Commit | Line | Data |
---|---|---|
2541517c | 1 | /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com |
0515e599 | 2 | * Copyright (c) 2016 Facebook |
2541517c AS |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | */ | |
8 | #include <linux/kernel.h> | |
9 | #include <linux/types.h> | |
10 | #include <linux/slab.h> | |
11 | #include <linux/bpf.h> | |
0515e599 | 12 | #include <linux/bpf_perf_event.h> |
2541517c AS |
13 | #include <linux/filter.h> |
14 | #include <linux/uaccess.h> | |
9c959c86 | 15 | #include <linux/ctype.h> |
9802d865 JB |
16 | #include <linux/kprobes.h> |
17 | #include <asm/kprobes.h> | |
18 | ||
19 | #include "trace_probe.h" | |
2541517c AS |
20 | #include "trace.h" |
21 | ||
035226b9 GB |
22 | u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
23 | ||
2541517c AS |
24 | /** |
25 | * trace_call_bpf - invoke BPF program | |
e87c6bc3 | 26 | * @call: tracepoint event |
2541517c AS |
27 | * @ctx: opaque context pointer |
28 | * | |
29 | * kprobe handlers execute BPF programs via this helper. | |
30 | * Can be used from static tracepoints in the future. | |
31 | * | |
32 | * Return: BPF programs always return an integer which is interpreted by | |
33 | * kprobe handler as: | |
34 | * 0 - return from kprobe (event is filtered out) | |
35 | * 1 - store kprobe event into ring buffer | |
36 | * Other values are reserved and currently alias to 1 | |
37 | */ | |
e87c6bc3 | 38 | unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) |
2541517c AS |
39 | { |
40 | unsigned int ret; | |
41 | ||
42 | if (in_nmi()) /* not supported yet */ | |
43 | return 1; | |
44 | ||
45 | preempt_disable(); | |
46 | ||
47 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
48 | /* | |
49 | * since some bpf program is already running on this cpu, | |
50 | * don't call into another bpf program (same or different) | |
51 | * and don't send kprobe event into ring-buffer, | |
52 | * so return zero here | |
53 | */ | |
54 | ret = 0; | |
55 | goto out; | |
56 | } | |
57 | ||
e87c6bc3 YS |
58 | /* |
59 | * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock | |
60 | * to all call sites, we did a bpf_prog_array_valid() there to check | |
61 | * whether call->prog_array is empty or not, which is | |
62 | * a heurisitc to speed up execution. | |
63 | * | |
64 | * If bpf_prog_array_valid() fetched prog_array was | |
65 | * non-NULL, we go into trace_call_bpf() and do the actual | |
66 | * proper rcu_dereference() under RCU lock. | |
67 | * If it turns out that prog_array is NULL then, we bail out. | |
68 | * For the opposite, if the bpf_prog_array_valid() fetched pointer | |
69 | * was NULL, you'll skip the prog_array with the risk of missing | |
70 | * out of events when it was updated in between this and the | |
71 | * rcu_dereference() which is accepted risk. | |
72 | */ | |
73 | ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN); | |
2541517c AS |
74 | |
75 | out: | |
76 | __this_cpu_dec(bpf_prog_active); | |
77 | preempt_enable(); | |
78 | ||
79 | return ret; | |
80 | } | |
81 | EXPORT_SYMBOL_GPL(trace_call_bpf); | |
82 | ||
9802d865 JB |
83 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
84 | BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) | |
85 | { | |
9802d865 | 86 | regs_set_return_value(regs, rc); |
b4da3340 | 87 | arch_kprobe_override_function(regs); |
9802d865 JB |
88 | return 0; |
89 | } | |
90 | ||
91 | static const struct bpf_func_proto bpf_override_return_proto = { | |
92 | .func = bpf_override_return, | |
93 | .gpl_only = true, | |
94 | .ret_type = RET_INTEGER, | |
95 | .arg1_type = ARG_PTR_TO_CTX, | |
96 | .arg2_type = ARG_ANYTHING, | |
97 | }; | |
98 | #endif | |
99 | ||
f3694e00 | 100 | BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr) |
2541517c | 101 | { |
eb33f2cc | 102 | int ret; |
2541517c | 103 | |
074f528e DB |
104 | ret = probe_kernel_read(dst, unsafe_ptr, size); |
105 | if (unlikely(ret < 0)) | |
106 | memset(dst, 0, size); | |
107 | ||
108 | return ret; | |
2541517c AS |
109 | } |
110 | ||
111 | static const struct bpf_func_proto bpf_probe_read_proto = { | |
112 | .func = bpf_probe_read, | |
113 | .gpl_only = true, | |
114 | .ret_type = RET_INTEGER, | |
39f19ebb | 115 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
9c019e2b | 116 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, |
2541517c AS |
117 | .arg3_type = ARG_ANYTHING, |
118 | }; | |
119 | ||
f3694e00 DB |
120 | BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src, |
121 | u32, size) | |
96ae5227 | 122 | { |
96ae5227 SD |
123 | /* |
124 | * Ensure we're in user context which is safe for the helper to | |
125 | * run. This helper has no business in a kthread. | |
126 | * | |
127 | * access_ok() should prevent writing to non-user memory, but in | |
128 | * some situations (nommu, temporary switch, etc) access_ok() does | |
129 | * not provide enough validation, hence the check on KERNEL_DS. | |
130 | */ | |
131 | ||
132 | if (unlikely(in_interrupt() || | |
133 | current->flags & (PF_KTHREAD | PF_EXITING))) | |
134 | return -EPERM; | |
db68ce10 | 135 | if (unlikely(uaccess_kernel())) |
96ae5227 SD |
136 | return -EPERM; |
137 | if (!access_ok(VERIFY_WRITE, unsafe_ptr, size)) | |
138 | return -EPERM; | |
139 | ||
140 | return probe_kernel_write(unsafe_ptr, src, size); | |
141 | } | |
142 | ||
143 | static const struct bpf_func_proto bpf_probe_write_user_proto = { | |
144 | .func = bpf_probe_write_user, | |
145 | .gpl_only = true, | |
146 | .ret_type = RET_INTEGER, | |
147 | .arg1_type = ARG_ANYTHING, | |
39f19ebb AS |
148 | .arg2_type = ARG_PTR_TO_MEM, |
149 | .arg3_type = ARG_CONST_SIZE, | |
96ae5227 SD |
150 | }; |
151 | ||
152 | static const struct bpf_func_proto *bpf_get_probe_write_proto(void) | |
153 | { | |
154 | pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", | |
155 | current->comm, task_pid_nr(current)); | |
156 | ||
157 | return &bpf_probe_write_user_proto; | |
158 | } | |
159 | ||
9c959c86 | 160 | /* |
7bda4b40 JF |
161 | * Only limited trace_printk() conversion specifiers allowed: |
162 | * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s | |
9c959c86 | 163 | */ |
f3694e00 DB |
164 | BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, |
165 | u64, arg2, u64, arg3) | |
9c959c86 | 166 | { |
8d3b7dce | 167 | bool str_seen = false; |
9c959c86 AS |
168 | int mod[3] = {}; |
169 | int fmt_cnt = 0; | |
8d3b7dce AS |
170 | u64 unsafe_addr; |
171 | char buf[64]; | |
9c959c86 AS |
172 | int i; |
173 | ||
174 | /* | |
175 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
176 | * guarantees that fmt points to bpf program stack, | |
177 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
178 | */ | |
179 | if (fmt[--fmt_size] != 0) | |
180 | return -EINVAL; | |
181 | ||
182 | /* check format string for allowed specifiers */ | |
183 | for (i = 0; i < fmt_size; i++) { | |
184 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) | |
185 | return -EINVAL; | |
186 | ||
187 | if (fmt[i] != '%') | |
188 | continue; | |
189 | ||
190 | if (fmt_cnt >= 3) | |
191 | return -EINVAL; | |
192 | ||
193 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
194 | i++; | |
195 | if (fmt[i] == 'l') { | |
196 | mod[fmt_cnt]++; | |
197 | i++; | |
8d3b7dce | 198 | } else if (fmt[i] == 'p' || fmt[i] == 's') { |
9c959c86 AS |
199 | mod[fmt_cnt]++; |
200 | i++; | |
201 | if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0) | |
202 | return -EINVAL; | |
203 | fmt_cnt++; | |
8d3b7dce AS |
204 | if (fmt[i - 1] == 's') { |
205 | if (str_seen) | |
206 | /* allow only one '%s' per fmt string */ | |
207 | return -EINVAL; | |
208 | str_seen = true; | |
209 | ||
210 | switch (fmt_cnt) { | |
211 | case 1: | |
f3694e00 DB |
212 | unsafe_addr = arg1; |
213 | arg1 = (long) buf; | |
8d3b7dce AS |
214 | break; |
215 | case 2: | |
f3694e00 DB |
216 | unsafe_addr = arg2; |
217 | arg2 = (long) buf; | |
8d3b7dce AS |
218 | break; |
219 | case 3: | |
f3694e00 DB |
220 | unsafe_addr = arg3; |
221 | arg3 = (long) buf; | |
8d3b7dce AS |
222 | break; |
223 | } | |
224 | buf[0] = 0; | |
225 | strncpy_from_unsafe(buf, | |
226 | (void *) (long) unsafe_addr, | |
227 | sizeof(buf)); | |
228 | } | |
9c959c86 AS |
229 | continue; |
230 | } | |
231 | ||
232 | if (fmt[i] == 'l') { | |
233 | mod[fmt_cnt]++; | |
234 | i++; | |
235 | } | |
236 | ||
7bda4b40 JF |
237 | if (fmt[i] != 'i' && fmt[i] != 'd' && |
238 | fmt[i] != 'u' && fmt[i] != 'x') | |
9c959c86 AS |
239 | return -EINVAL; |
240 | fmt_cnt++; | |
241 | } | |
242 | ||
88a5c690 DB |
243 | /* Horrid workaround for getting va_list handling working with different |
244 | * argument type combinations generically for 32 and 64 bit archs. | |
245 | */ | |
246 | #define __BPF_TP_EMIT() __BPF_ARG3_TP() | |
247 | #define __BPF_TP(...) \ | |
248 | __trace_printk(1 /* Fake ip will not be printed. */, \ | |
249 | fmt, ##__VA_ARGS__) | |
250 | ||
251 | #define __BPF_ARG1_TP(...) \ | |
252 | ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \ | |
253 | ? __BPF_TP(arg1, ##__VA_ARGS__) \ | |
254 | : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \ | |
255 | ? __BPF_TP((long)arg1, ##__VA_ARGS__) \ | |
256 | : __BPF_TP((u32)arg1, ##__VA_ARGS__))) | |
257 | ||
258 | #define __BPF_ARG2_TP(...) \ | |
259 | ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \ | |
260 | ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \ | |
261 | : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \ | |
262 | ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \ | |
263 | : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__))) | |
264 | ||
265 | #define __BPF_ARG3_TP(...) \ | |
266 | ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \ | |
267 | ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \ | |
268 | : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \ | |
269 | ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \ | |
270 | : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__))) | |
271 | ||
272 | return __BPF_TP_EMIT(); | |
9c959c86 AS |
273 | } |
274 | ||
275 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
276 | .func = bpf_trace_printk, | |
277 | .gpl_only = true, | |
278 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
279 | .arg1_type = ARG_PTR_TO_MEM, |
280 | .arg2_type = ARG_CONST_SIZE, | |
9c959c86 AS |
281 | }; |
282 | ||
0756ea3e AS |
283 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
284 | { | |
285 | /* | |
286 | * this program might be calling bpf_trace_printk, | |
287 | * so allocate per-cpu printk buffers | |
288 | */ | |
289 | trace_printk_init_buffers(); | |
290 | ||
291 | return &bpf_trace_printk_proto; | |
292 | } | |
293 | ||
908432ca YS |
294 | static __always_inline int |
295 | get_map_perf_counter(struct bpf_map *map, u64 flags, | |
296 | u64 *value, u64 *enabled, u64 *running) | |
35578d79 | 297 | { |
35578d79 | 298 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
6816a7ff DB |
299 | unsigned int cpu = smp_processor_id(); |
300 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 301 | struct bpf_event_entry *ee; |
35578d79 | 302 | |
6816a7ff DB |
303 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
304 | return -EINVAL; | |
305 | if (index == BPF_F_CURRENT_CPU) | |
306 | index = cpu; | |
35578d79 KX |
307 | if (unlikely(index >= array->map.max_entries)) |
308 | return -E2BIG; | |
309 | ||
3b1efb19 | 310 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 311 | if (!ee) |
35578d79 KX |
312 | return -ENOENT; |
313 | ||
908432ca YS |
314 | return perf_event_read_local(ee->event, value, enabled, running); |
315 | } | |
316 | ||
317 | BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) | |
318 | { | |
319 | u64 value = 0; | |
320 | int err; | |
321 | ||
322 | err = get_map_perf_counter(map, flags, &value, NULL, NULL); | |
35578d79 | 323 | /* |
f91840a3 AS |
324 | * this api is ugly since we miss [-22..-2] range of valid |
325 | * counter values, but that's uapi | |
35578d79 | 326 | */ |
f91840a3 AS |
327 | if (err) |
328 | return err; | |
329 | return value; | |
35578d79 KX |
330 | } |
331 | ||
62544ce8 | 332 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 333 | .func = bpf_perf_event_read, |
1075ef59 | 334 | .gpl_only = true, |
35578d79 KX |
335 | .ret_type = RET_INTEGER, |
336 | .arg1_type = ARG_CONST_MAP_PTR, | |
337 | .arg2_type = ARG_ANYTHING, | |
338 | }; | |
339 | ||
908432ca YS |
340 | BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, |
341 | struct bpf_perf_event_value *, buf, u32, size) | |
342 | { | |
343 | int err = -EINVAL; | |
344 | ||
345 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
346 | goto clear; | |
347 | err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, | |
348 | &buf->running); | |
349 | if (unlikely(err)) | |
350 | goto clear; | |
351 | return 0; | |
352 | clear: | |
353 | memset(buf, 0, size); | |
354 | return err; | |
355 | } | |
356 | ||
357 | static const struct bpf_func_proto bpf_perf_event_read_value_proto = { | |
358 | .func = bpf_perf_event_read_value, | |
359 | .gpl_only = true, | |
360 | .ret_type = RET_INTEGER, | |
361 | .arg1_type = ARG_CONST_MAP_PTR, | |
362 | .arg2_type = ARG_ANYTHING, | |
363 | .arg3_type = ARG_PTR_TO_UNINIT_MEM, | |
364 | .arg4_type = ARG_CONST_SIZE, | |
365 | }; | |
366 | ||
283ca526 | 367 | static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd); |
20b9d7ac | 368 | |
8e7a3920 DB |
369 | static __always_inline u64 |
370 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
283ca526 | 371 | u64 flags, struct perf_sample_data *sd) |
a43eec30 | 372 | { |
a43eec30 | 373 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
d7931330 | 374 | unsigned int cpu = smp_processor_id(); |
1e33759c | 375 | u64 index = flags & BPF_F_INDEX_MASK; |
3b1efb19 | 376 | struct bpf_event_entry *ee; |
a43eec30 | 377 | struct perf_event *event; |
a43eec30 | 378 | |
1e33759c | 379 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 380 | index = cpu; |
a43eec30 AS |
381 | if (unlikely(index >= array->map.max_entries)) |
382 | return -E2BIG; | |
383 | ||
3b1efb19 | 384 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 385 | if (!ee) |
a43eec30 AS |
386 | return -ENOENT; |
387 | ||
3b1efb19 | 388 | event = ee->event; |
a43eec30 AS |
389 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
390 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
391 | return -EINVAL; | |
392 | ||
d7931330 | 393 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
394 | return -EOPNOTSUPP; |
395 | ||
20b9d7ac | 396 | perf_event_output(event, sd, regs); |
a43eec30 AS |
397 | return 0; |
398 | } | |
399 | ||
f3694e00 DB |
400 | BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, |
401 | u64, flags, void *, data, u64, size) | |
8e7a3920 | 402 | { |
283ca526 | 403 | struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd); |
8e7a3920 DB |
404 | struct perf_raw_record raw = { |
405 | .frag = { | |
406 | .size = size, | |
407 | .data = data, | |
408 | }, | |
409 | }; | |
410 | ||
411 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) | |
412 | return -EINVAL; | |
413 | ||
283ca526 DB |
414 | perf_sample_data_init(sd, 0, 0); |
415 | sd->raw = &raw; | |
416 | ||
417 | return __bpf_perf_event_output(regs, map, flags, sd); | |
8e7a3920 DB |
418 | } |
419 | ||
a43eec30 AS |
420 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
421 | .func = bpf_perf_event_output, | |
1075ef59 | 422 | .gpl_only = true, |
a43eec30 AS |
423 | .ret_type = RET_INTEGER, |
424 | .arg1_type = ARG_PTR_TO_CTX, | |
425 | .arg2_type = ARG_CONST_MAP_PTR, | |
426 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 427 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 428 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
a43eec30 AS |
429 | }; |
430 | ||
bd570ff9 | 431 | static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); |
283ca526 | 432 | static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd); |
bd570ff9 | 433 | |
555c8a86 DB |
434 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
435 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 436 | { |
283ca526 | 437 | struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd); |
bd570ff9 | 438 | struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); |
555c8a86 DB |
439 | struct perf_raw_frag frag = { |
440 | .copy = ctx_copy, | |
441 | .size = ctx_size, | |
442 | .data = ctx, | |
443 | }; | |
444 | struct perf_raw_record raw = { | |
445 | .frag = { | |
183fc153 AM |
446 | { |
447 | .next = ctx_size ? &frag : NULL, | |
448 | }, | |
555c8a86 DB |
449 | .size = meta_size, |
450 | .data = meta, | |
451 | }, | |
452 | }; | |
bd570ff9 DB |
453 | |
454 | perf_fetch_caller_regs(regs); | |
283ca526 DB |
455 | perf_sample_data_init(sd, 0, 0); |
456 | sd->raw = &raw; | |
bd570ff9 | 457 | |
283ca526 | 458 | return __bpf_perf_event_output(regs, map, flags, sd); |
bd570ff9 DB |
459 | } |
460 | ||
f3694e00 | 461 | BPF_CALL_0(bpf_get_current_task) |
606274c5 AS |
462 | { |
463 | return (long) current; | |
464 | } | |
465 | ||
466 | static const struct bpf_func_proto bpf_get_current_task_proto = { | |
467 | .func = bpf_get_current_task, | |
468 | .gpl_only = true, | |
469 | .ret_type = RET_INTEGER, | |
470 | }; | |
471 | ||
f3694e00 | 472 | BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) |
60d20f91 | 473 | { |
60d20f91 SD |
474 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
475 | struct cgroup *cgrp; | |
60d20f91 SD |
476 | |
477 | if (unlikely(in_interrupt())) | |
478 | return -EINVAL; | |
60d20f91 SD |
479 | if (unlikely(idx >= array->map.max_entries)) |
480 | return -E2BIG; | |
481 | ||
482 | cgrp = READ_ONCE(array->ptrs[idx]); | |
483 | if (unlikely(!cgrp)) | |
484 | return -EAGAIN; | |
485 | ||
486 | return task_under_cgroup_hierarchy(current, cgrp); | |
487 | } | |
488 | ||
489 | static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { | |
490 | .func = bpf_current_task_under_cgroup, | |
491 | .gpl_only = false, | |
492 | .ret_type = RET_INTEGER, | |
493 | .arg1_type = ARG_CONST_MAP_PTR, | |
494 | .arg2_type = ARG_ANYTHING, | |
495 | }; | |
496 | ||
a5e8c070 GB |
497 | BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size, |
498 | const void *, unsafe_ptr) | |
499 | { | |
500 | int ret; | |
501 | ||
502 | /* | |
503 | * The strncpy_from_unsafe() call will likely not fill the entire | |
504 | * buffer, but that's okay in this circumstance as we're probing | |
505 | * arbitrary memory anyway similar to bpf_probe_read() and might | |
506 | * as well probe the stack. Thus, memory is explicitly cleared | |
507 | * only in error case, so that improper users ignoring return | |
508 | * code altogether don't copy garbage; otherwise length of string | |
509 | * is returned that can be used for bpf_perf_event_output() et al. | |
510 | */ | |
511 | ret = strncpy_from_unsafe(dst, unsafe_ptr, size); | |
512 | if (unlikely(ret < 0)) | |
513 | memset(dst, 0, size); | |
514 | ||
515 | return ret; | |
516 | } | |
517 | ||
518 | static const struct bpf_func_proto bpf_probe_read_str_proto = { | |
519 | .func = bpf_probe_read_str, | |
520 | .gpl_only = true, | |
521 | .ret_type = RET_INTEGER, | |
522 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
5c4e1201 | 523 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, |
a5e8c070 GB |
524 | .arg3_type = ARG_ANYTHING, |
525 | }; | |
526 | ||
9fd82b61 | 527 | static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) |
2541517c AS |
528 | { |
529 | switch (func_id) { | |
530 | case BPF_FUNC_map_lookup_elem: | |
531 | return &bpf_map_lookup_elem_proto; | |
532 | case BPF_FUNC_map_update_elem: | |
533 | return &bpf_map_update_elem_proto; | |
534 | case BPF_FUNC_map_delete_elem: | |
535 | return &bpf_map_delete_elem_proto; | |
536 | case BPF_FUNC_probe_read: | |
537 | return &bpf_probe_read_proto; | |
d9847d31 AS |
538 | case BPF_FUNC_ktime_get_ns: |
539 | return &bpf_ktime_get_ns_proto; | |
04fd61ab AS |
540 | case BPF_FUNC_tail_call: |
541 | return &bpf_tail_call_proto; | |
ffeedafb AS |
542 | case BPF_FUNC_get_current_pid_tgid: |
543 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
544 | case BPF_FUNC_get_current_task: |
545 | return &bpf_get_current_task_proto; | |
ffeedafb AS |
546 | case BPF_FUNC_get_current_uid_gid: |
547 | return &bpf_get_current_uid_gid_proto; | |
548 | case BPF_FUNC_get_current_comm: | |
549 | return &bpf_get_current_comm_proto; | |
9c959c86 | 550 | case BPF_FUNC_trace_printk: |
0756ea3e | 551 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
552 | case BPF_FUNC_get_smp_processor_id: |
553 | return &bpf_get_smp_processor_id_proto; | |
2d0e30c3 DB |
554 | case BPF_FUNC_get_numa_node_id: |
555 | return &bpf_get_numa_node_id_proto; | |
35578d79 KX |
556 | case BPF_FUNC_perf_event_read: |
557 | return &bpf_perf_event_read_proto; | |
96ae5227 SD |
558 | case BPF_FUNC_probe_write_user: |
559 | return bpf_get_probe_write_proto(); | |
60d20f91 SD |
560 | case BPF_FUNC_current_task_under_cgroup: |
561 | return &bpf_current_task_under_cgroup_proto; | |
8937bd80 AS |
562 | case BPF_FUNC_get_prandom_u32: |
563 | return &bpf_get_prandom_u32_proto; | |
a5e8c070 GB |
564 | case BPF_FUNC_probe_read_str: |
565 | return &bpf_probe_read_str_proto; | |
9fd82b61 AS |
566 | default: |
567 | return NULL; | |
568 | } | |
569 | } | |
570 | ||
571 | static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) | |
572 | { | |
573 | switch (func_id) { | |
a43eec30 AS |
574 | case BPF_FUNC_perf_event_output: |
575 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
576 | case BPF_FUNC_get_stackid: |
577 | return &bpf_get_stackid_proto; | |
908432ca YS |
578 | case BPF_FUNC_perf_event_read_value: |
579 | return &bpf_perf_event_read_value_proto; | |
9802d865 JB |
580 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
581 | case BPF_FUNC_override_return: | |
582 | return &bpf_override_return_proto; | |
583 | #endif | |
2541517c | 584 | default: |
9fd82b61 | 585 | return tracing_func_proto(func_id); |
2541517c AS |
586 | } |
587 | } | |
588 | ||
589 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 | 590 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 591 | struct bpf_insn_access_aux *info) |
2541517c | 592 | { |
2541517c AS |
593 | if (off < 0 || off >= sizeof(struct pt_regs)) |
594 | return false; | |
2541517c AS |
595 | if (type != BPF_READ) |
596 | return false; | |
2541517c AS |
597 | if (off % size != 0) |
598 | return false; | |
2d071c64 DB |
599 | /* |
600 | * Assertion for 32 bit to make sure last 8 byte access | |
601 | * (BPF_DW) to the last 4 byte member is disallowed. | |
602 | */ | |
603 | if (off + size > sizeof(struct pt_regs)) | |
604 | return false; | |
605 | ||
2541517c AS |
606 | return true; |
607 | } | |
608 | ||
7de16e3a | 609 | const struct bpf_verifier_ops kprobe_verifier_ops = { |
2541517c AS |
610 | .get_func_proto = kprobe_prog_func_proto, |
611 | .is_valid_access = kprobe_prog_is_valid_access, | |
612 | }; | |
613 | ||
7de16e3a JK |
614 | const struct bpf_prog_ops kprobe_prog_ops = { |
615 | }; | |
616 | ||
f3694e00 DB |
617 | BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, |
618 | u64, flags, void *, data, u64, size) | |
9940d67c | 619 | { |
f3694e00 DB |
620 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
621 | ||
9940d67c AS |
622 | /* |
623 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
624 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
f3694e00 | 625 | * from there and call the same bpf_perf_event_output() helper inline. |
9940d67c | 626 | */ |
f3694e00 | 627 | return ____bpf_perf_event_output(regs, map, flags, data, size); |
9940d67c AS |
628 | } |
629 | ||
630 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
631 | .func = bpf_perf_event_output_tp, | |
632 | .gpl_only = true, | |
633 | .ret_type = RET_INTEGER, | |
634 | .arg1_type = ARG_PTR_TO_CTX, | |
635 | .arg2_type = ARG_CONST_MAP_PTR, | |
636 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 637 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 638 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
9940d67c AS |
639 | }; |
640 | ||
f3694e00 DB |
641 | BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, |
642 | u64, flags) | |
9940d67c | 643 | { |
f3694e00 | 644 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
9940d67c | 645 | |
f3694e00 DB |
646 | /* |
647 | * Same comment as in bpf_perf_event_output_tp(), only that this time | |
648 | * the other helper's function body cannot be inlined due to being | |
649 | * external, thus we need to call raw helper function. | |
650 | */ | |
651 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
652 | flags, 0, 0); | |
9940d67c AS |
653 | } |
654 | ||
655 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
656 | .func = bpf_get_stackid_tp, | |
657 | .gpl_only = true, | |
658 | .ret_type = RET_INTEGER, | |
659 | .arg1_type = ARG_PTR_TO_CTX, | |
660 | .arg2_type = ARG_CONST_MAP_PTR, | |
661 | .arg3_type = ARG_ANYTHING, | |
662 | }; | |
663 | ||
4bebdc7a YS |
664 | BPF_CALL_3(bpf_perf_prog_read_value_tp, struct bpf_perf_event_data_kern *, ctx, |
665 | struct bpf_perf_event_value *, buf, u32, size) | |
666 | { | |
667 | int err = -EINVAL; | |
668 | ||
669 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
670 | goto clear; | |
671 | err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, | |
672 | &buf->running); | |
673 | if (unlikely(err)) | |
674 | goto clear; | |
675 | return 0; | |
676 | clear: | |
677 | memset(buf, 0, size); | |
678 | return err; | |
679 | } | |
680 | ||
681 | static const struct bpf_func_proto bpf_perf_prog_read_value_proto_tp = { | |
682 | .func = bpf_perf_prog_read_value_tp, | |
683 | .gpl_only = true, | |
684 | .ret_type = RET_INTEGER, | |
685 | .arg1_type = ARG_PTR_TO_CTX, | |
686 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
687 | .arg3_type = ARG_CONST_SIZE, | |
688 | }; | |
689 | ||
9fd82b61 AS |
690 | static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) |
691 | { | |
692 | switch (func_id) { | |
693 | case BPF_FUNC_perf_event_output: | |
9940d67c | 694 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 695 | case BPF_FUNC_get_stackid: |
9940d67c | 696 | return &bpf_get_stackid_proto_tp; |
4bebdc7a YS |
697 | case BPF_FUNC_perf_prog_read_value: |
698 | return &bpf_perf_prog_read_value_proto_tp; | |
9fd82b61 AS |
699 | default: |
700 | return tracing_func_proto(func_id); | |
701 | } | |
702 | } | |
703 | ||
19de99f7 | 704 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 705 | struct bpf_insn_access_aux *info) |
9fd82b61 AS |
706 | { |
707 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
708 | return false; | |
709 | if (type != BPF_READ) | |
710 | return false; | |
711 | if (off % size != 0) | |
712 | return false; | |
2d071c64 DB |
713 | |
714 | BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); | |
9fd82b61 AS |
715 | return true; |
716 | } | |
717 | ||
7de16e3a | 718 | const struct bpf_verifier_ops tracepoint_verifier_ops = { |
9fd82b61 AS |
719 | .get_func_proto = tp_prog_func_proto, |
720 | .is_valid_access = tp_prog_is_valid_access, | |
721 | }; | |
722 | ||
7de16e3a JK |
723 | const struct bpf_prog_ops tracepoint_prog_ops = { |
724 | }; | |
725 | ||
0515e599 | 726 | static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 727 | struct bpf_insn_access_aux *info) |
0515e599 | 728 | { |
f96da094 DB |
729 | const int size_sp = FIELD_SIZEOF(struct bpf_perf_event_data, |
730 | sample_period); | |
31fd8581 | 731 | |
0515e599 AS |
732 | if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) |
733 | return false; | |
734 | if (type != BPF_READ) | |
735 | return false; | |
736 | if (off % size != 0) | |
737 | return false; | |
31fd8581 | 738 | |
f96da094 DB |
739 | switch (off) { |
740 | case bpf_ctx_range(struct bpf_perf_event_data, sample_period): | |
741 | bpf_ctx_record_field_size(info, size_sp); | |
742 | if (!bpf_ctx_narrow_access_ok(off, size, size_sp)) | |
23994631 | 743 | return false; |
f96da094 DB |
744 | break; |
745 | default: | |
0515e599 AS |
746 | if (size != sizeof(long)) |
747 | return false; | |
748 | } | |
f96da094 | 749 | |
0515e599 AS |
750 | return true; |
751 | } | |
752 | ||
6b8cc1d1 DB |
753 | static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, |
754 | const struct bpf_insn *si, | |
0515e599 | 755 | struct bpf_insn *insn_buf, |
f96da094 | 756 | struct bpf_prog *prog, u32 *target_size) |
0515e599 AS |
757 | { |
758 | struct bpf_insn *insn = insn_buf; | |
759 | ||
6b8cc1d1 | 760 | switch (si->off) { |
0515e599 | 761 | case offsetof(struct bpf_perf_event_data, sample_period): |
f035a515 | 762 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 763 | data), si->dst_reg, si->src_reg, |
0515e599 | 764 | offsetof(struct bpf_perf_event_data_kern, data)); |
6b8cc1d1 | 765 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, |
f96da094 DB |
766 | bpf_target_off(struct perf_sample_data, period, 8, |
767 | target_size)); | |
0515e599 AS |
768 | break; |
769 | default: | |
f035a515 | 770 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 771 | regs), si->dst_reg, si->src_reg, |
0515e599 | 772 | offsetof(struct bpf_perf_event_data_kern, regs)); |
6b8cc1d1 DB |
773 | *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, |
774 | si->off); | |
0515e599 AS |
775 | break; |
776 | } | |
777 | ||
778 | return insn - insn_buf; | |
779 | } | |
780 | ||
7de16e3a | 781 | const struct bpf_verifier_ops perf_event_verifier_ops = { |
0515e599 AS |
782 | .get_func_proto = tp_prog_func_proto, |
783 | .is_valid_access = pe_prog_is_valid_access, | |
784 | .convert_ctx_access = pe_prog_convert_ctx_access, | |
785 | }; | |
7de16e3a JK |
786 | |
787 | const struct bpf_prog_ops perf_event_prog_ops = { | |
788 | }; | |
e87c6bc3 YS |
789 | |
790 | static DEFINE_MUTEX(bpf_event_mutex); | |
791 | ||
c8c088ba YS |
792 | #define BPF_TRACE_MAX_PROGS 64 |
793 | ||
e87c6bc3 YS |
794 | int perf_event_attach_bpf_prog(struct perf_event *event, |
795 | struct bpf_prog *prog) | |
796 | { | |
797 | struct bpf_prog_array __rcu *old_array; | |
798 | struct bpf_prog_array *new_array; | |
799 | int ret = -EEXIST; | |
800 | ||
9802d865 | 801 | /* |
b4da3340 MH |
802 | * Kprobe override only works if they are on the function entry, |
803 | * and only if they are on the opt-in list. | |
9802d865 JB |
804 | */ |
805 | if (prog->kprobe_override && | |
b4da3340 | 806 | (!trace_kprobe_on_func_entry(event->tp_event) || |
9802d865 JB |
807 | !trace_kprobe_error_injectable(event->tp_event))) |
808 | return -EINVAL; | |
809 | ||
e87c6bc3 YS |
810 | mutex_lock(&bpf_event_mutex); |
811 | ||
812 | if (event->prog) | |
07c41a29 | 813 | goto unlock; |
e87c6bc3 | 814 | |
07c41a29 | 815 | old_array = event->tp_event->prog_array; |
c8c088ba YS |
816 | if (old_array && |
817 | bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { | |
818 | ret = -E2BIG; | |
819 | goto unlock; | |
820 | } | |
821 | ||
e87c6bc3 YS |
822 | ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array); |
823 | if (ret < 0) | |
07c41a29 | 824 | goto unlock; |
e87c6bc3 YS |
825 | |
826 | /* set the new array to event->tp_event and set event->prog */ | |
827 | event->prog = prog; | |
828 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
829 | bpf_prog_array_free(old_array); | |
830 | ||
07c41a29 | 831 | unlock: |
e87c6bc3 YS |
832 | mutex_unlock(&bpf_event_mutex); |
833 | return ret; | |
834 | } | |
835 | ||
836 | void perf_event_detach_bpf_prog(struct perf_event *event) | |
837 | { | |
838 | struct bpf_prog_array __rcu *old_array; | |
839 | struct bpf_prog_array *new_array; | |
840 | int ret; | |
841 | ||
842 | mutex_lock(&bpf_event_mutex); | |
843 | ||
844 | if (!event->prog) | |
07c41a29 | 845 | goto unlock; |
e87c6bc3 | 846 | |
07c41a29 | 847 | old_array = event->tp_event->prog_array; |
e87c6bc3 YS |
848 | ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array); |
849 | if (ret < 0) { | |
850 | bpf_prog_array_delete_safe(old_array, event->prog); | |
851 | } else { | |
852 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
853 | bpf_prog_array_free(old_array); | |
854 | } | |
855 | ||
856 | bpf_prog_put(event->prog); | |
857 | event->prog = NULL; | |
858 | ||
07c41a29 | 859 | unlock: |
e87c6bc3 YS |
860 | mutex_unlock(&bpf_event_mutex); |
861 | } | |
f371b304 | 862 | |
f4e2298e | 863 | int perf_event_query_prog_array(struct perf_event *event, void __user *info) |
f371b304 YS |
864 | { |
865 | struct perf_event_query_bpf __user *uquery = info; | |
866 | struct perf_event_query_bpf query = {}; | |
867 | int ret; | |
868 | ||
869 | if (!capable(CAP_SYS_ADMIN)) | |
870 | return -EPERM; | |
871 | if (event->attr.type != PERF_TYPE_TRACEPOINT) | |
872 | return -EINVAL; | |
873 | if (copy_from_user(&query, uquery, sizeof(query))) | |
874 | return -EFAULT; | |
875 | ||
876 | mutex_lock(&bpf_event_mutex); | |
877 | ret = bpf_prog_array_copy_info(event->tp_event->prog_array, | |
878 | uquery->ids, | |
879 | query.ids_len, | |
880 | &uquery->prog_cnt); | |
881 | mutex_unlock(&bpf_event_mutex); | |
882 | ||
883 | return ret; | |
884 | } |