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