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;
 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         c = blobmsg_open_array(&b, "whitelist");
 138
 139         for (i = 0; i < SYSCALL_COUNT; i++) {
 140                 int sc = sorted[i].syscall;
 141                 if (!sorted[i].count)
 142                         break;
 143                 if (syscall_name(sc)) {
 144                         if (debug)
 145                                 printf("syscall %d (%s) was called %d times\n",
 146                                        sc, syscall_name(sc), sorted[i].count);
 147                         blobmsg_add_string(&b, NULL, syscall_name(sc));
 148                 } else {
 149                         ULOG_ERR("no name found for syscall(%d)\n", sc);
 150                 }
 151         }
 152         blobmsg_close_array(&b, c);
 153         blobmsg_add_u32(&b, "policy", policy);
 154         if (json) {
 155                 FILE *fp = fopen(json, "w");
 156                 if (fp) {
 157                         fprintf(fp, "%s", blobmsg_format_json_indent(b.head, true, 0));
 158                         fclose(fp);
 159                         ULOG_INFO("saving syscall trace to %s\n", json);
 160                 } else {
 161                         ULOG_ERR("failed to open %s\n", json);
 162                 }
 163         } else {
 164                 printf("%s\n",
 165                         blobmsg_format_json_indent(b.head, true, 0));
 166         }
 167
 168 }
 169
 170 static void report_seccomp_vialation(pid_t pid, unsigned syscall)
 171 {
 172         char buf[200];
 173         snprintf(buf, sizeof(buf), "/proc/%d/cmdline", pid);
 174         int f = open(buf, O_RDONLY);
 175         int r = read(f, buf, sizeof(buf) - 1);
 176         if (r >= 0)
 177                 buf[r] = 0;
 178         else
 179                 strcpy(buf, "unknown?");
 180         close(f);
 181
 182         if (violation_count < INT_MAX)
 183                 violation_count++;
 184         int i = syscall_index(syscall);
 185         if (i >= 0) {
 186                 syscall_count[i]++;
 187                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %s (see %s)\n",
 188                          buf, pid,  syscall_name(syscall), json);
 189         } else {
 190                 ULOG_ERR("%s[%u] tried to call non-whitelisted syscall: %d (see %s)\n",
 191                          buf, pid,  syscall, json);
 192         }
 193 }
 194
 195 static void tracer_cb(struct uloop_process *c, int ret)
 196 {
 197         struct tracee *tracee = container_of(c, struct tracee, proc);
 198         int inject_signal = 0;
 199
 200         /* We explicitely check for events in upper 16 bits, because
 201          * musl (as opposed to glibc) does not report
 202          * PTRACE_EVENT_STOP as WIFSTOPPED */
 203         if (WIFSTOPPED(ret) || (ret >> 16)) {
 204                 if (WSTOPSIG(ret) & 0x80) {
 205                         if (!tracee->in_syscall) {
 206                                 int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 207                                 int i = syscall_index(syscall);
 208                                 if (i >= 0) {
 209                                         syscall_count[i]++;
 210                                         if (debug)
 211                                                 fprintf(stderr, "%s()\n", syscall_name(syscall));
 212                                 } else if (debug) {
 213                                         fprintf(stderr, "syscal(%d)\n", syscall);
 214                                 }
 215                         }
 216                         tracee->in_syscall = !tracee->in_syscall;
 217                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_FORK << 8)) ||
 218                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_VFORK << 8)) ||
 219                            (ret >> 8) == (SIGTRAP | (PTRACE_EVENT_CLONE << 8))) {
 220                         struct tracee *child = calloc(1, sizeof(struct tracee));
 221
 222                         unsigned long msg;
 223                         ptrace(PTRACE_GETEVENTMSG, c->pid, 0, &msg);
 224                         child->proc.pid = msg;
 225                         child->proc.cb = tracer_cb;
 226                         ptrace(ptrace_restart, child->proc.pid, 0, 0);
 227                         uloop_process_add(&child->proc);
 228                         if (debug)
 229                                 fprintf(stderr, "Tracing new child %d\n", child->proc.pid);
 230                 } else if ((ret >> 16) == PTRACE_EVENT_STOP) {
 231                         /* Nothing special to do here */
 232                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_SECCOMP << 8))) {
 233                         int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 234 #if defined(__arm__)
 235                         ptrace(PTRACE_SET_SYSCALL, c->pid, 0, -1);
 236                         ptrace(PTRACE_POKEUSER, c->pid, reg_retval_nr, -ENOSYS);
 237 #else
 238                         ptrace(PTRACE_POKEUSER, c->pid, reg_syscall_nr, -1);
 239 #endif
 240                         report_seccomp_vialation(c->pid, syscall);
 241                 } else {
 242                         inject_signal = WSTOPSIG(ret);
 243                         if (debug)
 244                                 fprintf(stderr, "Injecting signal %d into pid %d\n",
 245                                         inject_signal, tracee->proc.pid);
 246                 }
 247         } else if (WIFEXITED(ret) || (WIFSIGNALED(ret) && WTERMSIG(ret))) {
 248                 if (tracee == &tracer) {
 249                         uloop_end(); /* Main process exit */
 250                 } else {
 251                         if (debug)
 252                                 fprintf(stderr, "Child %d exited\n", tracee->proc.pid);
 253                         free(tracee);
 254                 }
 255                 return;
 256         }
 257
 258         ptrace(ptrace_restart, c->pid, 0, inject_signal);
 259         uloop_process_add(c);
 260 }
 261
 262 static void sigterm_handler(int signum)
 263 {
 264         /* When we receive SIGTERM, we forward it to the tracee. After
 265          * the tracee exits, trace_cb() will be called and make us
 266          * exit too. */
 267         kill(tracer.proc.pid, SIGTERM);
 268 }
 269
 270
 271 int main(int argc, char **argv, char **envp)
 272 {
 273         int status, ch, policy = EPERM;
 274         pid_t child;
 275
 276         /* When invoked via seccomp-trace symlink, work as seccomp
 277          * violation logger rather than as syscall tracer */
 278         if (strstr(argv[0], "seccomp-trace"))
 279                 mode = SECCOMP_TRACE;
 280
 281         while ((ch = getopt(argc, argv, "f:p:")) != -1) {
 282                 switch (ch) {
 283                 case 'f':
 284                         json = optarg;
 285                         break;
 286                 case 'p':
 287                         policy = atoi(optarg);
 288                         break;
 289                 }
 290         }
 291
 292         if (!json)
 293                 json = getenv("SECCOMP_FILE");
 294
 295         argc -= optind;
 296         argv += optind;
 297
 298         if (!argc)
 299                 return -1;
 300
 301         if (getenv("TRACE_DEBUG"))
 302                 debug = 1;
 303         unsetenv("TRACE_DEBUG");
 304
 305         child = fork();
 306
 307         if (child == 0) {
 308                 char **_argv = calloc(argc + 1, sizeof(char *));
 309                 char **_envp;
 310                 char *preload = NULL;
 311                 const char *old_preload = getenv("LD_PRELOAD");
 312                 int newenv = 0;
 313                 int envc = 0;
 314                 int ret;
 315
 316                 memcpy(_argv, argv, argc * sizeof(char *));
 317
 318                 while (envp[envc++])
 319                         ;
 320
 321                 _envp = calloc(envc + 2, sizeof(char *));
 322                 switch (mode) {
 323                 case UTRACE:
 324                         preload = "/lib/libpreload-trace.so";
 325                         newenv = 1;
 326                         break;
 327                 case SECCOMP_TRACE:
 328                         preload = "/lib/libpreload-seccomp.so";
 329                         newenv = 2;
 330                         if (asprintf(&_envp[1], "SECCOMP_FILE=%s", json ? json : "") < 0)
 331                                 ULOG_ERR("failed to allocate SECCOMP_FILE env: %m\n");
 332
 333                         kill(getpid(), SIGSTOP);
 334                         break;
 335                 }
 336                 if (asprintf(&_envp[0], "LD_PRELOAD=%s%s%s", preload,
 337                              old_preload ? ":" : "",
 338                               old_preload ? old_preload : "") < 0)
 339                         ULOG_ERR("failed to allocate LD_PRELOAD env: %m\n");
 340
 341                 memcpy(&_envp[newenv], envp, envc * sizeof(char *));
 342
 343                 ret = execve(_argv[0], _argv, _envp);
 344                 ULOG_ERR("failed to exec %s: %m\n", _argv[0]);
 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                 ULOG_ERR("failed to start %s\n", *argv);
 357                 return -1;
 358         }
 359
 360         /* Initialize uloop to catch all ptrace stops from now on. */
 361         uloop_init();
 362
 363         int ptrace_options = PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK | PTRACE_O_TRACECLONE;
 364         switch (mode) {
 365         case UTRACE:
 366                 ptrace_options |= PTRACE_O_TRACESYSGOOD;
 367                 ptrace_restart = PTRACE_SYSCALL;
 368                 break;
 369         case SECCOMP_TRACE:
 370                 ptrace_options |= PTRACE_O_TRACESECCOMP;
 371                 ptrace_restart = PTRACE_CONT;
 372                 break;
 373         }
 374         if (ptrace(PTRACE_SEIZE, child, 0, ptrace_options) == -1) {
 375                 ULOG_ERR("PTRACE_SEIZE: %m\n");
 376                 return -1;
 377         }
 378         if (ptrace(ptrace_restart, child, 0, SIGCONT) == -1) {
 379                 ULOG_ERR("ptrace_restart: %m\n");
 380                 return -1;
 381         }
 382
 383         tracer.proc.pid = child;
 384         tracer.proc.cb = tracer_cb;
 385         uloop_process_add(&tracer.proc);
 386         signal(SIGTERM, sigterm_handler); /* Override uloop's SIGTERM handler */
 387         uloop_run();
 388         uloop_done();
 389
 390
 391         switch (mode) {
 392         case UTRACE:
 393                 if (!json)
 394                         if (asprintf(&json, "/tmp/%s.%u.json", basename(*argv), child) < 0)
 395                                 ULOG_ERR("failed to allocate output path: %m\n");
 396                 break;
 397         case SECCOMP_TRACE:
 398                 if (!violation_count)
 399                         return 0;
 400                 if (asprintf(&json, "/tmp/%s.%u.violations.json", basename(*argv), child) < 0)
 401                         ULOG_ERR("failed to allocate violations output path: %m\n");
 402                 break;
 403         }
 404         print_syscalls(policy, json);
 405         return 0;
 406 }