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> |
2541517c AS |
16 | #include "trace.h" |
17 | ||
2541517c AS |
18 | /** |
19 | * trace_call_bpf - invoke BPF program | |
20 | * @prog: BPF program | |
21 | * @ctx: opaque context pointer | |
22 | * | |
23 | * kprobe handlers execute BPF programs via this helper. | |
24 | * Can be used from static tracepoints in the future. | |
25 | * | |
26 | * Return: BPF programs always return an integer which is interpreted by | |
27 | * kprobe handler as: | |
28 | * 0 - return from kprobe (event is filtered out) | |
29 | * 1 - store kprobe event into ring buffer | |
30 | * Other values are reserved and currently alias to 1 | |
31 | */ | |
32 | unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx) | |
33 | { | |
34 | unsigned int ret; | |
35 | ||
36 | if (in_nmi()) /* not supported yet */ | |
37 | return 1; | |
38 | ||
39 | preempt_disable(); | |
40 | ||
41 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
42 | /* | |
43 | * since some bpf program is already running on this cpu, | |
44 | * don't call into another bpf program (same or different) | |
45 | * and don't send kprobe event into ring-buffer, | |
46 | * so return zero here | |
47 | */ | |
48 | ret = 0; | |
49 | goto out; | |
50 | } | |
51 | ||
52 | rcu_read_lock(); | |
53 | ret = BPF_PROG_RUN(prog, ctx); | |
54 | rcu_read_unlock(); | |
55 | ||
56 | out: | |
57 | __this_cpu_dec(bpf_prog_active); | |
58 | preempt_enable(); | |
59 | ||
60 | return ret; | |
61 | } | |
62 | EXPORT_SYMBOL_GPL(trace_call_bpf); | |
63 | ||
f3694e00 | 64 | BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr) |
2541517c | 65 | { |
f3694e00 | 66 | int ret; |
2541517c | 67 | |
074f528e DB |
68 | ret = probe_kernel_read(dst, unsafe_ptr, size); |
69 | if (unlikely(ret < 0)) | |
70 | memset(dst, 0, size); | |
71 | ||
72 | return ret; | |
2541517c AS |
73 | } |
74 | ||
75 | static const struct bpf_func_proto bpf_probe_read_proto = { | |
76 | .func = bpf_probe_read, | |
77 | .gpl_only = true, | |
78 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
79 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
80 | .arg2_type = ARG_CONST_SIZE, | |
2541517c AS |
81 | .arg3_type = ARG_ANYTHING, |
82 | }; | |
83 | ||
f3694e00 DB |
84 | BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src, |
85 | u32, size) | |
96ae5227 | 86 | { |
96ae5227 SD |
87 | /* |
88 | * Ensure we're in user context which is safe for the helper to | |
89 | * run. This helper has no business in a kthread. | |
90 | * | |
91 | * access_ok() should prevent writing to non-user memory, but in | |
92 | * some situations (nommu, temporary switch, etc) access_ok() does | |
93 | * not provide enough validation, hence the check on KERNEL_DS. | |
94 | */ | |
95 | ||
96 | if (unlikely(in_interrupt() || | |
97 | current->flags & (PF_KTHREAD | PF_EXITING))) | |
98 | return -EPERM; | |
db68ce10 | 99 | if (unlikely(uaccess_kernel())) |
96ae5227 SD |
100 | return -EPERM; |
101 | if (!access_ok(VERIFY_WRITE, unsafe_ptr, size)) | |
102 | return -EPERM; | |
103 | ||
104 | return probe_kernel_write(unsafe_ptr, src, size); | |
105 | } | |
106 | ||
107 | static const struct bpf_func_proto bpf_probe_write_user_proto = { | |
108 | .func = bpf_probe_write_user, | |
109 | .gpl_only = true, | |
110 | .ret_type = RET_INTEGER, | |
111 | .arg1_type = ARG_ANYTHING, | |
39f19ebb AS |
112 | .arg2_type = ARG_PTR_TO_MEM, |
113 | .arg3_type = ARG_CONST_SIZE, | |
96ae5227 SD |
114 | }; |
115 | ||
116 | static const struct bpf_func_proto *bpf_get_probe_write_proto(void) | |
117 | { | |
118 | pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", | |
119 | current->comm, task_pid_nr(current)); | |
120 | ||
121 | return &bpf_probe_write_user_proto; | |
122 | } | |
123 | ||
9c959c86 | 124 | /* |
7bda4b40 JF |
125 | * Only limited trace_printk() conversion specifiers allowed: |
126 | * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s | |
9c959c86 | 127 | */ |
f3694e00 DB |
128 | BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, |
129 | u64, arg2, u64, arg3) | |
9c959c86 | 130 | { |
8d3b7dce | 131 | bool str_seen = false; |
9c959c86 AS |
132 | int mod[3] = {}; |
133 | int fmt_cnt = 0; | |
8d3b7dce AS |
134 | u64 unsafe_addr; |
135 | char buf[64]; | |
9c959c86 AS |
136 | int i; |
137 | ||
138 | /* | |
139 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
140 | * guarantees that fmt points to bpf program stack, | |
141 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
142 | */ | |
143 | if (fmt[--fmt_size] != 0) | |
144 | return -EINVAL; | |
145 | ||
146 | /* check format string for allowed specifiers */ | |
147 | for (i = 0; i < fmt_size; i++) { | |
148 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) | |
149 | return -EINVAL; | |
150 | ||
151 | if (fmt[i] != '%') | |
152 | continue; | |
153 | ||
154 | if (fmt_cnt >= 3) | |
155 | return -EINVAL; | |
156 | ||
157 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
158 | i++; | |
159 | if (fmt[i] == 'l') { | |
160 | mod[fmt_cnt]++; | |
161 | i++; | |
8d3b7dce | 162 | } else if (fmt[i] == 'p' || fmt[i] == 's') { |
9c959c86 AS |
163 | mod[fmt_cnt]++; |
164 | i++; | |
165 | if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0) | |
166 | return -EINVAL; | |
167 | fmt_cnt++; | |
8d3b7dce AS |
168 | if (fmt[i - 1] == 's') { |
169 | if (str_seen) | |
170 | /* allow only one '%s' per fmt string */ | |
171 | return -EINVAL; | |
172 | str_seen = true; | |
173 | ||
174 | switch (fmt_cnt) { | |
175 | case 1: | |
f3694e00 DB |
176 | unsafe_addr = arg1; |
177 | arg1 = (long) buf; | |
8d3b7dce AS |
178 | break; |
179 | case 2: | |
f3694e00 DB |
180 | unsafe_addr = arg2; |
181 | arg2 = (long) buf; | |
8d3b7dce AS |
182 | break; |
183 | case 3: | |
f3694e00 DB |
184 | unsafe_addr = arg3; |
185 | arg3 = (long) buf; | |
8d3b7dce AS |
186 | break; |
187 | } | |
188 | buf[0] = 0; | |
189 | strncpy_from_unsafe(buf, | |
190 | (void *) (long) unsafe_addr, | |
191 | sizeof(buf)); | |
192 | } | |
9c959c86 AS |
193 | continue; |
194 | } | |
195 | ||
196 | if (fmt[i] == 'l') { | |
197 | mod[fmt_cnt]++; | |
198 | i++; | |
199 | } | |
200 | ||
7bda4b40 JF |
201 | if (fmt[i] != 'i' && fmt[i] != 'd' && |
202 | fmt[i] != 'u' && fmt[i] != 'x') | |
9c959c86 AS |
203 | return -EINVAL; |
204 | fmt_cnt++; | |
205 | } | |
206 | ||
88a5c690 DB |
207 | /* Horrid workaround for getting va_list handling working with different |
208 | * argument type combinations generically for 32 and 64 bit archs. | |
209 | */ | |
210 | #define __BPF_TP_EMIT() __BPF_ARG3_TP() | |
211 | #define __BPF_TP(...) \ | |
212 | __trace_printk(1 /* Fake ip will not be printed. */, \ | |
213 | fmt, ##__VA_ARGS__) | |
214 | ||
215 | #define __BPF_ARG1_TP(...) \ | |
216 | ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \ | |
217 | ? __BPF_TP(arg1, ##__VA_ARGS__) \ | |
218 | : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \ | |
219 | ? __BPF_TP((long)arg1, ##__VA_ARGS__) \ | |
220 | : __BPF_TP((u32)arg1, ##__VA_ARGS__))) | |
221 | ||
222 | #define __BPF_ARG2_TP(...) \ | |
223 | ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \ | |
224 | ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \ | |
225 | : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \ | |
226 | ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \ | |
227 | : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__))) | |
228 | ||
229 | #define __BPF_ARG3_TP(...) \ | |
230 | ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \ | |
231 | ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \ | |
232 | : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \ | |
233 | ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \ | |
234 | : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__))) | |
235 | ||
236 | return __BPF_TP_EMIT(); | |
9c959c86 AS |
237 | } |
238 | ||
239 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
240 | .func = bpf_trace_printk, | |
241 | .gpl_only = true, | |
242 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
243 | .arg1_type = ARG_PTR_TO_MEM, |
244 | .arg2_type = ARG_CONST_SIZE, | |
9c959c86 AS |
245 | }; |
246 | ||
0756ea3e AS |
247 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
248 | { | |
249 | /* | |
250 | * this program might be calling bpf_trace_printk, | |
251 | * so allocate per-cpu printk buffers | |
252 | */ | |
253 | trace_printk_init_buffers(); | |
254 | ||
255 | return &bpf_trace_printk_proto; | |
256 | } | |
257 | ||
f3694e00 | 258 | BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) |
35578d79 | 259 | { |
35578d79 | 260 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
6816a7ff DB |
261 | unsigned int cpu = smp_processor_id(); |
262 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 263 | struct bpf_event_entry *ee; |
f91840a3 AS |
264 | u64 value = 0; |
265 | int err; | |
35578d79 | 266 | |
6816a7ff DB |
267 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
268 | return -EINVAL; | |
269 | if (index == BPF_F_CURRENT_CPU) | |
270 | index = cpu; | |
35578d79 KX |
271 | if (unlikely(index >= array->map.max_entries)) |
272 | return -E2BIG; | |
273 | ||
3b1efb19 | 274 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 275 | if (!ee) |
35578d79 KX |
276 | return -ENOENT; |
277 | ||
97562633 | 278 | err = perf_event_read_local(ee->event, &value, NULL, NULL); |
35578d79 | 279 | /* |
f91840a3 AS |
280 | * this api is ugly since we miss [-22..-2] range of valid |
281 | * counter values, but that's uapi | |
35578d79 | 282 | */ |
f91840a3 AS |
283 | if (err) |
284 | return err; | |
285 | return value; | |
35578d79 KX |
286 | } |
287 | ||
62544ce8 | 288 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 289 | .func = bpf_perf_event_read, |
1075ef59 | 290 | .gpl_only = true, |
35578d79 KX |
291 | .ret_type = RET_INTEGER, |
292 | .arg1_type = ARG_CONST_MAP_PTR, | |
293 | .arg2_type = ARG_ANYTHING, | |
294 | }; | |
295 | ||
20b9d7ac DB |
296 | static DEFINE_PER_CPU(struct perf_sample_data, bpf_sd); |
297 | ||
8e7a3920 DB |
298 | static __always_inline u64 |
299 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
300 | u64 flags, struct perf_raw_record *raw) | |
a43eec30 | 301 | { |
a43eec30 | 302 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
20b9d7ac | 303 | struct perf_sample_data *sd = this_cpu_ptr(&bpf_sd); |
d7931330 | 304 | unsigned int cpu = smp_processor_id(); |
1e33759c | 305 | u64 index = flags & BPF_F_INDEX_MASK; |
3b1efb19 | 306 | struct bpf_event_entry *ee; |
a43eec30 | 307 | struct perf_event *event; |
a43eec30 | 308 | |
1e33759c | 309 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 310 | index = cpu; |
a43eec30 AS |
311 | if (unlikely(index >= array->map.max_entries)) |
312 | return -E2BIG; | |
313 | ||
3b1efb19 | 314 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 315 | if (!ee) |
a43eec30 AS |
316 | return -ENOENT; |
317 | ||
3b1efb19 | 318 | event = ee->event; |
a43eec30 AS |
319 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
320 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
321 | return -EINVAL; | |
322 | ||
d7931330 | 323 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
324 | return -EOPNOTSUPP; |
325 | ||
20b9d7ac DB |
326 | perf_sample_data_init(sd, 0, 0); |
327 | sd->raw = raw; | |
328 | perf_event_output(event, sd, regs); | |
a43eec30 AS |
329 | return 0; |
330 | } | |
331 | ||
f3694e00 DB |
332 | BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, |
333 | u64, flags, void *, data, u64, size) | |
8e7a3920 | 334 | { |
8e7a3920 DB |
335 | struct perf_raw_record raw = { |
336 | .frag = { | |
337 | .size = size, | |
338 | .data = data, | |
339 | }, | |
340 | }; | |
341 | ||
342 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) | |
343 | return -EINVAL; | |
344 | ||
345 | return __bpf_perf_event_output(regs, map, flags, &raw); | |
346 | } | |
347 | ||
a43eec30 AS |
348 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
349 | .func = bpf_perf_event_output, | |
1075ef59 | 350 | .gpl_only = true, |
a43eec30 AS |
351 | .ret_type = RET_INTEGER, |
352 | .arg1_type = ARG_PTR_TO_CTX, | |
353 | .arg2_type = ARG_CONST_MAP_PTR, | |
354 | .arg3_type = ARG_ANYTHING, | |
39f19ebb AS |
355 | .arg4_type = ARG_PTR_TO_MEM, |
356 | .arg5_type = ARG_CONST_SIZE, | |
a43eec30 AS |
357 | }; |
358 | ||
bd570ff9 DB |
359 | static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); |
360 | ||
555c8a86 DB |
361 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
362 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 DB |
363 | { |
364 | struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); | |
555c8a86 DB |
365 | struct perf_raw_frag frag = { |
366 | .copy = ctx_copy, | |
367 | .size = ctx_size, | |
368 | .data = ctx, | |
369 | }; | |
370 | struct perf_raw_record raw = { | |
371 | .frag = { | |
183fc153 AM |
372 | { |
373 | .next = ctx_size ? &frag : NULL, | |
374 | }, | |
555c8a86 DB |
375 | .size = meta_size, |
376 | .data = meta, | |
377 | }, | |
378 | }; | |
bd570ff9 DB |
379 | |
380 | perf_fetch_caller_regs(regs); | |
381 | ||
555c8a86 | 382 | return __bpf_perf_event_output(regs, map, flags, &raw); |
bd570ff9 DB |
383 | } |
384 | ||
f3694e00 | 385 | BPF_CALL_0(bpf_get_current_task) |
606274c5 AS |
386 | { |
387 | return (long) current; | |
388 | } | |
389 | ||
390 | static const struct bpf_func_proto bpf_get_current_task_proto = { | |
391 | .func = bpf_get_current_task, | |
392 | .gpl_only = true, | |
393 | .ret_type = RET_INTEGER, | |
394 | }; | |
395 | ||
f3694e00 | 396 | BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) |
60d20f91 | 397 | { |
60d20f91 SD |
398 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
399 | struct cgroup *cgrp; | |
60d20f91 SD |
400 | |
401 | if (unlikely(in_interrupt())) | |
402 | return -EINVAL; | |
60d20f91 SD |
403 | if (unlikely(idx >= array->map.max_entries)) |
404 | return -E2BIG; | |
405 | ||
406 | cgrp = READ_ONCE(array->ptrs[idx]); | |
407 | if (unlikely(!cgrp)) | |
408 | return -EAGAIN; | |
409 | ||
410 | return task_under_cgroup_hierarchy(current, cgrp); | |
411 | } | |
412 | ||
413 | static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { | |
414 | .func = bpf_current_task_under_cgroup, | |
415 | .gpl_only = false, | |
416 | .ret_type = RET_INTEGER, | |
417 | .arg1_type = ARG_CONST_MAP_PTR, | |
418 | .arg2_type = ARG_ANYTHING, | |
419 | }; | |
420 | ||
a5e8c070 GB |
421 | BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size, |
422 | const void *, unsafe_ptr) | |
423 | { | |
424 | int ret; | |
425 | ||
426 | /* | |
427 | * The strncpy_from_unsafe() call will likely not fill the entire | |
428 | * buffer, but that's okay in this circumstance as we're probing | |
429 | * arbitrary memory anyway similar to bpf_probe_read() and might | |
430 | * as well probe the stack. Thus, memory is explicitly cleared | |
431 | * only in error case, so that improper users ignoring return | |
432 | * code altogether don't copy garbage; otherwise length of string | |
433 | * is returned that can be used for bpf_perf_event_output() et al. | |
434 | */ | |
435 | ret = strncpy_from_unsafe(dst, unsafe_ptr, size); | |
436 | if (unlikely(ret < 0)) | |
437 | memset(dst, 0, size); | |
438 | ||
439 | return ret; | |
440 | } | |
441 | ||
442 | static const struct bpf_func_proto bpf_probe_read_str_proto = { | |
443 | .func = bpf_probe_read_str, | |
444 | .gpl_only = true, | |
445 | .ret_type = RET_INTEGER, | |
446 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
447 | .arg2_type = ARG_CONST_SIZE, | |
448 | .arg3_type = ARG_ANYTHING, | |
449 | }; | |
450 | ||
9fd82b61 | 451 | static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) |
2541517c AS |
452 | { |
453 | switch (func_id) { | |
454 | case BPF_FUNC_map_lookup_elem: | |
455 | return &bpf_map_lookup_elem_proto; | |
456 | case BPF_FUNC_map_update_elem: | |
457 | return &bpf_map_update_elem_proto; | |
458 | case BPF_FUNC_map_delete_elem: | |
459 | return &bpf_map_delete_elem_proto; | |
460 | case BPF_FUNC_probe_read: | |
461 | return &bpf_probe_read_proto; | |
d9847d31 AS |
462 | case BPF_FUNC_ktime_get_ns: |
463 | return &bpf_ktime_get_ns_proto; | |
04fd61ab AS |
464 | case BPF_FUNC_tail_call: |
465 | return &bpf_tail_call_proto; | |
ffeedafb AS |
466 | case BPF_FUNC_get_current_pid_tgid: |
467 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
468 | case BPF_FUNC_get_current_task: |
469 | return &bpf_get_current_task_proto; | |
ffeedafb AS |
470 | case BPF_FUNC_get_current_uid_gid: |
471 | return &bpf_get_current_uid_gid_proto; | |
472 | case BPF_FUNC_get_current_comm: | |
473 | return &bpf_get_current_comm_proto; | |
9c959c86 | 474 | case BPF_FUNC_trace_printk: |
0756ea3e | 475 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
476 | case BPF_FUNC_get_smp_processor_id: |
477 | return &bpf_get_smp_processor_id_proto; | |
2d0e30c3 DB |
478 | case BPF_FUNC_get_numa_node_id: |
479 | return &bpf_get_numa_node_id_proto; | |
35578d79 KX |
480 | case BPF_FUNC_perf_event_read: |
481 | return &bpf_perf_event_read_proto; | |
96ae5227 SD |
482 | case BPF_FUNC_probe_write_user: |
483 | return bpf_get_probe_write_proto(); | |
60d20f91 SD |
484 | case BPF_FUNC_current_task_under_cgroup: |
485 | return &bpf_current_task_under_cgroup_proto; | |
8937bd80 AS |
486 | case BPF_FUNC_get_prandom_u32: |
487 | return &bpf_get_prandom_u32_proto; | |
a5e8c070 GB |
488 | case BPF_FUNC_probe_read_str: |
489 | return &bpf_probe_read_str_proto; | |
9fd82b61 AS |
490 | default: |
491 | return NULL; | |
492 | } | |
493 | } | |
494 | ||
495 | static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) | |
496 | { | |
497 | switch (func_id) { | |
a43eec30 AS |
498 | case BPF_FUNC_perf_event_output: |
499 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
500 | case BPF_FUNC_get_stackid: |
501 | return &bpf_get_stackid_proto; | |
2541517c | 502 | default: |
9fd82b61 | 503 | return tracing_func_proto(func_id); |
2541517c AS |
504 | } |
505 | } | |
506 | ||
507 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 | 508 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 509 | struct bpf_insn_access_aux *info) |
2541517c | 510 | { |
2541517c AS |
511 | if (off < 0 || off >= sizeof(struct pt_regs)) |
512 | return false; | |
2541517c AS |
513 | if (type != BPF_READ) |
514 | return false; | |
2541517c AS |
515 | if (off % size != 0) |
516 | return false; | |
2d071c64 DB |
517 | /* |
518 | * Assertion for 32 bit to make sure last 8 byte access | |
519 | * (BPF_DW) to the last 4 byte member is disallowed. | |
520 | */ | |
521 | if (off + size > sizeof(struct pt_regs)) | |
522 | return false; | |
523 | ||
2541517c AS |
524 | return true; |
525 | } | |
526 | ||
be9370a7 | 527 | const struct bpf_verifier_ops kprobe_prog_ops = { |
2541517c AS |
528 | .get_func_proto = kprobe_prog_func_proto, |
529 | .is_valid_access = kprobe_prog_is_valid_access, | |
530 | }; | |
531 | ||
f3694e00 DB |
532 | BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, |
533 | u64, flags, void *, data, u64, size) | |
9940d67c | 534 | { |
f3694e00 DB |
535 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
536 | ||
9940d67c AS |
537 | /* |
538 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
539 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
f3694e00 | 540 | * from there and call the same bpf_perf_event_output() helper inline. |
9940d67c | 541 | */ |
f3694e00 | 542 | return ____bpf_perf_event_output(regs, map, flags, data, size); |
9940d67c AS |
543 | } |
544 | ||
545 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
546 | .func = bpf_perf_event_output_tp, | |
547 | .gpl_only = true, | |
548 | .ret_type = RET_INTEGER, | |
549 | .arg1_type = ARG_PTR_TO_CTX, | |
550 | .arg2_type = ARG_CONST_MAP_PTR, | |
551 | .arg3_type = ARG_ANYTHING, | |
39f19ebb AS |
552 | .arg4_type = ARG_PTR_TO_MEM, |
553 | .arg5_type = ARG_CONST_SIZE, | |
9940d67c AS |
554 | }; |
555 | ||
f3694e00 DB |
556 | BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, |
557 | u64, flags) | |
9940d67c | 558 | { |
f3694e00 | 559 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
9940d67c | 560 | |
f3694e00 DB |
561 | /* |
562 | * Same comment as in bpf_perf_event_output_tp(), only that this time | |
563 | * the other helper's function body cannot be inlined due to being | |
564 | * external, thus we need to call raw helper function. | |
565 | */ | |
566 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
567 | flags, 0, 0); | |
9940d67c AS |
568 | } |
569 | ||
570 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
571 | .func = bpf_get_stackid_tp, | |
572 | .gpl_only = true, | |
573 | .ret_type = RET_INTEGER, | |
574 | .arg1_type = ARG_PTR_TO_CTX, | |
575 | .arg2_type = ARG_CONST_MAP_PTR, | |
576 | .arg3_type = ARG_ANYTHING, | |
577 | }; | |
578 | ||
9fd82b61 AS |
579 | static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) |
580 | { | |
581 | switch (func_id) { | |
582 | case BPF_FUNC_perf_event_output: | |
9940d67c | 583 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 584 | case BPF_FUNC_get_stackid: |
9940d67c | 585 | return &bpf_get_stackid_proto_tp; |
9fd82b61 AS |
586 | default: |
587 | return tracing_func_proto(func_id); | |
588 | } | |
589 | } | |
590 | ||
19de99f7 | 591 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 592 | struct bpf_insn_access_aux *info) |
9fd82b61 AS |
593 | { |
594 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
595 | return false; | |
596 | if (type != BPF_READ) | |
597 | return false; | |
598 | if (off % size != 0) | |
599 | return false; | |
2d071c64 DB |
600 | |
601 | BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); | |
9fd82b61 AS |
602 | return true; |
603 | } | |
604 | ||
be9370a7 | 605 | const struct bpf_verifier_ops tracepoint_prog_ops = { |
9fd82b61 AS |
606 | .get_func_proto = tp_prog_func_proto, |
607 | .is_valid_access = tp_prog_is_valid_access, | |
608 | }; | |
609 | ||
0515e599 | 610 | static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
23994631 | 611 | struct bpf_insn_access_aux *info) |
0515e599 | 612 | { |
f96da094 DB |
613 | const int size_sp = FIELD_SIZEOF(struct bpf_perf_event_data, |
614 | sample_period); | |
31fd8581 | 615 | |
0515e599 AS |
616 | if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) |
617 | return false; | |
618 | if (type != BPF_READ) | |
619 | return false; | |
620 | if (off % size != 0) | |
621 | return false; | |
31fd8581 | 622 | |
f96da094 DB |
623 | switch (off) { |
624 | case bpf_ctx_range(struct bpf_perf_event_data, sample_period): | |
625 | bpf_ctx_record_field_size(info, size_sp); | |
626 | if (!bpf_ctx_narrow_access_ok(off, size, size_sp)) | |
23994631 | 627 | return false; |
f96da094 DB |
628 | break; |
629 | default: | |
0515e599 AS |
630 | if (size != sizeof(long)) |
631 | return false; | |
632 | } | |
f96da094 | 633 | |
0515e599 AS |
634 | return true; |
635 | } | |
636 | ||
6b8cc1d1 DB |
637 | static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, |
638 | const struct bpf_insn *si, | |
0515e599 | 639 | struct bpf_insn *insn_buf, |
f96da094 | 640 | struct bpf_prog *prog, u32 *target_size) |
0515e599 AS |
641 | { |
642 | struct bpf_insn *insn = insn_buf; | |
643 | ||
6b8cc1d1 | 644 | switch (si->off) { |
0515e599 | 645 | case offsetof(struct bpf_perf_event_data, sample_period): |
f035a515 | 646 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 647 | data), si->dst_reg, si->src_reg, |
0515e599 | 648 | offsetof(struct bpf_perf_event_data_kern, data)); |
6b8cc1d1 | 649 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, |
f96da094 DB |
650 | bpf_target_off(struct perf_sample_data, period, 8, |
651 | target_size)); | |
0515e599 AS |
652 | break; |
653 | default: | |
f035a515 | 654 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 655 | regs), si->dst_reg, si->src_reg, |
0515e599 | 656 | offsetof(struct bpf_perf_event_data_kern, regs)); |
6b8cc1d1 DB |
657 | *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, |
658 | si->off); | |
0515e599 AS |
659 | break; |
660 | } | |
661 | ||
662 | return insn - insn_buf; | |
663 | } | |
664 | ||
be9370a7 | 665 | const struct bpf_verifier_ops perf_event_prog_ops = { |
0515e599 AS |
666 | .get_func_proto = tp_prog_func_proto, |
667 | .is_valid_access = pe_prog_is_valid_access, | |
668 | .convert_ctx_access = pe_prog_convert_ctx_access, | |
669 | }; |