/*
- * A Process Isolator based in Linux Containers
+ * A Process Isolator based on Linux Containers
*
* (c) 2012 Martin Mares <mj@ucw.cz>
+ * (c) 2012 Bernard Blackham <bernard@blackham.com.au>
*/
#define _GNU_SOURCE
#include "autoconf.h"
-// FIXME: prune
#include <errno.h>
#include <stdio.h>
#include <fcntl.h>
#include <time.h>
#include <grp.h>
#include <sys/wait.h>
-#include <sys/user.h>
#include <sys/time.h>
-#include <sys/ptrace.h>
#include <sys/signal.h>
-#include <sys/sysinfo.h>
#include <sys/resource.h>
#include <sys/mount.h>
#include <sys/stat.h>
#define UNUSED __attribute__((unused))
#define ARRAY_SIZE(a) (int)(sizeof(a)/sizeof(a[0]))
-// FIXME: Make configurable, probably in compile time
-#define BOX_DIR "/tmp/box"
-#define BOX_UID 60000
-#define BOX_GID 60000
+#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 gid_t orig_gid;
static pid_t box_pid;
-static volatile sig_atomic_t timer_tick;
-static struct timeval start_time;
-static int ticks_per_sec;
static int partial_line;
static char cleanup_cmd[256];
+static struct timeval start_time;
+static int ticks_per_sec;
static int total_ms, wall_ms;
+static volatile sig_atomic_t timer_tick;
+
+static int error_pipes[2];
+static int write_errors_to_fd;
+static int read_errors_from_fd;
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_list args;
va_start(args, msg);
- flush_line();
char buf[1024];
- vsnprintf(buf, sizeof(buf), msg, args);
+ int n = vsnprintf(buf, sizeof(buf), msg, args);
+
+ if (write_errors_to_fd)
+ {
+ // We are inside the box, have to use error pipe for error reporting.
+ // We hope that the whole error message fits in PIPE_BUF bytes.
+ write(write_errors_to_fd, buf, n);
+ exit(2);
+ }
+
+ // Otherwise, we in the box keeper process, so we report errors normally
+ flush_line();
meta_printf("status:XX\nmessage:%s\n", buf);
fputs(buf, stderr);
fputc('\n', stderr);
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;
}
+/*** Mount 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,
+};
+
+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)
+ {
+ struct dir_rule *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)
+{
+ if (!strcmp(opt, "rw"))
+ return DIR_FLAG_RW;
+ if (!strcmp(opt, "noexec"))
+ return DIR_FLAG_NOEXEC;
+ if (!strcmp(opt, "fs"))
+ return DIR_FLAG_FS;
+ if (!strcmp(opt, "maybe"))
+ return DIR_FLAG_MAYBE;
+ 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)
+ {
+ char *opt = colon + 1;
+ char *next = strchr(opt, ':');
+ if (next)
+ *next = 0;
+ flags |= parse_dir_option(opt);
+ 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");
+ 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;
+ 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_FS)
+ {
+ msg("Mounting %s on %s\n", out, in);
+ if (mount("none", root_in, out, mount_flags, "") < 0)
+ die("Cannot mount %s on %s: %m", out, in);
+ }
+ else
+ {
+ msg("Binding %s on %s\n", out, in);
+ if (mount(out, root_in, "none", MS_BIND | MS_NOSUID | MS_NODEV | mount_flags, "") < 0)
+ die("Cannot bind %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))
+ 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);
if (buf[0])
die("Some tasks left in control group %s, failed to remove it", cg_path);
- // FIXME: Is this needed?
- // cg_write("memory.force_empty", "0\n");
-
if (rmdir(cg_path) < 0)
die("Cannot remove control group %s: %m", cg_path);
}
}
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)
static void
box_keeper(void)
{
- struct sigaction sa;
+ read_errors_from_fd = error_pipes[0];
+ close(error_pipes[1]);
+ struct sigaction sa;
bzero(&sa, sizeof(sa));
sa.sa_handler = signal_int;
sigaction(SIGINT, &sa, NULL);
}
if (p != box_pid)
die("wait4: unknown pid %d exited!", p);
+ box_pid = 0;
+
+ // Check error pipe if there is an internal error passed from inside the box
+ char interr[1024];
+ int n = read(read_errors_from_fd, interr, sizeof(interr) - 1);
+ if (n > 0)
+ {
+ interr[n] = 0;
+ die("%s", interr);
+ }
+
if (WIFEXITED(stat))
{
- box_pid = 0;
final_stats(&rus);
if (WEXITSTATUS(stat))
{
- // FIXME: Recognize internal errors during setup
meta_printf("exitcode:%d\n", WEXITSTATUS(stat));
err("RE: Exited with error status %d", WEXITSTATUS(stat));
}
wall_ms/1000, wall_ms%1000);
box_exit(0);
}
- if (WIFSIGNALED(stat))
+ else if (WIFSIGNALED(stat))
{
- box_pid = 0;
meta_printf("exitsig:%d\n", WTERMSIG(stat));
final_stats(&rus);
err("SG: Caught fatal signal %d", WTERMSIG(stat));
}
- if (WIFSTOPPED(stat))
+ else if (WIFSTOPPED(stat))
{
- box_pid = 0;
meta_printf("exitsig:%d\n", WSTOPSIG(stat));
final_stats(&rus);
err("SG: Stopped by signal %d", WSTOPSIG(stat));
if (mount("none", "root", "tmpfs", 0, "mode=755") < 0)
die("Cannot mount root ramdisk: %m");
- // FIXME: Make the list of bind-mounts configurable
- // FIXME: Virtual /dev?
- // FIXME: Read-only mounts?
-
- 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]; // FIXME
- sprintf(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");
die("Cannot change current directory: %m");
}
-static int
-box_inside(void *arg)
+static void
+setup_credentials(void)
{
- char **argv = arg;
- int argc = 0;
- while (argv[argc])
- argc++;
-
- struct rlimit rl;
- char *args[argc+1];
-
- memcpy(args, argv, argc * sizeof(char *));
- args[argc] = NULL;
-
- cg_enter();
- setup_root();
-
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)
die("setresuid: %m");
+ setpgrp();
+}
+static void
+setup_fds(void)
+{
if (redir_stdin)
{
close(0);
}
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(RLIMIT_AS): %m");
- }
+static void
+setup_rlim(const char *res_name, int res, rlim_t limit)
+{
+ struct rlimit rl = { .rlim_cur = limit, .rlim_max = limit };
+ if (setrlimit(res, &rl) < 0)
+ die("setrlimit(%s, %jd)", res_name, (intmax_t) limit);
+}
- 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");
+static void
+setup_rlimits(void)
+{
+#define RLIM(res, val) setup_rlim("RLIMIT_" #res, RLIMIT_##res, val)
+
+ if (memory_limit)
+ RLIM(AS, memory_limit * 1024);
- rl.rlim_cur = rl.rlim_max = 64;
- if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
- die("setrlimit(RLIMIT_NOFILE): %m");
+ RLIM(STACK, (stack_limit ? (rlim_t)stack_limit * 1024 : RLIM_INFINITY));
+ RLIM(NOFILE, 64);
+ RLIM(MEMLOCK, 0);
if (max_processes)
- {
- rl.rlim_cur = rl.rlim_max = max_processes;
- if (setrlimit(RLIMIT_NPROC, &rl) < 0)
- die("setrlimit(RLIMIT_NPROC): %m");
- }
+ RLIM(NPROC, max_processes);
- rl.rlim_cur = rl.rlim_max = 0;
- if (setrlimit(RLIMIT_MEMLOCK, &rl) < 0)
- die("setrlimit(RLIMIT_MEMLOCK): %m");
+#undef RLIM
+}
+static int
+box_inside(void *arg)
+{
+ char **args = arg;
+ write_errors_to_fd = error_pipes[1];
+ close(error_pipes[0]);
+
+ cg_enter();
+ setup_root();
+ setup_credentials();
+ setup_fds();
+ setup_rlimits();
char **env = setup_environment();
+
execve(args[0], args, env);
die("execve(\"%s\"): %m", args[0]);
}
+/*** Commands ***/
+
static void
init(void)
{
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");
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];
xsystem(cmd);
snprintf(cleanup_cmd, sizeof(cleanup_cmd), "chown -R %d.%d box", orig_uid, orig_gid);
+ if (pipe(error_pipes) < 0)
+ die("pipe: %m");
+ for (int i=0; i<2; i++)
+ if (fcntl(error_pipes[i], F_SETFD, fcntl(error_pipes[i], F_GETFD) | FD_CLOEXEC) < 0 ||
+ fcntl(error_pipes[i], F_SETFL, fcntl(error_pipes[i], F_GETFL) | O_NONBLOCK) < 0)
+ die("fcntl on pipe: %m");
+
box_pid = clone(
box_inside, // Function to execute as the body of the new process
argv, // Pass our stack
static void
show_version(void)
{
- // FIXME
- printf("Process isolator 0.0\n");
- printf("(c) 2012 Martin Mares <mj@ucw.cz>\n\n");
+ 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);
}
+/*** Options ***/
+
static void
usage(void)
{
-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: rw, noexec, fs, maybe\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\
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[] = "c::d:eE:i:k:m:M:o:p::r:t:vw:x:";
static const struct option long_opts[] = {
{ "cg", 2, 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' },
int c;
enum opt_code mode = 0;
+ init_dir_rules();
+
while ((c = getopt_long(argc, argv, short_opts, long_opts, NULL)) >= 0)
switch (c)
{
cg_root = optarg;
cg_enable = 1;
break;
+ case 'd':
+ if (!set_dir_action(optarg))
+ usage();
+ break;
case 'e':
pass_environ = 1;
break;