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 | 16 | #include <linux/kprobes.h> |
41bdc4b4 | 17 | #include <linux/syscalls.h> |
540adea3 | 18 | #include <linux/error-injection.h> |
9802d865 JB |
19 | |
20 | #include "trace_probe.h" | |
2541517c AS |
21 | #include "trace.h" |
22 | ||
035226b9 | 23 | u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
c195651e | 24 | u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
035226b9 | 25 | |
2541517c AS |
26 | /** |
27 | * trace_call_bpf - invoke BPF program | |
e87c6bc3 | 28 | * @call: tracepoint event |
2541517c AS |
29 | * @ctx: opaque context pointer |
30 | * | |
31 | * kprobe handlers execute BPF programs via this helper. | |
32 | * Can be used from static tracepoints in the future. | |
33 | * | |
34 | * Return: BPF programs always return an integer which is interpreted by | |
35 | * kprobe handler as: | |
36 | * 0 - return from kprobe (event is filtered out) | |
37 | * 1 - store kprobe event into ring buffer | |
38 | * Other values are reserved and currently alias to 1 | |
39 | */ | |
e87c6bc3 | 40 | unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) |
2541517c AS |
41 | { |
42 | unsigned int ret; | |
43 | ||
44 | if (in_nmi()) /* not supported yet */ | |
45 | return 1; | |
46 | ||
47 | preempt_disable(); | |
48 | ||
49 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
50 | /* | |
51 | * since some bpf program is already running on this cpu, | |
52 | * don't call into another bpf program (same or different) | |
53 | * and don't send kprobe event into ring-buffer, | |
54 | * so return zero here | |
55 | */ | |
56 | ret = 0; | |
57 | goto out; | |
58 | } | |
59 | ||
e87c6bc3 YS |
60 | /* |
61 | * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock | |
62 | * to all call sites, we did a bpf_prog_array_valid() there to check | |
63 | * whether call->prog_array is empty or not, which is | |
64 | * a heurisitc to speed up execution. | |
65 | * | |
66 | * If bpf_prog_array_valid() fetched prog_array was | |
67 | * non-NULL, we go into trace_call_bpf() and do the actual | |
68 | * proper rcu_dereference() under RCU lock. | |
69 | * If it turns out that prog_array is NULL then, we bail out. | |
70 | * For the opposite, if the bpf_prog_array_valid() fetched pointer | |
71 | * was NULL, you'll skip the prog_array with the risk of missing | |
72 | * out of events when it was updated in between this and the | |
73 | * rcu_dereference() which is accepted risk. | |
74 | */ | |
75 | ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN); | |
2541517c AS |
76 | |
77 | out: | |
78 | __this_cpu_dec(bpf_prog_active); | |
79 | preempt_enable(); | |
80 | ||
81 | return ret; | |
82 | } | |
83 | EXPORT_SYMBOL_GPL(trace_call_bpf); | |
84 | ||
9802d865 JB |
85 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
86 | BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) | |
87 | { | |
9802d865 | 88 | regs_set_return_value(regs, rc); |
540adea3 | 89 | override_function_with_return(regs); |
9802d865 JB |
90 | return 0; |
91 | } | |
92 | ||
93 | static const struct bpf_func_proto bpf_override_return_proto = { | |
94 | .func = bpf_override_return, | |
95 | .gpl_only = true, | |
96 | .ret_type = RET_INTEGER, | |
97 | .arg1_type = ARG_PTR_TO_CTX, | |
98 | .arg2_type = ARG_ANYTHING, | |
99 | }; | |
100 | #endif | |
101 | ||
f3694e00 | 102 | BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr) |
2541517c | 103 | { |
eb33f2cc | 104 | int ret; |
2541517c | 105 | |
074f528e DB |
106 | ret = probe_kernel_read(dst, unsafe_ptr, size); |
107 | if (unlikely(ret < 0)) | |
108 | memset(dst, 0, size); | |
109 | ||
110 | return ret; | |
2541517c AS |
111 | } |
112 | ||
113 | static const struct bpf_func_proto bpf_probe_read_proto = { | |
114 | .func = bpf_probe_read, | |
115 | .gpl_only = true, | |
116 | .ret_type = RET_INTEGER, | |
39f19ebb | 117 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
9c019e2b | 118 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, |
2541517c AS |
119 | .arg3_type = ARG_ANYTHING, |
120 | }; | |
121 | ||
f3694e00 DB |
122 | BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src, |
123 | u32, size) | |
96ae5227 | 124 | { |
96ae5227 SD |
125 | /* |
126 | * Ensure we're in user context which is safe for the helper to | |
127 | * run. This helper has no business in a kthread. | |
128 | * | |
129 | * access_ok() should prevent writing to non-user memory, but in | |
130 | * some situations (nommu, temporary switch, etc) access_ok() does | |
131 | * not provide enough validation, hence the check on KERNEL_DS. | |
132 | */ | |
133 | ||
134 | if (unlikely(in_interrupt() || | |
135 | current->flags & (PF_KTHREAD | PF_EXITING))) | |
136 | return -EPERM; | |
db68ce10 | 137 | if (unlikely(uaccess_kernel())) |
96ae5227 SD |
138 | return -EPERM; |
139 | if (!access_ok(VERIFY_WRITE, unsafe_ptr, size)) | |
140 | return -EPERM; | |
141 | ||
142 | return probe_kernel_write(unsafe_ptr, src, size); | |
143 | } | |
144 | ||
145 | static const struct bpf_func_proto bpf_probe_write_user_proto = { | |
146 | .func = bpf_probe_write_user, | |
147 | .gpl_only = true, | |
148 | .ret_type = RET_INTEGER, | |
149 | .arg1_type = ARG_ANYTHING, | |
39f19ebb AS |
150 | .arg2_type = ARG_PTR_TO_MEM, |
151 | .arg3_type = ARG_CONST_SIZE, | |
96ae5227 SD |
152 | }; |
153 | ||
154 | static const struct bpf_func_proto *bpf_get_probe_write_proto(void) | |
155 | { | |
156 | pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", | |
157 | current->comm, task_pid_nr(current)); | |
158 | ||
159 | return &bpf_probe_write_user_proto; | |
160 | } | |
161 | ||
9c959c86 | 162 | /* |
7bda4b40 JF |
163 | * Only limited trace_printk() conversion specifiers allowed: |
164 | * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s | |
9c959c86 | 165 | */ |
f3694e00 DB |
166 | BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, |
167 | u64, arg2, u64, arg3) | |
9c959c86 | 168 | { |
8d3b7dce | 169 | bool str_seen = false; |
9c959c86 AS |
170 | int mod[3] = {}; |
171 | int fmt_cnt = 0; | |
8d3b7dce AS |
172 | u64 unsafe_addr; |
173 | char buf[64]; | |
9c959c86 AS |
174 | int i; |
175 | ||
176 | /* | |
177 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
178 | * guarantees that fmt points to bpf program stack, | |
179 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
180 | */ | |
181 | if (fmt[--fmt_size] != 0) | |
182 | return -EINVAL; | |
183 | ||
184 | /* check format string for allowed specifiers */ | |
185 | for (i = 0; i < fmt_size; i++) { | |
186 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) | |
187 | return -EINVAL; | |
188 | ||
189 | if (fmt[i] != '%') | |
190 | continue; | |
191 | ||
192 | if (fmt_cnt >= 3) | |
193 | return -EINVAL; | |
194 | ||
195 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
196 | i++; | |
197 | if (fmt[i] == 'l') { | |
198 | mod[fmt_cnt]++; | |
199 | i++; | |
8d3b7dce | 200 | } else if (fmt[i] == 'p' || fmt[i] == 's') { |
9c959c86 AS |
201 | mod[fmt_cnt]++; |
202 | i++; | |
203 | if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0) | |
204 | return -EINVAL; | |
205 | fmt_cnt++; | |
8d3b7dce AS |
206 | if (fmt[i - 1] == 's') { |
207 | if (str_seen) | |
208 | /* allow only one '%s' per fmt string */ | |
209 | return -EINVAL; | |
210 | str_seen = true; | |
211 | ||
212 | switch (fmt_cnt) { | |
213 | case 1: | |
f3694e00 DB |
214 | unsafe_addr = arg1; |
215 | arg1 = (long) buf; | |
8d3b7dce AS |
216 | break; |
217 | case 2: | |
f3694e00 DB |
218 | unsafe_addr = arg2; |
219 | arg2 = (long) buf; | |
8d3b7dce AS |
220 | break; |
221 | case 3: | |
f3694e00 DB |
222 | unsafe_addr = arg3; |
223 | arg3 = (long) buf; | |
8d3b7dce AS |
224 | break; |
225 | } | |
226 | buf[0] = 0; | |
227 | strncpy_from_unsafe(buf, | |
228 | (void *) (long) unsafe_addr, | |
229 | sizeof(buf)); | |
230 | } | |
9c959c86 AS |
231 | continue; |
232 | } | |
233 | ||
234 | if (fmt[i] == 'l') { | |
235 | mod[fmt_cnt]++; | |
236 | i++; | |
237 | } | |
238 | ||
7bda4b40 JF |
239 | if (fmt[i] != 'i' && fmt[i] != 'd' && |
240 | fmt[i] != 'u' && fmt[i] != 'x') | |
9c959c86 AS |
241 | return -EINVAL; |
242 | fmt_cnt++; | |
243 | } | |
244 | ||
88a5c690 DB |
245 | /* Horrid workaround for getting va_list handling working with different |
246 | * argument type combinations generically for 32 and 64 bit archs. | |
247 | */ | |
248 | #define __BPF_TP_EMIT() __BPF_ARG3_TP() | |
249 | #define __BPF_TP(...) \ | |
eefa864a | 250 | __trace_printk(0 /* Fake ip */, \ |
88a5c690 DB |
251 | fmt, ##__VA_ARGS__) |
252 | ||
253 | #define __BPF_ARG1_TP(...) \ | |
254 | ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \ | |
255 | ? __BPF_TP(arg1, ##__VA_ARGS__) \ | |
256 | : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \ | |
257 | ? __BPF_TP((long)arg1, ##__VA_ARGS__) \ | |
258 | : __BPF_TP((u32)arg1, ##__VA_ARGS__))) | |
259 | ||
260 | #define __BPF_ARG2_TP(...) \ | |
261 | ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \ | |
262 | ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \ | |
263 | : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \ | |
264 | ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \ | |
265 | : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__))) | |
266 | ||
267 | #define __BPF_ARG3_TP(...) \ | |
268 | ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \ | |
269 | ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \ | |
270 | : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \ | |
271 | ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \ | |
272 | : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__))) | |
273 | ||
274 | return __BPF_TP_EMIT(); | |
9c959c86 AS |
275 | } |
276 | ||
277 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
278 | .func = bpf_trace_printk, | |
279 | .gpl_only = true, | |
280 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
281 | .arg1_type = ARG_PTR_TO_MEM, |
282 | .arg2_type = ARG_CONST_SIZE, | |
9c959c86 AS |
283 | }; |
284 | ||
0756ea3e AS |
285 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
286 | { | |
287 | /* | |
288 | * this program might be calling bpf_trace_printk, | |
289 | * so allocate per-cpu printk buffers | |
290 | */ | |
291 | trace_printk_init_buffers(); | |
292 | ||
293 | return &bpf_trace_printk_proto; | |
294 | } | |
295 | ||
908432ca YS |
296 | static __always_inline int |
297 | get_map_perf_counter(struct bpf_map *map, u64 flags, | |
298 | u64 *value, u64 *enabled, u64 *running) | |
35578d79 | 299 | { |
35578d79 | 300 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
6816a7ff DB |
301 | unsigned int cpu = smp_processor_id(); |
302 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 303 | struct bpf_event_entry *ee; |
35578d79 | 304 | |
6816a7ff DB |
305 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
306 | return -EINVAL; | |
307 | if (index == BPF_F_CURRENT_CPU) | |
308 | index = cpu; | |
35578d79 KX |
309 | if (unlikely(index >= array->map.max_entries)) |
310 | return -E2BIG; | |
311 | ||
3b1efb19 | 312 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 313 | if (!ee) |
35578d79 KX |
314 | return -ENOENT; |
315 | ||
908432ca YS |
316 | return perf_event_read_local(ee->event, value, enabled, running); |
317 | } | |
318 | ||
319 | BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) | |
320 | { | |
321 | u64 value = 0; | |
322 | int err; | |
323 | ||
324 | err = get_map_perf_counter(map, flags, &value, NULL, NULL); | |
35578d79 | 325 | /* |
f91840a3 AS |
326 | * this api is ugly since we miss [-22..-2] range of valid |
327 | * counter values, but that's uapi | |
35578d79 | 328 | */ |
f91840a3 AS |
329 | if (err) |
330 | return err; | |
331 | return value; | |
35578d79 KX |
332 | } |
333 | ||
62544ce8 | 334 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 335 | .func = bpf_perf_event_read, |
1075ef59 | 336 | .gpl_only = true, |
35578d79 KX |
337 | .ret_type = RET_INTEGER, |
338 | .arg1_type = ARG_CONST_MAP_PTR, | |
339 | .arg2_type = ARG_ANYTHING, | |
340 | }; | |
341 | ||
908432ca YS |
342 | BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, |
343 | struct bpf_perf_event_value *, buf, u32, size) | |
344 | { | |
345 | int err = -EINVAL; | |
346 | ||
347 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
348 | goto clear; | |
349 | err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, | |
350 | &buf->running); | |
351 | if (unlikely(err)) | |
352 | goto clear; | |
353 | return 0; | |
354 | clear: | |
355 | memset(buf, 0, size); | |
356 | return err; | |
357 | } | |
358 | ||
359 | static const struct bpf_func_proto bpf_perf_event_read_value_proto = { | |
360 | .func = bpf_perf_event_read_value, | |
361 | .gpl_only = true, | |
362 | .ret_type = RET_INTEGER, | |
363 | .arg1_type = ARG_CONST_MAP_PTR, | |
364 | .arg2_type = ARG_ANYTHING, | |
365 | .arg3_type = ARG_PTR_TO_UNINIT_MEM, | |
366 | .arg4_type = ARG_CONST_SIZE, | |
367 | }; | |
368 | ||
283ca526 | 369 | static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd); |
20b9d7ac | 370 | |
8e7a3920 DB |
371 | static __always_inline u64 |
372 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
283ca526 | 373 | u64 flags, struct perf_sample_data *sd) |
a43eec30 | 374 | { |
a43eec30 | 375 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
d7931330 | 376 | unsigned int cpu = smp_processor_id(); |
1e33759c | 377 | u64 index = flags & BPF_F_INDEX_MASK; |
3b1efb19 | 378 | struct bpf_event_entry *ee; |
a43eec30 | 379 | struct perf_event *event; |
a43eec30 | 380 | |
1e33759c | 381 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 382 | index = cpu; |
a43eec30 AS |
383 | if (unlikely(index >= array->map.max_entries)) |
384 | return -E2BIG; | |
385 | ||
3b1efb19 | 386 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 387 | if (!ee) |
a43eec30 AS |
388 | return -ENOENT; |
389 | ||
3b1efb19 | 390 | event = ee->event; |
a43eec30 AS |
391 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
392 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
393 | return -EINVAL; | |
394 | ||
d7931330 | 395 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
396 | return -EOPNOTSUPP; |
397 | ||
20b9d7ac | 398 | perf_event_output(event, sd, regs); |
a43eec30 AS |
399 | return 0; |
400 | } | |
401 | ||
f3694e00 DB |
402 | BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, |
403 | u64, flags, void *, data, u64, size) | |
8e7a3920 | 404 | { |
283ca526 | 405 | struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd); |
8e7a3920 DB |
406 | struct perf_raw_record raw = { |
407 | .frag = { | |
408 | .size = size, | |
409 | .data = data, | |
410 | }, | |
411 | }; | |
412 | ||
413 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) | |
414 | return -EINVAL; | |
415 | ||
283ca526 DB |
416 | perf_sample_data_init(sd, 0, 0); |
417 | sd->raw = &raw; | |
418 | ||
419 | return __bpf_perf_event_output(regs, map, flags, sd); | |
8e7a3920 DB |
420 | } |
421 | ||
a43eec30 AS |
422 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
423 | .func = bpf_perf_event_output, | |
1075ef59 | 424 | .gpl_only = true, |
a43eec30 AS |
425 | .ret_type = RET_INTEGER, |
426 | .arg1_type = ARG_PTR_TO_CTX, | |
427 | .arg2_type = ARG_CONST_MAP_PTR, | |
428 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 429 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 430 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
a43eec30 AS |
431 | }; |
432 | ||
bd570ff9 | 433 | static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); |
283ca526 | 434 | static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd); |
bd570ff9 | 435 | |
555c8a86 DB |
436 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
437 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 438 | { |
283ca526 | 439 | struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd); |
bd570ff9 | 440 | struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); |
555c8a86 DB |
441 | struct perf_raw_frag frag = { |
442 | .copy = ctx_copy, | |
443 | .size = ctx_size, | |
444 | .data = ctx, | |
445 | }; | |
446 | struct perf_raw_record raw = { | |
447 | .frag = { | |
183fc153 AM |
448 | { |
449 | .next = ctx_size ? &frag : NULL, | |
450 | }, | |
555c8a86 DB |
451 | .size = meta_size, |
452 | .data = meta, | |
453 | }, | |
454 | }; | |
bd570ff9 DB |
455 | |
456 | perf_fetch_caller_regs(regs); | |
283ca526 DB |
457 | perf_sample_data_init(sd, 0, 0); |
458 | sd->raw = &raw; | |
bd570ff9 | 459 | |
283ca526 | 460 | return __bpf_perf_event_output(regs, map, flags, sd); |
bd570ff9 DB |
461 | } |
462 | ||
f3694e00 | 463 | BPF_CALL_0(bpf_get_current_task) |
606274c5 AS |
464 | { |
465 | return (long) current; | |
466 | } | |
467 | ||
468 | static const struct bpf_func_proto bpf_get_current_task_proto = { | |
469 | .func = bpf_get_current_task, | |
470 | .gpl_only = true, | |
471 | .ret_type = RET_INTEGER, | |
472 | }; | |
473 | ||
f3694e00 | 474 | BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) |
60d20f91 | 475 | { |
60d20f91 SD |
476 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
477 | struct cgroup *cgrp; | |
60d20f91 | 478 | |
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 | ||
5e43f899 AI |
527 | static const struct bpf_func_proto * |
528 | tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
2541517c AS |
529 | { |
530 | switch (func_id) { | |
531 | case BPF_FUNC_map_lookup_elem: | |
532 | return &bpf_map_lookup_elem_proto; | |
533 | case BPF_FUNC_map_update_elem: | |
534 | return &bpf_map_update_elem_proto; | |
535 | case BPF_FUNC_map_delete_elem: | |
536 | return &bpf_map_delete_elem_proto; | |
537 | case BPF_FUNC_probe_read: | |
538 | return &bpf_probe_read_proto; | |
d9847d31 AS |
539 | case BPF_FUNC_ktime_get_ns: |
540 | return &bpf_ktime_get_ns_proto; | |
04fd61ab AS |
541 | case BPF_FUNC_tail_call: |
542 | return &bpf_tail_call_proto; | |
ffeedafb AS |
543 | case BPF_FUNC_get_current_pid_tgid: |
544 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
545 | case BPF_FUNC_get_current_task: |
546 | return &bpf_get_current_task_proto; | |
ffeedafb AS |
547 | case BPF_FUNC_get_current_uid_gid: |
548 | return &bpf_get_current_uid_gid_proto; | |
549 | case BPF_FUNC_get_current_comm: | |
550 | return &bpf_get_current_comm_proto; | |
9c959c86 | 551 | case BPF_FUNC_trace_printk: |
0756ea3e | 552 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
553 | case BPF_FUNC_get_smp_processor_id: |
554 | return &bpf_get_smp_processor_id_proto; | |
2d0e30c3 DB |
555 | case BPF_FUNC_get_numa_node_id: |
556 | return &bpf_get_numa_node_id_proto; | |
35578d79 KX |
557 | case BPF_FUNC_perf_event_read: |
558 | return &bpf_perf_event_read_proto; | |
96ae5227 SD |
559 | case BPF_FUNC_probe_write_user: |
560 | return bpf_get_probe_write_proto(); | |
60d20f91 SD |
561 | case BPF_FUNC_current_task_under_cgroup: |
562 | return &bpf_current_task_under_cgroup_proto; | |
8937bd80 AS |
563 | case BPF_FUNC_get_prandom_u32: |
564 | return &bpf_get_prandom_u32_proto; | |
a5e8c070 GB |
565 | case BPF_FUNC_probe_read_str: |
566 | return &bpf_probe_read_str_proto; | |
34ea38ca | 567 | #ifdef CONFIG_CGROUPS |
bf6fa2c8 YS |
568 | case BPF_FUNC_get_current_cgroup_id: |
569 | return &bpf_get_current_cgroup_id_proto; | |
34ea38ca | 570 | #endif |
9fd82b61 AS |
571 | default: |
572 | return NULL; | |
573 | } | |
574 | } | |
575 | ||
5e43f899 AI |
576 | static const struct bpf_func_proto * |
577 | kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
578 | { |
579 | switch (func_id) { | |
a43eec30 AS |
580 | case BPF_FUNC_perf_event_output: |
581 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
582 | case BPF_FUNC_get_stackid: |
583 | return &bpf_get_stackid_proto; | |
c195651e YS |
584 | case BPF_FUNC_get_stack: |
585 | return &bpf_get_stack_proto; | |
908432ca YS |
586 | case BPF_FUNC_perf_event_read_value: |
587 | return &bpf_perf_event_read_value_proto; | |
9802d865 JB |
588 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
589 | case BPF_FUNC_override_return: | |
590 | return &bpf_override_return_proto; | |
591 | #endif | |
2541517c | 592 | default: |
5e43f899 | 593 | return tracing_func_proto(func_id, prog); |
2541517c AS |
594 | } |
595 | } | |
596 | ||
597 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 | 598 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 599 | const struct bpf_prog *prog, |
23994631 | 600 | struct bpf_insn_access_aux *info) |
2541517c | 601 | { |
2541517c AS |
602 | if (off < 0 || off >= sizeof(struct pt_regs)) |
603 | return false; | |
2541517c AS |
604 | if (type != BPF_READ) |
605 | return false; | |
2541517c AS |
606 | if (off % size != 0) |
607 | return false; | |
2d071c64 DB |
608 | /* |
609 | * Assertion for 32 bit to make sure last 8 byte access | |
610 | * (BPF_DW) to the last 4 byte member is disallowed. | |
611 | */ | |
612 | if (off + size > sizeof(struct pt_regs)) | |
613 | return false; | |
614 | ||
2541517c AS |
615 | return true; |
616 | } | |
617 | ||
7de16e3a | 618 | const struct bpf_verifier_ops kprobe_verifier_ops = { |
2541517c AS |
619 | .get_func_proto = kprobe_prog_func_proto, |
620 | .is_valid_access = kprobe_prog_is_valid_access, | |
621 | }; | |
622 | ||
7de16e3a JK |
623 | const struct bpf_prog_ops kprobe_prog_ops = { |
624 | }; | |
625 | ||
f3694e00 DB |
626 | BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, |
627 | u64, flags, void *, data, u64, size) | |
9940d67c | 628 | { |
f3694e00 DB |
629 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
630 | ||
9940d67c AS |
631 | /* |
632 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
633 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
f3694e00 | 634 | * from there and call the same bpf_perf_event_output() helper inline. |
9940d67c | 635 | */ |
f3694e00 | 636 | return ____bpf_perf_event_output(regs, map, flags, data, size); |
9940d67c AS |
637 | } |
638 | ||
639 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
640 | .func = bpf_perf_event_output_tp, | |
641 | .gpl_only = true, | |
642 | .ret_type = RET_INTEGER, | |
643 | .arg1_type = ARG_PTR_TO_CTX, | |
644 | .arg2_type = ARG_CONST_MAP_PTR, | |
645 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 646 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 647 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
9940d67c AS |
648 | }; |
649 | ||
f3694e00 DB |
650 | BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, |
651 | u64, flags) | |
9940d67c | 652 | { |
f3694e00 | 653 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
9940d67c | 654 | |
f3694e00 DB |
655 | /* |
656 | * Same comment as in bpf_perf_event_output_tp(), only that this time | |
657 | * the other helper's function body cannot be inlined due to being | |
658 | * external, thus we need to call raw helper function. | |
659 | */ | |
660 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
661 | flags, 0, 0); | |
9940d67c AS |
662 | } |
663 | ||
664 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
665 | .func = bpf_get_stackid_tp, | |
666 | .gpl_only = true, | |
667 | .ret_type = RET_INTEGER, | |
668 | .arg1_type = ARG_PTR_TO_CTX, | |
669 | .arg2_type = ARG_CONST_MAP_PTR, | |
670 | .arg3_type = ARG_ANYTHING, | |
671 | }; | |
672 | ||
c195651e YS |
673 | BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size, |
674 | u64, flags) | |
675 | { | |
676 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; | |
677 | ||
678 | return bpf_get_stack((unsigned long) regs, (unsigned long) buf, | |
679 | (unsigned long) size, flags, 0); | |
680 | } | |
681 | ||
682 | static const struct bpf_func_proto bpf_get_stack_proto_tp = { | |
683 | .func = bpf_get_stack_tp, | |
684 | .gpl_only = true, | |
685 | .ret_type = RET_INTEGER, | |
686 | .arg1_type = ARG_PTR_TO_CTX, | |
687 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
688 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
689 | .arg4_type = ARG_ANYTHING, | |
690 | }; | |
691 | ||
5e43f899 AI |
692 | static const struct bpf_func_proto * |
693 | tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
f005afed YS |
694 | { |
695 | switch (func_id) { | |
696 | case BPF_FUNC_perf_event_output: | |
697 | return &bpf_perf_event_output_proto_tp; | |
698 | case BPF_FUNC_get_stackid: | |
699 | return &bpf_get_stackid_proto_tp; | |
c195651e YS |
700 | case BPF_FUNC_get_stack: |
701 | return &bpf_get_stack_proto_tp; | |
f005afed | 702 | default: |
5e43f899 | 703 | return tracing_func_proto(func_id, prog); |
f005afed YS |
704 | } |
705 | } | |
706 | ||
707 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, | |
5e43f899 | 708 | const struct bpf_prog *prog, |
f005afed YS |
709 | struct bpf_insn_access_aux *info) |
710 | { | |
711 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
712 | return false; | |
713 | if (type != BPF_READ) | |
714 | return false; | |
715 | if (off % size != 0) | |
716 | return false; | |
717 | ||
718 | BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); | |
719 | return true; | |
720 | } | |
721 | ||
722 | const struct bpf_verifier_ops tracepoint_verifier_ops = { | |
723 | .get_func_proto = tp_prog_func_proto, | |
724 | .is_valid_access = tp_prog_is_valid_access, | |
725 | }; | |
726 | ||
727 | const struct bpf_prog_ops tracepoint_prog_ops = { | |
728 | }; | |
729 | ||
730 | BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx, | |
4bebdc7a YS |
731 | struct bpf_perf_event_value *, buf, u32, size) |
732 | { | |
733 | int err = -EINVAL; | |
734 | ||
735 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
736 | goto clear; | |
737 | err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, | |
738 | &buf->running); | |
739 | if (unlikely(err)) | |
740 | goto clear; | |
741 | return 0; | |
742 | clear: | |
743 | memset(buf, 0, size); | |
744 | return err; | |
745 | } | |
746 | ||
f005afed YS |
747 | static const struct bpf_func_proto bpf_perf_prog_read_value_proto = { |
748 | .func = bpf_perf_prog_read_value, | |
4bebdc7a YS |
749 | .gpl_only = true, |
750 | .ret_type = RET_INTEGER, | |
751 | .arg1_type = ARG_PTR_TO_CTX, | |
752 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
753 | .arg3_type = ARG_CONST_SIZE, | |
754 | }; | |
755 | ||
5e43f899 AI |
756 | static const struct bpf_func_proto * |
757 | pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
758 | { |
759 | switch (func_id) { | |
760 | case BPF_FUNC_perf_event_output: | |
9940d67c | 761 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 762 | case BPF_FUNC_get_stackid: |
9940d67c | 763 | return &bpf_get_stackid_proto_tp; |
c195651e YS |
764 | case BPF_FUNC_get_stack: |
765 | return &bpf_get_stack_proto_tp; | |
4bebdc7a | 766 | case BPF_FUNC_perf_prog_read_value: |
f005afed | 767 | return &bpf_perf_prog_read_value_proto; |
9fd82b61 | 768 | default: |
5e43f899 | 769 | return tracing_func_proto(func_id, prog); |
9fd82b61 AS |
770 | } |
771 | } | |
772 | ||
c4f6699d AS |
773 | /* |
774 | * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp | |
775 | * to avoid potential recursive reuse issue when/if tracepoints are added | |
c195651e | 776 | * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack |
c4f6699d AS |
777 | */ |
778 | static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs); | |
779 | BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args, | |
780 | struct bpf_map *, map, u64, flags, void *, data, u64, size) | |
781 | { | |
782 | struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); | |
783 | ||
784 | perf_fetch_caller_regs(regs); | |
785 | return ____bpf_perf_event_output(regs, map, flags, data, size); | |
786 | } | |
787 | ||
788 | static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = { | |
789 | .func = bpf_perf_event_output_raw_tp, | |
790 | .gpl_only = true, | |
791 | .ret_type = RET_INTEGER, | |
792 | .arg1_type = ARG_PTR_TO_CTX, | |
793 | .arg2_type = ARG_CONST_MAP_PTR, | |
794 | .arg3_type = ARG_ANYTHING, | |
795 | .arg4_type = ARG_PTR_TO_MEM, | |
796 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, | |
797 | }; | |
798 | ||
799 | BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args, | |
800 | struct bpf_map *, map, u64, flags) | |
801 | { | |
802 | struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); | |
803 | ||
804 | perf_fetch_caller_regs(regs); | |
805 | /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */ | |
806 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
807 | flags, 0, 0); | |
808 | } | |
809 | ||
810 | static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = { | |
811 | .func = bpf_get_stackid_raw_tp, | |
812 | .gpl_only = true, | |
813 | .ret_type = RET_INTEGER, | |
814 | .arg1_type = ARG_PTR_TO_CTX, | |
815 | .arg2_type = ARG_CONST_MAP_PTR, | |
816 | .arg3_type = ARG_ANYTHING, | |
817 | }; | |
818 | ||
c195651e YS |
819 | BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args, |
820 | void *, buf, u32, size, u64, flags) | |
821 | { | |
822 | struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); | |
823 | ||
824 | perf_fetch_caller_regs(regs); | |
825 | return bpf_get_stack((unsigned long) regs, (unsigned long) buf, | |
826 | (unsigned long) size, flags, 0); | |
827 | } | |
828 | ||
829 | static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = { | |
830 | .func = bpf_get_stack_raw_tp, | |
831 | .gpl_only = true, | |
832 | .ret_type = RET_INTEGER, | |
833 | .arg1_type = ARG_PTR_TO_CTX, | |
834 | .arg2_type = ARG_PTR_TO_MEM, | |
835 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
836 | .arg4_type = ARG_ANYTHING, | |
837 | }; | |
838 | ||
5e43f899 AI |
839 | static const struct bpf_func_proto * |
840 | raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
c4f6699d AS |
841 | { |
842 | switch (func_id) { | |
843 | case BPF_FUNC_perf_event_output: | |
844 | return &bpf_perf_event_output_proto_raw_tp; | |
845 | case BPF_FUNC_get_stackid: | |
846 | return &bpf_get_stackid_proto_raw_tp; | |
c195651e YS |
847 | case BPF_FUNC_get_stack: |
848 | return &bpf_get_stack_proto_raw_tp; | |
c4f6699d | 849 | default: |
5e43f899 | 850 | return tracing_func_proto(func_id, prog); |
c4f6699d AS |
851 | } |
852 | } | |
853 | ||
854 | static bool raw_tp_prog_is_valid_access(int off, int size, | |
855 | enum bpf_access_type type, | |
5e43f899 | 856 | const struct bpf_prog *prog, |
c4f6699d AS |
857 | struct bpf_insn_access_aux *info) |
858 | { | |
859 | /* largest tracepoint in the kernel has 12 args */ | |
860 | if (off < 0 || off >= sizeof(__u64) * 12) | |
861 | return false; | |
862 | if (type != BPF_READ) | |
863 | return false; | |
864 | if (off % size != 0) | |
865 | return false; | |
866 | return true; | |
867 | } | |
868 | ||
869 | const struct bpf_verifier_ops raw_tracepoint_verifier_ops = { | |
870 | .get_func_proto = raw_tp_prog_func_proto, | |
871 | .is_valid_access = raw_tp_prog_is_valid_access, | |
872 | }; | |
873 | ||
874 | const struct bpf_prog_ops raw_tracepoint_prog_ops = { | |
875 | }; | |
876 | ||
0515e599 | 877 | static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 878 | const struct bpf_prog *prog, |
23994631 | 879 | struct bpf_insn_access_aux *info) |
0515e599 | 880 | { |
95da0cdb | 881 | const int size_u64 = sizeof(u64); |
31fd8581 | 882 | |
0515e599 AS |
883 | if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) |
884 | return false; | |
885 | if (type != BPF_READ) | |
886 | return false; | |
bc23105c DB |
887 | if (off % size != 0) { |
888 | if (sizeof(unsigned long) != 4) | |
889 | return false; | |
890 | if (size != 8) | |
891 | return false; | |
892 | if (off % size != 4) | |
893 | return false; | |
894 | } | |
31fd8581 | 895 | |
f96da094 DB |
896 | switch (off) { |
897 | case bpf_ctx_range(struct bpf_perf_event_data, sample_period): | |
95da0cdb TQ |
898 | bpf_ctx_record_field_size(info, size_u64); |
899 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
900 | return false; | |
901 | break; | |
902 | case bpf_ctx_range(struct bpf_perf_event_data, addr): | |
903 | bpf_ctx_record_field_size(info, size_u64); | |
904 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
23994631 | 905 | return false; |
f96da094 DB |
906 | break; |
907 | default: | |
0515e599 AS |
908 | if (size != sizeof(long)) |
909 | return false; | |
910 | } | |
f96da094 | 911 | |
0515e599 AS |
912 | return true; |
913 | } | |
914 | ||
6b8cc1d1 DB |
915 | static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, |
916 | const struct bpf_insn *si, | |
0515e599 | 917 | struct bpf_insn *insn_buf, |
f96da094 | 918 | struct bpf_prog *prog, u32 *target_size) |
0515e599 AS |
919 | { |
920 | struct bpf_insn *insn = insn_buf; | |
921 | ||
6b8cc1d1 | 922 | switch (si->off) { |
0515e599 | 923 | case offsetof(struct bpf_perf_event_data, sample_period): |
f035a515 | 924 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 925 | data), si->dst_reg, si->src_reg, |
0515e599 | 926 | offsetof(struct bpf_perf_event_data_kern, data)); |
6b8cc1d1 | 927 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, |
f96da094 DB |
928 | bpf_target_off(struct perf_sample_data, period, 8, |
929 | target_size)); | |
0515e599 | 930 | break; |
95da0cdb TQ |
931 | case offsetof(struct bpf_perf_event_data, addr): |
932 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, | |
933 | data), si->dst_reg, si->src_reg, | |
934 | offsetof(struct bpf_perf_event_data_kern, data)); | |
935 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, | |
936 | bpf_target_off(struct perf_sample_data, addr, 8, | |
937 | target_size)); | |
938 | break; | |
0515e599 | 939 | default: |
f035a515 | 940 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 941 | regs), si->dst_reg, si->src_reg, |
0515e599 | 942 | offsetof(struct bpf_perf_event_data_kern, regs)); |
6b8cc1d1 DB |
943 | *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, |
944 | si->off); | |
0515e599 AS |
945 | break; |
946 | } | |
947 | ||
948 | return insn - insn_buf; | |
949 | } | |
950 | ||
7de16e3a | 951 | const struct bpf_verifier_ops perf_event_verifier_ops = { |
f005afed | 952 | .get_func_proto = pe_prog_func_proto, |
0515e599 AS |
953 | .is_valid_access = pe_prog_is_valid_access, |
954 | .convert_ctx_access = pe_prog_convert_ctx_access, | |
955 | }; | |
7de16e3a JK |
956 | |
957 | const struct bpf_prog_ops perf_event_prog_ops = { | |
958 | }; | |
e87c6bc3 YS |
959 | |
960 | static DEFINE_MUTEX(bpf_event_mutex); | |
961 | ||
c8c088ba YS |
962 | #define BPF_TRACE_MAX_PROGS 64 |
963 | ||
e87c6bc3 YS |
964 | int perf_event_attach_bpf_prog(struct perf_event *event, |
965 | struct bpf_prog *prog) | |
966 | { | |
967 | struct bpf_prog_array __rcu *old_array; | |
968 | struct bpf_prog_array *new_array; | |
969 | int ret = -EEXIST; | |
970 | ||
9802d865 | 971 | /* |
b4da3340 MH |
972 | * Kprobe override only works if they are on the function entry, |
973 | * and only if they are on the opt-in list. | |
9802d865 JB |
974 | */ |
975 | if (prog->kprobe_override && | |
b4da3340 | 976 | (!trace_kprobe_on_func_entry(event->tp_event) || |
9802d865 JB |
977 | !trace_kprobe_error_injectable(event->tp_event))) |
978 | return -EINVAL; | |
979 | ||
e87c6bc3 YS |
980 | mutex_lock(&bpf_event_mutex); |
981 | ||
982 | if (event->prog) | |
07c41a29 | 983 | goto unlock; |
e87c6bc3 | 984 | |
07c41a29 | 985 | old_array = event->tp_event->prog_array; |
c8c088ba YS |
986 | if (old_array && |
987 | bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { | |
988 | ret = -E2BIG; | |
989 | goto unlock; | |
990 | } | |
991 | ||
e87c6bc3 YS |
992 | ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array); |
993 | if (ret < 0) | |
07c41a29 | 994 | goto unlock; |
e87c6bc3 YS |
995 | |
996 | /* set the new array to event->tp_event and set event->prog */ | |
997 | event->prog = prog; | |
998 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
999 | bpf_prog_array_free(old_array); | |
1000 | ||
07c41a29 | 1001 | unlock: |
e87c6bc3 YS |
1002 | mutex_unlock(&bpf_event_mutex); |
1003 | return ret; | |
1004 | } | |
1005 | ||
1006 | void perf_event_detach_bpf_prog(struct perf_event *event) | |
1007 | { | |
1008 | struct bpf_prog_array __rcu *old_array; | |
1009 | struct bpf_prog_array *new_array; | |
1010 | int ret; | |
1011 | ||
1012 | mutex_lock(&bpf_event_mutex); | |
1013 | ||
1014 | if (!event->prog) | |
07c41a29 | 1015 | goto unlock; |
e87c6bc3 | 1016 | |
07c41a29 | 1017 | old_array = event->tp_event->prog_array; |
e87c6bc3 | 1018 | ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array); |
170a7e3e SY |
1019 | if (ret == -ENOENT) |
1020 | goto unlock; | |
e87c6bc3 YS |
1021 | if (ret < 0) { |
1022 | bpf_prog_array_delete_safe(old_array, event->prog); | |
1023 | } else { | |
1024 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
1025 | bpf_prog_array_free(old_array); | |
1026 | } | |
1027 | ||
1028 | bpf_prog_put(event->prog); | |
1029 | event->prog = NULL; | |
1030 | ||
07c41a29 | 1031 | unlock: |
e87c6bc3 YS |
1032 | mutex_unlock(&bpf_event_mutex); |
1033 | } | |
f371b304 | 1034 | |
f4e2298e | 1035 | int perf_event_query_prog_array(struct perf_event *event, void __user *info) |
f371b304 YS |
1036 | { |
1037 | struct perf_event_query_bpf __user *uquery = info; | |
1038 | struct perf_event_query_bpf query = {}; | |
3a38bb98 | 1039 | u32 *ids, prog_cnt, ids_len; |
f371b304 YS |
1040 | int ret; |
1041 | ||
1042 | if (!capable(CAP_SYS_ADMIN)) | |
1043 | return -EPERM; | |
1044 | if (event->attr.type != PERF_TYPE_TRACEPOINT) | |
1045 | return -EINVAL; | |
1046 | if (copy_from_user(&query, uquery, sizeof(query))) | |
1047 | return -EFAULT; | |
3a38bb98 YS |
1048 | |
1049 | ids_len = query.ids_len; | |
1050 | if (ids_len > BPF_TRACE_MAX_PROGS) | |
9c481b90 | 1051 | return -E2BIG; |
3a38bb98 YS |
1052 | ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN); |
1053 | if (!ids) | |
1054 | return -ENOMEM; | |
1055 | /* | |
1056 | * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which | |
1057 | * is required when user only wants to check for uquery->prog_cnt. | |
1058 | * There is no need to check for it since the case is handled | |
1059 | * gracefully in bpf_prog_array_copy_info. | |
1060 | */ | |
f371b304 YS |
1061 | |
1062 | mutex_lock(&bpf_event_mutex); | |
1063 | ret = bpf_prog_array_copy_info(event->tp_event->prog_array, | |
3a38bb98 YS |
1064 | ids, |
1065 | ids_len, | |
1066 | &prog_cnt); | |
f371b304 YS |
1067 | mutex_unlock(&bpf_event_mutex); |
1068 | ||
3a38bb98 YS |
1069 | if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) || |
1070 | copy_to_user(uquery->ids, ids, ids_len * sizeof(u32))) | |
1071 | ret = -EFAULT; | |
1072 | ||
1073 | kfree(ids); | |
f371b304 YS |
1074 | return ret; |
1075 | } | |
c4f6699d AS |
1076 | |
1077 | extern struct bpf_raw_event_map __start__bpf_raw_tp[]; | |
1078 | extern struct bpf_raw_event_map __stop__bpf_raw_tp[]; | |
1079 | ||
1080 | struct bpf_raw_event_map *bpf_find_raw_tracepoint(const char *name) | |
1081 | { | |
1082 | struct bpf_raw_event_map *btp = __start__bpf_raw_tp; | |
1083 | ||
1084 | for (; btp < __stop__bpf_raw_tp; btp++) { | |
1085 | if (!strcmp(btp->tp->name, name)) | |
1086 | return btp; | |
1087 | } | |
1088 | return NULL; | |
1089 | } | |
1090 | ||
1091 | static __always_inline | |
1092 | void __bpf_trace_run(struct bpf_prog *prog, u64 *args) | |
1093 | { | |
1094 | rcu_read_lock(); | |
1095 | preempt_disable(); | |
1096 | (void) BPF_PROG_RUN(prog, args); | |
1097 | preempt_enable(); | |
1098 | rcu_read_unlock(); | |
1099 | } | |
1100 | ||
1101 | #define UNPACK(...) __VA_ARGS__ | |
1102 | #define REPEAT_1(FN, DL, X, ...) FN(X) | |
1103 | #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__) | |
1104 | #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__) | |
1105 | #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__) | |
1106 | #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__) | |
1107 | #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__) | |
1108 | #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__) | |
1109 | #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__) | |
1110 | #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__) | |
1111 | #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__) | |
1112 | #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__) | |
1113 | #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__) | |
1114 | #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__) | |
1115 | ||
1116 | #define SARG(X) u64 arg##X | |
1117 | #define COPY(X) args[X] = arg##X | |
1118 | ||
1119 | #define __DL_COM (,) | |
1120 | #define __DL_SEM (;) | |
1121 | ||
1122 | #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 | |
1123 | ||
1124 | #define BPF_TRACE_DEFN_x(x) \ | |
1125 | void bpf_trace_run##x(struct bpf_prog *prog, \ | |
1126 | REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \ | |
1127 | { \ | |
1128 | u64 args[x]; \ | |
1129 | REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \ | |
1130 | __bpf_trace_run(prog, args); \ | |
1131 | } \ | |
1132 | EXPORT_SYMBOL_GPL(bpf_trace_run##x) | |
1133 | BPF_TRACE_DEFN_x(1); | |
1134 | BPF_TRACE_DEFN_x(2); | |
1135 | BPF_TRACE_DEFN_x(3); | |
1136 | BPF_TRACE_DEFN_x(4); | |
1137 | BPF_TRACE_DEFN_x(5); | |
1138 | BPF_TRACE_DEFN_x(6); | |
1139 | BPF_TRACE_DEFN_x(7); | |
1140 | BPF_TRACE_DEFN_x(8); | |
1141 | BPF_TRACE_DEFN_x(9); | |
1142 | BPF_TRACE_DEFN_x(10); | |
1143 | BPF_TRACE_DEFN_x(11); | |
1144 | BPF_TRACE_DEFN_x(12); | |
1145 | ||
1146 | static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1147 | { | |
1148 | struct tracepoint *tp = btp->tp; | |
1149 | ||
1150 | /* | |
1151 | * check that program doesn't access arguments beyond what's | |
1152 | * available in this tracepoint | |
1153 | */ | |
1154 | if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64)) | |
1155 | return -EINVAL; | |
1156 | ||
1157 | return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog); | |
1158 | } | |
1159 | ||
1160 | int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1161 | { | |
1162 | int err; | |
1163 | ||
1164 | mutex_lock(&bpf_event_mutex); | |
1165 | err = __bpf_probe_register(btp, prog); | |
1166 | mutex_unlock(&bpf_event_mutex); | |
1167 | return err; | |
1168 | } | |
1169 | ||
1170 | int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1171 | { | |
1172 | int err; | |
1173 | ||
1174 | mutex_lock(&bpf_event_mutex); | |
1175 | err = tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog); | |
1176 | mutex_unlock(&bpf_event_mutex); | |
1177 | return err; | |
1178 | } | |
41bdc4b4 YS |
1179 | |
1180 | int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id, | |
1181 | u32 *fd_type, const char **buf, | |
1182 | u64 *probe_offset, u64 *probe_addr) | |
1183 | { | |
1184 | bool is_tracepoint, is_syscall_tp; | |
1185 | struct bpf_prog *prog; | |
1186 | int flags, err = 0; | |
1187 | ||
1188 | prog = event->prog; | |
1189 | if (!prog) | |
1190 | return -ENOENT; | |
1191 | ||
1192 | /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */ | |
1193 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) | |
1194 | return -EOPNOTSUPP; | |
1195 | ||
1196 | *prog_id = prog->aux->id; | |
1197 | flags = event->tp_event->flags; | |
1198 | is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT; | |
1199 | is_syscall_tp = is_syscall_trace_event(event->tp_event); | |
1200 | ||
1201 | if (is_tracepoint || is_syscall_tp) { | |
1202 | *buf = is_tracepoint ? event->tp_event->tp->name | |
1203 | : event->tp_event->name; | |
1204 | *fd_type = BPF_FD_TYPE_TRACEPOINT; | |
1205 | *probe_offset = 0x0; | |
1206 | *probe_addr = 0x0; | |
1207 | } else { | |
1208 | /* kprobe/uprobe */ | |
1209 | err = -EOPNOTSUPP; | |
1210 | #ifdef CONFIG_KPROBE_EVENTS | |
1211 | if (flags & TRACE_EVENT_FL_KPROBE) | |
1212 | err = bpf_get_kprobe_info(event, fd_type, buf, | |
1213 | probe_offset, probe_addr, | |
1214 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1215 | #endif | |
1216 | #ifdef CONFIG_UPROBE_EVENTS | |
1217 | if (flags & TRACE_EVENT_FL_UPROBE) | |
1218 | err = bpf_get_uprobe_info(event, fd_type, buf, | |
1219 | probe_offset, | |
1220 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1221 | #endif | |
1222 | } | |
1223 | ||
1224 | return err; | |
1225 | } |