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 static void die(char *msg, ...) NONRET;
54 static FILE *metafile;
57 meta_open(const char *name)
59 if (!strcmp(name, "-"))
64 metafile = fopen(name, "w");
66 die("Failed to open metafile '%s'",name);
72 if (metafile && metafile != stdout)
76 static void __attribute__((format(printf,1,2)))
77 meta_printf(const char *fmt, ...)
84 vfprintf(metafile, fmt, args);
88 static int total_ms, wall_ms;
91 final_stats(struct rusage *rus)
93 struct timeval total, now, wall;
94 timeradd(&rus->ru_utime, &rus->ru_stime, &total);
95 total_ms = total.tv_sec*1000 + total.tv_usec/1000;
96 gettimeofday(&now, NULL);
97 timersub(&now, &start_time, &wall);
98 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
100 meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000);
101 meta_printf("time-wall:%d.%03d\n", wall_ms/1000, wall_ms%1000);
104 /*** Messages and exits ***/
112 ptrace(PTRACE_KILL, box_pid);
113 kill(-box_pid, SIGKILL);
114 kill(box_pid, SIGKILL);
118 int p = wait4(box_pid, &stat, 0, &rus);
120 fprintf(stderr, "UGH: Lost track of the process\n");
136 /* Report an error of the sandbox itself */
137 static void NONRET __attribute__((format(printf,1,2)))
144 vsnprintf(buf, sizeof(buf), msg, args);
145 meta_printf("status:XX\nmessage:%s\n", buf);
151 /* Report an error of the program inside the sandbox */
152 static void NONRET __attribute__((format(printf,1,2)))
158 if (msg[0] && msg[1] && msg[2] == ':' && msg[3] == ' ')
160 meta_printf("status:%c%c\n", msg[0], msg[1]);
164 vsnprintf(buf, sizeof(buf), msg, args);
165 meta_printf("message:%s\n", buf);
171 /* Write a message, but only if in verbose mode */
172 static void __attribute__((format(printf,1,2)))
179 int len = strlen(msg);
181 partial_line = (msg[len-1] != '\n');
182 vfprintf(stderr, msg, args);
191 void *p = malloc(size);
193 die("Out of memory");
197 /*** Syscall rules ***/
199 static const char * const syscall_names[] = {
200 #include "box/syscall-table.h"
202 #define NUM_SYSCALLS ARRAY_SIZE(syscall_names)
203 #define NUM_ACTIONS (NUM_SYSCALLS+64)
206 A_DEFAULT, // Use the default action
207 A_NO, // Always forbid
208 A_YES, // Always permit
209 A_FILENAME, // Permit if arg1 is a known filename
210 A_LIBERAL = 128, // Valid only in liberal mode
213 static unsigned char syscall_action[NUM_ACTIONS] = {
214 #define S(x) [__NR_##x]
216 // Syscalls permitted for specific file names
217 S(open) = A_FILENAME,
218 S(creat) = A_FILENAME,
219 S(unlink) = A_FILENAME,
220 S(oldstat) = A_FILENAME,
221 S(access) = A_FILENAME,
222 S(oldlstat) = A_FILENAME,
223 S(truncate) = A_FILENAME,
224 S(stat) = A_FILENAME,
225 S(lstat) = A_FILENAME,
226 S(truncate64) = A_FILENAME,
227 S(stat64) = A_FILENAME,
228 S(lstat64) = A_FILENAME,
229 S(readlink) = A_FILENAME,
231 // Syscalls permitted always
246 S(ftruncate) = A_YES,
248 S(personality) = A_YES,
252 S(getresuid) = A_YES,
260 S(ftruncate64) = A_YES,
270 S(set_thread_area) = A_YES,
271 S(get_thread_area) = A_YES,
272 S(set_tid_address) = A_YES,
273 S(exit_group) = A_YES,
275 // Syscalls permitted only in liberal mode
276 S(time) = A_YES | A_LIBERAL,
277 S(alarm) = A_YES | A_LIBERAL,
278 S(pause) = A_YES | A_LIBERAL,
279 S(signal) = A_YES | A_LIBERAL,
280 S(fchmod) = A_YES | A_LIBERAL,
281 S(sigaction) = A_YES | A_LIBERAL,
282 S(sgetmask) = A_YES | A_LIBERAL,
283 S(ssetmask) = A_YES | A_LIBERAL,
284 S(sigsuspend) = A_YES | A_LIBERAL,
285 S(sigpending) = A_YES | A_LIBERAL,
286 S(getrlimit) = A_YES | A_LIBERAL,
287 S(getrusage) = A_YES | A_LIBERAL,
288 S(ugetrlimit) = A_YES | A_LIBERAL,
289 S(gettimeofday) = A_YES | A_LIBERAL,
290 S(select) = A_YES | A_LIBERAL,
291 S(readdir) = A_YES | A_LIBERAL,
292 S(setitimer) = A_YES | A_LIBERAL,
293 S(getitimer) = A_YES | A_LIBERAL,
294 S(sigreturn) = A_YES | A_LIBERAL,
295 S(mprotect) = A_YES | A_LIBERAL,
296 S(sigprocmask) = A_YES | A_LIBERAL,
297 S(getdents) = A_YES | A_LIBERAL,
298 S(getdents64) = A_YES | A_LIBERAL,
299 S(_newselect) = A_YES | A_LIBERAL,
300 S(fdatasync) = A_YES | A_LIBERAL,
301 S(mremap) = A_YES | A_LIBERAL,
302 S(poll) = A_YES | A_LIBERAL,
303 S(getcwd) = A_YES | A_LIBERAL,
304 S(nanosleep) = A_YES | A_LIBERAL,
305 S(rt_sigreturn) = A_YES | A_LIBERAL,
306 S(rt_sigaction) = A_YES | A_LIBERAL,
307 S(rt_sigprocmask) = A_YES | A_LIBERAL,
308 S(rt_sigpending) = A_YES | A_LIBERAL,
309 S(rt_sigtimedwait) = A_YES | A_LIBERAL,
310 S(rt_sigqueueinfo) = A_YES | A_LIBERAL,
311 S(rt_sigsuspend) = A_YES | A_LIBERAL,
312 S(_sysctl) = A_YES | A_LIBERAL,
317 syscall_name(unsigned int id, char *buf)
319 if (id < NUM_SYSCALLS && syscall_names[id])
320 return syscall_names[id];
323 sprintf(buf, "#%d", id);
329 syscall_by_name(char *name)
331 for (unsigned int i=0; i<NUM_SYSCALLS; i++)
332 if (syscall_names[i] && !strcmp(syscall_names[i], name))
339 unsigned long l = strtoul(name, &ep, 0);
342 if (l >= NUM_ACTIONS)
348 set_syscall_action(char *a)
350 char *sep = strchr(a, '=');
351 enum action act = A_YES;
355 if (!strcmp(sep, "yes"))
357 else if (!strcmp(sep, "no"))
359 else if (!strcmp(sep, "file"))
365 int sys = syscall_by_name(a);
367 die("Unknown syscall `%s'", a);
368 if (sys >= NUM_ACTIONS)
369 die("Syscall `%s' out of range", a);
370 syscall_action[sys] = act;
379 struct path_rule *next;
382 static struct path_rule default_path_rules[] = {
385 { "/usr/lib/", A_YES },
386 { "/opt/lib/", A_YES },
387 { "/usr/share/zoneinfo/", A_YES },
388 { "/usr/share/locale/", A_YES },
389 { "/dev/null", A_YES },
390 { "/dev/zero", A_YES },
391 { "/proc/meminfo", A_YES },
392 { "/proc/self/stat", A_YES },
393 { "/proc/self/exe", A_YES }, // Needed by FPC 2.0.x runtime
396 static struct path_rule *user_path_rules;
397 static struct path_rule **last_path_rule = &user_path_rules;
400 set_path_action(char *a)
402 char *sep = strchr(a, '=');
403 enum action act = A_YES;
407 if (!strcmp(sep, "yes"))
409 else if (!strcmp(sep, "no"))
415 struct path_rule *r = xmalloc(sizeof(*r) + strlen(a) + 1);
416 r->path = (char *)(r+1);
421 last_path_rule = &r->next;
426 match_path_rule(struct path_rule *r, char *path)
430 if (*rr++ != *path++)
432 if (rr[-1] == '/' && !path[-1])
436 if (rr > r->path && rr[-1] != '/' && *path)
441 /*** Environment rules ***/
444 char *var; // Variable to match
445 char *val; // ""=clear, NULL=inherit
447 struct env_rule *next;
450 static struct env_rule *first_env_rule;
451 static struct env_rule **last_env_rule = &first_env_rule;
453 static struct env_rule default_env_rules[] = {
454 { "LIBC_FATAL_STDERR_", "1" }
458 set_env_action(char *a0)
460 struct env_rule *r = xmalloc(sizeof(*r) + strlen(a0) + 1);
461 char *a = (char *)(r+1);
464 char *sep = strchr(a, '=');
476 last_env_rule = &r->next;
482 match_env_var(char *env_entry, struct env_rule *r)
484 if (strncmp(env_entry, r->var, r->var_len))
486 return (env_entry[r->var_len] == '=');
490 apply_env_rule(char **env, int *env_sizep, struct env_rule *r)
492 // First remove the variable if already set
494 while (pos < *env_sizep && !match_env_var(env[pos], r))
496 if (pos < *env_sizep)
499 env[pos] = env[*env_sizep];
500 env[*env_sizep] = NULL;
503 // What is the new value?
509 new = xmalloc(r->var_len + 1 + strlen(r->val) + 1);
510 sprintf(new, "%s=%s", r->var, r->val);
515 while (environ[pos] && !match_env_var(environ[pos], r))
517 if (!(new = environ[pos]))
521 // Add it at the end of the array
522 env[(*env_sizep)++] = new;
523 env[*env_sizep] = NULL;
527 setup_environment(void)
529 // Link built-in rules with user rules
530 for (int i=ARRAY_SIZE(default_env_rules)-1; i >= 0; i--)
532 default_env_rules[i].next = first_env_rule;
533 first_env_rule = &default_env_rules[i];
536 // Scan the original environment
537 char **orig_env = environ;
539 while (orig_env[orig_size])
542 // For each rule, reserve one more slot and calculate length
544 for (struct env_rule *r = first_env_rule; r; r=r->next)
547 r->var_len = strlen(r->var);
550 // Create a new environment
551 char **env = xmalloc((orig_size + num_rules + 1) * sizeof(char *));
555 memcpy(env, environ, orig_size * sizeof(char *));
562 // Apply the rules one by one
563 for (struct env_rule *r = first_env_rule; r; r=r->next)
564 apply_env_rule(env, &size, r);
566 // Return the new env and pass some gossip
569 fprintf(stderr, "Passing environment:\n");
570 for (int i=0; env[i]; i++)
571 fprintf(stderr, "\t%s\n", env[i]);
576 /*** Syscall checks ***/
579 valid_filename(unsigned long addr)
581 char namebuf[4096], *p, *end;
585 err("FA: File access forbidden");
586 if (file_access >= 9)
591 sprintf(namebuf, "/proc/%d/mem", (int) box_pid);
592 mem_fd = open(namebuf, O_RDONLY);
594 die("open(%s): %m", namebuf);
601 int remains = PAGE_SIZE - (addr & (PAGE_SIZE-1));
602 int l = namebuf + sizeof(namebuf) - end;
606 err("FA: Access to file with name too long");
607 if (lseek64(mem_fd, addr, SEEK_SET) < 0)
608 die("lseek64(mem): %m");
609 remains = read(mem_fd, end, l);
611 die("read(mem): %m");
613 err("FA: Access to file with name out of memory");
620 msg("[%s] ", namebuf);
621 if (file_access >= 3)
624 // Everything in current directory is permitted
625 if (!strchr(namebuf, '/') && strcmp(namebuf, ".."))
628 // ".." anywhere in the path is forbidden
629 enum action act = A_DEFAULT;
630 if (strstr(namebuf, ".."))
634 for (struct path_rule *r = user_path_rules; r && !act; r=r->next)
635 act = match_path_rule(r, namebuf);
637 // Scan built-in rules
638 if (file_access >= 2)
639 for (int i=0; i<ARRAY_SIZE(default_path_rules) && !act; i++)
640 act = match_path_rule(&default_path_rules[i], namebuf);
643 err("FA: Forbidden access to file `%s'", namebuf);
647 valid_syscall(struct user *u)
649 unsigned int sys = u->regs.orig_eax;
650 enum action act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT;
654 if (filter_syscalls == 1)
666 valid_filename(u->regs.ebx);
674 if (u->regs.ebx == box_pid)
676 meta_printf("exitsig:%d\n", (int)u->regs.ecx);
677 err("SG: Committed suicide by signal %d", (int)u->regs.ecx);
681 if (u->regs.ebx == box_pid && u->regs.ecx == box_pid)
683 meta_printf("exitsig:%d\n", (int)u->regs.edx);
684 err("SG: Committed suicide by signal %d", (int)u->regs.edx);
693 signal_alarm(int unused UNUSED)
695 /* Time limit checks are synchronous, so we only schedule them there. */
701 signal_int(int unused UNUSED)
703 /* Interrupts are fatal, so no synchronization requirements. */
704 meta_printf("exitsig:%d\n", SIGINT);
705 err("SG: Interrupted");
713 struct timeval now, wall;
715 gettimeofday(&now, NULL);
716 timersub(&now, &start_time, &wall);
717 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
718 if (wall_ms > wall_timeout)
719 err("TO: Time limit exceeded (wall clock)");
721 fprintf(stderr, "[wall time check: %d msec]\n", wall_ms);
726 int c, utime, stime, ms;
727 static int proc_status_fd;
730 sprintf(buf, "/proc/%d/stat", (int) box_pid);
731 proc_status_fd = open(buf, O_RDONLY);
732 if (proc_status_fd < 0)
733 die("open(%s): %m", buf);
735 lseek(proc_status_fd, 0, SEEK_SET);
736 if ((c = read(proc_status_fd, buf, sizeof(buf)-1)) < 0)
737 die("read on /proc/$pid/stat: %m");
738 if (c >= (int) sizeof(buf) - 1)
739 die("/proc/$pid/stat too long");
742 while (*x && *x != ' ')
747 die("proc syntax error 1");
748 while (*x && (*x != ')' || x[1] != ' '))
750 while (*x == ')' || *x == ' ')
752 if (sscanf(x, "%*c %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d %d %d", &utime, &stime) != 2)
753 die("proc syntax error 2");
754 ms = (utime + stime) * 1000 / ticks_per_sec;
756 fprintf(stderr, "[time check: %d msec]\n", ms);
758 err("TO: Time limit exceeded");
765 int syscall_count = 0;
769 bzero(&sa, sizeof(sa));
770 sa.sa_handler = signal_int;
771 sigaction(SIGINT, &sa, NULL);
772 gettimeofday(&start_time, NULL);
773 ticks_per_sec = sysconf(_SC_CLK_TCK);
774 if (ticks_per_sec <= 0)
775 die("Invalid ticks_per_sec!");
776 if (timeout || wall_timeout)
778 sa.sa_handler = signal_alarm;
779 sigaction(SIGALRM, &sa, NULL);
792 p = wait4(box_pid, &stat, WUNTRACED, &rus);
800 die("wait4: unknown pid %d exited!", p);
805 if (WEXITSTATUS(stat))
809 meta_printf("exitcode:%d\n", WEXITSTATUS(stat));
810 err("RE: Exited with error status %d", WEXITSTATUS(stat));
814 // Internal error happened inside the child process and it has been already reported.
818 if (timeout && total_ms > timeout)
819 err("TO: Time limit exceeded");
820 if (wall_timeout && wall_ms > wall_timeout)
821 err("TO: Time limit exceeded (wall clock)");
823 fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n",
824 total_ms/1000, total_ms%1000,
825 wall_ms/1000, wall_ms%1000,
829 if (WIFSIGNALED(stat))
832 meta_printf("exitsig:%d\n", WTERMSIG(stat));
833 err("SG: Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup"));
835 if (WIFSTOPPED(stat))
837 int sig = WSTOPSIG(stat);
841 static int stop_count = -1;
842 if (ptrace(PTRACE_GETREGS, box_pid, NULL, &u) < 0)
843 die("ptrace(PTRACE_GETREGS): %m");
845 if (!stop_count) /* Traceme request */
846 msg(">> Traceme request caught\n");
847 else if (stop_count & 1) /* Syscall entry */
850 msg(">> Syscall %-12s (%08lx,%08lx,%08lx) ", syscall_name(u.regs.orig_eax, namebuf), u.regs.ebx, u.regs.ecx, u.regs.edx);
854 if (u.regs.orig_eax == __NR_execve)
857 else if (valid_syscall(&u))
862 * Unfortunately, PTRACE_KILL kills _after_ the syscall completes,
863 * so we have to change it to something harmless (e.g., an undefined
864 * syscall) and make the program continue.
866 unsigned int sys = u.regs.orig_eax;
867 u.regs.orig_eax = 0xffffffff;
868 if (ptrace(PTRACE_SETREGS, box_pid, NULL, &u) < 0)
869 die("ptrace(PTRACE_SETREGS): %m");
870 err("FO: Forbidden syscall %s", syscall_name(sys, namebuf));
873 else /* Syscall return */
874 msg("= %ld\n", u.regs.eax);
875 ptrace(PTRACE_SYSCALL, box_pid, 0, 0);
877 else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ)
879 msg(">> Signal %d\n", sig);
880 ptrace(PTRACE_SYSCALL, box_pid, 0, sig);
884 meta_printf("exitsig:%d", sig);
885 err("SG: Received signal %d", sig);
889 die("wait4: unknown status %x, giving up!", stat);
894 box_inside(int argc, char **argv)
899 memcpy(args, argv, argc * sizeof(char *));
901 if (set_cwd && chdir(set_cwd))
906 if (open(redir_stdin, O_RDONLY) != 0)
907 die("open(\"%s\"): %m", redir_stdin);
912 if (open(redir_stdout, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 1)
913 die("open(\"%s\"): %m", redir_stdout);
918 if (open(redir_stderr, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 2)
919 die("open(\"%s\"): %m", redir_stderr);
926 rl.rlim_cur = rl.rlim_max = memory_limit * 1024;
927 if (setrlimit(RLIMIT_AS, &rl) < 0)
928 die("setrlimit: %m");
930 rl.rlim_cur = rl.rlim_max = 64;
931 if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
932 die("setrlimit: %m");
935 if (ptrace(PTRACE_TRACEME) < 0)
936 die("ptrace(PTRACE_TRACEME): %m");
937 /* Trick: Make sure that we are stopped until the boxkeeper wakes up. */
938 signal(SIGCHLD, SIG_IGN);
941 execve(args[0], args, setup_environment());
942 die("execve(\"%s\"): %m", args[0]);
948 fprintf(stderr, "Invalid arguments!\n");
950 Usage: box [<options>] -- <command> <arguments>\n\
953 -a <level>\tSet file access level (0=none, 1=cwd, 2=/etc,/lib,..., 3=whole fs, 9=no checks; needs -f)\n\
954 -c <dir>\tChange directory to <dir> first\n\
955 -e\t\tInherit full environment of the parent process\n\
956 -E <var>\tInherit the environment variable <var> from the parent process\n\
957 -E <var>=<val>\tSet the environment variable <var> to <val>; unset it if <var> is empty\n\
958 -f\t\tFilter system calls (-ff=very restricted)\n\
959 -i <file>\tRedirect stdin from <file>\n\
960 -m <size>\tLimit address space to <size> KB\n\
961 -M <file>\tOutput process information to <file> (name:value)\n\
962 -o <file>\tRedirect stdout to <file>\n\
963 -p <path>\tPermit access to the specified path (or subtree if it ends with a `/')\n\
964 -p <path>=<act>\tDefine action for the specified path (<act>=yes/no)\n\
965 -r <file>\tRedirect stderr to <file>\n\
966 -s <sys>\tPermit the specified syscall (be careful)\n\
967 -s <sys>=<act>\tDefine action for the specified syscall (<act>=yes/no/file)\n\
968 -t <time>\tSet run time limit (seconds, fractions allowed)\n\
969 -T\t\tAllow syscalls for measuring run time\n\
970 -v\t\tBe verbose (use multiple times for even more verbosity)\n\
971 -w <time>\tSet wall clock time limit (seconds, fractions allowed)\n\
977 main(int argc, char **argv)
982 while ((c = getopt(argc, argv, "a:c:eE:fi:m:M:o:p:r:s:t:Tvw:")) >= 0)
986 file_access = atol(optarg);
995 if (!set_env_action(optarg))
1002 redir_stdin = optarg;
1005 memory_limit = atol(optarg);
1011 redir_stdout = optarg;
1014 if (!set_path_action(optarg))
1018 redir_stderr = optarg;
1021 if (!set_syscall_action(optarg))
1025 timeout = 1000*atof(optarg);
1028 syscall_action[__NR_times] = A_YES;
1034 wall_timeout = 1000*atof(optarg);
1043 if (setreuid(uid, uid) < 0)
1044 die("setreuid: %m");
1049 box_inside(argc-optind, argv+optind);
1052 die("Internal error: fell over edge of the world");