr(&bpf_seq_printf_buf);
594
595 /*
596 * bpf_check()->check_func_arg()->check_stack_boundary()
597 * guarantees that fmt points to bpf program stack,
598 * fmt_size bytes of it were initialized and fmt_size > 0
599 */
600 if (fmt[--fmt_size] != 0)
601 goto out;
602
603 if (data_len & 7)
604 goto out;
605
606 for (i = 0; i < fmt_size; i++) {
607 if (fmt[i] == '%') {
608 if (fmt[i + 1] == '%')
609 i++;
610 else if (!data || !data_len)
611 goto out;
612 }
613 }
614
615 num_args = data_len / 8;
616
617 /* check format string for allowed specifiers */
618 for (i = 0; i < fmt_size; i++) {
619 /* only printable ascii for now. */
620 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
621 err = -EINVAL;
622 goto out;
623 }
624
625 if (fmt[i] != '%')
626 continue;
627
628 if (fmt[i + 1] == '%') {
629 i++;
630 continue;
631 }
632
633 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
634 err = -E2BIG;
635 goto out;
636 }
637
638 if (fmt_cnt >= num_args) {
639 err = -EINVAL;
640 goto out;
641 }
642
643 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
644 i++;
645
646 /* skip optional "[0 +-][num]" width formating field */
647 while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' ||
648 fmt[i] == ' ')
649 i++;
650 if (fmt[i] >= '1' && fmt[i] <= '9') {
651 i++;
652 while (fmt[i] >= '0' && fmt[i] <= '9')
653 i++;
654 }
655
656 if (fmt[i] == 's') {
657 void *unsafe_ptr;
658
659 /* try our best to copy */
660 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
661 err = -E2BIG;
662 goto out;
663 }
664
665 unsafe_ptr = (void *)(long)args[fmt_cnt];
666 err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
667 unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
668 if (err < 0)
669 bufs->buf[memcpy_cnt][0] = '\0';
670 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
671
672 fmt_cnt++;
673 memcpy_cnt++;
674 continue;
675 }
676
677 if (fmt[i] == 'p') {
678 if (fmt[i + 1] == 0 ||
679 fmt[i + 1] == 'K' ||
680 fmt[i + 1] == 'x' ||
681 fmt[i + 1] == 'B') {
682 /* just kernel pointers */
683 params[fmt_cnt] = args[fmt_cnt];
684 fmt_cnt++;
685 continue;
686 }
687
688 /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
689 if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
690 err = -EINVAL;
691 goto out;
692 }
693 if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
694 err = -EINVAL;
695 goto out;
696 }
697
698 if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
699 err = -E2BIG;
700 goto out;
701 }
702
703
704 copy_size = (fmt[i + 2] == '4') ? 4 : 16;
705
706 err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
707 (void *) (long) args[fmt_cnt],
708 copy_size);
709 if (err < 0)
710 memset(bufs->buf[memcpy_cnt], 0, copy_size);
711 params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
712
713 i += 2;
714 fmt_cnt++;
715 memcpy_cnt++;
716 continue;
717 }
718
719 if (fmt[i] == 'l') {
720 i++;
721 if (fmt[i] == 'l')
722 i++;
723 }
724
725 if (fmt[i] != 'i' && fmt[i] != 'd' &&
726 fmt[i] != 'u' && fmt[i] != 'x' &&
727 fmt[i] != 'X') {
728 err = -EINVAL;
729 goto out;
730 }
731
732 params[fmt_cnt] = args[fmt_cnt];
733 fmt_cnt++;
734 }
735
736 /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
737 * all of them to seq_printf().
738 */
739 seq_printf(m, fmt, params[0], params[1], params[2], params[3],
740 params[4], params[5], params[6], params[7], params[8],
741 params[9], params[10], params[11]);
742
743 err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
744 out:
745 this_cpu_dec(bpf_seq_printf_buf_used);
746 return err;
747 }
748
749 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
750
751 static const struct bpf_func_proto bpf_seq_printf_proto = {
752 .func = bpf_seq_printf,
753 .gpl_only = true,
754 .ret_type = RET_INTEGER,
755 .arg1_type = ARG_PTR_TO_BTF_ID,
756 .arg1_btf_id = &btf_seq_file_ids[0],
757 .arg2_type = ARG_PTR_TO_MEM,
758 .arg3_type = ARG_CONST_SIZE,
759 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
760 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
761 };
762
BPF_CALL_3(bpf_seq_write,struct seq_file *,m,const void *,data,u32,len)763 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
764 {
765 return seq_write(m, data, len) ? -EOVERFLOW : 0;
766 }
767
768 static const struct bpf_func_proto bpf_seq_write_proto = {
769 .func = bpf_seq_write,
770 .gpl_only = true,
771 .ret_type = RET_INTEGER,
772 .arg1_type = ARG_PTR_TO_BTF_ID,
773 .arg1_btf_id = &btf_seq_file_ids[0],
774 .arg2_type = ARG_PTR_TO_MEM,
775 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
776 };
777
BPF_CALL_4(bpf_seq_printf_btf,struct seq_file *,m,struct btf_ptr *,ptr,u32,btf_ptr_size,u64,flags)778 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
779 u32, btf_ptr_size, u64, flags)
780 {
781 const struct btf *btf;
782 s32 btf_id;
783 int ret;
784
785 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
786 if (ret)
787 return ret;
788
789 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
790 }
791
792 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
793 .func = bpf_seq_printf_btf,
794 .gpl_only = true,
795 .ret_type = RET_INTEGER,
796 .arg1_type = ARG_PTR_TO_BTF_ID,
797 .arg1_btf_id = &btf_seq_file_ids[0],
798 .arg2_type = ARG_PTR_TO_MEM,
799 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
800 .arg4_type = ARG_ANYTHING,
801 };
802
803 static __always_inline int
get_map_perf_counter(struct bpf_map * map,u64 flags,u64 * value,u64 * enabled,u64 * running)804 get_map_perf_counter(struct bpf_map *map, u64 flags,
805 u64 *value, u64 *enabled, u64 *running)
806 {
807 struct bpf_array *array = container_of(map, struct bpf_array, map);
808 unsigned int cpu = smp_processor_id();
809 u64 index = flags & BPF_F_INDEX_MASK;
810 struct bpf_event_entry *ee;
811
812 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
813 return -EINVAL;
814 if (index == BPF_F_CURRENT_CPU)
815 index = cpu;
816 if (unlikely(index >= array->map.max_entries))
817 return -E2BIG;
818
819 ee = READ_ONCE(array->ptrs[index]);
820 if (!ee)
821 return -ENOENT;
822
823 return perf_event_read_local(ee->event, value, enabled, running);
824 }
825
BPF_CALL_2(bpf_perf_event_read,struct bpf_map *,map,u64,flags)826 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
827 {
828 u64 value = 0;
829 int err;
830
831 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
832 /*
833 * this api is ugly since we miss [-22..-2] range of valid
834 * counter values, but that's uapi
835 */
836 if (err)
837 return err;
838 return value;
839 }
840
841 static const struct bpf_func_proto bpf_perf_event_read_proto = {
842 .func = bpf_perf_event_read,
843 .gpl_only = true,
844 .ret_type = RET_INTEGER,
845 .arg1_type = ARG_CONST_MAP_PTR,
846 .arg2_type = ARG_ANYTHING,
847 };
848
BPF_CALL_4(bpf_perf_event_read_value,struct bpf_map *,map,u64,flags,struct bpf_perf_event_value *,buf,u32,size)849 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
850 struct bpf_perf_event_value *, buf, u32, size)
851 {
852 int err = -EINVAL;
853
854 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
855 goto clear;
856 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
857 &buf->running);
858 if (unlikely(err))
859 goto clear;
860 return 0;
861 clear:
862 memset(buf, 0, size);
863 return err;
864 }
865
866 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
867 .func = bpf_perf_event_read_value,
868 .gpl_only = true,
869 .ret_type = RET_INTEGER,
870 .arg1_type = ARG_CONST_MAP_PTR,
871 .arg2_type = ARG_ANYTHING,
872 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
873 .arg4_type = ARG_CONST_SIZE,
874 };
875
876 static __always_inline u64
__bpf_perf_event_output(struct pt_regs * regs,struct bpf_map * map,u64 flags,struct perf_sample_data * sd)877 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
878 u64 flags, struct perf_sample_data *sd)
879 {
880 struct bpf_array *array = container_of(map, struct bpf_array, map);
881 unsigned int cpu = smp_processor_id();
882 u64 index = flags & BPF_F_INDEX_MASK;
883 struct bpf_event_entry *ee;
884 struct perf_event *event;
885
886 if (index == BPF_F_CURRENT_CPU)
887 index = cpu;
888 if (unlikely(index >= array->map.max_entries))
889 return -E2BIG;
890
891 ee = READ_ONCE(array->ptrs[index]);
892 if (!ee)
893 return -ENOENT;
894
895 event = ee->event;
896 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
897 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
898 return -EINVAL;
899
900 if (unlikely(event->oncpu != cpu))
901 return -EOPNOTSUPP;
902
903 return perf_event_output(event, sd, regs);
904 }
905
906 /*
907 * Support executing tracepoints in normal, irq, and nmi context that each call
908 * bpf_perf_event_output
909 */
910 struct bpf_trace_sample_data {
911 struct perf_sample_data sds[3];
912 };
913
914 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
915 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
BPF_CALL_5(bpf_perf_event_output,struct pt_regs *,regs,struct bpf_map *,map,u64,flags,void *,data,u64,size)916 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
917 u64, flags, void *, data, u64, size)
918 {
919 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
920 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
921 struct perf_raw_record raw = {
922 .frag = {
923 .size = size,
924 .data = data,
925 },
926 };
927 struct perf_sample_data *sd;
928 int err;
929
930 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
931 err = -EBUSY;
932 goto out;
933 }
934
935 sd = &sds->sds[nest_level - 1];
936
937 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
938 err = -EINVAL;
939 goto out;
940 }
941
942 perf_sample_data_init(sd, 0, 0);
943 sd->raw = &raw;
944
945 err = __bpf_perf_event_output(regs, map, flags, sd);
946
947 out:
948 this_cpu_dec(bpf_trace_nest_level);
949 return err;
950 }
951
952 static const struct bpf_func_proto bpf_perf_event_output_proto = {
953 .func = bpf_perf_event_output,
954 .gpl_only = true,
955 .ret_type = RET_INTEGER,
956 .arg1_type = ARG_PTR_TO_CTX,
957 .arg2_type = ARG_CONST_MAP_PTR,
958 .arg3_type = ARG_ANYTHING,
959 .arg4_type = ARG_PTR_TO_MEM,
960 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
961 };
962
963 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
964 struct bpf_nested_pt_regs {
965 struct pt_regs regs[3];
966 };
967 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
968 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
969
bpf_event_output(struct bpf_map * map,u64 flags,void * meta,u64 meta_size,void * ctx,u64 ctx_size,bpf_ctx_copy_t ctx_copy)970 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
971 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
972 {
973 struct perf_raw_frag frag = {
974 .copy = ctx_copy,
975 .size = ctx_size,
976 .data = ctx,
977 };
978 struct perf_raw_record raw = {
979 .frag = {
980 {
981 .next = ctx_size ? &frag : NULL,
982 },
983 .size = meta_size,
984 .data = meta,
985 },
986 };
987 struct perf_sample_data *sd;
988 struct pt_regs *regs;
989 int nest_level;
990 u64 ret;
991
992 preempt_disable();
993 nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
994
995 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
996 ret = -EBUSY;
997 goto out;
998 }
999 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
1000 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
1001
1002 perf_fetch_caller_regs(regs);
1003 perf_sample_data_init(sd, 0, 0);
1004 sd->raw = &raw;
1005
1006 ret = __bpf_perf_event_output(regs, map, flags, sd);
1007 out:
1008 this_cpu_dec(bpf_event_output_nest_level);
1009 preempt_enable();
1010 return ret;
1011 }
1012
BPF_CALL_0(bpf_get_current_task)1013 BPF_CALL_0(bpf_get_current_task)
1014 {
1015 return (long) current;
1016 }
1017
1018 const struct bpf_func_proto bpf_get_current_task_proto = {
1019 .func = bpf_get_current_task,
1020 .gpl_only = true,
1021 .ret_type = RET_INTEGER,
1022 };
1023
BPF_CALL_2(bpf_current_task_under_cgroup,struct bpf_map *,map,u32,idx)1024 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
1025 {
1026 struct bpf_array *array = container_of(map, struct bpf_array, map);
1027 struct cgroup *cgrp;
1028
1029 if (unlikely(idx >= array->map.max_entries))
1030 return -E2BIG;
1031
1032 cgrp = READ_ONCE(array->ptrs[idx]);
1033 if (unlikely(!cgrp))
1034 return -EAGAIN;
1035
1036 return task_under_cgroup_hierarchy(current, cgrp);
1037 }
1038
1039 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
1040 .func = bpf_current_task_under_cgroup,
1041 .gpl_only = false,
1042 .ret_type = RET_INTEGER,
1043 .arg1_type = ARG_CONST_MAP_PTR,
1044 .arg2_type = ARG_ANYTHING,
1045 };
1046
1047 struct send_signal_irq_work {
1048 struct irq_work irq_work;
1049 struct task_struct *task;
1050 u32 sig;
1051 enum pid_type type;
1052 };
1053
1054 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
1055
do_bpf_send_signal(struct irq_work * entry)1056 static void do_bpf_send_signal(struct irq_work *entry)
1057 {
1058 struct send_signal_irq_work *work;
1059
1060 work = container_of(entry, struct send_signal_irq_work, irq_work);
1061 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
1062 put_task_struct(work->task);
1063 }
1064
bpf_send_signal_common(u32 sig,enum pid_type type)1065 static int bpf_send_signal_common(u32 sig, enum pid_type type)
1066 {
1067 struct send_signal_irq_work *work = NULL;
1068
1069 /* Similar to bpf_probe_write_user, task needs to be
1070 * in a sound condition and kernel memory access be
1071 * permitted in order to send signal to the current
1072 * task.
1073 */
1074 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1075 return -EPERM;
1076 if (unlikely(uaccess_kernel()))
1077 return -EPERM;
1078 if (unlikely(!nmi_uaccess_okay()))
1079 return -EPERM;
1080 /* Task should not be pid=1 to avoid kernel panic. */
1081 if (unlikely(is_global_init(current)))
1082 return -EPERM;
1083
1084 if (irqs_disabled()) {
1085 /* Do an early check on signal validity. Otherwise,
1086 * the error is lost in deferred irq_work.
1087 */
1088 if (unlikely(!valid_signal(sig)))
1089 return -EINVAL;
1090
1091 work = this_cpu_ptr(&send_signal_work);
1092 if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
1093 return -EBUSY;
1094
1095 /* Add the current task, which is the target of sending signal,
1096 * to the irq_work. The current task may change when queued
1097 * irq works get executed.
1098 */
1099 work->task = get_task_struct(current);
1100 work->sig = sig;
1101 work->type = type;
1102 irq_work_queue(&work->irq_work);
1103 return 0;
1104 }
1105
1106 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1107 }
1108
BPF_CALL_1(bpf_send_signal,u32,sig)1109 BPF_CALL_1(bpf_send_signal, u32, sig)
1110 {
1111 return bpf_send_signal_common(sig, PIDTYPE_TGID);
1112 }
1113
1114 static const struct bpf_func_proto bpf_send_signal_proto = {
1115 .func = bpf_send_signal,
1116 .gpl_only = false,
1117 .ret_type = RET_INTEGER,
1118 .arg1_type = ARG_ANYTHING,
1119 };
1120
BPF_CALL_1(bpf_send_signal_thread,u32,sig)1121 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1122 {
1123 return bpf_send_signal_common(sig, PIDTYPE_PID);
1124 }
1125
1126 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1127 .func = bpf_send_signal_thread,
1128 .gpl_only = false,
1129 .ret_type = RET_INTEGER,
1130 .arg1_type = ARG_ANYTHING,
1131 };
1132
BPF_CALL_3(bpf_d_path,struct path *,path,char *,buf,u32,sz)1133 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
1134 {
1135 struct path copy;
1136 long len;
1137 char *p;
1138
1139 if (!sz)
1140 return 0;
1141
1142 /*
1143 * The path pointer is verified as trusted and safe to use,
1144 * but let's double check it's valid anyway to workaround
1145 * potentially broken verifier.
1146 */
1147 len = copy_from_kernel_nofault(©, path, sizeof(*path));
1148 if (len < 0)
1149 return len;
1150
1151 p = d_path(©, buf, sz);
1152 if (IS_ERR(p)) {
1153 len = PTR_ERR(p);
1154 } else {
1155 len = buf + sz - p;
1156 memmove(buf, p, len);
1157 }
1158
1159 return len;
1160 }
1161
1162 BTF_SET_START(btf_allowlist_d_path)
1163 #ifdef CONFIG_SECURITY
BTF_ID(func,security_file_permission)1164 BTF_ID(func, security_file_permission)
1165 BTF_ID(func, security_inode_getattr)
1166 BTF_ID(func, security_file_open)
1167 #endif
1168 #ifdef CONFIG_SECURITY_PATH
1169 BTF_ID(func, security_path_truncate)
1170 #endif
1171 BTF_ID(func, vfs_truncate)
1172 BTF_ID(func, vfs_fallocate)
1173 BTF_ID(func, dentry_open)
1174 BTF_ID(func, vfs_getattr)
1175 BTF_ID(func, filp_close)
1176 BTF_SET_END(btf_allowlist_d_path)
1177
1178 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
1179 {
1180 return btf_id_set_contains(&btf_allowlist_d_path, prog->aux->attach_btf_id);
1181 }
1182
1183 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
1184
1185 static const struct bpf_func_proto bpf_d_path_proto = {
1186 .func = bpf_d_path,
1187 .gpl_only = false,
1188 .ret_type = RET_INTEGER,
1189 .arg1_type = ARG_PTR_TO_BTF_ID,
1190 .arg1_btf_id = &bpf_d_path_btf_ids[0],
1191 .arg2_type = ARG_PTR_TO_MEM,
1192 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1193 .allowed = bpf_d_path_allowed,
1194 };
1195
1196 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
1197 BTF_F_PTR_RAW | BTF_F_ZERO)
1198
bpf_btf_printf_prepare(struct btf_ptr * ptr,u32 btf_ptr_size,u64 flags,const struct btf ** btf,s32 * btf_id)1199 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
1200 u64 flags, const struct btf **btf,
1201 s32 *btf_id)
1202 {
1203 const struct btf_type *t;
1204
1205 if (unlikely(flags & ~(BTF_F_ALL)))
1206 return -EINVAL;
1207
1208 if (btf_ptr_size != sizeof(struct btf_ptr))
1209 return -EINVAL;
1210
1211 *btf = bpf_get_btf_vmlinux();
1212
1213 if (IS_ERR_OR_NULL(*btf))
1214 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
1215
1216 if (ptr->type_id > 0)
1217 *btf_id = ptr->type_id;
1218 else
1219 return -EINVAL;
1220
1221 if (*btf_id > 0)
1222 t = btf_type_by_id(*btf, *btf_id);
1223 if (*btf_id <= 0 || !t)
1224 return -ENOENT;
1225
1226 return 0;
1227 }
1228
BPF_CALL_5(bpf_snprintf_btf,char *,str,u32,str_size,struct btf_ptr *,ptr,u32,btf_ptr_size,u64,flags)1229 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1230 u32, btf_ptr_size, u64, flags)
1231 {
1232 const struct btf *btf;
1233 s32 btf_id;
1234 int ret;
1235
1236 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1237 if (ret)
1238 return ret;
1239
1240 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1241 flags);
1242 }
1243
1244 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1245 .func = bpf_snprintf_btf,
1246 .gpl_only = false,
1247 .ret_type = RET_INTEGER,
1248 .arg1_type = ARG_PTR_TO_MEM,
1249 .arg2_type = ARG_CONST_SIZE,
1250 .arg3_type = ARG_PTR_TO_MEM,
1251 .arg4_type = ARG_CONST_SIZE,
1252 .arg5_type = ARG_ANYTHING,
1253 };
1254
1255 const struct bpf_func_proto *
bpf_tracing_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1256 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1257 {
1258 switch (func_id) {
1259 case BPF_FUNC_map_lookup_elem:
1260 return &bpf_map_lookup_elem_proto;
1261 case BPF_FUNC_map_update_elem:
1262 return &bpf_map_update_elem_proto;
1263 case BPF_FUNC_map_delete_elem:
1264 return &bpf_map_delete_elem_proto;
1265 case BPF_FUNC_map_push_elem:
1266 return &bpf_map_push_elem_proto;
1267 case BPF_FUNC_map_pop_elem:
1268 return &bpf_map_pop_elem_proto;
1269 case BPF_FUNC_map_peek_elem:
1270 return &bpf_map_peek_elem_proto;
1271 case BPF_FUNC_ktime_get_ns:
1272 return &bpf_ktime_get_ns_proto;
1273 case BPF_FUNC_ktime_get_boot_ns:
1274 return &bpf_ktime_get_boot_ns_proto;
1275 case BPF_FUNC_tail_call:
1276 return &bpf_tail_call_proto;
1277 case BPF_FUNC_get_current_pid_tgid:
1278 return &bpf_get_current_pid_tgid_proto;
1279 case BPF_FUNC_get_current_task:
1280 return &bpf_get_current_task_proto;
1281 case BPF_FUNC_get_current_uid_gid:
1282 return &bpf_get_current_uid_gid_proto;
1283 case BPF_FUNC_get_current_comm:
1284 return &bpf_get_current_comm_proto;
1285 case BPF_FUNC_trace_printk:
1286 return bpf_get_trace_printk_proto();
1287 case BPF_FUNC_get_smp_processor_id:
1288 return &bpf_get_smp_processor_id_proto;
1289 case BPF_FUNC_get_numa_node_id:
1290 return &bpf_get_numa_node_id_proto;
1291 case BPF_FUNC_perf_event_read:
1292 return &bpf_perf_event_read_proto;
1293 case BPF_FUNC_current_task_under_cgroup:
1294 return &bpf_current_task_under_cgroup_proto;
1295 case BPF_FUNC_get_prandom_u32:
1296 return &bpf_get_prandom_u32_proto;
1297 case BPF_FUNC_probe_write_user:
1298 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1299 NULL : bpf_get_probe_write_proto();
1300 case BPF_FUNC_probe_read_user:
1301 return &bpf_probe_read_user_proto;
1302 case BPF_FUNC_probe_read_kernel:
1303 return security_locked_down(LOCKDOWN_BPF_READ) < 0 ?
1304 NULL : &bpf_probe_read_kernel_proto;
1305 case BPF_FUNC_probe_read_user_str:
1306 return &bpf_probe_read_user_str_proto;
1307 case BPF_FUNC_probe_read_kernel_str:
1308 return security_locked_down(LOCKDOWN_BPF_READ) < 0 ?
1309 NULL : &bpf_probe_read_kernel_str_proto;
1310 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1311 case BPF_FUNC_probe_read:
1312 return security_locked_down(LOCKDOWN_BPF_READ) < 0 ?
1313 NULL : &bpf_probe_read_compat_proto;
1314 case BPF_FUNC_probe_read_str:
1315 return security_locked_down(LOCKDOWN_BPF_READ) < 0 ?
1316 NULL : &bpf_probe_read_compat_str_proto;
1317 #endif
1318 #ifdef CONFIG_CGROUPS
1319 case BPF_FUNC_get_current_cgroup_id:
1320 return &bpf_get_current_cgroup_id_proto;
1321 #endif
1322 case BPF_FUNC_send_signal:
1323 return &bpf_send_signal_proto;
1324 case BPF_FUNC_send_signal_thread:
1325 return &bpf_send_signal_thread_proto;
1326 case BPF_FUNC_perf_event_read_value:
1327 return &bpf_perf_event_read_value_proto;
1328 case BPF_FUNC_get_ns_current_pid_tgid:
1329 return &bpf_get_ns_current_pid_tgid_proto;
1330 case BPF_FUNC_ringbuf_output:
1331 return &bpf_ringbuf_output_proto;
1332 case BPF_FUNC_ringbuf_reserve:
1333 return &bpf_ringbuf_reserve_proto;
1334 case BPF_FUNC_ringbuf_submit:
1335 return &bpf_ringbuf_submit_proto;
1336 case BPF_FUNC_ringbuf_discard:
1337 return &bpf_ringbuf_discard_proto;
1338 case BPF_FUNC_ringbuf_query:
1339 return &bpf_ringbuf_query_proto;
1340 case BPF_FUNC_jiffies64:
1341 return &bpf_jiffies64_proto;
1342 case BPF_FUNC_get_task_stack:
1343 return &bpf_get_task_stack_proto;
1344 case BPF_FUNC_copy_from_user:
1345 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1346 case BPF_FUNC_snprintf_btf:
1347 return &bpf_snprintf_btf_proto;
1348 case BPF_FUNC_per_cpu_ptr:
1349 return &bpf_per_cpu_ptr_proto;
1350 case BPF_FUNC_this_cpu_ptr:
1351 return &bpf_this_cpu_ptr_proto;
1352 default:
1353 return NULL;
1354 }
1355 }
1356
1357 static const struct bpf_func_proto *
kprobe_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1358 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1359 {
1360 switch (func_id) {
1361 case BPF_FUNC_perf_event_output:
1362 return &bpf_perf_event_output_proto;
1363 case BPF_FUNC_get_stackid:
1364 return &bpf_get_stackid_proto;
1365 case BPF_FUNC_get_stack:
1366 return &bpf_get_stack_proto;
1367 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1368 case BPF_FUNC_override_return:
1369 return &bpf_override_return_proto;
1370 #endif
1371 default:
1372 return bpf_tracing_func_proto(func_id, prog);
1373 }
1374 }
1375
1376 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
kprobe_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1377 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1378 const struct bpf_prog *prog,
1379 struct bpf_insn_access_aux *info)
1380 {
1381 if (off < 0 || off >= sizeof(struct pt_regs))
1382 return false;
1383 if (type != BPF_READ)
1384 return false;
1385 if (off % size != 0)
1386 return false;
1387 /*
1388 * Assertion for 32 bit to make sure last 8 byte access
1389 * (BPF_DW) to the last 4 byte member is disallowed.
1390 */
1391 if (off + size > sizeof(struct pt_regs))
1392 return false;
1393
1394 return true;
1395 }
1396
1397 const struct bpf_verifier_ops kprobe_verifier_ops = {
1398 .get_func_proto = kprobe_prog_func_proto,
1399 .is_valid_access = kprobe_prog_is_valid_access,
1400 };
1401
1402 const struct bpf_prog_ops kprobe_prog_ops = {
1403 };
1404
BPF_CALL_5(bpf_perf_event_output_tp,void *,tp_buff,struct bpf_map *,map,u64,flags,void *,data,u64,size)1405 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1406 u64, flags, void *, data, u64, size)
1407 {
1408 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1409
1410 /*
1411 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1412 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1413 * from there and call the same bpf_perf_event_output() helper inline.
1414 */
1415 return ____bpf_perf_event_output(regs, map, flags, data, size);
1416 }
1417
1418 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1419 .func = bpf_perf_event_output_tp,
1420 .gpl_only = true,
1421 .ret_type = RET_INTEGER,
1422 .arg1_type = ARG_PTR_TO_CTX,
1423 .arg2_type = ARG_CONST_MAP_PTR,
1424 .arg3_type = ARG_ANYTHING,
1425 .arg4_type = ARG_PTR_TO_MEM,
1426 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1427 };
1428
BPF_CALL_3(bpf_get_stackid_tp,void *,tp_buff,struct bpf_map *,map,u64,flags)1429 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1430 u64, flags)
1431 {
1432 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1433
1434 /*
1435 * Same comment as in bpf_perf_event_output_tp(), only that this time
1436 * the other helper's function body cannot be inlined due to being
1437 * external, thus we need to call raw helper function.
1438 */
1439 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1440 flags, 0, 0);
1441 }
1442
1443 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1444 .func = bpf_get_stackid_tp,
1445 .gpl_only = true,
1446 .ret_type = RET_INTEGER,
1447 .arg1_type = ARG_PTR_TO_CTX,
1448 .arg2_type = ARG_CONST_MAP_PTR,
1449 .arg3_type = ARG_ANYTHING,
1450 };
1451
BPF_CALL_4(bpf_get_stack_tp,void *,tp_buff,void *,buf,u32,size,u64,flags)1452 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1453 u64, flags)
1454 {
1455 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1456
1457 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1458 (unsigned long) size, flags, 0);
1459 }
1460
1461 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1462 .func = bpf_get_stack_tp,
1463 .gpl_only = true,
1464 .ret_type = RET_INTEGER,
1465 .arg1_type = ARG_PTR_TO_CTX,
1466 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1467 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1468 .arg4_type = ARG_ANYTHING,
1469 };
1470
1471 static const struct bpf_func_proto *
tp_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1472 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1473 {
1474 switch (func_id) {
1475 case BPF_FUNC_perf_event_output:
1476 return &bpf_perf_event_output_proto_tp;
1477 case BPF_FUNC_get_stackid:
1478 return &bpf_get_stackid_proto_tp;
1479 case BPF_FUNC_get_stack:
1480 return &bpf_get_stack_proto_tp;
1481 default:
1482 return bpf_tracing_func_proto(func_id, prog);
1483 }
1484 }
1485
tp_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1486 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1487 const struct bpf_prog *prog,
1488 struct bpf_insn_access_aux *info)
1489 {
1490 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1491 return false;
1492 if (type != BPF_READ)
1493 return false;
1494 if (off % size != 0)
1495 return false;
1496
1497 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1498 return true;
1499 }
1500
1501 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1502 .get_func_proto = tp_prog_func_proto,
1503 .is_valid_access = tp_prog_is_valid_access,
1504 };
1505
1506 const struct bpf_prog_ops tracepoint_prog_ops = {
1507 };
1508
BPF_CALL_3(bpf_perf_prog_read_value,struct bpf_perf_event_data_kern *,ctx,struct bpf_perf_event_value *,buf,u32,size)1509 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1510 struct bpf_perf_event_value *, buf, u32, size)
1511 {
1512 int err = -EINVAL;
1513
1514 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1515 goto clear;
1516 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1517 &buf->running);
1518 if (unlikely(err))
1519 goto clear;
1520 return 0;
1521 clear:
1522 memset(buf, 0, size);
1523 return err;
1524 }
1525
1526 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1527 .func = bpf_perf_prog_read_value,
1528 .gpl_only = true,
1529 .ret_type = RET_INTEGER,
1530 .arg1_type = ARG_PTR_TO_CTX,
1531 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1532 .arg3_type = ARG_CONST_SIZE,
1533 };
1534
BPF_CALL_4(bpf_read_branch_records,struct bpf_perf_event_data_kern *,ctx,void *,buf,u32,size,u64,flags)1535 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1536 void *, buf, u32, size, u64, flags)
1537 {
1538 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1539 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1540 u32 to_copy;
1541
1542 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1543 return -EINVAL;
1544
1545 if (unlikely(!br_stack))
1546 return -ENOENT;
1547
1548 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1549 return br_stack->nr * br_entry_size;
1550
1551 if (!buf || (size % br_entry_size != 0))
1552 return -EINVAL;
1553
1554 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1555 memcpy(buf, br_stack->entries, to_copy);
1556
1557 return to_copy;
1558 }
1559
1560 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1561 .func = bpf_read_branch_records,
1562 .gpl_only = true,
1563 .ret_type = RET_INTEGER,
1564 .arg1_type = ARG_PTR_TO_CTX,
1565 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1566 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1567 .arg4_type = ARG_ANYTHING,
1568 };
1569
1570 static const struct bpf_func_proto *
pe_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1571 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1572 {
1573 switch (func_id) {
1574 case BPF_FUNC_perf_event_output:
1575 return &bpf_perf_event_output_proto_tp;
1576 case BPF_FUNC_get_stackid:
1577 return &bpf_get_stackid_proto_pe;
1578 case BPF_FUNC_get_stack:
1579 return &bpf_get_stack_proto_pe;
1580 case BPF_FUNC_perf_prog_read_value:
1581 return &bpf_perf_prog_read_value_proto;
1582 case BPF_FUNC_read_branch_records:
1583 return &bpf_read_branch_records_proto;
1584 default:
1585 return bpf_tracing_func_proto(func_id, prog);
1586 }
1587 }
1588
1589 /*
1590 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1591 * to avoid potential recursive reuse issue when/if tracepoints are added
1592 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1593 *
1594 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1595 * in normal, irq, and nmi context.
1596 */
1597 struct bpf_raw_tp_regs {
1598 struct pt_regs regs[3];
1599 };
1600 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1601 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
get_bpf_raw_tp_regs(void)1602 static struct pt_regs *get_bpf_raw_tp_regs(void)
1603 {
1604 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1605 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1606
1607 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1608 this_cpu_dec(bpf_raw_tp_nest_level);
1609 return ERR_PTR(-EBUSY);
1610 }
1611
1612 return &tp_regs->regs[nest_level - 1];
1613 }
1614
put_bpf_raw_tp_regs(void)1615 static void put_bpf_raw_tp_regs(void)
1616 {
1617 this_cpu_dec(bpf_raw_tp_nest_level);
1618 }
1619
BPF_CALL_5(bpf_perf_event_output_raw_tp,struct bpf_raw_tracepoint_args *,args,struct bpf_map *,map,u64,flags,void *,data,u64,size)1620 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1621 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1622 {
1623 struct pt_regs *regs = get_bpf_raw_tp_regs();
1624 int ret;
1625
1626 if (IS_ERR(regs))
1627 return PTR_ERR(regs);
1628
1629 perf_fetch_caller_regs(regs);
1630 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1631
1632 put_bpf_raw_tp_regs();
1633 return ret;
1634 }
1635
1636 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1637 .func = bpf_perf_event_output_raw_tp,
1638 .gpl_only = true,
1639 .ret_type = RET_INTEGER,
1640 .arg1_type = ARG_PTR_TO_CTX,
1641 .arg2_type = ARG_CONST_MAP_PTR,
1642 .arg3_type = ARG_ANYTHING,
1643 .arg4_type = ARG_PTR_TO_MEM,
1644 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1645 };
1646
1647 extern const struct bpf_func_proto bpf_skb_output_proto;
1648 extern const struct bpf_func_proto bpf_xdp_output_proto;
1649
BPF_CALL_3(bpf_get_stackid_raw_tp,struct bpf_raw_tracepoint_args *,args,struct bpf_map *,map,u64,flags)1650 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1651 struct bpf_map *, map, u64, flags)
1652 {
1653 struct pt_regs *regs = get_bpf_raw_tp_regs();
1654 int ret;
1655
1656 if (IS_ERR(regs))
1657 return PTR_ERR(regs);
1658
1659 perf_fetch_caller_regs(regs);
1660 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1661 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1662 flags, 0, 0);
1663 put_bpf_raw_tp_regs();
1664 return ret;
1665 }
1666
1667 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1668 .func = bpf_get_stackid_raw_tp,
1669 .gpl_only = true,
1670 .ret_type = RET_INTEGER,
1671 .arg1_type = ARG_PTR_TO_CTX,
1672 .arg2_type = ARG_CONST_MAP_PTR,
1673 .arg3_type = ARG_ANYTHING,
1674 };
1675
BPF_CALL_4(bpf_get_stack_raw_tp,struct bpf_raw_tracepoint_args *,args,void *,buf,u32,size,u64,flags)1676 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1677 void *, buf, u32, size, u64, flags)
1678 {
1679 struct pt_regs *regs = get_bpf_raw_tp_regs();
1680 int ret;
1681
1682 if (IS_ERR(regs))
1683 return PTR_ERR(regs);
1684
1685 perf_fetch_caller_regs(regs);
1686 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1687 (unsigned long) size, flags, 0);
1688 put_bpf_raw_tp_regs();
1689 return ret;
1690 }
1691
1692 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1693 .func = bpf_get_stack_raw_tp,
1694 .gpl_only = true,
1695 .ret_type = RET_INTEGER,
1696 .arg1_type = ARG_PTR_TO_CTX,
1697 .arg2_type = ARG_PTR_TO_MEM,
1698 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1699 .arg4_type = ARG_ANYTHING,
1700 };
1701
1702 static const struct bpf_func_proto *
raw_tp_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1703 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1704 {
1705 switch (func_id) {
1706 case BPF_FUNC_perf_event_output:
1707 return &bpf_perf_event_output_proto_raw_tp;
1708 case BPF_FUNC_get_stackid:
1709 return &bpf_get_stackid_proto_raw_tp;
1710 case BPF_FUNC_get_stack:
1711 return &bpf_get_stack_proto_raw_tp;
1712 default:
1713 return bpf_tracing_func_proto(func_id, prog);
1714 }
1715 }
1716
1717 const struct bpf_func_proto *
tracing_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1718 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1719 {
1720 switch (func_id) {
1721 #ifdef CONFIG_NET
1722 case BPF_FUNC_skb_output:
1723 return &bpf_skb_output_proto;
1724 case BPF_FUNC_xdp_output:
1725 return &bpf_xdp_output_proto;
1726 case BPF_FUNC_skc_to_tcp6_sock:
1727 return &bpf_skc_to_tcp6_sock_proto;
1728 case BPF_FUNC_skc_to_tcp_sock:
1729 return &bpf_skc_to_tcp_sock_proto;
1730 case BPF_FUNC_skc_to_tcp_timewait_sock:
1731 return &bpf_skc_to_tcp_timewait_sock_proto;
1732 case BPF_FUNC_skc_to_tcp_request_sock:
1733 return &bpf_skc_to_tcp_request_sock_proto;
1734 case BPF_FUNC_skc_to_udp6_sock:
1735 return &bpf_skc_to_udp6_sock_proto;
1736 #endif
1737 case BPF_FUNC_seq_printf:
1738 return prog->expected_attach_type == BPF_TRACE_ITER ?
1739 &bpf_seq_printf_proto :
1740 NULL;
1741 case BPF_FUNC_seq_write:
1742 return prog->expected_attach_type == BPF_TRACE_ITER ?
1743 &bpf_seq_write_proto :
1744 NULL;
1745 case BPF_FUNC_seq_printf_btf:
1746 return prog->expected_attach_type == BPF_TRACE_ITER ?
1747 &bpf_seq_printf_btf_proto :
1748 NULL;
1749 case BPF_FUNC_d_path:
1750 return &bpf_d_path_proto;
1751 default:
1752 return raw_tp_prog_func_proto(func_id, prog);
1753 }
1754 }
1755
raw_tp_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1756 static bool raw_tp_prog_is_valid_access(int off, int size,
1757 enum bpf_access_type type,
1758 const struct bpf_prog *prog,
1759 struct bpf_insn_access_aux *info)
1760 {
1761 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1762 return false;
1763 if (type != BPF_READ)
1764 return false;
1765 if (off % size != 0)
1766 return false;
1767 return true;
1768 }
1769
tracing_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1770 static bool tracing_prog_is_valid_access(int off, int size,
1771 enum bpf_access_type type,
1772 const struct bpf_prog *prog,
1773 struct bpf_insn_access_aux *info)
1774 {
1775 if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1776 return false;
1777 if (type != BPF_READ)
1778 return false;
1779 if (off % size != 0)
1780 return false;
1781 return btf_ctx_access(off, size, type, prog, info);
1782 }
1783
bpf_prog_test_run_tracing(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)1784 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1785 const union bpf_attr *kattr,
1786 union bpf_attr __user *uattr)
1787 {
1788 return -ENOTSUPP;
1789 }
1790
1791 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1792 .get_func_proto = raw_tp_prog_func_proto,
1793 .is_valid_access = raw_tp_prog_is_valid_access,
1794 };
1795
1796 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1797 #ifdef CONFIG_NET
1798 .test_run = bpf_prog_test_run_raw_tp,
1799 #endif
1800 };
1801
1802 const struct bpf_verifier_ops tracing_verifier_ops = {
1803 .get_func_proto = tracing_prog_func_proto,
1804 .is_valid_access = tracing_prog_is_valid_access,
1805 };
1806
1807 const struct bpf_prog_ops tracing_prog_ops = {
1808 .test_run = bpf_prog_test_run_tracing,
1809 };
1810
raw_tp_writable_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1811 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1812 enum bpf_access_type type,
1813 const struct bpf_prog *prog,
1814 struct bpf_insn_access_aux *info)
1815 {
1816 if (off == 0) {
1817 if (size != sizeof(u64) || type != BPF_READ)
1818 return false;
1819 info->reg_type = PTR_TO_TP_BUFFER;
1820 }
1821 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1822 }
1823
1824 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1825 .get_func_proto = raw_tp_prog_func_proto,
1826 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1827 };
1828
1829 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1830 };
1831
pe_prog_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)1832 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1833 const struct bpf_prog *prog,
1834 struct bpf_insn_access_aux *info)
1835 {
1836 const int size_u64 = sizeof(u64);
1837
1838 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1839 return false;
1840 if (type != BPF_READ)
1841 return false;
1842 if (off % size != 0) {
1843 if (sizeof(unsigned long) != 4)
1844 return false;
1845 if (size != 8)
1846 return false;
1847 if (off % size != 4)
1848 return false;
1849 }
1850
1851 switch (off) {
1852 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1853 bpf_ctx_record_field_size(info, size_u64);
1854 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1855 return false;
1856 break;
1857 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1858 bpf_ctx_record_field_size(info, size_u64);
1859 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1860 return false;
1861 break;
1862 default:
1863 if (size != sizeof(long))
1864 return false;
1865 }
1866
1867 return true;
1868 }
1869
pe_prog_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)1870 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1871 const struct bpf_insn *si,
1872 struct bpf_insn *insn_buf,
1873 struct bpf_prog *prog, u32 *target_size)
1874 {
1875 struct bpf_insn *insn = insn_buf;
1876
1877 switch (si->off) {
1878 case offsetof(struct bpf_perf_event_data, sample_period):
1879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1880 data), si->dst_reg, si->src_reg,
1881 offsetof(struct bpf_perf_event_data_kern, data));
1882 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1883 bpf_target_off(struct perf_sample_data, period, 8,
1884 target_size));
1885 break;
1886 case offsetof(struct bpf_perf_event_data, addr):
1887 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1888 data), si->dst_reg, si->src_reg,
1889 offsetof(struct bpf_perf_event_data_kern, data));
1890 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1891 bpf_target_off(struct perf_sample_data, addr, 8,
1892 target_size));
1893 break;
1894 default:
1895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1896 regs), si->dst_reg, si->src_reg,
1897 offsetof(struct bpf_perf_event_data_kern, regs));
1898 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1899 si->off);
1900 break;
1901 }
1902
1903 return insn - insn_buf;
1904 }
1905
1906 const struct bpf_verifier_ops perf_event_verifier_ops = {
1907 .get_func_proto = pe_prog_func_proto,
1908 .is_valid_access = pe_prog_is_valid_access,
1909 .convert_ctx_access = pe_prog_convert_ctx_access,
1910 };
1911
1912 const struct bpf_prog_ops perf_event_prog_ops = {
1913 };
1914
1915 static DEFINE_MUTEX(bpf_event_mutex);
1916
1917 #define BPF_TRACE_MAX_PROGS 64
1918
perf_event_attach_bpf_prog(struct perf_event * event,struct bpf_prog * prog)1919 int perf_event_attach_bpf_prog(struct perf_event *event,
1920 struct bpf_prog *prog)
1921 {
1922 struct bpf_prog_array *old_array;
1923 struct bpf_prog_array *new_array;
1924 int ret = -EEXIST;
1925
1926 /*
1927 * Kprobe override only works if they are on the function entry,
1928 * and only if they are on the opt-in list.
1929 */
1930 if (prog->kprobe_override &&
1931 (!trace_kprobe_on_func_entry(event->tp_event) ||
1932 !trace_kprobe_error_injectable(event->tp_event)))
1933 return -EINVAL;
1934
1935 mutex_lock(&bpf_event_mutex);
1936
1937 if (event->prog)
1938 goto unlock;
1939
1940 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1941 if (old_array &&
1942 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1943 ret = -E2BIG;
1944 goto unlock;
1945 }
1946
1947 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1948 if (ret < 0)
1949 goto unlock;
1950
1951 /* set the new array to event->tp_event and set event->prog */
1952 event->prog = prog;
1953 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1954 bpf_prog_array_free(old_array);
1955
1956 unlock:
1957 mutex_unlock(&bpf_event_mutex);
1958 return ret;
1959 }
1960
perf_event_detach_bpf_prog(struct perf_event * event)1961 void perf_event_detach_bpf_prog(struct perf_event *event)
1962 {
1963 struct bpf_prog_array *old_array;
1964 struct bpf_prog_array *new_array;
1965 int ret;
1966
1967 mutex_lock(&bpf_event_mutex);
1968
1969 if (!event->prog)
1970 goto unlock;
1971
1972 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1973 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1974 if (ret == -ENOENT)
1975 goto unlock;
1976 if (ret < 0) {
1977 bpf_prog_array_delete_safe(old_array, event->prog);
1978 } else {
1979 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1980 bpf_prog_array_free(old_array);
1981 }
1982
1983 bpf_prog_put(event->prog);
1984 event->prog = NULL;
1985
1986 unlock:
1987 mutex_unlock(&bpf_event_mutex);
1988 }
1989
perf_event_query_prog_array(struct perf_event * event,void __user * info)1990 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1991 {
1992 struct perf_event_query_bpf __user *uquery = info;
1993 struct perf_event_query_bpf query = {};
1994 struct bpf_prog_array *progs;
1995 u32 *ids, prog_cnt, ids_len;
1996 int ret;
1997
1998 if (!perfmon_capable())
1999 return -EPERM;
2000 if (event->attr.type != PERF_TYPE_TRACEPOINT)
2001 return -EINVAL;
2002 if (copy_from_user(&query, uquery, sizeof(query)))
2003 return -EFAULT;
2004
2005 ids_len = query.ids_len;
2006 if (ids_len > BPF_TRACE_MAX_PROGS)
2007 return -E2BIG;
2008 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2009 if (!ids)
2010 return -ENOMEM;
2011 /*
2012 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2013 * is required when user only wants to check for uquery->prog_cnt.
2014 * There is no need to check for it since the case is handled
2015 * gracefully in bpf_prog_array_copy_info.
2016 */
2017
2018 mutex_lock(&bpf_event_mutex);
2019 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2020 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2021 mutex_unlock(&bpf_event_mutex);
2022
2023 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2024 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2025 ret = -EFAULT;
2026
2027 kfree(ids);
2028 return ret;
2029 }
2030
2031 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2032 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2033
bpf_get_raw_tracepoint(const char * name)2034 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2035 {
2036 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2037
2038 for (; btp < __stop__bpf_raw_tp; btp++) {
2039 if (!strcmp(btp->tp->name, name))
2040 return btp;
2041 }
2042
2043 return bpf_get_raw_tracepoint_module(name);
2044 }
2045
bpf_put_raw_tracepoint(struct bpf_raw_event_map * btp)2046 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2047 {
2048 struct module *mod;
2049
2050 preempt_disable();
2051 mod = __module_address((unsigned long)btp);
2052 module_put(mod);
2053 preempt_enable();
2054 }
2055
2056 static __always_inline
__bpf_trace_run(struct bpf_prog * prog,u64 * args)2057 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2058 {
2059 cant_sleep();
2060 rcu_read_lock();
2061 (void) BPF_PROG_RUN(prog, args);
2062 rcu_read_unlock();
2063 }
2064
2065 #define UNPACK(...) __VA_ARGS__
2066 #define REPEAT_1(FN, DL, X, ...) FN(X)
2067 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2068 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2069 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2070 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2071 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2072 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2073 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2074 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2075 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2076 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2077 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2078 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2079
2080 #define SARG(X) u64 arg##X
2081 #define COPY(X) args[X] = arg##X
2082
2083 #define __DL_COM (,)
2084 #define __DL_SEM (;)
2085
2086 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2087
2088 #define BPF_TRACE_DEFN_x(x) \
2089 void bpf_trace_run##x(struct bpf_prog *prog, \
2090 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2091 { \
2092 u64 args[x]; \
2093 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2094 __bpf_trace_run(prog, args); \
2095 } \
2096 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2097 BPF_TRACE_DEFN_x(1);
2098 BPF_TRACE_DEFN_x(2);
2099 BPF_TRACE_DEFN_x(3);
2100 BPF_TRACE_DEFN_x(4);
2101 BPF_TRACE_DEFN_x(5);
2102 BPF_TRACE_DEFN_x(6);
2103 BPF_TRACE_DEFN_x(7);
2104 BPF_TRACE_DEFN_x(8);
2105 BPF_TRACE_DEFN_x(9);
2106 BPF_TRACE_DEFN_x(10);
2107 BPF_TRACE_DEFN_x(11);
2108 BPF_TRACE_DEFN_x(12);
2109
__bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_prog * prog)2110 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2111 {
2112 struct tracepoint *tp = btp->tp;
2113
2114 /*
2115 * check that program doesn't access arguments beyond what's
2116 * available in this tracepoint
2117 */
2118 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2119 return -EINVAL;
2120
2121 if (prog->aux->max_tp_access > btp->writable_size)
2122 return -EINVAL;
2123
2124 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2125 prog);
2126 }
2127
bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_prog * prog)2128 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2129 {
2130 return __bpf_probe_register(btp, prog);
2131 }
2132
bpf_probe_unregister(struct bpf_raw_event_map * btp,struct bpf_prog * prog)2133 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2134 {
2135 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2136 }
2137
bpf_get_perf_event_info(const struct perf_event * event,u32 * prog_id,u32 * fd_type,const char ** buf,u64 * probe_offset,u64 * probe_addr)2138 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2139 u32 *fd_type, const char **buf,
2140 u64 *probe_offset, u64 *probe_addr)
2141 {
2142 bool is_tracepoint, is_syscall_tp;
2143 struct bpf_prog *prog;
2144 int flags, err = 0;
2145
2146 prog = event->prog;
2147 if (!prog)
2148 return -ENOENT;
2149
2150 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2151 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2152 return -EOPNOTSUPP;
2153
2154 *prog_id = prog->aux->id;
2155 flags = event->tp_event->flags;
2156 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2157 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2158
2159 if (is_tracepoint || is_syscall_tp) {
2160 *buf = is_tracepoint ? event->tp_event->tp->name
2161 : event->tp_event->name;
2162 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2163 *probe_offset = 0x0;
2164 *probe_addr = 0x0;
2165 } else {
2166 /* kprobe/uprobe */
2167 err = -EOPNOTSUPP;
2168 #ifdef CONFIG_KPROBE_EVENTS
2169 if (flags & TRACE_EVENT_FL_KPROBE)
2170 err = bpf_get_kprobe_info(event, fd_type, buf,
2171 probe_offset, probe_addr,
2172 event->attr.type == PERF_TYPE_TRACEPOINT);
2173 #endif
2174 #ifdef CONFIG_UPROBE_EVENTS
2175 if (flags & TRACE_EVENT_FL_UPROBE)
2176 err = bpf_get_uprobe_info(event, fd_type, buf,
2177 probe_offset, probe_addr,
2178 event->attr.type == PERF_TYPE_TRACEPOINT);
2179 #endif
2180 }
2181
2182 return err;
2183 }
2184
send_signal_irq_work_init(void)2185 static int __init send_signal_irq_work_init(void)
2186 {
2187 int cpu;
2188 struct send_signal_irq_work *work;
2189
2190 for_each_possible_cpu(cpu) {
2191 work = per_cpu_ptr(&send_signal_work, cpu);
2192 init_irq_work(&work->irq_work, do_bpf_send_signal);
2193 }
2194 return 0;
2195 }
2196
2197 subsys_initcall(send_signal_irq_work_init);
2198
2199 #ifdef CONFIG_MODULES
bpf_event_notify(struct notifier_block * nb,unsigned long op,void * module)2200 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2201 void *module)
2202 {
2203 struct bpf_trace_module *btm, *tmp;
2204 struct module *mod = module;
2205 int ret = 0;
2206
2207 if (mod->num_bpf_raw_events == 0 ||
2208 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2209 goto out;
2210
2211 mutex_lock(&bpf_module_mutex);
2212
2213 switch (op) {
2214 case MODULE_STATE_COMING:
2215 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2216 if (btm) {
2217 btm->module = module;
2218 list_add(&btm->list, &bpf_trace_modules);
2219 } else {
2220 ret = -ENOMEM;
2221 }
2222 break;
2223 case MODULE_STATE_GOING:
2224 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2225 if (btm->module == module) {
2226 list_del(&btm->list);
2227 kfree(btm);
2228 break;
2229 }
2230 }
2231 break;
2232 }
2233
2234 mutex_unlock(&bpf_module_mutex);
2235
2236 out:
2237 return notifier_from_errno(ret);
2238 }
2239
2240 static struct notifier_block bpf_module_nb = {
2241 .notifier_call = bpf_event_notify,
2242 };
2243
bpf_event_init(void)2244 static int __init bpf_event_init(void)
2245 {
2246 register_module_notifier(&bpf_module_nb);
2247 return 0;
2248 }
2249
2250 fs_initcall(bpf_event_init);
2251 #endif /* CONFIG_MODULES */
2252