1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
4 */
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/slab.h>
8 #include <linux/bpf.h>
9 #include <linux/bpf_verifier.h>
10 #include <linux/bpf_perf_event.h>
11 #include <linux/btf.h>
12 #include <linux/filter.h>
13 #include <linux/uaccess.h>
14 #include <linux/ctype.h>
15 #include <linux/kprobes.h>
16 #include <linux/spinlock.h>
17 #include <linux/syscalls.h>
18 #include <linux/error-injection.h>
19 #include <linux/btf_ids.h>
20 #include <linux/bpf_lsm.h>
21 #include <linux/fprobe.h>
22 #include <linux/bsearch.h>
23 #include <linux/sort.h>
24 #include <linux/key.h>
25 #include <linux/verification.h>
26 #include <linux/namei.h>
27
28 #include <net/bpf_sk_storage.h>
29
30 #include <uapi/linux/bpf.h>
31 #include <uapi/linux/btf.h>
32
33 #include <asm/tlb.h>
34
35 #include "trace_probe.h"
36 #include "trace.h"
37
38 #define CREATE_TRACE_POINTS
39 #include "bpf_trace.h"
40
41 #define bpf_event_rcu_dereference(p) \
42 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
43
44 #define MAX_UPROBE_MULTI_CNT (1U << 20)
45 #define MAX_KPROBE_MULTI_CNT (1U << 20)
46
47 #ifdef CONFIG_MODULES
48 struct bpf_trace_module {
49 struct module *module;
50 struct list_head list;
51 };
52
53 static LIST_HEAD(bpf_trace_modules);
54 static DEFINE_MUTEX(bpf_module_mutex);
55
bpf_get_raw_tracepoint_module(const char * name)56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
57 {
58 struct bpf_raw_event_map *btp, *ret = NULL;
59 struct bpf_trace_module *btm;
60 unsigned int i;
61
62 mutex_lock(&bpf_module_mutex);
63 list_for_each_entry(btm, &bpf_trace_modules, list) {
64 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
65 btp = &btm->module->bpf_raw_events[i];
66 if (!strcmp(btp->tp->name, name)) {
67 if (try_module_get(btm->module))
68 ret = btp;
69 goto out;
70 }
71 }
72 }
73 out:
74 mutex_unlock(&bpf_module_mutex);
75 return ret;
76 }
77 #else
bpf_get_raw_tracepoint_module(const char * name)78 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
79 {
80 return NULL;
81 }
82 #endif /* CONFIG_MODULES */
83
84 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
85 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
86
87 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
88 u64 flags, const struct btf **btf,
89 s32 *btf_id);
90 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
91 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
92
93 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx);
94 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
95
96 /**
97 * trace_call_bpf - invoke BPF program
98 * @call: tracepoint event
99 * @ctx: opaque context pointer
100 *
101 * kprobe handlers execute BPF programs via this helper.
102 * Can be used from static tracepoints in the future.
103 *
104 * Return: BPF programs always return an integer which is interpreted by
105 * kprobe handler as:
106 * 0 - return from kprobe (event is filtered out)
107 * 1 - store kprobe event into ring buffer
108 * Other values are reserved and currently alias to 1
109 */
trace_call_bpf(struct trace_event_call * call,void * ctx)110 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
111 {
112 unsigned int ret;
113
114 cant_sleep();
115
116 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
117 /*
118 * since some bpf program is already running on this cpu,
119 * don't call into another bpf program (same or different)
120 * and don't send kprobe event into ring-buffer,
121 * so return zero here
122 */
123 rcu_read_lock();
124 bpf_prog_inc_misses_counters(rcu_dereference(call->prog_array));
125 rcu_read_unlock();
126 ret = 0;
127 goto out;
128 }
129
130 /*
131 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
132 * to all call sites, we did a bpf_prog_array_valid() there to check
133 * whether call->prog_array is empty or not, which is
134 * a heuristic to speed up execution.
135 *
136 * If bpf_prog_array_valid() fetched prog_array was
137 * non-NULL, we go into trace_call_bpf() and do the actual
138 * proper rcu_dereference() under RCU lock.
139 * If it turns out that prog_array is NULL then, we bail out.
140 * For the opposite, if the bpf_prog_array_valid() fetched pointer
141 * was NULL, you'll skip the prog_array with the risk of missing
142 * out of events when it was updated in between this and the
143 * rcu_dereference() which is accepted risk.
144 */
145 rcu_read_lock();
146 ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
147 ctx, bpf_prog_run);
148 rcu_read_unlock();
149
150 out:
151 __this_cpu_dec(bpf_prog_active);
152
153 return ret;
154 }
155
156 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
BPF_CALL_2(bpf_override_return,struct pt_regs *,regs,unsigned long,rc)157 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
158 {
159 regs_set_return_value(regs, rc);
160 override_function_with_return(regs);
161 return 0;
162 }
163
164 static const struct bpf_func_proto bpf_override_return_proto = {
165 .func = bpf_override_return,
166 .gpl_only = true,
167 .ret_type = RET_INTEGER,
168 .arg1_type = ARG_PTR_TO_CTX,
169 .arg2_type = ARG_ANYTHING,
170 };
171 #endif
172
173 static __always_inline int
bpf_probe_read_user_common(void * dst,u32 size,const void __user * unsafe_ptr)174 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
175 {
176 int ret;
177
178 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
179 if (unlikely(ret < 0))
180 memset(dst, 0, size);
181 return ret;
182 }
183
BPF_CALL_3(bpf_probe_read_user,void *,dst,u32,size,const void __user *,unsafe_ptr)184 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
185 const void __user *, unsafe_ptr)
186 {
187 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
188 }
189
190 const struct bpf_func_proto bpf_probe_read_user_proto = {
191 .func = bpf_probe_read_user,
192 .gpl_only = true,
193 .ret_type = RET_INTEGER,
194 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
195 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
196 .arg3_type = ARG_ANYTHING,
197 };
198
199 static __always_inline int
bpf_probe_read_user_str_common(void * dst,u32 size,const void __user * unsafe_ptr)200 bpf_probe_read_user_str_common(void *dst, u32 size,
201 const void __user *unsafe_ptr)
202 {
203 int ret;
204
205 /*
206 * NB: We rely on strncpy_from_user() not copying junk past the NUL
207 * terminator into `dst`.
208 *
209 * strncpy_from_user() does long-sized strides in the fast path. If the
210 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
211 * then there could be junk after the NUL in `dst`. If user takes `dst`
212 * and keys a hash map with it, then semantically identical strings can
213 * occupy multiple entries in the map.
214 */
215 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
216 if (unlikely(ret < 0))
217 memset(dst, 0, size);
218 return ret;
219 }
220
BPF_CALL_3(bpf_probe_read_user_str,void *,dst,u32,size,const void __user *,unsafe_ptr)221 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
222 const void __user *, unsafe_ptr)
223 {
224 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
225 }
226
227 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
228 .func = bpf_probe_read_user_str,
229 .gpl_only = true,
230 .ret_type = RET_INTEGER,
231 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
232 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
233 .arg3_type = ARG_ANYTHING,
234 };
235
BPF_CALL_3(bpf_probe_read_kernel,void *,dst,u32,size,const void *,unsafe_ptr)236 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
237 const void *, unsafe_ptr)
238 {
239 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
240 }
241
242 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
243 .func = bpf_probe_read_kernel,
244 .gpl_only = true,
245 .ret_type = RET_INTEGER,
246 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
247 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
248 .arg3_type = ARG_ANYTHING,
249 };
250
251 static __always_inline int
bpf_probe_read_kernel_str_common(void * dst,u32 size,const void * unsafe_ptr)252 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
253 {
254 int ret;
255
256 /*
257 * The strncpy_from_kernel_nofault() call will likely not fill the
258 * entire buffer, but that's okay in this circumstance as we're probing
259 * arbitrary memory anyway similar to bpf_probe_read_*() and might
260 * as well probe the stack. Thus, memory is explicitly cleared
261 * only in error case, so that improper users ignoring return
262 * code altogether don't copy garbage; otherwise length of string
263 * is returned that can be used for bpf_perf_event_output() et al.
264 */
265 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
266 if (unlikely(ret < 0))
267 memset(dst, 0, size);
268 return ret;
269 }
270
BPF_CALL_3(bpf_probe_read_kernel_str,void *,dst,u32,size,const void *,unsafe_ptr)271 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
272 const void *, unsafe_ptr)
273 {
274 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
275 }
276
277 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
278 .func = bpf_probe_read_kernel_str,
279 .gpl_only = true,
280 .ret_type = RET_INTEGER,
281 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
282 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
283 .arg3_type = ARG_ANYTHING,
284 };
285
286 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
BPF_CALL_3(bpf_probe_read_compat,void *,dst,u32,size,const void *,unsafe_ptr)287 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
288 const void *, unsafe_ptr)
289 {
290 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
291 return bpf_probe_read_user_common(dst, size,
292 (__force void __user *)unsafe_ptr);
293 }
294 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
295 }
296
297 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
298 .func = bpf_probe_read_compat,
299 .gpl_only = true,
300 .ret_type = RET_INTEGER,
301 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
302 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
303 .arg3_type = ARG_ANYTHING,
304 };
305
BPF_CALL_3(bpf_probe_read_compat_str,void *,dst,u32,size,const void *,unsafe_ptr)306 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
307 const void *, unsafe_ptr)
308 {
309 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
310 return bpf_probe_read_user_str_common(dst, size,
311 (__force void __user *)unsafe_ptr);
312 }
313 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
314 }
315
316 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
317 .func = bpf_probe_read_compat_str,
318 .gpl_only = true,
319 .ret_type = RET_INTEGER,
320 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
321 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
322 .arg3_type = ARG_ANYTHING,
323 };
324 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
325
BPF_CALL_3(bpf_probe_write_user,void __user *,unsafe_ptr,const void *,src,u32,size)326 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
327 u32, size)
328 {
329 /*
330 * Ensure we're in user context which is safe for the helper to
331 * run. This helper has no business in a kthread.
332 *
333 * access_ok() should prevent writing to non-user memory, but in
334 * some situations (nommu, temporary switch, etc) access_ok() does
335 * not provide enough validation, hence the check on KERNEL_DS.
336 *
337 * nmi_uaccess_okay() ensures the probe is not run in an interim
338 * state, when the task or mm are switched. This is specifically
339 * required to prevent the use of temporary mm.
340 */
341
342 if (unlikely(in_interrupt() ||
343 current->flags & (PF_KTHREAD | PF_EXITING)))
344 return -EPERM;
345 if (unlikely(!nmi_uaccess_okay()))
346 return -EPERM;
347
348 return copy_to_user_nofault(unsafe_ptr, src, size);
349 }
350
351 static const struct bpf_func_proto bpf_probe_write_user_proto = {
352 .func = bpf_probe_write_user,
353 .gpl_only = true,
354 .ret_type = RET_INTEGER,
355 .arg1_type = ARG_ANYTHING,
356 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
357 .arg3_type = ARG_CONST_SIZE,
358 };
359
360 #define MAX_TRACE_PRINTK_VARARGS 3
361 #define BPF_TRACE_PRINTK_SIZE 1024
362
BPF_CALL_5(bpf_trace_printk,char *,fmt,u32,fmt_size,u64,arg1,u64,arg2,u64,arg3)363 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
364 u64, arg2, u64, arg3)
365 {
366 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
367 struct bpf_bprintf_data data = {
368 .get_bin_args = true,
369 .get_buf = true,
370 };
371 int ret;
372
373 ret = bpf_bprintf_prepare(fmt, fmt_size, args,
374 MAX_TRACE_PRINTK_VARARGS, &data);
375 if (ret < 0)
376 return ret;
377
378 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
379
380 trace_bpf_trace_printk(data.buf);
381
382 bpf_bprintf_cleanup(&data);
383
384 return ret;
385 }
386
387 static const struct bpf_func_proto bpf_trace_printk_proto = {
388 .func = bpf_trace_printk,
389 .gpl_only = true,
390 .ret_type = RET_INTEGER,
391 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
392 .arg2_type = ARG_CONST_SIZE,
393 };
394
__set_printk_clr_event(void)395 static void __set_printk_clr_event(void)
396 {
397 /*
398 * This program might be calling bpf_trace_printk,
399 * so enable the associated bpf_trace/bpf_trace_printk event.
400 * Repeat this each time as it is possible a user has
401 * disabled bpf_trace_printk events. By loading a program
402 * calling bpf_trace_printk() however the user has expressed
403 * the intent to see such events.
404 */
405 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
406 pr_warn_ratelimited("could not enable bpf_trace_printk events");
407 }
408
bpf_get_trace_printk_proto(void)409 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
410 {
411 __set_printk_clr_event();
412 return &bpf_trace_printk_proto;
413 }
414
BPF_CALL_4(bpf_trace_vprintk,char *,fmt,u32,fmt_size,const void *,args,u32,data_len)415 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args,
416 u32, data_len)
417 {
418 struct bpf_bprintf_data data = {
419 .get_bin_args = true,
420 .get_buf = true,
421 };
422 int ret, num_args;
423
424 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
425 (data_len && !args))
426 return -EINVAL;
427 num_args = data_len / 8;
428
429 ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
430 if (ret < 0)
431 return ret;
432
433 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
434
435 trace_bpf_trace_printk(data.buf);
436
437 bpf_bprintf_cleanup(&data);
438
439 return ret;
440 }
441
442 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
443 .func = bpf_trace_vprintk,
444 .gpl_only = true,
445 .ret_type = RET_INTEGER,
446 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
447 .arg2_type = ARG_CONST_SIZE,
448 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
449 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
450 };
451
bpf_get_trace_vprintk_proto(void)452 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
453 {
454 __set_printk_clr_event();
455 return &bpf_trace_vprintk_proto;
456 }
457
BPF_CALL_5(bpf_seq_printf,struct seq_file *,m,char *,fmt,u32,fmt_size,const void *,args,u32,data_len)458 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
459 const void *, args, u32, data_len)
460 {
461 struct bpf_bprintf_data data = {
462 .get_bin_args = true,
463 };
464 int err, num_args;
465
466 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
467 (data_len && !args))
468 return -EINVAL;
469 num_args = data_len / 8;
470
471 err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
472 if (err < 0)
473 return err;
474
475 seq_bprintf(m, fmt, data.bin_args);
476
477 bpf_bprintf_cleanup(&data);
478
479 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
480 }
481
482 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
483
484 static const struct bpf_func_proto bpf_seq_printf_proto = {
485 .func = bpf_seq_printf,
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 | MEM_RDONLY,
491 .arg3_type = ARG_CONST_SIZE,
492 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
493 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
494 };
495
BPF_CALL_3(bpf_seq_write,struct seq_file *,m,const void *,data,u32,len)496 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
497 {
498 return seq_write(m, data, len) ? -EOVERFLOW : 0;
499 }
500
501 static const struct bpf_func_proto bpf_seq_write_proto = {
502 .func = bpf_seq_write,
503 .gpl_only = true,
504 .ret_type = RET_INTEGER,
505 .arg1_type = ARG_PTR_TO_BTF_ID,
506 .arg1_btf_id = &btf_seq_file_ids[0],
507 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
508 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
509 };
510
BPF_CALL_4(bpf_seq_printf_btf,struct seq_file *,m,struct btf_ptr *,ptr,u32,btf_ptr_size,u64,flags)511 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
512 u32, btf_ptr_size, u64, flags)
513 {
514 const struct btf *btf;
515 s32 btf_id;
516 int ret;
517
518 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
519 if (ret)
520 return ret;
521
522 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
523 }
524
525 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
526 .func = bpf_seq_printf_btf,
527 .gpl_only = true,
528 .ret_type = RET_INTEGER,
529 .arg1_type = ARG_PTR_TO_BTF_ID,
530 .arg1_btf_id = &btf_seq_file_ids[0],
531 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
532 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
533 .arg4_type = ARG_ANYTHING,
534 };
535
536 static __always_inline int
get_map_perf_counter(struct bpf_map * map,u64 flags,u64 * value,u64 * enabled,u64 * running)537 get_map_perf_counter(struct bpf_map *map, u64 flags,
538 u64 *value, u64 *enabled, u64 *running)
539 {
540 struct bpf_array *array = container_of(map, struct bpf_array, map);
541 unsigned int cpu = smp_processor_id();
542 u64 index = flags & BPF_F_INDEX_MASK;
543 struct bpf_event_entry *ee;
544
545 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
546 return -EINVAL;
547 if (index == BPF_F_CURRENT_CPU)
548 index = cpu;
549 if (unlikely(index >= array->map.max_entries))
550 return -E2BIG;
551
552 ee = READ_ONCE(array->ptrs[index]);
553 if (!ee)
554 return -ENOENT;
555
556 return perf_event_read_local(ee->event, value, enabled, running);
557 }
558
BPF_CALL_2(bpf_perf_event_read,struct bpf_map *,map,u64,flags)559 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
560 {
561 u64 value = 0;
562 int err;
563
564 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
565 /*
566 * this api is ugly since we miss [-22..-2] range of valid
567 * counter values, but that's uapi
568 */
569 if (err)
570 return err;
571 return value;
572 }
573
574 static const struct bpf_func_proto bpf_perf_event_read_proto = {
575 .func = bpf_perf_event_read,
576 .gpl_only = true,
577 .ret_type = RET_INTEGER,
578 .arg1_type = ARG_CONST_MAP_PTR,
579 .arg2_type = ARG_ANYTHING,
580 };
581
BPF_CALL_4(bpf_perf_event_read_value,struct bpf_map *,map,u64,flags,struct bpf_perf_event_value *,buf,u32,size)582 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
583 struct bpf_perf_event_value *, buf, u32, size)
584 {
585 int err = -EINVAL;
586
587 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
588 goto clear;
589 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
590 &buf->running);
591 if (unlikely(err))
592 goto clear;
593 return 0;
594 clear:
595 memset(buf, 0, size);
596 return err;
597 }
598
599 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
600 .func = bpf_perf_event_read_value,
601 .gpl_only = true,
602 .ret_type = RET_INTEGER,
603 .arg1_type = ARG_CONST_MAP_PTR,
604 .arg2_type = ARG_ANYTHING,
605 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
606 .arg4_type = ARG_CONST_SIZE,
607 };
608
609 static __always_inline u64
__bpf_perf_event_output(struct pt_regs * regs,struct bpf_map * map,u64 flags,struct perf_raw_record * raw,struct perf_sample_data * sd)610 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
611 u64 flags, struct perf_raw_record *raw,
612 struct perf_sample_data *sd)
613 {
614 struct bpf_array *array = container_of(map, struct bpf_array, map);
615 unsigned int cpu = smp_processor_id();
616 u64 index = flags & BPF_F_INDEX_MASK;
617 struct bpf_event_entry *ee;
618 struct perf_event *event;
619
620 if (index == BPF_F_CURRENT_CPU)
621 index = cpu;
622 if (unlikely(index >= array->map.max_entries))
623 return -E2BIG;
624
625 ee = READ_ONCE(array->ptrs[index]);
626 if (!ee)
627 return -ENOENT;
628
629 event = ee->event;
630 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
631 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
632 return -EINVAL;
633
634 if (unlikely(event->oncpu != cpu))
635 return -EOPNOTSUPP;
636
637 perf_sample_save_raw_data(sd, event, raw);
638
639 return perf_event_output(event, sd, regs);
640 }
641
642 /*
643 * Support executing tracepoints in normal, irq, and nmi context that each call
644 * bpf_perf_event_output
645 */
646 struct bpf_trace_sample_data {
647 struct perf_sample_data sds[3];
648 };
649
650 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
651 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
BPF_CALL_5(bpf_perf_event_output,struct pt_regs *,regs,struct bpf_map *,map,u64,flags,void *,data,u64,size)652 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
653 u64, flags, void *, data, u64, size)
654 {
655 struct bpf_trace_sample_data *sds;
656 struct perf_raw_record raw = {
657 .frag = {
658 .size = size,
659 .data = data,
660 },
661 };
662 struct perf_sample_data *sd;
663 int nest_level, err;
664
665 preempt_disable();
666 sds = this_cpu_ptr(&bpf_trace_sds);
667 nest_level = this_cpu_inc_return(bpf_trace_nest_level);
668
669 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
670 err = -EBUSY;
671 goto out;
672 }
673
674 sd = &sds->sds[nest_level - 1];
675
676 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
677 err = -EINVAL;
678 goto out;
679 }
680
681 perf_sample_data_init(sd, 0, 0);
682
683 err = __bpf_perf_event_output(regs, map, flags, &raw, sd);
684 out:
685 this_cpu_dec(bpf_trace_nest_level);
686 preempt_enable();
687 return err;
688 }
689
690 static const struct bpf_func_proto bpf_perf_event_output_proto = {
691 .func = bpf_perf_event_output,
692 .gpl_only = true,
693 .ret_type = RET_INTEGER,
694 .arg1_type = ARG_PTR_TO_CTX,
695 .arg2_type = ARG_CONST_MAP_PTR,
696 .arg3_type = ARG_ANYTHING,
697 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
698 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
699 };
700
701 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
702 struct bpf_nested_pt_regs {
703 struct pt_regs regs[3];
704 };
705 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
706 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
707
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)708 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
709 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
710 {
711 struct perf_raw_frag frag = {
712 .copy = ctx_copy,
713 .size = ctx_size,
714 .data = ctx,
715 };
716 struct perf_raw_record raw = {
717 .frag = {
718 {
719 .next = ctx_size ? &frag : NULL,
720 },
721 .size = meta_size,
722 .data = meta,
723 },
724 };
725 struct perf_sample_data *sd;
726 struct pt_regs *regs;
727 int nest_level;
728 u64 ret;
729
730 preempt_disable();
731 nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
732
733 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
734 ret = -EBUSY;
735 goto out;
736 }
737 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
738 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
739
740 perf_fetch_caller_regs(regs);
741 perf_sample_data_init(sd, 0, 0);
742
743 ret = __bpf_perf_event_output(regs, map, flags, &raw, sd);
744 out:
745 this_cpu_dec(bpf_event_output_nest_level);
746 preempt_enable();
747 return ret;
748 }
749
BPF_CALL_0(bpf_get_current_task)750 BPF_CALL_0(bpf_get_current_task)
751 {
752 return (long) current;
753 }
754
755 const struct bpf_func_proto bpf_get_current_task_proto = {
756 .func = bpf_get_current_task,
757 .gpl_only = true,
758 .ret_type = RET_INTEGER,
759 };
760
BPF_CALL_0(bpf_get_current_task_btf)761 BPF_CALL_0(bpf_get_current_task_btf)
762 {
763 return (unsigned long) current;
764 }
765
766 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
767 .func = bpf_get_current_task_btf,
768 .gpl_only = true,
769 .ret_type = RET_PTR_TO_BTF_ID_TRUSTED,
770 .ret_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
771 };
772
BPF_CALL_1(bpf_task_pt_regs,struct task_struct *,task)773 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
774 {
775 return (unsigned long) task_pt_regs(task);
776 }
777
778 BTF_ID_LIST(bpf_task_pt_regs_ids)
779 BTF_ID(struct, pt_regs)
780
781 const struct bpf_func_proto bpf_task_pt_regs_proto = {
782 .func = bpf_task_pt_regs,
783 .gpl_only = true,
784 .arg1_type = ARG_PTR_TO_BTF_ID,
785 .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
786 .ret_type = RET_PTR_TO_BTF_ID,
787 .ret_btf_id = &bpf_task_pt_regs_ids[0],
788 };
789
790 struct send_signal_irq_work {
791 struct irq_work irq_work;
792 struct task_struct *task;
793 u32 sig;
794 enum pid_type type;
795 bool has_siginfo;
796 struct kernel_siginfo info;
797 };
798
799 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
800
do_bpf_send_signal(struct irq_work * entry)801 static void do_bpf_send_signal(struct irq_work *entry)
802 {
803 struct send_signal_irq_work *work;
804 struct kernel_siginfo *siginfo;
805
806 work = container_of(entry, struct send_signal_irq_work, irq_work);
807 siginfo = work->has_siginfo ? &work->info : SEND_SIG_PRIV;
808
809 group_send_sig_info(work->sig, siginfo, work->task, work->type);
810 put_task_struct(work->task);
811 }
812
bpf_send_signal_common(u32 sig,enum pid_type type,struct task_struct * task,u64 value)813 static int bpf_send_signal_common(u32 sig, enum pid_type type, struct task_struct *task, u64 value)
814 {
815 struct send_signal_irq_work *work = NULL;
816 struct kernel_siginfo info;
817 struct kernel_siginfo *siginfo;
818
819 if (!task) {
820 task = current;
821 siginfo = SEND_SIG_PRIV;
822 } else {
823 clear_siginfo(&info);
824 info.si_signo = sig;
825 info.si_errno = 0;
826 info.si_code = SI_KERNEL;
827 info.si_pid = 0;
828 info.si_uid = 0;
829 info.si_value.sival_ptr = (void *)(unsigned long)value;
830 siginfo = &info;
831 }
832
833 /* Similar to bpf_probe_write_user, task needs to be
834 * in a sound condition and kernel memory access be
835 * permitted in order to send signal to the current
836 * task.
837 */
838 if (unlikely(task->flags & (PF_KTHREAD | PF_EXITING)))
839 return -EPERM;
840 if (unlikely(!nmi_uaccess_okay()))
841 return -EPERM;
842 /* Task should not be pid=1 to avoid kernel panic. */
843 if (unlikely(is_global_init(task)))
844 return -EPERM;
845
846 if (!preemptible()) {
847 /* Do an early check on signal validity. Otherwise,
848 * the error is lost in deferred irq_work.
849 */
850 if (unlikely(!valid_signal(sig)))
851 return -EINVAL;
852
853 work = this_cpu_ptr(&send_signal_work);
854 if (irq_work_is_busy(&work->irq_work))
855 return -EBUSY;
856
857 /* Add the current task, which is the target of sending signal,
858 * to the irq_work. The current task may change when queued
859 * irq works get executed.
860 */
861 work->task = get_task_struct(task);
862 work->has_siginfo = siginfo == &info;
863 if (work->has_siginfo)
864 copy_siginfo(&work->info, &info);
865 work->sig = sig;
866 work->type = type;
867 irq_work_queue(&work->irq_work);
868 return 0;
869 }
870
871 return group_send_sig_info(sig, siginfo, task, type);
872 }
873
BPF_CALL_1(bpf_send_signal,u32,sig)874 BPF_CALL_1(bpf_send_signal, u32, sig)
875 {
876 return bpf_send_signal_common(sig, PIDTYPE_TGID, NULL, 0);
877 }
878
879 static const struct bpf_func_proto bpf_send_signal_proto = {
880 .func = bpf_send_signal,
881 .gpl_only = false,
882 .ret_type = RET_INTEGER,
883 .arg1_type = ARG_ANYTHING,
884 };
885
BPF_CALL_1(bpf_send_signal_thread,u32,sig)886 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
887 {
888 return bpf_send_signal_common(sig, PIDTYPE_PID, NULL, 0);
889 }
890
891 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
892 .func = bpf_send_signal_thread,
893 .gpl_only = false,
894 .ret_type = RET_INTEGER,
895 .arg1_type = ARG_ANYTHING,
896 };
897
BPF_CALL_3(bpf_d_path,struct path *,path,char *,buf,u32,sz)898 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
899 {
900 struct path copy;
901 long len;
902 char *p;
903
904 if (!sz)
905 return 0;
906
907 /*
908 * The path pointer is verified as trusted and safe to use,
909 * but let's double check it's valid anyway to workaround
910 * potentially broken verifier.
911 */
912 len = copy_from_kernel_nofault(©, path, sizeof(*path));
913 if (len < 0)
914 return len;
915
916 p = d_path(©, buf, sz);
917 if (IS_ERR(p)) {
918 len = PTR_ERR(p);
919 } else {
920 len = buf + sz - p;
921 memmove(buf, p, len);
922 }
923
924 return len;
925 }
926
927 BTF_SET_START(btf_allowlist_d_path)
928 #ifdef CONFIG_SECURITY
BTF_ID(func,security_file_permission)929 BTF_ID(func, security_file_permission)
930 BTF_ID(func, security_inode_getattr)
931 BTF_ID(func, security_file_open)
932 #endif
933 #ifdef CONFIG_SECURITY_PATH
934 BTF_ID(func, security_path_truncate)
935 #endif
936 BTF_ID(func, vfs_truncate)
937 BTF_ID(func, vfs_fallocate)
938 BTF_ID(func, dentry_open)
939 BTF_ID(func, vfs_getattr)
940 BTF_ID(func, filp_close)
941 BTF_SET_END(btf_allowlist_d_path)
942
943 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
944 {
945 if (prog->type == BPF_PROG_TYPE_TRACING &&
946 prog->expected_attach_type == BPF_TRACE_ITER)
947 return true;
948
949 if (prog->type == BPF_PROG_TYPE_LSM)
950 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
951
952 return btf_id_set_contains(&btf_allowlist_d_path,
953 prog->aux->attach_btf_id);
954 }
955
956 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
957
958 static const struct bpf_func_proto bpf_d_path_proto = {
959 .func = bpf_d_path,
960 .gpl_only = false,
961 .ret_type = RET_INTEGER,
962 .arg1_type = ARG_PTR_TO_BTF_ID,
963 .arg1_btf_id = &bpf_d_path_btf_ids[0],
964 .arg2_type = ARG_PTR_TO_MEM,
965 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
966 .allowed = bpf_d_path_allowed,
967 };
968
969 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
970 BTF_F_PTR_RAW | BTF_F_ZERO)
971
bpf_btf_printf_prepare(struct btf_ptr * ptr,u32 btf_ptr_size,u64 flags,const struct btf ** btf,s32 * btf_id)972 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
973 u64 flags, const struct btf **btf,
974 s32 *btf_id)
975 {
976 const struct btf_type *t;
977
978 if (unlikely(flags & ~(BTF_F_ALL)))
979 return -EINVAL;
980
981 if (btf_ptr_size != sizeof(struct btf_ptr))
982 return -EINVAL;
983
984 *btf = bpf_get_btf_vmlinux();
985
986 if (IS_ERR_OR_NULL(*btf))
987 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
988
989 if (ptr->type_id > 0)
990 *btf_id = ptr->type_id;
991 else
992 return -EINVAL;
993
994 if (*btf_id > 0)
995 t = btf_type_by_id(*btf, *btf_id);
996 if (*btf_id <= 0 || !t)
997 return -ENOENT;
998
999 return 0;
1000 }
1001
BPF_CALL_5(bpf_snprintf_btf,char *,str,u32,str_size,struct btf_ptr *,ptr,u32,btf_ptr_size,u64,flags)1002 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1003 u32, btf_ptr_size, u64, flags)
1004 {
1005 const struct btf *btf;
1006 s32 btf_id;
1007 int ret;
1008
1009 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1010 if (ret)
1011 return ret;
1012
1013 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1014 flags);
1015 }
1016
1017 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1018 .func = bpf_snprintf_btf,
1019 .gpl_only = false,
1020 .ret_type = RET_INTEGER,
1021 .arg1_type = ARG_PTR_TO_MEM,
1022 .arg2_type = ARG_CONST_SIZE,
1023 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1024 .arg4_type = ARG_CONST_SIZE,
1025 .arg5_type = ARG_ANYTHING,
1026 };
1027
BPF_CALL_1(bpf_get_func_ip_tracing,void *,ctx)1028 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1029 {
1030 /* This helper call is inlined by verifier. */
1031 return ((u64 *)ctx)[-2];
1032 }
1033
1034 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1035 .func = bpf_get_func_ip_tracing,
1036 .gpl_only = true,
1037 .ret_type = RET_INTEGER,
1038 .arg1_type = ARG_PTR_TO_CTX,
1039 };
1040
get_entry_ip(unsigned long fentry_ip)1041 static inline unsigned long get_entry_ip(unsigned long fentry_ip)
1042 {
1043 #ifdef CONFIG_X86_KERNEL_IBT
1044 if (is_endbr((void *)(fentry_ip - ENDBR_INSN_SIZE)))
1045 fentry_ip -= ENDBR_INSN_SIZE;
1046 #endif
1047 return fentry_ip;
1048 }
1049
BPF_CALL_1(bpf_get_func_ip_kprobe,struct pt_regs *,regs)1050 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1051 {
1052 struct bpf_trace_run_ctx *run_ctx __maybe_unused;
1053 struct kprobe *kp;
1054
1055 #ifdef CONFIG_UPROBES
1056 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1057 if (run_ctx->is_uprobe)
1058 return ((struct uprobe_dispatch_data *)current->utask->vaddr)->bp_addr;
1059 #endif
1060
1061 kp = kprobe_running();
1062
1063 if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1064 return 0;
1065
1066 return get_entry_ip((uintptr_t)kp->addr);
1067 }
1068
1069 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1070 .func = bpf_get_func_ip_kprobe,
1071 .gpl_only = true,
1072 .ret_type = RET_INTEGER,
1073 .arg1_type = ARG_PTR_TO_CTX,
1074 };
1075
BPF_CALL_1(bpf_get_func_ip_kprobe_multi,struct pt_regs *,regs)1076 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1077 {
1078 return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1079 }
1080
1081 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1082 .func = bpf_get_func_ip_kprobe_multi,
1083 .gpl_only = false,
1084 .ret_type = RET_INTEGER,
1085 .arg1_type = ARG_PTR_TO_CTX,
1086 };
1087
BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi,struct pt_regs *,regs)1088 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1089 {
1090 return bpf_kprobe_multi_cookie(current->bpf_ctx);
1091 }
1092
1093 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1094 .func = bpf_get_attach_cookie_kprobe_multi,
1095 .gpl_only = false,
1096 .ret_type = RET_INTEGER,
1097 .arg1_type = ARG_PTR_TO_CTX,
1098 };
1099
BPF_CALL_1(bpf_get_func_ip_uprobe_multi,struct pt_regs *,regs)1100 BPF_CALL_1(bpf_get_func_ip_uprobe_multi, struct pt_regs *, regs)
1101 {
1102 return bpf_uprobe_multi_entry_ip(current->bpf_ctx);
1103 }
1104
1105 static const struct bpf_func_proto bpf_get_func_ip_proto_uprobe_multi = {
1106 .func = bpf_get_func_ip_uprobe_multi,
1107 .gpl_only = false,
1108 .ret_type = RET_INTEGER,
1109 .arg1_type = ARG_PTR_TO_CTX,
1110 };
1111
BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi,struct pt_regs *,regs)1112 BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi, struct pt_regs *, regs)
1113 {
1114 return bpf_uprobe_multi_cookie(current->bpf_ctx);
1115 }
1116
1117 static const struct bpf_func_proto bpf_get_attach_cookie_proto_umulti = {
1118 .func = bpf_get_attach_cookie_uprobe_multi,
1119 .gpl_only = false,
1120 .ret_type = RET_INTEGER,
1121 .arg1_type = ARG_PTR_TO_CTX,
1122 };
1123
BPF_CALL_1(bpf_get_attach_cookie_trace,void *,ctx)1124 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1125 {
1126 struct bpf_trace_run_ctx *run_ctx;
1127
1128 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1129 return run_ctx->bpf_cookie;
1130 }
1131
1132 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1133 .func = bpf_get_attach_cookie_trace,
1134 .gpl_only = false,
1135 .ret_type = RET_INTEGER,
1136 .arg1_type = ARG_PTR_TO_CTX,
1137 };
1138
BPF_CALL_1(bpf_get_attach_cookie_pe,struct bpf_perf_event_data_kern *,ctx)1139 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1140 {
1141 return ctx->event->bpf_cookie;
1142 }
1143
1144 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1145 .func = bpf_get_attach_cookie_pe,
1146 .gpl_only = false,
1147 .ret_type = RET_INTEGER,
1148 .arg1_type = ARG_PTR_TO_CTX,
1149 };
1150
BPF_CALL_1(bpf_get_attach_cookie_tracing,void *,ctx)1151 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1152 {
1153 struct bpf_trace_run_ctx *run_ctx;
1154
1155 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1156 return run_ctx->bpf_cookie;
1157 }
1158
1159 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1160 .func = bpf_get_attach_cookie_tracing,
1161 .gpl_only = false,
1162 .ret_type = RET_INTEGER,
1163 .arg1_type = ARG_PTR_TO_CTX,
1164 };
1165
BPF_CALL_3(bpf_get_branch_snapshot,void *,buf,u32,size,u64,flags)1166 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1167 {
1168 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1169 u32 entry_cnt = size / br_entry_size;
1170
1171 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1172
1173 if (unlikely(flags))
1174 return -EINVAL;
1175
1176 if (!entry_cnt)
1177 return -ENOENT;
1178
1179 return entry_cnt * br_entry_size;
1180 }
1181
1182 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1183 .func = bpf_get_branch_snapshot,
1184 .gpl_only = true,
1185 .ret_type = RET_INTEGER,
1186 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1187 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1188 };
1189
BPF_CALL_3(get_func_arg,void *,ctx,u32,n,u64 *,value)1190 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1191 {
1192 /* This helper call is inlined by verifier. */
1193 u64 nr_args = ((u64 *)ctx)[-1];
1194
1195 if ((u64) n >= nr_args)
1196 return -EINVAL;
1197 *value = ((u64 *)ctx)[n];
1198 return 0;
1199 }
1200
1201 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1202 .func = get_func_arg,
1203 .ret_type = RET_INTEGER,
1204 .arg1_type = ARG_PTR_TO_CTX,
1205 .arg2_type = ARG_ANYTHING,
1206 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
1207 .arg3_size = sizeof(u64),
1208 };
1209
BPF_CALL_2(get_func_ret,void *,ctx,u64 *,value)1210 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1211 {
1212 /* This helper call is inlined by verifier. */
1213 u64 nr_args = ((u64 *)ctx)[-1];
1214
1215 *value = ((u64 *)ctx)[nr_args];
1216 return 0;
1217 }
1218
1219 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1220 .func = get_func_ret,
1221 .ret_type = RET_INTEGER,
1222 .arg1_type = ARG_PTR_TO_CTX,
1223 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
1224 .arg2_size = sizeof(u64),
1225 };
1226
BPF_CALL_1(get_func_arg_cnt,void *,ctx)1227 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1228 {
1229 /* This helper call is inlined by verifier. */
1230 return ((u64 *)ctx)[-1];
1231 }
1232
1233 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1234 .func = get_func_arg_cnt,
1235 .ret_type = RET_INTEGER,
1236 .arg1_type = ARG_PTR_TO_CTX,
1237 };
1238
1239 #ifdef CONFIG_KEYS
1240 __bpf_kfunc_start_defs();
1241
1242 /**
1243 * bpf_lookup_user_key - lookup a key by its serial
1244 * @serial: key handle serial number
1245 * @flags: lookup-specific flags
1246 *
1247 * Search a key with a given *serial* and the provided *flags*.
1248 * If found, increment the reference count of the key by one, and
1249 * return it in the bpf_key structure.
1250 *
1251 * The bpf_key structure must be passed to bpf_key_put() when done
1252 * with it, so that the key reference count is decremented and the
1253 * bpf_key structure is freed.
1254 *
1255 * Permission checks are deferred to the time the key is used by
1256 * one of the available key-specific kfuncs.
1257 *
1258 * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1259 * special keyring (e.g. session keyring), if it doesn't yet exist.
1260 * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1261 * for the key construction, and to retrieve uninstantiated keys (keys
1262 * without data attached to them).
1263 *
1264 * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1265 * NULL pointer otherwise.
1266 */
bpf_lookup_user_key(u32 serial,u64 flags)1267 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1268 {
1269 key_ref_t key_ref;
1270 struct bpf_key *bkey;
1271
1272 if (flags & ~KEY_LOOKUP_ALL)
1273 return NULL;
1274
1275 /*
1276 * Permission check is deferred until the key is used, as the
1277 * intent of the caller is unknown here.
1278 */
1279 key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1280 if (IS_ERR(key_ref))
1281 return NULL;
1282
1283 bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1284 if (!bkey) {
1285 key_put(key_ref_to_ptr(key_ref));
1286 return NULL;
1287 }
1288
1289 bkey->key = key_ref_to_ptr(key_ref);
1290 bkey->has_ref = true;
1291
1292 return bkey;
1293 }
1294
1295 /**
1296 * bpf_lookup_system_key - lookup a key by a system-defined ID
1297 * @id: key ID
1298 *
1299 * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1300 * The key pointer is marked as invalid, to prevent bpf_key_put() from
1301 * attempting to decrement the key reference count on that pointer. The key
1302 * pointer set in such way is currently understood only by
1303 * verify_pkcs7_signature().
1304 *
1305 * Set *id* to one of the values defined in include/linux/verification.h:
1306 * 0 for the primary keyring (immutable keyring of system keys);
1307 * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1308 * (where keys can be added only if they are vouched for by existing keys
1309 * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1310 * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1311 * kerned image and, possibly, the initramfs signature).
1312 *
1313 * Return: a bpf_key pointer with an invalid key pointer set from the
1314 * pre-determined ID on success, a NULL pointer otherwise
1315 */
bpf_lookup_system_key(u64 id)1316 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id)
1317 {
1318 struct bpf_key *bkey;
1319
1320 if (system_keyring_id_check(id) < 0)
1321 return NULL;
1322
1323 bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1324 if (!bkey)
1325 return NULL;
1326
1327 bkey->key = (struct key *)(unsigned long)id;
1328 bkey->has_ref = false;
1329
1330 return bkey;
1331 }
1332
1333 /**
1334 * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1335 * @bkey: bpf_key structure
1336 *
1337 * Decrement the reference count of the key inside *bkey*, if the pointer
1338 * is valid, and free *bkey*.
1339 */
bpf_key_put(struct bpf_key * bkey)1340 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey)
1341 {
1342 if (bkey->has_ref)
1343 key_put(bkey->key);
1344
1345 kfree(bkey);
1346 }
1347
1348 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1349 /**
1350 * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1351 * @data_p: data to verify
1352 * @sig_p: signature of the data
1353 * @trusted_keyring: keyring with keys trusted for signature verification
1354 *
1355 * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1356 * with keys in a keyring referenced by *trusted_keyring*.
1357 *
1358 * Return: 0 on success, a negative value on error.
1359 */
bpf_verify_pkcs7_signature(struct bpf_dynptr * data_p,struct bpf_dynptr * sig_p,struct bpf_key * trusted_keyring)1360 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_p,
1361 struct bpf_dynptr *sig_p,
1362 struct bpf_key *trusted_keyring)
1363 {
1364 struct bpf_dynptr_kern *data_ptr = (struct bpf_dynptr_kern *)data_p;
1365 struct bpf_dynptr_kern *sig_ptr = (struct bpf_dynptr_kern *)sig_p;
1366 const void *data, *sig;
1367 u32 data_len, sig_len;
1368 int ret;
1369
1370 if (trusted_keyring->has_ref) {
1371 /*
1372 * Do the permission check deferred in bpf_lookup_user_key().
1373 * See bpf_lookup_user_key() for more details.
1374 *
1375 * A call to key_task_permission() here would be redundant, as
1376 * it is already done by keyring_search() called by
1377 * find_asymmetric_key().
1378 */
1379 ret = key_validate(trusted_keyring->key);
1380 if (ret < 0)
1381 return ret;
1382 }
1383
1384 data_len = __bpf_dynptr_size(data_ptr);
1385 data = __bpf_dynptr_data(data_ptr, data_len);
1386 sig_len = __bpf_dynptr_size(sig_ptr);
1387 sig = __bpf_dynptr_data(sig_ptr, sig_len);
1388
1389 return verify_pkcs7_signature(data, data_len, sig, sig_len,
1390 trusted_keyring->key,
1391 VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1392 NULL);
1393 }
1394 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1395
1396 __bpf_kfunc_end_defs();
1397
1398 BTF_KFUNCS_START(key_sig_kfunc_set)
1399 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1400 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1401 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1402 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1403 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1404 #endif
1405 BTF_KFUNCS_END(key_sig_kfunc_set)
1406
1407 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1408 .owner = THIS_MODULE,
1409 .set = &key_sig_kfunc_set,
1410 };
1411
bpf_key_sig_kfuncs_init(void)1412 static int __init bpf_key_sig_kfuncs_init(void)
1413 {
1414 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1415 &bpf_key_sig_kfunc_set);
1416 }
1417
1418 late_initcall(bpf_key_sig_kfuncs_init);
1419 #endif /* CONFIG_KEYS */
1420
1421 static const struct bpf_func_proto *
bpf_tracing_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1422 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1423 {
1424 const struct bpf_func_proto *func_proto;
1425
1426 switch (func_id) {
1427 case BPF_FUNC_map_lookup_elem:
1428 return &bpf_map_lookup_elem_proto;
1429 case BPF_FUNC_map_update_elem:
1430 return &bpf_map_update_elem_proto;
1431 case BPF_FUNC_map_delete_elem:
1432 return &bpf_map_delete_elem_proto;
1433 case BPF_FUNC_map_push_elem:
1434 return &bpf_map_push_elem_proto;
1435 case BPF_FUNC_map_pop_elem:
1436 return &bpf_map_pop_elem_proto;
1437 case BPF_FUNC_map_peek_elem:
1438 return &bpf_map_peek_elem_proto;
1439 case BPF_FUNC_map_lookup_percpu_elem:
1440 return &bpf_map_lookup_percpu_elem_proto;
1441 case BPF_FUNC_ktime_get_ns:
1442 return &bpf_ktime_get_ns_proto;
1443 case BPF_FUNC_ktime_get_boot_ns:
1444 return &bpf_ktime_get_boot_ns_proto;
1445 case BPF_FUNC_tail_call:
1446 return &bpf_tail_call_proto;
1447 case BPF_FUNC_get_current_task:
1448 return &bpf_get_current_task_proto;
1449 case BPF_FUNC_get_current_task_btf:
1450 return &bpf_get_current_task_btf_proto;
1451 case BPF_FUNC_task_pt_regs:
1452 return &bpf_task_pt_regs_proto;
1453 case BPF_FUNC_get_current_uid_gid:
1454 return &bpf_get_current_uid_gid_proto;
1455 case BPF_FUNC_get_current_comm:
1456 return &bpf_get_current_comm_proto;
1457 case BPF_FUNC_trace_printk:
1458 return bpf_get_trace_printk_proto();
1459 case BPF_FUNC_get_smp_processor_id:
1460 return &bpf_get_smp_processor_id_proto;
1461 case BPF_FUNC_get_numa_node_id:
1462 return &bpf_get_numa_node_id_proto;
1463 case BPF_FUNC_perf_event_read:
1464 return &bpf_perf_event_read_proto;
1465 case BPF_FUNC_get_prandom_u32:
1466 return &bpf_get_prandom_u32_proto;
1467 case BPF_FUNC_probe_read_user:
1468 return &bpf_probe_read_user_proto;
1469 case BPF_FUNC_probe_read_kernel:
1470 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1471 NULL : &bpf_probe_read_kernel_proto;
1472 case BPF_FUNC_probe_read_user_str:
1473 return &bpf_probe_read_user_str_proto;
1474 case BPF_FUNC_probe_read_kernel_str:
1475 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1476 NULL : &bpf_probe_read_kernel_str_proto;
1477 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1478 case BPF_FUNC_probe_read:
1479 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1480 NULL : &bpf_probe_read_compat_proto;
1481 case BPF_FUNC_probe_read_str:
1482 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1483 NULL : &bpf_probe_read_compat_str_proto;
1484 #endif
1485 #ifdef CONFIG_CGROUPS
1486 case BPF_FUNC_cgrp_storage_get:
1487 return &bpf_cgrp_storage_get_proto;
1488 case BPF_FUNC_cgrp_storage_delete:
1489 return &bpf_cgrp_storage_delete_proto;
1490 case BPF_FUNC_current_task_under_cgroup:
1491 return &bpf_current_task_under_cgroup_proto;
1492 #endif
1493 case BPF_FUNC_send_signal:
1494 return &bpf_send_signal_proto;
1495 case BPF_FUNC_send_signal_thread:
1496 return &bpf_send_signal_thread_proto;
1497 case BPF_FUNC_perf_event_read_value:
1498 return &bpf_perf_event_read_value_proto;
1499 case BPF_FUNC_ringbuf_output:
1500 return &bpf_ringbuf_output_proto;
1501 case BPF_FUNC_ringbuf_reserve:
1502 return &bpf_ringbuf_reserve_proto;
1503 case BPF_FUNC_ringbuf_submit:
1504 return &bpf_ringbuf_submit_proto;
1505 case BPF_FUNC_ringbuf_discard:
1506 return &bpf_ringbuf_discard_proto;
1507 case BPF_FUNC_ringbuf_query:
1508 return &bpf_ringbuf_query_proto;
1509 case BPF_FUNC_jiffies64:
1510 return &bpf_jiffies64_proto;
1511 case BPF_FUNC_get_task_stack:
1512 return prog->sleepable ? &bpf_get_task_stack_sleepable_proto
1513 : &bpf_get_task_stack_proto;
1514 case BPF_FUNC_copy_from_user:
1515 return &bpf_copy_from_user_proto;
1516 case BPF_FUNC_copy_from_user_task:
1517 return &bpf_copy_from_user_task_proto;
1518 case BPF_FUNC_snprintf_btf:
1519 return &bpf_snprintf_btf_proto;
1520 case BPF_FUNC_per_cpu_ptr:
1521 return &bpf_per_cpu_ptr_proto;
1522 case BPF_FUNC_this_cpu_ptr:
1523 return &bpf_this_cpu_ptr_proto;
1524 case BPF_FUNC_task_storage_get:
1525 if (bpf_prog_check_recur(prog))
1526 return &bpf_task_storage_get_recur_proto;
1527 return &bpf_task_storage_get_proto;
1528 case BPF_FUNC_task_storage_delete:
1529 if (bpf_prog_check_recur(prog))
1530 return &bpf_task_storage_delete_recur_proto;
1531 return &bpf_task_storage_delete_proto;
1532 case BPF_FUNC_for_each_map_elem:
1533 return &bpf_for_each_map_elem_proto;
1534 case BPF_FUNC_snprintf:
1535 return &bpf_snprintf_proto;
1536 case BPF_FUNC_get_func_ip:
1537 return &bpf_get_func_ip_proto_tracing;
1538 case BPF_FUNC_get_branch_snapshot:
1539 return &bpf_get_branch_snapshot_proto;
1540 case BPF_FUNC_find_vma:
1541 return &bpf_find_vma_proto;
1542 case BPF_FUNC_trace_vprintk:
1543 return bpf_get_trace_vprintk_proto();
1544 default:
1545 break;
1546 }
1547
1548 func_proto = bpf_base_func_proto(func_id, prog);
1549 if (func_proto)
1550 return func_proto;
1551
1552 if (!bpf_token_capable(prog->aux->token, CAP_SYS_ADMIN))
1553 return NULL;
1554
1555 switch (func_id) {
1556 case BPF_FUNC_probe_write_user:
1557 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1558 NULL : &bpf_probe_write_user_proto;
1559 default:
1560 return NULL;
1561 }
1562 }
1563
is_kprobe_multi(const struct bpf_prog * prog)1564 static bool is_kprobe_multi(const struct bpf_prog *prog)
1565 {
1566 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ||
1567 prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1568 }
1569
is_kprobe_session(const struct bpf_prog * prog)1570 static inline bool is_kprobe_session(const struct bpf_prog *prog)
1571 {
1572 return prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1573 }
1574
is_uprobe_multi(const struct bpf_prog * prog)1575 static inline bool is_uprobe_multi(const struct bpf_prog *prog)
1576 {
1577 return prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI ||
1578 prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION;
1579 }
1580
is_uprobe_session(const struct bpf_prog * prog)1581 static inline bool is_uprobe_session(const struct bpf_prog *prog)
1582 {
1583 return prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION;
1584 }
1585
1586 static const struct bpf_func_proto *
kprobe_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1587 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1588 {
1589 switch (func_id) {
1590 case BPF_FUNC_perf_event_output:
1591 return &bpf_perf_event_output_proto;
1592 case BPF_FUNC_get_stackid:
1593 return &bpf_get_stackid_proto;
1594 case BPF_FUNC_get_stack:
1595 return prog->sleepable ? &bpf_get_stack_sleepable_proto : &bpf_get_stack_proto;
1596 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1597 case BPF_FUNC_override_return:
1598 return &bpf_override_return_proto;
1599 #endif
1600 case BPF_FUNC_get_func_ip:
1601 if (is_kprobe_multi(prog))
1602 return &bpf_get_func_ip_proto_kprobe_multi;
1603 if (is_uprobe_multi(prog))
1604 return &bpf_get_func_ip_proto_uprobe_multi;
1605 return &bpf_get_func_ip_proto_kprobe;
1606 case BPF_FUNC_get_attach_cookie:
1607 if (is_kprobe_multi(prog))
1608 return &bpf_get_attach_cookie_proto_kmulti;
1609 if (is_uprobe_multi(prog))
1610 return &bpf_get_attach_cookie_proto_umulti;
1611 return &bpf_get_attach_cookie_proto_trace;
1612 default:
1613 return bpf_tracing_func_proto(func_id, prog);
1614 }
1615 }
1616
1617 /* 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)1618 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1619 const struct bpf_prog *prog,
1620 struct bpf_insn_access_aux *info)
1621 {
1622 if (off < 0 || off >= sizeof(struct pt_regs))
1623 return false;
1624 if (type != BPF_READ)
1625 return false;
1626 if (off % size != 0)
1627 return false;
1628 /*
1629 * Assertion for 32 bit to make sure last 8 byte access
1630 * (BPF_DW) to the last 4 byte member is disallowed.
1631 */
1632 if (off + size > sizeof(struct pt_regs))
1633 return false;
1634
1635 return true;
1636 }
1637
1638 const struct bpf_verifier_ops kprobe_verifier_ops = {
1639 .get_func_proto = kprobe_prog_func_proto,
1640 .is_valid_access = kprobe_prog_is_valid_access,
1641 };
1642
1643 const struct bpf_prog_ops kprobe_prog_ops = {
1644 };
1645
BPF_CALL_5(bpf_perf_event_output_tp,void *,tp_buff,struct bpf_map *,map,u64,flags,void *,data,u64,size)1646 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1647 u64, flags, void *, data, u64, size)
1648 {
1649 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1650
1651 /*
1652 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1653 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1654 * from there and call the same bpf_perf_event_output() helper inline.
1655 */
1656 return ____bpf_perf_event_output(regs, map, flags, data, size);
1657 }
1658
1659 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1660 .func = bpf_perf_event_output_tp,
1661 .gpl_only = true,
1662 .ret_type = RET_INTEGER,
1663 .arg1_type = ARG_PTR_TO_CTX,
1664 .arg2_type = ARG_CONST_MAP_PTR,
1665 .arg3_type = ARG_ANYTHING,
1666 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1667 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1668 };
1669
BPF_CALL_3(bpf_get_stackid_tp,void *,tp_buff,struct bpf_map *,map,u64,flags)1670 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1671 u64, flags)
1672 {
1673 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1674
1675 /*
1676 * Same comment as in bpf_perf_event_output_tp(), only that this time
1677 * the other helper's function body cannot be inlined due to being
1678 * external, thus we need to call raw helper function.
1679 */
1680 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1681 flags, 0, 0);
1682 }
1683
1684 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1685 .func = bpf_get_stackid_tp,
1686 .gpl_only = true,
1687 .ret_type = RET_INTEGER,
1688 .arg1_type = ARG_PTR_TO_CTX,
1689 .arg2_type = ARG_CONST_MAP_PTR,
1690 .arg3_type = ARG_ANYTHING,
1691 };
1692
BPF_CALL_4(bpf_get_stack_tp,void *,tp_buff,void *,buf,u32,size,u64,flags)1693 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1694 u64, flags)
1695 {
1696 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1697
1698 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1699 (unsigned long) size, flags, 0);
1700 }
1701
1702 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1703 .func = bpf_get_stack_tp,
1704 .gpl_only = true,
1705 .ret_type = RET_INTEGER,
1706 .arg1_type = ARG_PTR_TO_CTX,
1707 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1708 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1709 .arg4_type = ARG_ANYTHING,
1710 };
1711
1712 static const struct bpf_func_proto *
tp_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1713 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1714 {
1715 switch (func_id) {
1716 case BPF_FUNC_perf_event_output:
1717 return &bpf_perf_event_output_proto_tp;
1718 case BPF_FUNC_get_stackid:
1719 return &bpf_get_stackid_proto_tp;
1720 case BPF_FUNC_get_stack:
1721 return &bpf_get_stack_proto_tp;
1722 case BPF_FUNC_get_attach_cookie:
1723 return &bpf_get_attach_cookie_proto_trace;
1724 default:
1725 return bpf_tracing_func_proto(func_id, prog);
1726 }
1727 }
1728
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)1729 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1730 const struct bpf_prog *prog,
1731 struct bpf_insn_access_aux *info)
1732 {
1733 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1734 return false;
1735 if (type != BPF_READ)
1736 return false;
1737 if (off % size != 0)
1738 return false;
1739
1740 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1741 return true;
1742 }
1743
1744 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1745 .get_func_proto = tp_prog_func_proto,
1746 .is_valid_access = tp_prog_is_valid_access,
1747 };
1748
1749 const struct bpf_prog_ops tracepoint_prog_ops = {
1750 };
1751
BPF_CALL_3(bpf_perf_prog_read_value,struct bpf_perf_event_data_kern *,ctx,struct bpf_perf_event_value *,buf,u32,size)1752 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1753 struct bpf_perf_event_value *, buf, u32, size)
1754 {
1755 int err = -EINVAL;
1756
1757 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1758 goto clear;
1759 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1760 &buf->running);
1761 if (unlikely(err))
1762 goto clear;
1763 return 0;
1764 clear:
1765 memset(buf, 0, size);
1766 return err;
1767 }
1768
1769 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1770 .func = bpf_perf_prog_read_value,
1771 .gpl_only = true,
1772 .ret_type = RET_INTEGER,
1773 .arg1_type = ARG_PTR_TO_CTX,
1774 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1775 .arg3_type = ARG_CONST_SIZE,
1776 };
1777
BPF_CALL_4(bpf_read_branch_records,struct bpf_perf_event_data_kern *,ctx,void *,buf,u32,size,u64,flags)1778 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1779 void *, buf, u32, size, u64, flags)
1780 {
1781 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1782 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1783 u32 to_copy;
1784
1785 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1786 return -EINVAL;
1787
1788 if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1789 return -ENOENT;
1790
1791 if (unlikely(!br_stack))
1792 return -ENOENT;
1793
1794 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1795 return br_stack->nr * br_entry_size;
1796
1797 if (!buf || (size % br_entry_size != 0))
1798 return -EINVAL;
1799
1800 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1801 memcpy(buf, br_stack->entries, to_copy);
1802
1803 return to_copy;
1804 }
1805
1806 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1807 .func = bpf_read_branch_records,
1808 .gpl_only = true,
1809 .ret_type = RET_INTEGER,
1810 .arg1_type = ARG_PTR_TO_CTX,
1811 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1812 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1813 .arg4_type = ARG_ANYTHING,
1814 };
1815
1816 static const struct bpf_func_proto *
pe_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1817 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1818 {
1819 switch (func_id) {
1820 case BPF_FUNC_perf_event_output:
1821 return &bpf_perf_event_output_proto_tp;
1822 case BPF_FUNC_get_stackid:
1823 return &bpf_get_stackid_proto_pe;
1824 case BPF_FUNC_get_stack:
1825 return &bpf_get_stack_proto_pe;
1826 case BPF_FUNC_perf_prog_read_value:
1827 return &bpf_perf_prog_read_value_proto;
1828 case BPF_FUNC_read_branch_records:
1829 return &bpf_read_branch_records_proto;
1830 case BPF_FUNC_get_attach_cookie:
1831 return &bpf_get_attach_cookie_proto_pe;
1832 default:
1833 return bpf_tracing_func_proto(func_id, prog);
1834 }
1835 }
1836
1837 /*
1838 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1839 * to avoid potential recursive reuse issue when/if tracepoints are added
1840 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1841 *
1842 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1843 * in normal, irq, and nmi context.
1844 */
1845 struct bpf_raw_tp_regs {
1846 struct pt_regs regs[3];
1847 };
1848 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1849 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
get_bpf_raw_tp_regs(void)1850 static struct pt_regs *get_bpf_raw_tp_regs(void)
1851 {
1852 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1853 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1854
1855 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1856 this_cpu_dec(bpf_raw_tp_nest_level);
1857 return ERR_PTR(-EBUSY);
1858 }
1859
1860 return &tp_regs->regs[nest_level - 1];
1861 }
1862
put_bpf_raw_tp_regs(void)1863 static void put_bpf_raw_tp_regs(void)
1864 {
1865 this_cpu_dec(bpf_raw_tp_nest_level);
1866 }
1867
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)1868 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1869 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1870 {
1871 struct pt_regs *regs = get_bpf_raw_tp_regs();
1872 int ret;
1873
1874 if (IS_ERR(regs))
1875 return PTR_ERR(regs);
1876
1877 perf_fetch_caller_regs(regs);
1878 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1879
1880 put_bpf_raw_tp_regs();
1881 return ret;
1882 }
1883
1884 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1885 .func = bpf_perf_event_output_raw_tp,
1886 .gpl_only = true,
1887 .ret_type = RET_INTEGER,
1888 .arg1_type = ARG_PTR_TO_CTX,
1889 .arg2_type = ARG_CONST_MAP_PTR,
1890 .arg3_type = ARG_ANYTHING,
1891 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1892 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1893 };
1894
1895 extern const struct bpf_func_proto bpf_skb_output_proto;
1896 extern const struct bpf_func_proto bpf_xdp_output_proto;
1897 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1898
BPF_CALL_3(bpf_get_stackid_raw_tp,struct bpf_raw_tracepoint_args *,args,struct bpf_map *,map,u64,flags)1899 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1900 struct bpf_map *, map, u64, flags)
1901 {
1902 struct pt_regs *regs = get_bpf_raw_tp_regs();
1903 int ret;
1904
1905 if (IS_ERR(regs))
1906 return PTR_ERR(regs);
1907
1908 perf_fetch_caller_regs(regs);
1909 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1910 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1911 flags, 0, 0);
1912 put_bpf_raw_tp_regs();
1913 return ret;
1914 }
1915
1916 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1917 .func = bpf_get_stackid_raw_tp,
1918 .gpl_only = true,
1919 .ret_type = RET_INTEGER,
1920 .arg1_type = ARG_PTR_TO_CTX,
1921 .arg2_type = ARG_CONST_MAP_PTR,
1922 .arg3_type = ARG_ANYTHING,
1923 };
1924
BPF_CALL_4(bpf_get_stack_raw_tp,struct bpf_raw_tracepoint_args *,args,void *,buf,u32,size,u64,flags)1925 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1926 void *, buf, u32, size, u64, flags)
1927 {
1928 struct pt_regs *regs = get_bpf_raw_tp_regs();
1929 int ret;
1930
1931 if (IS_ERR(regs))
1932 return PTR_ERR(regs);
1933
1934 perf_fetch_caller_regs(regs);
1935 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1936 (unsigned long) size, flags, 0);
1937 put_bpf_raw_tp_regs();
1938 return ret;
1939 }
1940
1941 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1942 .func = bpf_get_stack_raw_tp,
1943 .gpl_only = true,
1944 .ret_type = RET_INTEGER,
1945 .arg1_type = ARG_PTR_TO_CTX,
1946 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1947 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1948 .arg4_type = ARG_ANYTHING,
1949 };
1950
1951 static const struct bpf_func_proto *
raw_tp_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1952 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1953 {
1954 switch (func_id) {
1955 case BPF_FUNC_perf_event_output:
1956 return &bpf_perf_event_output_proto_raw_tp;
1957 case BPF_FUNC_get_stackid:
1958 return &bpf_get_stackid_proto_raw_tp;
1959 case BPF_FUNC_get_stack:
1960 return &bpf_get_stack_proto_raw_tp;
1961 case BPF_FUNC_get_attach_cookie:
1962 return &bpf_get_attach_cookie_proto_tracing;
1963 default:
1964 return bpf_tracing_func_proto(func_id, prog);
1965 }
1966 }
1967
1968 const struct bpf_func_proto *
tracing_prog_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)1969 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1970 {
1971 const struct bpf_func_proto *fn;
1972
1973 switch (func_id) {
1974 #ifdef CONFIG_NET
1975 case BPF_FUNC_skb_output:
1976 return &bpf_skb_output_proto;
1977 case BPF_FUNC_xdp_output:
1978 return &bpf_xdp_output_proto;
1979 case BPF_FUNC_skc_to_tcp6_sock:
1980 return &bpf_skc_to_tcp6_sock_proto;
1981 case BPF_FUNC_skc_to_tcp_sock:
1982 return &bpf_skc_to_tcp_sock_proto;
1983 case BPF_FUNC_skc_to_tcp_timewait_sock:
1984 return &bpf_skc_to_tcp_timewait_sock_proto;
1985 case BPF_FUNC_skc_to_tcp_request_sock:
1986 return &bpf_skc_to_tcp_request_sock_proto;
1987 case BPF_FUNC_skc_to_udp6_sock:
1988 return &bpf_skc_to_udp6_sock_proto;
1989 case BPF_FUNC_skc_to_unix_sock:
1990 return &bpf_skc_to_unix_sock_proto;
1991 case BPF_FUNC_skc_to_mptcp_sock:
1992 return &bpf_skc_to_mptcp_sock_proto;
1993 case BPF_FUNC_sk_storage_get:
1994 return &bpf_sk_storage_get_tracing_proto;
1995 case BPF_FUNC_sk_storage_delete:
1996 return &bpf_sk_storage_delete_tracing_proto;
1997 case BPF_FUNC_sock_from_file:
1998 return &bpf_sock_from_file_proto;
1999 case BPF_FUNC_get_socket_cookie:
2000 return &bpf_get_socket_ptr_cookie_proto;
2001 case BPF_FUNC_xdp_get_buff_len:
2002 return &bpf_xdp_get_buff_len_trace_proto;
2003 #endif
2004 case BPF_FUNC_seq_printf:
2005 return prog->expected_attach_type == BPF_TRACE_ITER ?
2006 &bpf_seq_printf_proto :
2007 NULL;
2008 case BPF_FUNC_seq_write:
2009 return prog->expected_attach_type == BPF_TRACE_ITER ?
2010 &bpf_seq_write_proto :
2011 NULL;
2012 case BPF_FUNC_seq_printf_btf:
2013 return prog->expected_attach_type == BPF_TRACE_ITER ?
2014 &bpf_seq_printf_btf_proto :
2015 NULL;
2016 case BPF_FUNC_d_path:
2017 return &bpf_d_path_proto;
2018 case BPF_FUNC_get_func_arg:
2019 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
2020 case BPF_FUNC_get_func_ret:
2021 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
2022 case BPF_FUNC_get_func_arg_cnt:
2023 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
2024 case BPF_FUNC_get_attach_cookie:
2025 if (prog->type == BPF_PROG_TYPE_TRACING &&
2026 prog->expected_attach_type == BPF_TRACE_RAW_TP)
2027 return &bpf_get_attach_cookie_proto_tracing;
2028 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
2029 default:
2030 fn = raw_tp_prog_func_proto(func_id, prog);
2031 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
2032 fn = bpf_iter_get_func_proto(func_id, prog);
2033 return fn;
2034 }
2035 }
2036
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)2037 static bool raw_tp_prog_is_valid_access(int off, int size,
2038 enum bpf_access_type type,
2039 const struct bpf_prog *prog,
2040 struct bpf_insn_access_aux *info)
2041 {
2042 return bpf_tracing_ctx_access(off, size, type);
2043 }
2044
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)2045 static bool tracing_prog_is_valid_access(int off, int size,
2046 enum bpf_access_type type,
2047 const struct bpf_prog *prog,
2048 struct bpf_insn_access_aux *info)
2049 {
2050 return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
2051 }
2052
bpf_prog_test_run_tracing(struct bpf_prog * prog,const union bpf_attr * kattr,union bpf_attr __user * uattr)2053 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
2054 const union bpf_attr *kattr,
2055 union bpf_attr __user *uattr)
2056 {
2057 return -ENOTSUPP;
2058 }
2059
2060 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
2061 .get_func_proto = raw_tp_prog_func_proto,
2062 .is_valid_access = raw_tp_prog_is_valid_access,
2063 };
2064
2065 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
2066 #ifdef CONFIG_NET
2067 .test_run = bpf_prog_test_run_raw_tp,
2068 #endif
2069 };
2070
2071 const struct bpf_verifier_ops tracing_verifier_ops = {
2072 .get_func_proto = tracing_prog_func_proto,
2073 .is_valid_access = tracing_prog_is_valid_access,
2074 };
2075
2076 const struct bpf_prog_ops tracing_prog_ops = {
2077 .test_run = bpf_prog_test_run_tracing,
2078 };
2079
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)2080 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2081 enum bpf_access_type type,
2082 const struct bpf_prog *prog,
2083 struct bpf_insn_access_aux *info)
2084 {
2085 if (off == 0) {
2086 if (size != sizeof(u64) || type != BPF_READ)
2087 return false;
2088 info->reg_type = PTR_TO_TP_BUFFER;
2089 }
2090 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2091 }
2092
2093 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2094 .get_func_proto = raw_tp_prog_func_proto,
2095 .is_valid_access = raw_tp_writable_prog_is_valid_access,
2096 };
2097
2098 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2099 };
2100
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)2101 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2102 const struct bpf_prog *prog,
2103 struct bpf_insn_access_aux *info)
2104 {
2105 const int size_u64 = sizeof(u64);
2106
2107 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2108 return false;
2109 if (type != BPF_READ)
2110 return false;
2111 if (off % size != 0) {
2112 if (sizeof(unsigned long) != 4)
2113 return false;
2114 if (size != 8)
2115 return false;
2116 if (off % size != 4)
2117 return false;
2118 }
2119
2120 switch (off) {
2121 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2122 bpf_ctx_record_field_size(info, size_u64);
2123 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2124 return false;
2125 break;
2126 case bpf_ctx_range(struct bpf_perf_event_data, addr):
2127 bpf_ctx_record_field_size(info, size_u64);
2128 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2129 return false;
2130 break;
2131 default:
2132 if (size != sizeof(long))
2133 return false;
2134 }
2135
2136 return true;
2137 }
2138
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)2139 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2140 const struct bpf_insn *si,
2141 struct bpf_insn *insn_buf,
2142 struct bpf_prog *prog, u32 *target_size)
2143 {
2144 struct bpf_insn *insn = insn_buf;
2145
2146 switch (si->off) {
2147 case offsetof(struct bpf_perf_event_data, sample_period):
2148 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2149 data), si->dst_reg, si->src_reg,
2150 offsetof(struct bpf_perf_event_data_kern, data));
2151 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2152 bpf_target_off(struct perf_sample_data, period, 8,
2153 target_size));
2154 break;
2155 case offsetof(struct bpf_perf_event_data, addr):
2156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2157 data), si->dst_reg, si->src_reg,
2158 offsetof(struct bpf_perf_event_data_kern, data));
2159 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2160 bpf_target_off(struct perf_sample_data, addr, 8,
2161 target_size));
2162 break;
2163 default:
2164 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2165 regs), si->dst_reg, si->src_reg,
2166 offsetof(struct bpf_perf_event_data_kern, regs));
2167 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2168 si->off);
2169 break;
2170 }
2171
2172 return insn - insn_buf;
2173 }
2174
2175 const struct bpf_verifier_ops perf_event_verifier_ops = {
2176 .get_func_proto = pe_prog_func_proto,
2177 .is_valid_access = pe_prog_is_valid_access,
2178 .convert_ctx_access = pe_prog_convert_ctx_access,
2179 };
2180
2181 const struct bpf_prog_ops perf_event_prog_ops = {
2182 };
2183
2184 static DEFINE_MUTEX(bpf_event_mutex);
2185
2186 #define BPF_TRACE_MAX_PROGS 64
2187
perf_event_attach_bpf_prog(struct perf_event * event,struct bpf_prog * prog,u64 bpf_cookie)2188 int perf_event_attach_bpf_prog(struct perf_event *event,
2189 struct bpf_prog *prog,
2190 u64 bpf_cookie)
2191 {
2192 struct bpf_prog_array *old_array;
2193 struct bpf_prog_array *new_array;
2194 int ret = -EEXIST;
2195
2196 /*
2197 * Kprobe override only works if they are on the function entry,
2198 * and only if they are on the opt-in list.
2199 */
2200 if (prog->kprobe_override &&
2201 (!trace_kprobe_on_func_entry(event->tp_event) ||
2202 !trace_kprobe_error_injectable(event->tp_event)))
2203 return -EINVAL;
2204
2205 mutex_lock(&bpf_event_mutex);
2206
2207 if (event->prog)
2208 goto unlock;
2209
2210 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2211 if (old_array &&
2212 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2213 ret = -E2BIG;
2214 goto unlock;
2215 }
2216
2217 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2218 if (ret < 0)
2219 goto unlock;
2220
2221 /* set the new array to event->tp_event and set event->prog */
2222 event->prog = prog;
2223 event->bpf_cookie = bpf_cookie;
2224 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2225 bpf_prog_array_free_sleepable(old_array);
2226
2227 unlock:
2228 mutex_unlock(&bpf_event_mutex);
2229 return ret;
2230 }
2231
perf_event_detach_bpf_prog(struct perf_event * event)2232 void perf_event_detach_bpf_prog(struct perf_event *event)
2233 {
2234 struct bpf_prog_array *old_array;
2235 struct bpf_prog_array *new_array;
2236 struct bpf_prog *prog = NULL;
2237 int ret;
2238
2239 mutex_lock(&bpf_event_mutex);
2240
2241 if (!event->prog)
2242 goto unlock;
2243
2244 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2245 if (!old_array)
2246 goto put;
2247
2248 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2249 if (ret < 0) {
2250 bpf_prog_array_delete_safe(old_array, event->prog);
2251 } else {
2252 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2253 bpf_prog_array_free_sleepable(old_array);
2254 }
2255
2256 put:
2257 prog = event->prog;
2258 event->prog = NULL;
2259
2260 unlock:
2261 mutex_unlock(&bpf_event_mutex);
2262
2263 if (prog) {
2264 /*
2265 * It could be that the bpf_prog is not sleepable (and will be freed
2266 * via normal RCU), but is called from a point that supports sleepable
2267 * programs and uses tasks-trace-RCU.
2268 */
2269 synchronize_rcu_tasks_trace();
2270
2271 bpf_prog_put(prog);
2272 }
2273 }
2274
perf_event_query_prog_array(struct perf_event * event,void __user * info)2275 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2276 {
2277 struct perf_event_query_bpf __user *uquery = info;
2278 struct perf_event_query_bpf query = {};
2279 struct bpf_prog_array *progs;
2280 u32 *ids, prog_cnt, ids_len;
2281 int ret;
2282
2283 if (!perfmon_capable())
2284 return -EPERM;
2285 if (event->attr.type != PERF_TYPE_TRACEPOINT)
2286 return -EINVAL;
2287 if (copy_from_user(&query, uquery, sizeof(query)))
2288 return -EFAULT;
2289
2290 ids_len = query.ids_len;
2291 if (ids_len > BPF_TRACE_MAX_PROGS)
2292 return -E2BIG;
2293 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2294 if (!ids)
2295 return -ENOMEM;
2296 /*
2297 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2298 * is required when user only wants to check for uquery->prog_cnt.
2299 * There is no need to check for it since the case is handled
2300 * gracefully in bpf_prog_array_copy_info.
2301 */
2302
2303 mutex_lock(&bpf_event_mutex);
2304 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2305 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2306 mutex_unlock(&bpf_event_mutex);
2307
2308 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2309 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2310 ret = -EFAULT;
2311
2312 kfree(ids);
2313 return ret;
2314 }
2315
2316 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2317 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2318
bpf_get_raw_tracepoint(const char * name)2319 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2320 {
2321 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2322
2323 for (; btp < __stop__bpf_raw_tp; btp++) {
2324 if (!strcmp(btp->tp->name, name))
2325 return btp;
2326 }
2327
2328 return bpf_get_raw_tracepoint_module(name);
2329 }
2330
bpf_put_raw_tracepoint(struct bpf_raw_event_map * btp)2331 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2332 {
2333 struct module *mod;
2334
2335 preempt_disable();
2336 mod = __module_address((unsigned long)btp);
2337 module_put(mod);
2338 preempt_enable();
2339 }
2340
2341 static __always_inline
__bpf_trace_run(struct bpf_raw_tp_link * link,u64 * args)2342 void __bpf_trace_run(struct bpf_raw_tp_link *link, u64 *args)
2343 {
2344 struct bpf_prog *prog = link->link.prog;
2345 struct bpf_run_ctx *old_run_ctx;
2346 struct bpf_trace_run_ctx run_ctx;
2347
2348 cant_sleep();
2349 if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2350 bpf_prog_inc_misses_counter(prog);
2351 goto out;
2352 }
2353
2354 run_ctx.bpf_cookie = link->cookie;
2355 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2356
2357 rcu_read_lock();
2358 (void) bpf_prog_run(prog, args);
2359 rcu_read_unlock();
2360
2361 bpf_reset_run_ctx(old_run_ctx);
2362 out:
2363 this_cpu_dec(*(prog->active));
2364 }
2365
2366 #define UNPACK(...) __VA_ARGS__
2367 #define REPEAT_1(FN, DL, X, ...) FN(X)
2368 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2369 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2370 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2371 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2372 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2373 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2374 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2375 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2376 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2377 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2378 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2379 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2380
2381 #define SARG(X) u64 arg##X
2382 #define COPY(X) args[X] = arg##X
2383
2384 #define __DL_COM (,)
2385 #define __DL_SEM (;)
2386
2387 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2388
2389 #define BPF_TRACE_DEFN_x(x) \
2390 void bpf_trace_run##x(struct bpf_raw_tp_link *link, \
2391 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2392 { \
2393 u64 args[x]; \
2394 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2395 __bpf_trace_run(link, args); \
2396 } \
2397 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2398 BPF_TRACE_DEFN_x(1);
2399 BPF_TRACE_DEFN_x(2);
2400 BPF_TRACE_DEFN_x(3);
2401 BPF_TRACE_DEFN_x(4);
2402 BPF_TRACE_DEFN_x(5);
2403 BPF_TRACE_DEFN_x(6);
2404 BPF_TRACE_DEFN_x(7);
2405 BPF_TRACE_DEFN_x(8);
2406 BPF_TRACE_DEFN_x(9);
2407 BPF_TRACE_DEFN_x(10);
2408 BPF_TRACE_DEFN_x(11);
2409 BPF_TRACE_DEFN_x(12);
2410
bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_raw_tp_link * link)2411 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2412 {
2413 struct tracepoint *tp = btp->tp;
2414 struct bpf_prog *prog = link->link.prog;
2415
2416 /*
2417 * check that program doesn't access arguments beyond what's
2418 * available in this tracepoint
2419 */
2420 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2421 return -EINVAL;
2422
2423 if (prog->aux->max_tp_access > btp->writable_size)
2424 return -EINVAL;
2425
2426 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func, link);
2427 }
2428
bpf_probe_unregister(struct bpf_raw_event_map * btp,struct bpf_raw_tp_link * link)2429 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2430 {
2431 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, link);
2432 }
2433
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,unsigned long * missed)2434 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2435 u32 *fd_type, const char **buf,
2436 u64 *probe_offset, u64 *probe_addr,
2437 unsigned long *missed)
2438 {
2439 bool is_tracepoint, is_syscall_tp;
2440 struct bpf_prog *prog;
2441 int flags, err = 0;
2442
2443 prog = event->prog;
2444 if (!prog)
2445 return -ENOENT;
2446
2447 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2448 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2449 return -EOPNOTSUPP;
2450
2451 *prog_id = prog->aux->id;
2452 flags = event->tp_event->flags;
2453 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2454 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2455
2456 if (is_tracepoint || is_syscall_tp) {
2457 *buf = is_tracepoint ? event->tp_event->tp->name
2458 : event->tp_event->name;
2459 /* We allow NULL pointer for tracepoint */
2460 if (fd_type)
2461 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2462 if (probe_offset)
2463 *probe_offset = 0x0;
2464 if (probe_addr)
2465 *probe_addr = 0x0;
2466 } else {
2467 /* kprobe/uprobe */
2468 err = -EOPNOTSUPP;
2469 #ifdef CONFIG_KPROBE_EVENTS
2470 if (flags & TRACE_EVENT_FL_KPROBE)
2471 err = bpf_get_kprobe_info(event, fd_type, buf,
2472 probe_offset, probe_addr, missed,
2473 event->attr.type == PERF_TYPE_TRACEPOINT);
2474 #endif
2475 #ifdef CONFIG_UPROBE_EVENTS
2476 if (flags & TRACE_EVENT_FL_UPROBE)
2477 err = bpf_get_uprobe_info(event, fd_type, buf,
2478 probe_offset, probe_addr,
2479 event->attr.type == PERF_TYPE_TRACEPOINT);
2480 #endif
2481 }
2482
2483 return err;
2484 }
2485
send_signal_irq_work_init(void)2486 static int __init send_signal_irq_work_init(void)
2487 {
2488 int cpu;
2489 struct send_signal_irq_work *work;
2490
2491 for_each_possible_cpu(cpu) {
2492 work = per_cpu_ptr(&send_signal_work, cpu);
2493 init_irq_work(&work->irq_work, do_bpf_send_signal);
2494 }
2495 return 0;
2496 }
2497
2498 subsys_initcall(send_signal_irq_work_init);
2499
2500 #ifdef CONFIG_MODULES
bpf_event_notify(struct notifier_block * nb,unsigned long op,void * module)2501 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2502 void *module)
2503 {
2504 struct bpf_trace_module *btm, *tmp;
2505 struct module *mod = module;
2506 int ret = 0;
2507
2508 if (mod->num_bpf_raw_events == 0 ||
2509 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2510 goto out;
2511
2512 mutex_lock(&bpf_module_mutex);
2513
2514 switch (op) {
2515 case MODULE_STATE_COMING:
2516 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2517 if (btm) {
2518 btm->module = module;
2519 list_add(&btm->list, &bpf_trace_modules);
2520 } else {
2521 ret = -ENOMEM;
2522 }
2523 break;
2524 case MODULE_STATE_GOING:
2525 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2526 if (btm->module == module) {
2527 list_del(&btm->list);
2528 kfree(btm);
2529 break;
2530 }
2531 }
2532 break;
2533 }
2534
2535 mutex_unlock(&bpf_module_mutex);
2536
2537 out:
2538 return notifier_from_errno(ret);
2539 }
2540
2541 static struct notifier_block bpf_module_nb = {
2542 .notifier_call = bpf_event_notify,
2543 };
2544
bpf_event_init(void)2545 static int __init bpf_event_init(void)
2546 {
2547 register_module_notifier(&bpf_module_nb);
2548 return 0;
2549 }
2550
2551 fs_initcall(bpf_event_init);
2552 #endif /* CONFIG_MODULES */
2553
2554 struct bpf_session_run_ctx {
2555 struct bpf_run_ctx run_ctx;
2556 bool is_return;
2557 void *data;
2558 };
2559
2560 #ifdef CONFIG_FPROBE
2561 struct bpf_kprobe_multi_link {
2562 struct bpf_link link;
2563 struct fprobe fp;
2564 unsigned long *addrs;
2565 u64 *cookies;
2566 u32 cnt;
2567 u32 mods_cnt;
2568 struct module **mods;
2569 u32 flags;
2570 };
2571
2572 struct bpf_kprobe_multi_run_ctx {
2573 struct bpf_session_run_ctx session_ctx;
2574 struct bpf_kprobe_multi_link *link;
2575 unsigned long entry_ip;
2576 };
2577
2578 struct user_syms {
2579 const char **syms;
2580 char *buf;
2581 };
2582
2583 #ifndef CONFIG_HAVE_FTRACE_REGS_HAVING_PT_REGS
2584 static DEFINE_PER_CPU(struct pt_regs, bpf_kprobe_multi_pt_regs);
2585 #define bpf_kprobe_multi_pt_regs_ptr() this_cpu_ptr(&bpf_kprobe_multi_pt_regs)
2586 #else
2587 #define bpf_kprobe_multi_pt_regs_ptr() (NULL)
2588 #endif
2589
ftrace_get_entry_ip(unsigned long fentry_ip)2590 static unsigned long ftrace_get_entry_ip(unsigned long fentry_ip)
2591 {
2592 unsigned long ip = ftrace_get_symaddr(fentry_ip);
2593
2594 return ip ? : fentry_ip;
2595 }
2596
copy_user_syms(struct user_syms * us,unsigned long __user * usyms,u32 cnt)2597 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2598 {
2599 unsigned long __user usymbol;
2600 const char **syms = NULL;
2601 char *buf = NULL, *p;
2602 int err = -ENOMEM;
2603 unsigned int i;
2604
2605 syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2606 if (!syms)
2607 goto error;
2608
2609 buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2610 if (!buf)
2611 goto error;
2612
2613 for (p = buf, i = 0; i < cnt; i++) {
2614 if (__get_user(usymbol, usyms + i)) {
2615 err = -EFAULT;
2616 goto error;
2617 }
2618 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2619 if (err == KSYM_NAME_LEN)
2620 err = -E2BIG;
2621 if (err < 0)
2622 goto error;
2623 syms[i] = p;
2624 p += err + 1;
2625 }
2626
2627 us->syms = syms;
2628 us->buf = buf;
2629 return 0;
2630
2631 error:
2632 if (err) {
2633 kvfree(syms);
2634 kvfree(buf);
2635 }
2636 return err;
2637 }
2638
kprobe_multi_put_modules(struct module ** mods,u32 cnt)2639 static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2640 {
2641 u32 i;
2642
2643 for (i = 0; i < cnt; i++)
2644 module_put(mods[i]);
2645 }
2646
free_user_syms(struct user_syms * us)2647 static void free_user_syms(struct user_syms *us)
2648 {
2649 kvfree(us->syms);
2650 kvfree(us->buf);
2651 }
2652
bpf_kprobe_multi_link_release(struct bpf_link * link)2653 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2654 {
2655 struct bpf_kprobe_multi_link *kmulti_link;
2656
2657 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2658 unregister_fprobe(&kmulti_link->fp);
2659 kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2660 }
2661
bpf_kprobe_multi_link_dealloc(struct bpf_link * link)2662 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2663 {
2664 struct bpf_kprobe_multi_link *kmulti_link;
2665
2666 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2667 kvfree(kmulti_link->addrs);
2668 kvfree(kmulti_link->cookies);
2669 kfree(kmulti_link->mods);
2670 kfree(kmulti_link);
2671 }
2672
bpf_kprobe_multi_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)2673 static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
2674 struct bpf_link_info *info)
2675 {
2676 u64 __user *ucookies = u64_to_user_ptr(info->kprobe_multi.cookies);
2677 u64 __user *uaddrs = u64_to_user_ptr(info->kprobe_multi.addrs);
2678 struct bpf_kprobe_multi_link *kmulti_link;
2679 u32 ucount = info->kprobe_multi.count;
2680 int err = 0, i;
2681
2682 if (!uaddrs ^ !ucount)
2683 return -EINVAL;
2684 if (ucookies && !ucount)
2685 return -EINVAL;
2686
2687 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2688 info->kprobe_multi.count = kmulti_link->cnt;
2689 info->kprobe_multi.flags = kmulti_link->flags;
2690 info->kprobe_multi.missed = kmulti_link->fp.nmissed;
2691
2692 if (!uaddrs)
2693 return 0;
2694 if (ucount < kmulti_link->cnt)
2695 err = -ENOSPC;
2696 else
2697 ucount = kmulti_link->cnt;
2698
2699 if (ucookies) {
2700 if (kmulti_link->cookies) {
2701 if (copy_to_user(ucookies, kmulti_link->cookies, ucount * sizeof(u64)))
2702 return -EFAULT;
2703 } else {
2704 for (i = 0; i < ucount; i++) {
2705 if (put_user(0, ucookies + i))
2706 return -EFAULT;
2707 }
2708 }
2709 }
2710
2711 if (kallsyms_show_value(current_cred())) {
2712 if (copy_to_user(uaddrs, kmulti_link->addrs, ucount * sizeof(u64)))
2713 return -EFAULT;
2714 } else {
2715 for (i = 0; i < ucount; i++) {
2716 if (put_user(0, uaddrs + i))
2717 return -EFAULT;
2718 }
2719 }
2720 return err;
2721 }
2722
2723 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2724 .release = bpf_kprobe_multi_link_release,
2725 .dealloc_deferred = bpf_kprobe_multi_link_dealloc,
2726 .fill_link_info = bpf_kprobe_multi_link_fill_link_info,
2727 };
2728
bpf_kprobe_multi_cookie_swap(void * a,void * b,int size,const void * priv)2729 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2730 {
2731 const struct bpf_kprobe_multi_link *link = priv;
2732 unsigned long *addr_a = a, *addr_b = b;
2733 u64 *cookie_a, *cookie_b;
2734
2735 cookie_a = link->cookies + (addr_a - link->addrs);
2736 cookie_b = link->cookies + (addr_b - link->addrs);
2737
2738 /* swap addr_a/addr_b and cookie_a/cookie_b values */
2739 swap(*addr_a, *addr_b);
2740 swap(*cookie_a, *cookie_b);
2741 }
2742
bpf_kprobe_multi_addrs_cmp(const void * a,const void * b)2743 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2744 {
2745 const unsigned long *addr_a = a, *addr_b = b;
2746
2747 if (*addr_a == *addr_b)
2748 return 0;
2749 return *addr_a < *addr_b ? -1 : 1;
2750 }
2751
bpf_kprobe_multi_cookie_cmp(const void * a,const void * b,const void * priv)2752 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2753 {
2754 return bpf_kprobe_multi_addrs_cmp(a, b);
2755 }
2756
bpf_kprobe_multi_cookie(struct bpf_run_ctx * ctx)2757 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2758 {
2759 struct bpf_kprobe_multi_run_ctx *run_ctx;
2760 struct bpf_kprobe_multi_link *link;
2761 u64 *cookie, entry_ip;
2762 unsigned long *addr;
2763
2764 if (WARN_ON_ONCE(!ctx))
2765 return 0;
2766 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2767 session_ctx.run_ctx);
2768 link = run_ctx->link;
2769 if (!link->cookies)
2770 return 0;
2771 entry_ip = run_ctx->entry_ip;
2772 addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2773 bpf_kprobe_multi_addrs_cmp);
2774 if (!addr)
2775 return 0;
2776 cookie = link->cookies + (addr - link->addrs);
2777 return *cookie;
2778 }
2779
bpf_kprobe_multi_entry_ip(struct bpf_run_ctx * ctx)2780 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2781 {
2782 struct bpf_kprobe_multi_run_ctx *run_ctx;
2783
2784 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2785 session_ctx.run_ctx);
2786 return run_ctx->entry_ip;
2787 }
2788
2789 static int
kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link * link,unsigned long entry_ip,struct ftrace_regs * fregs,bool is_return,void * data)2790 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2791 unsigned long entry_ip, struct ftrace_regs *fregs,
2792 bool is_return, void *data)
2793 {
2794 struct bpf_kprobe_multi_run_ctx run_ctx = {
2795 .session_ctx = {
2796 .is_return = is_return,
2797 .data = data,
2798 },
2799 .link = link,
2800 .entry_ip = entry_ip,
2801 };
2802 struct bpf_run_ctx *old_run_ctx;
2803 struct pt_regs *regs;
2804 int err;
2805
2806 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2807 bpf_prog_inc_misses_counter(link->link.prog);
2808 err = 1;
2809 goto out;
2810 }
2811
2812 migrate_disable();
2813 rcu_read_lock();
2814 regs = ftrace_partial_regs(fregs, bpf_kprobe_multi_pt_regs_ptr());
2815 old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
2816 err = bpf_prog_run(link->link.prog, regs);
2817 bpf_reset_run_ctx(old_run_ctx);
2818 rcu_read_unlock();
2819 migrate_enable();
2820
2821 out:
2822 __this_cpu_dec(bpf_prog_active);
2823 return err;
2824 }
2825
2826 static int
kprobe_multi_link_handler(struct fprobe * fp,unsigned long fentry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * data)2827 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2828 unsigned long ret_ip, struct ftrace_regs *fregs,
2829 void *data)
2830 {
2831 struct bpf_kprobe_multi_link *link;
2832 int err;
2833
2834 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2835 err = kprobe_multi_link_prog_run(link, ftrace_get_entry_ip(fentry_ip),
2836 fregs, false, data);
2837 return is_kprobe_session(link->link.prog) ? err : 0;
2838 }
2839
2840 static void
kprobe_multi_link_exit_handler(struct fprobe * fp,unsigned long fentry_ip,unsigned long ret_ip,struct ftrace_regs * fregs,void * data)2841 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip,
2842 unsigned long ret_ip, struct ftrace_regs *fregs,
2843 void *data)
2844 {
2845 struct bpf_kprobe_multi_link *link;
2846
2847 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2848 kprobe_multi_link_prog_run(link, ftrace_get_entry_ip(fentry_ip),
2849 fregs, true, data);
2850 }
2851
symbols_cmp_r(const void * a,const void * b,const void * priv)2852 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2853 {
2854 const char **str_a = (const char **) a;
2855 const char **str_b = (const char **) b;
2856
2857 return strcmp(*str_a, *str_b);
2858 }
2859
2860 struct multi_symbols_sort {
2861 const char **funcs;
2862 u64 *cookies;
2863 };
2864
symbols_swap_r(void * a,void * b,int size,const void * priv)2865 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2866 {
2867 const struct multi_symbols_sort *data = priv;
2868 const char **name_a = a, **name_b = b;
2869
2870 swap(*name_a, *name_b);
2871
2872 /* If defined, swap also related cookies. */
2873 if (data->cookies) {
2874 u64 *cookie_a, *cookie_b;
2875
2876 cookie_a = data->cookies + (name_a - data->funcs);
2877 cookie_b = data->cookies + (name_b - data->funcs);
2878 swap(*cookie_a, *cookie_b);
2879 }
2880 }
2881
2882 struct modules_array {
2883 struct module **mods;
2884 int mods_cnt;
2885 int mods_cap;
2886 };
2887
add_module(struct modules_array * arr,struct module * mod)2888 static int add_module(struct modules_array *arr, struct module *mod)
2889 {
2890 struct module **mods;
2891
2892 if (arr->mods_cnt == arr->mods_cap) {
2893 arr->mods_cap = max(16, arr->mods_cap * 3 / 2);
2894 mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL);
2895 if (!mods)
2896 return -ENOMEM;
2897 arr->mods = mods;
2898 }
2899
2900 arr->mods[arr->mods_cnt] = mod;
2901 arr->mods_cnt++;
2902 return 0;
2903 }
2904
has_module(struct modules_array * arr,struct module * mod)2905 static bool has_module(struct modules_array *arr, struct module *mod)
2906 {
2907 int i;
2908
2909 for (i = arr->mods_cnt - 1; i >= 0; i--) {
2910 if (arr->mods[i] == mod)
2911 return true;
2912 }
2913 return false;
2914 }
2915
get_modules_for_addrs(struct module *** mods,unsigned long * addrs,u32 addrs_cnt)2916 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2917 {
2918 struct modules_array arr = {};
2919 u32 i, err = 0;
2920
2921 for (i = 0; i < addrs_cnt; i++) {
2922 struct module *mod;
2923
2924 preempt_disable();
2925 mod = __module_address(addrs[i]);
2926 /* Either no module or we it's already stored */
2927 if (!mod || has_module(&arr, mod)) {
2928 preempt_enable();
2929 continue;
2930 }
2931 if (!try_module_get(mod))
2932 err = -EINVAL;
2933 preempt_enable();
2934 if (err)
2935 break;
2936 err = add_module(&arr, mod);
2937 if (err) {
2938 module_put(mod);
2939 break;
2940 }
2941 }
2942
2943 /* We return either err < 0 in case of error, ... */
2944 if (err) {
2945 kprobe_multi_put_modules(arr.mods, arr.mods_cnt);
2946 kfree(arr.mods);
2947 return err;
2948 }
2949
2950 /* or number of modules found if everything is ok. */
2951 *mods = arr.mods;
2952 return arr.mods_cnt;
2953 }
2954
addrs_check_error_injection_list(unsigned long * addrs,u32 cnt)2955 static int addrs_check_error_injection_list(unsigned long *addrs, u32 cnt)
2956 {
2957 u32 i;
2958
2959 for (i = 0; i < cnt; i++) {
2960 if (!within_error_injection_list(addrs[i]))
2961 return -EINVAL;
2962 }
2963 return 0;
2964 }
2965
bpf_kprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)2966 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2967 {
2968 struct bpf_kprobe_multi_link *link = NULL;
2969 struct bpf_link_primer link_primer;
2970 void __user *ucookies;
2971 unsigned long *addrs;
2972 u32 flags, cnt, size;
2973 void __user *uaddrs;
2974 u64 *cookies = NULL;
2975 void __user *usyms;
2976 int err;
2977
2978 /* no support for 32bit archs yet */
2979 if (sizeof(u64) != sizeof(void *))
2980 return -EOPNOTSUPP;
2981
2982 if (!is_kprobe_multi(prog))
2983 return -EINVAL;
2984
2985 flags = attr->link_create.kprobe_multi.flags;
2986 if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2987 return -EINVAL;
2988
2989 uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2990 usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2991 if (!!uaddrs == !!usyms)
2992 return -EINVAL;
2993
2994 cnt = attr->link_create.kprobe_multi.cnt;
2995 if (!cnt)
2996 return -EINVAL;
2997 if (cnt > MAX_KPROBE_MULTI_CNT)
2998 return -E2BIG;
2999
3000 size = cnt * sizeof(*addrs);
3001 addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
3002 if (!addrs)
3003 return -ENOMEM;
3004
3005 ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
3006 if (ucookies) {
3007 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
3008 if (!cookies) {
3009 err = -ENOMEM;
3010 goto error;
3011 }
3012 if (copy_from_user(cookies, ucookies, size)) {
3013 err = -EFAULT;
3014 goto error;
3015 }
3016 }
3017
3018 if (uaddrs) {
3019 if (copy_from_user(addrs, uaddrs, size)) {
3020 err = -EFAULT;
3021 goto error;
3022 }
3023 } else {
3024 struct multi_symbols_sort data = {
3025 .cookies = cookies,
3026 };
3027 struct user_syms us;
3028
3029 err = copy_user_syms(&us, usyms, cnt);
3030 if (err)
3031 goto error;
3032
3033 if (cookies)
3034 data.funcs = us.syms;
3035
3036 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
3037 symbols_swap_r, &data);
3038
3039 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
3040 free_user_syms(&us);
3041 if (err)
3042 goto error;
3043 }
3044
3045 if (prog->kprobe_override && addrs_check_error_injection_list(addrs, cnt)) {
3046 err = -EINVAL;
3047 goto error;
3048 }
3049
3050 link = kzalloc(sizeof(*link), GFP_KERNEL);
3051 if (!link) {
3052 err = -ENOMEM;
3053 goto error;
3054 }
3055
3056 bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
3057 &bpf_kprobe_multi_link_lops, prog);
3058
3059 err = bpf_link_prime(&link->link, &link_primer);
3060 if (err)
3061 goto error;
3062
3063 if (!(flags & BPF_F_KPROBE_MULTI_RETURN))
3064 link->fp.entry_handler = kprobe_multi_link_handler;
3065 if ((flags & BPF_F_KPROBE_MULTI_RETURN) || is_kprobe_session(prog))
3066 link->fp.exit_handler = kprobe_multi_link_exit_handler;
3067 if (is_kprobe_session(prog))
3068 link->fp.entry_data_size = sizeof(u64);
3069
3070 link->addrs = addrs;
3071 link->cookies = cookies;
3072 link->cnt = cnt;
3073 link->flags = flags;
3074
3075 if (cookies) {
3076 /*
3077 * Sorting addresses will trigger sorting cookies as well
3078 * (check bpf_kprobe_multi_cookie_swap). This way we can
3079 * find cookie based on the address in bpf_get_attach_cookie
3080 * helper.
3081 */
3082 sort_r(addrs, cnt, sizeof(*addrs),
3083 bpf_kprobe_multi_cookie_cmp,
3084 bpf_kprobe_multi_cookie_swap,
3085 link);
3086 }
3087
3088 err = get_modules_for_addrs(&link->mods, addrs, cnt);
3089 if (err < 0) {
3090 bpf_link_cleanup(&link_primer);
3091 return err;
3092 }
3093 link->mods_cnt = err;
3094
3095 err = register_fprobe_ips(&link->fp, addrs, cnt);
3096 if (err) {
3097 kprobe_multi_put_modules(link->mods, link->mods_cnt);
3098 bpf_link_cleanup(&link_primer);
3099 return err;
3100 }
3101
3102 return bpf_link_settle(&link_primer);
3103
3104 error:
3105 kfree(link);
3106 kvfree(addrs);
3107 kvfree(cookies);
3108 return err;
3109 }
3110 #else /* !CONFIG_FPROBE */
bpf_kprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)3111 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3112 {
3113 return -EOPNOTSUPP;
3114 }
bpf_kprobe_multi_cookie(struct bpf_run_ctx * ctx)3115 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
3116 {
3117 return 0;
3118 }
bpf_kprobe_multi_entry_ip(struct bpf_run_ctx * ctx)3119 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3120 {
3121 return 0;
3122 }
3123 #endif
3124
3125 #ifdef CONFIG_UPROBES
3126 struct bpf_uprobe_multi_link;
3127
3128 struct bpf_uprobe {
3129 struct bpf_uprobe_multi_link *link;
3130 loff_t offset;
3131 unsigned long ref_ctr_offset;
3132 u64 cookie;
3133 struct uprobe *uprobe;
3134 struct uprobe_consumer consumer;
3135 bool session;
3136 };
3137
3138 struct bpf_uprobe_multi_link {
3139 struct path path;
3140 struct bpf_link link;
3141 u32 cnt;
3142 u32 flags;
3143 struct bpf_uprobe *uprobes;
3144 struct task_struct *task;
3145 };
3146
3147 struct bpf_uprobe_multi_run_ctx {
3148 struct bpf_session_run_ctx session_ctx;
3149 unsigned long entry_ip;
3150 struct bpf_uprobe *uprobe;
3151 };
3152
bpf_uprobe_unregister(struct bpf_uprobe * uprobes,u32 cnt)3153 static void bpf_uprobe_unregister(struct bpf_uprobe *uprobes, u32 cnt)
3154 {
3155 u32 i;
3156
3157 for (i = 0; i < cnt; i++)
3158 uprobe_unregister_nosync(uprobes[i].uprobe, &uprobes[i].consumer);
3159
3160 if (cnt)
3161 uprobe_unregister_sync();
3162 }
3163
bpf_uprobe_multi_link_release(struct bpf_link * link)3164 static void bpf_uprobe_multi_link_release(struct bpf_link *link)
3165 {
3166 struct bpf_uprobe_multi_link *umulti_link;
3167
3168 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3169 bpf_uprobe_unregister(umulti_link->uprobes, umulti_link->cnt);
3170 if (umulti_link->task)
3171 put_task_struct(umulti_link->task);
3172 path_put(&umulti_link->path);
3173 }
3174
bpf_uprobe_multi_link_dealloc(struct bpf_link * link)3175 static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)
3176 {
3177 struct bpf_uprobe_multi_link *umulti_link;
3178
3179 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3180 kvfree(umulti_link->uprobes);
3181 kfree(umulti_link);
3182 }
3183
bpf_uprobe_multi_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)3184 static int bpf_uprobe_multi_link_fill_link_info(const struct bpf_link *link,
3185 struct bpf_link_info *info)
3186 {
3187 u64 __user *uref_ctr_offsets = u64_to_user_ptr(info->uprobe_multi.ref_ctr_offsets);
3188 u64 __user *ucookies = u64_to_user_ptr(info->uprobe_multi.cookies);
3189 u64 __user *uoffsets = u64_to_user_ptr(info->uprobe_multi.offsets);
3190 u64 __user *upath = u64_to_user_ptr(info->uprobe_multi.path);
3191 u32 upath_size = info->uprobe_multi.path_size;
3192 struct bpf_uprobe_multi_link *umulti_link;
3193 u32 ucount = info->uprobe_multi.count;
3194 int err = 0, i;
3195 char *p, *buf;
3196 long left = 0;
3197
3198 if (!upath ^ !upath_size)
3199 return -EINVAL;
3200
3201 if ((uoffsets || uref_ctr_offsets || ucookies) && !ucount)
3202 return -EINVAL;
3203
3204 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3205 info->uprobe_multi.count = umulti_link->cnt;
3206 info->uprobe_multi.flags = umulti_link->flags;
3207 info->uprobe_multi.pid = umulti_link->task ?
3208 task_pid_nr_ns(umulti_link->task, task_active_pid_ns(current)) : 0;
3209
3210 upath_size = upath_size ? min_t(u32, upath_size, PATH_MAX) : PATH_MAX;
3211 buf = kmalloc(upath_size, GFP_KERNEL);
3212 if (!buf)
3213 return -ENOMEM;
3214 p = d_path(&umulti_link->path, buf, upath_size);
3215 if (IS_ERR(p)) {
3216 kfree(buf);
3217 return PTR_ERR(p);
3218 }
3219 upath_size = buf + upath_size - p;
3220
3221 if (upath)
3222 left = copy_to_user(upath, p, upath_size);
3223 kfree(buf);
3224 if (left)
3225 return -EFAULT;
3226 info->uprobe_multi.path_size = upath_size;
3227
3228 if (!uoffsets && !ucookies && !uref_ctr_offsets)
3229 return 0;
3230
3231 if (ucount < umulti_link->cnt)
3232 err = -ENOSPC;
3233 else
3234 ucount = umulti_link->cnt;
3235
3236 for (i = 0; i < ucount; i++) {
3237 if (uoffsets &&
3238 put_user(umulti_link->uprobes[i].offset, uoffsets + i))
3239 return -EFAULT;
3240 if (uref_ctr_offsets &&
3241 put_user(umulti_link->uprobes[i].ref_ctr_offset, uref_ctr_offsets + i))
3242 return -EFAULT;
3243 if (ucookies &&
3244 put_user(umulti_link->uprobes[i].cookie, ucookies + i))
3245 return -EFAULT;
3246 }
3247
3248 return err;
3249 }
3250
3251 static const struct bpf_link_ops bpf_uprobe_multi_link_lops = {
3252 .release = bpf_uprobe_multi_link_release,
3253 .dealloc_deferred = bpf_uprobe_multi_link_dealloc,
3254 .fill_link_info = bpf_uprobe_multi_link_fill_link_info,
3255 };
3256
uprobe_prog_run(struct bpf_uprobe * uprobe,unsigned long entry_ip,struct pt_regs * regs,bool is_return,void * data)3257 static int uprobe_prog_run(struct bpf_uprobe *uprobe,
3258 unsigned long entry_ip,
3259 struct pt_regs *regs,
3260 bool is_return, void *data)
3261 {
3262 struct bpf_uprobe_multi_link *link = uprobe->link;
3263 struct bpf_uprobe_multi_run_ctx run_ctx = {
3264 .session_ctx = {
3265 .is_return = is_return,
3266 .data = data,
3267 },
3268 .entry_ip = entry_ip,
3269 .uprobe = uprobe,
3270 };
3271 struct bpf_prog *prog = link->link.prog;
3272 bool sleepable = prog->sleepable;
3273 struct bpf_run_ctx *old_run_ctx;
3274 int err;
3275
3276 if (link->task && !same_thread_group(current, link->task))
3277 return 0;
3278
3279 if (sleepable)
3280 rcu_read_lock_trace();
3281 else
3282 rcu_read_lock();
3283
3284 migrate_disable();
3285
3286 old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
3287 err = bpf_prog_run(link->link.prog, regs);
3288 bpf_reset_run_ctx(old_run_ctx);
3289
3290 migrate_enable();
3291
3292 if (sleepable)
3293 rcu_read_unlock_trace();
3294 else
3295 rcu_read_unlock();
3296 return err;
3297 }
3298
3299 static bool
uprobe_multi_link_filter(struct uprobe_consumer * con,struct mm_struct * mm)3300 uprobe_multi_link_filter(struct uprobe_consumer *con, struct mm_struct *mm)
3301 {
3302 struct bpf_uprobe *uprobe;
3303
3304 uprobe = container_of(con, struct bpf_uprobe, consumer);
3305 return uprobe->link->task->mm == mm;
3306 }
3307
3308 static int
uprobe_multi_link_handler(struct uprobe_consumer * con,struct pt_regs * regs,__u64 * data)3309 uprobe_multi_link_handler(struct uprobe_consumer *con, struct pt_regs *regs,
3310 __u64 *data)
3311 {
3312 struct bpf_uprobe *uprobe;
3313 int ret;
3314
3315 uprobe = container_of(con, struct bpf_uprobe, consumer);
3316 ret = uprobe_prog_run(uprobe, instruction_pointer(regs), regs, false, data);
3317 if (uprobe->session)
3318 return ret ? UPROBE_HANDLER_IGNORE : 0;
3319 return 0;
3320 }
3321
3322 static int
uprobe_multi_link_ret_handler(struct uprobe_consumer * con,unsigned long func,struct pt_regs * regs,__u64 * data)3323 uprobe_multi_link_ret_handler(struct uprobe_consumer *con, unsigned long func, struct pt_regs *regs,
3324 __u64 *data)
3325 {
3326 struct bpf_uprobe *uprobe;
3327
3328 uprobe = container_of(con, struct bpf_uprobe, consumer);
3329 uprobe_prog_run(uprobe, func, regs, true, data);
3330 return 0;
3331 }
3332
bpf_uprobe_multi_entry_ip(struct bpf_run_ctx * ctx)3333 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3334 {
3335 struct bpf_uprobe_multi_run_ctx *run_ctx;
3336
3337 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx,
3338 session_ctx.run_ctx);
3339 return run_ctx->entry_ip;
3340 }
3341
bpf_uprobe_multi_cookie(struct bpf_run_ctx * ctx)3342 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3343 {
3344 struct bpf_uprobe_multi_run_ctx *run_ctx;
3345
3346 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx,
3347 session_ctx.run_ctx);
3348 return run_ctx->uprobe->cookie;
3349 }
3350
bpf_uprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)3351 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3352 {
3353 struct bpf_uprobe_multi_link *link = NULL;
3354 unsigned long __user *uref_ctr_offsets;
3355 struct bpf_link_primer link_primer;
3356 struct bpf_uprobe *uprobes = NULL;
3357 struct task_struct *task = NULL;
3358 unsigned long __user *uoffsets;
3359 u64 __user *ucookies;
3360 void __user *upath;
3361 u32 flags, cnt, i;
3362 struct path path;
3363 char *name;
3364 pid_t pid;
3365 int err;
3366
3367 /* no support for 32bit archs yet */
3368 if (sizeof(u64) != sizeof(void *))
3369 return -EOPNOTSUPP;
3370
3371 if (!is_uprobe_multi(prog))
3372 return -EINVAL;
3373
3374 flags = attr->link_create.uprobe_multi.flags;
3375 if (flags & ~BPF_F_UPROBE_MULTI_RETURN)
3376 return -EINVAL;
3377
3378 /*
3379 * path, offsets and cnt are mandatory,
3380 * ref_ctr_offsets and cookies are optional
3381 */
3382 upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
3383 uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
3384 cnt = attr->link_create.uprobe_multi.cnt;
3385 pid = attr->link_create.uprobe_multi.pid;
3386
3387 if (!upath || !uoffsets || !cnt || pid < 0)
3388 return -EINVAL;
3389 if (cnt > MAX_UPROBE_MULTI_CNT)
3390 return -E2BIG;
3391
3392 uref_ctr_offsets = u64_to_user_ptr(attr->link_create.uprobe_multi.ref_ctr_offsets);
3393 ucookies = u64_to_user_ptr(attr->link_create.uprobe_multi.cookies);
3394
3395 name = strndup_user(upath, PATH_MAX);
3396 if (IS_ERR(name)) {
3397 err = PTR_ERR(name);
3398 return err;
3399 }
3400
3401 err = kern_path(name, LOOKUP_FOLLOW, &path);
3402 kfree(name);
3403 if (err)
3404 return err;
3405
3406 if (!d_is_reg(path.dentry)) {
3407 err = -EBADF;
3408 goto error_path_put;
3409 }
3410
3411 if (pid) {
3412 task = get_pid_task(find_vpid(pid), PIDTYPE_TGID);
3413 if (!task) {
3414 err = -ESRCH;
3415 goto error_path_put;
3416 }
3417 }
3418
3419 err = -ENOMEM;
3420
3421 link = kzalloc(sizeof(*link), GFP_KERNEL);
3422 uprobes = kvcalloc(cnt, sizeof(*uprobes), GFP_KERNEL);
3423
3424 if (!uprobes || !link)
3425 goto error_free;
3426
3427 for (i = 0; i < cnt; i++) {
3428 if (__get_user(uprobes[i].offset, uoffsets + i)) {
3429 err = -EFAULT;
3430 goto error_free;
3431 }
3432 if (uprobes[i].offset < 0) {
3433 err = -EINVAL;
3434 goto error_free;
3435 }
3436 if (uref_ctr_offsets && __get_user(uprobes[i].ref_ctr_offset, uref_ctr_offsets + i)) {
3437 err = -EFAULT;
3438 goto error_free;
3439 }
3440 if (ucookies && __get_user(uprobes[i].cookie, ucookies + i)) {
3441 err = -EFAULT;
3442 goto error_free;
3443 }
3444
3445 uprobes[i].link = link;
3446
3447 if (!(flags & BPF_F_UPROBE_MULTI_RETURN))
3448 uprobes[i].consumer.handler = uprobe_multi_link_handler;
3449 if (flags & BPF_F_UPROBE_MULTI_RETURN || is_uprobe_session(prog))
3450 uprobes[i].consumer.ret_handler = uprobe_multi_link_ret_handler;
3451 if (is_uprobe_session(prog))
3452 uprobes[i].session = true;
3453 if (pid)
3454 uprobes[i].consumer.filter = uprobe_multi_link_filter;
3455 }
3456
3457 link->cnt = cnt;
3458 link->uprobes = uprobes;
3459 link->path = path;
3460 link->task = task;
3461 link->flags = flags;
3462
3463 bpf_link_init(&link->link, BPF_LINK_TYPE_UPROBE_MULTI,
3464 &bpf_uprobe_multi_link_lops, prog);
3465
3466 for (i = 0; i < cnt; i++) {
3467 uprobes[i].uprobe = uprobe_register(d_real_inode(link->path.dentry),
3468 uprobes[i].offset,
3469 uprobes[i].ref_ctr_offset,
3470 &uprobes[i].consumer);
3471 if (IS_ERR(uprobes[i].uprobe)) {
3472 err = PTR_ERR(uprobes[i].uprobe);
3473 link->cnt = i;
3474 goto error_unregister;
3475 }
3476 }
3477
3478 err = bpf_link_prime(&link->link, &link_primer);
3479 if (err)
3480 goto error_unregister;
3481
3482 return bpf_link_settle(&link_primer);
3483
3484 error_unregister:
3485 bpf_uprobe_unregister(uprobes, link->cnt);
3486
3487 error_free:
3488 kvfree(uprobes);
3489 kfree(link);
3490 if (task)
3491 put_task_struct(task);
3492 error_path_put:
3493 path_put(&path);
3494 return err;
3495 }
3496 #else /* !CONFIG_UPROBES */
bpf_uprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)3497 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3498 {
3499 return -EOPNOTSUPP;
3500 }
bpf_uprobe_multi_cookie(struct bpf_run_ctx * ctx)3501 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3502 {
3503 return 0;
3504 }
bpf_uprobe_multi_entry_ip(struct bpf_run_ctx * ctx)3505 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3506 {
3507 return 0;
3508 }
3509 #endif /* CONFIG_UPROBES */
3510
3511 __bpf_kfunc_start_defs();
3512
bpf_session_is_return(void)3513 __bpf_kfunc bool bpf_session_is_return(void)
3514 {
3515 struct bpf_session_run_ctx *session_ctx;
3516
3517 session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3518 return session_ctx->is_return;
3519 }
3520
bpf_session_cookie(void)3521 __bpf_kfunc __u64 *bpf_session_cookie(void)
3522 {
3523 struct bpf_session_run_ctx *session_ctx;
3524
3525 session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3526 return session_ctx->data;
3527 }
3528
3529 __bpf_kfunc_end_defs();
3530
3531 BTF_KFUNCS_START(kprobe_multi_kfunc_set_ids)
BTF_ID_FLAGS(func,bpf_session_is_return)3532 BTF_ID_FLAGS(func, bpf_session_is_return)
3533 BTF_ID_FLAGS(func, bpf_session_cookie)
3534 BTF_KFUNCS_END(kprobe_multi_kfunc_set_ids)
3535
3536 static int bpf_kprobe_multi_filter(const struct bpf_prog *prog, u32 kfunc_id)
3537 {
3538 if (!btf_id_set8_contains(&kprobe_multi_kfunc_set_ids, kfunc_id))
3539 return 0;
3540
3541 if (!is_kprobe_session(prog) && !is_uprobe_session(prog))
3542 return -EACCES;
3543
3544 return 0;
3545 }
3546
3547 static const struct btf_kfunc_id_set bpf_kprobe_multi_kfunc_set = {
3548 .owner = THIS_MODULE,
3549 .set = &kprobe_multi_kfunc_set_ids,
3550 .filter = bpf_kprobe_multi_filter,
3551 };
3552
bpf_kprobe_multi_kfuncs_init(void)3553 static int __init bpf_kprobe_multi_kfuncs_init(void)
3554 {
3555 return register_btf_kfunc_id_set(BPF_PROG_TYPE_KPROBE, &bpf_kprobe_multi_kfunc_set);
3556 }
3557
3558 late_initcall(bpf_kprobe_multi_kfuncs_init);
3559
3560 __bpf_kfunc_start_defs();
3561
bpf_send_signal_task(struct task_struct * task,int sig,enum pid_type type,u64 value)3562 __bpf_kfunc int bpf_send_signal_task(struct task_struct *task, int sig, enum pid_type type,
3563 u64 value)
3564 {
3565 if (type != PIDTYPE_PID && type != PIDTYPE_TGID)
3566 return -EINVAL;
3567
3568 return bpf_send_signal_common(sig, type, task, value);
3569 }
3570
3571 __bpf_kfunc_end_defs();
3572