static int exec_seen;
static int partial_line;
+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 ***/
va_end(args);
}
-static int total_ms, wall_ms;
-
static void
final_stats(struct rusage *rus)
{
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 ***/
{
if (box_pid > 0)
{
+ sample_mem_peak();
if (is_ptraced)
ptrace(PTRACE_KILL, box_pid);
kill(-box_pid, SIGKILL);
partial_line = 0;
}
+/* Report an error of the sandbox itself */
static void NONRET __attribute__((format(printf,1,2)))
die(char *msg, ...)
{
box_exit(2);
}
+/* Report an error of the program inside the sandbox */
static void NONRET __attribute__((format(printf,1,2)))
err(char *msg, ...)
{
box_exit(1);
}
+/* Write a message, but only if in verbose mode */
static void __attribute__((format(printf,1,2)))
msg(char *msg, ...)
{
A_NO, // Always forbid
A_YES, // Always permit
A_FILENAME, // Permit if arg1 is a known filename
+ A_ACTION_MASK = 15,
+ 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] = {
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,
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,
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
};
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: ;
}
meta_printf("exitsig:%d\n", (int)u->regs.ecx);
err("SG: Committed suicide by signal %d", (int)u->regs.ecx);
}
- return 0;
+ return -1;
case __NR_tgkill:
if (u->regs.ebx == box_pid && u->regs.ecx == box_pid)
{
meta_printf("exitsig:%d\n", (int)u->regs.edx);
err("SG: Committed suicide by signal %d", (int)u->regs.edx);
}
- return 0;
+ return -1;
default:
- return 0;
+ return -1;
}
}
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
check_timeout(void)
{
}
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);
}
}
+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)
{
if (wall_timeout && wall_ms > wall_timeout)
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);
}
{
box_pid = 0;
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))
static int stop_count = -1;
if (ptrace(PTRACE_GETREGS, box_pid, NULL, &u) < 0)
die("ptrace(PTRACE_GETREGS): %m");
+ if (u.regs.orig_eax < 0) /* Process issued a breakpoint instruction */
+ err("SG: Breakpoint");
stop_count++;
if (!stop_count) /* Traceme request */
msg(">> Traceme request caught\n");
else if (stop_count & 1) /* Syscall entry */
{
char namebuf[32];
+ int act;
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)
{
if (u.regs.orig_eax == __NR_execve)
exec_seen = 1;
}
- else if (valid_syscall(&u))
- syscall_count++;
+ else if ((act = valid_syscall(&u)) >= 0)
+ {
+ syscall_count++;
+ if (act & A_SAMPLE_MEM)
+ sample_mem_peak();
+ }
else
{
/*
else if (sig != SIGSTOP && 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
projects / moe.git / blobdiff