/*
- * A Process Isolator based in Linux Containers
+ * A Process Isolator based on Linux Containers
*
- * (c) 2012 Martin Mares <mj@ucw.cz>
+ * (c) 2012-2013 Martin Mares <mj@ucw.cz>
+ * (c) 2012-2013 Bernard Blackham <bernard@blackham.com.au>
*/
#define _GNU_SOURCE
#include "autoconf.h"
-// FIXME: prune
#include <errno.h>
#include <stdio.h>
#include <fcntl.h>
#include <getopt.h>
#include <sched.h>
#include <time.h>
+#include <ftw.h>
#include <grp.h>
+#include <mntent.h>
+#include <limits.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>
+#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]))
-// FIXME: Make configurable, probably in compile time
-#define BOX_DIR "/tmp/box"
-#define BOX_UID 60000
-#define BOX_GID 60000
-
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 char *set_cwd;
+static int cg_enable;
+static int cg_memory_limit;
+static int cg_timing;
+
+static int box_id;
+static char box_dir[1024];
static pid_t box_pid;
-static volatile int timer_tick;
-static struct timeval start_time;
-static int ticks_per_sec;
+
+static uid_t box_uid;
+static gid_t box_gid;
+static uid_t orig_uid;
+static gid_t orig_gid;
+
static int partial_line;
+static int cleanup_ownership;
-static int mem_peak_kb;
+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 sample_mem_peak(void);
+static void cg_stats(void);
+static int get_wall_time_ms(void);
+static int get_run_time_ms(struct rusage *rus);
+
+static void chowntree(char *path, uid_t uid, gid_t gid);
/*** Meta-files ***/
static void
final_stats(struct rusage *rus)
{
- struct timeval total, now, wall;
- timeradd(&rus->ru_utime, &rus->ru_stime, &total);
- total_ms = total.tv_sec*1000 + total.tv_usec/1000;
- gettimeofday(&now, NULL);
- timersub(&now, &start_time, &wall);
- wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
+ 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);
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);
+ meta_printf("max-rss:%ld\n", rus->ru_maxrss);
+ meta_printf("csw-voluntary:%ld\n", rus->ru_nvcsw);
+ meta_printf("csw-forced:%ld\n", rus->ru_nivcsw);
+
+ cg_stats();
}
/*** Messages and exits ***/
{
if (box_pid > 0)
{
- sample_mem_peak();
kill(-box_pid, SIGKILL);
kill(box_pid, SIGKILL);
meta_printf("killed:1\n");
else
final_stats(&rus);
}
+
+ if (rc < 2 && cleanup_ownership)
+ {
+ chowntree("box", orig_uid, orig_gid);
+ }
+
meta_close();
exit(rc);
}
{
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));
+}
+
+static int rmtree_helper(const char *fpath, const struct stat *sb,
+ int typeflag, struct FTW *ftwbuf)
+{
+ if (S_ISDIR(sb->st_mode))
+ {
+ if (rmdir(fpath) < 0)
+ die("Cannot rmdir %s: %m", fpath);
+ }
+ else
+ {
+ if (unlink(fpath) < 0)
+ die("Cannot unlink %s: %m", fpath);
+ }
+ return FTW_CONTINUE;
+}
+
+static void
+rmtree(char *path)
+{
+ nftw(path, rmtree_helper, 32, FTW_MOUNT | FTW_PHYS | FTW_DEPTH);
+}
+
+static uid_t chown_uid;
+static gid_t chown_gid;
+static int chowntree_helper(const char *fpath, const struct stat *sb,
+ int typeflag, struct FTW *ftwbuf)
+{
+ if (lchown(fpath, chown_uid, chown_gid) < 0)
+ die("Cannot chown %s: %m", fpath);
+ else
+ return FTW_CONTINUE;
+}
+
+static void
+chowntree(char *path, uid_t uid, gid_t gid)
+{
+ chown_uid = uid;
+ chown_gid = gid;
+ nftw(path, chowntree_helper, 32, FTW_MOUNT | FTW_PHYS);
+}
+
/*** Environment rules ***/
struct env_rule {
return env;
}
-/*** FIXME ***/
+/*** 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 ***/
+
+struct cg_controller_desc {
+ const char *name;
+ int optional;
+};
+
+typedef enum {
+ CG_MEMORY = 0,
+ CG_CPUACCT,
+ CG_CPUSET,
+ CG_NUM_CONTROLLERS,
+} cg_controller;
+
+static const struct cg_controller_desc cg_controllers[CG_NUM_CONTROLLERS+1] = {
+ [CG_MEMORY] = { "memory", 0 },
+ [CG_CPUACCT] = { "cpuacct", 0 },
+ [CG_CPUSET] = { "cpuset", 1 },
+ [CG_NUM_CONTROLLERS] = { NULL, 0 },
+};
+
+#define FOREACH_CG_CONTROLLER(_controller) \
+ for (cg_controller (_controller) = 0; \
+ (_controller) < CG_NUM_CONTROLLERS; (_controller)++)
+
+static const char *cg_controller_name(cg_controller c)
+{
+ return cg_controllers[c].name;
+}
+
+static const int cg_controller_optional(cg_controller c)
+{
+ return cg_controllers[c].optional;
+}
+
+static char cg_name[256];
+
+#define CG_BUFSIZE 1024
+
+static void
+cg_makepath(char *buf, size_t len, cg_controller c, const char *attr)
+{
+ const char *cg_root = CONFIG_ISOLATE_CGROUP_ROOT;
+ snprintf(buf, len, "%s/%s/%s/%s", cg_root, cg_controller_name(c), cg_name, attr);
+}
+
+static int
+cg_read(cg_controller controller, const char *attr, char *buf)
+{
+ int maybe = 0;
+ if (attr[0] == '?')
+ {
+ attr++;
+ maybe = 1;
+ }
+
+ char path[256];
+ cg_makepath(path, sizeof(path), controller, attr);
+
+ int fd = open(path, O_RDONLY);
+ if (fd < 0)
+ {
+ if (maybe)
+ return 0;
+ die("Cannot read %s: %m", path);
+ }
+
+ int n = read(fd, buf, CG_BUFSIZE);
+ if (n < 0)
+ {
+ if (maybe)
+ return 0;
+ die("Cannot read %s: %m", path);
+ }
+ if (n >= CG_BUFSIZE - 1)
+ die("Attribute %s too long", path);
+ if (n > 0 && buf[n-1] == '\n')
+ n--;
+ buf[n] = 0;
+
+ if (verbose > 1)
+ msg("CG: Read %s = %s\n", attr, buf);
+
+ close(fd);
+ return 1;
+}
+
+static void __attribute__((format(printf,3,4)))
+cg_write(cg_controller controller, const char *attr, const char *fmt, ...)
+{
+ int maybe = 0;
+ if (attr[0] == '?')
+ {
+ attr++;
+ maybe = 1;
+ }
+
+ va_list args;
+ va_start(args, fmt);
+
+ char buf[CG_BUFSIZE];
+ int n = vsnprintf(buf, sizeof(buf), fmt, args);
+ if (n >= CG_BUFSIZE)
+ die("cg_write: Value for attribute %s is too long", attr);
+
+ if (verbose > 1)
+ msg("CG: Write %s = %s", attr, buf);
+
+ char path[256];
+ cg_makepath(path, sizeof(path), controller, attr);
+
+ int fd = open(path, O_WRONLY | O_TRUNC);
+ if (fd < 0)
+ {
+ if (maybe)
+ return;
+ else
+ die("Cannot write %s: %m", path);
+ }
+
+ int written = write(fd, buf, n);
+ if (written < 0)
+ {
+ if (maybe)
+ return;
+ else
+ die("Cannot set %s to %s: %m", path, buf);
+ }
+ if (written != n)
+ die("Short write to %s (%d out of %d bytes)", path, written, n);
+
+ close(fd);
+ va_end(args);
+}
+
+static void
+cg_init(void)
+{
+ if (!cg_enable)
+ return;
+
+ char *cg_root = CONFIG_ISOLATE_CGROUP_ROOT;
+ if (!dir_exists(cg_root))
+ die("Control group filesystem at %s not mounted", cg_root);
+
+ snprintf(cg_name, sizeof(cg_name), "box-%d", box_id);
+ msg("Using control group %s\n", cg_name);
+}
+
+static void
+cg_prepare(void)
+{
+ if (!cg_enable)
+ return;
+
+ struct stat st;
+ char buf[CG_BUFSIZE];
+ char path[256];
+
+ FOREACH_CG_CONTROLLER(controller)
+ {
+ cg_makepath(path, sizeof(path), controller, "");
+ if (stat(path, &st) >= 0 || errno != ENOENT)
+ {
+ msg("Control group %s already exists, trying to empty it.\n", path);
+ if (rmdir(path) < 0)
+ die("Failed to reset control group %s: %m", path);
+ }
+
+ if (mkdir(path, 0777) < 0 && !cg_controller_optional(controller))
+ die("Failed to create control group %s: %m", path);
+ }
+
+ // If cpuset module is enabled, copy allowed cpus and memory nodes from parent group
+ if (cg_read(CG_CPUSET, "?cpuset.cpus", buf))
+ cg_write(CG_CPUSET, "cpuset.cpus", "%s", buf);
+ if (cg_read(CG_CPUSET, "?cpuset.mems", buf))
+ cg_write(CG_CPUSET, "cpuset.mems", "%s", buf);
+}
+
+static void
+cg_enter(void)
+{
+ if (!cg_enable)
+ return;
+
+ msg("Entering control group %s\n", cg_name);
+
+ FOREACH_CG_CONTROLLER(controller)
+ {
+ if (cg_controller_optional(controller))
+ cg_write(controller, "?tasks", "%d\n", (int) getpid());
+ else
+ cg_write(controller, "tasks", "%d\n", (int) getpid());
+ }
+
+ if (cg_memory_limit)
+ {
+ cg_write(CG_MEMORY, "memory.limit_in_bytes", "%lld\n", (long long) cg_memory_limit << 10);
+ cg_write(CG_MEMORY, "?memory.memsw.limit_in_bytes", "%lld\n", (long long) cg_memory_limit << 10);
+ }
+
+ if (cg_timing)
+ cg_write(CG_CPUACCT, "cpuacct.usage", "0\n");
+}
+
+static int
+cg_get_run_time_ms(void)
+{
+ if (!cg_enable)
+ return 0;
+
+ char buf[CG_BUFSIZE];
+ cg_read(CG_CPUACCT, "cpuacct.usage", buf);
+ unsigned long long ns = atoll(buf);
+ return ns / 1000000;
+}
+
+static void
+cg_stats(void)
+{
+ if (!cg_enable)
+ return;
+
+ char buf[CG_BUFSIZE];
+
+ // Memory usage statistics
+ unsigned long long mem=0, memsw=0;
+ if (cg_read(CG_MEMORY, "?memory.max_usage_in_bytes", buf))
+ mem = atoll(buf);
+ if (cg_read(CG_MEMORY, "?memory.memsw.max_usage_in_bytes", buf))
+ {
+ memsw = atoll(buf);
+ if (memsw > mem)
+ mem = memsw;
+ }
+ if (mem)
+ meta_printf("cg-mem:%lld\n", mem >> 10);
+}
+
+static void
+cg_remove(void)
+{
+ char buf[CG_BUFSIZE];
+
+ if (!cg_enable)
+ return;
+
+ FOREACH_CG_CONTROLLER(controller)
+ {
+ if (cg_controller_optional(controller)) {
+ if (!cg_read(controller, "?tasks", buf))
+ continue;
+ } else
+ cg_read(controller, "tasks", buf);
+
+ if (buf[0])
+ die("Some tasks left in controller %s of cgroup %s, failed to remove it",
+ cg_controller_name(controller), cg_name);
+
+ char path[256];
+ cg_makepath(path, sizeof(path), controller, "");
+
+ if (rmdir(path) < 0)
+ die("Cannot remove control group %s: %m", 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
signal_alarm(int unused UNUSED)
buf[c] = 0;
}
+static int
+get_wall_time_ms(void)
+{
+ struct timeval now, wall;
+ gettimeofday(&now, NULL);
+ timersub(&now, &start_time, &wall);
+ return wall.tv_sec*1000 + wall.tv_usec/1000;
+}
+
+static int
+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;
+
+ read_proc_file(buf, "stat", &proc_stat_fd);
+ x = buf;
+ while (*x && *x != ' ')
+ x++;
+ while (*x == ' ')
+ x++;
+ if (*x++ != '(')
+ 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 stat syntax error 2");
+
+ return (utime + stime) * 1000 / ticks_per_sec;
+}
+
static void
check_timeout(void)
{
if (wall_timeout)
{
- struct timeval now, wall;
- int wall_ms;
- gettimeofday(&now, NULL);
- timersub(&now, &start_time, &wall);
- wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
+ int wall_ms = get_wall_time_ms();
if (wall_ms > wall_timeout)
err("TO: Time limit exceeded (wall clock)");
if (verbose > 1)
}
if (timeout)
{
- 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 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 stat syntax error 2");
- ms = (utime + stime) * 1000 / ticks_per_sec;
+ 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
-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
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));
}
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 MB)\n",
+ fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall)\n",
total_ms/1000, total_ms%1000,
- wall_ms/1000, wall_ms%1000,
- (mem_peak_kb + 1023) / 1024);
+ 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));
}
}
+/*** The process running inside the box ***/
+
static void
setup_root(void)
{
- umask(0022);
-
- if (mkdir("root", 0777) < 0 && errno != EEXIST)
+ if (mkdir("root", 0750) < 0 && errno != EEXIST)
die("mkdir('root'): %m");
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));
- printf("Binding %s on %s\n", d, buf);
- if (mkdir(buf, 0777) < 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", 0777) < 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;
-
- setup_root();
-
- 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();
+}
+static void
+setup_fds(void)
+{
if (redir_stdin)
{
close(0);
}
else
dup2(1, 2);
- setpgrp();
+}
+
+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);
+}
+
+static void
+setup_rlimits(void)
+{
+#define RLIM(res, val) setup_rlim("RLIMIT_" #res, RLIMIT_##res, val)
if (memory_limit)
- {
- rl.rlim_cur = rl.rlim_max = memory_limit * 1024;
- if (setrlimit(RLIMIT_AS, &rl) < 0)
- die("setrlimit(RLIMIT_AS): %m");
- }
+ RLIM(AS, memory_limit * 1024);
- 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");
+ RLIM(STACK, (stack_limit ? (rlim_t)stack_limit * 1024 : RLIM_INFINITY));
+ RLIM(NOFILE, 64);
+ RLIM(MEMLOCK, 0);
- rl.rlim_cur = rl.rlim_max = 64;
- if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
- die("setrlimit(RLIMIT_NOFILE): %m");
+ if (max_processes)
+ RLIM(NPROC, max_processes);
- // FIXME: Create multi-process mode
- rl.rlim_cur = rl.rlim_max = 1;
- if (setrlimit(RLIMIT_NPROC, &rl) < 0)
- die("setrlimit(RLIMIT_NPROC): %m");
+#undef RLIM
+}
- rl.rlim_cur = rl.rlim_max = 0;
- if (setrlimit(RLIMIT_MEMLOCK, &rl) < 0)
- die("setrlimit(RLIMIT_MEMLOCK): %m");
+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();
+
+ if (set_cwd && chdir(set_cwd))
+ die("chdir: %m");
+
execve(args[0], args, env);
die("execve(\"%s\"): %m", args[0]);
}
static void
-prepare(void)
+box_init(void)
{
- // FIXME: Move chdir to common code?
- if (chdir(BOX_DIR) < 0)
- die("chdir(%s): %m", BOX_DIR);
+ 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 ***/
- if (system("./prepare"))
- die("Prepare hook failed");
+static void
+init(void)
+{
+ msg("Preparing sandbox directory\n");
+ rmtree("box");
+ if (mkdir("box", 0700) < 0)
+ die("Cannot create box: %m");
+ if (chown("box", orig_uid, orig_gid) < 0)
+ die("Cannot chown box: %m");
+
+ cg_prepare();
+ set_quota();
+
+ puts(box_dir);
}
static void
cleanup(void)
{
- if (chdir(BOX_DIR) < 0)
- die("chdir(%s): %m", BOX_DIR);
+ if (!dir_exists("box"))
+ die("Box directory not found, there isn't anything to clean up");
- if (system("./cleanup"))
- die("Prepare hook failed");
+ msg("Deleting sandbox directory\n");
+ rmtree(box_dir);
+ cg_remove();
}
static void
run(char **argv)
{
- if (chdir(BOX_DIR) < 0)
- die("chdir(%s): %m", BOX_DIR);
+ if (!dir_exists("box"))
+ die("Box directory not found, did you run `isolate --init'?");
- struct stat st;
- if (stat("box", &st) < 0 || !S_ISDIR(st.st_mode))
- die("Box directory not found, did you run `isolate --prepare'?");
+ chowntree("box", box_uid, box_gid);
+ cleanup_ownership = 1;
+
+ 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
box_keeper();
}
-// FIXME: Prune (and also the option list)
static void
-usage(void)
+show_version(void)
+{
+ 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", 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 ***/
+
+static void __attribute__((format(printf,1,2)))
+usage(const char *msg, ...)
{
- fprintf(stderr, "Invalid arguments!\n");
+ if (msg != NULL)
+ {
+ va_list args;
+ va_start(args, msg);
+ vfprintf(stderr, msg, args);
+ va_end(args);
+ }
printf("\
-Usage: box [<options>] -- <command> <arguments>\n\
+Usage: isolate [<options>] <command>\n\
\n\
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\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\
--k <size>\tLimit stack size to <size> KB (default: 0=unlimited)\n\
--m <size>\tLimit address space to <size> KB\n\
--M <file>\tOutput process information to <file> (name:value)\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\
--r <file>\tRedirect stderr to <file>\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 (use multiple times for even more verbosity)\n\
--w <time>\tSet wall clock time limit (seconds, fractions allowed)\n\
--x <time>\tSet extra timeout, before which a timing-out program is not yet killed,\n\
-\t\tso that its real execution time is reported (seconds, fractions allowed)\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\
+-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\
+-o, --stdout=<file>\tRedirect stdout to <file>\n\
+-p, --processes[=<max>]\tEnable multiple processes (at most <max> of them); needs --cg\n\
+-t, --time=<time>\tSet run time limit (seconds, fractions allowed)\n\
+-v, --verbose\t\tBe verbose (use multiple times for even more verbosity)\n\
+-w, --wall-time=<time>\tSet wall clock time limit (seconds, fractions allowed)\n\
+\n\
+Commands:\n\
+ --init\t\tInitialize sandbox (and its control group when --cg is used)\n\
+ --run -- <cmd> ...\tRun given command within sandbox\n\
+ --cleanup\t\tClean up sandbox\n\
+ --version\t\tDisplay program version and configuration\n\
");
exit(2);
}
enum opt_code {
- OPT_PREPARE,
+ OPT_INIT = 256,
OPT_RUN,
OPT_CLEANUP,
+ OPT_VERSION,
+ OPT_CG,
+ OPT_CG_MEM,
+ OPT_CG_TIMING,
};
-static const char short_opts[] = "a:c:eE:fi:k:m:M:o:p:r:s:t:Tvw: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[] = {
- { "prepare", 0, NULL, OPT_PREPARE },
- { "run", 0, NULL, OPT_RUN },
+ { "box-id", 1, NULL, 'b' },
+ { "chdir", 1, NULL, 'c' },
+ { "cg", 0, NULL, OPT_CG },
+ { "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' },
+ { "init", 0, NULL, OPT_INIT },
+ { "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' },
+ { "stdin", 1, NULL, 'i' },
+ { "stdout", 1, NULL, 'o' },
+ { "time", 1, NULL, 't' },
+ { "verbose", 0, NULL, 'v' },
+ { "version", 0, NULL, OPT_VERSION },
+ { "wall-time", 1, NULL, 'w' },
{ NULL, 0, NULL, 0 }
};
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':
set_cwd = optarg;
break;
+ case OPT_CG:
+ cg_enable = 1;
+ break;
+ case 'd':
+ if (!set_dir_action(optarg))
+ usage("Invalid directory specified: %s\n", optarg);
+ break;
case 'e':
pass_environ = 1;
break;
case 'E':
if (!set_env_action(optarg))
- usage();
+ usage("Invalid environment specified: %s\n", optarg);
break;
case 'k':
- stack_limit = atol(optarg);
+ stack_limit = atoi(optarg);
break;
case 'i':
redir_stdin = optarg;
break;
case 'm':
- memory_limit = atol(optarg);
+ memory_limit = atoi(optarg);
break;
case 'M':
meta_open(optarg);
case 'o':
redir_stdout = optarg;
break;
+ case 'p':
+ if (optarg)
+ max_processes = atoi(optarg);
+ else
+ max_processes = 0;
+ break;
+ case 'q':
+ sep = strchr(optarg, ',');
+ if (!sep)
+ usage("Invalid quota specified: %s\n", optarg);
+ block_quota = atoi(optarg);
+ inode_quota = atoi(sep+1);
+ break;
case 'r':
redir_stderr = optarg;
break;
case 'x':
extra_timeout = 1000*atof(optarg);
break;
- case OPT_PREPARE:
+ case OPT_INIT:
case OPT_RUN:
case OPT_CLEANUP:
- mode = c;
+ case OPT_VERSION:
+ if (!mode || mode == c)
+ mode = c;
+ else
+ usage("Only one command is allowed.\n");
+ break;
+ case OPT_CG_MEM:
+ cg_memory_limit = atoi(optarg);
+ break;
+ case OPT_CG_TIMING:
+ cg_timing = 1;
break;
default:
- usage();
+ usage(NULL);
}
+ if (!mode)
+ usage("Please specify an isolate command (e.g. --init, --run).\n");
+ if (mode == OPT_VERSION)
+ {
+ show_version();
+ return 0;
+ }
+
if (geteuid())
die("Must be started as root");
+ orig_uid = getuid();
+ orig_gid = getgid();
- // FIXME: Copying of files into the box
+ umask(022);
+ box_init();
+ cg_init();
switch (mode)
{
- case OPT_PREPARE:
+ case OPT_INIT:
if (optind < argc)
- usage();
- prepare();
+ usage("--init mode takes no parameters\n");
+ init();
break;
case OPT_RUN:
if (optind >= argc)
- usage();
+ usage("--run mode requires a command to run\n");
run(argv+optind);
break;
case OPT_CLEANUP:
if (optind < argc)
- usage();
+ usage("--cleanup mode takes no parameters\n");
cleanup();
break;
default: