utrace: Support non-contiguous syscall numbers
[project/procd.git] / trace / trace.c
1 /*
2  * Copyright (C) 2015 John Crispin <blogic@openwrt.org>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU Lesser General Public License version 2.1
6  * as published by the Free Software Foundation
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  */
13
14 #define _GNU_SOURCE
15 #include <fcntl.h>
16 #include <stddef.h>
17 #include <sys/ptrace.h>
18 #include <sys/stat.h>
19 #include <sys/types.h>
20 #include <sys/user.h>
21 #include <sys/wait.h>
22 #include <unistd.h>
23 #include <stdlib.h>
24 #include <stdio.h>
25 #include <errno.h>
26 #include <string.h>
27 #include <syslog.h>
28 #include <err.h>
29
30 #ifndef PTRACE_EVENT_STOP
31 /* PTRACE_EVENT_STOP is defined in linux/ptrace.h, but this header
32  * collides with musl's sys/ptrace.h */
33 #define PTRACE_EVENT_STOP 128
34 #endif
35
36 #include <libubox/uloop.h>
37 #include <libubox/blobmsg.h>
38 #include <libubox/blobmsg_json.h>
39
40 #include "../syscall-names.h"
41
42 #define _offsetof(a, b) __builtin_offsetof(a,b)
43 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
44
45 #ifdef __amd64__
46 #define reg_syscall_nr  _offsetof(struct user, regs.orig_rax)
47 #elif defined(__i386__)
48 #define reg_syscall_nr  _offsetof(struct user, regs.orig_eax)
49 #elif defined(__mips)
50 # ifndef EF_REG2
51 # define EF_REG2        8
52 # endif
53 #define reg_syscall_nr  (EF_REG2 / 4)
54 #elif defined(__arm__)
55 #include <asm/ptrace.h>         /* for PTRACE_SET_SYSCALL */
56 #define reg_syscall_nr  _offsetof(struct user, regs.uregs[7])
57 # if defined(__ARM_EABI__)
58 # define reg_retval_nr  _offsetof(struct user, regs.uregs[0])
59 # endif
60 #else
61 #error tracing is not supported on this architecture
62 #endif
63
64 enum mode {
65         UTRACE,
66         SECCOMP_TRACE,
67 } mode = UTRACE;
68
69 #define PROC_NAME(mode) (mode == UTRACE ? "utrace" : "seccomp-trace")
70
71 #define INFO(fmt, ...) do { \
72         fprintf(stderr, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
73 } while (0)
74
75 #define ERROR(fmt, ...) do { \
76         syslog(LOG_ERR, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
77         fprintf(stderr, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__);  \
78 } while (0)
79
80 #define LOGERR(fmt, ...) do { \
81         syslog(LOG_ERR, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
82 } while (0)
83
84 struct tracee {
85         struct uloop_process proc;
86         int in_syscall;
87 };
88
89 static struct tracee tracer;
90 static int syscall_count[SYSCALL_COUNT];
91 static int violation_count;
92 static struct blob_buf b;
93 static int debug;
94 char *json = NULL;
95 int ptrace_restart;
96
97 static void set_syscall(const char *name, int val)
98 {
99         int i;
100
101         for (i = 0; i < SYSCALL_COUNT; i++) {
102                 int sc = syscall_index_to_number(i);
103                 if (syscall_name(sc) && !strcmp(syscall_name(sc), name)) {
104                         syscall_count[i] = val;
105                         return;
106                 }
107         }
108 }
109
110 struct syscall {
111         int syscall;
112         int count;
113 };
114
115 static int cmp_count(const void *a, const void *b)
116 {
117         return ((struct syscall*)b)->count - ((struct syscall*)a)->count;
118 }
119
120 static void print_syscalls(int policy, const char *json)
121 {
122         void *c;
123         int i;
124
125         if (mode == UTRACE) {
126                 set_syscall("rt_sigaction", 1);
127                 set_syscall("sigreturn", 1);
128                 set_syscall("rt_sigreturn", 1);
129                 set_syscall("exit_group", 1);
130                 set_syscall("exit", 1);
131         }
132
133         struct syscall sorted[SYSCALL_COUNT];
134
135         for (i = 0; i < SYSCALL_COUNT; i++) {
136                 sorted[i].syscall = syscall_index_to_number(i);
137                 sorted[i].count = syscall_count[i];
138         }
139
140         qsort(sorted, SYSCALL_COUNT, sizeof(sorted[0]), cmp_count);
141
142         blob_buf_init(&b, 0);
143         c = blobmsg_open_array(&b, "whitelist");
144
145         for (i = 0; i < SYSCALL_COUNT; i++) {
146                 int sc = sorted[i].syscall;
147                 if (!sorted[i].count)
148                         break;
149                 if (syscall_name(sc)) {
150                         if (debug)
151                                 printf("syscall %d (%s) was called %d times\n",
152                                        sc, syscall_name(sc), sorted[i].count);
153                         blobmsg_add_string(&b, NULL, syscall_name(sc));
154                 } else {
155                         ERROR("no name found for syscall(%d)\n", sc);
156                 }
157         }
158         blobmsg_close_array(&b, c);
159         blobmsg_add_u32(&b, "policy", policy);
160         if (json) {
161                 FILE *fp = fopen(json, "w");
162                 if (fp) {
163                         fprintf(fp, "%s", blobmsg_format_json_indent(b.head, true, 0));
164                         fclose(fp);
165                         INFO("saving syscall trace to %s\n", json);
166                 } else {
167                         ERROR("failed to open %s\n", json);
168                 }
169         } else {
170                 printf("%s\n",
171                         blobmsg_format_json_indent(b.head, true, 0));
172         }
173
174 }
175
176 static void report_seccomp_vialation(pid_t pid, unsigned syscall)
177 {
178         char buf[200];
179         snprintf(buf, sizeof(buf), "/proc/%d/cmdline", pid);
180         int f = open(buf, O_RDONLY);
181         int r = read(f, buf, sizeof(buf) - 1);
182         if (r >= 0)
183                 buf[r] = 0;
184         else
185                 strcpy(buf, "unknown?");
186         close(f);
187
188         if (violation_count < INT_MAX)
189                 violation_count++;
190         int i = syscall_index(syscall);
191         if (i >= 0) {
192                 syscall_count[i]++;
193                 LOGERR("%s[%u] tried to call non-whitelisted syscall: %s (see %s)\n",
194                        buf, pid,  syscall_name(syscall), json);
195         } else {
196                 LOGERR("%s[%u] tried to call non-whitelisted syscall: %d (see %s)\n",
197                        buf, pid,  syscall, json);
198         }
199 }
200
201 static void tracer_cb(struct uloop_process *c, int ret)
202 {
203         struct tracee *tracee = container_of(c, struct tracee, proc);
204         int inject_signal = 0;
205
206         /* We explicitely check for events in upper 16 bits, because
207          * musl (as opposed to glibc) does not report
208          * PTRACE_EVENT_STOP as WIFSTOPPED */
209         if (WIFSTOPPED(ret) || (ret >> 16)) {
210                 if (WSTOPSIG(ret) & 0x80) {
211                         if (!tracee->in_syscall) {
212                                 int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
213                                 int i = syscall_index(syscall);
214                                 if (i >= 0) {
215                                         syscall_count[i]++;
216                                         if (debug)
217                                                 fprintf(stderr, "%s()\n", syscall_name(syscall));
218                                 } else if (debug) {
219                                         fprintf(stderr, "syscal(%d)\n", syscall);
220                                 }
221                         }
222                         tracee->in_syscall = !tracee->in_syscall;
223                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_FORK << 8)) ||
224                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_VFORK << 8)) ||
225                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_CLONE << 8))) {
226                         struct tracee *child = calloc(1, sizeof(struct tracee));
227
228                         ptrace(PTRACE_GETEVENTMSG, c->pid, 0, &child->proc.pid);
229                         child->proc.cb = tracer_cb;
230                         ptrace(ptrace_restart, child->proc.pid, 0, 0);
231                         uloop_process_add(&child->proc);
232                         if (debug)
233                                 fprintf(stderr, "Tracing new child %d\n", child->proc.pid);
234                 } else if ((ret >> 16) == PTRACE_EVENT_STOP) {
235                         /* Nothing special to do here */
236                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_SECCOMP << 8))) {
237                         int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
238 #if defined(__arm__)
239                         ptrace(PTRACE_SET_SYSCALL, c->pid, 0, -1);
240                         ptrace(PTRACE_POKEUSER, c->pid, reg_retval_nr, -ENOSYS);
241 #else
242                         ptrace(PTRACE_POKEUSER, c->pid, reg_syscall_nr, -1);
243 #endif
244                         report_seccomp_vialation(c->pid, syscall);
245                 } else {
246                         inject_signal = WSTOPSIG(ret);
247                         if (debug)
248                                 fprintf(stderr, "Injecting signal %d into pid %d\n",
249                                         inject_signal, tracee->proc.pid);
250                 }
251         } else if (WIFEXITED(ret) || (WIFSIGNALED(ret) && WTERMSIG(ret))) {
252                 if (tracee == &tracer) {
253                         uloop_end(); /* Main process exit */
254                 } else {
255                         if (debug)
256                                 fprintf(stderr, "Child %d exited\n", tracee->proc.pid);
257                         free(tracee);
258                 }
259                 return;
260         }
261
262         ptrace(ptrace_restart, c->pid, 0, inject_signal);
263         uloop_process_add(c);
264 }
265
266 static void sigterm_handler(int signum)
267 {
268         /* When we receive SIGTERM, we forward it to the tracee. After
269          * the tracee exits, trace_cb() will be called and make us
270          * exit too. */
271         kill(tracer.proc.pid, SIGTERM);
272 }
273
274
275 int main(int argc, char **argv, char **envp)
276 {
277         int status, ch, policy = EPERM;
278         pid_t child;
279
280         /* When invoked via seccomp-trace symlink, work as seccomp
281          * violation logger rather than as syscall tracer */
282         if (strstr(argv[0], "seccomp-trace"))
283                 mode = SECCOMP_TRACE;
284
285         while ((ch = getopt(argc, argv, "f:p:")) != -1) {
286                 switch (ch) {
287                 case 'f':
288                         json = optarg;
289                         break;
290                 case 'p':
291                         policy = atoi(optarg);
292                         break;
293                 }
294         }
295
296         if (!json)
297                 json = getenv("SECCOMP_FILE");
298
299         argc -= optind;
300         argv += optind;
301
302         if (!argc)
303                 return -1;
304
305         if (getenv("TRACE_DEBUG"))
306                 debug = 1;
307         unsetenv("TRACE_DEBUG");
308
309         child = fork();
310
311         if (child == 0) {
312                 char **_argv = calloc(argc + 1, sizeof(char *));
313                 char **_envp;
314                 char *preload = NULL;
315                 const char *old_preload = getenv("LD_PRELOAD");
316                 int newenv = 0;
317                 int envc = 0;
318                 int ret;
319
320                 memcpy(_argv, argv, argc * sizeof(char *));
321
322                 while (envp[envc++])
323                         ;
324
325                 _envp = calloc(envc + 2, sizeof(char *));
326                 switch (mode) {
327                 case UTRACE:
328                         preload = "/lib/libpreload-trace.so";
329                         newenv = 1;
330                         break;
331                 case SECCOMP_TRACE:
332                         preload = "/lib/libpreload-seccomp.so";
333                         newenv = 2;
334                         asprintf(&_envp[1], "SECCOMP_FILE=%s", json ? json : "");
335                         kill(getpid(), SIGSTOP);
336                         break;
337                 }
338                 asprintf(&_envp[0], "LD_PRELOAD=%s%s%s", preload,
339                          old_preload ? ":" : "",
340                          old_preload ? old_preload : "");
341                 memcpy(&_envp[newenv], envp, envc * sizeof(char *));
342
343                 ret = execve(_argv[0], _argv, _envp);
344                 ERROR("failed to exec %s: %s\n", _argv[0], strerror(errno));
345
346                 free(_argv);
347                 free(_envp);
348                 return ret;
349         }
350
351         if (child < 0)
352                 return -1;
353
354         waitpid(child, &status, WUNTRACED);
355         if (!WIFSTOPPED(status)) {
356                 ERROR("failed to start %s\n", *argv);
357                 return -1;
358         }
359
360         int ptrace_options = PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK | PTRACE_O_TRACECLONE;
361         switch (mode) {
362         case UTRACE:
363                 ptrace_options |= PTRACE_O_TRACESYSGOOD;
364                 ptrace_restart = PTRACE_SYSCALL;
365                 break;
366         case SECCOMP_TRACE:
367                 ptrace_options |= PTRACE_O_TRACESECCOMP;
368                 ptrace_restart = PTRACE_CONT;
369                 break;
370         }
371         if (ptrace(PTRACE_SEIZE, child, 0, ptrace_options) == -1)
372                 err(1, "PTRACE_SEIZE");
373         if (ptrace(ptrace_restart, child, 0, SIGCONT) == -1)
374                 err(1, "ptrace restart");
375
376         uloop_init();
377         tracer.proc.pid = child;
378         tracer.proc.cb = tracer_cb;
379         uloop_process_add(&tracer.proc);
380         signal(SIGTERM, sigterm_handler); /* Override uloop's SIGTERM handler */
381         uloop_run();
382         uloop_done();
383
384
385         switch (mode) {
386         case UTRACE:
387                 if (!json)
388                         if (asprintf(&json, "/tmp/%s.%u.json", basename(*argv), child) < 0)
389                                 ERROR("failed to allocate output path: %s\n", strerror(errno));
390                 break;
391         case SECCOMP_TRACE:
392                 if (!violation_count)
393                         return 0;
394                 asprintf(&json, "/tmp/%s.%u.violations.json", basename(*argv), child);
395                 break;
396         }
397         print_syscalls(policy, json);
398         return 0;
399 }