1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Augment the raw_syscalls tracepoints with the contents of the pointer arguments.
4 *
5 * Test it with:
6 *
7 * perf trace -e tools/perf/examples/bpf/augmented_raw_syscalls.c cat /etc/passwd > /dev/null
8 *
9 * This exactly matches what is marshalled into the raw_syscall:sys_enter
10 * payload expected by the 'perf trace' beautifiers.
11 *
12 * For now it just uses the existing tracepoint augmentation code in 'perf
13 * trace', in the next csets we'll hook up these with the sys_enter/sys_exit
14 * code that will combine entry/exit in a strace like way.
15 */
16
17 #include <unistd.h>
18 #include <linux/limits.h>
19 #include <linux/socket.h>
20 #include <pid_filter.h>
21
22 /* bpf-output associated map */
23 bpf_map(__augmented_syscalls__, PERF_EVENT_ARRAY, int, u32, __NR_CPUS__);
24
25 /*
26 * string_args_len: one per syscall arg, 0 means not a string or don't copy it,
27 * PATH_MAX for copying everything, any other value to limit
28 * it a la 'strace -s strsize'.
29 */
30 struct syscall {
31 bool enabled;
32 u16 string_args_len[6];
33 };
34
35 bpf_map(syscalls, ARRAY, int, struct syscall, 512);
36
37 /*
38 * What to augment at entry?
39 *
40 * Pointer arg payloads (filenames, etc) passed from userspace to the kernel
41 */
42 bpf_map(syscalls_sys_enter, PROG_ARRAY, u32, u32, 512);
43
44 /*
45 * What to augment at exit?
46 *
47 * Pointer arg payloads returned from the kernel (struct stat, etc) to userspace.
48 */
49 bpf_map(syscalls_sys_exit, PROG_ARRAY, u32, u32, 512);
50
51 struct syscall_enter_args {
52 unsigned long long common_tp_fields;
53 long syscall_nr;
54 unsigned long args[6];
55 };
56
57 struct syscall_exit_args {
58 unsigned long long common_tp_fields;
59 long syscall_nr;
60 long ret;
61 };
62
63 struct augmented_arg {
64 unsigned int size;
65 int err;
66 char value[PATH_MAX];
67 };
68
69 pid_filter(pids_filtered);
70
71 struct augmented_args_payload {
72 struct syscall_enter_args args;
73 union {
74 struct {
75 struct augmented_arg arg, arg2;
76 };
77 struct sockaddr_storage saddr;
78 };
79 };
80
81 // We need more tmp space than the BPF stack can give us
82 bpf_map(augmented_args_tmp, PERCPU_ARRAY, int, struct augmented_args_payload, 1);
83
augmented_args_payload(void)84 static inline struct augmented_args_payload *augmented_args_payload(void)
85 {
86 int key = 0;
87 return bpf_map_lookup_elem(&augmented_args_tmp, &key);
88 }
89
augmented__output(void * ctx,struct augmented_args_payload * args,int len)90 static inline int augmented__output(void *ctx, struct augmented_args_payload *args, int len)
91 {
92 /* If perf_event_output fails, return non-zero so that it gets recorded unaugmented */
93 return perf_event_output(ctx, &__augmented_syscalls__, BPF_F_CURRENT_CPU, args, len);
94 }
95
96 static inline
augmented_arg__read_str(struct augmented_arg * augmented_arg,const void * arg,unsigned int arg_len)97 unsigned int augmented_arg__read_str(struct augmented_arg *augmented_arg, const void *arg, unsigned int arg_len)
98 {
99 unsigned int augmented_len = sizeof(*augmented_arg);
100 int string_len = probe_read_str(&augmented_arg->value, arg_len, arg);
101
102 augmented_arg->size = augmented_arg->err = 0;
103 /*
104 * probe_read_str may return < 0, e.g. -EFAULT
105 * So we leave that in the augmented_arg->size that userspace will
106 */
107 if (string_len > 0) {
108 augmented_len -= sizeof(augmented_arg->value) - string_len;
109 augmented_len &= sizeof(augmented_arg->value) - 1;
110 augmented_arg->size = string_len;
111 } else {
112 /*
113 * So that username notice the error while still being able
114 * to skip this augmented arg record
115 */
116 augmented_arg->err = string_len;
117 augmented_len = offsetof(struct augmented_arg, value);
118 }
119
120 return augmented_len;
121 }
122
123 SEC("!raw_syscalls:unaugmented")
syscall_unaugmented(struct syscall_enter_args * args)124 int syscall_unaugmented(struct syscall_enter_args *args)
125 {
126 return 1;
127 }
128
129 /*
130 * These will be tail_called from SEC("raw_syscalls:sys_enter"), so will find in
131 * augmented_args_tmp what was read by that raw_syscalls:sys_enter and go
132 * on from there, reading the first syscall arg as a string, i.e. open's
133 * filename.
134 */
135 SEC("!syscalls:sys_enter_connect")
sys_enter_connect(struct syscall_enter_args * args)136 int sys_enter_connect(struct syscall_enter_args *args)
137 {
138 struct augmented_args_payload *augmented_args = augmented_args_payload();
139 const void *sockaddr_arg = (const void *)args->args[1];
140 unsigned int socklen = args->args[2];
141 unsigned int len = sizeof(augmented_args->args);
142
143 if (augmented_args == NULL)
144 return 1; /* Failure: don't filter */
145
146 if (socklen > sizeof(augmented_args->saddr))
147 socklen = sizeof(augmented_args->saddr);
148
149 probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
150
151 return augmented__output(args, augmented_args, len + socklen);
152 }
153
154 SEC("!syscalls:sys_enter_sendto")
sys_enter_sendto(struct syscall_enter_args * args)155 int sys_enter_sendto(struct syscall_enter_args *args)
156 {
157 struct augmented_args_payload *augmented_args = augmented_args_payload();
158 const void *sockaddr_arg = (const void *)args->args[4];
159 unsigned int socklen = args->args[5];
160 unsigned int len = sizeof(augmented_args->args);
161
162 if (augmented_args == NULL)
163 return 1; /* Failure: don't filter */
164
165 if (socklen > sizeof(augmented_args->saddr))
166 socklen = sizeof(augmented_args->saddr);
167
168 probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
169
170 return augmented__output(args, augmented_args, len + socklen);
171 }
172
173 SEC("!syscalls:sys_enter_open")
sys_enter_open(struct syscall_enter_args * args)174 int sys_enter_open(struct syscall_enter_args *args)
175 {
176 struct augmented_args_payload *augmented_args = augmented_args_payload();
177 const void *filename_arg = (const void *)args->args[0];
178 unsigned int len = sizeof(augmented_args->args);
179
180 if (augmented_args == NULL)
181 return 1; /* Failure: don't filter */
182
183 len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
184
185 return augmented__output(args, augmented_args, len);
186 }
187
188 SEC("!syscalls:sys_enter_openat")
sys_enter_openat(struct syscall_enter_args * args)189 int sys_enter_openat(struct syscall_enter_args *args)
190 {
191 struct augmented_args_payload *augmented_args = augmented_args_payload();
192 const void *filename_arg = (const void *)args->args[1];
193 unsigned int len = sizeof(augmented_args->args);
194
195 if (augmented_args == NULL)
196 return 1; /* Failure: don't filter */
197
198 len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
199
200 return augmented__output(args, augmented_args, len);
201 }
202
203 SEC("!syscalls:sys_enter_rename")
sys_enter_rename(struct syscall_enter_args * args)204 int sys_enter_rename(struct syscall_enter_args *args)
205 {
206 struct augmented_args_payload *augmented_args = augmented_args_payload();
207 const void *oldpath_arg = (const void *)args->args[0],
208 *newpath_arg = (const void *)args->args[1];
209 unsigned int len = sizeof(augmented_args->args), oldpath_len;
210
211 if (augmented_args == NULL)
212 return 1; /* Failure: don't filter */
213
214 oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
215 len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
216
217 return augmented__output(args, augmented_args, len);
218 }
219
220 SEC("!syscalls:sys_enter_renameat")
sys_enter_renameat(struct syscall_enter_args * args)221 int sys_enter_renameat(struct syscall_enter_args *args)
222 {
223 struct augmented_args_payload *augmented_args = augmented_args_payload();
224 const void *oldpath_arg = (const void *)args->args[1],
225 *newpath_arg = (const void *)args->args[3];
226 unsigned int len = sizeof(augmented_args->args), oldpath_len;
227
228 if (augmented_args == NULL)
229 return 1; /* Failure: don't filter */
230
231 oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
232 len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
233
234 return augmented__output(args, augmented_args, len);
235 }
236
237 SEC("raw_syscalls:sys_enter")
sys_enter(struct syscall_enter_args * args)238 int sys_enter(struct syscall_enter_args *args)
239 {
240 struct augmented_args_payload *augmented_args;
241 /*
242 * We start len, the amount of data that will be in the perf ring
243 * buffer, if this is not filtered out by one of pid_filter__has(),
244 * syscall->enabled, etc, with the non-augmented raw syscall payload,
245 * i.e. sizeof(augmented_args->args).
246 *
247 * We'll add to this as we add augmented syscalls right after that
248 * initial, non-augmented raw_syscalls:sys_enter payload.
249 */
250 unsigned int len = sizeof(augmented_args->args);
251 struct syscall *syscall;
252
253 if (pid_filter__has(&pids_filtered, getpid()))
254 return 0;
255
256 augmented_args = augmented_args_payload();
257 if (augmented_args == NULL)
258 return 1;
259
260 probe_read(&augmented_args->args, sizeof(augmented_args->args), args);
261
262 /*
263 * Jump to syscall specific augmenter, even if the default one,
264 * "!raw_syscalls:unaugmented" that will just return 1 to return the
265 * unagmented tracepoint payload.
266 */
267 bpf_tail_call(args, &syscalls_sys_enter, augmented_args->args.syscall_nr);
268
269 // If not found on the PROG_ARRAY syscalls map, then we're filtering it:
270 return 0;
271 }
272
273 SEC("raw_syscalls:sys_exit")
sys_exit(struct syscall_exit_args * args)274 int sys_exit(struct syscall_exit_args *args)
275 {
276 struct syscall_exit_args exit_args;
277
278 if (pid_filter__has(&pids_filtered, getpid()))
279 return 0;
280
281 probe_read(&exit_args, sizeof(exit_args), args);
282 /*
283 * Jump to syscall specific return augmenter, even if the default one,
284 * "!raw_syscalls:unaugmented" that will just return 1 to return the
285 * unagmented tracepoint payload.
286 */
287 bpf_tail_call(args, &syscalls_sys_exit, exit_args.syscall_nr);
288 /*
289 * If not found on the PROG_ARRAY syscalls map, then we're filtering it:
290 */
291 return 0;
292 }
293
294 license(GPL);
295
