1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Augment the raw_syscalls tracepoints with the contents of the pointer arguments.
4  *
5  * This exactly matches what is marshalled into the raw_syscall:sys_enter
6  * payload expected by the 'perf trace' beautifiers.
7  */
8 
9 #include <linux/bpf.h>
10 #include <bpf/bpf_helpers.h>
11 #include <linux/limits.h>
12 
13 /**
14  * is_power_of_2() - check if a value is a power of two
15  * @n: the value to check
16  *
17  * Determine whether some value is a power of two, where zero is *not*
18  * considered a power of two.  Return: true if @n is a power of 2, otherwise
19  * false.
20  */
21 #define is_power_of_2(n) (n != 0 && ((n & (n - 1)) == 0))
22 
23 #define MAX_CPUS  4096
24 
25 // FIXME: These should come from system headers
26 typedef char bool;
27 typedef int pid_t;
28 typedef long long int __s64;
29 typedef __s64 time64_t;
30 
31 struct timespec64 {
32 	time64_t	tv_sec;
33 	long int	tv_nsec;
34 };
35 
36 /* bpf-output associated map */
37 struct __augmented_syscalls__ {
38 	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
39 	__type(key, int);
40 	__type(value, __u32);
41 	__uint(max_entries, MAX_CPUS);
42 } __augmented_syscalls__ SEC(".maps");
43 
44 /*
45  * What to augment at entry?
46  *
47  * Pointer arg payloads (filenames, etc) passed from userspace to the kernel
48  */
49 struct syscalls_sys_enter {
50 	__uint(type, BPF_MAP_TYPE_PROG_ARRAY);
51 	__type(key, __u32);
52 	__type(value, __u32);
53 	__uint(max_entries, 512);
54 } syscalls_sys_enter SEC(".maps");
55 
56 /*
57  * What to augment at exit?
58  *
59  * Pointer arg payloads returned from the kernel (struct stat, etc) to userspace.
60  */
61 struct syscalls_sys_exit {
62 	__uint(type, BPF_MAP_TYPE_PROG_ARRAY);
63 	__type(key, __u32);
64 	__type(value, __u32);
65 	__uint(max_entries, 512);
66 } syscalls_sys_exit SEC(".maps");
67 
68 struct syscall_enter_args {
69 	unsigned long long common_tp_fields;
70 	long		   syscall_nr;
71 	unsigned long	   args[6];
72 };
73 
74 struct syscall_exit_args {
75 	unsigned long long common_tp_fields;
76 	long		   syscall_nr;
77 	long		   ret;
78 };
79 
80 struct augmented_arg {
81 	unsigned int	size;
82 	int		err;
83 	char		value[PATH_MAX];
84 };
85 
86 struct pids_filtered {
87 	__uint(type, BPF_MAP_TYPE_HASH);
88 	__type(key, pid_t);
89 	__type(value, bool);
90 	__uint(max_entries, 64);
91 } pids_filtered SEC(".maps");
92 
93 /*
94  * Desired design of maximum size and alignment (see RFC2553)
95  */
96 #define SS_MAXSIZE   128     /* Implementation specific max size */
97 
98 typedef unsigned short sa_family_t;
99 
100 /*
101  * FIXME: Should come from system headers
102  *
103  * The definition uses anonymous union and struct in order to control the
104  * default alignment.
105  */
106 struct sockaddr_storage {
107 	union {
108 		struct {
109 			sa_family_t    ss_family; /* address family */
110 			/* Following field(s) are implementation specific */
111 			char __data[SS_MAXSIZE - sizeof(unsigned short)];
112 				/* space to achieve desired size, */
113 				/* _SS_MAXSIZE value minus size of ss_family */
114 		};
115 		void *__align; /* implementation specific desired alignment */
116 	};
117 };
118 
119 struct augmented_args_payload {
120        struct syscall_enter_args args;
121        union {
122 		struct {
123 			struct augmented_arg arg, arg2;
124 		};
125 		struct sockaddr_storage saddr;
126 		char   __data[sizeof(struct augmented_arg)];
127 	};
128 };
129 
130 // We need more tmp space than the BPF stack can give us
131 struct augmented_args_tmp {
132 	__uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
133 	__type(key, int);
134 	__type(value, struct augmented_args_payload);
135 	__uint(max_entries, 1);
136 } augmented_args_tmp SEC(".maps");
137 
138 static inline struct augmented_args_payload *augmented_args_payload(void)
139 {
140 	int key = 0;
141 	return bpf_map_lookup_elem(&augmented_args_tmp, &key);
142 }
143 
144 static inline int augmented__output(void *ctx, struct augmented_args_payload *args, int len)
145 {
146 	/* If perf_event_output fails, return non-zero so that it gets recorded unaugmented */
147 	return bpf_perf_event_output(ctx, &__augmented_syscalls__, BPF_F_CURRENT_CPU, args, len);
148 }
149 
150 static inline
151 unsigned int augmented_arg__read_str(struct augmented_arg *augmented_arg, const void *arg, unsigned int arg_len)
152 {
153 	unsigned int augmented_len = sizeof(*augmented_arg);
154 	int string_len = bpf_probe_read_str(&augmented_arg->value, arg_len, arg);
155 
156 	augmented_arg->size = augmented_arg->err = 0;
157 	/*
158 	 * probe_read_str may return < 0, e.g. -EFAULT
159 	 * So we leave that in the augmented_arg->size that userspace will
160 	 */
161 	if (string_len > 0) {
162 		augmented_len -= sizeof(augmented_arg->value) - string_len;
163 		_Static_assert(is_power_of_2(sizeof(augmented_arg->value)), "sizeof(augmented_arg->value) needs to be a power of two");
164 		augmented_len &= sizeof(augmented_arg->value) - 1;
165 		augmented_arg->size = string_len;
166 	} else {
167 		/*
168 		 * So that username notice the error while still being able
169 		 * to skip this augmented arg record
170 		 */
171 		augmented_arg->err = string_len;
172 		augmented_len = offsetof(struct augmented_arg, value);
173 	}
174 
175 	return augmented_len;
176 }
177 
178 SEC("tp/raw_syscalls/sys_enter")
179 int syscall_unaugmented(struct syscall_enter_args *args)
180 {
181 	return 1;
182 }
183 
184 /*
185  * These will be tail_called from SEC("raw_syscalls:sys_enter"), so will find in
186  * augmented_args_tmp what was read by that raw_syscalls:sys_enter and go
187  * on from there, reading the first syscall arg as a string, i.e. open's
188  * filename.
189  */
190 SEC("tp/syscalls/sys_enter_connect")
191 int sys_enter_connect(struct syscall_enter_args *args)
192 {
193 	struct augmented_args_payload *augmented_args = augmented_args_payload();
194 	const void *sockaddr_arg = (const void *)args->args[1];
195 	unsigned int socklen = args->args[2];
196 	unsigned int len = sizeof(augmented_args->args);
197 
198         if (augmented_args == NULL)
199                 return 1; /* Failure: don't filter */
200 
201 	_Static_assert(is_power_of_2(sizeof(augmented_args->saddr)), "sizeof(augmented_args->saddr) needs to be a power of two");
202 	socklen &= sizeof(augmented_args->saddr) - 1;
203 
204 	bpf_probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
205 
206 	return augmented__output(args, augmented_args, len + socklen);
207 }
208 
209 SEC("tp/syscalls/sys_enter_sendto")
210 int sys_enter_sendto(struct syscall_enter_args *args)
211 {
212 	struct augmented_args_payload *augmented_args = augmented_args_payload();
213 	const void *sockaddr_arg = (const void *)args->args[4];
214 	unsigned int socklen = args->args[5];
215 	unsigned int len = sizeof(augmented_args->args);
216 
217         if (augmented_args == NULL)
218                 return 1; /* Failure: don't filter */
219 
220 	socklen &= sizeof(augmented_args->saddr) - 1;
221 
222 	bpf_probe_read(&augmented_args->saddr, socklen, sockaddr_arg);
223 
224 	return augmented__output(args, augmented_args, len + socklen);
225 }
226 
227 SEC("tp/syscalls/sys_enter_open")
228 int sys_enter_open(struct syscall_enter_args *args)
229 {
230 	struct augmented_args_payload *augmented_args = augmented_args_payload();
231 	const void *filename_arg = (const void *)args->args[0];
232 	unsigned int len = sizeof(augmented_args->args);
233 
234         if (augmented_args == NULL)
235                 return 1; /* Failure: don't filter */
236 
237 	len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
238 
239 	return augmented__output(args, augmented_args, len);
240 }
241 
242 SEC("tp/syscalls/sys_enter_openat")
243 int sys_enter_openat(struct syscall_enter_args *args)
244 {
245 	struct augmented_args_payload *augmented_args = augmented_args_payload();
246 	const void *filename_arg = (const void *)args->args[1];
247 	unsigned int len = sizeof(augmented_args->args);
248 
249         if (augmented_args == NULL)
250                 return 1; /* Failure: don't filter */
251 
252 	len += augmented_arg__read_str(&augmented_args->arg, filename_arg, sizeof(augmented_args->arg.value));
253 
254 	return augmented__output(args, augmented_args, len);
255 }
256 
257 SEC("tp/syscalls/sys_enter_rename")
258 int sys_enter_rename(struct syscall_enter_args *args)
259 {
260 	struct augmented_args_payload *augmented_args = augmented_args_payload();
261 	const void *oldpath_arg = (const void *)args->args[0],
262 		   *newpath_arg = (const void *)args->args[1];
263 	unsigned int len = sizeof(augmented_args->args), oldpath_len;
264 
265         if (augmented_args == NULL)
266                 return 1; /* Failure: don't filter */
267 
268 	oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
269 	len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
270 
271 	return augmented__output(args, augmented_args, len);
272 }
273 
274 SEC("tp/syscalls/sys_enter_renameat")
275 int sys_enter_renameat(struct syscall_enter_args *args)
276 {
277 	struct augmented_args_payload *augmented_args = augmented_args_payload();
278 	const void *oldpath_arg = (const void *)args->args[1],
279 		   *newpath_arg = (const void *)args->args[3];
280 	unsigned int len = sizeof(augmented_args->args), oldpath_len;
281 
282         if (augmented_args == NULL)
283                 return 1; /* Failure: don't filter */
284 
285 	oldpath_len = augmented_arg__read_str(&augmented_args->arg, oldpath_arg, sizeof(augmented_args->arg.value));
286 	len += oldpath_len + augmented_arg__read_str((void *)(&augmented_args->arg) + oldpath_len, newpath_arg, sizeof(augmented_args->arg.value));
287 
288 	return augmented__output(args, augmented_args, len);
289 }
290 
291 #define PERF_ATTR_SIZE_VER0     64      /* sizeof first published struct */
292 
293 // we need just the start, get the size to then copy it
294 struct perf_event_attr_size {
295         __u32                   type;
296         /*
297          * Size of the attr structure, for fwd/bwd compat.
298          */
299         __u32                   size;
300 };
301 
302 SEC("tp/syscalls/sys_enter_perf_event_open")
303 int sys_enter_perf_event_open(struct syscall_enter_args *args)
304 {
305 	struct augmented_args_payload *augmented_args = augmented_args_payload();
306 	const struct perf_event_attr_size *attr = (const struct perf_event_attr_size *)args->args[0], *attr_read;
307 	unsigned int len = sizeof(augmented_args->args);
308 
309         if (augmented_args == NULL)
310 		goto failure;
311 
312 	if (bpf_probe_read(&augmented_args->__data, sizeof(*attr), attr) < 0)
313 		goto failure;
314 
315 	attr_read = (const struct perf_event_attr_size *)augmented_args->__data;
316 
317 	__u32 size = attr_read->size;
318 
319 	if (!size)
320 		size = PERF_ATTR_SIZE_VER0;
321 
322 	if (size > sizeof(augmented_args->__data))
323                 goto failure;
324 
325 	// Now that we read attr->size and tested it against the size limits, read it completely
326 	if (bpf_probe_read(&augmented_args->__data, size, attr) < 0)
327 		goto failure;
328 
329 	return augmented__output(args, augmented_args, len + size);
330 failure:
331 	return 1; /* Failure: don't filter */
332 }
333 
334 SEC("tp/syscalls/sys_enter_clock_nanosleep")
335 int sys_enter_clock_nanosleep(struct syscall_enter_args *args)
336 {
337 	struct augmented_args_payload *augmented_args = augmented_args_payload();
338 	const void *rqtp_arg = (const void *)args->args[2];
339 	unsigned int len = sizeof(augmented_args->args);
340 	__u32 size = sizeof(struct timespec64);
341 
342         if (augmented_args == NULL)
343 		goto failure;
344 
345 	if (size > sizeof(augmented_args->__data))
346                 goto failure;
347 
348 	bpf_probe_read(&augmented_args->__data, size, rqtp_arg);
349 
350 	return augmented__output(args, augmented_args, len + size);
351 failure:
352 	return 1; /* Failure: don't filter */
353 }
354 
355 static pid_t getpid(void)
356 {
357 	return bpf_get_current_pid_tgid();
358 }
359 
360 static bool pid_filter__has(struct pids_filtered *pids, pid_t pid)
361 {
362 	return bpf_map_lookup_elem(pids, &pid) != NULL;
363 }
364 
365 SEC("tp/raw_syscalls/sys_enter")
366 int sys_enter(struct syscall_enter_args *args)
367 {
368 	struct augmented_args_payload *augmented_args;
369 	/*
370 	 * We start len, the amount of data that will be in the perf ring
371 	 * buffer, if this is not filtered out by one of pid_filter__has(),
372 	 * syscall->enabled, etc, with the non-augmented raw syscall payload,
373 	 * i.e. sizeof(augmented_args->args).
374 	 *
375 	 * We'll add to this as we add augmented syscalls right after that
376 	 * initial, non-augmented raw_syscalls:sys_enter payload.
377 	 */
378 
379 	if (pid_filter__has(&pids_filtered, getpid()))
380 		return 0;
381 
382 	augmented_args = augmented_args_payload();
383 	if (augmented_args == NULL)
384 		return 1;
385 
386 	bpf_probe_read(&augmented_args->args, sizeof(augmented_args->args), args);
387 
388 	/*
389 	 * Jump to syscall specific augmenter, even if the default one,
390 	 * "!raw_syscalls:unaugmented" that will just return 1 to return the
391 	 * unaugmented tracepoint payload.
392 	 */
393 	bpf_tail_call(args, &syscalls_sys_enter, augmented_args->args.syscall_nr);
394 
395 	// If not found on the PROG_ARRAY syscalls map, then we're filtering it:
396 	return 0;
397 }
398 
399 SEC("tp/raw_syscalls/sys_exit")
400 int sys_exit(struct syscall_exit_args *args)
401 {
402 	struct syscall_exit_args exit_args;
403 
404 	if (pid_filter__has(&pids_filtered, getpid()))
405 		return 0;
406 
407 	bpf_probe_read(&exit_args, sizeof(exit_args), args);
408 	/*
409 	 * Jump to syscall specific return augmenter, even if the default one,
410 	 * "!raw_syscalls:unaugmented" that will just return 1 to return the
411 	 * unaugmented tracepoint payload.
412 	 */
413 	bpf_tail_call(args, &syscalls_sys_exit, exit_args.syscall_nr);
414 	/*
415 	 * If not found on the PROG_ARRAY syscalls map, then we're filtering it:
416 	 */
417 	return 0;
418 }
419 
420 char _license[] SEC("license") = "GPL";
421