xref: /linux/kernel/trace/bpf_trace.c (revision f02e58f91a121ec909efad06b0a7aa806e1f7a84)
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 static DEFINE_PER_CPU(int, bpf_prog_active);
17 
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 
64 static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
65 {
66 	void *dst = (void *) (long) r1;
67 	int size = (int) r2;
68 	void *unsafe_ptr = (void *) (long) r3;
69 
70 	return probe_kernel_read(dst, unsafe_ptr, size);
71 }
72 
73 static const struct bpf_func_proto bpf_probe_read_proto = {
74 	.func		= bpf_probe_read,
75 	.gpl_only	= true,
76 	.ret_type	= RET_INTEGER,
77 	.arg1_type	= ARG_PTR_TO_STACK,
78 	.arg2_type	= ARG_CONST_STACK_SIZE,
79 	.arg3_type	= ARG_ANYTHING,
80 };
81 
82 /*
83  * limited trace_printk()
84  * only %d %u %x %ld %lu %lx %lld %llu %llx %p conversion specifiers allowed
85  */
86 static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5)
87 {
88 	char *fmt = (char *) (long) r1;
89 	int mod[3] = {};
90 	int fmt_cnt = 0;
91 	int i;
92 
93 	/*
94 	 * bpf_check()->check_func_arg()->check_stack_boundary()
95 	 * guarantees that fmt points to bpf program stack,
96 	 * fmt_size bytes of it were initialized and fmt_size > 0
97 	 */
98 	if (fmt[--fmt_size] != 0)
99 		return -EINVAL;
100 
101 	/* check format string for allowed specifiers */
102 	for (i = 0; i < fmt_size; i++) {
103 		if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
104 			return -EINVAL;
105 
106 		if (fmt[i] != '%')
107 			continue;
108 
109 		if (fmt_cnt >= 3)
110 			return -EINVAL;
111 
112 		/* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
113 		i++;
114 		if (fmt[i] == 'l') {
115 			mod[fmt_cnt]++;
116 			i++;
117 		} else if (fmt[i] == 'p') {
118 			mod[fmt_cnt]++;
119 			i++;
120 			if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
121 				return -EINVAL;
122 			fmt_cnt++;
123 			continue;
124 		}
125 
126 		if (fmt[i] == 'l') {
127 			mod[fmt_cnt]++;
128 			i++;
129 		}
130 
131 		if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
132 			return -EINVAL;
133 		fmt_cnt++;
134 	}
135 
136 	return __trace_printk(1/* fake ip will not be printed */, fmt,
137 			      mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3,
138 			      mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4,
139 			      mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5);
140 }
141 
142 static const struct bpf_func_proto bpf_trace_printk_proto = {
143 	.func		= bpf_trace_printk,
144 	.gpl_only	= true,
145 	.ret_type	= RET_INTEGER,
146 	.arg1_type	= ARG_PTR_TO_STACK,
147 	.arg2_type	= ARG_CONST_STACK_SIZE,
148 };
149 
150 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
151 {
152 	/*
153 	 * this program might be calling bpf_trace_printk,
154 	 * so allocate per-cpu printk buffers
155 	 */
156 	trace_printk_init_buffers();
157 
158 	return &bpf_trace_printk_proto;
159 }
160 
161 static u64 bpf_perf_event_read(u64 r1, u64 index, u64 r3, u64 r4, u64 r5)
162 {
163 	struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
164 	struct bpf_array *array = container_of(map, struct bpf_array, map);
165 	struct perf_event *event;
166 
167 	if (unlikely(index >= array->map.max_entries))
168 		return -E2BIG;
169 
170 	event = (struct perf_event *)array->ptrs[index];
171 	if (!event)
172 		return -ENOENT;
173 
174 	/*
175 	 * we don't know if the function is run successfully by the
176 	 * return value. It can be judged in other places, such as
177 	 * eBPF programs.
178 	 */
179 	return perf_event_read_local(event);
180 }
181 
182 const struct bpf_func_proto bpf_perf_event_read_proto = {
183 	.func		= bpf_perf_event_read,
184 	.gpl_only	= false,
185 	.ret_type	= RET_INTEGER,
186 	.arg1_type	= ARG_CONST_MAP_PTR,
187 	.arg2_type	= ARG_ANYTHING,
188 };
189 
190 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
191 {
192 	switch (func_id) {
193 	case BPF_FUNC_map_lookup_elem:
194 		return &bpf_map_lookup_elem_proto;
195 	case BPF_FUNC_map_update_elem:
196 		return &bpf_map_update_elem_proto;
197 	case BPF_FUNC_map_delete_elem:
198 		return &bpf_map_delete_elem_proto;
199 	case BPF_FUNC_probe_read:
200 		return &bpf_probe_read_proto;
201 	case BPF_FUNC_ktime_get_ns:
202 		return &bpf_ktime_get_ns_proto;
203 	case BPF_FUNC_tail_call:
204 		return &bpf_tail_call_proto;
205 	case BPF_FUNC_get_current_pid_tgid:
206 		return &bpf_get_current_pid_tgid_proto;
207 	case BPF_FUNC_get_current_uid_gid:
208 		return &bpf_get_current_uid_gid_proto;
209 	case BPF_FUNC_get_current_comm:
210 		return &bpf_get_current_comm_proto;
211 	case BPF_FUNC_trace_printk:
212 		return bpf_get_trace_printk_proto();
213 	case BPF_FUNC_get_smp_processor_id:
214 		return &bpf_get_smp_processor_id_proto;
215 	case BPF_FUNC_perf_event_read:
216 		return &bpf_perf_event_read_proto;
217 	default:
218 		return NULL;
219 	}
220 }
221 
222 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
223 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type)
224 {
225 	/* check bounds */
226 	if (off < 0 || off >= sizeof(struct pt_regs))
227 		return false;
228 
229 	/* only read is allowed */
230 	if (type != BPF_READ)
231 		return false;
232 
233 	/* disallow misaligned access */
234 	if (off % size != 0)
235 		return false;
236 
237 	return true;
238 }
239 
240 static struct bpf_verifier_ops kprobe_prog_ops = {
241 	.get_func_proto  = kprobe_prog_func_proto,
242 	.is_valid_access = kprobe_prog_is_valid_access,
243 };
244 
245 static struct bpf_prog_type_list kprobe_tl = {
246 	.ops	= &kprobe_prog_ops,
247 	.type	= BPF_PROG_TYPE_KPROBE,
248 };
249 
250 static int __init register_kprobe_prog_ops(void)
251 {
252 	bpf_register_prog_type(&kprobe_tl);
253 	return 0;
254 }
255 late_initcall(register_kprobe_prog_ops);
256