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
  29 #ifndef PTRACE_EVENT_STOP
  30 /* PTRACE_EVENT_STOP is defined in linux/ptrace.h, but this header
  31  * collides with musl's sys/ptrace.h */
  32 #define PTRACE_EVENT_STOP 128
  33 #endif
  34
  35 #include <libubox/uloop.h>
  36 #include <libubox/blobmsg.h>
  37 #include <libubox/blobmsg_json.h>
  38
  39 #include "../syscall-names.h"
  40
  41 #define _offsetof(a, b) __builtin_offsetof(a,b)
  42 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
  43
  44 #ifdef __amd64__
  45 #define reg_syscall_nr  _offsetof(struct user, regs.orig_rax)
  46 #elif defined(__i386__)
  47 #define reg_syscall_nr  _offsetof(struct user, regs.orig_eax)
  48 #elif defined(__mips)
  49 # ifndef EF_REG2
  50 # define EF_REG2        8
  51 # endif
  52 #define reg_syscall_nr  (EF_REG2 / 4)
  53 #elif defined(__arm__)
  54 #include <asm/ptrace.h>         /* for PTRACE_SET_SYSCALL */
  55 #define reg_syscall_nr  _offsetof(struct user, regs.uregs[7])
  56 # if defined(__ARM_EABI__)
  57 # define reg_retval_nr  _offsetof(struct user, regs.uregs[0])
  58 # endif
  59 #else
  60 #error tracing is not supported on this architecture
  61 #endif
  62
  63 enum mode {
  64         UTRACE,
  65         SECCOMP_TRACE,
  66 } mode = UTRACE;
  67
  68 #define PROC_NAME(mode) (mode == UTRACE ? "utrace" : "seccomp-trace")
  69
  70 #define INFO(fmt, ...) do { \
  71         fprintf(stderr, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
  72 } while (0)
  73
  74 #define ERROR(fmt, ...) do { \
  75         syslog(LOG_ERR, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
  76         fprintf(stderr, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__);  \
  77 } while (0)
  78
  79 #define LOGERR(fmt, ...) do { \
  80         syslog(LOG_ERR, "%s: "fmt, PROC_NAME(mode), ## __VA_ARGS__); \
  81 } while (0)
  82
  83 struct tracee {
  84         struct uloop_process proc;
  85         int in_syscall;
  86 };
  87
  88 static struct tracee tracer;
  89 static int *syscall_count;
  90 static int violation_count;
  91 static struct blob_buf b;
  92 static int syscall_max;
  93 static int debug;
  94 char *json = NULL;
  95 int ptrace_restart;
  96
  97 static int max_syscall = ARRAY_SIZE(syscall_names);
  98
  99 static void set_syscall(const char *name, int val)
 100 {
 101         int i;
 102
 103         for (i = 0; i < max_syscall; i++)
 104                 if (syscall_names[i] && !strcmp(syscall_names[i], name)) {
 105                         syscall_count[i] = val;
 106                         return;
 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[ARRAY_SIZE(syscall_names)];
 134
 135         for (i = 0; i < ARRAY_SIZE(syscall_names); i++) {
 136                 sorted[i].syscall = i;
 137                 sorted[i].count = syscall_count[i];
 138         }
 139
 140         qsort(sorted, ARRAY_SIZE(syscall_names), 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 < ARRAY_SIZE(syscall_names); i++) {
 146                 int sc = sorted[i].syscall;
 147                 if (!sorted[i].count)
 148                         break;
 149                 if (syscall_names[sc]) {
 150                         if (debug)
 151                                 printf("syscall %d (%s) was called %d times\n",
 152                                         sc, syscall_names[sc], sorted[i].count);
 153                         blobmsg_add_string(&b, NULL, syscall_names[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         if (syscall < ARRAY_SIZE(syscall_names)) {
 191                 syscall_count[syscall]++;
 192                 LOGERR("%s[%u] tried to call non-whitelisted syscall: %s (see %s)\n",
 193                        buf, pid,  syscall_names[syscall], json);
 194         } else {
 195                 LOGERR("%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
 213                                 if (syscall < syscall_max) {
 214                                         syscall_count[syscall]++;
 215                                         if (debug)
 216                                                 fprintf(stderr, "%s()\n", syscall_names[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                         ptrace(PTRACE_GETEVENTMSG, c->pid, 0, &child->proc.pid);
 228                         child->proc.cb = tracer_cb;
 229                         ptrace(ptrace_restart, child->proc.pid, 0, 0);
 230                         uloop_process_add(&child->proc);
 231                         if (debug)
 232                                 fprintf(stderr, "Tracing new child %d\n", child->proc.pid);
 233                 } else if ((ret >> 16) == PTRACE_EVENT_STOP) {
 234                         /* Nothing special to do here */
 235                 } else if ((ret >> 8) == (SIGTRAP | (PTRACE_EVENT_SECCOMP << 8))) {
 236                         int syscall = ptrace(PTRACE_PEEKUSER, c->pid, reg_syscall_nr);
 237 #if defined(__arm__)
 238                         ptrace(PTRACE_SET_SYSCALL, c->pid, 0, -1);
 239                         ptrace(PTRACE_POKEUSER, c->pid, reg_retval_nr, -ENOSYS);
 240 #else
 241                         ptrace(PTRACE_POKEUSER, c->pid, reg_syscall_nr, -1);
 242 #endif
 243                         report_seccomp_vialation(c->pid, syscall);
 244                 } else {
 245                         inject_signal = WSTOPSIG(ret);
 246                         if (debug)
 247                                 fprintf(stderr, "Injecting signal %d into pid %d\n",
 248                                         inject_signal, tracee->proc.pid);
 249                 }
 250         } else if (WIFEXITED(ret) || (WIFSIGNALED(ret) && WTERMSIG(ret))) {
 251                 if (tracee == &tracer) {
 252                         uloop_end(); /* Main process exit */
 253                 } else {
 254                         if (debug)
 255                                 fprintf(stderr, "Child %d exited\n", tracee->proc.pid);
 256                         free(tracee);
 257                 }
 258                 return;
 259         }
 260
 261         ptrace(ptrace_restart, c->pid, 0, inject_signal);
 262         uloop_process_add(c);
 263 }
 264
 265 int main(int argc, char **argv, char **envp)
 266 {
 267         int status, ch, policy = EPERM;
 268         pid_t child;
 269
 270         /* When invoked via seccomp-trace symlink, work as seccomp
 271          * violation logger rather than as syscall tracer */
 272         if (strstr(argv[0], "seccomp-trace"))
 273                 mode = SECCOMP_TRACE;
 274
 275         while ((ch = getopt(argc, argv, "f:p:")) != -1) {
 276                 switch (ch) {
 277                 case 'f':
 278                         json = optarg;
 279                         break;
 280                 case 'p':
 281                         policy = atoi(optarg);
 282                         break;
 283                 }
 284         }
 285
 286         if (!json)
 287                 json = getenv("SECCOMP_FILE");
 288
 289         argc -= optind;
 290         argv += optind;
 291
 292         if (!argc)
 293                 return -1;
 294
 295         if (getenv("TRACE_DEBUG"))
 296                 debug = 1;
 297         unsetenv("TRACE_DEBUG");
 298
 299         child = fork();
 300
 301         if (child == 0) {
 302                 char **_argv = calloc(argc + 1, sizeof(char *));
 303                 char **_envp;
 304                 char *preload = NULL;
 305                 const char *old_preload = getenv("LD_PRELOAD");
 306                 int newenv = 0;
 307                 int envc = 0;
 308                 int ret;
 309
 310                 memcpy(_argv, argv, argc * sizeof(char *));
 311
 312                 while (envp[envc++])
 313                         ;
 314
 315                 _envp = calloc(envc + 2, sizeof(char *));
 316                 switch (mode) {
 317                 case UTRACE:
 318                         preload = "/lib/libpreload-trace.so";
 319                         newenv = 1;
 320                         break;
 321                 case SECCOMP_TRACE:
 322                         preload = "/lib/libpreload-seccomp.so";
 323                         newenv = 2;
 324                         asprintf(&_envp[1], "SECCOMP_FILE=%s", json ? json : "");
 325                         kill(getpid(), SIGSTOP);
 326                         break;
 327                 }
 328                 asprintf(&_envp[0], "LD_PRELOAD=%s%s%s", preload,
 329                          old_preload ? ":" : "",
 330                          old_preload ? old_preload : "");
 331                 memcpy(&_envp[newenv], envp, envc * sizeof(char *));
 332
 333                 ret = execve(_argv[0], _argv, _envp);
 334                 ERROR("failed to exec %s: %s\n", _argv[0], strerror(errno));
 335
 336                 free(_argv);
 337                 free(_envp);
 338                 return ret;
 339         }
 340
 341         if (child < 0)
 342                 return -1;
 343
 344         syscall_max = ARRAY_SIZE(syscall_names);
 345         syscall_count = calloc(syscall_max, sizeof(int));
 346         waitpid(child, &status, WUNTRACED);
 347         if (!WIFSTOPPED(status)) {
 348                 ERROR("failed to start %s\n", *argv);
 349                 return -1;
 350         }
 351
 352         int ptrace_options = PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK | PTRACE_O_TRACECLONE;
 353         switch (mode) {
 354         case UTRACE:
 355                 ptrace_options |= PTRACE_O_TRACESYSGOOD;
 356                 ptrace_restart = PTRACE_SYSCALL;
 357                 break;
 358         case SECCOMP_TRACE:
 359                 ptrace_options |= PTRACE_O_TRACESECCOMP;
 360                 ptrace_restart = PTRACE_CONT;
 361                 break;
 362         }
 363         ptrace(PTRACE_SEIZE, child, 0, ptrace_options);
 364         ptrace(ptrace_restart, child, 0, SIGCONT);
 365
 366         uloop_init();
 367         tracer.proc.pid = child;
 368         tracer.proc.cb = tracer_cb;
 369         uloop_process_add(&tracer.proc);
 370         uloop_run();
 371         uloop_done();
 372
 373
 374         switch (mode) {
 375         case UTRACE:
 376                 if (!json)
 377                         if (asprintf(&json, "/tmp/%s.%u.json", basename(*argv), child) < 0)
 378                                 ERROR("failed to allocate output path: %s\n", strerror(errno));
 379                 break;
 380         case SECCOMP_TRACE:
 381                 if (!violation_count)
 382                         return 0;
 383                 asprintf(&json, "/tmp/%s.%u.violations.json", basename(*argv), child);
 384                 break;
 385         }
 386         print_syscalls(policy, json);
 387         return 0;
 388 }