Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
2/* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3#ifndef __USDT_BPF_H__
4#define __USDT_BPF_H__
5
6#include <linux/errno.h>
7#include "bpf_helpers.h"
8#include "bpf_tracing.h"
9
10/* Below types and maps are internal implementation details of libbpf's USDT
11 * support and are subjects to change. Also, bpf_usdt_xxx() API helpers should
12 * be considered an unstable API as well and might be adjusted based on user
13 * feedback from using libbpf's USDT support in production.
14 */
15
16/* User can override BPF_USDT_MAX_SPEC_CNT to change default size of internal
17 * map that keeps track of USDT argument specifications. This might be
18 * necessary if there are a lot of USDT attachments.
19 */
20#ifndef BPF_USDT_MAX_SPEC_CNT
21#define BPF_USDT_MAX_SPEC_CNT 256
22#endif
23/* User can override BPF_USDT_MAX_IP_CNT to change default size of internal
24 * map that keeps track of IP (memory address) mapping to USDT argument
25 * specification.
26 * Note, if kernel supports BPF cookies, this map is not used and could be
27 * resized all the way to 1 to save a bit of memory.
28 */
29#ifndef BPF_USDT_MAX_IP_CNT
30#define BPF_USDT_MAX_IP_CNT (4 * BPF_USDT_MAX_SPEC_CNT)
31#endif
32
33enum __bpf_usdt_arg_type {
34 BPF_USDT_ARG_CONST,
35 BPF_USDT_ARG_REG,
36 BPF_USDT_ARG_REG_DEREF,
37 BPF_USDT_ARG_SIB,
38};
39
40/*
41 * This struct layout is designed specifically to be backwards/forward
42 * compatible between libbpf versions for ARG_CONST, ARG_REG, and
43 * ARG_REG_DEREF modes. ARG_SIB requires libbpf v1.7+.
44 */
45struct __bpf_usdt_arg_spec {
46 /* u64 scalar interpreted depending on arg_type, see below */
47 __u64 val_off;
48#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
49 /* arg location case, see bpf_usdt_arg() for details */
50 enum __bpf_usdt_arg_type arg_type: 8;
51 /* index register offset within struct pt_regs */
52 __u16 idx_reg_off: 12;
53 /* scale factor for index register (1, 2, 4, or 8) */
54 __u16 scale_bitshift: 4;
55 /* reserved for future use, keeps reg_off offset stable */
56 __u8 __reserved: 8;
57#else
58 __u8 __reserved: 8;
59 __u16 idx_reg_off: 12;
60 __u16 scale_bitshift: 4;
61 enum __bpf_usdt_arg_type arg_type: 8;
62#endif
63 /* offset of referenced register within struct pt_regs */
64 short reg_off;
65 /* whether arg should be interpreted as signed value */
66 bool arg_signed;
67 /* number of bits that need to be cleared and, optionally,
68 * sign-extended to cast arguments that are 1, 2, or 4 bytes
69 * long into final 8-byte u64/s64 value returned to user
70 */
71 char arg_bitshift;
72};
73
74/* should match USDT_MAX_ARG_CNT in usdt.c exactly */
75#define BPF_USDT_MAX_ARG_CNT 12
76struct __bpf_usdt_spec {
77 struct __bpf_usdt_arg_spec args[BPF_USDT_MAX_ARG_CNT];
78 __u64 usdt_cookie;
79 short arg_cnt;
80};
81
82struct {
83 __uint(type, BPF_MAP_TYPE_ARRAY);
84 __uint(max_entries, BPF_USDT_MAX_SPEC_CNT);
85 __type(key, int);
86 __type(value, struct __bpf_usdt_spec);
87} __bpf_usdt_specs SEC(".maps") __weak;
88
89struct {
90 __uint(type, BPF_MAP_TYPE_HASH);
91 __uint(max_entries, BPF_USDT_MAX_IP_CNT);
92 __type(key, long);
93 __type(value, __u32);
94} __bpf_usdt_ip_to_spec_id SEC(".maps") __weak;
95
96extern const _Bool LINUX_HAS_BPF_COOKIE __kconfig;
97
98static __always_inline
99int __bpf_usdt_spec_id(struct pt_regs *ctx)
100{
101 if (!LINUX_HAS_BPF_COOKIE) {
102 long ip = PT_REGS_IP(ctx);
103 int *spec_id_ptr;
104
105 spec_id_ptr = bpf_map_lookup_elem(&__bpf_usdt_ip_to_spec_id, &ip);
106 return spec_id_ptr ? *spec_id_ptr : -ESRCH;
107 }
108
109 return bpf_get_attach_cookie(ctx);
110}
111
112/* Return number of USDT arguments defined for currently traced USDT. */
113__weak __hidden
114int bpf_usdt_arg_cnt(struct pt_regs *ctx)
115{
116 struct __bpf_usdt_spec *spec;
117 int spec_id;
118
119 spec_id = __bpf_usdt_spec_id(ctx);
120 if (spec_id < 0)
121 return -ESRCH;
122
123 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
124 if (!spec)
125 return -ESRCH;
126
127 return spec->arg_cnt;
128}
129
130/* Returns the size in bytes of the #*arg_num* (zero-indexed) USDT argument.
131 * Returns negative error if argument is not found or arg_num is invalid.
132 */
133static __always_inline
134int bpf_usdt_arg_size(struct pt_regs *ctx, __u64 arg_num)
135{
136 struct __bpf_usdt_arg_spec *arg_spec;
137 struct __bpf_usdt_spec *spec;
138 int spec_id;
139
140 spec_id = __bpf_usdt_spec_id(ctx);
141 if (spec_id < 0)
142 return -ESRCH;
143
144 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
145 if (!spec)
146 return -ESRCH;
147
148 if (arg_num >= BPF_USDT_MAX_ARG_CNT)
149 return -ENOENT;
150 barrier_var(arg_num);
151 if (arg_num >= spec->arg_cnt)
152 return -ENOENT;
153
154 arg_spec = &spec->args[arg_num];
155
156 /* arg_spec->arg_bitshift = 64 - arg_sz * 8
157 * so: arg_sz = (64 - arg_spec->arg_bitshift) / 8
158 */
159 return (unsigned int)(64 - arg_spec->arg_bitshift) / 8;
160}
161
162/* Fetch USDT argument #*arg_num* (zero-indexed) and put its value into *res.
163 * Returns 0 on success; negative error, otherwise.
164 * On error *res is guaranteed to be set to zero.
165 */
166__weak __hidden
167int bpf_usdt_arg(struct pt_regs *ctx, __u64 arg_num, long *res)
168{
169 struct __bpf_usdt_spec *spec;
170 struct __bpf_usdt_arg_spec *arg_spec;
171 unsigned long val, idx;
172 int err, spec_id;
173
174 *res = 0;
175
176 spec_id = __bpf_usdt_spec_id(ctx);
177 if (spec_id < 0)
178 return -ESRCH;
179
180 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
181 if (!spec)
182 return -ESRCH;
183
184 if (arg_num >= BPF_USDT_MAX_ARG_CNT)
185 return -ENOENT;
186 barrier_var(arg_num);
187 if (arg_num >= spec->arg_cnt)
188 return -ENOENT;
189
190 arg_spec = &spec->args[arg_num];
191 switch (arg_spec->arg_type) {
192 case BPF_USDT_ARG_CONST:
193 /* Arg is just a constant ("-4@$-9" in USDT arg spec).
194 * value is recorded in arg_spec->val_off directly.
195 */
196 val = arg_spec->val_off;
197 break;
198 case BPF_USDT_ARG_REG:
199 /* Arg is in a register (e.g, "8@%rax" in USDT arg spec),
200 * so we read the contents of that register directly from
201 * struct pt_regs. To keep things simple user-space parts
202 * record offsetof(struct pt_regs, <regname>) in arg_spec->reg_off.
203 */
204 err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
205 if (err)
206 return err;
207 break;
208 case BPF_USDT_ARG_REG_DEREF:
209 /* Arg is in memory addressed by register, plus some offset
210 * (e.g., "-4@-1204(%rbp)" in USDT arg spec). Register is
211 * identified like with BPF_USDT_ARG_REG case, and the offset
212 * is in arg_spec->val_off. We first fetch register contents
213 * from pt_regs, then do another user-space probe read to
214 * fetch argument value itself.
215 */
216 err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
217 if (err)
218 return err;
219 err = bpf_probe_read_user(&val, sizeof(val), (void *)val + arg_spec->val_off);
220 if (err)
221 return err;
222#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
223 val >>= arg_spec->arg_bitshift;
224#endif
225 break;
226 case BPF_USDT_ARG_SIB:
227 /* Arg is in memory addressed by SIB (Scale-Index-Base) mode
228 * (e.g., "-1@-96(%rbp,%rax,8)" in USDT arg spec). We first
229 * fetch the base register contents and the index register
230 * contents from pt_regs. Then we calculate the final address
231 * as base + (index * scale) + offset, and do a user-space
232 * probe read to fetch the argument value.
233 */
234 err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
235 if (err)
236 return err;
237 err = bpf_probe_read_kernel(&idx, sizeof(idx), (void *)ctx + arg_spec->idx_reg_off);
238 if (err)
239 return err;
240 err = bpf_probe_read_user(&val, sizeof(val), (void *)(val + (idx << arg_spec->scale_bitshift) + arg_spec->val_off));
241 if (err)
242 return err;
243#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
244 val >>= arg_spec->arg_bitshift;
245#endif
246 break;
247 default:
248 return -EINVAL;
249 }
250
251 /* cast arg from 1, 2, or 4 bytes to final 8 byte size clearing
252 * necessary upper arg_bitshift bits, with sign extension if argument
253 * is signed
254 */
255 val <<= arg_spec->arg_bitshift;
256 if (arg_spec->arg_signed)
257 val = ((long)val) >> arg_spec->arg_bitshift;
258 else
259 val = val >> arg_spec->arg_bitshift;
260 *res = val;
261 return 0;
262}
263
264/* Retrieve user-specified cookie value provided during attach as
265 * bpf_usdt_opts.usdt_cookie. This serves the same purpose as BPF cookie
266 * returned by bpf_get_attach_cookie(). Libbpf's support for USDT is itself
267 * utilizing BPF cookies internally, so user can't use BPF cookie directly
268 * for USDT programs and has to use bpf_usdt_cookie() API instead.
269 */
270__weak __hidden
271long bpf_usdt_cookie(struct pt_regs *ctx)
272{
273 struct __bpf_usdt_spec *spec;
274 int spec_id;
275
276 spec_id = __bpf_usdt_spec_id(ctx);
277 if (spec_id < 0)
278 return 0;
279
280 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
281 if (!spec)
282 return 0;
283
284 return spec->usdt_cookie;
285}
286
287/* we rely on ___bpf_apply() and ___bpf_narg() macros already defined in bpf_tracing.h */
288#define ___bpf_usdt_args0() ctx
289#define ___bpf_usdt_args1(x) ___bpf_usdt_args0(), ({ long _x; bpf_usdt_arg(ctx, 0, &_x); _x; })
290#define ___bpf_usdt_args2(x, args...) ___bpf_usdt_args1(args), ({ long _x; bpf_usdt_arg(ctx, 1, &_x); _x; })
291#define ___bpf_usdt_args3(x, args...) ___bpf_usdt_args2(args), ({ long _x; bpf_usdt_arg(ctx, 2, &_x); _x; })
292#define ___bpf_usdt_args4(x, args...) ___bpf_usdt_args3(args), ({ long _x; bpf_usdt_arg(ctx, 3, &_x); _x; })
293#define ___bpf_usdt_args5(x, args...) ___bpf_usdt_args4(args), ({ long _x; bpf_usdt_arg(ctx, 4, &_x); _x; })
294#define ___bpf_usdt_args6(x, args...) ___bpf_usdt_args5(args), ({ long _x; bpf_usdt_arg(ctx, 5, &_x); _x; })
295#define ___bpf_usdt_args7(x, args...) ___bpf_usdt_args6(args), ({ long _x; bpf_usdt_arg(ctx, 6, &_x); _x; })
296#define ___bpf_usdt_args8(x, args...) ___bpf_usdt_args7(args), ({ long _x; bpf_usdt_arg(ctx, 7, &_x); _x; })
297#define ___bpf_usdt_args9(x, args...) ___bpf_usdt_args8(args), ({ long _x; bpf_usdt_arg(ctx, 8, &_x); _x; })
298#define ___bpf_usdt_args10(x, args...) ___bpf_usdt_args9(args), ({ long _x; bpf_usdt_arg(ctx, 9, &_x); _x; })
299#define ___bpf_usdt_args11(x, args...) ___bpf_usdt_args10(args), ({ long _x; bpf_usdt_arg(ctx, 10, &_x); _x; })
300#define ___bpf_usdt_args12(x, args...) ___bpf_usdt_args11(args), ({ long _x; bpf_usdt_arg(ctx, 11, &_x); _x; })
301#define ___bpf_usdt_args(args...) ___bpf_apply(___bpf_usdt_args, ___bpf_narg(args))(args)
302
303/*
304 * BPF_USDT serves the same purpose for USDT handlers as BPF_PROG for
305 * tp_btf/fentry/fexit BPF programs and BPF_KPROBE for kprobes.
306 * Original struct pt_regs * context is preserved as 'ctx' argument.
307 */
308#define BPF_USDT(name, args...) \
309name(struct pt_regs *ctx); \
310static __always_inline typeof(name(0)) \
311____##name(struct pt_regs *ctx, ##args); \
312typeof(name(0)) name(struct pt_regs *ctx) \
313{ \
314 _Pragma("GCC diagnostic push") \
315 _Pragma("GCC diagnostic ignored \"-Wint-conversion\"") \
316 return ____##name(___bpf_usdt_args(args)); \
317 _Pragma("GCC diagnostic pop") \
318} \
319static __always_inline typeof(name(0)) \
320____##name(struct pt_regs *ctx, ##args)
321
322#endif /* __USDT_BPF_H__ */