xref: /linux/kernel/trace/bpf_trace.c (revision 0d240e7811c4ec1965760ee4643b5bbc9cfacbb3)
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>
13 #include <linux/ctype.h>
14 #include "trace.h"
15 
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;
65 	int ret, size = (int) r2;
66 	void *unsafe_ptr = (void *) (long) r3;
67 
68 	ret = probe_kernel_read(dst, unsafe_ptr, size);
69 	if (unlikely(ret < 0))
70 		memset(dst, 0, size);
71 
72 	return ret;
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,
79 	.arg1_type	= ARG_PTR_TO_RAW_STACK,
80 	.arg2_type	= ARG_CONST_STACK_SIZE,
81 	.arg3_type	= ARG_ANYTHING,
82 };
83 
84 /*
85  * limited trace_printk()
86  * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
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;
91 	bool str_seen = false;
92 	int mod[3] = {};
93 	int fmt_cnt = 0;
94 	u64 unsafe_addr;
95 	char buf[64];
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++;
122 		} else if (fmt[i] == 'p' || fmt[i] == 's') {
123 			mod[fmt_cnt]++;
124 			i++;
125 			if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
126 				return -EINVAL;
127 			fmt_cnt++;
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 			}
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 
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 
191 static u64 bpf_perf_event_read(u64 r1, u64 index, u64 r3, u64 r4, u64 r5)
192 {
193 	struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
194 	struct bpf_array *array = container_of(map, struct bpf_array, map);
195 	struct bpf_event_entry *ee;
196 	struct perf_event *event;
197 
198 	if (unlikely(index >= array->map.max_entries))
199 		return -E2BIG;
200 
201 	ee = READ_ONCE(array->ptrs[index]);
202 	if (unlikely(!ee))
203 		return -ENOENT;
204 
205 	event = ee->event;
206 	/* make sure event is local and doesn't have pmu::count */
207 	if (event->oncpu != smp_processor_id() ||
208 	    event->pmu->count)
209 		return -EINVAL;
210 
211 	/*
212 	 * we don't know if the function is run successfully by the
213 	 * return value. It can be judged in other places, such as
214 	 * eBPF programs.
215 	 */
216 	return perf_event_read_local(event);
217 }
218 
219 static const struct bpf_func_proto bpf_perf_event_read_proto = {
220 	.func		= bpf_perf_event_read,
221 	.gpl_only	= true,
222 	.ret_type	= RET_INTEGER,
223 	.arg1_type	= ARG_CONST_MAP_PTR,
224 	.arg2_type	= ARG_ANYTHING,
225 };
226 
227 static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size)
228 {
229 	struct pt_regs *regs = (struct pt_regs *) (long) r1;
230 	struct bpf_map *map = (struct bpf_map *) (long) r2;
231 	struct bpf_array *array = container_of(map, struct bpf_array, map);
232 	u64 index = flags & BPF_F_INDEX_MASK;
233 	void *data = (void *) (long) r4;
234 	struct perf_sample_data sample_data;
235 	struct bpf_event_entry *ee;
236 	struct perf_event *event;
237 	struct perf_raw_record raw = {
238 		.size = size,
239 		.data = data,
240 	};
241 
242 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
243 		return -EINVAL;
244 	if (index == BPF_F_CURRENT_CPU)
245 		index = raw_smp_processor_id();
246 	if (unlikely(index >= array->map.max_entries))
247 		return -E2BIG;
248 
249 	ee = READ_ONCE(array->ptrs[index]);
250 	if (unlikely(!ee))
251 		return -ENOENT;
252 
253 	event = ee->event;
254 	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
255 		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
256 		return -EINVAL;
257 
258 	if (unlikely(event->oncpu != smp_processor_id()))
259 		return -EOPNOTSUPP;
260 
261 	perf_sample_data_init(&sample_data, 0, 0);
262 	sample_data.raw = &raw;
263 	perf_event_output(event, &sample_data, regs);
264 	return 0;
265 }
266 
267 static const struct bpf_func_proto bpf_perf_event_output_proto = {
268 	.func		= bpf_perf_event_output,
269 	.gpl_only	= true,
270 	.ret_type	= RET_INTEGER,
271 	.arg1_type	= ARG_PTR_TO_CTX,
272 	.arg2_type	= ARG_CONST_MAP_PTR,
273 	.arg3_type	= ARG_ANYTHING,
274 	.arg4_type	= ARG_PTR_TO_STACK,
275 	.arg5_type	= ARG_CONST_STACK_SIZE,
276 };
277 
278 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
279 
280 static u64 bpf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size)
281 {
282 	struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
283 
284 	perf_fetch_caller_regs(regs);
285 
286 	return bpf_perf_event_output((long)regs, r2, flags, r4, size);
287 }
288 
289 static const struct bpf_func_proto bpf_event_output_proto = {
290 	.func		= bpf_event_output,
291 	.gpl_only	= true,
292 	.ret_type	= RET_INTEGER,
293 	.arg1_type	= ARG_PTR_TO_CTX,
294 	.arg2_type	= ARG_CONST_MAP_PTR,
295 	.arg3_type	= ARG_ANYTHING,
296 	.arg4_type	= ARG_PTR_TO_STACK,
297 	.arg5_type	= ARG_CONST_STACK_SIZE,
298 };
299 
300 const struct bpf_func_proto *bpf_get_event_output_proto(void)
301 {
302 	return &bpf_event_output_proto;
303 }
304 
305 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
306 {
307 	switch (func_id) {
308 	case BPF_FUNC_map_lookup_elem:
309 		return &bpf_map_lookup_elem_proto;
310 	case BPF_FUNC_map_update_elem:
311 		return &bpf_map_update_elem_proto;
312 	case BPF_FUNC_map_delete_elem:
313 		return &bpf_map_delete_elem_proto;
314 	case BPF_FUNC_probe_read:
315 		return &bpf_probe_read_proto;
316 	case BPF_FUNC_ktime_get_ns:
317 		return &bpf_ktime_get_ns_proto;
318 	case BPF_FUNC_tail_call:
319 		return &bpf_tail_call_proto;
320 	case BPF_FUNC_get_current_pid_tgid:
321 		return &bpf_get_current_pid_tgid_proto;
322 	case BPF_FUNC_get_current_uid_gid:
323 		return &bpf_get_current_uid_gid_proto;
324 	case BPF_FUNC_get_current_comm:
325 		return &bpf_get_current_comm_proto;
326 	case BPF_FUNC_trace_printk:
327 		return bpf_get_trace_printk_proto();
328 	case BPF_FUNC_get_smp_processor_id:
329 		return &bpf_get_smp_processor_id_proto;
330 	case BPF_FUNC_perf_event_read:
331 		return &bpf_perf_event_read_proto;
332 	default:
333 		return NULL;
334 	}
335 }
336 
337 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
338 {
339 	switch (func_id) {
340 	case BPF_FUNC_perf_event_output:
341 		return &bpf_perf_event_output_proto;
342 	case BPF_FUNC_get_stackid:
343 		return &bpf_get_stackid_proto;
344 	default:
345 		return tracing_func_proto(func_id);
346 	}
347 }
348 
349 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
350 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type)
351 {
352 	/* check bounds */
353 	if (off < 0 || off >= sizeof(struct pt_regs))
354 		return false;
355 
356 	/* only read is allowed */
357 	if (type != BPF_READ)
358 		return false;
359 
360 	/* disallow misaligned access */
361 	if (off % size != 0)
362 		return false;
363 
364 	return true;
365 }
366 
367 static const struct bpf_verifier_ops kprobe_prog_ops = {
368 	.get_func_proto  = kprobe_prog_func_proto,
369 	.is_valid_access = kprobe_prog_is_valid_access,
370 };
371 
372 static struct bpf_prog_type_list kprobe_tl = {
373 	.ops	= &kprobe_prog_ops,
374 	.type	= BPF_PROG_TYPE_KPROBE,
375 };
376 
377 static u64 bpf_perf_event_output_tp(u64 r1, u64 r2, u64 index, u64 r4, u64 size)
378 {
379 	/*
380 	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
381 	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
382 	 * from there and call the same bpf_perf_event_output() helper
383 	 */
384 	u64 ctx = *(long *)(uintptr_t)r1;
385 
386 	return bpf_perf_event_output(ctx, r2, index, r4, size);
387 }
388 
389 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
390 	.func		= bpf_perf_event_output_tp,
391 	.gpl_only	= true,
392 	.ret_type	= RET_INTEGER,
393 	.arg1_type	= ARG_PTR_TO_CTX,
394 	.arg2_type	= ARG_CONST_MAP_PTR,
395 	.arg3_type	= ARG_ANYTHING,
396 	.arg4_type	= ARG_PTR_TO_STACK,
397 	.arg5_type	= ARG_CONST_STACK_SIZE,
398 };
399 
400 static u64 bpf_get_stackid_tp(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
401 {
402 	u64 ctx = *(long *)(uintptr_t)r1;
403 
404 	return bpf_get_stackid(ctx, r2, r3, r4, r5);
405 }
406 
407 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
408 	.func		= bpf_get_stackid_tp,
409 	.gpl_only	= true,
410 	.ret_type	= RET_INTEGER,
411 	.arg1_type	= ARG_PTR_TO_CTX,
412 	.arg2_type	= ARG_CONST_MAP_PTR,
413 	.arg3_type	= ARG_ANYTHING,
414 };
415 
416 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
417 {
418 	switch (func_id) {
419 	case BPF_FUNC_perf_event_output:
420 		return &bpf_perf_event_output_proto_tp;
421 	case BPF_FUNC_get_stackid:
422 		return &bpf_get_stackid_proto_tp;
423 	default:
424 		return tracing_func_proto(func_id);
425 	}
426 }
427 
428 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type)
429 {
430 	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
431 		return false;
432 	if (type != BPF_READ)
433 		return false;
434 	if (off % size != 0)
435 		return false;
436 	return true;
437 }
438 
439 static const struct bpf_verifier_ops tracepoint_prog_ops = {
440 	.get_func_proto  = tp_prog_func_proto,
441 	.is_valid_access = tp_prog_is_valid_access,
442 };
443 
444 static struct bpf_prog_type_list tracepoint_tl = {
445 	.ops	= &tracepoint_prog_ops,
446 	.type	= BPF_PROG_TYPE_TRACEPOINT,
447 };
448 
449 static int __init register_kprobe_prog_ops(void)
450 {
451 	bpf_register_prog_type(&kprobe_tl);
452 	bpf_register_prog_type(&tracepoint_tl);
453 	return 0;
454 }
455 late_initcall(register_kprobe_prog_ops);
456