#include <sched.h>
#include <time.h>
#include <grp.h>
+#include <mntent.h>
+#include <limits.h>
#include <sys/wait.h>
#include <sys/time.h>
#include <sys/signal.h>
#include <sys/resource.h>
#include <sys/mount.h>
#include <sys/stat.h>
+#include <sys/quota.h>
+#include <sys/vfs.h>
#define NONRET __attribute__((noreturn))
#define UNUSED __attribute__((unused))
#define ARRAY_SIZE(a) (int)(sizeof(a)/sizeof(a[0]))
-#define BOX_DIR CONFIG_ISOLATE_BOX_DIR
-#define BOX_UID CONFIG_ISOLATE_BOX_UID
-#define BOX_GID CONFIG_ISOLATE_BOX_GID
-
static int timeout; /* milliseconds */
static int wall_timeout;
static int extra_timeout;
static int verbose;
static int memory_limit;
static int stack_limit;
+static int block_quota;
+static int inode_quota;
static int max_processes = 1;
static char *redir_stdin, *redir_stdout, *redir_stderr;
static int cg_enable;
static int cg_memory_limit;
static int cg_timing;
-static char *cg_root = "/sys/fs/cgroup";
+static int box_id;
+static char box_dir[1024];
+static pid_t box_pid;
+
+static uid_t box_uid;
+static gid_t box_gid;
static uid_t orig_uid;
static gid_t orig_gid;
-static pid_t box_pid;
static int partial_line;
static char cleanup_cmd[256];
static void die(char *msg, ...) NONRET;
static void cg_stats(void);
static int get_wall_time_ms(void);
-static int get_run_time_ms(void);
+static int get_run_time_ms(struct rusage *rus);
/*** Meta-files ***/
static void
final_stats(struct rusage *rus)
{
- total_ms = get_run_time_ms();
+ total_ms = get_run_time_ms(rus);
wall_ms = get_wall_time_ms();
meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000);
va_end(args);
}
+/*** Utility functions ***/
+
static void *
xmalloc(size_t size)
{
return p;
}
+static char *
+xstrdup(char *str)
+{
+ char *p = strdup(str);
+ if (!p)
+ die("Out of memory");
+ return p;
+}
+
+static int dir_exists(char *path)
+{
+ struct stat st;
+ return (stat(path, &st) >= 0 && S_ISDIR(st.st_mode));
+}
+
/*** Environment rules ***/
struct env_rule {
return env;
}
+/*** Directory rules ***/
+
+struct dir_rule {
+ char *inside; // A relative path
+ char *outside; // This can be an absolute path or a relative path starting with "./"
+ unsigned int flags; // DIR_FLAG_xxx
+ struct dir_rule *next;
+};
+
+enum dir_rule_flags {
+ DIR_FLAG_RW = 1,
+ DIR_FLAG_NOEXEC = 2,
+ DIR_FLAG_FS = 4,
+ DIR_FLAG_MAYBE = 8,
+ DIR_FLAG_DEV = 16,
+};
+
+static const char * const dir_flag_names[] = { "rw", "noexec", "fs", "maybe", "dev" };
+
+static struct dir_rule *first_dir_rule;
+static struct dir_rule **last_dir_rule = &first_dir_rule;
+
+static int add_dir_rule(char *in, char *out, unsigned int flags)
+{
+ // Make sure that "in" is relative
+ while (in[0] == '/')
+ in++;
+ if (!*in)
+ return 0;
+
+ // Check "out"
+ if (flags & DIR_FLAG_FS)
+ {
+ if (!out || out[0] == '/')
+ return 0;
+ }
+ else
+ {
+ if (out && out[0] != '/' && strncmp(out, "./", 2))
+ return 0;
+ }
+
+ // Override an existing rule
+ struct dir_rule *r;
+ for (r = first_dir_rule; r; r = r->next)
+ if (!strcmp(r->inside, in))
+ break;
+
+ // Add a new rule
+ if (!r)
+ {
+ r = xmalloc(sizeof(*r));
+ r->inside = in;
+ *last_dir_rule = r;
+ last_dir_rule = &r->next;
+ r->next = NULL;
+ }
+ r->outside = out;
+ r->flags = flags;
+ return 1;
+}
+
+static unsigned int parse_dir_option(char *opt)
+{
+ for (unsigned int i = 0; i < ARRAY_SIZE(dir_flag_names); i++)
+ if (!strcmp(opt, dir_flag_names[i]))
+ return 1U << i;
+ die("Unknown directory option %s", opt);
+}
+
+static int set_dir_action(char *arg)
+{
+ arg = xstrdup(arg);
+
+ char *colon = strchr(arg, ':');
+ unsigned int flags = 0;
+ while (colon)
+ {
+ *colon++ = 0;
+ char *next = strchr(colon, ':');
+ if (next)
+ *next = 0;
+ flags |= parse_dir_option(colon);
+ colon = next;
+ }
+
+ char *eq = strchr(arg, '=');
+ if (eq)
+ {
+ *eq++ = 0;
+ return add_dir_rule(arg, (*eq ? eq : NULL), flags);
+ }
+ else
+ {
+ char *out = xmalloc(1 + strlen(arg) + 1);
+ sprintf(out, "/%s", arg);
+ return add_dir_rule(arg, out, flags);
+ }
+}
+
+static void init_dir_rules(void)
+{
+ set_dir_action("box=./box:rw");
+ set_dir_action("bin");
+ set_dir_action("dev:dev");
+ set_dir_action("lib");
+ set_dir_action("lib64:maybe");
+ set_dir_action("proc=proc:fs");
+ set_dir_action("usr");
+}
+
+static void make_dir(char *path)
+{
+ char *sep = (path[0] == '/' ? path+1 : path);
+
+ for (;;)
+ {
+ sep = strchr(sep, '/');
+ if (sep)
+ *sep = 0;
+
+ if (!dir_exists(path) && mkdir(path, 0777) < 0)
+ die("Cannot create directory %s: %m\n", path);
+
+ if (!sep)
+ return;
+ *sep++ = '/';
+ }
+}
+
+static void apply_dir_rules(void)
+{
+ for (struct dir_rule *r = first_dir_rule; r; r=r->next)
+ {
+ char *in = r->inside;
+ char *out = r->outside;
+ if (!out)
+ {
+ msg("Not binding anything on %s\n", r->inside);
+ continue;
+ }
+
+ if ((r->flags & DIR_FLAG_MAYBE) && !dir_exists(out))
+ {
+ msg("Not binding %s on %s (does not exist)\n", out, r->inside);
+ continue;
+ }
+
+ char root_in[1024];
+ snprintf(root_in, sizeof(root_in), "root/%s", in);
+ make_dir(root_in);
+
+ unsigned long mount_flags = 0;
+ if (!(r->flags & DIR_FLAG_RW))
+ mount_flags |= MS_RDONLY;
+ if (r->flags & DIR_FLAG_NOEXEC)
+ mount_flags |= MS_NOEXEC;
+ if (!(r->flags & DIR_FLAG_DEV))
+ mount_flags |= MS_NODEV;
+
+ if (r->flags & DIR_FLAG_FS)
+ {
+ msg("Mounting %s on %s (flags %lx)\n", out, in, mount_flags);
+ if (mount("none", root_in, out, mount_flags, "") < 0)
+ die("Cannot mount %s on %s: %m", out, in);
+ }
+ else
+ {
+ mount_flags |= MS_BIND | MS_NOSUID;
+ msg("Binding %s on %s (flags %lx)\n", out, in, mount_flags);
+ // Most mount flags need remount to work
+ if (mount(out, root_in, "none", mount_flags, "") < 0 ||
+ mount(out, root_in, "none", MS_REMOUNT | mount_flags, "") < 0)
+ die("Cannot mount %s on %s: %m", out, in);
+ }
+ }
+}
+
/*** Control groups ***/
static char cg_path[256];
if (!cg_enable)
return;
- struct stat st;
- if (stat(cg_root, &st) < 0 || !S_ISDIR(st.st_mode))
+ char *cg_root = CONFIG_ISOLATE_CGROUP_ROOT;
+ if (!dir_exists(cg_root))
die("Control group filesystem at %s not mounted", cg_root);
- snprintf(cg_path, sizeof(cg_path), "%s/box-%d", cg_root, BOX_UID);
+ snprintf(cg_path, sizeof(cg_path), "%s/box-%d", cg_root, box_id);
msg("Using control group %s\n", cg_path);
}
die("Cannot remove control group %s: %m", cg_path);
}
+/*** Disk quotas ***/
+
+static int
+path_begins_with(char *path, char *with)
+{
+ while (*with)
+ if (*path++ != *with++)
+ return 0;
+ return (!*with || *with == '/');
+}
+
+static char *
+find_device(char *path)
+{
+ FILE *f = setmntent("/proc/mounts", "r");
+ if (!f)
+ die("Cannot open /proc/mounts: %m");
+
+ struct mntent *me;
+ int best_len = 0;
+ char *best_dev = NULL;
+ while (me = getmntent(f))
+ {
+ if (!path_begins_with(me->mnt_fsname, "/dev"))
+ continue;
+ if (path_begins_with(path, me->mnt_dir))
+ {
+ int len = strlen(me->mnt_dir);
+ if (len > best_len)
+ {
+ best_len = len;
+ free(best_dev);
+ best_dev = xstrdup(me->mnt_fsname);
+ }
+ }
+ }
+ endmntent(f);
+ return best_dev;
+}
+
+static void
+set_quota(void)
+{
+ if (!block_quota)
+ return;
+
+ char cwd[PATH_MAX];
+ if (!getcwd(cwd, sizeof(cwd)))
+ die("getcwd: %m");
+
+ char *dev = find_device(cwd);
+ if (!dev)
+ die("Cannot identify filesystem which contains %s", cwd);
+ msg("Quota: Mapped path %s to a filesystem on %s\n", cwd, dev);
+
+ // Sanity check
+ struct stat dev_st, cwd_st;
+ if (stat(dev, &dev_st) < 0)
+ die("Cannot identify block device %s: %m", dev);
+ if (!S_ISBLK(dev_st.st_mode))
+ die("Expected that %s is a block device", dev);
+ if (stat(".", &cwd_st) < 0)
+ die("Cannot stat cwd: %m");
+ if (cwd_st.st_dev != dev_st.st_rdev)
+ die("Identified %s as a filesystem on %s, but it is obviously false", cwd, dev);
+
+ struct dqblk dq = {
+ .dqb_bhardlimit = block_quota,
+ .dqb_bsoftlimit = block_quota,
+ .dqb_ihardlimit = inode_quota,
+ .dqb_isoftlimit = inode_quota,
+ .dqb_valid = QIF_LIMITS,
+ };
+ if (quotactl(QCMD(Q_SETQUOTA, USRQUOTA), dev, box_uid, (caddr_t) &dq) < 0)
+ die("Cannot set disk quota: %m");
+ msg("Quota: Set block quota %d and inode quota %d\n", block_quota, inode_quota);
+
+ free(dev);
+}
+
/*** The keeper process ***/
static void
}
static int
-get_run_time_ms(void)
+get_run_time_ms(struct rusage *rus)
{
if (cg_timing)
return cg_get_run_time_ms();
+ if (rus)
+ {
+ struct timeval total;
+ timeradd(&rus->ru_utime, &rus->ru_stime, &total);
+ return total.tv_sec*1000 + total.tv_usec/1000;
+ }
+
char buf[PROC_BUF_SIZE], *x;
int utime, stime;
static int proc_stat_fd;
}
if (timeout)
{
- int ms = get_run_time_ms();
+ int ms = get_run_time_ms(NULL);
if (verbose > 1)
fprintf(stderr, "[time check: %d msec]\n", ms);
if (ms > timeout && ms > extra_timeout)
if (mount("none", "root", "tmpfs", 0, "mode=755") < 0)
die("Cannot mount root ramdisk: %m");
- static const char * const dirs[] = { "box", "/bin", "/lib", "/usr", "/dev" };
- for (int i=0; i < ARRAY_SIZE(dirs); i++)
- {
- const char *d = dirs[i];
- char buf[1024];
- snprintf(buf, sizeof(buf), "root/%s", (d[0] == '/' ? d+1 : d));
- msg("Binding %s on %s\n", d, buf);
- if (mkdir(buf, 0755) < 0)
- die("mkdir(%s): %m", buf);
- if (mount(d, buf, "none", MS_BIND | MS_NOSUID | MS_NODEV, "") < 0)
- die("Cannot bind %s on %s: %m", d, buf);
- }
-
- if (mkdir("root/proc", 0755) < 0)
- die("Cannot create proc: %m");
- if (mount("none", "root/proc", "proc", 0, "") < 0)
- die("Cannot mount proc: %m");
+ apply_dir_rules();
if (chroot("root") < 0)
die("Chroot failed: %m");
static void
setup_credentials(void)
{
- if (setresgid(BOX_GID, BOX_GID, BOX_GID) < 0)
+ if (setresgid(box_gid, box_gid, box_gid) < 0)
die("setresgid: %m");
if (setgroups(0, NULL) < 0)
die("setgroups: %m");
- if (setresuid(BOX_UID, BOX_UID, BOX_UID) < 0)
+ if (setresuid(box_uid, box_uid, box_uid) < 0)
die("setresuid: %m");
setpgrp();
}
die("execve(\"%s\"): %m", args[0]);
}
+static void
+box_init(void)
+{
+ if (box_id < 0 || box_id >= CONFIG_ISOLATE_NUM_BOXES)
+ die("Sandbox ID out of range (allowed: 0-%d)", CONFIG_ISOLATE_NUM_BOXES-1);
+ box_uid = CONFIG_ISOLATE_FIRST_UID + box_id;
+ box_gid = CONFIG_ISOLATE_FIRST_GID + box_id;
+
+ snprintf(box_dir, sizeof(box_dir), "%s/%d", CONFIG_ISOLATE_BOX_DIR, box_id);
+ make_dir(box_dir);
+ if (chdir(box_dir) < 0)
+ die("chdir(%s): %m", box_dir);
+}
+
/*** Commands ***/
static void
die("Cannot chown box: %m");
cg_prepare();
+ set_quota();
+
+ puts(box_dir);
}
static void
cleanup(void)
{
- struct stat st;
- if (stat("box", &st) < 0 || !S_ISDIR(st.st_mode))
+ if (!dir_exists("box"))
die("Box directory not found, there isn't anything to clean up");
msg("Deleting sandbox directory\n");
- xsystem("rm -rf box");
+ xsystem("rm -rf *");
+ if (rmdir(box_dir) < 0)
+ die("Cannot remove %s: %m", box_dir);
cg_remove();
}
static void
run(char **argv)
{
- struct stat st;
- if (stat("box", &st) < 0 || !S_ISDIR(st.st_mode))
+ if (!dir_exists("box"))
die("Box directory not found, did you run `isolate --init'?");
char cmd[256];
- snprintf(cmd, sizeof(cmd), "chown -R %d.%d box", BOX_UID, BOX_GID);
+ snprintf(cmd, sizeof(cmd), "chown -R %d.%d box", box_uid, box_gid);
xsystem(cmd);
snprintf(cleanup_cmd, sizeof(cleanup_cmd), "chown -R %d.%d box", orig_uid, orig_gid);
printf("Process isolator 1.0\n");
printf("(c) 2012 Martin Mares and Bernard Blackham\n");
printf("\nCompile-time configuration:\n");
- printf("Sandbox directory: %s\n", BOX_DIR);
- printf("Sandbox credentials: uid=%u gid=%u\n", BOX_UID, BOX_GID);
+ printf("Sandbox directory: %s\n", CONFIG_ISOLATE_BOX_DIR);
+ printf("Sandbox credentials: uid=%u-%u gid=%u-%u\n",
+ CONFIG_ISOLATE_FIRST_UID,
+ CONFIG_ISOLATE_FIRST_UID + CONFIG_ISOLATE_NUM_BOXES - 1,
+ CONFIG_ISOLATE_FIRST_GID,
+ CONFIG_ISOLATE_FIRST_GID + CONFIG_ISOLATE_NUM_BOXES - 1);
}
/*** Options ***/
Usage: isolate [<options>] <command>\n\
\n\
Options:\n\
+-b, --box-id=<id>\tWhen multiple sandboxes are used in parallel, each must get a unique ID\n\
-c, --cg[=<parent>]\tPut process in a control group (optionally a sub-group of <parent>)\n\
--cg-mem=<size>\tLimit memory usage of the control group to <size> KB\n\
--cg-timing\t\tTime limits affects total run time of the control group\n\
--E, --env=<var>\tInherit the environment variable <var> from the parent process\n\
+-d, --dir=<dir>\t\tMake a directory <dir> visible inside the sandbox\n\
+ --dir=<in>=<out>\tMake a directory <out> outside visible as <in> inside\n\
+ --dir=<in>=\t\tDelete a previously defined directory rule (even a default one)\n\
+ --dir=...:<opt>\tSpecify options for a rule:\n\
+\t\t\t\tdev\tAllow access to special files\n\
+\t\t\t\tfs\tMount a filesystem (e.g., --dir=/proc:proc:fs)\n\
+\t\t\t\tmaybe\tSkip the rule if <out> does not exist\n\
+\t\t\t\tnoexec\tDo not allow execution of binaries\n\
+\t\t\t\trw\tAllow read-write access\n\
+-E, --env=<var>\t\tInherit the environment variable <var> from the parent process\n\
-E, --env=<var>=<val>\tSet the environment variable <var> to <val>; unset it if <var> is empty\n\
-x, --extra-time=<time>\tSet extra timeout, before which a timing-out program is not yet killed,\n\
\t\t\tso that its real execution time is reported (seconds, fractions allowed)\n\
-e, --full-env\t\tInherit full environment of the parent process\n\
-m, --mem=<size>\tLimit address space to <size> KB\n\
-M, --meta=<file>\tOutput process information to <file> (name:value)\n\
+-q, --quota=<blk>,<ino>\tSet disk quota to <blk> blocks and <ino> inodes\n\
-k, --stack=<size>\tLimit stack size to <size> KB (default: 0=unlimited)\n\
-r, --stderr=<file>\tRedirect stderr to <file>\n\
-i, --stdin=<file>\tRedirect stdin from <file>\n\
OPT_CG_TIMING,
};
-static const char short_opts[] = "c::eE:i:k:m:M:o:p::r:t:vw:x:";
+static const char short_opts[] = "b:c:d:eE:i:k:m:M:o:p::q:r:t:vw:x:";
static const struct option long_opts[] = {
- { "cg", 2, NULL, 'c' },
+ { "box-id", 1, NULL, 'b' },
+ { "cg", 1, NULL, 'c' },
{ "cg-mem", 1, NULL, OPT_CG_MEM },
{ "cg-timing", 0, NULL, OPT_CG_TIMING },
{ "cleanup", 0, NULL, OPT_CLEANUP },
+ { "dir", 1, NULL, 'd' },
{ "env", 1, NULL, 'E' },
{ "extra-time", 1, NULL, 'x' },
{ "full-env", 0, NULL, 'e' },
{ "mem", 1, NULL, 'm' },
{ "meta", 1, NULL, 'M' },
{ "processes", 2, NULL, 'p' },
+ { "quota", 1, NULL, 'q' },
{ "run", 0, NULL, OPT_RUN },
{ "stack", 1, NULL, 'k' },
{ "stderr", 1, NULL, 'r' },
main(int argc, char **argv)
{
int c;
+ char *sep;
enum opt_code mode = 0;
+ init_dir_rules();
+
while ((c = getopt_long(argc, argv, short_opts, long_opts, NULL)) >= 0)
switch (c)
{
+ case 'b':
+ box_id = atoi(optarg);
+ break;
case 'c':
- if (optarg)
- cg_root = optarg;
cg_enable = 1;
break;
+ case 'd':
+ if (!set_dir_action(optarg))
+ usage();
+ break;
case 'e':
pass_environ = 1;
break;
else
max_processes = 0;
break;
+ case 'q':
+ sep = strchr(optarg, ',');
+ if (!sep)
+ usage();
+ block_quota = atoi(optarg);
+ inode_quota = atoi(sep+1);
+ break;
case 'r':
redir_stderr = optarg;
break;
orig_gid = getgid();
umask(022);
- if (chdir(BOX_DIR) < 0)
- die("chdir(%s): %m", BOX_DIR);
+ box_init();
cg_init();
switch (mode)