/*
* A Process Isolator based on Linux Containers
*
- * (c) 2012 Martin Mares <mj@ucw.cz>
- * (c) 2012 Bernard Blackham <bernard@blackham.com.au>
+ * (c) 2012-2013 Martin Mares <mj@ucw.cz>
+ * (c) 2012-2013 Bernard Blackham <bernard@blackham.com.au>
*/
#define _GNU_SOURCE
#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 char *set_cwd;
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];
return env;
}
-/*** Mount rules ***/
+/*** Directory rules ***/
struct dir_rule {
char *inside; // A relative path
static void make_dir(char *path)
{
- char *sep = path;
+ char *sep = (path[0] == '/' ? path+1 : path);
+
for (;;)
{
sep = strchr(sep, '/');
/*** Control groups ***/
-static char cg_path[256];
+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(char *attr, char *buf)
+cg_read(cg_controller controller, const char *attr, char *buf)
{
int maybe = 0;
if (attr[0] == '?')
}
char path[256];
- snprintf(path, sizeof(path), "%s/%s", cg_path, attr);
+ cg_makepath(path, sizeof(path), controller, attr);
int fd = open(path, O_RDONLY);
if (fd < 0)
return 1;
}
-static void __attribute__((format(printf,2,3)))
-cg_write(char *attr, char *fmt, ...)
+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_writef: Value for attribute %s is too long", attr);
+ die("cg_write: Value for attribute %s is too long", attr);
if (verbose > 1)
msg("CG: Write %s = %s", attr, buf);
char path[256];
- snprintf(path, sizeof(path), "%s/%s", cg_path, attr);
+ cg_makepath(path, sizeof(path), controller, attr);
int fd = open(path, O_WRONLY | O_TRUNC);
if (fd < 0)
- die("Cannot write %s: %m", path);
+ {
+ if (maybe)
+ return;
+ else
+ die("Cannot write %s: %m", path);
+ }
int written = write(fd, buf, n);
if (written < 0)
- die("Cannot set %s to %s: %m", path, buf);
+ {
+ 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);
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_path, sizeof(cg_path), "%s/box-%d", cg_root, BOX_UID);
- msg("Using control group %s\n", cg_path);
+ snprintf(cg_name, sizeof(cg_name), "box-%d", box_id);
+ msg("Using control group %s\n", cg_name);
}
static void
struct stat st;
char buf[CG_BUFSIZE];
+ char path[256];
- if (stat(cg_path, &st) >= 0 || errno != ENOENT)
+ FOREACH_CG_CONTROLLER(controller)
{
- msg("Control group %s already exists, trying to empty it.\n", cg_path);
- if (rmdir(cg_path) < 0)
- die("Failed to reset control group %s: %m", cg_path);
- }
+ 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(cg_path, 0777) < 0)
- die("Failed to create control group %s: %m", cg_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("?../cpuset.cpus", buf))
- cg_write("cpuset.cpus", "%s", buf);
- if (cg_read("?../cpuset.mems", buf))
- cg_write("cpuset.mems", "%s", buf);
+ 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
if (!cg_enable)
return;
- msg("Entering control group %s\n", cg_path);
+ msg("Entering control group %s\n", cg_name);
- struct stat st;
- if (stat(cg_path, &st) < 0)
- die("Control group %s does not exist: %m", cg_path);
+ 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("memory.limit_in_bytes", "%lld\n", (long long) cg_memory_limit << 10);
- cg_write("memory.memsw.limit_in_bytes", "%lld\n", (long long) cg_memory_limit << 10);
+ 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("cpuacct.usage", "0\n");
-
- cg_write("tasks", "%d\n", (int) getpid());
+ cg_write(CG_CPUACCT, "cpuacct.usage", "0\n");
}
static int
return 0;
char buf[CG_BUFSIZE];
- cg_read("cpuacct.usage", buf);
+ cg_read(CG_CPUACCT, "cpuacct.usage", buf);
unsigned long long ns = atoll(buf);
return ns / 1000000;
}
// Memory usage statistics
unsigned long long mem=0, memsw=0;
- if (cg_read("?memory.max_usage_in_bytes", buf))
+ if (cg_read(CG_MEMORY, "?memory.max_usage_in_bytes", buf))
mem = atoll(buf);
- if (cg_read("?memory.memsw.max_usage_in_bytes", buf))
+ if (cg_read(CG_MEMORY, "?memory.memsw.max_usage_in_bytes", buf))
{
memsw = atoll(buf);
if (memsw > mem)
if (!cg_enable)
return;
- cg_read("tasks", buf);
- if (buf[0])
- die("Some tasks left in control group %s, failed to remove it", cg_path);
+ 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 ***/
- if (rmdir(cg_path) < 0)
- die("Cannot remove control group %s: %m", cg_path);
+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
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();
}
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
+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
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();
}
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 ***/
-static void
-usage(void)
+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: 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, --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_RUN,
OPT_CLEANUP,
OPT_VERSION,
+ OPT_CG,
OPT_CG_MEM,
OPT_CG_TIMING,
};
-static const char short_opts[] = "c::d: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' },
+ { "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 },
{ "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;
+ set_cwd = optarg;
+ break;
+ case OPT_CG:
cg_enable = 1;
break;
case 'd':
if (!set_dir_action(optarg))
- usage();
+ 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 = 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 OPT_RUN:
case OPT_CLEANUP:
case OPT_VERSION:
- mode = c;
+ if (!mode || mode == c)
+ mode = c;
+ else
+ usage("Only one command is allowed.\n");
break;
case OPT_CG_MEM:
cg_memory_limit = atoi(optarg);
cg_timing = 1;
break;
default:
- usage();
+ usage(NULL);
}
if (!mode)
- usage();
+ usage("Please specify an isolate command (e.g. --init, --run).\n");
if (mode == OPT_VERSION)
{
show_version();
orig_gid = getgid();
umask(022);
- if (chdir(BOX_DIR) < 0)
- die("chdir(%s): %m", BOX_DIR);
+ box_init();
cg_init();
switch (mode)
{
case OPT_INIT:
if (optind < argc)
- usage();
+ 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: