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 #if defined (__aarch64__)
  52 #include <linux/ptrace.h>
  53 #elif defined(__amd64__)
  54 #define reg_syscall_nr  _offsetof(struct user, regs.orig_rax)
  55 #elif defined(__arm__)
  56 #include <asm/ptrace.h>         /* for PTRACE_SET_SYSCALL */
  57 #define reg_syscall_nr  _offsetof(struct user, regs.uregs[7])
  58 # if defined(__ARM_EABI__)
  59 # define reg_retval_nr  _offsetof(struct user, regs.uregs[0])
  60 # endif
  61 #elif defined(__i386__)
  62 #define reg_syscall_nr  _offsetof(struct user, regs.orig_eax)
  63 #elif defined(__mips)
  64 # ifndef EF_REG2
  65 # define EF_REG2        8
  66 # endif
  67 #define reg_syscall_nr  (EF_REG2 / 4)
  68 #elif defined(__PPC__)
  69 #define reg_syscall_nr  _offsetof(struct user, regs.gpr[0])
  70 #define reg_retval_nr   _offsetof(struct user, regs.gpr[3])
  71 #else
  72 #error tracing is not supported on this architecture
  73 #endif
  74
  75 enum mode {
  76         UTRACE,
  77         SECCOMP_TRACE,
  78 } mode = UTRACE;
  79
  80 struct tracee {
  81         struct uloop_process proc;
  82         int in_syscall;
  83 };
  84
  85 static struct tracee tracer;
  86 static int syscall_count[SYSCALL_COUNT];
  87 static int violation_count;
  88 static struct blob_buf b;
  89 static int debug;
  90 char *json = NULL;
  91 int ptrace_restart;
  92
  93 static void set_syscall(const char *name, int val)
  94 {
  95         int i;
  96
  97         for (i = 0; i < SYSCALL_COUNT; i++) {
  98                 int sc = syscall_index_to_number(i);
  99                 if (syscall_name(sc) && !strcmp(syscall_name(sc), name)) {
 100                         syscall_count[i] = val;
 101                         return;
 102                 }
 103         }
 104 }
 105
 106 struct syscall {
 107         int syscall;
 108         int count;
 109 };
 110
 111 static int cmp_count(const void *a, const void *b)
 112 {
 113         return ((struct syscall*)b)->count - ((struct syscall*)a)->count;
 114 }
 115
 116 static void print_syscalls(int policy, const char *json)
 117 {
 118         void *c, *d, *e;
 119         int i;
 120         char *tmp;
 121
 122         if (mode == UTRACE) {
 123                 set_syscall("rt_sigaction", 1);
 124                 set_syscall("sigreturn", 1);
 125                 set_syscall("rt_sigreturn", 1);
 126                 set_syscall("exit_group", 1);
 127                 set_syscall("exit", 1);
 128         }
 129
 130         struct syscall sorted[SYSCALL_COUNT];
 131
 132         for (i = 0; i < SYSCALL_COUNT; i++) {
 133                 sorted[i].syscall = syscall_index_to_number(i);
 134                 sorted[i].count = syscall_count[i];
 135         }
 136
 137         qsort(sorted, SYSCALL_COUNT, sizeof(sorted[0]), cmp_count);
 138
 139         blob_buf_init(&b, 0);
 140         blobmsg_add_string(&b, "defaultAction", "SCMP_ACT_KILL_PROCESS");
 141         c = blobmsg_open_array(&b, "syscalls");
 142         d = blobmsg_open_table(&b, "");
 143         e = blobmsg_open_array(&b, "names");
 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                         ULOG_ERR("no name found for syscall(%d)\n", sc);
 156                 }
 157         }
 158         blobmsg_close_array(&b, e);
 159         blobmsg_add_string(&b, "action", "SCMP_ACT_ALLOW");
 160         blobmsg_close_table(&b, d);
 161         blobmsg_close_array(&b, c);
 162         if (json) {
 163                 FILE *fp = fopen(json, "w");
 164                 if (fp) {
 165                         tmp = blobmsg_format_json_indent(b.head, true, 0);
 166                         if (!tmp) {
 167                                 fclose(fp);
 168                                 return;
 169                         }
 170
 171                         fprintf(fp, "%s\n", tmp);
 172                         free(tmp);
 173                         fclose(fp);
 174                         ULOG_INFO("saving syscall trace to %s\n", json);
 175                 } else {
 176                         ULOG_ERR("failed to open %s\n", json);
 177                 }
 178         } else {
 179                 tmp = blobmsg_format_json_indent(b.head, true, 0);
 180                 if (!tmp)
 181                         return;
 182
 183                 printf("%s\n", tmp);
 184                 free(tmp);
 185         }
 186 }
 187
 188 static void report_seccomp_vialation(pid_t pid, unsigned syscall)
 189 {
 190         char buf[200];
 191         snprintf(buf, sizeof(buf), "/proc/%d/cmdline", pid);
 192         int f = open(buf, O_RDONLY);
 193         if (f < 0)
 194                 return;
 195
 196         int r = read(f, buf, sizeof(buf) - 1);
 197         buf[sizeof(buf) - 1] = '\0';
 198
 199         if (r >= 0)
 200                 buf[r] = 0;
 201         else
 202                 strcpy(buf, "unknown?");
 203         close(f);
 204
 205         if (violation_count < INT_MAX)
 206                 violation_count++;
 207         int i = syscall_index(syscall);
 208         if (i >= 0) {
 209                 syscall_count[i]++;
 210                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %s (see %s)\n",
 211                          buf, pid,  syscall_name(syscall), json);
 212         } else {
 213                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %d (see %s)\n",
 214                          buf, pid,  syscall, json);
 215         }
 216 }
 217
 218 static void tracer_cb(struct uloop_process *c, int ret)
 219 {
 220         struct tracee *tracee = container_of(c, struct tracee, proc);
 221         int inject_signal = 0;
 222
 223         /* We explicitely check for events in upper 16 bits, because
 224          * musl (as opposed to glibc) does not report
 225          * PTRACE_EVENT_STOP as WIFSTOPPED */
 226         if (WIFSTOPPED(ret) || (ret >> 16)) {
 227                 if (WSTOPSIG(ret) & 0x80) {
 228                         if (!tracee->in_syscall) {
 229 #ifdef __aarch64__
 230                                 int syscall = -1;
 231                                 struct ptrace_syscall_info ptsi = {.op=PTRACE_SYSCALL_INFO_ENTRY};
 232                                 if (ptrace(PTRACE_GET_SYSCALL_INFO, c->pid, sizeof(ptsi), &ptsi) != -1)
 233                                         syscall = ptsi.entry.nr;
 234 #else
 235                                 int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 236 #endif
 237                                 int i = syscall_index(syscall);
 238                                 if (i >= 0) {
 239                                         syscall_count[i]++;
 240                                         if (debug)
 241                                                 fprintf(stderr, "%s()\n", syscall_name(syscall));
 242                                 } else if (debug) {
 243                                         fprintf(stderr, "syscal(%d)\n", syscall);
 244                                 }
 245                         }
 246                         tracee->in_syscall = !tracee->in_syscall;
 247                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_FORK << 8)) ||
 248                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_VFORK << 8)) ||
 249                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_CLONE << 8))) {
 250                         struct tracee *child = calloc(1, sizeof(struct tracee));
 251
 252                         unsigned long msg;
 253                         ptrace(PTRACE_GETEVENTMSG, c->pid, 0, &msg);
 254                         child->proc.pid = msg;
 255                         child->proc.cb = tracer_cb;
 256                         ptrace(ptrace_restart, child->proc.pid, 0, 0);
 257                         uloop_process_add(&child->proc);
 258                         if (debug)
 259                                 fprintf(stderr, "Tracing new child %d\n", child->proc.pid);
 260                 } else if ((ret >> 16) == PTRACE_EVENT_STOP) {
 261                         /* Nothing special to do here */
 262                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_SECCOMP << 8))) {
 263 #ifdef __aarch64__
 264                         int syscall = -1;
 265                         struct ptrace_syscall_info ptsi = {.op=PTRACE_SYSCALL_INFO_SECCOMP};
 266                         if (ptrace(PTRACE_GET_SYSCALL_INFO, c->pid, sizeof(ptsi), &ptsi) != -1)
 267                                 syscall = ptsi.entry.nr;
 268 #else
 269                         int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 270 #if defined(__arm__)
 271                         ptrace(PTRACE_SET_SYSCALL, c->pid, 0, -1);
 272                         ptrace(PTRACE_POKEUSER, c->pid, reg_retval_nr, -ENOSYS);
 273 #else
 274                         ptrace(PTRACE_POKEUSER, c->pid, reg_syscall_nr, -1);
 275 #endif
 276 #endif
 277                         report_seccomp_vialation(c->pid, syscall);
 278                 } else {
 279                         inject_signal = WSTOPSIG(ret);
 280                         if (debug)
 281                                 fprintf(stderr, "Injecting signal %d into pid %d\n",
 282                                         inject_signal, tracee->proc.pid);
 283                 }
 284         } else if (WIFEXITED(ret) || (WIFSIGNALED(ret) && WTERMSIG(ret))) {
 285                 if (tracee == &tracer) {
 286                         uloop_end(); /* Main process exit */
 287                 } else {
 288                         if (debug)
 289                                 fprintf(stderr, "Child %d exited\n", tracee->proc.pid);
 290                         free(tracee);
 291                 }
 292                 return;
 293         }
 294
 295         ptrace(ptrace_restart, c->pid, 0, inject_signal);
 296         uloop_process_add(c);
 297 }
 298
 299 static void sigterm_handler(int signum)
 300 {
 301         /* When we receive SIGTERM, we forward it to the tracee. After
 302          * the tracee exits, trace_cb() will be called and make us
 303          * exit too. */
 304         kill(tracer.proc.pid, SIGTERM);
 305 }
 306
 307
 308 int main(int argc, char **argv, char **envp)
 309 {
 310         int status, ch, policy = EPERM;
 311         pid_t child;
 312
 313         /* When invoked via seccomp-trace symlink, work as seccomp
 314          * violation logger rather than as syscall tracer */
 315         if (strstr(argv[0], "seccomp-trace"))
 316                 mode = SECCOMP_TRACE;
 317
 318         while ((ch = getopt(argc, argv, "f:p:")) != -1) {
 319                 switch (ch) {
 320                 case 'f':
 321                         json = optarg;
 322                         break;
 323                 case 'p':
 324                         policy = atoi(optarg);
 325                         break;
 326                 }
 327         }
 328
 329         if (!json)
 330                 json = getenv("SECCOMP_FILE");
 331
 332         argc -= optind;
 333         argv += optind;
 334
 335         if (!argc)
 336                 return -1;
 337
 338         if (getenv("TRACE_DEBUG"))
 339                 debug = 1;
 340         unsetenv("TRACE_DEBUG");
 341
 342         child = fork();
 343
 344         if (child == 0) {
 345                 char **_argv = calloc(argc + 1, sizeof(char *));
 346                 char **_envp;
 347                 char *preload = NULL;
 348                 const char *old_preload = getenv("LD_PRELOAD");
 349                 int newenv = 0;
 350                 int envc = 0;
 351                 int ret;
 352
 353                 memcpy(_argv, argv, argc * sizeof(char *));
 354
 355                 while (envp[envc++])
 356                         ;
 357
 358                 _envp = calloc(envc + 2, sizeof(char *));
 359                 switch (mode) {
 360                 case UTRACE:
 361                         preload = "/lib/libpreload-trace.so";
 362                         newenv = 1;
 363                         break;
 364                 case SECCOMP_TRACE:
 365                         preload = "/lib/libpreload-seccomp.so";
 366                         newenv = 2;
 367                         if (asprintf(&_envp[1], "SECCOMP_FILE=%s", json ? json : "") < 0)
 368                                 ULOG_ERR("failed to allocate SECCOMP_FILE env: %m\n");
 369
 370                         kill(getpid(), SIGSTOP);
 371                         break;
 372                 }
 373                 if (asprintf(&_envp[0], "LD_PRELOAD=%s%s%s", preload,
 374                              old_preload ? ":" : "",
 375                               old_preload ? old_preload : "") < 0)
 376                         ULOG_ERR("failed to allocate LD_PRELOAD env: %m\n");
 377
 378                 memcpy(&_envp[newenv], envp, envc * sizeof(char *));
 379
 380                 ret = execve(_argv[0], _argv, _envp);
 381                 ULOG_ERR("failed to exec %s: %m\n", _argv[0]);
 382
 383                 free(_argv);
 384                 free(_envp);
 385                 return ret;
 386         }
 387
 388         if (child < 0)
 389                 return -1;
 390
 391         waitpid(child, &status, WUNTRACED);
 392         if (!WIFSTOPPED(status)) {
 393                 ULOG_ERR("failed to start %s\n", *argv);
 394                 return -1;
 395         }
 396
 397         /* Initialize uloop to catch all ptrace stops from now on. */
 398         uloop_init();
 399
 400         int ptrace_options = PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK | PTRACE_O_TRACECLONE;
 401         switch (mode) {
 402         case UTRACE:
 403                 ptrace_options |= PTRACE_O_TRACESYSGOOD;
 404                 ptrace_restart = PTRACE_SYSCALL;
 405                 break;
 406         case SECCOMP_TRACE:
 407                 ptrace_options |= PTRACE_O_TRACESECCOMP;
 408                 ptrace_restart = PTRACE_CONT;
 409                 break;
 410         }
 411         if (ptrace(PTRACE_SEIZE, child, 0, ptrace_options) == -1) {
 412                 ULOG_ERR("PTRACE_SEIZE: %m\n");
 413                 return -1;
 414         }
 415         if (ptrace(ptrace_restart, child, 0, SIGCONT) == -1) {
 416                 ULOG_ERR("ptrace_restart: %m\n");
 417                 return -1;
 418         }
 419
 420         tracer.proc.pid = child;
 421         tracer.proc.cb = tracer_cb;
 422         uloop_process_add(&tracer.proc);
 423         signal(SIGTERM, sigterm_handler); /* Override uloop's SIGTERM handler */
 424         uloop_run();
 425         uloop_done();
 426
 427
 428         switch (mode) {
 429         case UTRACE:
 430                 if (!json)
 431                         if (asprintf(&json, "/tmp/%s.%u.json", basename(*argv), child) < 0)
 432                                 ULOG_ERR("failed to allocate output path: %m\n");
 433                 break;
 434         case SECCOMP_TRACE:
 435                 if (!violation_count)
 436                         return 0;
 437                 if (asprintf(&json, "/tmp/%s.%u.violations.json", basename(*argv), child) < 0)
 438                         ULOG_ERR("failed to allocate violations output path: %m\n");
 439                 break;
 440         }
 441         print_syscalls(policy, json);
 442         return 0;
 443 }