1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2 1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
2 /* Copyright (c) 2022 Meta Platforms, Inc. and 2 /* Copyright (c) 2022 Meta Platforms, Inc. and affiliates. */
3 #ifndef __USDT_BPF_H__ 3 #ifndef __USDT_BPF_H__
4 #define __USDT_BPF_H__ 4 #define __USDT_BPF_H__
5 5
6 #include <linux/errno.h> 6 #include <linux/errno.h>
7 #include "bpf_helpers.h" 7 #include "bpf_helpers.h"
8 #include "bpf_tracing.h" 8 #include "bpf_tracing.h"
9 9
10 /* Below types and maps are internal implement 10 /* Below types and maps are internal implementation details of libbpf's USDT
11 * support and are subjects to change. Also, b 11 * support and are subjects to change. Also, bpf_usdt_xxx() API helpers should
12 * be considered an unstable API as well and m 12 * be considered an unstable API as well and might be adjusted based on user
13 * feedback from using libbpf's USDT support i 13 * feedback from using libbpf's USDT support in production.
14 */ 14 */
15 15
16 /* User can override BPF_USDT_MAX_SPEC_CNT to 16 /* User can override BPF_USDT_MAX_SPEC_CNT to change default size of internal
17 * map that keeps track of USDT argument speci 17 * map that keeps track of USDT argument specifications. This might be
18 * necessary if there are a lot of USDT attach 18 * necessary if there are a lot of USDT attachments.
19 */ 19 */
20 #ifndef BPF_USDT_MAX_SPEC_CNT 20 #ifndef BPF_USDT_MAX_SPEC_CNT
21 #define BPF_USDT_MAX_SPEC_CNT 256 21 #define BPF_USDT_MAX_SPEC_CNT 256
22 #endif 22 #endif
23 /* User can override BPF_USDT_MAX_IP_CNT to ch 23 /* User can override BPF_USDT_MAX_IP_CNT to change default size of internal
24 * map that keeps track of IP (memory address) 24 * map that keeps track of IP (memory address) mapping to USDT argument
25 * specification. 25 * specification.
26 * Note, if kernel supports BPF cookies, this 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 m 27 * resized all the way to 1 to save a bit of memory.
28 */ 28 */
29 #ifndef BPF_USDT_MAX_IP_CNT 29 #ifndef BPF_USDT_MAX_IP_CNT
30 #define BPF_USDT_MAX_IP_CNT (4 * BPF_USDT_MAX_ 30 #define BPF_USDT_MAX_IP_CNT (4 * BPF_USDT_MAX_SPEC_CNT)
31 #endif 31 #endif
32 32
33 enum __bpf_usdt_arg_type { 33 enum __bpf_usdt_arg_type {
34 BPF_USDT_ARG_CONST, 34 BPF_USDT_ARG_CONST,
35 BPF_USDT_ARG_REG, 35 BPF_USDT_ARG_REG,
36 BPF_USDT_ARG_REG_DEREF, 36 BPF_USDT_ARG_REG_DEREF,
37 }; 37 };
38 38
39 struct __bpf_usdt_arg_spec { 39 struct __bpf_usdt_arg_spec {
40 /* u64 scalar interpreted depending on 40 /* u64 scalar interpreted depending on arg_type, see below */
41 __u64 val_off; 41 __u64 val_off;
42 /* arg location case, see bpf_usdt_arg 42 /* arg location case, see bpf_usdt_arg() for details */
43 enum __bpf_usdt_arg_type arg_type; 43 enum __bpf_usdt_arg_type arg_type;
44 /* offset of referenced register withi 44 /* offset of referenced register within struct pt_regs */
45 short reg_off; 45 short reg_off;
46 /* whether arg should be interpreted a 46 /* whether arg should be interpreted as signed value */
47 bool arg_signed; 47 bool arg_signed;
48 /* number of bits that need to be clea 48 /* number of bits that need to be cleared and, optionally,
49 * sign-extended to cast arguments tha 49 * sign-extended to cast arguments that are 1, 2, or 4 bytes
50 * long into final 8-byte u64/s64 valu 50 * long into final 8-byte u64/s64 value returned to user
51 */ 51 */
52 char arg_bitshift; 52 char arg_bitshift;
53 }; 53 };
54 54
55 /* should match USDT_MAX_ARG_CNT in usdt.c exa 55 /* should match USDT_MAX_ARG_CNT in usdt.c exactly */
56 #define BPF_USDT_MAX_ARG_CNT 12 56 #define BPF_USDT_MAX_ARG_CNT 12
57 struct __bpf_usdt_spec { 57 struct __bpf_usdt_spec {
58 struct __bpf_usdt_arg_spec args[BPF_US 58 struct __bpf_usdt_arg_spec args[BPF_USDT_MAX_ARG_CNT];
59 __u64 usdt_cookie; 59 __u64 usdt_cookie;
60 short arg_cnt; 60 short arg_cnt;
61 }; 61 };
62 62
63 struct { 63 struct {
64 __uint(type, BPF_MAP_TYPE_ARRAY); 64 __uint(type, BPF_MAP_TYPE_ARRAY);
65 __uint(max_entries, BPF_USDT_MAX_SPEC_ 65 __uint(max_entries, BPF_USDT_MAX_SPEC_CNT);
66 __type(key, int); 66 __type(key, int);
67 __type(value, struct __bpf_usdt_spec); 67 __type(value, struct __bpf_usdt_spec);
68 } __bpf_usdt_specs SEC(".maps") __weak; 68 } __bpf_usdt_specs SEC(".maps") __weak;
69 69
70 struct { 70 struct {
71 __uint(type, BPF_MAP_TYPE_HASH); 71 __uint(type, BPF_MAP_TYPE_HASH);
72 __uint(max_entries, BPF_USDT_MAX_IP_CN 72 __uint(max_entries, BPF_USDT_MAX_IP_CNT);
73 __type(key, long); 73 __type(key, long);
74 __type(value, __u32); 74 __type(value, __u32);
75 } __bpf_usdt_ip_to_spec_id SEC(".maps") __weak 75 } __bpf_usdt_ip_to_spec_id SEC(".maps") __weak;
76 76
77 extern const _Bool LINUX_HAS_BPF_COOKIE __kcon 77 extern const _Bool LINUX_HAS_BPF_COOKIE __kconfig;
78 78
79 static __always_inline 79 static __always_inline
80 int __bpf_usdt_spec_id(struct pt_regs *ctx) 80 int __bpf_usdt_spec_id(struct pt_regs *ctx)
81 { 81 {
82 if (!LINUX_HAS_BPF_COOKIE) { 82 if (!LINUX_HAS_BPF_COOKIE) {
83 long ip = PT_REGS_IP(ctx); 83 long ip = PT_REGS_IP(ctx);
84 int *spec_id_ptr; 84 int *spec_id_ptr;
85 85
86 spec_id_ptr = bpf_map_lookup_e 86 spec_id_ptr = bpf_map_lookup_elem(&__bpf_usdt_ip_to_spec_id, &ip);
87 return spec_id_ptr ? *spec_id_ 87 return spec_id_ptr ? *spec_id_ptr : -ESRCH;
88 } 88 }
89 89
90 return bpf_get_attach_cookie(ctx); 90 return bpf_get_attach_cookie(ctx);
91 } 91 }
92 92
93 /* Return number of USDT arguments defined for 93 /* Return number of USDT arguments defined for currently traced USDT. */
94 __weak __hidden 94 __weak __hidden
95 int bpf_usdt_arg_cnt(struct pt_regs *ctx) 95 int bpf_usdt_arg_cnt(struct pt_regs *ctx)
96 { 96 {
97 struct __bpf_usdt_spec *spec; 97 struct __bpf_usdt_spec *spec;
98 int spec_id; 98 int spec_id;
99 99
100 spec_id = __bpf_usdt_spec_id(ctx); 100 spec_id = __bpf_usdt_spec_id(ctx);
101 if (spec_id < 0) 101 if (spec_id < 0)
102 return -ESRCH; 102 return -ESRCH;
103 103
104 spec = bpf_map_lookup_elem(&__bpf_usdt 104 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
105 if (!spec) 105 if (!spec)
106 return -ESRCH; 106 return -ESRCH;
107 107
108 return spec->arg_cnt; 108 return spec->arg_cnt;
109 } 109 }
110 110
111 /* Fetch USDT argument #*arg_num* (zero-indexe 111 /* Fetch USDT argument #*arg_num* (zero-indexed) and put its value into *res.
112 * Returns 0 on success; negative error, other 112 * Returns 0 on success; negative error, otherwise.
113 * On error *res is guaranteed to be set to ze 113 * On error *res is guaranteed to be set to zero.
114 */ 114 */
115 __weak __hidden 115 __weak __hidden
116 int bpf_usdt_arg(struct pt_regs *ctx, __u64 ar 116 int bpf_usdt_arg(struct pt_regs *ctx, __u64 arg_num, long *res)
117 { 117 {
118 struct __bpf_usdt_spec *spec; 118 struct __bpf_usdt_spec *spec;
119 struct __bpf_usdt_arg_spec *arg_spec; 119 struct __bpf_usdt_arg_spec *arg_spec;
120 unsigned long val; 120 unsigned long val;
121 int err, spec_id; 121 int err, spec_id;
122 122
123 *res = 0; 123 *res = 0;
124 124
125 spec_id = __bpf_usdt_spec_id(ctx); 125 spec_id = __bpf_usdt_spec_id(ctx);
126 if (spec_id < 0) 126 if (spec_id < 0)
127 return -ESRCH; 127 return -ESRCH;
128 128
129 spec = bpf_map_lookup_elem(&__bpf_usdt 129 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
130 if (!spec) 130 if (!spec)
131 return -ESRCH; 131 return -ESRCH;
132 132
133 if (arg_num >= BPF_USDT_MAX_ARG_CNT) 133 if (arg_num >= BPF_USDT_MAX_ARG_CNT)
134 return -ENOENT; 134 return -ENOENT;
135 barrier_var(arg_num); 135 barrier_var(arg_num);
136 if (arg_num >= spec->arg_cnt) 136 if (arg_num >= spec->arg_cnt)
137 return -ENOENT; 137 return -ENOENT;
138 138
139 arg_spec = &spec->args[arg_num]; 139 arg_spec = &spec->args[arg_num];
140 switch (arg_spec->arg_type) { 140 switch (arg_spec->arg_type) {
141 case BPF_USDT_ARG_CONST: 141 case BPF_USDT_ARG_CONST:
142 /* Arg is just a constant ("-4 142 /* Arg is just a constant ("-4@$-9" in USDT arg spec).
143 * value is recorded in arg_sp 143 * value is recorded in arg_spec->val_off directly.
144 */ 144 */
145 val = arg_spec->val_off; 145 val = arg_spec->val_off;
146 break; 146 break;
147 case BPF_USDT_ARG_REG: 147 case BPF_USDT_ARG_REG:
148 /* Arg is in a register (e.g, 148 /* Arg is in a register (e.g, "8@%rax" in USDT arg spec),
149 * so we read the contents of 149 * so we read the contents of that register directly from
150 * struct pt_regs. To keep thi 150 * struct pt_regs. To keep things simple user-space parts
151 * record offsetof(struct pt_r 151 * record offsetof(struct pt_regs, <regname>) in arg_spec->reg_off.
152 */ 152 */
153 err = bpf_probe_read_kernel(&v 153 err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
154 if (err) 154 if (err)
155 return err; 155 return err;
156 break; 156 break;
157 case BPF_USDT_ARG_REG_DEREF: 157 case BPF_USDT_ARG_REG_DEREF:
158 /* Arg is in memory addressed 158 /* Arg is in memory addressed by register, plus some offset
159 * (e.g., "-4@-1204(%rbp)" in 159 * (e.g., "-4@-1204(%rbp)" in USDT arg spec). Register is
160 * identified like with BPF_US 160 * identified like with BPF_USDT_ARG_REG case, and the offset
161 * is in arg_spec->val_off. We 161 * is in arg_spec->val_off. We first fetch register contents
162 * from pt_regs, then do anoth 162 * from pt_regs, then do another user-space probe read to
163 * fetch argument value itself 163 * fetch argument value itself.
164 */ 164 */
165 err = bpf_probe_read_kernel(&v 165 err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
166 if (err) 166 if (err)
167 return err; 167 return err;
168 err = bpf_probe_read_user(&val 168 err = bpf_probe_read_user(&val, sizeof(val), (void *)val + arg_spec->val_off);
169 if (err) 169 if (err)
170 return err; 170 return err;
171 #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 171 #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
172 val >>= arg_spec->arg_bitshift 172 val >>= arg_spec->arg_bitshift;
173 #endif 173 #endif
174 break; 174 break;
175 default: 175 default:
176 return -EINVAL; 176 return -EINVAL;
177 } 177 }
178 178
179 /* cast arg from 1, 2, or 4 bytes to f 179 /* cast arg from 1, 2, or 4 bytes to final 8 byte size clearing
180 * necessary upper arg_bitshift bits, 180 * necessary upper arg_bitshift bits, with sign extension if argument
181 * is signed 181 * is signed
182 */ 182 */
183 val <<= arg_spec->arg_bitshift; 183 val <<= arg_spec->arg_bitshift;
184 if (arg_spec->arg_signed) 184 if (arg_spec->arg_signed)
185 val = ((long)val) >> arg_spec- 185 val = ((long)val) >> arg_spec->arg_bitshift;
186 else 186 else
187 val = val >> arg_spec->arg_bit 187 val = val >> arg_spec->arg_bitshift;
188 *res = val; 188 *res = val;
189 return 0; 189 return 0;
190 } 190 }
191 191
192 /* Retrieve user-specified cookie value provid 192 /* Retrieve user-specified cookie value provided during attach as
193 * bpf_usdt_opts.usdt_cookie. This serves the 193 * bpf_usdt_opts.usdt_cookie. This serves the same purpose as BPF cookie
194 * returned by bpf_get_attach_cookie(). Libbpf 194 * returned by bpf_get_attach_cookie(). Libbpf's support for USDT is itself
195 * utilizing BPF cookies internally, so user c 195 * utilizing BPF cookies internally, so user can't use BPF cookie directly
196 * for USDT programs and has to use bpf_usdt_c 196 * for USDT programs and has to use bpf_usdt_cookie() API instead.
197 */ 197 */
198 __weak __hidden 198 __weak __hidden
199 long bpf_usdt_cookie(struct pt_regs *ctx) 199 long bpf_usdt_cookie(struct pt_regs *ctx)
200 { 200 {
201 struct __bpf_usdt_spec *spec; 201 struct __bpf_usdt_spec *spec;
202 int spec_id; 202 int spec_id;
203 203
204 spec_id = __bpf_usdt_spec_id(ctx); 204 spec_id = __bpf_usdt_spec_id(ctx);
205 if (spec_id < 0) 205 if (spec_id < 0)
206 return 0; 206 return 0;
207 207
208 spec = bpf_map_lookup_elem(&__bpf_usdt 208 spec = bpf_map_lookup_elem(&__bpf_usdt_specs, &spec_id);
209 if (!spec) 209 if (!spec)
210 return 0; 210 return 0;
211 211
212 return spec->usdt_cookie; 212 return spec->usdt_cookie;
213 } 213 }
214 214
215 /* we rely on ___bpf_apply() and ___bpf_narg() 215 /* we rely on ___bpf_apply() and ___bpf_narg() macros already defined in bpf_tracing.h */
216 #define ___bpf_usdt_args0() ctx 216 #define ___bpf_usdt_args0() ctx
217 #define ___bpf_usdt_args1(x) ___bpf_usdt_args0 217 #define ___bpf_usdt_args1(x) ___bpf_usdt_args0(), ({ long _x; bpf_usdt_arg(ctx, 0, &_x); _x; })
218 #define ___bpf_usdt_args2(x, args...) ___bpf_u 218 #define ___bpf_usdt_args2(x, args...) ___bpf_usdt_args1(args), ({ long _x; bpf_usdt_arg(ctx, 1, &_x); _x; })
219 #define ___bpf_usdt_args3(x, args...) ___bpf_u 219 #define ___bpf_usdt_args3(x, args...) ___bpf_usdt_args2(args), ({ long _x; bpf_usdt_arg(ctx, 2, &_x); _x; })
220 #define ___bpf_usdt_args4(x, args...) ___bpf_u 220 #define ___bpf_usdt_args4(x, args...) ___bpf_usdt_args3(args), ({ long _x; bpf_usdt_arg(ctx, 3, &_x); _x; })
221 #define ___bpf_usdt_args5(x, args...) ___bpf_u 221 #define ___bpf_usdt_args5(x, args...) ___bpf_usdt_args4(args), ({ long _x; bpf_usdt_arg(ctx, 4, &_x); _x; })
222 #define ___bpf_usdt_args6(x, args...) ___bpf_u 222 #define ___bpf_usdt_args6(x, args...) ___bpf_usdt_args5(args), ({ long _x; bpf_usdt_arg(ctx, 5, &_x); _x; })
223 #define ___bpf_usdt_args7(x, args...) ___bpf_u 223 #define ___bpf_usdt_args7(x, args...) ___bpf_usdt_args6(args), ({ long _x; bpf_usdt_arg(ctx, 6, &_x); _x; })
224 #define ___bpf_usdt_args8(x, args...) ___bpf_u 224 #define ___bpf_usdt_args8(x, args...) ___bpf_usdt_args7(args), ({ long _x; bpf_usdt_arg(ctx, 7, &_x); _x; })
225 #define ___bpf_usdt_args9(x, args...) ___bpf_u 225 #define ___bpf_usdt_args9(x, args...) ___bpf_usdt_args8(args), ({ long _x; bpf_usdt_arg(ctx, 8, &_x); _x; })
226 #define ___bpf_usdt_args10(x, args...) ___bpf_ 226 #define ___bpf_usdt_args10(x, args...) ___bpf_usdt_args9(args), ({ long _x; bpf_usdt_arg(ctx, 9, &_x); _x; })
227 #define ___bpf_usdt_args11(x, args...) ___bpf_ 227 #define ___bpf_usdt_args11(x, args...) ___bpf_usdt_args10(args), ({ long _x; bpf_usdt_arg(ctx, 10, &_x); _x; })
228 #define ___bpf_usdt_args12(x, args...) ___bpf_ 228 #define ___bpf_usdt_args12(x, args...) ___bpf_usdt_args11(args), ({ long _x; bpf_usdt_arg(ctx, 11, &_x); _x; })
229 #define ___bpf_usdt_args(args...) ___bpf_apply 229 #define ___bpf_usdt_args(args...) ___bpf_apply(___bpf_usdt_args, ___bpf_narg(args))(args)
230 230
231 /* 231 /*
232 * BPF_USDT serves the same purpose for USDT h 232 * BPF_USDT serves the same purpose for USDT handlers as BPF_PROG for
233 * tp_btf/fentry/fexit BPF programs and BPF_KP 233 * tp_btf/fentry/fexit BPF programs and BPF_KPROBE for kprobes.
234 * Original struct pt_regs * context is preser 234 * Original struct pt_regs * context is preserved as 'ctx' argument.
235 */ 235 */
236 #define BPF_USDT(name, args...) 236 #define BPF_USDT(name, args...) \
237 name(struct pt_regs *ctx); 237 name(struct pt_regs *ctx); \
238 static __always_inline typeof(name(0)) 238 static __always_inline typeof(name(0)) \
239 ____##name(struct pt_regs *ctx, ##args); 239 ____##name(struct pt_regs *ctx, ##args); \
240 typeof(name(0)) name(struct pt_regs *ctx) 240 typeof(name(0)) name(struct pt_regs *ctx) \
241 { 241 { \
242 _Pragma("GCC diagnostic push") 242 _Pragma("GCC diagnostic push") \
243 _Pragma("GCC diagnostic ignored \"-Win 243 _Pragma("GCC diagnostic ignored \"-Wint-conversion\"") \
244 return ____##name(___bpf_usdt_args(arg 244 return ____##name(___bpf_usdt_args(args)); \
245 _Pragma("GCC diagnostic pop") 245 _Pragma("GCC diagnostic pop") \
246 } 246 } \
247 static __always_inline typeof(name(0)) 247 static __always_inline typeof(name(0)) \
248 ____##name(struct pt_regs *ctx, ##args) 248 ____##name(struct pt_regs *ctx, ##args)
249 249
250 #endif /* __USDT_BPF_H__ */ 250 #endif /* __USDT_BPF_H__ */
251 251
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.