]>
Commit | Line | Data |
---|---|---|
179a0cc4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
2541517c | 2 | /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com |
0515e599 | 3 | * Copyright (c) 2016 Facebook |
2541517c AS |
4 | */ |
5 | #include <linux/kernel.h> | |
6 | #include <linux/types.h> | |
7 | #include <linux/slab.h> | |
8 | #include <linux/bpf.h> | |
0515e599 | 9 | #include <linux/bpf_perf_event.h> |
2541517c AS |
10 | #include <linux/filter.h> |
11 | #include <linux/uaccess.h> | |
9c959c86 | 12 | #include <linux/ctype.h> |
9802d865 | 13 | #include <linux/kprobes.h> |
41bdc4b4 | 14 | #include <linux/syscalls.h> |
540adea3 | 15 | #include <linux/error-injection.h> |
9802d865 | 16 | |
c7b6f29b NA |
17 | #include <asm/tlb.h> |
18 | ||
9802d865 | 19 | #include "trace_probe.h" |
2541517c AS |
20 | #include "trace.h" |
21 | ||
e672db03 SF |
22 | #define bpf_event_rcu_dereference(p) \ |
23 | rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex)) | |
24 | ||
a38d1107 MM |
25 | #ifdef CONFIG_MODULES |
26 | struct bpf_trace_module { | |
27 | struct module *module; | |
28 | struct list_head list; | |
29 | }; | |
30 | ||
31 | static LIST_HEAD(bpf_trace_modules); | |
32 | static DEFINE_MUTEX(bpf_module_mutex); | |
33 | ||
34 | static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) | |
35 | { | |
36 | struct bpf_raw_event_map *btp, *ret = NULL; | |
37 | struct bpf_trace_module *btm; | |
38 | unsigned int i; | |
39 | ||
40 | mutex_lock(&bpf_module_mutex); | |
41 | list_for_each_entry(btm, &bpf_trace_modules, list) { | |
42 | for (i = 0; i < btm->module->num_bpf_raw_events; ++i) { | |
43 | btp = &btm->module->bpf_raw_events[i]; | |
44 | if (!strcmp(btp->tp->name, name)) { | |
45 | if (try_module_get(btm->module)) | |
46 | ret = btp; | |
47 | goto out; | |
48 | } | |
49 | } | |
50 | } | |
51 | out: | |
52 | mutex_unlock(&bpf_module_mutex); | |
53 | return ret; | |
54 | } | |
55 | #else | |
56 | static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) | |
57 | { | |
58 | return NULL; | |
59 | } | |
60 | #endif /* CONFIG_MODULES */ | |
61 | ||
035226b9 | 62 | u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
c195651e | 63 | u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
035226b9 | 64 | |
2541517c AS |
65 | /** |
66 | * trace_call_bpf - invoke BPF program | |
e87c6bc3 | 67 | * @call: tracepoint event |
2541517c AS |
68 | * @ctx: opaque context pointer |
69 | * | |
70 | * kprobe handlers execute BPF programs via this helper. | |
71 | * Can be used from static tracepoints in the future. | |
72 | * | |
73 | * Return: BPF programs always return an integer which is interpreted by | |
74 | * kprobe handler as: | |
75 | * 0 - return from kprobe (event is filtered out) | |
76 | * 1 - store kprobe event into ring buffer | |
77 | * Other values are reserved and currently alias to 1 | |
78 | */ | |
e87c6bc3 | 79 | unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) |
2541517c AS |
80 | { |
81 | unsigned int ret; | |
82 | ||
83 | if (in_nmi()) /* not supported yet */ | |
84 | return 1; | |
85 | ||
b0a81b94 | 86 | cant_sleep(); |
2541517c AS |
87 | |
88 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
89 | /* | |
90 | * since some bpf program is already running on this cpu, | |
91 | * don't call into another bpf program (same or different) | |
92 | * and don't send kprobe event into ring-buffer, | |
93 | * so return zero here | |
94 | */ | |
95 | ret = 0; | |
96 | goto out; | |
97 | } | |
98 | ||
e87c6bc3 YS |
99 | /* |
100 | * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock | |
101 | * to all call sites, we did a bpf_prog_array_valid() there to check | |
102 | * whether call->prog_array is empty or not, which is | |
103 | * a heurisitc to speed up execution. | |
104 | * | |
105 | * If bpf_prog_array_valid() fetched prog_array was | |
106 | * non-NULL, we go into trace_call_bpf() and do the actual | |
107 | * proper rcu_dereference() under RCU lock. | |
108 | * If it turns out that prog_array is NULL then, we bail out. | |
109 | * For the opposite, if the bpf_prog_array_valid() fetched pointer | |
110 | * was NULL, you'll skip the prog_array with the risk of missing | |
111 | * out of events when it was updated in between this and the | |
112 | * rcu_dereference() which is accepted risk. | |
113 | */ | |
114 | ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN); | |
2541517c AS |
115 | |
116 | out: | |
117 | __this_cpu_dec(bpf_prog_active); | |
2541517c AS |
118 | |
119 | return ret; | |
120 | } | |
2541517c | 121 | |
9802d865 JB |
122 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
123 | BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) | |
124 | { | |
9802d865 | 125 | regs_set_return_value(regs, rc); |
540adea3 | 126 | override_function_with_return(regs); |
9802d865 JB |
127 | return 0; |
128 | } | |
129 | ||
130 | static const struct bpf_func_proto bpf_override_return_proto = { | |
131 | .func = bpf_override_return, | |
132 | .gpl_only = true, | |
133 | .ret_type = RET_INTEGER, | |
134 | .arg1_type = ARG_PTR_TO_CTX, | |
135 | .arg2_type = ARG_ANYTHING, | |
136 | }; | |
137 | #endif | |
138 | ||
6ae08ae3 DB |
139 | BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size, |
140 | const void __user *, unsafe_ptr) | |
2541517c | 141 | { |
6ae08ae3 | 142 | int ret = probe_user_read(dst, unsafe_ptr, size); |
2541517c | 143 | |
6ae08ae3 DB |
144 | if (unlikely(ret < 0)) |
145 | memset(dst, 0, size); | |
146 | ||
147 | return ret; | |
148 | } | |
149 | ||
f470378c | 150 | const struct bpf_func_proto bpf_probe_read_user_proto = { |
6ae08ae3 DB |
151 | .func = bpf_probe_read_user, |
152 | .gpl_only = true, | |
153 | .ret_type = RET_INTEGER, | |
154 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
155 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
156 | .arg3_type = ARG_ANYTHING, | |
157 | }; | |
158 | ||
159 | BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size, | |
160 | const void __user *, unsafe_ptr) | |
161 | { | |
bd88bb5d | 162 | int ret = strncpy_from_user_nofault(dst, unsafe_ptr, size); |
6ae08ae3 DB |
163 | |
164 | if (unlikely(ret < 0)) | |
165 | memset(dst, 0, size); | |
166 | ||
167 | return ret; | |
168 | } | |
169 | ||
f470378c | 170 | const struct bpf_func_proto bpf_probe_read_user_str_proto = { |
6ae08ae3 DB |
171 | .func = bpf_probe_read_user_str, |
172 | .gpl_only = true, | |
173 | .ret_type = RET_INTEGER, | |
174 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
175 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
176 | .arg3_type = ARG_ANYTHING, | |
177 | }; | |
178 | ||
179 | static __always_inline int | |
180 | bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr, | |
181 | const bool compat) | |
182 | { | |
183 | int ret = security_locked_down(LOCKDOWN_BPF_READ); | |
9d1f8be5 | 184 | |
6ae08ae3 DB |
185 | if (unlikely(ret < 0)) |
186 | goto out; | |
187 | ret = compat ? probe_kernel_read(dst, unsafe_ptr, size) : | |
188 | probe_kernel_read_strict(dst, unsafe_ptr, size); | |
074f528e | 189 | if (unlikely(ret < 0)) |
9d1f8be5 | 190 | out: |
074f528e | 191 | memset(dst, 0, size); |
6ae08ae3 DB |
192 | return ret; |
193 | } | |
074f528e | 194 | |
6ae08ae3 DB |
195 | BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size, |
196 | const void *, unsafe_ptr) | |
197 | { | |
198 | return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, false); | |
199 | } | |
200 | ||
f470378c | 201 | const struct bpf_func_proto bpf_probe_read_kernel_proto = { |
6ae08ae3 DB |
202 | .func = bpf_probe_read_kernel, |
203 | .gpl_only = true, | |
204 | .ret_type = RET_INTEGER, | |
205 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
206 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
207 | .arg3_type = ARG_ANYTHING, | |
208 | }; | |
209 | ||
210 | BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size, | |
211 | const void *, unsafe_ptr) | |
212 | { | |
213 | return bpf_probe_read_kernel_common(dst, size, unsafe_ptr, true); | |
214 | } | |
215 | ||
216 | static const struct bpf_func_proto bpf_probe_read_compat_proto = { | |
217 | .func = bpf_probe_read_compat, | |
218 | .gpl_only = true, | |
219 | .ret_type = RET_INTEGER, | |
220 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
221 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
222 | .arg3_type = ARG_ANYTHING, | |
223 | }; | |
224 | ||
225 | static __always_inline int | |
226 | bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr, | |
227 | const bool compat) | |
228 | { | |
229 | int ret = security_locked_down(LOCKDOWN_BPF_READ); | |
230 | ||
231 | if (unlikely(ret < 0)) | |
232 | goto out; | |
233 | /* | |
234 | * The strncpy_from_unsafe_*() call will likely not fill the entire | |
235 | * buffer, but that's okay in this circumstance as we're probing | |
236 | * arbitrary memory anyway similar to bpf_probe_read_*() and might | |
237 | * as well probe the stack. Thus, memory is explicitly cleared | |
238 | * only in error case, so that improper users ignoring return | |
239 | * code altogether don't copy garbage; otherwise length of string | |
240 | * is returned that can be used for bpf_perf_event_output() et al. | |
241 | */ | |
242 | ret = compat ? strncpy_from_unsafe(dst, unsafe_ptr, size) : | |
c4cb1644 | 243 | strncpy_from_kernel_nofault(dst, unsafe_ptr, size); |
6ae08ae3 DB |
244 | if (unlikely(ret < 0)) |
245 | out: | |
246 | memset(dst, 0, size); | |
074f528e | 247 | return ret; |
2541517c AS |
248 | } |
249 | ||
6ae08ae3 DB |
250 | BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size, |
251 | const void *, unsafe_ptr) | |
252 | { | |
253 | return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, false); | |
254 | } | |
255 | ||
f470378c | 256 | const struct bpf_func_proto bpf_probe_read_kernel_str_proto = { |
6ae08ae3 DB |
257 | .func = bpf_probe_read_kernel_str, |
258 | .gpl_only = true, | |
259 | .ret_type = RET_INTEGER, | |
260 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
261 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
262 | .arg3_type = ARG_ANYTHING, | |
263 | }; | |
264 | ||
265 | BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size, | |
266 | const void *, unsafe_ptr) | |
267 | { | |
268 | return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr, true); | |
269 | } | |
270 | ||
271 | static const struct bpf_func_proto bpf_probe_read_compat_str_proto = { | |
272 | .func = bpf_probe_read_compat_str, | |
2541517c AS |
273 | .gpl_only = true, |
274 | .ret_type = RET_INTEGER, | |
39f19ebb | 275 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
9c019e2b | 276 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, |
2541517c AS |
277 | .arg3_type = ARG_ANYTHING, |
278 | }; | |
279 | ||
eb1b6688 | 280 | BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src, |
f3694e00 | 281 | u32, size) |
96ae5227 | 282 | { |
96ae5227 SD |
283 | /* |
284 | * Ensure we're in user context which is safe for the helper to | |
285 | * run. This helper has no business in a kthread. | |
286 | * | |
287 | * access_ok() should prevent writing to non-user memory, but in | |
288 | * some situations (nommu, temporary switch, etc) access_ok() does | |
289 | * not provide enough validation, hence the check on KERNEL_DS. | |
c7b6f29b NA |
290 | * |
291 | * nmi_uaccess_okay() ensures the probe is not run in an interim | |
292 | * state, when the task or mm are switched. This is specifically | |
293 | * required to prevent the use of temporary mm. | |
96ae5227 SD |
294 | */ |
295 | ||
296 | if (unlikely(in_interrupt() || | |
297 | current->flags & (PF_KTHREAD | PF_EXITING))) | |
298 | return -EPERM; | |
db68ce10 | 299 | if (unlikely(uaccess_kernel())) |
96ae5227 | 300 | return -EPERM; |
c7b6f29b NA |
301 | if (unlikely(!nmi_uaccess_okay())) |
302 | return -EPERM; | |
96ae5227 | 303 | |
eb1b6688 | 304 | return probe_user_write(unsafe_ptr, src, size); |
96ae5227 SD |
305 | } |
306 | ||
307 | static const struct bpf_func_proto bpf_probe_write_user_proto = { | |
308 | .func = bpf_probe_write_user, | |
309 | .gpl_only = true, | |
310 | .ret_type = RET_INTEGER, | |
311 | .arg1_type = ARG_ANYTHING, | |
39f19ebb AS |
312 | .arg2_type = ARG_PTR_TO_MEM, |
313 | .arg3_type = ARG_CONST_SIZE, | |
96ae5227 SD |
314 | }; |
315 | ||
316 | static const struct bpf_func_proto *bpf_get_probe_write_proto(void) | |
317 | { | |
2c78ee89 AS |
318 | if (!capable(CAP_SYS_ADMIN)) |
319 | return NULL; | |
320 | ||
96ae5227 SD |
321 | pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", |
322 | current->comm, task_pid_nr(current)); | |
323 | ||
324 | return &bpf_probe_write_user_proto; | |
325 | } | |
326 | ||
d7b2977b CH |
327 | static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype, |
328 | size_t bufsz) | |
329 | { | |
330 | void __user *user_ptr = (__force void __user *)unsafe_ptr; | |
331 | ||
332 | buf[0] = 0; | |
333 | ||
334 | switch (fmt_ptype) { | |
335 | case 's': | |
336 | #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE | |
aec6ce59 CH |
337 | if ((unsigned long)unsafe_ptr < TASK_SIZE) { |
338 | strncpy_from_user_nofault(buf, user_ptr, bufsz); | |
339 | break; | |
340 | } | |
341 | fallthrough; | |
d7b2977b CH |
342 | #endif |
343 | case 'k': | |
344 | strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz); | |
345 | break; | |
346 | case 'u': | |
347 | strncpy_from_user_nofault(buf, user_ptr, bufsz); | |
348 | break; | |
349 | } | |
350 | } | |
351 | ||
9c959c86 | 352 | /* |
7bda4b40 | 353 | * Only limited trace_printk() conversion specifiers allowed: |
b2a5212f | 354 | * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pks %pus %s |
9c959c86 | 355 | */ |
f3694e00 DB |
356 | BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, |
357 | u64, arg2, u64, arg3) | |
9c959c86 | 358 | { |
b2a5212f DB |
359 | int i, mod[3] = {}, fmt_cnt = 0; |
360 | char buf[64], fmt_ptype; | |
361 | void *unsafe_ptr = NULL; | |
8d3b7dce | 362 | bool str_seen = false; |
9c959c86 AS |
363 | |
364 | /* | |
365 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
366 | * guarantees that fmt points to bpf program stack, | |
367 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
368 | */ | |
369 | if (fmt[--fmt_size] != 0) | |
370 | return -EINVAL; | |
371 | ||
372 | /* check format string for allowed specifiers */ | |
373 | for (i = 0; i < fmt_size; i++) { | |
374 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) | |
375 | return -EINVAL; | |
376 | ||
377 | if (fmt[i] != '%') | |
378 | continue; | |
379 | ||
380 | if (fmt_cnt >= 3) | |
381 | return -EINVAL; | |
382 | ||
383 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
384 | i++; | |
385 | if (fmt[i] == 'l') { | |
386 | mod[fmt_cnt]++; | |
387 | i++; | |
b2a5212f | 388 | } else if (fmt[i] == 'p') { |
9c959c86 | 389 | mod[fmt_cnt]++; |
b2a5212f DB |
390 | if ((fmt[i + 1] == 'k' || |
391 | fmt[i + 1] == 'u') && | |
392 | fmt[i + 2] == 's') { | |
393 | fmt_ptype = fmt[i + 1]; | |
394 | i += 2; | |
395 | goto fmt_str; | |
396 | } | |
397 | ||
1efb6ee3 MP |
398 | /* disallow any further format extensions */ |
399 | if (fmt[i + 1] != 0 && | |
400 | !isspace(fmt[i + 1]) && | |
401 | !ispunct(fmt[i + 1])) | |
9c959c86 | 402 | return -EINVAL; |
b2a5212f DB |
403 | |
404 | goto fmt_next; | |
405 | } else if (fmt[i] == 's') { | |
406 | mod[fmt_cnt]++; | |
407 | fmt_ptype = fmt[i]; | |
408 | fmt_str: | |
409 | if (str_seen) | |
410 | /* allow only one '%s' per fmt string */ | |
411 | return -EINVAL; | |
412 | str_seen = true; | |
413 | ||
414 | if (fmt[i + 1] != 0 && | |
415 | !isspace(fmt[i + 1]) && | |
416 | !ispunct(fmt[i + 1])) | |
417 | return -EINVAL; | |
418 | ||
419 | switch (fmt_cnt) { | |
420 | case 0: | |
421 | unsafe_ptr = (void *)(long)arg1; | |
422 | arg1 = (long)buf; | |
423 | break; | |
424 | case 1: | |
425 | unsafe_ptr = (void *)(long)arg2; | |
426 | arg2 = (long)buf; | |
427 | break; | |
428 | case 2: | |
429 | unsafe_ptr = (void *)(long)arg3; | |
430 | arg3 = (long)buf; | |
431 | break; | |
432 | } | |
433 | ||
d7b2977b CH |
434 | bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype, |
435 | sizeof(buf)); | |
b2a5212f | 436 | goto fmt_next; |
9c959c86 AS |
437 | } |
438 | ||
439 | if (fmt[i] == 'l') { | |
440 | mod[fmt_cnt]++; | |
441 | i++; | |
442 | } | |
443 | ||
7bda4b40 JF |
444 | if (fmt[i] != 'i' && fmt[i] != 'd' && |
445 | fmt[i] != 'u' && fmt[i] != 'x') | |
9c959c86 | 446 | return -EINVAL; |
b2a5212f | 447 | fmt_next: |
9c959c86 AS |
448 | fmt_cnt++; |
449 | } | |
450 | ||
88a5c690 DB |
451 | /* Horrid workaround for getting va_list handling working with different |
452 | * argument type combinations generically for 32 and 64 bit archs. | |
453 | */ | |
454 | #define __BPF_TP_EMIT() __BPF_ARG3_TP() | |
455 | #define __BPF_TP(...) \ | |
eefa864a | 456 | __trace_printk(0 /* Fake ip */, \ |
88a5c690 DB |
457 | fmt, ##__VA_ARGS__) |
458 | ||
459 | #define __BPF_ARG1_TP(...) \ | |
460 | ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \ | |
461 | ? __BPF_TP(arg1, ##__VA_ARGS__) \ | |
462 | : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \ | |
463 | ? __BPF_TP((long)arg1, ##__VA_ARGS__) \ | |
464 | : __BPF_TP((u32)arg1, ##__VA_ARGS__))) | |
465 | ||
466 | #define __BPF_ARG2_TP(...) \ | |
467 | ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \ | |
468 | ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \ | |
469 | : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \ | |
470 | ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \ | |
471 | : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__))) | |
472 | ||
473 | #define __BPF_ARG3_TP(...) \ | |
474 | ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \ | |
475 | ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \ | |
476 | : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \ | |
477 | ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \ | |
478 | : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__))) | |
479 | ||
480 | return __BPF_TP_EMIT(); | |
9c959c86 AS |
481 | } |
482 | ||
483 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
484 | .func = bpf_trace_printk, | |
485 | .gpl_only = true, | |
486 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
487 | .arg1_type = ARG_PTR_TO_MEM, |
488 | .arg2_type = ARG_CONST_SIZE, | |
9c959c86 AS |
489 | }; |
490 | ||
0756ea3e AS |
491 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
492 | { | |
493 | /* | |
494 | * this program might be calling bpf_trace_printk, | |
495 | * so allocate per-cpu printk buffers | |
496 | */ | |
497 | trace_printk_init_buffers(); | |
498 | ||
499 | return &bpf_trace_printk_proto; | |
500 | } | |
501 | ||
492e639f YS |
502 | #define MAX_SEQ_PRINTF_VARARGS 12 |
503 | #define MAX_SEQ_PRINTF_MAX_MEMCPY 6 | |
504 | #define MAX_SEQ_PRINTF_STR_LEN 128 | |
505 | ||
506 | struct bpf_seq_printf_buf { | |
507 | char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN]; | |
508 | }; | |
509 | static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf); | |
510 | static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used); | |
511 | ||
512 | BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size, | |
513 | const void *, data, u32, data_len) | |
514 | { | |
515 | int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0; | |
516 | int i, buf_used, copy_size, num_args; | |
517 | u64 params[MAX_SEQ_PRINTF_VARARGS]; | |
518 | struct bpf_seq_printf_buf *bufs; | |
519 | const u64 *args = data; | |
520 | ||
521 | buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used); | |
522 | if (WARN_ON_ONCE(buf_used > 1)) { | |
523 | err = -EBUSY; | |
524 | goto out; | |
525 | } | |
526 | ||
527 | bufs = this_cpu_ptr(&bpf_seq_printf_buf); | |
528 | ||
529 | /* | |
530 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
531 | * guarantees that fmt points to bpf program stack, | |
532 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
533 | */ | |
534 | if (fmt[--fmt_size] != 0) | |
535 | goto out; | |
536 | ||
537 | if (data_len & 7) | |
538 | goto out; | |
539 | ||
540 | for (i = 0; i < fmt_size; i++) { | |
541 | if (fmt[i] == '%') { | |
542 | if (fmt[i + 1] == '%') | |
543 | i++; | |
544 | else if (!data || !data_len) | |
545 | goto out; | |
546 | } | |
547 | } | |
548 | ||
549 | num_args = data_len / 8; | |
550 | ||
551 | /* check format string for allowed specifiers */ | |
552 | for (i = 0; i < fmt_size; i++) { | |
553 | /* only printable ascii for now. */ | |
554 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) { | |
555 | err = -EINVAL; | |
556 | goto out; | |
557 | } | |
558 | ||
559 | if (fmt[i] != '%') | |
560 | continue; | |
561 | ||
562 | if (fmt[i + 1] == '%') { | |
563 | i++; | |
564 | continue; | |
565 | } | |
566 | ||
567 | if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) { | |
568 | err = -E2BIG; | |
569 | goto out; | |
570 | } | |
571 | ||
572 | if (fmt_cnt >= num_args) { | |
573 | err = -EINVAL; | |
574 | goto out; | |
575 | } | |
576 | ||
577 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
578 | i++; | |
579 | ||
580 | /* skip optional "[0 +-][num]" width formating field */ | |
581 | while (fmt[i] == '0' || fmt[i] == '+' || fmt[i] == '-' || | |
582 | fmt[i] == ' ') | |
583 | i++; | |
584 | if (fmt[i] >= '1' && fmt[i] <= '9') { | |
585 | i++; | |
586 | while (fmt[i] >= '0' && fmt[i] <= '9') | |
587 | i++; | |
588 | } | |
589 | ||
590 | if (fmt[i] == 's') { | |
19c8d8ac AM |
591 | void *unsafe_ptr; |
592 | ||
492e639f YS |
593 | /* try our best to copy */ |
594 | if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) { | |
595 | err = -E2BIG; | |
596 | goto out; | |
597 | } | |
598 | ||
19c8d8ac AM |
599 | unsafe_ptr = (void *)(long)args[fmt_cnt]; |
600 | err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt], | |
601 | unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN); | |
492e639f YS |
602 | if (err < 0) |
603 | bufs->buf[memcpy_cnt][0] = '\0'; | |
604 | params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt]; | |
605 | ||
606 | fmt_cnt++; | |
607 | memcpy_cnt++; | |
608 | continue; | |
609 | } | |
610 | ||
611 | if (fmt[i] == 'p') { | |
612 | if (fmt[i + 1] == 0 || | |
613 | fmt[i + 1] == 'K' || | |
614 | fmt[i + 1] == 'x') { | |
615 | /* just kernel pointers */ | |
616 | params[fmt_cnt] = args[fmt_cnt]; | |
617 | fmt_cnt++; | |
618 | continue; | |
619 | } | |
620 | ||
621 | /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */ | |
622 | if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') { | |
623 | err = -EINVAL; | |
624 | goto out; | |
625 | } | |
626 | if (fmt[i + 2] != '4' && fmt[i + 2] != '6') { | |
627 | err = -EINVAL; | |
628 | goto out; | |
629 | } | |
630 | ||
631 | if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) { | |
632 | err = -E2BIG; | |
633 | goto out; | |
634 | } | |
635 | ||
636 | ||
637 | copy_size = (fmt[i + 2] == '4') ? 4 : 16; | |
638 | ||
639 | err = probe_kernel_read(bufs->buf[memcpy_cnt], | |
640 | (void *) (long) args[fmt_cnt], | |
641 | copy_size); | |
642 | if (err < 0) | |
643 | memset(bufs->buf[memcpy_cnt], 0, copy_size); | |
644 | params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt]; | |
645 | ||
646 | i += 2; | |
647 | fmt_cnt++; | |
648 | memcpy_cnt++; | |
649 | continue; | |
650 | } | |
651 | ||
652 | if (fmt[i] == 'l') { | |
653 | i++; | |
654 | if (fmt[i] == 'l') | |
655 | i++; | |
656 | } | |
657 | ||
658 | if (fmt[i] != 'i' && fmt[i] != 'd' && | |
659 | fmt[i] != 'u' && fmt[i] != 'x') { | |
660 | err = -EINVAL; | |
661 | goto out; | |
662 | } | |
663 | ||
664 | params[fmt_cnt] = args[fmt_cnt]; | |
665 | fmt_cnt++; | |
666 | } | |
667 | ||
668 | /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give | |
669 | * all of them to seq_printf(). | |
670 | */ | |
671 | seq_printf(m, fmt, params[0], params[1], params[2], params[3], | |
672 | params[4], params[5], params[6], params[7], params[8], | |
673 | params[9], params[10], params[11]); | |
674 | ||
675 | err = seq_has_overflowed(m) ? -EOVERFLOW : 0; | |
676 | out: | |
677 | this_cpu_dec(bpf_seq_printf_buf_used); | |
678 | return err; | |
679 | } | |
680 | ||
681 | static int bpf_seq_printf_btf_ids[5]; | |
682 | static const struct bpf_func_proto bpf_seq_printf_proto = { | |
683 | .func = bpf_seq_printf, | |
684 | .gpl_only = true, | |
685 | .ret_type = RET_INTEGER, | |
686 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
687 | .arg2_type = ARG_PTR_TO_MEM, | |
688 | .arg3_type = ARG_CONST_SIZE, | |
689 | .arg4_type = ARG_PTR_TO_MEM_OR_NULL, | |
690 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, | |
691 | .btf_id = bpf_seq_printf_btf_ids, | |
692 | }; | |
693 | ||
694 | BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len) | |
695 | { | |
696 | return seq_write(m, data, len) ? -EOVERFLOW : 0; | |
697 | } | |
698 | ||
699 | static int bpf_seq_write_btf_ids[5]; | |
700 | static const struct bpf_func_proto bpf_seq_write_proto = { | |
701 | .func = bpf_seq_write, | |
702 | .gpl_only = true, | |
703 | .ret_type = RET_INTEGER, | |
704 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
705 | .arg2_type = ARG_PTR_TO_MEM, | |
706 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
707 | .btf_id = bpf_seq_write_btf_ids, | |
708 | }; | |
709 | ||
908432ca YS |
710 | static __always_inline int |
711 | get_map_perf_counter(struct bpf_map *map, u64 flags, | |
712 | u64 *value, u64 *enabled, u64 *running) | |
35578d79 | 713 | { |
35578d79 | 714 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
6816a7ff DB |
715 | unsigned int cpu = smp_processor_id(); |
716 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 717 | struct bpf_event_entry *ee; |
35578d79 | 718 | |
6816a7ff DB |
719 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
720 | return -EINVAL; | |
721 | if (index == BPF_F_CURRENT_CPU) | |
722 | index = cpu; | |
35578d79 KX |
723 | if (unlikely(index >= array->map.max_entries)) |
724 | return -E2BIG; | |
725 | ||
3b1efb19 | 726 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 727 | if (!ee) |
35578d79 KX |
728 | return -ENOENT; |
729 | ||
908432ca YS |
730 | return perf_event_read_local(ee->event, value, enabled, running); |
731 | } | |
732 | ||
733 | BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) | |
734 | { | |
735 | u64 value = 0; | |
736 | int err; | |
737 | ||
738 | err = get_map_perf_counter(map, flags, &value, NULL, NULL); | |
35578d79 | 739 | /* |
f91840a3 AS |
740 | * this api is ugly since we miss [-22..-2] range of valid |
741 | * counter values, but that's uapi | |
35578d79 | 742 | */ |
f91840a3 AS |
743 | if (err) |
744 | return err; | |
745 | return value; | |
35578d79 KX |
746 | } |
747 | ||
62544ce8 | 748 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 749 | .func = bpf_perf_event_read, |
1075ef59 | 750 | .gpl_only = true, |
35578d79 KX |
751 | .ret_type = RET_INTEGER, |
752 | .arg1_type = ARG_CONST_MAP_PTR, | |
753 | .arg2_type = ARG_ANYTHING, | |
754 | }; | |
755 | ||
908432ca YS |
756 | BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, |
757 | struct bpf_perf_event_value *, buf, u32, size) | |
758 | { | |
759 | int err = -EINVAL; | |
760 | ||
761 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
762 | goto clear; | |
763 | err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, | |
764 | &buf->running); | |
765 | if (unlikely(err)) | |
766 | goto clear; | |
767 | return 0; | |
768 | clear: | |
769 | memset(buf, 0, size); | |
770 | return err; | |
771 | } | |
772 | ||
773 | static const struct bpf_func_proto bpf_perf_event_read_value_proto = { | |
774 | .func = bpf_perf_event_read_value, | |
775 | .gpl_only = true, | |
776 | .ret_type = RET_INTEGER, | |
777 | .arg1_type = ARG_CONST_MAP_PTR, | |
778 | .arg2_type = ARG_ANYTHING, | |
779 | .arg3_type = ARG_PTR_TO_UNINIT_MEM, | |
780 | .arg4_type = ARG_CONST_SIZE, | |
781 | }; | |
782 | ||
8e7a3920 DB |
783 | static __always_inline u64 |
784 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
283ca526 | 785 | u64 flags, struct perf_sample_data *sd) |
a43eec30 | 786 | { |
a43eec30 | 787 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
d7931330 | 788 | unsigned int cpu = smp_processor_id(); |
1e33759c | 789 | u64 index = flags & BPF_F_INDEX_MASK; |
3b1efb19 | 790 | struct bpf_event_entry *ee; |
a43eec30 | 791 | struct perf_event *event; |
a43eec30 | 792 | |
1e33759c | 793 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 794 | index = cpu; |
a43eec30 AS |
795 | if (unlikely(index >= array->map.max_entries)) |
796 | return -E2BIG; | |
797 | ||
3b1efb19 | 798 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 799 | if (!ee) |
a43eec30 AS |
800 | return -ENOENT; |
801 | ||
3b1efb19 | 802 | event = ee->event; |
a43eec30 AS |
803 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
804 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
805 | return -EINVAL; | |
806 | ||
d7931330 | 807 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
808 | return -EOPNOTSUPP; |
809 | ||
56201969 | 810 | return perf_event_output(event, sd, regs); |
a43eec30 AS |
811 | } |
812 | ||
9594dc3c MM |
813 | /* |
814 | * Support executing tracepoints in normal, irq, and nmi context that each call | |
815 | * bpf_perf_event_output | |
816 | */ | |
817 | struct bpf_trace_sample_data { | |
818 | struct perf_sample_data sds[3]; | |
819 | }; | |
820 | ||
821 | static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds); | |
822 | static DEFINE_PER_CPU(int, bpf_trace_nest_level); | |
f3694e00 DB |
823 | BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, |
824 | u64, flags, void *, data, u64, size) | |
8e7a3920 | 825 | { |
9594dc3c MM |
826 | struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds); |
827 | int nest_level = this_cpu_inc_return(bpf_trace_nest_level); | |
8e7a3920 DB |
828 | struct perf_raw_record raw = { |
829 | .frag = { | |
830 | .size = size, | |
831 | .data = data, | |
832 | }, | |
833 | }; | |
9594dc3c MM |
834 | struct perf_sample_data *sd; |
835 | int err; | |
8e7a3920 | 836 | |
9594dc3c MM |
837 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) { |
838 | err = -EBUSY; | |
839 | goto out; | |
840 | } | |
841 | ||
842 | sd = &sds->sds[nest_level - 1]; | |
843 | ||
844 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) { | |
845 | err = -EINVAL; | |
846 | goto out; | |
847 | } | |
8e7a3920 | 848 | |
283ca526 DB |
849 | perf_sample_data_init(sd, 0, 0); |
850 | sd->raw = &raw; | |
851 | ||
9594dc3c MM |
852 | err = __bpf_perf_event_output(regs, map, flags, sd); |
853 | ||
854 | out: | |
855 | this_cpu_dec(bpf_trace_nest_level); | |
856 | return err; | |
8e7a3920 DB |
857 | } |
858 | ||
a43eec30 AS |
859 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
860 | .func = bpf_perf_event_output, | |
1075ef59 | 861 | .gpl_only = true, |
a43eec30 AS |
862 | .ret_type = RET_INTEGER, |
863 | .arg1_type = ARG_PTR_TO_CTX, | |
864 | .arg2_type = ARG_CONST_MAP_PTR, | |
865 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 866 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 867 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
a43eec30 AS |
868 | }; |
869 | ||
768fb61f AZ |
870 | static DEFINE_PER_CPU(int, bpf_event_output_nest_level); |
871 | struct bpf_nested_pt_regs { | |
872 | struct pt_regs regs[3]; | |
873 | }; | |
874 | static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs); | |
875 | static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds); | |
bd570ff9 | 876 | |
555c8a86 DB |
877 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
878 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 879 | { |
768fb61f | 880 | int nest_level = this_cpu_inc_return(bpf_event_output_nest_level); |
555c8a86 DB |
881 | struct perf_raw_frag frag = { |
882 | .copy = ctx_copy, | |
883 | .size = ctx_size, | |
884 | .data = ctx, | |
885 | }; | |
886 | struct perf_raw_record raw = { | |
887 | .frag = { | |
183fc153 AM |
888 | { |
889 | .next = ctx_size ? &frag : NULL, | |
890 | }, | |
555c8a86 DB |
891 | .size = meta_size, |
892 | .data = meta, | |
893 | }, | |
894 | }; | |
768fb61f AZ |
895 | struct perf_sample_data *sd; |
896 | struct pt_regs *regs; | |
897 | u64 ret; | |
898 | ||
899 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) { | |
900 | ret = -EBUSY; | |
901 | goto out; | |
902 | } | |
903 | sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]); | |
904 | regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]); | |
bd570ff9 DB |
905 | |
906 | perf_fetch_caller_regs(regs); | |
283ca526 DB |
907 | perf_sample_data_init(sd, 0, 0); |
908 | sd->raw = &raw; | |
bd570ff9 | 909 | |
768fb61f AZ |
910 | ret = __bpf_perf_event_output(regs, map, flags, sd); |
911 | out: | |
912 | this_cpu_dec(bpf_event_output_nest_level); | |
913 | return ret; | |
bd570ff9 DB |
914 | } |
915 | ||
f3694e00 | 916 | BPF_CALL_0(bpf_get_current_task) |
606274c5 AS |
917 | { |
918 | return (long) current; | |
919 | } | |
920 | ||
f470378c | 921 | const struct bpf_func_proto bpf_get_current_task_proto = { |
606274c5 AS |
922 | .func = bpf_get_current_task, |
923 | .gpl_only = true, | |
924 | .ret_type = RET_INTEGER, | |
925 | }; | |
926 | ||
f3694e00 | 927 | BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) |
60d20f91 | 928 | { |
60d20f91 SD |
929 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
930 | struct cgroup *cgrp; | |
60d20f91 | 931 | |
60d20f91 SD |
932 | if (unlikely(idx >= array->map.max_entries)) |
933 | return -E2BIG; | |
934 | ||
935 | cgrp = READ_ONCE(array->ptrs[idx]); | |
936 | if (unlikely(!cgrp)) | |
937 | return -EAGAIN; | |
938 | ||
939 | return task_under_cgroup_hierarchy(current, cgrp); | |
940 | } | |
941 | ||
942 | static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { | |
943 | .func = bpf_current_task_under_cgroup, | |
944 | .gpl_only = false, | |
945 | .ret_type = RET_INTEGER, | |
946 | .arg1_type = ARG_CONST_MAP_PTR, | |
947 | .arg2_type = ARG_ANYTHING, | |
948 | }; | |
949 | ||
8b401f9e YS |
950 | struct send_signal_irq_work { |
951 | struct irq_work irq_work; | |
952 | struct task_struct *task; | |
953 | u32 sig; | |
8482941f | 954 | enum pid_type type; |
8b401f9e YS |
955 | }; |
956 | ||
957 | static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work); | |
958 | ||
959 | static void do_bpf_send_signal(struct irq_work *entry) | |
960 | { | |
961 | struct send_signal_irq_work *work; | |
962 | ||
963 | work = container_of(entry, struct send_signal_irq_work, irq_work); | |
8482941f | 964 | group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type); |
8b401f9e YS |
965 | } |
966 | ||
8482941f | 967 | static int bpf_send_signal_common(u32 sig, enum pid_type type) |
8b401f9e YS |
968 | { |
969 | struct send_signal_irq_work *work = NULL; | |
970 | ||
971 | /* Similar to bpf_probe_write_user, task needs to be | |
972 | * in a sound condition and kernel memory access be | |
973 | * permitted in order to send signal to the current | |
974 | * task. | |
975 | */ | |
976 | if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING))) | |
977 | return -EPERM; | |
978 | if (unlikely(uaccess_kernel())) | |
979 | return -EPERM; | |
980 | if (unlikely(!nmi_uaccess_okay())) | |
981 | return -EPERM; | |
982 | ||
1bc7896e | 983 | if (irqs_disabled()) { |
e1afb702 YS |
984 | /* Do an early check on signal validity. Otherwise, |
985 | * the error is lost in deferred irq_work. | |
986 | */ | |
987 | if (unlikely(!valid_signal(sig))) | |
988 | return -EINVAL; | |
989 | ||
8b401f9e | 990 | work = this_cpu_ptr(&send_signal_work); |
153bedba | 991 | if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY) |
8b401f9e YS |
992 | return -EBUSY; |
993 | ||
994 | /* Add the current task, which is the target of sending signal, | |
995 | * to the irq_work. The current task may change when queued | |
996 | * irq works get executed. | |
997 | */ | |
998 | work->task = current; | |
999 | work->sig = sig; | |
8482941f | 1000 | work->type = type; |
8b401f9e YS |
1001 | irq_work_queue(&work->irq_work); |
1002 | return 0; | |
1003 | } | |
1004 | ||
8482941f YS |
1005 | return group_send_sig_info(sig, SEND_SIG_PRIV, current, type); |
1006 | } | |
1007 | ||
1008 | BPF_CALL_1(bpf_send_signal, u32, sig) | |
1009 | { | |
1010 | return bpf_send_signal_common(sig, PIDTYPE_TGID); | |
8b401f9e YS |
1011 | } |
1012 | ||
1013 | static const struct bpf_func_proto bpf_send_signal_proto = { | |
1014 | .func = bpf_send_signal, | |
1015 | .gpl_only = false, | |
1016 | .ret_type = RET_INTEGER, | |
1017 | .arg1_type = ARG_ANYTHING, | |
1018 | }; | |
1019 | ||
8482941f YS |
1020 | BPF_CALL_1(bpf_send_signal_thread, u32, sig) |
1021 | { | |
1022 | return bpf_send_signal_common(sig, PIDTYPE_PID); | |
1023 | } | |
1024 | ||
1025 | static const struct bpf_func_proto bpf_send_signal_thread_proto = { | |
1026 | .func = bpf_send_signal_thread, | |
1027 | .gpl_only = false, | |
1028 | .ret_type = RET_INTEGER, | |
1029 | .arg1_type = ARG_ANYTHING, | |
1030 | }; | |
1031 | ||
fc611f47 KS |
1032 | const struct bpf_func_proto * |
1033 | bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
2541517c AS |
1034 | { |
1035 | switch (func_id) { | |
1036 | case BPF_FUNC_map_lookup_elem: | |
1037 | return &bpf_map_lookup_elem_proto; | |
1038 | case BPF_FUNC_map_update_elem: | |
1039 | return &bpf_map_update_elem_proto; | |
1040 | case BPF_FUNC_map_delete_elem: | |
1041 | return &bpf_map_delete_elem_proto; | |
02a8c817 AC |
1042 | case BPF_FUNC_map_push_elem: |
1043 | return &bpf_map_push_elem_proto; | |
1044 | case BPF_FUNC_map_pop_elem: | |
1045 | return &bpf_map_pop_elem_proto; | |
1046 | case BPF_FUNC_map_peek_elem: | |
1047 | return &bpf_map_peek_elem_proto; | |
d9847d31 AS |
1048 | case BPF_FUNC_ktime_get_ns: |
1049 | return &bpf_ktime_get_ns_proto; | |
71d19214 MÅ» |
1050 | case BPF_FUNC_ktime_get_boot_ns: |
1051 | return &bpf_ktime_get_boot_ns_proto; | |
04fd61ab AS |
1052 | case BPF_FUNC_tail_call: |
1053 | return &bpf_tail_call_proto; | |
ffeedafb AS |
1054 | case BPF_FUNC_get_current_pid_tgid: |
1055 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
1056 | case BPF_FUNC_get_current_task: |
1057 | return &bpf_get_current_task_proto; | |
ffeedafb AS |
1058 | case BPF_FUNC_get_current_uid_gid: |
1059 | return &bpf_get_current_uid_gid_proto; | |
1060 | case BPF_FUNC_get_current_comm: | |
1061 | return &bpf_get_current_comm_proto; | |
9c959c86 | 1062 | case BPF_FUNC_trace_printk: |
0756ea3e | 1063 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
1064 | case BPF_FUNC_get_smp_processor_id: |
1065 | return &bpf_get_smp_processor_id_proto; | |
2d0e30c3 DB |
1066 | case BPF_FUNC_get_numa_node_id: |
1067 | return &bpf_get_numa_node_id_proto; | |
35578d79 KX |
1068 | case BPF_FUNC_perf_event_read: |
1069 | return &bpf_perf_event_read_proto; | |
96ae5227 SD |
1070 | case BPF_FUNC_probe_write_user: |
1071 | return bpf_get_probe_write_proto(); | |
60d20f91 SD |
1072 | case BPF_FUNC_current_task_under_cgroup: |
1073 | return &bpf_current_task_under_cgroup_proto; | |
8937bd80 AS |
1074 | case BPF_FUNC_get_prandom_u32: |
1075 | return &bpf_get_prandom_u32_proto; | |
6ae08ae3 DB |
1076 | case BPF_FUNC_probe_read_user: |
1077 | return &bpf_probe_read_user_proto; | |
1078 | case BPF_FUNC_probe_read_kernel: | |
1079 | return &bpf_probe_read_kernel_proto; | |
6ae08ae3 DB |
1080 | case BPF_FUNC_probe_read_user_str: |
1081 | return &bpf_probe_read_user_str_proto; | |
1082 | case BPF_FUNC_probe_read_kernel_str: | |
1083 | return &bpf_probe_read_kernel_str_proto; | |
0ebeea8c DB |
1084 | #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE |
1085 | case BPF_FUNC_probe_read: | |
1086 | return &bpf_probe_read_compat_proto; | |
a5e8c070 | 1087 | case BPF_FUNC_probe_read_str: |
6ae08ae3 | 1088 | return &bpf_probe_read_compat_str_proto; |
0ebeea8c | 1089 | #endif |
34ea38ca | 1090 | #ifdef CONFIG_CGROUPS |
bf6fa2c8 YS |
1091 | case BPF_FUNC_get_current_cgroup_id: |
1092 | return &bpf_get_current_cgroup_id_proto; | |
34ea38ca | 1093 | #endif |
8b401f9e YS |
1094 | case BPF_FUNC_send_signal: |
1095 | return &bpf_send_signal_proto; | |
8482941f YS |
1096 | case BPF_FUNC_send_signal_thread: |
1097 | return &bpf_send_signal_thread_proto; | |
b80b033b SL |
1098 | case BPF_FUNC_perf_event_read_value: |
1099 | return &bpf_perf_event_read_value_proto; | |
b4490c5c CN |
1100 | case BPF_FUNC_get_ns_current_pid_tgid: |
1101 | return &bpf_get_ns_current_pid_tgid_proto; | |
457f4436 AN |
1102 | case BPF_FUNC_ringbuf_output: |
1103 | return &bpf_ringbuf_output_proto; | |
1104 | case BPF_FUNC_ringbuf_reserve: | |
1105 | return &bpf_ringbuf_reserve_proto; | |
1106 | case BPF_FUNC_ringbuf_submit: | |
1107 | return &bpf_ringbuf_submit_proto; | |
1108 | case BPF_FUNC_ringbuf_discard: | |
1109 | return &bpf_ringbuf_discard_proto; | |
1110 | case BPF_FUNC_ringbuf_query: | |
1111 | return &bpf_ringbuf_query_proto; | |
9fd82b61 AS |
1112 | default: |
1113 | return NULL; | |
1114 | } | |
1115 | } | |
1116 | ||
5e43f899 AI |
1117 | static const struct bpf_func_proto * |
1118 | kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
1119 | { |
1120 | switch (func_id) { | |
a43eec30 AS |
1121 | case BPF_FUNC_perf_event_output: |
1122 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
1123 | case BPF_FUNC_get_stackid: |
1124 | return &bpf_get_stackid_proto; | |
c195651e YS |
1125 | case BPF_FUNC_get_stack: |
1126 | return &bpf_get_stack_proto; | |
9802d865 JB |
1127 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
1128 | case BPF_FUNC_override_return: | |
1129 | return &bpf_override_return_proto; | |
1130 | #endif | |
2541517c | 1131 | default: |
fc611f47 | 1132 | return bpf_tracing_func_proto(func_id, prog); |
2541517c AS |
1133 | } |
1134 | } | |
1135 | ||
1136 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 | 1137 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 1138 | const struct bpf_prog *prog, |
23994631 | 1139 | struct bpf_insn_access_aux *info) |
2541517c | 1140 | { |
2541517c AS |
1141 | if (off < 0 || off >= sizeof(struct pt_regs)) |
1142 | return false; | |
2541517c AS |
1143 | if (type != BPF_READ) |
1144 | return false; | |
2541517c AS |
1145 | if (off % size != 0) |
1146 | return false; | |
2d071c64 DB |
1147 | /* |
1148 | * Assertion for 32 bit to make sure last 8 byte access | |
1149 | * (BPF_DW) to the last 4 byte member is disallowed. | |
1150 | */ | |
1151 | if (off + size > sizeof(struct pt_regs)) | |
1152 | return false; | |
1153 | ||
2541517c AS |
1154 | return true; |
1155 | } | |
1156 | ||
7de16e3a | 1157 | const struct bpf_verifier_ops kprobe_verifier_ops = { |
2541517c AS |
1158 | .get_func_proto = kprobe_prog_func_proto, |
1159 | .is_valid_access = kprobe_prog_is_valid_access, | |
1160 | }; | |
1161 | ||
7de16e3a JK |
1162 | const struct bpf_prog_ops kprobe_prog_ops = { |
1163 | }; | |
1164 | ||
f3694e00 DB |
1165 | BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, |
1166 | u64, flags, void *, data, u64, size) | |
9940d67c | 1167 | { |
f3694e00 DB |
1168 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
1169 | ||
9940d67c AS |
1170 | /* |
1171 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
1172 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
f3694e00 | 1173 | * from there and call the same bpf_perf_event_output() helper inline. |
9940d67c | 1174 | */ |
f3694e00 | 1175 | return ____bpf_perf_event_output(regs, map, flags, data, size); |
9940d67c AS |
1176 | } |
1177 | ||
1178 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
1179 | .func = bpf_perf_event_output_tp, | |
1180 | .gpl_only = true, | |
1181 | .ret_type = RET_INTEGER, | |
1182 | .arg1_type = ARG_PTR_TO_CTX, | |
1183 | .arg2_type = ARG_CONST_MAP_PTR, | |
1184 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 1185 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 1186 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
9940d67c AS |
1187 | }; |
1188 | ||
f3694e00 DB |
1189 | BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, |
1190 | u64, flags) | |
9940d67c | 1191 | { |
f3694e00 | 1192 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
9940d67c | 1193 | |
f3694e00 DB |
1194 | /* |
1195 | * Same comment as in bpf_perf_event_output_tp(), only that this time | |
1196 | * the other helper's function body cannot be inlined due to being | |
1197 | * external, thus we need to call raw helper function. | |
1198 | */ | |
1199 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
1200 | flags, 0, 0); | |
9940d67c AS |
1201 | } |
1202 | ||
1203 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
1204 | .func = bpf_get_stackid_tp, | |
1205 | .gpl_only = true, | |
1206 | .ret_type = RET_INTEGER, | |
1207 | .arg1_type = ARG_PTR_TO_CTX, | |
1208 | .arg2_type = ARG_CONST_MAP_PTR, | |
1209 | .arg3_type = ARG_ANYTHING, | |
1210 | }; | |
1211 | ||
c195651e YS |
1212 | BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size, |
1213 | u64, flags) | |
1214 | { | |
1215 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; | |
1216 | ||
1217 | return bpf_get_stack((unsigned long) regs, (unsigned long) buf, | |
1218 | (unsigned long) size, flags, 0); | |
1219 | } | |
1220 | ||
1221 | static const struct bpf_func_proto bpf_get_stack_proto_tp = { | |
1222 | .func = bpf_get_stack_tp, | |
1223 | .gpl_only = true, | |
1224 | .ret_type = RET_INTEGER, | |
1225 | .arg1_type = ARG_PTR_TO_CTX, | |
1226 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
1227 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1228 | .arg4_type = ARG_ANYTHING, | |
1229 | }; | |
1230 | ||
5e43f899 AI |
1231 | static const struct bpf_func_proto * |
1232 | tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
f005afed YS |
1233 | { |
1234 | switch (func_id) { | |
1235 | case BPF_FUNC_perf_event_output: | |
1236 | return &bpf_perf_event_output_proto_tp; | |
1237 | case BPF_FUNC_get_stackid: | |
1238 | return &bpf_get_stackid_proto_tp; | |
c195651e YS |
1239 | case BPF_FUNC_get_stack: |
1240 | return &bpf_get_stack_proto_tp; | |
f005afed | 1241 | default: |
fc611f47 | 1242 | return bpf_tracing_func_proto(func_id, prog); |
f005afed YS |
1243 | } |
1244 | } | |
1245 | ||
1246 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, | |
5e43f899 | 1247 | const struct bpf_prog *prog, |
f005afed YS |
1248 | struct bpf_insn_access_aux *info) |
1249 | { | |
1250 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
1251 | return false; | |
1252 | if (type != BPF_READ) | |
1253 | return false; | |
1254 | if (off % size != 0) | |
1255 | return false; | |
1256 | ||
1257 | BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); | |
1258 | return true; | |
1259 | } | |
1260 | ||
1261 | const struct bpf_verifier_ops tracepoint_verifier_ops = { | |
1262 | .get_func_proto = tp_prog_func_proto, | |
1263 | .is_valid_access = tp_prog_is_valid_access, | |
1264 | }; | |
1265 | ||
1266 | const struct bpf_prog_ops tracepoint_prog_ops = { | |
1267 | }; | |
1268 | ||
1269 | BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx, | |
4bebdc7a YS |
1270 | struct bpf_perf_event_value *, buf, u32, size) |
1271 | { | |
1272 | int err = -EINVAL; | |
1273 | ||
1274 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
1275 | goto clear; | |
1276 | err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, | |
1277 | &buf->running); | |
1278 | if (unlikely(err)) | |
1279 | goto clear; | |
1280 | return 0; | |
1281 | clear: | |
1282 | memset(buf, 0, size); | |
1283 | return err; | |
1284 | } | |
1285 | ||
f005afed YS |
1286 | static const struct bpf_func_proto bpf_perf_prog_read_value_proto = { |
1287 | .func = bpf_perf_prog_read_value, | |
4bebdc7a YS |
1288 | .gpl_only = true, |
1289 | .ret_type = RET_INTEGER, | |
1290 | .arg1_type = ARG_PTR_TO_CTX, | |
1291 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
1292 | .arg3_type = ARG_CONST_SIZE, | |
1293 | }; | |
1294 | ||
fff7b643 DX |
1295 | BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx, |
1296 | void *, buf, u32, size, u64, flags) | |
1297 | { | |
1298 | #ifndef CONFIG_X86 | |
1299 | return -ENOENT; | |
1300 | #else | |
1301 | static const u32 br_entry_size = sizeof(struct perf_branch_entry); | |
1302 | struct perf_branch_stack *br_stack = ctx->data->br_stack; | |
1303 | u32 to_copy; | |
1304 | ||
1305 | if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE)) | |
1306 | return -EINVAL; | |
1307 | ||
1308 | if (unlikely(!br_stack)) | |
1309 | return -EINVAL; | |
1310 | ||
1311 | if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE) | |
1312 | return br_stack->nr * br_entry_size; | |
1313 | ||
1314 | if (!buf || (size % br_entry_size != 0)) | |
1315 | return -EINVAL; | |
1316 | ||
1317 | to_copy = min_t(u32, br_stack->nr * br_entry_size, size); | |
1318 | memcpy(buf, br_stack->entries, to_copy); | |
1319 | ||
1320 | return to_copy; | |
1321 | #endif | |
1322 | } | |
1323 | ||
1324 | static const struct bpf_func_proto bpf_read_branch_records_proto = { | |
1325 | .func = bpf_read_branch_records, | |
1326 | .gpl_only = true, | |
1327 | .ret_type = RET_INTEGER, | |
1328 | .arg1_type = ARG_PTR_TO_CTX, | |
1329 | .arg2_type = ARG_PTR_TO_MEM_OR_NULL, | |
1330 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1331 | .arg4_type = ARG_ANYTHING, | |
1332 | }; | |
1333 | ||
5e43f899 AI |
1334 | static const struct bpf_func_proto * |
1335 | pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
1336 | { |
1337 | switch (func_id) { | |
1338 | case BPF_FUNC_perf_event_output: | |
9940d67c | 1339 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 1340 | case BPF_FUNC_get_stackid: |
9940d67c | 1341 | return &bpf_get_stackid_proto_tp; |
c195651e YS |
1342 | case BPF_FUNC_get_stack: |
1343 | return &bpf_get_stack_proto_tp; | |
4bebdc7a | 1344 | case BPF_FUNC_perf_prog_read_value: |
f005afed | 1345 | return &bpf_perf_prog_read_value_proto; |
fff7b643 DX |
1346 | case BPF_FUNC_read_branch_records: |
1347 | return &bpf_read_branch_records_proto; | |
9fd82b61 | 1348 | default: |
fc611f47 | 1349 | return bpf_tracing_func_proto(func_id, prog); |
9fd82b61 AS |
1350 | } |
1351 | } | |
1352 | ||
c4f6699d AS |
1353 | /* |
1354 | * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp | |
1355 | * to avoid potential recursive reuse issue when/if tracepoints are added | |
9594dc3c MM |
1356 | * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack. |
1357 | * | |
1358 | * Since raw tracepoints run despite bpf_prog_active, support concurrent usage | |
1359 | * in normal, irq, and nmi context. | |
c4f6699d | 1360 | */ |
9594dc3c MM |
1361 | struct bpf_raw_tp_regs { |
1362 | struct pt_regs regs[3]; | |
1363 | }; | |
1364 | static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs); | |
1365 | static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level); | |
1366 | static struct pt_regs *get_bpf_raw_tp_regs(void) | |
1367 | { | |
1368 | struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs); | |
1369 | int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level); | |
1370 | ||
1371 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) { | |
1372 | this_cpu_dec(bpf_raw_tp_nest_level); | |
1373 | return ERR_PTR(-EBUSY); | |
1374 | } | |
1375 | ||
1376 | return &tp_regs->regs[nest_level - 1]; | |
1377 | } | |
1378 | ||
1379 | static void put_bpf_raw_tp_regs(void) | |
1380 | { | |
1381 | this_cpu_dec(bpf_raw_tp_nest_level); | |
1382 | } | |
1383 | ||
c4f6699d AS |
1384 | BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1385 | struct bpf_map *, map, u64, flags, void *, data, u64, size) | |
1386 | { | |
9594dc3c MM |
1387 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1388 | int ret; | |
1389 | ||
1390 | if (IS_ERR(regs)) | |
1391 | return PTR_ERR(regs); | |
c4f6699d AS |
1392 | |
1393 | perf_fetch_caller_regs(regs); | |
9594dc3c MM |
1394 | ret = ____bpf_perf_event_output(regs, map, flags, data, size); |
1395 | ||
1396 | put_bpf_raw_tp_regs(); | |
1397 | return ret; | |
c4f6699d AS |
1398 | } |
1399 | ||
1400 | static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = { | |
1401 | .func = bpf_perf_event_output_raw_tp, | |
1402 | .gpl_only = true, | |
1403 | .ret_type = RET_INTEGER, | |
1404 | .arg1_type = ARG_PTR_TO_CTX, | |
1405 | .arg2_type = ARG_CONST_MAP_PTR, | |
1406 | .arg3_type = ARG_ANYTHING, | |
1407 | .arg4_type = ARG_PTR_TO_MEM, | |
1408 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, | |
1409 | }; | |
1410 | ||
a7658e1a | 1411 | extern const struct bpf_func_proto bpf_skb_output_proto; |
d831ee84 | 1412 | extern const struct bpf_func_proto bpf_xdp_output_proto; |
a7658e1a | 1413 | |
c4f6699d AS |
1414 | BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1415 | struct bpf_map *, map, u64, flags) | |
1416 | { | |
9594dc3c MM |
1417 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1418 | int ret; | |
1419 | ||
1420 | if (IS_ERR(regs)) | |
1421 | return PTR_ERR(regs); | |
c4f6699d AS |
1422 | |
1423 | perf_fetch_caller_regs(regs); | |
1424 | /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */ | |
9594dc3c MM |
1425 | ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map, |
1426 | flags, 0, 0); | |
1427 | put_bpf_raw_tp_regs(); | |
1428 | return ret; | |
c4f6699d AS |
1429 | } |
1430 | ||
1431 | static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = { | |
1432 | .func = bpf_get_stackid_raw_tp, | |
1433 | .gpl_only = true, | |
1434 | .ret_type = RET_INTEGER, | |
1435 | .arg1_type = ARG_PTR_TO_CTX, | |
1436 | .arg2_type = ARG_CONST_MAP_PTR, | |
1437 | .arg3_type = ARG_ANYTHING, | |
1438 | }; | |
1439 | ||
c195651e YS |
1440 | BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1441 | void *, buf, u32, size, u64, flags) | |
1442 | { | |
9594dc3c MM |
1443 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1444 | int ret; | |
1445 | ||
1446 | if (IS_ERR(regs)) | |
1447 | return PTR_ERR(regs); | |
c195651e YS |
1448 | |
1449 | perf_fetch_caller_regs(regs); | |
9594dc3c MM |
1450 | ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf, |
1451 | (unsigned long) size, flags, 0); | |
1452 | put_bpf_raw_tp_regs(); | |
1453 | return ret; | |
c195651e YS |
1454 | } |
1455 | ||
1456 | static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = { | |
1457 | .func = bpf_get_stack_raw_tp, | |
1458 | .gpl_only = true, | |
1459 | .ret_type = RET_INTEGER, | |
1460 | .arg1_type = ARG_PTR_TO_CTX, | |
1461 | .arg2_type = ARG_PTR_TO_MEM, | |
1462 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1463 | .arg4_type = ARG_ANYTHING, | |
1464 | }; | |
1465 | ||
5e43f899 AI |
1466 | static const struct bpf_func_proto * |
1467 | raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
c4f6699d AS |
1468 | { |
1469 | switch (func_id) { | |
1470 | case BPF_FUNC_perf_event_output: | |
1471 | return &bpf_perf_event_output_proto_raw_tp; | |
1472 | case BPF_FUNC_get_stackid: | |
1473 | return &bpf_get_stackid_proto_raw_tp; | |
c195651e YS |
1474 | case BPF_FUNC_get_stack: |
1475 | return &bpf_get_stack_proto_raw_tp; | |
c4f6699d | 1476 | default: |
fc611f47 | 1477 | return bpf_tracing_func_proto(func_id, prog); |
c4f6699d AS |
1478 | } |
1479 | } | |
1480 | ||
958a3f2d | 1481 | const struct bpf_func_proto * |
f1b9509c AS |
1482 | tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) |
1483 | { | |
1484 | switch (func_id) { | |
1485 | #ifdef CONFIG_NET | |
1486 | case BPF_FUNC_skb_output: | |
1487 | return &bpf_skb_output_proto; | |
d831ee84 EC |
1488 | case BPF_FUNC_xdp_output: |
1489 | return &bpf_xdp_output_proto; | |
f1b9509c | 1490 | #endif |
492e639f YS |
1491 | case BPF_FUNC_seq_printf: |
1492 | return prog->expected_attach_type == BPF_TRACE_ITER ? | |
1493 | &bpf_seq_printf_proto : | |
1494 | NULL; | |
1495 | case BPF_FUNC_seq_write: | |
1496 | return prog->expected_attach_type == BPF_TRACE_ITER ? | |
1497 | &bpf_seq_write_proto : | |
1498 | NULL; | |
f1b9509c AS |
1499 | default: |
1500 | return raw_tp_prog_func_proto(func_id, prog); | |
1501 | } | |
1502 | } | |
1503 | ||
c4f6699d AS |
1504 | static bool raw_tp_prog_is_valid_access(int off, int size, |
1505 | enum bpf_access_type type, | |
5e43f899 | 1506 | const struct bpf_prog *prog, |
c4f6699d AS |
1507 | struct bpf_insn_access_aux *info) |
1508 | { | |
f1b9509c AS |
1509 | if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) |
1510 | return false; | |
1511 | if (type != BPF_READ) | |
1512 | return false; | |
1513 | if (off % size != 0) | |
1514 | return false; | |
1515 | return true; | |
1516 | } | |
1517 | ||
1518 | static bool tracing_prog_is_valid_access(int off, int size, | |
1519 | enum bpf_access_type type, | |
1520 | const struct bpf_prog *prog, | |
1521 | struct bpf_insn_access_aux *info) | |
1522 | { | |
1523 | if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) | |
c4f6699d AS |
1524 | return false; |
1525 | if (type != BPF_READ) | |
1526 | return false; | |
1527 | if (off % size != 0) | |
1528 | return false; | |
9e15db66 | 1529 | return btf_ctx_access(off, size, type, prog, info); |
c4f6699d AS |
1530 | } |
1531 | ||
3e7c67d9 KS |
1532 | int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog, |
1533 | const union bpf_attr *kattr, | |
1534 | union bpf_attr __user *uattr) | |
1535 | { | |
1536 | return -ENOTSUPP; | |
1537 | } | |
1538 | ||
c4f6699d AS |
1539 | const struct bpf_verifier_ops raw_tracepoint_verifier_ops = { |
1540 | .get_func_proto = raw_tp_prog_func_proto, | |
1541 | .is_valid_access = raw_tp_prog_is_valid_access, | |
1542 | }; | |
1543 | ||
1544 | const struct bpf_prog_ops raw_tracepoint_prog_ops = { | |
1545 | }; | |
1546 | ||
f1b9509c AS |
1547 | const struct bpf_verifier_ops tracing_verifier_ops = { |
1548 | .get_func_proto = tracing_prog_func_proto, | |
1549 | .is_valid_access = tracing_prog_is_valid_access, | |
1550 | }; | |
1551 | ||
1552 | const struct bpf_prog_ops tracing_prog_ops = { | |
da00d2f1 | 1553 | .test_run = bpf_prog_test_run_tracing, |
f1b9509c AS |
1554 | }; |
1555 | ||
9df1c28b MM |
1556 | static bool raw_tp_writable_prog_is_valid_access(int off, int size, |
1557 | enum bpf_access_type type, | |
1558 | const struct bpf_prog *prog, | |
1559 | struct bpf_insn_access_aux *info) | |
1560 | { | |
1561 | if (off == 0) { | |
1562 | if (size != sizeof(u64) || type != BPF_READ) | |
1563 | return false; | |
1564 | info->reg_type = PTR_TO_TP_BUFFER; | |
1565 | } | |
1566 | return raw_tp_prog_is_valid_access(off, size, type, prog, info); | |
1567 | } | |
1568 | ||
1569 | const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = { | |
1570 | .get_func_proto = raw_tp_prog_func_proto, | |
1571 | .is_valid_access = raw_tp_writable_prog_is_valid_access, | |
1572 | }; | |
1573 | ||
1574 | const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = { | |
1575 | }; | |
1576 | ||
0515e599 | 1577 | static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 1578 | const struct bpf_prog *prog, |
23994631 | 1579 | struct bpf_insn_access_aux *info) |
0515e599 | 1580 | { |
95da0cdb | 1581 | const int size_u64 = sizeof(u64); |
31fd8581 | 1582 | |
0515e599 AS |
1583 | if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) |
1584 | return false; | |
1585 | if (type != BPF_READ) | |
1586 | return false; | |
bc23105c DB |
1587 | if (off % size != 0) { |
1588 | if (sizeof(unsigned long) != 4) | |
1589 | return false; | |
1590 | if (size != 8) | |
1591 | return false; | |
1592 | if (off % size != 4) | |
1593 | return false; | |
1594 | } | |
31fd8581 | 1595 | |
f96da094 DB |
1596 | switch (off) { |
1597 | case bpf_ctx_range(struct bpf_perf_event_data, sample_period): | |
95da0cdb TQ |
1598 | bpf_ctx_record_field_size(info, size_u64); |
1599 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
1600 | return false; | |
1601 | break; | |
1602 | case bpf_ctx_range(struct bpf_perf_event_data, addr): | |
1603 | bpf_ctx_record_field_size(info, size_u64); | |
1604 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
23994631 | 1605 | return false; |
f96da094 DB |
1606 | break; |
1607 | default: | |
0515e599 AS |
1608 | if (size != sizeof(long)) |
1609 | return false; | |
1610 | } | |
f96da094 | 1611 | |
0515e599 AS |
1612 | return true; |
1613 | } | |
1614 | ||
6b8cc1d1 DB |
1615 | static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, |
1616 | const struct bpf_insn *si, | |
0515e599 | 1617 | struct bpf_insn *insn_buf, |
f96da094 | 1618 | struct bpf_prog *prog, u32 *target_size) |
0515e599 AS |
1619 | { |
1620 | struct bpf_insn *insn = insn_buf; | |
1621 | ||
6b8cc1d1 | 1622 | switch (si->off) { |
0515e599 | 1623 | case offsetof(struct bpf_perf_event_data, sample_period): |
f035a515 | 1624 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 1625 | data), si->dst_reg, si->src_reg, |
0515e599 | 1626 | offsetof(struct bpf_perf_event_data_kern, data)); |
6b8cc1d1 | 1627 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, |
f96da094 DB |
1628 | bpf_target_off(struct perf_sample_data, period, 8, |
1629 | target_size)); | |
0515e599 | 1630 | break; |
95da0cdb TQ |
1631 | case offsetof(struct bpf_perf_event_data, addr): |
1632 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, | |
1633 | data), si->dst_reg, si->src_reg, | |
1634 | offsetof(struct bpf_perf_event_data_kern, data)); | |
1635 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, | |
1636 | bpf_target_off(struct perf_sample_data, addr, 8, | |
1637 | target_size)); | |
1638 | break; | |
0515e599 | 1639 | default: |
f035a515 | 1640 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 1641 | regs), si->dst_reg, si->src_reg, |
0515e599 | 1642 | offsetof(struct bpf_perf_event_data_kern, regs)); |
6b8cc1d1 DB |
1643 | *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, |
1644 | si->off); | |
0515e599 AS |
1645 | break; |
1646 | } | |
1647 | ||
1648 | return insn - insn_buf; | |
1649 | } | |
1650 | ||
7de16e3a | 1651 | const struct bpf_verifier_ops perf_event_verifier_ops = { |
f005afed | 1652 | .get_func_proto = pe_prog_func_proto, |
0515e599 AS |
1653 | .is_valid_access = pe_prog_is_valid_access, |
1654 | .convert_ctx_access = pe_prog_convert_ctx_access, | |
1655 | }; | |
7de16e3a JK |
1656 | |
1657 | const struct bpf_prog_ops perf_event_prog_ops = { | |
1658 | }; | |
e87c6bc3 YS |
1659 | |
1660 | static DEFINE_MUTEX(bpf_event_mutex); | |
1661 | ||
c8c088ba YS |
1662 | #define BPF_TRACE_MAX_PROGS 64 |
1663 | ||
e87c6bc3 YS |
1664 | int perf_event_attach_bpf_prog(struct perf_event *event, |
1665 | struct bpf_prog *prog) | |
1666 | { | |
e672db03 | 1667 | struct bpf_prog_array *old_array; |
e87c6bc3 YS |
1668 | struct bpf_prog_array *new_array; |
1669 | int ret = -EEXIST; | |
1670 | ||
9802d865 | 1671 | /* |
b4da3340 MH |
1672 | * Kprobe override only works if they are on the function entry, |
1673 | * and only if they are on the opt-in list. | |
9802d865 JB |
1674 | */ |
1675 | if (prog->kprobe_override && | |
b4da3340 | 1676 | (!trace_kprobe_on_func_entry(event->tp_event) || |
9802d865 JB |
1677 | !trace_kprobe_error_injectable(event->tp_event))) |
1678 | return -EINVAL; | |
1679 | ||
e87c6bc3 YS |
1680 | mutex_lock(&bpf_event_mutex); |
1681 | ||
1682 | if (event->prog) | |
07c41a29 | 1683 | goto unlock; |
e87c6bc3 | 1684 | |
e672db03 | 1685 | old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); |
c8c088ba YS |
1686 | if (old_array && |
1687 | bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { | |
1688 | ret = -E2BIG; | |
1689 | goto unlock; | |
1690 | } | |
1691 | ||
e87c6bc3 YS |
1692 | ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array); |
1693 | if (ret < 0) | |
07c41a29 | 1694 | goto unlock; |
e87c6bc3 YS |
1695 | |
1696 | /* set the new array to event->tp_event and set event->prog */ | |
1697 | event->prog = prog; | |
1698 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
1699 | bpf_prog_array_free(old_array); | |
1700 | ||
07c41a29 | 1701 | unlock: |
e87c6bc3 YS |
1702 | mutex_unlock(&bpf_event_mutex); |
1703 | return ret; | |
1704 | } | |
1705 | ||
1706 | void perf_event_detach_bpf_prog(struct perf_event *event) | |
1707 | { | |
e672db03 | 1708 | struct bpf_prog_array *old_array; |
e87c6bc3 YS |
1709 | struct bpf_prog_array *new_array; |
1710 | int ret; | |
1711 | ||
1712 | mutex_lock(&bpf_event_mutex); | |
1713 | ||
1714 | if (!event->prog) | |
07c41a29 | 1715 | goto unlock; |
e87c6bc3 | 1716 | |
e672db03 | 1717 | old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); |
e87c6bc3 | 1718 | ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array); |
170a7e3e SY |
1719 | if (ret == -ENOENT) |
1720 | goto unlock; | |
e87c6bc3 YS |
1721 | if (ret < 0) { |
1722 | bpf_prog_array_delete_safe(old_array, event->prog); | |
1723 | } else { | |
1724 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
1725 | bpf_prog_array_free(old_array); | |
1726 | } | |
1727 | ||
1728 | bpf_prog_put(event->prog); | |
1729 | event->prog = NULL; | |
1730 | ||
07c41a29 | 1731 | unlock: |
e87c6bc3 YS |
1732 | mutex_unlock(&bpf_event_mutex); |
1733 | } | |
f371b304 | 1734 | |
f4e2298e | 1735 | int perf_event_query_prog_array(struct perf_event *event, void __user *info) |
f371b304 YS |
1736 | { |
1737 | struct perf_event_query_bpf __user *uquery = info; | |
1738 | struct perf_event_query_bpf query = {}; | |
e672db03 | 1739 | struct bpf_prog_array *progs; |
3a38bb98 | 1740 | u32 *ids, prog_cnt, ids_len; |
f371b304 YS |
1741 | int ret; |
1742 | ||
031258da | 1743 | if (!perfmon_capable()) |
f371b304 YS |
1744 | return -EPERM; |
1745 | if (event->attr.type != PERF_TYPE_TRACEPOINT) | |
1746 | return -EINVAL; | |
1747 | if (copy_from_user(&query, uquery, sizeof(query))) | |
1748 | return -EFAULT; | |
3a38bb98 YS |
1749 | |
1750 | ids_len = query.ids_len; | |
1751 | if (ids_len > BPF_TRACE_MAX_PROGS) | |
9c481b90 | 1752 | return -E2BIG; |
3a38bb98 YS |
1753 | ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN); |
1754 | if (!ids) | |
1755 | return -ENOMEM; | |
1756 | /* | |
1757 | * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which | |
1758 | * is required when user only wants to check for uquery->prog_cnt. | |
1759 | * There is no need to check for it since the case is handled | |
1760 | * gracefully in bpf_prog_array_copy_info. | |
1761 | */ | |
f371b304 YS |
1762 | |
1763 | mutex_lock(&bpf_event_mutex); | |
e672db03 SF |
1764 | progs = bpf_event_rcu_dereference(event->tp_event->prog_array); |
1765 | ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt); | |
f371b304 YS |
1766 | mutex_unlock(&bpf_event_mutex); |
1767 | ||
3a38bb98 YS |
1768 | if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) || |
1769 | copy_to_user(uquery->ids, ids, ids_len * sizeof(u32))) | |
1770 | ret = -EFAULT; | |
1771 | ||
1772 | kfree(ids); | |
f371b304 YS |
1773 | return ret; |
1774 | } | |
c4f6699d AS |
1775 | |
1776 | extern struct bpf_raw_event_map __start__bpf_raw_tp[]; | |
1777 | extern struct bpf_raw_event_map __stop__bpf_raw_tp[]; | |
1778 | ||
a38d1107 | 1779 | struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name) |
c4f6699d AS |
1780 | { |
1781 | struct bpf_raw_event_map *btp = __start__bpf_raw_tp; | |
1782 | ||
1783 | for (; btp < __stop__bpf_raw_tp; btp++) { | |
1784 | if (!strcmp(btp->tp->name, name)) | |
1785 | return btp; | |
1786 | } | |
a38d1107 MM |
1787 | |
1788 | return bpf_get_raw_tracepoint_module(name); | |
1789 | } | |
1790 | ||
1791 | void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp) | |
1792 | { | |
1793 | struct module *mod = __module_address((unsigned long)btp); | |
1794 | ||
1795 | if (mod) | |
1796 | module_put(mod); | |
c4f6699d AS |
1797 | } |
1798 | ||
1799 | static __always_inline | |
1800 | void __bpf_trace_run(struct bpf_prog *prog, u64 *args) | |
1801 | { | |
f03efe49 | 1802 | cant_sleep(); |
c4f6699d | 1803 | rcu_read_lock(); |
c4f6699d | 1804 | (void) BPF_PROG_RUN(prog, args); |
c4f6699d AS |
1805 | rcu_read_unlock(); |
1806 | } | |
1807 | ||
1808 | #define UNPACK(...) __VA_ARGS__ | |
1809 | #define REPEAT_1(FN, DL, X, ...) FN(X) | |
1810 | #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__) | |
1811 | #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__) | |
1812 | #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__) | |
1813 | #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__) | |
1814 | #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__) | |
1815 | #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__) | |
1816 | #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__) | |
1817 | #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__) | |
1818 | #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__) | |
1819 | #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__) | |
1820 | #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__) | |
1821 | #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__) | |
1822 | ||
1823 | #define SARG(X) u64 arg##X | |
1824 | #define COPY(X) args[X] = arg##X | |
1825 | ||
1826 | #define __DL_COM (,) | |
1827 | #define __DL_SEM (;) | |
1828 | ||
1829 | #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 | |
1830 | ||
1831 | #define BPF_TRACE_DEFN_x(x) \ | |
1832 | void bpf_trace_run##x(struct bpf_prog *prog, \ | |
1833 | REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \ | |
1834 | { \ | |
1835 | u64 args[x]; \ | |
1836 | REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \ | |
1837 | __bpf_trace_run(prog, args); \ | |
1838 | } \ | |
1839 | EXPORT_SYMBOL_GPL(bpf_trace_run##x) | |
1840 | BPF_TRACE_DEFN_x(1); | |
1841 | BPF_TRACE_DEFN_x(2); | |
1842 | BPF_TRACE_DEFN_x(3); | |
1843 | BPF_TRACE_DEFN_x(4); | |
1844 | BPF_TRACE_DEFN_x(5); | |
1845 | BPF_TRACE_DEFN_x(6); | |
1846 | BPF_TRACE_DEFN_x(7); | |
1847 | BPF_TRACE_DEFN_x(8); | |
1848 | BPF_TRACE_DEFN_x(9); | |
1849 | BPF_TRACE_DEFN_x(10); | |
1850 | BPF_TRACE_DEFN_x(11); | |
1851 | BPF_TRACE_DEFN_x(12); | |
1852 | ||
1853 | static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1854 | { | |
1855 | struct tracepoint *tp = btp->tp; | |
1856 | ||
1857 | /* | |
1858 | * check that program doesn't access arguments beyond what's | |
1859 | * available in this tracepoint | |
1860 | */ | |
1861 | if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64)) | |
1862 | return -EINVAL; | |
1863 | ||
9df1c28b MM |
1864 | if (prog->aux->max_tp_access > btp->writable_size) |
1865 | return -EINVAL; | |
1866 | ||
c4f6699d AS |
1867 | return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog); |
1868 | } | |
1869 | ||
1870 | int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1871 | { | |
e16ec340 | 1872 | return __bpf_probe_register(btp, prog); |
c4f6699d AS |
1873 | } |
1874 | ||
1875 | int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1876 | { | |
e16ec340 | 1877 | return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog); |
c4f6699d | 1878 | } |
41bdc4b4 YS |
1879 | |
1880 | int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id, | |
1881 | u32 *fd_type, const char **buf, | |
1882 | u64 *probe_offset, u64 *probe_addr) | |
1883 | { | |
1884 | bool is_tracepoint, is_syscall_tp; | |
1885 | struct bpf_prog *prog; | |
1886 | int flags, err = 0; | |
1887 | ||
1888 | prog = event->prog; | |
1889 | if (!prog) | |
1890 | return -ENOENT; | |
1891 | ||
1892 | /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */ | |
1893 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) | |
1894 | return -EOPNOTSUPP; | |
1895 | ||
1896 | *prog_id = prog->aux->id; | |
1897 | flags = event->tp_event->flags; | |
1898 | is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT; | |
1899 | is_syscall_tp = is_syscall_trace_event(event->tp_event); | |
1900 | ||
1901 | if (is_tracepoint || is_syscall_tp) { | |
1902 | *buf = is_tracepoint ? event->tp_event->tp->name | |
1903 | : event->tp_event->name; | |
1904 | *fd_type = BPF_FD_TYPE_TRACEPOINT; | |
1905 | *probe_offset = 0x0; | |
1906 | *probe_addr = 0x0; | |
1907 | } else { | |
1908 | /* kprobe/uprobe */ | |
1909 | err = -EOPNOTSUPP; | |
1910 | #ifdef CONFIG_KPROBE_EVENTS | |
1911 | if (flags & TRACE_EVENT_FL_KPROBE) | |
1912 | err = bpf_get_kprobe_info(event, fd_type, buf, | |
1913 | probe_offset, probe_addr, | |
1914 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1915 | #endif | |
1916 | #ifdef CONFIG_UPROBE_EVENTS | |
1917 | if (flags & TRACE_EVENT_FL_UPROBE) | |
1918 | err = bpf_get_uprobe_info(event, fd_type, buf, | |
1919 | probe_offset, | |
1920 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1921 | #endif | |
1922 | } | |
1923 | ||
1924 | return err; | |
1925 | } | |
a38d1107 | 1926 | |
9db1ff0a YS |
1927 | static int __init send_signal_irq_work_init(void) |
1928 | { | |
1929 | int cpu; | |
1930 | struct send_signal_irq_work *work; | |
1931 | ||
1932 | for_each_possible_cpu(cpu) { | |
1933 | work = per_cpu_ptr(&send_signal_work, cpu); | |
1934 | init_irq_work(&work->irq_work, do_bpf_send_signal); | |
1935 | } | |
1936 | return 0; | |
1937 | } | |
1938 | ||
1939 | subsys_initcall(send_signal_irq_work_init); | |
1940 | ||
a38d1107 | 1941 | #ifdef CONFIG_MODULES |
390e99cf SF |
1942 | static int bpf_event_notify(struct notifier_block *nb, unsigned long op, |
1943 | void *module) | |
a38d1107 MM |
1944 | { |
1945 | struct bpf_trace_module *btm, *tmp; | |
1946 | struct module *mod = module; | |
1947 | ||
1948 | if (mod->num_bpf_raw_events == 0 || | |
1949 | (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING)) | |
1950 | return 0; | |
1951 | ||
1952 | mutex_lock(&bpf_module_mutex); | |
1953 | ||
1954 | switch (op) { | |
1955 | case MODULE_STATE_COMING: | |
1956 | btm = kzalloc(sizeof(*btm), GFP_KERNEL); | |
1957 | if (btm) { | |
1958 | btm->module = module; | |
1959 | list_add(&btm->list, &bpf_trace_modules); | |
1960 | } | |
1961 | break; | |
1962 | case MODULE_STATE_GOING: | |
1963 | list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) { | |
1964 | if (btm->module == module) { | |
1965 | list_del(&btm->list); | |
1966 | kfree(btm); | |
1967 | break; | |
1968 | } | |
1969 | } | |
1970 | break; | |
1971 | } | |
1972 | ||
1973 | mutex_unlock(&bpf_module_mutex); | |
1974 | ||
1975 | return 0; | |
1976 | } | |
1977 | ||
1978 | static struct notifier_block bpf_module_nb = { | |
1979 | .notifier_call = bpf_event_notify, | |
1980 | }; | |
1981 | ||
390e99cf | 1982 | static int __init bpf_event_init(void) |
a38d1107 MM |
1983 | { |
1984 | register_module_notifier(&bpf_module_nb); | |
1985 | return 0; | |
1986 | } | |
1987 | ||
1988 | fs_initcall(bpf_event_init); | |
1989 | #endif /* CONFIG_MODULES */ |