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 <limits.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 #ifndef PTRACE_EVENT_SECCOMP
  37 /* undefined with uClibc-ng */
  38 #define PTRACE_EVENT_SECCOMP 7
  39 #endif
  40
  41 #include <libubox/ulog.h>
  42 #include <libubox/uloop.h>
  43 #include <libubox/blobmsg.h>
  44 #include <libubox/blobmsg_json.h>
  45
  46 #include "../syscall-names.h"
  47
  48 #define _offsetof(a, b) __builtin_offsetof(a,b)
  49 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
  50
  51 #ifdef __amd64__
  52 #define reg_syscall_nr  _offsetof(struct user, regs.orig_rax)
  53 #elif defined(__i386__)
  54 #define reg_syscall_nr  _offsetof(struct user, regs.orig_eax)
  55 #elif defined(__mips)
  56 # ifndef EF_REG2
  57 # define EF_REG2        8
  58 # endif
  59 #define reg_syscall_nr  (EF_REG2 / 4)
  60 #elif defined(__arm__)
  61 #include <asm/ptrace.h>         /* for PTRACE_SET_SYSCALL */
  62 #define reg_syscall_nr  _offsetof(struct user, regs.uregs[7])
  63 # if defined(__ARM_EABI__)
  64 # define reg_retval_nr  _offsetof(struct user, regs.uregs[0])
  65 # endif
  66 #elif defined(__PPC__)
  67 #define reg_syscall_nr  _offsetof(struct user, regs.gpr[0])
  68 #define reg_retval_nr   _offsetof(struct user, regs.gpr[3])
  69 #else
  70 #error tracing is not supported on this architecture
  71 #endif
  72
  73 enum mode {
  74         UTRACE,
  75         SECCOMP_TRACE,
  76 } mode = UTRACE;
  77
  78 struct tracee {
  79         struct uloop_process proc;
  80         int in_syscall;
  81 };
  82
  83 static struct tracee tracer;
  84 static int syscall_count[SYSCALL_COUNT];
  85 static int violation_count;
  86 static struct blob_buf b;
  87 static int debug;
  88 char *json = NULL;
  89 int ptrace_restart;
  90
  91 static void set_syscall(const char *name, int val)
  92 {
  93         int i;
  94
  95         for (i = 0; i < SYSCALL_COUNT; i++) {
  96                 int sc = syscall_index_to_number(i);
  97                 if (syscall_name(sc) && !strcmp(syscall_name(sc), name)) {
  98                         syscall_count[i] = val;
  99                         return;
 100                 }
 101         }
 102 }
 103
 104 struct syscall {
 105         int syscall;
 106         int count;
 107 };
 108
 109 static int cmp_count(const void *a, const void *b)
 110 {
 111         return ((struct syscall*)b)->count - ((struct syscall*)a)->count;
 112 }
 113
 114 static void print_syscalls(int policy, const char *json)
 115 {
 116         void *c, *d, *e;
 117         int i;
 118
 119         if (mode == UTRACE) {
 120                 set_syscall("rt_sigaction", 1);
 121                 set_syscall("sigreturn", 1);
 122                 set_syscall("rt_sigreturn", 1);
 123                 set_syscall("exit_group", 1);
 124                 set_syscall("exit", 1);
 125         }
 126
 127         struct syscall sorted[SYSCALL_COUNT];
 128
 129         for (i = 0; i < SYSCALL_COUNT; i++) {
 130                 sorted[i].syscall = syscall_index_to_number(i);
 131                 sorted[i].count = syscall_count[i];
 132         }
 133
 134         qsort(sorted, SYSCALL_COUNT, sizeof(sorted[0]), cmp_count);
 135
 136         blob_buf_init(&b, 0);
 137         blobmsg_add_string(&b, "defaultAction", "SCMP_ACT_KILL_PROCESS");
 138         c = blobmsg_open_array(&b, "syscalls");
 139         d = blobmsg_open_table(&b, "");
 140         e = blobmsg_open_array(&b, "names");
 141
 142         for (i = 0; i < SYSCALL_COUNT; i++) {
 143                 int sc = sorted[i].syscall;
 144                 if (!sorted[i].count)
 145                         break;
 146                 if (syscall_name(sc)) {
 147                         if (debug)
 148                                 printf("syscall %d (%s) was called %d times\n",
 149                                        sc, syscall_name(sc), sorted[i].count);
 150                         blobmsg_add_string(&b, NULL, syscall_name(sc));
 151                 } else {
 152                         ULOG_ERR("no name found for syscall(%d)\n", sc);
 153                 }
 154         }
 155         blobmsg_close_array(&b, e);
 156         blobmsg_add_string(&b, "action", "SCMP_ACT_ALLOW");
 157         blobmsg_close_table(&b, d);
 158         blobmsg_close_array(&b, c);
 159         if (json) {
 160                 FILE *fp = fopen(json, "w");
 161                 if (fp) {
 162                         fprintf(fp, "%s\n", blobmsg_format_json_indent(b.head, true, 0));
 163                         fclose(fp);
 164                         ULOG_INFO("saving syscall trace to %s\n", json);
 165                 } else {
 166                         ULOG_ERR("failed to open %s\n", json);
 167                 }
 168         } else {
 169                 printf("%s\n",
 170                         blobmsg_format_json_indent(b.head, true, 0));
 171         }
 172
 173 }
 174
 175 static void report_seccomp_vialation(pid_t pid, unsigned syscall)
 176 {
 177         char buf[200];
 178         snprintf(buf, sizeof(buf), "/proc/%d/cmdline", pid);
 179         int f = open(buf, O_RDONLY);
 180         int r = read(f, buf, sizeof(buf) - 1);
 181         if (r >= 0)
 182                 buf[r] = 0;
 183         else
 184                 strcpy(buf, "unknown?");
 185         close(f);
 186
 187         if (violation_count < INT_MAX)
 188                 violation_count++;
 189         int i = syscall_index(syscall);
 190         if (i >= 0) {
 191                 syscall_count[i]++;
 192                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %s (see %s)\n",
 193                          buf, pid,  syscall_name(syscall), json);
 194         } else {
 195                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %d (see %s)\n",
 196                          buf, pid,  syscall, json);
 197         }
 198 }
 199
 200 static void tracer_cb(struct uloop_process *c, int ret)
 201 {
 202         struct tracee *tracee = container_of(c, struct tracee, proc);
 203         int inject_signal = 0;
 204
 205         /* We explicitely check for events in upper 16 bits, because
 206          * musl (as opposed to glibc) does not report
 207          * PTRACE_EVENT_STOP as WIFSTOPPED */
 208         if (WIFSTOPPED(ret) || (ret >> 16)) {
 209                 if (WSTOPSIG(ret) & 0x80) {
 210                         if (!tracee->in_syscall) {
 211                                 int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 212                                 int i = syscall_index(syscall);
 213                                 if (i >= 0) {
 214                                         syscall_count[i]++;
 215                                         if (debug)
 216                                                 fprintf(stderr, "%s()\n", syscall_name(syscall));
 217                                 } else if (debug) {
 218                                         fprintf(stderr, "syscal(%d)\n", syscall);
 219                                 }
 220                         }
 221                         tracee->in_syscall = !tracee->in_syscall;
 222                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_FORK << 8)) ||
 223                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_VFORK << 8)) ||
 224                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_CLONE << 8))) {
 225                         struct tracee *child = calloc(1, sizeof(struct tracee));
 226
 227                         unsigned long msg;
 228                         ptrace(PTRACE_GETEVENTMSG, c->pid, 0, &msg);
 229                         child->proc.pid = msg;
 230                         child->proc.cb = tracer_cb;
 231                         ptrace(ptrace_restart, child->proc.pid, 0, 0);
 232                         uloop_process_add(&child->proc);
 233                         if (debug)
 234                                 fprintf(stderr, "Tracing new child %d\n", child->proc.pid);
 235                 } else if ((ret >> 16) == PTRACE_EVENT_STOP) {
 236                         /* Nothing special to do here */
 237                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_SECCOMP << 8))) {
 238                         int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 239 #if defined(__arm__)
 240                         ptrace(PTRACE_SET_SYSCALL, c->pid, 0, -1);
 241                         ptrace(PTRACE_POKEUSER, c->pid, reg_retval_nr, -ENOSYS);
 242 #else
 243                         ptrace(PTRACE_POKEUSER, c->pid, reg_syscall_nr, -1);
 244 #endif
 245                         report_seccomp_vialation(c->pid, syscall);
 246                 } else {
 247                         inject_signal = WSTOPSIG(ret);
 248                         if (debug)
 249                                 fprintf(stderr, "Injecting signal %d into pid %d\n",
 250                                         inject_signal, tracee->proc.pid);
 251                 }
 252         } else if (WIFEXITED(ret) || (WIFSIGNALED(ret) && WTERMSIG(ret))) {
 253                 if (tracee == &tracer) {
 254                         uloop_end(); /* Main process exit */
 255                 } else {
 256                         if (debug)
 257                                 fprintf(stderr, "Child %d exited\n", tracee->proc.pid);
 258                         free(tracee);
 259                 }
 260                 return;
 261         }
 262
 263         ptrace(ptrace_restart, c->pid, 0, inject_signal);
 264         uloop_process_add(c);
 265 }
 266
 267 static void sigterm_handler(int signum)
 268 {
 269         /* When we receive SIGTERM, we forward it to the tracee. After
 270          * the tracee exits, trace_cb() will be called and make us
 271          * exit too. */
 272         kill(tracer.proc.pid, SIGTERM);
 273 }
 274
 275
 276 int main(int argc, char **argv, char **envp)
 277 {
 278         int status, ch, policy = EPERM;
 279         pid_t child;
 280
 281         /* When invoked via seccomp-trace symlink, work as seccomp
 282          * violation logger rather than as syscall tracer */
 283         if (strstr(argv[0], "seccomp-trace"))
 284                 mode = SECCOMP_TRACE;
 285
 286         while ((ch = getopt(argc, argv, "f:p:")) != -1) {
 287                 switch (ch) {
 288                 case 'f':
 289                         json = optarg;
 290                         break;
 291                 case 'p':
 292                         policy = atoi(optarg);
 293                         break;
 294                 }
 295         }
 296
 297         if (!json)
 298                 json = getenv("SECCOMP_FILE");
 299
 300         argc -= optind;
 301         argv += optind;
 302
 303         if (!argc)
 304                 return -1;
 305
 306         if (getenv("TRACE_DEBUG"))
 307                 debug = 1;
 308         unsetenv("TRACE_DEBUG");
 309
 310         child = fork();
 311
 312         if (child == 0) {
 313                 char **_argv = calloc(argc + 1, sizeof(char *));
 314                 char **_envp;
 315                 char *preload = NULL;
 316                 const char *old_preload = getenv("LD_PRELOAD");
 317                 int newenv = 0;
 318                 int envc = 0;
 319                 int ret;
 320
 321                 memcpy(_argv, argv, argc * sizeof(char *));
 322
 323                 while (envp[envc++])
 324                         ;
 325
 326                 _envp = calloc(envc + 2, sizeof(char *));
 327                 switch (mode) {
 328                 case UTRACE:
 329                         preload = "/lib/libpreload-trace.so";
 330                         newenv = 1;
 331                         break;
 332                 case SECCOMP_TRACE:
 333                         preload = "/lib/libpreload-seccomp.so";
 334                         newenv = 2;
 335                         if (asprintf(&_envp[1], "SECCOMP_FILE=%s", json ? json : "") < 0)
 336                                 ULOG_ERR("failed to allocate SECCOMP_FILE env: %m\n");
 337
 338                         kill(getpid(), SIGSTOP);
 339                         break;
 340                 }
 341                 if (asprintf(&_envp[0], "LD_PRELOAD=%s%s%s", preload,
 342                              old_preload ? ":" : "",
 343                               old_preload ? old_preload : "") < 0)
 344                         ULOG_ERR("failed to allocate LD_PRELOAD env: %m\n");
 345
 346                 memcpy(&_envp[newenv], envp, envc * sizeof(char *));
 347
 348                 ret = execve(_argv[0], _argv, _envp);
 349                 ULOG_ERR("failed to exec %s: %m\n", _argv[0]);
 350
 351                 free(_argv);
 352                 free(_envp);
 353                 return ret;
 354         }
 355
 356         if (child < 0)
 357                 return -1;
 358
 359         waitpid(child, &status, WUNTRACED);
 360         if (!WIFSTOPPED(status)) {
 361                 ULOG_ERR("failed to start %s\n", *argv);
 362                 return -1;
 363         }
 364
 365         /* Initialize uloop to catch all ptrace stops from now on. */
 366         uloop_init();
 367
 368         int ptrace_options = PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK | PTRACE_O_TRACECLONE;
 369         switch (mode) {
 370         case UTRACE:
 371                 ptrace_options |= PTRACE_O_TRACESYSGOOD;
 372                 ptrace_restart = PTRACE_SYSCALL;
 373                 break;
 374         case SECCOMP_TRACE:
 375                 ptrace_options |= PTRACE_O_TRACESECCOMP;
 376                 ptrace_restart = PTRACE_CONT;
 377                 break;
 378         }
 379         if (ptrace(PTRACE_SEIZE, child, 0, ptrace_options) == -1) {
 380                 ULOG_ERR("PTRACE_SEIZE: %m\n");
 381                 return -1;
 382         }
 383         if (ptrace(ptrace_restart, child, 0, SIGCONT) == -1) {
 384                 ULOG_ERR("ptrace_restart: %m\n");
 385                 return -1;
 386         }
 387
 388         tracer.proc.pid = child;
 389         tracer.proc.cb = tracer_cb;
 390         uloop_process_add(&tracer.proc);
 391         signal(SIGTERM, sigterm_handler); /* Override uloop's SIGTERM handler */
 392         uloop_run();
 393         uloop_done();
 394
 395
 396         switch (mode) {
 397         case UTRACE:
 398                 if (!json)
 399                         if (asprintf(&json, "/tmp/%s.%u.json", basename(*argv), child) < 0)
 400                                 ULOG_ERR("failed to allocate output path: %m\n");
 401                 break;
 402         case SECCOMP_TRACE:
 403                 if (!violation_count)
 404                         return 0;
 405                 if (asprintf(&json, "/tmp/%s.%u.violations.json", basename(*argv), child) < 0)
 406                         ULOG_ERR("failed to allocate violations output path: %m\n");
 407                 break;
 408         }
 409         print_syscalls(policy, json);
 410         return 0;
 411 }