]>
Commit | Line | Data |
---|---|---|
2541517c AS |
1 | /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com |
2 | * | |
3 | * This program is free software; you can redistribute it and/or | |
4 | * modify it under the terms of version 2 of the GNU General Public | |
5 | * License as published by the Free Software Foundation. | |
6 | */ | |
7 | #include <linux/kernel.h> | |
8 | #include <linux/types.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/bpf.h> | |
11 | #include <linux/filter.h> | |
12 | #include <linux/uaccess.h> | |
9c959c86 | 13 | #include <linux/ctype.h> |
2541517c AS |
14 | #include "trace.h" |
15 | ||
2541517c AS |
16 | /** |
17 | * trace_call_bpf - invoke BPF program | |
18 | * @prog: BPF program | |
19 | * @ctx: opaque context pointer | |
20 | * | |
21 | * kprobe handlers execute BPF programs via this helper. | |
22 | * Can be used from static tracepoints in the future. | |
23 | * | |
24 | * Return: BPF programs always return an integer which is interpreted by | |
25 | * kprobe handler as: | |
26 | * 0 - return from kprobe (event is filtered out) | |
27 | * 1 - store kprobe event into ring buffer | |
28 | * Other values are reserved and currently alias to 1 | |
29 | */ | |
30 | unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx) | |
31 | { | |
32 | unsigned int ret; | |
33 | ||
34 | if (in_nmi()) /* not supported yet */ | |
35 | return 1; | |
36 | ||
37 | preempt_disable(); | |
38 | ||
39 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
40 | /* | |
41 | * since some bpf program is already running on this cpu, | |
42 | * don't call into another bpf program (same or different) | |
43 | * and don't send kprobe event into ring-buffer, | |
44 | * so return zero here | |
45 | */ | |
46 | ret = 0; | |
47 | goto out; | |
48 | } | |
49 | ||
50 | rcu_read_lock(); | |
51 | ret = BPF_PROG_RUN(prog, ctx); | |
52 | rcu_read_unlock(); | |
53 | ||
54 | out: | |
55 | __this_cpu_dec(bpf_prog_active); | |
56 | preempt_enable(); | |
57 | ||
58 | return ret; | |
59 | } | |
60 | EXPORT_SYMBOL_GPL(trace_call_bpf); | |
61 | ||
62 | static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
63 | { | |
64 | void *dst = (void *) (long) r1; | |
074f528e | 65 | int ret, size = (int) r2; |
2541517c AS |
66 | void *unsafe_ptr = (void *) (long) r3; |
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, | |
074f528e | 79 | .arg1_type = ARG_PTR_TO_RAW_STACK, |
2541517c AS |
80 | .arg2_type = ARG_CONST_STACK_SIZE, |
81 | .arg3_type = ARG_ANYTHING, | |
82 | }; | |
83 | ||
9c959c86 AS |
84 | /* |
85 | * limited trace_printk() | |
8d3b7dce | 86 | * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed |
9c959c86 AS |
87 | */ |
88 | static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5) | |
89 | { | |
90 | char *fmt = (char *) (long) r1; | |
8d3b7dce | 91 | bool str_seen = false; |
9c959c86 AS |
92 | int mod[3] = {}; |
93 | int fmt_cnt = 0; | |
8d3b7dce AS |
94 | u64 unsafe_addr; |
95 | char buf[64]; | |
9c959c86 AS |
96 | int i; |
97 | ||
98 | /* | |
99 | * bpf_check()->check_func_arg()->check_stack_boundary() | |
100 | * guarantees that fmt points to bpf program stack, | |
101 | * fmt_size bytes of it were initialized and fmt_size > 0 | |
102 | */ | |
103 | if (fmt[--fmt_size] != 0) | |
104 | return -EINVAL; | |
105 | ||
106 | /* check format string for allowed specifiers */ | |
107 | for (i = 0; i < fmt_size; i++) { | |
108 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) | |
109 | return -EINVAL; | |
110 | ||
111 | if (fmt[i] != '%') | |
112 | continue; | |
113 | ||
114 | if (fmt_cnt >= 3) | |
115 | return -EINVAL; | |
116 | ||
117 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ | |
118 | i++; | |
119 | if (fmt[i] == 'l') { | |
120 | mod[fmt_cnt]++; | |
121 | i++; | |
8d3b7dce | 122 | } else if (fmt[i] == 'p' || fmt[i] == 's') { |
9c959c86 AS |
123 | mod[fmt_cnt]++; |
124 | i++; | |
125 | if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0) | |
126 | return -EINVAL; | |
127 | fmt_cnt++; | |
8d3b7dce AS |
128 | if (fmt[i - 1] == 's') { |
129 | if (str_seen) | |
130 | /* allow only one '%s' per fmt string */ | |
131 | return -EINVAL; | |
132 | str_seen = true; | |
133 | ||
134 | switch (fmt_cnt) { | |
135 | case 1: | |
136 | unsafe_addr = r3; | |
137 | r3 = (long) buf; | |
138 | break; | |
139 | case 2: | |
140 | unsafe_addr = r4; | |
141 | r4 = (long) buf; | |
142 | break; | |
143 | case 3: | |
144 | unsafe_addr = r5; | |
145 | r5 = (long) buf; | |
146 | break; | |
147 | } | |
148 | buf[0] = 0; | |
149 | strncpy_from_unsafe(buf, | |
150 | (void *) (long) unsafe_addr, | |
151 | sizeof(buf)); | |
152 | } | |
9c959c86 AS |
153 | continue; |
154 | } | |
155 | ||
156 | if (fmt[i] == 'l') { | |
157 | mod[fmt_cnt]++; | |
158 | i++; | |
159 | } | |
160 | ||
161 | if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x') | |
162 | return -EINVAL; | |
163 | fmt_cnt++; | |
164 | } | |
165 | ||
166 | return __trace_printk(1/* fake ip will not be printed */, fmt, | |
167 | mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3, | |
168 | mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4, | |
169 | mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5); | |
170 | } | |
171 | ||
172 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
173 | .func = bpf_trace_printk, | |
174 | .gpl_only = true, | |
175 | .ret_type = RET_INTEGER, | |
176 | .arg1_type = ARG_PTR_TO_STACK, | |
177 | .arg2_type = ARG_CONST_STACK_SIZE, | |
178 | }; | |
179 | ||
0756ea3e AS |
180 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
181 | { | |
182 | /* | |
183 | * this program might be calling bpf_trace_printk, | |
184 | * so allocate per-cpu printk buffers | |
185 | */ | |
186 | trace_printk_init_buffers(); | |
187 | ||
188 | return &bpf_trace_printk_proto; | |
189 | } | |
190 | ||
6816a7ff | 191 | static u64 bpf_perf_event_read(u64 r1, u64 flags, u64 r3, u64 r4, u64 r5) |
35578d79 KX |
192 | { |
193 | struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
194 | struct bpf_array *array = container_of(map, struct bpf_array, map); | |
6816a7ff DB |
195 | unsigned int cpu = smp_processor_id(); |
196 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 197 | struct bpf_event_entry *ee; |
35578d79 KX |
198 | struct perf_event *event; |
199 | ||
6816a7ff DB |
200 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
201 | return -EINVAL; | |
202 | if (index == BPF_F_CURRENT_CPU) | |
203 | index = cpu; | |
35578d79 KX |
204 | if (unlikely(index >= array->map.max_entries)) |
205 | return -E2BIG; | |
206 | ||
3b1efb19 | 207 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 208 | if (!ee) |
35578d79 KX |
209 | return -ENOENT; |
210 | ||
3b1efb19 | 211 | event = ee->event; |
1ca1cc98 DB |
212 | if (unlikely(event->attr.type != PERF_TYPE_HARDWARE && |
213 | event->attr.type != PERF_TYPE_RAW)) | |
214 | return -EINVAL; | |
215 | ||
62544ce8 | 216 | /* make sure event is local and doesn't have pmu::count */ |
6816a7ff | 217 | if (unlikely(event->oncpu != cpu || event->pmu->count)) |
62544ce8 AS |
218 | return -EINVAL; |
219 | ||
35578d79 KX |
220 | /* |
221 | * we don't know if the function is run successfully by the | |
222 | * return value. It can be judged in other places, such as | |
223 | * eBPF programs. | |
224 | */ | |
225 | return perf_event_read_local(event); | |
226 | } | |
227 | ||
62544ce8 | 228 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 229 | .func = bpf_perf_event_read, |
1075ef59 | 230 | .gpl_only = true, |
35578d79 KX |
231 | .ret_type = RET_INTEGER, |
232 | .arg1_type = ARG_CONST_MAP_PTR, | |
233 | .arg2_type = ARG_ANYTHING, | |
234 | }; | |
235 | ||
8e7a3920 DB |
236 | static __always_inline u64 |
237 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
238 | u64 flags, struct perf_raw_record *raw) | |
a43eec30 | 239 | { |
a43eec30 | 240 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
d7931330 | 241 | unsigned int cpu = smp_processor_id(); |
1e33759c | 242 | u64 index = flags & BPF_F_INDEX_MASK; |
a43eec30 | 243 | struct perf_sample_data sample_data; |
3b1efb19 | 244 | struct bpf_event_entry *ee; |
a43eec30 | 245 | struct perf_event *event; |
a43eec30 | 246 | |
1e33759c | 247 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 248 | index = cpu; |
a43eec30 AS |
249 | if (unlikely(index >= array->map.max_entries)) |
250 | return -E2BIG; | |
251 | ||
3b1efb19 | 252 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 253 | if (!ee) |
a43eec30 AS |
254 | return -ENOENT; |
255 | ||
3b1efb19 | 256 | event = ee->event; |
a43eec30 AS |
257 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
258 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
259 | return -EINVAL; | |
260 | ||
d7931330 | 261 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
262 | return -EOPNOTSUPP; |
263 | ||
264 | perf_sample_data_init(&sample_data, 0, 0); | |
8e7a3920 | 265 | sample_data.raw = raw; |
a43eec30 AS |
266 | perf_event_output(event, &sample_data, regs); |
267 | return 0; | |
268 | } | |
269 | ||
8e7a3920 DB |
270 | static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size) |
271 | { | |
272 | struct pt_regs *regs = (struct pt_regs *)(long) r1; | |
273 | struct bpf_map *map = (struct bpf_map *)(long) r2; | |
274 | void *data = (void *)(long) r4; | |
275 | struct perf_raw_record raw = { | |
276 | .frag = { | |
277 | .size = size, | |
278 | .data = data, | |
279 | }, | |
280 | }; | |
281 | ||
282 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) | |
283 | return -EINVAL; | |
284 | ||
285 | return __bpf_perf_event_output(regs, map, flags, &raw); | |
286 | } | |
287 | ||
a43eec30 AS |
288 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
289 | .func = bpf_perf_event_output, | |
1075ef59 | 290 | .gpl_only = true, |
a43eec30 AS |
291 | .ret_type = RET_INTEGER, |
292 | .arg1_type = ARG_PTR_TO_CTX, | |
293 | .arg2_type = ARG_CONST_MAP_PTR, | |
294 | .arg3_type = ARG_ANYTHING, | |
295 | .arg4_type = ARG_PTR_TO_STACK, | |
296 | .arg5_type = ARG_CONST_STACK_SIZE, | |
297 | }; | |
298 | ||
bd570ff9 DB |
299 | static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); |
300 | ||
555c8a86 DB |
301 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
302 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 DB |
303 | { |
304 | struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); | |
555c8a86 DB |
305 | struct perf_raw_frag frag = { |
306 | .copy = ctx_copy, | |
307 | .size = ctx_size, | |
308 | .data = ctx, | |
309 | }; | |
310 | struct perf_raw_record raw = { | |
311 | .frag = { | |
312 | .next = ctx_size ? &frag : NULL, | |
313 | .size = meta_size, | |
314 | .data = meta, | |
315 | }, | |
316 | }; | |
bd570ff9 DB |
317 | |
318 | perf_fetch_caller_regs(regs); | |
319 | ||
555c8a86 | 320 | return __bpf_perf_event_output(regs, map, flags, &raw); |
bd570ff9 DB |
321 | } |
322 | ||
606274c5 AS |
323 | static u64 bpf_get_current_task(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
324 | { | |
325 | return (long) current; | |
326 | } | |
327 | ||
328 | static const struct bpf_func_proto bpf_get_current_task_proto = { | |
329 | .func = bpf_get_current_task, | |
330 | .gpl_only = true, | |
331 | .ret_type = RET_INTEGER, | |
332 | }; | |
333 | ||
9fd82b61 | 334 | static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) |
2541517c AS |
335 | { |
336 | switch (func_id) { | |
337 | case BPF_FUNC_map_lookup_elem: | |
338 | return &bpf_map_lookup_elem_proto; | |
339 | case BPF_FUNC_map_update_elem: | |
340 | return &bpf_map_update_elem_proto; | |
341 | case BPF_FUNC_map_delete_elem: | |
342 | return &bpf_map_delete_elem_proto; | |
343 | case BPF_FUNC_probe_read: | |
344 | return &bpf_probe_read_proto; | |
d9847d31 AS |
345 | case BPF_FUNC_ktime_get_ns: |
346 | return &bpf_ktime_get_ns_proto; | |
04fd61ab AS |
347 | case BPF_FUNC_tail_call: |
348 | return &bpf_tail_call_proto; | |
ffeedafb AS |
349 | case BPF_FUNC_get_current_pid_tgid: |
350 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
351 | case BPF_FUNC_get_current_task: |
352 | return &bpf_get_current_task_proto; | |
ffeedafb AS |
353 | case BPF_FUNC_get_current_uid_gid: |
354 | return &bpf_get_current_uid_gid_proto; | |
355 | case BPF_FUNC_get_current_comm: | |
356 | return &bpf_get_current_comm_proto; | |
9c959c86 | 357 | case BPF_FUNC_trace_printk: |
0756ea3e | 358 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
359 | case BPF_FUNC_get_smp_processor_id: |
360 | return &bpf_get_smp_processor_id_proto; | |
35578d79 KX |
361 | case BPF_FUNC_perf_event_read: |
362 | return &bpf_perf_event_read_proto; | |
9fd82b61 AS |
363 | default: |
364 | return NULL; | |
365 | } | |
366 | } | |
367 | ||
368 | static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) | |
369 | { | |
370 | switch (func_id) { | |
a43eec30 AS |
371 | case BPF_FUNC_perf_event_output: |
372 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
373 | case BPF_FUNC_get_stackid: |
374 | return &bpf_get_stackid_proto; | |
2541517c | 375 | default: |
9fd82b61 | 376 | return tracing_func_proto(func_id); |
2541517c AS |
377 | } |
378 | } | |
379 | ||
380 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 AS |
381 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
382 | enum bpf_reg_type *reg_type) | |
2541517c | 383 | { |
2541517c AS |
384 | if (off < 0 || off >= sizeof(struct pt_regs)) |
385 | return false; | |
2541517c AS |
386 | if (type != BPF_READ) |
387 | return false; | |
2541517c AS |
388 | if (off % size != 0) |
389 | return false; | |
2541517c AS |
390 | return true; |
391 | } | |
392 | ||
27dff4e0 | 393 | static const struct bpf_verifier_ops kprobe_prog_ops = { |
2541517c AS |
394 | .get_func_proto = kprobe_prog_func_proto, |
395 | .is_valid_access = kprobe_prog_is_valid_access, | |
396 | }; | |
397 | ||
398 | static struct bpf_prog_type_list kprobe_tl = { | |
399 | .ops = &kprobe_prog_ops, | |
400 | .type = BPF_PROG_TYPE_KPROBE, | |
401 | }; | |
402 | ||
9940d67c AS |
403 | static u64 bpf_perf_event_output_tp(u64 r1, u64 r2, u64 index, u64 r4, u64 size) |
404 | { | |
405 | /* | |
406 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
407 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
408 | * from there and call the same bpf_perf_event_output() helper | |
409 | */ | |
266a0a79 | 410 | u64 ctx = *(long *)(uintptr_t)r1; |
9940d67c AS |
411 | |
412 | return bpf_perf_event_output(ctx, r2, index, r4, size); | |
413 | } | |
414 | ||
415 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
416 | .func = bpf_perf_event_output_tp, | |
417 | .gpl_only = true, | |
418 | .ret_type = RET_INTEGER, | |
419 | .arg1_type = ARG_PTR_TO_CTX, | |
420 | .arg2_type = ARG_CONST_MAP_PTR, | |
421 | .arg3_type = ARG_ANYTHING, | |
422 | .arg4_type = ARG_PTR_TO_STACK, | |
423 | .arg5_type = ARG_CONST_STACK_SIZE, | |
424 | }; | |
425 | ||
426 | static u64 bpf_get_stackid_tp(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
427 | { | |
266a0a79 | 428 | u64 ctx = *(long *)(uintptr_t)r1; |
9940d67c AS |
429 | |
430 | return bpf_get_stackid(ctx, r2, r3, r4, r5); | |
431 | } | |
432 | ||
433 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
434 | .func = bpf_get_stackid_tp, | |
435 | .gpl_only = true, | |
436 | .ret_type = RET_INTEGER, | |
437 | .arg1_type = ARG_PTR_TO_CTX, | |
438 | .arg2_type = ARG_CONST_MAP_PTR, | |
439 | .arg3_type = ARG_ANYTHING, | |
440 | }; | |
441 | ||
9fd82b61 AS |
442 | static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) |
443 | { | |
444 | switch (func_id) { | |
445 | case BPF_FUNC_perf_event_output: | |
9940d67c | 446 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 447 | case BPF_FUNC_get_stackid: |
9940d67c | 448 | return &bpf_get_stackid_proto_tp; |
9fd82b61 AS |
449 | default: |
450 | return tracing_func_proto(func_id); | |
451 | } | |
452 | } | |
453 | ||
19de99f7 AS |
454 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
455 | enum bpf_reg_type *reg_type) | |
9fd82b61 AS |
456 | { |
457 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
458 | return false; | |
459 | if (type != BPF_READ) | |
460 | return false; | |
461 | if (off % size != 0) | |
462 | return false; | |
463 | return true; | |
464 | } | |
465 | ||
466 | static const struct bpf_verifier_ops tracepoint_prog_ops = { | |
467 | .get_func_proto = tp_prog_func_proto, | |
468 | .is_valid_access = tp_prog_is_valid_access, | |
469 | }; | |
470 | ||
471 | static struct bpf_prog_type_list tracepoint_tl = { | |
472 | .ops = &tracepoint_prog_ops, | |
473 | .type = BPF_PROG_TYPE_TRACEPOINT, | |
474 | }; | |
475 | ||
2541517c AS |
476 | static int __init register_kprobe_prog_ops(void) |
477 | { | |
478 | bpf_register_prog_type(&kprobe_tl); | |
9fd82b61 | 479 | bpf_register_prog_type(&tracepoint_tl); |
2541517c AS |
480 | return 0; |
481 | } | |
482 | late_initcall(register_kprobe_prog_ops); |