/*
- * A Simple Testing Sandbox
+ * A Simple Sandbox for MO-Eval
*
- * (c) 2001--2007 Martin Mares <mj@ucw.cz>
+ * (c) 2001--2008 Martin Mares <mj@ucw.cz>
*/
#define _LARGEFILE64_SOURCE
#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; /* milliseconds */
static int file_access;
static int verbose;
static int memory_limit;
-static int allow_times;
static char *redir_stdin, *redir_stdout;
static char *set_cwd;
va_end(args);
}
-struct syscall {
- const char *name;
- int override;
-};
+static void *
+xmalloc(size_t size)
+{
+ void *p = malloc(size);
+ if (!p)
+ die("Out of memory");
+ return p;
+}
+
+/*** Syscall rules ***/
-static struct syscall syscall_tab[] = {
+static const char * const syscall_names[] = {
#include "syscall-table.h"
};
-#define NUM_SYSCALLS (sizeof(syscall_tab)/sizeof(syscall_tab[0]))
+#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_tab[id].name)
- return syscall_tab[id].name;
+ if (id < NUM_SYSCALLS && syscall_names[id])
+ return syscall_names[id];
else
{
sprintf(buf, "#%d", id);
}
}
-#if 0
-static struct syscall *
+static int
syscall_by_name(char *name)
{
- for (unsigned int i=0; i<sizeof(syscall_tab)/sizeof(syscall_tab[0]); i++)
- if (!strcmp(syscall_tab[i].name, name))
- return &syscall_tab[i];
- return NULL;
+ for (unsigned int i=0; i<NUM_SYSCALLS; i++)
+ if (syscall_names[i] && !strcmp(syscall_names[i], name))
+ return i;
+ if (name[0] == '#')
+ name++;
+ if (!*name)
+ return -1;
+ char *ep;
+ unsigned long l = strtoul(name, &ep, 0);
+ if (*ep)
+ return -1;
+ if (l >= 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;
+}
+
+/*** Path rules ***/
+
+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) + 1);
+ 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;
+}
+
+/*** Environment rules ***/
+
+struct env_rule {
+ char *var; // Variable to match
+ char *val; // ""=clear, NULL=inherit
+ int var_len;
+ struct env_rule *next;
+};
+
+static struct env_rule *first_env_rule;
+static struct env_rule **last_env_rule = &first_env_rule;
+
+static struct env_rule default_env_rules[] = {
+ { "LIBC_FATAL_STDERR_", "1" }
+};
+
+static int
+set_env_action(char *a0)
+{
+ struct env_rule *r = xmalloc(sizeof(*r) + strlen(a0) + 1);
+ char *a = (char *)(r+1);
+ strcpy(a, a0);
+
+ char *sep = strchr(a, '=');
+ if (sep == a)
+ return 0;
+ r->var = a;
+ if (sep)
+ {
+ *sep++ = 0;
+ r->val = sep;
+ }
+ else
+ r->val = NULL;
+ *last_env_rule = r;
+ last_env_rule = &r->next;
+ r->next = NULL;
+ return 1;
+}
+
+static int
+match_env_var(char *env_entry, struct env_rule *r)
+{
+ if (strncmp(env_entry, r->var, r->var_len))
+ return 0;
+ return (env_entry[r->var_len] == '=');
}
-#endif
+
+static void
+apply_env_rule(char **env, int *env_sizep, struct env_rule *r)
+{
+ // First remove the variable if already set
+ int pos = 0;
+ while (pos < *env_sizep && !match_env_var(env[pos], r))
+ pos++;
+ if (pos < *env_sizep)
+ {
+ (*env_sizep)--;
+ env[pos] = env[*env_sizep];
+ env[*env_sizep] = NULL;
+ }
+
+ // What is the new value?
+ char *new;
+ if (r->val)
+ {
+ if (!r->val[0])
+ return;
+ new = xmalloc(r->var_len + 1 + strlen(r->val) + 1);
+ sprintf(new, "%s=%s", r->var, r->val);
+ }
+ else
+ {
+ pos = 0;
+ while (environ[pos] && !match_env_var(environ[pos], r))
+ pos++;
+ if (!(new = environ[pos]))
+ return;
+ }
+
+ // Add it at the end of the array
+ env[(*env_sizep)++] = new;
+ env[*env_sizep] = NULL;
+}
+
+static char **
+setup_environment(void)
+{
+ // Link built-in rules with user rules
+ for (int i=ARRAY_SIZE(default_env_rules)-1; i >= 0; i--)
+ {
+ default_env_rules[i].next = first_env_rule;
+ first_env_rule = &default_env_rules[i];
+ }
+
+ // Scan the original environment
+ char **orig_env = environ;
+ int orig_size = 0;
+ while (orig_env[orig_size])
+ orig_size++;
+
+ // For each rule, reserve one more slot and calculate length
+ int num_rules = 0;
+ for (struct env_rule *r = first_env_rule; r; r=r->next)
+ {
+ num_rules++;
+ r->var_len = strlen(r->var);
+ }
+
+ // Create a new environment
+ char **env = xmalloc((orig_size + num_rules + 1) * sizeof(char *));
+ int size;
+ if (pass_environ)
+ {
+ memcpy(env, environ, orig_size * sizeof(char *));
+ size = orig_size;
+ }
+ else
+ size = 0;
+ env[size] = NULL;
+
+ // Apply the rules one by one
+ for (struct env_rule *r = first_env_rule; r; r=r->next)
+ apply_env_rule(env, &size, r);
+
+ // Return the new env and pass some gossip
+ if (verbose > 1)
+ {
+ fprintf(stderr, "Passing environment:\n");
+ for (int i=0; env[i]; i++)
+ fprintf(stderr, "\t%s\n", env[i]);
+ }
+ return env;
+}
+
+/*** Syscall checks ***/
static void
valid_filename(unsigned long addr)
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) ||
- !strncmp(namebuf, "/opt/lib/", 9))
- && !strstr(namebuf, ".."))
- return;
- if (!strcmp(namebuf, "/dev/null") ||
- !strcmp(namebuf, "/dev/zero") ||
- !strcmp(namebuf, "/proc/meminfo") ||
- !strcmp(namebuf, "/proc/self/stat") ||
- !strcmp(namebuf, "/proc/self/exe") || /* Needed by FPC 2.0.x runtime */
- !strncmp(namebuf, "/usr/share/zoneinfo/", 20))
- return;
- }
- die("Forbidden access to file `%s'", namebuf);
+ for (int i=0; i<ARRAY_SIZE(default_path_rules) && !act; i++)
+ act = match_path_rule(&default_path_rules[i], namebuf);
+
+ if (act != A_YES)
+ die("Forbidden access to file `%s'", namebuf);
}
static int
valid_syscall(struct user *u)
{
- switch (u->regs.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 __NR_open:
- case __NR_creat:
- case __NR_unlink:
- case __NR_oldstat:
- case __NR_access:
- case __NR_oldlstat:
- case __NR_truncate:
- case __NR_stat:
- case __NR_lstat:
- case __NR_truncate64:
- case __NR_stat64:
- case __NR_lstat64:
- case __NR_readlink:
- valid_filename(u->regs.ebx);
+ if (filter_syscalls == 1)
+ act &= ~A_LIBERAL;
+ else
+ act = A_DEFAULT;
+ }
+ switch (act)
+ {
+ case A_YES:
return 1;
- case __NR_exit:
- case __NR_read:
- case __NR_write:
- case __NR_close:
- case __NR_lseek:
- case __NR_getpid:
- case __NR_getuid:
- case __NR_oldfstat:
- case __NR_dup:
- case __NR_brk:
- case __NR_getgid:
- case __NR_geteuid:
- case __NR_getegid:
- case __NR_dup2:
- case __NR_ftruncate:
- case __NR_fstat:
- case __NR_personality:
- case __NR__llseek:
- case __NR_readv:
- case __NR_writev:
- case __NR_getresuid:
-#ifdef __NR_pread64
- case __NR_pread64:
- case __NR_pwrite64:
-#else
- case __NR_pread:
- case __NR_pwrite:
-#endif
- case __NR_ftruncate64:
- case __NR_fstat64:
- case __NR_fcntl:
- case __NR_fcntl64:
- case __NR_mmap:
- case __NR_munmap:
- case __NR_ioctl:
- case __NR_uname:
- case __NR_gettid:
- case __NR_set_thread_area:
- case __NR_get_thread_area:
- case __NR_exit_group:
+ case A_NO:
+ return 0;
+ case A_FILENAME:
+ valid_filename(u->regs.ebx);
return 1;
- case __NR_time:
- case __NR_alarm:
- case __NR_pause:
- case __NR_signal:
- case __NR_fchmod:
- case __NR_sigaction:
- case __NR_sgetmask:
- case __NR_ssetmask:
- case __NR_sigsuspend:
- case __NR_sigpending:
- case __NR_getrlimit:
- case __NR_getrusage:
- case __NR_gettimeofday:
- case __NR_select:
- case __NR_readdir:
- case __NR_setitimer:
- case __NR_getitimer:
- case __NR_sigreturn:
- case __NR_mprotect:
- case __NR_sigprocmask:
- case __NR_getdents:
- case __NR_getdents64:
- case __NR__newselect:
- case __NR_fdatasync:
- case __NR_mremap:
- case __NR_poll:
- case __NR_getcwd:
- case __NR_nanosleep:
- case __NR_rt_sigreturn:
- case __NR_rt_sigaction:
- case __NR_rt_sigprocmask:
- case __NR_rt_sigpending:
- case __NR_rt_sigtimedwait:
- case __NR_rt_sigqueueinfo:
- case __NR_rt_sigsuspend:
- case __NR_mmap2:
- case __NR__sysctl:
- return (filter_syscalls == 1);
- case __NR_times:
- return allow_times;
+ default: ;
+ }
+
+ switch (sys)
+ {
case __NR_kill:
if (u->regs.ebx == box_pid)
die("Committed suicide by signal %d", (int)u->regs.ecx);
{
struct rlimit rl;
char *args[argc+1];
- char *env[] = { "LIBC_FATAL_STDERR_=1", NULL };
memcpy(args, argv, argc * sizeof(char *));
args[argc] = NULL;
signal(SIGCHLD, SIG_IGN);
raise(SIGCHLD);
}
- execve(args[0], args, (pass_environ ? environ : env));
+ execve(args[0], args, setup_environment());
die("execve(\"%s\"): %m", args[0]);
}
Options:\n\
-a <level>\tSet file access level (0=none, 1=cwd, 2=/etc,/lib,..., 3=whole fs, 9=no checks; needs -f)\n\
-c <dir>\tChange directory to <dir> first\n\
--e\t\tPass full environment of parent process\n\
+-e\t\tInherit full environment of the parent process\n\
+-E <var>\tInherit the environment variable <var> from the parent process\n\
+-E <var>=<val>\tSet the environment variable <var> to <val>; unset it if <var> is empty\n\
-f\t\tFilter system calls (-ff=very restricted)\n\
-i <file>\tRedirect stdin from <file>\n\
-m <size>\tLimit address space to <size> KB\n\
-o <file>\tRedirect stdout to <file>\n\
+-p <path>\tPermit access to the specified path (or subtree if it ends with a `/')\n\
+-p <path>=<act>\tDefine action for the specified path (<act>=yes/no)\n\
+-s <sys>\tPermit the specified syscall (be careful)\n\
+-s <sys>=<act>\tDefine action for the specified syscall (<act>=yes/no/file)\n\
-t <time>\tSet run time limit (seconds, fractions allowed)\n\
-T\t\tAllow syscalls for measuring run time\n\
--v\t\tBe verbose\n\
+-v\t\tBe verbose (use multiple times for even more verbosity)\n\
-w <time>\tSet wall clock time limit (seconds, fractions allowed)\n\
");
exit(1);
int c;
uid_t uid;
- while ((c = getopt(argc, argv, "a:c:efi:m:o:t:Tvw:")) >= 0)
+ while ((c = getopt(argc, argv, "a:c:eE:fi:m:o:p:s:t:Tvw:")) >= 0)
switch (c)
{
case 'a':
case 'e':
pass_environ = 1;
break;
+ case 'E':
+ if (!set_env_action(optarg))
+ usage();
+ break;
case 'f':
filter_syscalls++;
break;
case 'o':
redir_stdout = optarg;
break;
+ case 'p':
+ if (!set_path_action(optarg))
+ usage();
+ break;
+ case 's':
+ if (!set_syscall_action(optarg))
+ usage();
+ break;
case 't':
timeout = 1000*atof(optarg);
break;
case 'T':
- allow_times++;
+ syscall_action[__NR_times] = A_YES;
break;
case 'v':
verbose++;