X-Git-Url: http://mj.ucw.cz/gitweb/?a=blobdiff_plain;ds=sidebyside;f=box%2Fbox.c;h=892cc53eb9b8bee98fb369f7b4e8810a342375ad;hb=4ee9f1ead0d2ef8133a8f1f692816b78d038ab4e;hp=1a84e106aa2547b00cb2c8eba49376809cca9a89;hpb=0616d23216a49e2e6eeb6587f1caac8c2eac0514;p=moe.git diff --git a/box/box.c b/box/box.c index 1a84e10..892cc53 100644 --- a/box/box.c +++ b/box/box.c @@ -24,6 +24,7 @@ #include #include #include +#include #define NONRET __attribute__((noreturn)) #define UNUSED __attribute__((unused)) @@ -32,10 +33,12 @@ static int filter_syscalls; /* 0=off, 1=liberal, 2=totalitarian */ static int timeout; /* milliseconds */ static int wall_timeout; +static int extra_timeout; static int pass_environ; static int file_access; static int verbose; static int memory_limit; +static int stack_limit; static char *redir_stdin, *redir_stdout, *redir_stderr; static char *set_cwd; @@ -47,16 +50,11 @@ static int ticks_per_sec; static int exec_seen; static int partial_line; -#if defined(__GLIBC__) && __GLIBC__ == 2 && __GLIBC_MINOR__ > 0 -/* glibc 2.1 or newer -> has lseek64 */ -#define long_seek(f,o,w) lseek64(f,o,w) -#else -/* Touching clandestine places in glibc */ -extern loff_t llseek(int fd, loff_t pos, int whence); -#define long_seek(f,o,w) llseek(f,o,w) -#endif +static int mem_peak_kb; +static int total_ms, wall_ms; static void die(char *msg, ...) NONRET; +static void sample_mem_peak(void); /*** Meta-files ***/ @@ -94,8 +92,6 @@ meta_printf(const char *fmt, ...) va_end(args); } -static int total_ms, wall_ms; - static void final_stats(struct rusage *rus) { @@ -103,13 +99,12 @@ final_stats(struct rusage *rus) timeradd(&rus->ru_utime, &rus->ru_stime, &total); total_ms = total.tv_sec*1000 + total.tv_usec/1000; gettimeofday(&now, NULL); - // FIXME: We are not guaranteed to have start_time always initialized timersub(&now, &start_time, &wall); wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000; - meta_printf("time:%d.%03d\ntime_wall:%d.%03d\n", - total_ms/1000, total_ms%1000, - wall_ms/1000, wall_ms%1000); + meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000); + meta_printf("time-wall:%d.%03d\n", wall_ms/1000, wall_ms%1000); + meta_printf("mem:%llu\n", (unsigned long long) mem_peak_kb * 1024); } /*** Messages and exits ***/ @@ -119,16 +114,20 @@ box_exit(int rc) { if (box_pid > 0) { + sample_mem_peak(); if (is_ptraced) ptrace(PTRACE_KILL, box_pid); kill(-box_pid, SIGKILL); kill(box_pid, SIGKILL); + meta_printf("killed:1\n"); struct rusage rus; - int stat; - int p = wait4(box_pid, &stat, 0, &rus); + int p, stat; + do + p = wait4(box_pid, &stat, 0, &rus); + while (p < 0 && errno == EINTR); if (p < 0) - fprintf(stderr, "UGH: Lost track of the process\n"); + fprintf(stderr, "UGH: Lost track of the process (%m)\n"); else final_stats(&rus); } @@ -144,18 +143,42 @@ flush_line(void) partial_line = 0; } +/* Report an error of the sandbox itself */ static void NONRET __attribute__((format(printf,1,2))) die(char *msg, ...) { va_list args; va_start(args, msg); flush_line(); - vfprintf(stderr, msg, args); + char buf[1024]; + vsnprintf(buf, sizeof(buf), msg, args); + meta_printf("status:XX\nmessage:%s\n", buf); + fputs(buf, stderr); + fputc('\n', stderr); + box_exit(2); +} + +/* Report an error of the program inside the sandbox */ +static void NONRET __attribute__((format(printf,1,2))) +err(char *msg, ...) +{ + va_list args; + va_start(args, msg); + flush_line(); + if (msg[0] && msg[1] && msg[2] == ':' && msg[3] == ' ') + { + meta_printf("status:%c%c\n", msg[0], msg[1]); + msg += 4; + } + char buf[1024]; + vsnprintf(buf, sizeof(buf), msg, args); + meta_printf("message:%s\n", buf); + fputs(buf, stderr); fputc('\n', stderr); box_exit(1); - // FIXME: exit(2) for errors of the box itself } +/* Write a message, but only if in verbose mode */ static void __attribute__((format(printf,1,2))) msg(char *msg, ...) { @@ -194,7 +217,11 @@ enum action { A_NO, // Always forbid A_YES, // Always permit A_FILENAME, // Permit if arg1 is a known filename + A_ACTION_MASK = 15, + A_NO_RETVAL = 32, // Does not return a value + A_SAMPLE_MEM = 64, // Sample memory usage before the syscall A_LIBERAL = 128, // Valid only in liberal mode + // Must fit in a unsigned char }; static unsigned char syscall_action[NUM_ACTIONS] = { @@ -216,7 +243,7 @@ static unsigned char syscall_action[NUM_ACTIONS] = { S(readlink) = A_FILENAME, // Syscalls permitted always - S(exit) = A_YES, + S(exit) = A_YES | A_SAMPLE_MEM, S(read) = A_YES, S(write) = A_YES, S(close) = A_YES, @@ -249,13 +276,15 @@ static unsigned char syscall_action[NUM_ACTIONS] = { S(fcntl) = A_YES, S(fcntl64) = A_YES, S(mmap) = A_YES, + S(mmap2) = 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, + S(set_tid_address) = A_YES, + S(exit_group) = A_YES | A_SAMPLE_MEM, // Syscalls permitted only in liberal mode S(time) = A_YES | A_LIBERAL, @@ -276,7 +305,7 @@ static unsigned char syscall_action[NUM_ACTIONS] = { 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(sigreturn) = A_YES | A_LIBERAL | A_NO_RETVAL, S(mprotect) = A_YES | A_LIBERAL, S(sigprocmask) = A_YES | A_LIBERAL, S(getdents) = A_YES | A_LIBERAL, @@ -287,14 +316,13 @@ static unsigned char syscall_action[NUM_ACTIONS] = { 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_sigreturn) = A_YES | A_LIBERAL | A_NO_RETVAL, 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 }; @@ -568,7 +596,7 @@ valid_filename(unsigned long addr) static int mem_fd; if (!file_access) - die("File access forbidden"); + err("FA: File access forbidden"); if (file_access >= 9) return; @@ -589,14 +617,14 @@ valid_filename(unsigned long addr) if (l > remains) l = remains; if (!l) - die("Access to file with name too long"); - if (long_seek(mem_fd, addr, SEEK_SET) < 0) - die("long_seek(mem): %m"); + err("FA: Access to file with name too long"); + if (lseek64(mem_fd, addr, SEEK_SET) < 0) + die("lseek64(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"); + err("FA: Access to file with name out of memory"); end += l; addr += l; } @@ -626,31 +654,31 @@ valid_filename(unsigned long addr) act = match_path_rule(&default_path_rules[i], namebuf); if (act != A_YES) - die("Forbidden access to file `%s'", namebuf); + err("FA: Forbidden access to file `%s'", namebuf); } +// Check syscall. If invalid, return -1, otherwise return the action mask. static int valid_syscall(struct user *u) { unsigned int sys = u->regs.orig_eax; - enum action act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT; + unsigned int act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT; if (act & A_LIBERAL) { - if (filter_syscalls == 1) - act &= ~A_LIBERAL; - else + if (filter_syscalls != 1) act = A_DEFAULT; } - switch (act) + + switch (act & A_ACTION_MASK) { case A_YES: - return 1; + return act; case A_NO: - return 0; + return -1; case A_FILENAME: valid_filename(u->regs.ebx); - return 1; + return act; default: ; } @@ -658,14 +686,20 @@ valid_syscall(struct user *u) { case __NR_kill: if (u->regs.ebx == box_pid) - die("Committed suicide by signal %d", (int)u->regs.ecx); - return 0; + { + meta_printf("exitsig:%d\n", (int)u->regs.ecx); + err("SG: Committed suicide by signal %d", (int)u->regs.ecx); + } + return -1; 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; + { + meta_printf("exitsig:%d\n", (int)u->regs.edx); + err("SG: Committed suicide by signal %d", (int)u->regs.edx); + } + return -1; default: - return 0; + return -1; } } @@ -681,7 +715,29 @@ static void signal_int(int unused UNUSED) { /* Interrupts are fatal, so no synchronization requirements. */ - die("Interrupted"); + meta_printf("exitsig:%d\n", SIGINT); + err("SG: Interrupted"); +} + +#define PROC_BUF_SIZE 4096 +static void +read_proc_file(char *buf, char *name, int *fdp) +{ + int c; + + if (!*fdp) + { + sprintf(buf, "/proc/%d/%s", (int) box_pid, name); + *fdp = open(buf, O_RDONLY); + if (*fdp < 0) + die("open(%s): %m", buf); + } + lseek(*fdp, 0, SEEK_SET); + if ((c = read(*fdp, buf, PROC_BUF_SIZE-1)) < 0) + die("read on /proc/$pid/%s: %m", name); + if (c >= PROC_BUF_SIZE-1) + die("/proc/$pid/%s too long", name); + buf[c] = 0; } static void @@ -695,69 +751,106 @@ check_timeout(void) 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)"); + err("TO: 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, ms; - static int proc_status_fd; - if (!proc_status_fd) - { - sprintf(buf, "/proc/%d/stat", (int) box_pid); - proc_status_fd = open(buf, O_RDONLY); - if (proc_status_fd < 0) - die("open(%s): %m", buf); - } - lseek(proc_status_fd, 0, SEEK_SET); - if ((c = read(proc_status_fd, buf, sizeof(buf)-1)) < 0) - die("read on /proc/$pid/stat: %m"); - if (c >= (int) sizeof(buf) - 1) - die("/proc/$pid/stat too long"); - buf[c] = 0; + char buf[PROC_BUF_SIZE], *x; + int utime, stime, ms; + static int proc_stat_fd; + read_proc_file(buf, "stat", &proc_stat_fd); x = buf; while (*x && *x != ' ') x++; while (*x == ' ') x++; if (*x++ != '(') - die("proc syntax error 1"); + die("proc stat syntax error 1"); while (*x && (*x != ')' || x[1] != ' ')) x++; while (*x == ')' || *x == ' ') 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"); + die("proc stat syntax error 2"); 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 (ms > timeout && ms > extra_timeout) + err("TO: Time limit exceeded"); + } +} + +static void +sample_mem_peak(void) +{ + /* + * We want to find out the peak memory usage of the process, which is + * maintained by the kernel, but unforunately it gets lost when the + * process exits (it is not reported in struct rusage). Therefore we + * have to sample it whenever we suspect that the process is about + * to exit. + */ + char buf[PROC_BUF_SIZE], *x; + static int proc_status_fd; + read_proc_file(buf, "status", &proc_status_fd); + + x = buf; + while (*x) + { + char *key = x; + while (*x && *x != ':' && *x != '\n') + x++; + if (!*x || *x == '\n') + break; + *x++ = 0; + while (*x == ' ' || *x == '\t') + x++; + + char *val = x; + while (*x && *x != '\n') + x++; + if (!*x) + break; + *x++ = 0; + + if (!strcmp(key, "VmPeak")) + { + int peak = atoi(val); + if (peak > mem_peak_kb) + mem_peak_kb = peak; + } } + + if (verbose > 1) + msg("[mem-peak: %u KB]\n", mem_peak_kb); } static void boxkeeper(void) { - int syscall_count = 0; + int syscall_count = (filter_syscalls ? 0 : 1); struct sigaction sa; is_ptraced = 1; + bzero(&sa, sizeof(sa)); sa.sa_handler = signal_int; sigaction(SIGINT, &sa, NULL); + gettimeofday(&start_time, NULL); ticks_per_sec = sysconf(_SC_CLK_TCK); if (ticks_per_sec <= 0) die("Invalid ticks_per_sec!"); + if (timeout || wall_timeout) { sa.sa_handler = signal_alarm; sigaction(SIGALRM, &sa, NULL); alarm(1); } + for(;;) { struct rusage rus; @@ -781,49 +874,79 @@ boxkeeper(void) { box_pid = 0; final_stats(&rus); - // FIXME: If the process has exited before being ptraced, signal an internal error if (WEXITSTATUS(stat)) - die("Exited with error status %d", WEXITSTATUS(stat)); + { + if (syscall_count) + { + meta_printf("exitcode:%d\n", WEXITSTATUS(stat)); + err("RE: Exited with error status %d", WEXITSTATUS(stat)); + } + else + { + // Internal error happened inside the child process and it has been already reported. + box_exit(2); + } + } if (timeout && total_ms > timeout) - die("Time limit exceeded"); + err("TO: Time limit exceeded"); if (wall_timeout && wall_ms > wall_timeout) - die("Time limit exceeded (wall clock)"); + err("TO: Time limit exceeded (wall clock)"); flush_line(); - fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n", + fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d MB, %d syscalls)\n", total_ms/1000, total_ms%1000, wall_ms/1000, wall_ms%1000, + (mem_peak_kb + 1023) / 1024, syscall_count); box_exit(0); } if (WIFSIGNALED(stat)) { box_pid = 0; - die("Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup")); + meta_printf("exitsig:%d\n", WTERMSIG(stat)); + final_stats(&rus); + err("SG: Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup")); } if (WIFSTOPPED(stat)) { int sig = WSTOPSIG(stat); if (sig == SIGTRAP) { + if (verbose > 2) + msg("[ptrace status %08x] ", stat); + static int stop_count; + if (!stop_count++) /* Traceme request */ + msg(">> Traceme request caught\n"); + else + err("SG: Breakpoint"); + ptrace(PTRACE_SYSCALL, box_pid, 0, 0); + } + else if (sig == (SIGTRAP | 0x80)) + { + if (verbose > 2) + msg("[ptrace status %08x] ", stat); struct user u; - static int stop_count = -1; + static unsigned int sys_tick, last_sys, last_act; if (ptrace(PTRACE_GETREGS, box_pid, NULL, &u) < 0) die("ptrace(PTRACE_GETREGS): %m"); - stop_count++; - if (!stop_count) /* Traceme request */ - msg(">> Traceme request caught\n"); - else if (stop_count & 1) /* Syscall entry */ + unsigned int sys = u.regs.orig_eax; + if (++sys_tick & 1) /* Syscall entry */ { 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); + int act; + msg(">> Syscall %-12s (%08lx,%08lx,%08lx) ", syscall_name(sys, namebuf), u.regs.ebx, u.regs.ecx, u.regs.edx); if (!exec_seen) { msg("[master] "); - if (u.regs.orig_eax == __NR_execve) + if (sys == __NR_execve) exec_seen = 1; } - else if (valid_syscall(&u)) - syscall_count++; + else if ((act = valid_syscall(&u)) >= 0) + { + last_act = act; + syscall_count++; + if (act & A_SAMPLE_MEM) + sample_mem_peak(); + } else { /* @@ -831,24 +954,51 @@ boxkeeper(void) * so we have to change it to something harmless (e.g., an undefined * syscall) and make the program continue. */ - unsigned int sys = u.regs.orig_eax; u.regs.orig_eax = 0xffffffff; if (ptrace(PTRACE_SETREGS, box_pid, NULL, &u) < 0) die("ptrace(PTRACE_SETREGS): %m"); - die("Forbidden syscall %s", syscall_name(sys, namebuf)); + err("FO: Forbidden syscall %s", syscall_name(sys, namebuf)); } + last_sys = sys; } else /* Syscall return */ - msg("= %ld\n", u.regs.eax); + { + if (sys == 0xffffffff) + { + /* Some syscalls (sigreturn et al.) do not return a value */ + if (!(last_act & A_NO_RETVAL)) + err("XX: Syscall does not return, but it should"); + } + else + { + if (sys != last_sys) + err("XX: Mismatched syscall entry/exit"); + } + if (last_act & A_NO_RETVAL) + msg("= ?\n"); + else + msg("= %ld\n", u.regs.eax); + } ptrace(PTRACE_SYSCALL, box_pid, 0, 0); } - else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ) + else if (sig == SIGSTOP) + { + msg(">> SIGSTOP\n"); + if (ptrace(PTRACE_SETOPTIONS, box_pid, NULL, (void *) PTRACE_O_TRACESYSGOOD) < 0) + die("ptrace(PTRACE_SETOPTIONS): %m"); + ptrace(PTRACE_SYSCALL, box_pid, 0, 0); + } + else if (sig != SIGXCPU && sig != SIGXFSZ) { msg(">> Signal %d\n", sig); + sample_mem_peak(); /* Signal might be fatal, so update mem-peak */ ptrace(PTRACE_SYSCALL, box_pid, 0, sig); } else - die("Received signal %d", sig); + { + meta_printf("exitsig:%d", sig); + err("SG: Received signal %d", sig); + } } else die("wait4: unknown status %x, giving up!", stat); @@ -886,24 +1036,31 @@ box_inside(int argc, char **argv) else dup2(1, 2); setpgrp(); + if (memory_limit) { rl.rlim_cur = rl.rlim_max = memory_limit * 1024; if (setrlimit(RLIMIT_AS, &rl) < 0) - die("setrlimit: %m"); + die("setrlimit(RLIMIT_AS): %m"); } + + rl.rlim_cur = rl.rlim_max = (stack_limit ? (rlim_t)stack_limit * 1024 : RLIM_INFINITY); + if (setrlimit(RLIMIT_STACK, &rl) < 0) + die("setrlimit(RLIMIT_STACK): %m"); + rl.rlim_cur = rl.rlim_max = 64; if (setrlimit(RLIMIT_NOFILE, &rl) < 0) - die("setrlimit: %m"); + die("setrlimit(RLIMIT_NOFILE): %m"); + + char **env = setup_environment(); 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); + raise(SIGSTOP); } - execve(args[0], args, setup_environment()); + execve(args[0], args, env); die("execve(\"%s\"): %m", args[0]); } @@ -922,6 +1079,7 @@ Options:\n\ -E =\tSet the environment variable to ; unset it if is empty\n\ -f\t\tFilter system calls (-ff=very restricted)\n\ -i \tRedirect stdin from \n\ +-k \tLimit stack size to KB (default: 0=unlimited)\n\ -m \tLimit address space to KB\n\ -M \tOutput process information to (name:value)\n\ -o \tRedirect stdout to \n\ @@ -934,8 +1092,10 @@ Options:\n\ -T\t\tAllow syscalls for measuring run time\n\ -v\t\tBe verbose (use multiple times for even more verbosity)\n\ -w