2 * A Simple Sandbox for MO-Eval
4 * (c) 2001--2008 Martin Mares <mj@ucw.cz>
7 #define _LARGEFILE64_SOURCE
22 #include <sys/ptrace.h>
23 #include <sys/signal.h>
24 #include <sys/sysinfo.h>
25 #include <sys/syscall.h>
26 #include <sys/resource.h>
28 #define NONRET __attribute__((noreturn))
29 #define UNUSED __attribute__((unused))
30 #define ARRAY_SIZE(a) (int)(sizeof(a)/sizeof(a[0]))
32 static int filter_syscalls; /* 0=off, 1=liberal, 2=totalitarian */
33 static int timeout; /* milliseconds */
34 static int wall_timeout;
35 static int pass_environ;
36 static int file_access;
38 static int memory_limit;
39 static char *redir_stdin, *redir_stdout, *redir_stderr;
43 static int is_ptraced;
44 static volatile int timer_tick;
45 static struct timeval start_time;
46 static int ticks_per_sec;
48 static int partial_line;
50 #if defined(__GLIBC__) && __GLIBC__ == 2 && __GLIBC_MINOR__ > 0
51 /* glibc 2.1 or newer -> has lseek64 */
52 #define long_seek(f,o,w) lseek64(f,o,w)
54 /* Touching clandestine places in glibc */
55 extern loff_t llseek(int fd, loff_t pos, int whence);
56 #define long_seek(f,o,w) llseek(f,o,w)
59 static void die(char *msg, ...) NONRET;
63 static FILE *metafile;
66 meta_open(const char *name)
68 if (!strcmp(name, "-"))
73 metafile = fopen(name, "w");
75 die("Failed to open metafile '%s'",name);
81 if (metafile && metafile != stdout)
85 static void __attribute__((format(printf,1,2)))
86 meta_printf(const char *fmt, ...)
93 vfprintf(metafile, fmt, args);
97 static int total_ms, wall_ms;
100 final_stats(struct rusage *rus)
102 struct timeval total, now, wall;
103 timeradd(&rus->ru_utime, &rus->ru_stime, &total);
104 total_ms = total.tv_sec*1000 + total.tv_usec/1000;
105 gettimeofday(&now, NULL);
106 // FIXME: We are not guaranteed to have start_time always initialized
107 timersub(&now, &start_time, &wall);
108 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
110 meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000);
111 meta_printf("time-wall:%d.%03d\n", wall_ms/1000, wall_ms%1000);
114 /*** Messages and exits ***/
122 ptrace(PTRACE_KILL, box_pid);
123 kill(-box_pid, SIGKILL);
124 kill(box_pid, SIGKILL);
128 int p = wait4(box_pid, &stat, 0, &rus);
130 fprintf(stderr, "UGH: Lost track of the process\n");
146 static void NONRET __attribute__((format(printf,1,2)))
152 vfprintf(stderr, msg, args);
155 // FIXME: exit(2) for errors of the box itself
158 static void __attribute__((format(printf,1,2)))
165 int len = strlen(msg);
167 partial_line = (msg[len-1] != '\n');
168 vfprintf(stderr, msg, args);
177 void *p = malloc(size);
179 die("Out of memory");
183 /*** Syscall rules ***/
185 static const char * const syscall_names[] = {
186 #include "box/syscall-table.h"
188 #define NUM_SYSCALLS ARRAY_SIZE(syscall_names)
189 #define NUM_ACTIONS (NUM_SYSCALLS+64)
192 A_DEFAULT, // Use the default action
193 A_NO, // Always forbid
194 A_YES, // Always permit
195 A_FILENAME, // Permit if arg1 is a known filename
196 A_LIBERAL = 128, // Valid only in liberal mode
199 static unsigned char syscall_action[NUM_ACTIONS] = {
200 #define S(x) [__NR_##x]
202 // Syscalls permitted for specific file names
203 S(open) = A_FILENAME,
204 S(creat) = A_FILENAME,
205 S(unlink) = A_FILENAME,
206 S(oldstat) = A_FILENAME,
207 S(access) = A_FILENAME,
208 S(oldlstat) = A_FILENAME,
209 S(truncate) = A_FILENAME,
210 S(stat) = A_FILENAME,
211 S(lstat) = A_FILENAME,
212 S(truncate64) = A_FILENAME,
213 S(stat64) = A_FILENAME,
214 S(lstat64) = A_FILENAME,
215 S(readlink) = A_FILENAME,
217 // Syscalls permitted always
232 S(ftruncate) = A_YES,
234 S(personality) = A_YES,
238 S(getresuid) = A_YES,
246 S(ftruncate64) = A_YES,
255 S(set_thread_area) = A_YES,
256 S(get_thread_area) = A_YES,
257 S(exit_group) = A_YES,
259 // Syscalls permitted only in liberal mode
260 S(time) = A_YES | A_LIBERAL,
261 S(alarm) = A_YES | A_LIBERAL,
262 S(pause) = A_YES | A_LIBERAL,
263 S(signal) = A_YES | A_LIBERAL,
264 S(fchmod) = A_YES | A_LIBERAL,
265 S(sigaction) = A_YES | A_LIBERAL,
266 S(sgetmask) = A_YES | A_LIBERAL,
267 S(ssetmask) = A_YES | A_LIBERAL,
268 S(sigsuspend) = A_YES | A_LIBERAL,
269 S(sigpending) = A_YES | A_LIBERAL,
270 S(getrlimit) = A_YES | A_LIBERAL,
271 S(getrusage) = A_YES | A_LIBERAL,
272 S(ugetrlimit) = A_YES | A_LIBERAL,
273 S(gettimeofday) = A_YES | A_LIBERAL,
274 S(select) = A_YES | A_LIBERAL,
275 S(readdir) = A_YES | A_LIBERAL,
276 S(setitimer) = A_YES | A_LIBERAL,
277 S(getitimer) = A_YES | A_LIBERAL,
278 S(sigreturn) = A_YES | A_LIBERAL,
279 S(mprotect) = A_YES | A_LIBERAL,
280 S(sigprocmask) = A_YES | A_LIBERAL,
281 S(getdents) = A_YES | A_LIBERAL,
282 S(getdents64) = A_YES | A_LIBERAL,
283 S(_newselect) = A_YES | A_LIBERAL,
284 S(fdatasync) = A_YES | A_LIBERAL,
285 S(mremap) = A_YES | A_LIBERAL,
286 S(poll) = A_YES | A_LIBERAL,
287 S(getcwd) = A_YES | A_LIBERAL,
288 S(nanosleep) = A_YES | A_LIBERAL,
289 S(rt_sigreturn) = A_YES | A_LIBERAL,
290 S(rt_sigaction) = A_YES | A_LIBERAL,
291 S(rt_sigprocmask) = A_YES | A_LIBERAL,
292 S(rt_sigpending) = A_YES | A_LIBERAL,
293 S(rt_sigtimedwait) = A_YES | A_LIBERAL,
294 S(rt_sigqueueinfo) = A_YES | A_LIBERAL,
295 S(rt_sigsuspend) = A_YES | A_LIBERAL,
296 S(mmap2) = A_YES | A_LIBERAL,
297 S(_sysctl) = A_YES | A_LIBERAL,
302 syscall_name(unsigned int id, char *buf)
304 if (id < NUM_SYSCALLS && syscall_names[id])
305 return syscall_names[id];
308 sprintf(buf, "#%d", id);
314 syscall_by_name(char *name)
316 for (unsigned int i=0; i<NUM_SYSCALLS; i++)
317 if (syscall_names[i] && !strcmp(syscall_names[i], name))
324 unsigned long l = strtoul(name, &ep, 0);
327 if (l >= NUM_ACTIONS)
333 set_syscall_action(char *a)
335 char *sep = strchr(a, '=');
336 enum action act = A_YES;
340 if (!strcmp(sep, "yes"))
342 else if (!strcmp(sep, "no"))
344 else if (!strcmp(sep, "file"))
350 int sys = syscall_by_name(a);
352 die("Unknown syscall `%s'", a);
353 if (sys >= NUM_ACTIONS)
354 die("Syscall `%s' out of range", a);
355 syscall_action[sys] = act;
364 struct path_rule *next;
367 static struct path_rule default_path_rules[] = {
370 { "/usr/lib/", A_YES },
371 { "/opt/lib/", A_YES },
372 { "/usr/share/zoneinfo/", A_YES },
373 { "/usr/share/locale/", A_YES },
374 { "/dev/null", A_YES },
375 { "/dev/zero", A_YES },
376 { "/proc/meminfo", A_YES },
377 { "/proc/self/stat", A_YES },
378 { "/proc/self/exe", A_YES }, // Needed by FPC 2.0.x runtime
381 static struct path_rule *user_path_rules;
382 static struct path_rule **last_path_rule = &user_path_rules;
385 set_path_action(char *a)
387 char *sep = strchr(a, '=');
388 enum action act = A_YES;
392 if (!strcmp(sep, "yes"))
394 else if (!strcmp(sep, "no"))
400 struct path_rule *r = xmalloc(sizeof(*r) + strlen(a) + 1);
401 r->path = (char *)(r+1);
406 last_path_rule = &r->next;
411 match_path_rule(struct path_rule *r, char *path)
415 if (*rr++ != *path++)
417 if (rr[-1] == '/' && !path[-1])
421 if (rr > r->path && rr[-1] != '/' && *path)
426 /*** Environment rules ***/
429 char *var; // Variable to match
430 char *val; // ""=clear, NULL=inherit
432 struct env_rule *next;
435 static struct env_rule *first_env_rule;
436 static struct env_rule **last_env_rule = &first_env_rule;
438 static struct env_rule default_env_rules[] = {
439 { "LIBC_FATAL_STDERR_", "1" }
443 set_env_action(char *a0)
445 struct env_rule *r = xmalloc(sizeof(*r) + strlen(a0) + 1);
446 char *a = (char *)(r+1);
449 char *sep = strchr(a, '=');
461 last_env_rule = &r->next;
467 match_env_var(char *env_entry, struct env_rule *r)
469 if (strncmp(env_entry, r->var, r->var_len))
471 return (env_entry[r->var_len] == '=');
475 apply_env_rule(char **env, int *env_sizep, struct env_rule *r)
477 // First remove the variable if already set
479 while (pos < *env_sizep && !match_env_var(env[pos], r))
481 if (pos < *env_sizep)
484 env[pos] = env[*env_sizep];
485 env[*env_sizep] = NULL;
488 // What is the new value?
494 new = xmalloc(r->var_len + 1 + strlen(r->val) + 1);
495 sprintf(new, "%s=%s", r->var, r->val);
500 while (environ[pos] && !match_env_var(environ[pos], r))
502 if (!(new = environ[pos]))
506 // Add it at the end of the array
507 env[(*env_sizep)++] = new;
508 env[*env_sizep] = NULL;
512 setup_environment(void)
514 // Link built-in rules with user rules
515 for (int i=ARRAY_SIZE(default_env_rules)-1; i >= 0; i--)
517 default_env_rules[i].next = first_env_rule;
518 first_env_rule = &default_env_rules[i];
521 // Scan the original environment
522 char **orig_env = environ;
524 while (orig_env[orig_size])
527 // For each rule, reserve one more slot and calculate length
529 for (struct env_rule *r = first_env_rule; r; r=r->next)
532 r->var_len = strlen(r->var);
535 // Create a new environment
536 char **env = xmalloc((orig_size + num_rules + 1) * sizeof(char *));
540 memcpy(env, environ, orig_size * sizeof(char *));
547 // Apply the rules one by one
548 for (struct env_rule *r = first_env_rule; r; r=r->next)
549 apply_env_rule(env, &size, r);
551 // Return the new env and pass some gossip
554 fprintf(stderr, "Passing environment:\n");
555 for (int i=0; env[i]; i++)
556 fprintf(stderr, "\t%s\n", env[i]);
561 /*** Syscall checks ***/
564 valid_filename(unsigned long addr)
566 char namebuf[4096], *p, *end;
570 die("File access forbidden");
571 if (file_access >= 9)
576 sprintf(namebuf, "/proc/%d/mem", (int) box_pid);
577 mem_fd = open(namebuf, O_RDONLY);
579 die("open(%s): %m", namebuf);
586 int remains = PAGE_SIZE - (addr & (PAGE_SIZE-1));
587 int l = namebuf + sizeof(namebuf) - end;
591 die("Access to file with name too long");
592 if (long_seek(mem_fd, addr, SEEK_SET) < 0)
593 die("long_seek(mem): %m");
594 remains = read(mem_fd, end, l);
596 die("read(mem): %m");
598 die("Access to file with name out of memory");
605 msg("[%s] ", namebuf);
606 if (file_access >= 3)
609 // Everything in current directory is permitted
610 if (!strchr(namebuf, '/') && strcmp(namebuf, ".."))
613 // ".." anywhere in the path is forbidden
614 enum action act = A_DEFAULT;
615 if (strstr(namebuf, ".."))
619 for (struct path_rule *r = user_path_rules; r && !act; r=r->next)
620 act = match_path_rule(r, namebuf);
622 // Scan built-in rules
623 if (file_access >= 2)
624 for (int i=0; i<ARRAY_SIZE(default_path_rules) && !act; i++)
625 act = match_path_rule(&default_path_rules[i], namebuf);
628 die("Forbidden access to file `%s'", namebuf);
632 valid_syscall(struct user *u)
634 unsigned int sys = u->regs.orig_eax;
635 enum action act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT;
639 if (filter_syscalls == 1)
651 valid_filename(u->regs.ebx);
659 if (u->regs.ebx == box_pid)
660 die("Committed suicide by signal %d", (int)u->regs.ecx);
663 if (u->regs.ebx == box_pid && u->regs.ecx == box_pid)
664 die("Committed suicide by signal %d", (int)u->regs.edx);
672 signal_alarm(int unused UNUSED)
674 /* Time limit checks are synchronous, so we only schedule them there. */
680 signal_int(int unused UNUSED)
682 /* Interrupts are fatal, so no synchronization requirements. */
691 struct timeval now, wall;
693 gettimeofday(&now, NULL);
694 timersub(&now, &start_time, &wall);
695 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
696 if (wall_ms > wall_timeout)
697 die("Time limit exceeded (wall clock)");
699 fprintf(stderr, "[wall time check: %d msec]\n", wall_ms);
704 int c, utime, stime, ms;
705 static int proc_status_fd;
708 sprintf(buf, "/proc/%d/stat", (int) box_pid);
709 proc_status_fd = open(buf, O_RDONLY);
710 if (proc_status_fd < 0)
711 die("open(%s): %m", buf);
713 lseek(proc_status_fd, 0, SEEK_SET);
714 if ((c = read(proc_status_fd, buf, sizeof(buf)-1)) < 0)
715 die("read on /proc/$pid/stat: %m");
716 if (c >= (int) sizeof(buf) - 1)
717 die("/proc/$pid/stat too long");
720 while (*x && *x != ' ')
725 die("proc syntax error 1");
726 while (*x && (*x != ')' || x[1] != ' '))
728 while (*x == ')' || *x == ' ')
730 if (sscanf(x, "%*c %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d %d %d", &utime, &stime) != 2)
731 die("proc syntax error 2");
732 ms = (utime + stime) * 1000 / ticks_per_sec;
734 fprintf(stderr, "[time check: %d msec]\n", ms);
736 die("Time limit exceeded");
743 int syscall_count = 0;
747 bzero(&sa, sizeof(sa));
748 sa.sa_handler = signal_int;
749 sigaction(SIGINT, &sa, NULL);
750 gettimeofday(&start_time, NULL);
751 ticks_per_sec = sysconf(_SC_CLK_TCK);
752 if (ticks_per_sec <= 0)
753 die("Invalid ticks_per_sec!");
754 if (timeout || wall_timeout)
756 sa.sa_handler = signal_alarm;
757 sigaction(SIGALRM, &sa, NULL);
770 p = wait4(box_pid, &stat, WUNTRACED, &rus);
778 die("wait4: unknown pid %d exited!", p);
783 // FIXME: If the process has exited before being ptraced, signal an internal error
784 if (WEXITSTATUS(stat))
785 die("Exited with error status %d", WEXITSTATUS(stat));
786 if (timeout && total_ms > timeout)
787 die("Time limit exceeded");
788 if (wall_timeout && wall_ms > wall_timeout)
789 die("Time limit exceeded (wall clock)");
791 fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n",
792 total_ms/1000, total_ms%1000,
793 wall_ms/1000, wall_ms%1000,
797 if (WIFSIGNALED(stat))
800 die("Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup"));
802 if (WIFSTOPPED(stat))
804 int sig = WSTOPSIG(stat);
808 static int stop_count = -1;
809 if (ptrace(PTRACE_GETREGS, box_pid, NULL, &u) < 0)
810 die("ptrace(PTRACE_GETREGS): %m");
812 if (!stop_count) /* Traceme request */
813 msg(">> Traceme request caught\n");
814 else if (stop_count & 1) /* Syscall entry */
817 msg(">> Syscall %-12s (%08lx,%08lx,%08lx) ", syscall_name(u.regs.orig_eax, namebuf), u.regs.ebx, u.regs.ecx, u.regs.edx);
821 if (u.regs.orig_eax == __NR_execve)
824 else if (valid_syscall(&u))
829 * Unfortunately, PTRACE_KILL kills _after_ the syscall completes,
830 * so we have to change it to something harmless (e.g., an undefined
831 * syscall) and make the program continue.
833 unsigned int sys = u.regs.orig_eax;
834 u.regs.orig_eax = 0xffffffff;
835 if (ptrace(PTRACE_SETREGS, box_pid, NULL, &u) < 0)
836 die("ptrace(PTRACE_SETREGS): %m");
837 die("Forbidden syscall %s", syscall_name(sys, namebuf));
840 else /* Syscall return */
841 msg("= %ld\n", u.regs.eax);
842 ptrace(PTRACE_SYSCALL, box_pid, 0, 0);
844 else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ)
846 msg(">> Signal %d\n", sig);
847 ptrace(PTRACE_SYSCALL, box_pid, 0, sig);
850 die("Received signal %d", sig);
853 die("wait4: unknown status %x, giving up!", stat);
858 box_inside(int argc, char **argv)
863 memcpy(args, argv, argc * sizeof(char *));
865 if (set_cwd && chdir(set_cwd))
870 if (open(redir_stdin, O_RDONLY) != 0)
871 die("open(\"%s\"): %m", redir_stdin);
876 if (open(redir_stdout, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 1)
877 die("open(\"%s\"): %m", redir_stdout);
882 if (open(redir_stderr, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 2)
883 die("open(\"%s\"): %m", redir_stderr);
890 rl.rlim_cur = rl.rlim_max = memory_limit * 1024;
891 if (setrlimit(RLIMIT_AS, &rl) < 0)
892 die("setrlimit: %m");
894 rl.rlim_cur = rl.rlim_max = 64;
895 if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
896 die("setrlimit: %m");
899 if (ptrace(PTRACE_TRACEME) < 0)
900 die("ptrace(PTRACE_TRACEME): %m");
901 /* Trick: Make sure that we are stopped until the boxkeeper wakes up. */
902 signal(SIGCHLD, SIG_IGN);
905 execve(args[0], args, setup_environment());
906 die("execve(\"%s\"): %m", args[0]);
912 fprintf(stderr, "Invalid arguments!\n");
914 Usage: box [<options>] -- <command> <arguments>\n\
917 -a <level>\tSet file access level (0=none, 1=cwd, 2=/etc,/lib,..., 3=whole fs, 9=no checks; needs -f)\n\
918 -c <dir>\tChange directory to <dir> first\n\
919 -e\t\tInherit full environment of the parent process\n\
920 -E <var>\tInherit the environment variable <var> from the parent process\n\
921 -E <var>=<val>\tSet the environment variable <var> to <val>; unset it if <var> is empty\n\
922 -f\t\tFilter system calls (-ff=very restricted)\n\
923 -i <file>\tRedirect stdin from <file>\n\
924 -m <size>\tLimit address space to <size> KB\n\
925 -M <file>\tOutput process information to <file> (name:value)\n\
926 -o <file>\tRedirect stdout to <file>\n\
927 -p <path>\tPermit access to the specified path (or subtree if it ends with a `/')\n\
928 -p <path>=<act>\tDefine action for the specified path (<act>=yes/no)\n\
929 -r <file>\tRedirect stderr to <file>\n\
930 -s <sys>\tPermit the specified syscall (be careful)\n\
931 -s <sys>=<act>\tDefine action for the specified syscall (<act>=yes/no/file)\n\
932 -t <time>\tSet run time limit (seconds, fractions allowed)\n\
933 -T\t\tAllow syscalls for measuring run time\n\
934 -v\t\tBe verbose (use multiple times for even more verbosity)\n\
935 -w <time>\tSet wall clock time limit (seconds, fractions allowed)\n\
941 main(int argc, char **argv)
946 while ((c = getopt(argc, argv, "a:c:eE:fi:m:M:o:p:r:s:t:Tvw:")) >= 0)
950 file_access = atol(optarg);
959 if (!set_env_action(optarg))
966 redir_stdin = optarg;
969 memory_limit = atol(optarg);
975 redir_stdout = optarg;
978 if (!set_path_action(optarg))
982 redir_stderr = optarg;
985 if (!set_syscall_action(optarg))
989 timeout = 1000*atof(optarg);
992 syscall_action[__NR_times] = A_YES;
998 wall_timeout = 1000*atof(optarg);
1007 if (setreuid(uid, uid) < 0)
1008 die("setreuid: %m");
1013 box_inside(argc-optind, argv+optind);
1016 die("Internal error: fell over edge of the world");