X-Git-Url: http://mj.ucw.cz/gitweb/?a=blobdiff_plain;f=src%2Fbox.c;h=6c120f821f36436ddd478eb5f16fb6ac9d4be094;hb=62f031249f096887b62544f8e4513fd225ac86fd;hp=b76fd8a06037a8fc8fa34e3d8f9cb6b1cb40fb1c;hpb=666506e10535397d92733c7b6df0421c18708d04;p=moe.git diff --git a/src/box.c b/src/box.c index b76fd8a..6c120f8 100644 --- a/src/box.c +++ b/src/box.c @@ -1,7 +1,7 @@ /* - * A Simple Testing Sandbox + * A Simple Sandbox for MO-Eval * - * (c) 2001 Martin Mares + * (c) 2001--2008 Martin Mares */ #define _LARGEFILE64_SOURCE @@ -27,21 +27,24 @@ #define NONRET __attribute__((noreturn)) #define UNUSED __attribute__((unused)) +#define ARRAY_SIZE(a) (int)(sizeof(a)/sizeof(a[0])) static int filter_syscalls; /* 0=off, 1=liberal, 2=totalitarian */ -static int timeout; +static int timeout; /* milliseconds */ +static int wall_timeout; static int pass_environ; -static int use_wall_clock; static int file_access; static int verbose; static int memory_limit; static char *redir_stdin, *redir_stdout; +static char *set_cwd; static pid_t box_pid; static int is_ptraced; static volatile int timer_tick; -static time_t start_time; +static struct timeval start_time; static int ticks_per_sec; +static int exec_seen; #if defined(__GLIBC__) && __GLIBC__ == 2 && __GLIBC_MINOR__ > 0 /* glibc 2.1 or newer -> has lseek64 */ @@ -76,7 +79,7 @@ die(char *msg, ...) } static void __attribute__((format(printf,1,2))) -log(char *msg, ...) +msg(char *msg, ...) { va_list args; va_start(args, msg); @@ -88,6 +91,254 @@ log(char *msg, ...) va_end(args); } +static void * +xmalloc(size_t size) +{ + void *p = malloc(size); + if (!p) + die("Out of memory"); + return p; +} + +static const char * const syscall_names[] = { +#include "syscall-table.h" +}; +#define NUM_SYSCALLS ARRAY_SIZE(syscall_names) +#define NUM_ACTIONS (NUM_SYSCALLS+64) + +enum action { + A_DEFAULT, // Use the default action + A_NO, // Always forbid + A_YES, // Always permit + A_FILENAME, // Permit if arg1 is a known filename + A_LIBERAL = 128, // Valid only in liberal mode +}; + +static unsigned char syscall_action[NUM_ACTIONS] = { +#define S(x) [__NR_##x] + + // Syscalls permitted for specific file names + S(open) = A_FILENAME, + S(creat) = A_FILENAME, + S(unlink) = A_FILENAME, + S(oldstat) = A_FILENAME, + S(access) = A_FILENAME, + S(oldlstat) = A_FILENAME, + S(truncate) = A_FILENAME, + S(stat) = A_FILENAME, + S(lstat) = A_FILENAME, + S(truncate64) = A_FILENAME, + S(stat64) = A_FILENAME, + S(lstat64) = A_FILENAME, + S(readlink) = A_FILENAME, + + // Syscalls permitted always + S(exit) = A_YES, + S(read) = A_YES, + S(write) = A_YES, + S(close) = A_YES, + S(lseek) = A_YES, + S(getpid) = A_YES, + S(getuid) = A_YES, + S(oldfstat) = A_YES, + S(dup) = A_YES, + S(brk) = A_YES, + S(getgid) = A_YES, + S(geteuid) = A_YES, + S(getegid) = A_YES, + S(dup2) = A_YES, + S(ftruncate) = A_YES, + S(fstat) = A_YES, + S(personality) = A_YES, + S(_llseek) = A_YES, + S(readv) = A_YES, + S(writev) = A_YES, + S(getresuid) = A_YES, +#ifdef __NR_pread64 + S(pread64) = A_YES, + S(pwrite64) = A_YES, +#else + S(pread) = A_YES, + S(pwrite) = A_YES, +#endif + S(ftruncate64) = A_YES, + S(fstat64) = A_YES, + S(fcntl) = A_YES, + S(fcntl64) = A_YES, + S(mmap) = A_YES, + S(munmap) = A_YES, + S(ioctl) = A_YES, + S(uname) = A_YES, + S(gettid) = A_YES, + S(set_thread_area) = A_YES, + S(get_thread_area) = A_YES, + S(exit_group) = A_YES, + + // Syscalls permitted only in liberal mode + S(time) = A_YES | A_LIBERAL, + S(alarm) = A_YES | A_LIBERAL, + S(pause) = A_YES | A_LIBERAL, + S(signal) = A_YES | A_LIBERAL, + S(fchmod) = A_YES | A_LIBERAL, + S(sigaction) = A_YES | A_LIBERAL, + S(sgetmask) = A_YES | A_LIBERAL, + S(ssetmask) = A_YES | A_LIBERAL, + S(sigsuspend) = A_YES | A_LIBERAL, + S(sigpending) = A_YES | A_LIBERAL, + S(getrlimit) = A_YES | A_LIBERAL, + S(getrusage) = A_YES | A_LIBERAL, + S(ugetrlimit) = A_YES | A_LIBERAL, + S(gettimeofday) = A_YES | A_LIBERAL, + S(select) = A_YES | A_LIBERAL, + S(readdir) = A_YES | A_LIBERAL, + S(setitimer) = A_YES | A_LIBERAL, + S(getitimer) = A_YES | A_LIBERAL, + S(sigreturn) = A_YES | A_LIBERAL, + S(mprotect) = A_YES | A_LIBERAL, + S(sigprocmask) = A_YES | A_LIBERAL, + S(getdents) = A_YES | A_LIBERAL, + S(getdents64) = A_YES | A_LIBERAL, + S(_newselect) = A_YES | A_LIBERAL, + S(fdatasync) = A_YES | A_LIBERAL, + S(mremap) = A_YES | A_LIBERAL, + S(poll) = A_YES | A_LIBERAL, + S(getcwd) = A_YES | A_LIBERAL, + S(nanosleep) = A_YES | A_LIBERAL, + S(rt_sigreturn) = A_YES | A_LIBERAL, + S(rt_sigaction) = A_YES | A_LIBERAL, + S(rt_sigprocmask) = A_YES | A_LIBERAL, + S(rt_sigpending) = A_YES | A_LIBERAL, + S(rt_sigtimedwait) = A_YES | A_LIBERAL, + S(rt_sigqueueinfo) = A_YES | A_LIBERAL, + S(rt_sigsuspend) = A_YES | A_LIBERAL, + S(mmap2) = A_YES | A_LIBERAL, + S(_sysctl) = A_YES | A_LIBERAL, +#undef S +}; + +static const char * +syscall_name(unsigned int id, char *buf) +{ + if (id < NUM_SYSCALLS && syscall_names[id]) + return syscall_names[id]; + else + { + sprintf(buf, "#%d", id); + return buf; + } +} + +static int +syscall_by_name(char *name) +{ + for (unsigned int i=0; i= NUM_ACTIONS) + return NUM_ACTIONS; + return l; +} + +static int +set_syscall_action(char *a) +{ + char *sep = strchr(a, '='); + enum action act = A_YES; + if (sep) + { + *sep++ = 0; + if (!strcmp(sep, "yes")) + act = A_YES; + else if (!strcmp(sep, "no")) + act = A_NO; + else if (!strcmp(sep, "file")) + act = A_FILENAME; + else + return 0; + } + + int sys = syscall_by_name(a); + if (sys < 0) + die("Unknown syscall `%s'", a); + if (sys >= NUM_ACTIONS) + die("Syscall `%s' out of range", a); + syscall_action[sys] = act; + return 1; +} + +struct path_rule { + char *path; + enum action action; + struct path_rule *next; +}; + +static struct path_rule default_path_rules[] = { + { "/etc/", A_YES }, + { "/lib/", A_YES }, + { "/usr/lib/", A_YES }, + { "/opt/lib/", A_YES }, + { "/usr/share/zoneinfo/", A_YES }, + { "/usr/share/locale/", A_YES }, + { "/dev/null", A_YES }, + { "/dev/zero", A_YES }, + { "/proc/meminfo", A_YES }, + { "/proc/self/stat", A_YES }, + { "/proc/self/exe", A_YES }, // Needed by FPC 2.0.x runtime +}; + +static struct path_rule *user_path_rules; +static struct path_rule **last_path_rule = &user_path_rules; + +static int +set_path_action(char *a) +{ + char *sep = strchr(a, '='); + enum action act = A_YES; + if (sep) + { + *sep++ = 0; + if (!strcmp(sep, "yes")) + act = A_YES; + else if (!strcmp(sep, "no")) + act = A_NO; + else + return 0; + } + + struct path_rule *r = xmalloc(sizeof(*r) + strlen(a)); + r->path = (char *)(r+1); + strcpy(r->path, a); + r->action = act; + r->next = NULL; + *last_path_rule = r; + last_path_rule = &r->next; + return 1; +} + +static enum action +match_path_rule(struct path_rule *r, char *path) +{ + char *rr = r->path; + while (*rr) + if (*rr++ != *path++) + { + if (rr[-1] == '/' && !path[-1]) + break; + return A_DEFAULT; + } + if (rr > r->path && rr[-1] != '/' && *path) + return A_DEFAULT; + return r->action; +} + static void valid_filename(unsigned long addr) { @@ -95,7 +346,7 @@ valid_filename(unsigned long addr) static int mem_fd; if (!file_access) - die("File access forbidden."); + die("File access forbidden"); if (file_access >= 9) return; @@ -116,130 +367,81 @@ valid_filename(unsigned long addr) if (l > remains) l = remains; if (!l) - die("Access to file with name too long."); + die("Access to file with name too long"); if (long_seek(mem_fd, addr, SEEK_SET) < 0) die("long_seek(mem): %m"); remains = read(mem_fd, end, l); if (remains < 0) die("read(mem): %m"); if (!remains) - die("Access to file with name out of memory."); + die("Access to file with name out of memory"); end += l; addr += l; } } while (*p++); - log("[%s] ", namebuf); + msg("[%s] ", namebuf); if (file_access >= 3) return; + + // Everything in current directory is permitted if (!strchr(namebuf, '/') && strcmp(namebuf, "..")) return; + + // ".." anywhere in the path is forbidden + enum action act = A_DEFAULT; + if (strstr(namebuf, "..")) + act = A_NO; + + // Scan user rules + for (struct path_rule *r = user_path_rules; r && !act; r=r->next) + act = match_path_rule(r, namebuf); + + // Scan built-in rules if (file_access >= 2) - { - if ((!strncmp(namebuf, "/etc/", 5) || - !strncmp(namebuf, "/lib/", 5) || - !strncmp(namebuf, "/usr/lib/", 9)) - && !strstr(namebuf, "..")) - return; - if (!strcmp(namebuf, "/dev/null") || - !strcmp(namebuf, "/dev/zero")) - return; - } - die("Forbidden access to file `%s'.", namebuf); + for (int i=0; iregs.orig_eax) + unsigned int sys = u->regs.orig_eax; + enum action act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT; + + if (act & A_LIBERAL) { - case SYS_execve: - { - static int exec_counter; - return !exec_counter++; - } - case SYS_open: - case SYS_creat: - case SYS_unlink: - case SYS_oldstat: - case SYS_access: - case SYS_oldlstat: - case SYS_truncate: - case SYS_stat: - case SYS_lstat: - case SYS_truncate64: - case SYS_stat64: - case SYS_lstat64: - valid_filename(u->regs.ebx); + if (filter_syscalls == 1) + act &= ~A_LIBERAL; + else + act = A_DEFAULT; + } + switch (act) + { + case A_YES: return 1; - case SYS_exit: - case SYS_read: - case SYS_write: - case SYS_close: - case SYS_lseek: - case SYS_getpid: - case SYS_getuid: - case SYS_oldfstat: - case SYS_dup: - case SYS_brk: - case SYS_getgid: - case SYS_geteuid: - case SYS_getegid: - case SYS_dup2: - case SYS_ftruncate: - case SYS_fstat: - case SYS_personality: - case SYS__llseek: - case SYS_readv: - case SYS_writev: - case SYS_getresuid: - case SYS_pread: - case SYS_pwrite: - case SYS_ftruncate64: - case SYS_fstat64: - case SYS_fcntl: - case SYS_mmap: - case SYS_munmap: - case SYS_ioctl: - case SYS_uname: + case A_NO: + return 0; + case A_FILENAME: + valid_filename(u->regs.ebx); return 1; - case SYS_time: - case SYS_alarm: - case SYS_pause: - case SYS_signal: - case SYS_fchmod: - case SYS_sigaction: - case SYS_sgetmask: - case SYS_ssetmask: - case SYS_sigsuspend: - case SYS_sigpending: - case SYS_getrlimit: - case SYS_getrusage: - case SYS_gettimeofday: - case SYS_select: - case SYS_readdir: - case SYS_setitimer: - case SYS_getitimer: - case SYS_sigreturn: - case SYS_mprotect: - case SYS_sigprocmask: - case SYS_getdents: - case SYS__newselect: - case SYS_fdatasync: - case SYS_mremap: - case SYS_poll: - case SYS_getcwd: - case SYS_nanosleep: - case SYS_rt_sigreturn: - case SYS_rt_sigaction: - case SYS_rt_sigprocmask: - case SYS_rt_sigpending: - case SYS_rt_sigtimedwait: - case SYS_rt_sigqueueinfo: - case SYS_rt_sigsuspend: - case SYS_mmap2: - return (filter_syscalls == 1); + default: ; + } + + switch (sys) + { + case __NR_kill: + if (u->regs.ebx == box_pid) + die("Committed suicide by signal %d", (int)u->regs.ecx); + return 0; + case __NR_tgkill: + if (u->regs.ebx == box_pid && u->regs.ecx == box_pid) + die("Committed suicide by signal %d", (int)u->regs.edx); + return 0; default: return 0; } @@ -257,20 +459,28 @@ static void signal_int(int unused UNUSED) { /* Interrupts are fatal, so no synchronization requirements. */ - die("Interrupted."); + die("Interrupted"); } static void check_timeout(void) { - int sec; - - if (use_wall_clock) - sec = time(NULL) - start_time; - else + if (wall_timeout) + { + struct timeval now, wall; + int wall_ms; + gettimeofday(&now, NULL); + timersub(&now, &start_time, &wall); + wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000; + if (wall_ms > wall_timeout) + die("Time limit exceeded (wall clock)"); + if (verbose > 1) + fprintf(stderr, "[wall time check: %d msec]\n", wall_ms); + } + if (timeout) { char buf[4096], *x; - int c, utime, stime; + int c, utime, stime, ms; static int proc_status_fd; if (!proc_status_fd) { @@ -298,12 +508,12 @@ check_timeout(void) x++; if (sscanf(x, "%*c %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d %d %d", &utime, &stime) != 2) die("proc syntax error 2"); - sec = (utime + stime)/ticks_per_sec; + ms = (utime + stime) * 1000 / ticks_per_sec; + if (verbose > 1) + fprintf(stderr, "[time check: %d msec]\n", ms); + if (ms > timeout) + die("Time limit exceeded"); } - if (verbose > 1) - fprintf(stderr, "[timecheck: %d seconds]\n", sec); - if (sec > timeout) - die("Time limit exceeded."); } static void @@ -316,11 +526,11 @@ boxkeeper(void) bzero(&sa, sizeof(sa)); sa.sa_handler = signal_int; sigaction(SIGINT, &sa, NULL); - start_time = time(NULL); + gettimeofday(&start_time, NULL); ticks_per_sec = sysconf(_SC_CLK_TCK); if (ticks_per_sec <= 0) die("Invalid ticks_per_sec!"); - if (timeout) + if (timeout || wall_timeout) { sa.sa_handler = signal_alarm; sigaction(SIGALRM, &sa, NULL); @@ -347,22 +557,30 @@ boxkeeper(void) die("wait4: unknown pid %d exited!", p); if (WIFEXITED(stat)) { - struct timeval total; - int wall; + struct timeval total, now, wall; + int total_ms, wall_ms; box_pid = 0; if (WEXITSTATUS(stat)) - die("Exited with error status %d.", WEXITSTATUS(stat)); + die("Exited with error status %d", WEXITSTATUS(stat)); timeradd(&rus.ru_utime, &rus.ru_stime, &total); - wall = time(NULL) - start_time; - if ((use_wall_clock ? wall : total.tv_sec) > timeout) - die("Timeout exceeded (after exit)."); - fprintf(stderr, "OK (%d sec real, %d sec wall, %d syscalls)\n", (int) total.tv_sec, wall, syscall_count); + total_ms = total.tv_sec*1000 + total.tv_usec/1000; + gettimeofday(&now, NULL); + timersub(&now, &start_time, &wall); + wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000; + if (timeout && total_ms > timeout) + die("Time limit exceeded"); + if (wall_timeout && wall_ms > wall_timeout) + die("Time limit exceeded (wall clock)"); + fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n", + (int) total.tv_sec, (int) total.tv_usec/1000, + (int) wall.tv_sec, (int) wall.tv_usec/1000, + syscall_count); exit(0); } if (WIFSIGNALED(stat)) { box_pid = 0; - die("Caught fatal signal %d.", WTERMSIG(stat)); + die("Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup")); } if (WIFSTOPPED(stat)) { @@ -375,12 +593,20 @@ boxkeeper(void) die("ptrace(PTRACE_GETREGS): %m"); stop_count++; if (!stop_count) /* Traceme request */ - log(">> Traceme request caught\n"); + msg(">> Traceme request caught\n"); else if (stop_count & 1) /* Syscall entry */ { - log(">> Syscall %3ld (%08lx,%08lx,%08lx) ", u.regs.orig_eax, u.regs.ebx, u.regs.ecx, u.regs.edx); - syscall_count++; - if (!valid_syscall(&u)) + char namebuf[32]; + msg(">> Syscall %-12s (%08lx,%08lx,%08lx) ", syscall_name(u.regs.orig_eax, namebuf), u.regs.ebx, u.regs.ecx, u.regs.edx); + if (!exec_seen) + { + msg("[master] "); + if (u.regs.orig_eax == __NR_execve) + exec_seen = 1; + } + else if (valid_syscall(&u)) + syscall_count++; + else { /* * Unfortunately, PTRACE_KILL kills _after_ the syscall completes, @@ -391,20 +617,20 @@ boxkeeper(void) u.regs.orig_eax = 0xffffffff; if (ptrace(PTRACE_SETREGS, box_pid, NULL, &u) < 0) die("ptrace(PTRACE_SETREGS): %m"); - die("Forbidden syscall %d.", sys); + die("Forbidden syscall %s", syscall_name(sys, namebuf)); } } else /* Syscall return */ - log("= %ld\n", u.regs.eax); + msg("= %ld\n", u.regs.eax); ptrace(PTRACE_SYSCALL, box_pid, 0, 0); } else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ) { - log(">> Signal %d\n", sig); + msg(">> Signal %d\n", sig); ptrace(PTRACE_SYSCALL, box_pid, 0, sig); } else - die("Received signal %d.", sig); + die("Received signal %d", sig); } else die("wait4: unknown status %x, giving up!", stat); @@ -416,10 +642,12 @@ box_inside(int argc, char **argv) { struct rlimit rl; char *args[argc+1]; - char *env[1] = { NULL }; + char *env[] = { "LIBC_FATAL_STDERR_=1", NULL }; memcpy(args, argv, argc * sizeof(char *)); args[argc] = NULL; + if (set_cwd && chdir(set_cwd)) + die("chdir: %m"); if (redir_stdin) { close(0); @@ -432,8 +660,7 @@ box_inside(int argc, char **argv) if (open(redir_stdout, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 1) die("open(\"%s\"): %m", redir_stdout); } - close(2); - dup(1); + dup2(1, 2); setpgrp(); if (memory_limit) { @@ -444,8 +671,14 @@ box_inside(int argc, char **argv) rl.rlim_cur = rl.rlim_max = 64; if (setrlimit(RLIMIT_NOFILE, &rl) < 0) die("setrlimit: %m"); - if (filter_syscalls && ptrace(PTRACE_TRACEME) < 0) - die("ptrace(PTRACE_TRACEME): %m"); + if (filter_syscalls) + { + if (ptrace(PTRACE_TRACEME) < 0) + die("ptrace(PTRACE_TRACEME): %m"); + /* Trick: Make sure that we are stopped until the boxkeeper wakes up. */ + signal(SIGCHLD, SIG_IGN); + raise(SIGCHLD); + } execve(args[0], args, (pass_environ ? environ : env)); die("execve(\"%s\"): %m", args[0]); } @@ -465,9 +698,14 @@ Options:\n\ -i \tRedirect stdin from \n\ -m \tLimit address space to KB\n\ -o \tRedirect stdout to \n\ --t