xref: /openbmc/linux/tools/perf/util/bpf_skel/augmented_raw_syscalls.bpf.c (revision 5e6da6be3082f77be06894a1a94d52a90b4007dc)
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 <linux/bpf.h>
18 #include <bpf/bpf_helpers.h>
19 #include <linux/limits.h>
20 
21 #define MAX_CPUS  4096
22 
23 // FIXME: These should come from system headers
24 typedef char bool;
25 typedef int pid_t;
26 typedef long long int __s64;
27 typedef __s64 time64_t;
28 
29 struct timespec64 {
30 	time64_t	tv_sec;
31 	long int	tv_nsec;
32 };
33 
34 /* bpf-output associated map */
35 struct __augmented_syscalls__ {
36 	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
37 	__type(key, int);
38 	__type(value, __u32);
39 	__uint(max_entries, MAX_CPUS);
40 } __augmented_syscalls__ SEC(".maps");
41 
42 /*
43  * What to augment at entry?
44  *
45  * Pointer arg payloads (filenames, etc) passed from userspace to the kernel
46  */
47 struct syscalls_sys_enter {
48 	__uint(type, BPF_MAP_TYPE_PROG_ARRAY);
49 	__type(key, __u32);
50 	__type(value, __u32);
51 	__uint(max_entries, 512);
52 } syscalls_sys_enter SEC(".maps");
53 
54 /*
55  * What to augment at exit?
56  *
57  * Pointer arg payloads returned from the kernel (struct stat, etc) to userspace.
58  */
59 struct syscalls_sys_exit {
60 	__uint(type, BPF_MAP_TYPE_PROG_ARRAY);
61 	__type(key, __u32);
62 	__type(value, __u32);
63 	__uint(max_entries, 512);
64 } syscalls_sys_exit SEC(".maps");
65 
66 struct syscall_enter_args {
67 	unsigned long long common_tp_fields;
68 	long		   syscall_nr;
69 	unsigned long	   args[6];
70 };
71 
72 struct syscall_exit_args {
73 	unsigned long long common_tp_fields;
74 	long		   syscall_nr;
75 	long		   ret;
76 };
77 
78 struct augmented_arg {
79 	unsigned int	size;
80 	int		err;
81 	char		value[PATH_MAX];
82 };
83 
84 struct pids_filtered {
85 	__uint(type, BPF_MAP_TYPE_HASH);
86 	__type(key, pid_t);
87 	__type(value, bool);
88 	__uint(max_entries, 64);
89 } pids_filtered SEC(".maps");
90 
91 /*
92  * Desired design of maximum size and alignment (see RFC2553)
93  */
94 #define SS_MAXSIZE   128     /* Implementation specific max size */
95 
96 typedef unsigned short sa_family_t;
97 
98 /*
99  * FIXME: Should come from system headers
100  *
101  * The definition uses anonymous union and struct in order to control the
102  * default alignment.
103  */
104 struct sockaddr_storage {
105 	union {
106 		struct {
107 			sa_family_t    ss_family; /* address family */
108 			/* Following field(s) are implementation specific */
109 			char __data[SS_MAXSIZE - sizeof(unsigned short)];
110 				/* space to achieve desired size, */
111 				/* _SS_MAXSIZE value minus size of ss_family */
112 		};
113 		void *__align; /* implementation specific desired alignment */
114 	};
115 };
116 
117 struct augmented_args_payload {
118        struct syscall_enter_args args;
119        union {
120 		struct {
121 			struct augmented_arg arg, arg2;
122 		};
123 		struct sockaddr_storage saddr;
124 		char   __data[sizeof(struct augmented_arg)];
125 	};
126 };
127 
128 // We need more tmp space than the BPF stack can give us
129 struct augmented_args_tmp {
130 	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
131 	__type(key, int);
132 	__type(value, struct augmented_args_payload);
133 	__uint(max_entries, 1);
134 } augmented_args_tmp SEC(".maps");
135 
136 static inline struct augmented_args_payload *augmented_args_payload(void)
137 {
138 	int key = 0;
139 	return bpf_map_lookup_elem(&augmented_args_tmp, &key);
140 }
141 
142 static inline int augmented__output(void *ctx, struct augmented_args_payload *args, int len)
143 {
144 	/* If perf_event_output fails, return non-zero so that it gets recorded unaugmented */
145 	return bpf_perf_event_output(ctx, &__augmented_syscalls__, BPF_F_CURRENT_CPU, args, len);
146 }
147 
148 static inline
149 unsigned int augmented_arg__read_str(struct augmented_arg *augmented_arg, const void *arg, unsigned int arg_len)
150 {
151 	unsigned int augmented_len = sizeof(*augmented_arg);
152 	int string_len = bpf_probe_read_str(&augmented_arg->value, arg_len, arg);
153 
154 	augmented_arg->size = augmented_arg->err = 0;
155 	/*
156 	 * probe_read_str may return < 0, e.g. -EFAULT
157 	 * So we leave that in the augmented_arg->size that userspace will
158 	 */
159 	if (string_len > 0) {
160 		augmented_len -= sizeof(augmented_arg->value) - string_len;
161 		augmented_len &= sizeof(augmented_arg->value) - 1;
162 		augmented_arg->size = string_len;
163 	} else {
164 		/*
165 		 * So that username notice the error while still being able
166 		 * to skip this augmented arg record
167 		 */
168 		augmented_arg->err = string_len;
169 		augmented_len = offsetof(struct augmented_arg, value);
170 	}
171 
172 	return augmented_len;
173 }
174 
175 SEC("tp/raw_syscalls/sys_enter")
176 int syscall_unaugmented(struct syscall_enter_args *args)
177 {
178 	return 1;
179 }
180 
181 /*
182  * These will be tail_called from SEC("raw_syscalls:sys_enter"), so will find in
183  * augmented_args_tmp what was read by that raw_syscalls:sys_enter and go
184  * on from there, reading the first syscall arg as a string, i.e. open's
185  * filename.
186  */
187 SEC("tp/syscalls/sys_enter_connect")
188 int sys_enter_connect(struct syscall_enter_args *args)
189 {
190 	struct augmented_args_payload *augmented_args = augmented_args_payload();
191 	const void *sockaddr_arg = (const void *)args->args[1];
192 	unsigned int socklen = args->args[2];
193 	unsigned int len = sizeof(augmented_args->args);
194 
195         if (augmented_args == NULL)
196                 return 1; /* Failure: don't filter */
197 
198 	if (socklen > sizeof(augmented_args->saddr))
199 		socklen = sizeof(augmented_args->saddr);
200 
201 	bpf_probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
202 
203 	return augmented__output(args, augmented_args, len + socklen);
204 }
205 
206 SEC("tp/syscalls/sys_enter_sendto")
207 int sys_enter_sendto(struct syscall_enter_args *args)
208 {
209 	struct augmented_args_payload *augmented_args = augmented_args_payload();
210 	const void *sockaddr_arg = (const void *)args->args[4];
211 	unsigned int socklen = args->args[5];
212 	unsigned int len = sizeof(augmented_args->args);
213 
214         if (augmented_args == NULL)
215                 return 1; /* Failure: don't filter */
216 
217 	if (socklen > sizeof(augmented_args->saddr))
218 		socklen = sizeof(augmented_args->saddr);
219 
220 	bpf_probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
221 
222 	return augmented__output(args, augmented_args, len + socklen);
223 }
224 
225 SEC("tp/syscalls/sys_enter_open")
226 int sys_enter_open(struct syscall_enter_args *args)
227 {
228 	struct augmented_args_payload *augmented_args = augmented_args_payload();
229 	const void *filename_arg = (const void *)args->args[0];
230 	unsigned int len = sizeof(augmented_args->args);
231 
232         if (augmented_args == NULL)
233                 return 1; /* Failure: don't filter */
234 
235 	len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
236 
237 	return augmented__output(args, augmented_args, len);
238 }
239 
240 SEC("tp/syscalls/sys_enter_openat")
241 int sys_enter_openat(struct syscall_enter_args *args)
242 {
243 	struct augmented_args_payload *augmented_args = augmented_args_payload();
244 	const void *filename_arg = (const void *)args->args[1];
245 	unsigned int len = sizeof(augmented_args->args);
246 
247         if (augmented_args == NULL)
248                 return 1; /* Failure: don't filter */
249 
250 	len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
251 
252 	return augmented__output(args, augmented_args, len);
253 }
254 
255 SEC("tp/syscalls/sys_enter_rename")
256 int sys_enter_rename(struct syscall_enter_args *args)
257 {
258 	struct augmented_args_payload *augmented_args = augmented_args_payload();
259 	const void *oldpath_arg = (const void *)args->args[0],
260 		   *newpath_arg = (const void *)args->args[1];
261 	unsigned int len = sizeof(augmented_args->args), oldpath_len;
262 
263         if (augmented_args == NULL)
264                 return 1; /* Failure: don't filter */
265 
266 	oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
267 	len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
268 
269 	return augmented__output(args, augmented_args, len);
270 }
271 
272 SEC("tp/syscalls/sys_enter_renameat")
273 int sys_enter_renameat(struct syscall_enter_args *args)
274 {
275 	struct augmented_args_payload *augmented_args = augmented_args_payload();
276 	const void *oldpath_arg = (const void *)args->args[1],
277 		   *newpath_arg = (const void *)args->args[3];
278 	unsigned int len = sizeof(augmented_args->args), oldpath_len;
279 
280         if (augmented_args == NULL)
281                 return 1; /* Failure: don't filter */
282 
283 	oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
284 	len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
285 
286 	return augmented__output(args, augmented_args, len);
287 }
288 
289 #define PERF_ATTR_SIZE_VER0     64      /* sizeof first published struct */
290 
291 // we need just the start, get the size to then copy it
292 struct perf_event_attr_size {
293         __u32                   type;
294         /*
295          * Size of the attr structure, for fwd/bwd compat.
296          */
297         __u32                   size;
298 };
299 
300 SEC("tp/syscalls/sys_enter_perf_event_open")
301 int sys_enter_perf_event_open(struct syscall_enter_args *args)
302 {
303 	struct augmented_args_payload *augmented_args = augmented_args_payload();
304 	const struct perf_event_attr_size *attr = (const struct perf_event_attr_size *)args->args[0], *attr_read;
305 	unsigned int len = sizeof(augmented_args->args);
306 
307         if (augmented_args == NULL)
308 		goto failure;
309 
310 	if (bpf_probe_read(&augmented_args->__data, sizeof(*attr), attr) < 0)
311 		goto failure;
312 
313 	attr_read = (const struct perf_event_attr_size *)augmented_args->__data;
314 
315 	__u32 size = attr_read->size;
316 
317 	if (!size)
318 		size = PERF_ATTR_SIZE_VER0;
319 
320 	if (size > sizeof(augmented_args->__data))
321                 goto failure;
322 
323 	// Now that we read attr->size and tested it against the size limits, read it completely
324 	if (bpf_probe_read(&augmented_args->__data, size, attr) < 0)
325 		goto failure;
326 
327 	return augmented__output(args, augmented_args, len + size);
328 failure:
329 	return 1; /* Failure: don't filter */
330 }
331 
332 SEC("tp/syscalls/sys_enter_clock_nanosleep")
333 int sys_enter_clock_nanosleep(struct syscall_enter_args *args)
334 {
335 	struct augmented_args_payload *augmented_args = augmented_args_payload();
336 	const void *rqtp_arg = (const void *)args->args[2];
337 	unsigned int len = sizeof(augmented_args->args);
338 	__u32 size = sizeof(struct timespec64);
339 
340         if (augmented_args == NULL)
341 		goto failure;
342 
343 	if (size > sizeof(augmented_args->__data))
344                 goto failure;
345 
346 	bpf_probe_read(&augmented_args->__data, size, rqtp_arg);
347 
348 	return augmented__output(args, augmented_args, len + size);
349 failure:
350 	return 1; /* Failure: don't filter */
351 }
352 
353 static pid_t getpid(void)
354 {
355 	return bpf_get_current_pid_tgid();
356 }
357 
358 static bool pid_filter__has(struct pids_filtered *pids, pid_t pid)
359 {
360 	return bpf_map_lookup_elem(pids, &pid) != NULL;
361 }
362 
363 SEC("tp/raw_syscalls/sys_enter")
364 int sys_enter(struct syscall_enter_args *args)
365 {
366 	struct augmented_args_payload *augmented_args;
367 	/*
368 	 * We start len, the amount of data that will be in the perf ring
369 	 * buffer, if this is not filtered out by one of pid_filter__has(),
370 	 * syscall->enabled, etc, with the non-augmented raw syscall payload,
371 	 * i.e. sizeof(augmented_args->args).
372 	 *
373 	 * We'll add to this as we add augmented syscalls right after that
374 	 * initial, non-augmented raw_syscalls:sys_enter payload.
375 	 */
376 
377 	if (pid_filter__has(&pids_filtered, getpid()))
378 		return 0;
379 
380 	augmented_args = augmented_args_payload();
381 	if (augmented_args == NULL)
382 		return 1;
383 
384 	bpf_probe_read(&augmented_args->args, sizeof(augmented_args->args), args);
385 
386 	/*
387 	 * Jump to syscall specific augmenter, even if the default one,
388 	 * "!raw_syscalls:unaugmented" that will just return 1 to return the
389 	 * unaugmented tracepoint payload.
390 	 */
391 	bpf_tail_call(args, &syscalls_sys_enter, augmented_args->args.syscall_nr);
392 
393 	// If not found on the PROG_ARRAY syscalls map, then we're filtering it:
394 	return 0;
395 }
396 
397 SEC("tp/raw_syscalls/sys_exit")
398 int sys_exit(struct syscall_exit_args *args)
399 {
400 	struct syscall_exit_args exit_args;
401 
402 	if (pid_filter__has(&pids_filtered, getpid()))
403 		return 0;
404 
405 	bpf_probe_read(&exit_args, sizeof(exit_args), args);
406 	/*
407 	 * Jump to syscall specific return augmenter, even if the default one,
408 	 * "!raw_syscalls:unaugmented" that will just return 1 to return the
409 	 * unaugmented tracepoint payload.
410 	 */
411 	bpf_tail_call(args, &syscalls_sys_exit, exit_args.syscall_nr);
412 	/*
413 	 * If not found on the PROG_ARRAY syscalls map, then we're filtering it:
414 	 */
415 	return 0;
416 }
417 
418 char _license[] SEC("license") = "GPL";
419