/*
* A Process Isolator based on Linux Containers
*
- * (c) 2012 Martin Mares <mj@ucw.cz>
- * (c) 2012 Bernard Blackham <bernard@blackham.com.au>
+ * (c) 2012-2014 Martin Mares <mj@ucw.cz>
+ * (c) 2012-2014 Bernard Blackham <bernard@blackham.com.au>
*/
#define _GNU_SOURCE
#include <getopt.h>
#include <sched.h>
#include <time.h>
+#include <ftw.h>
#include <grp.h>
#include <mntent.h>
#include <limits.h>
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 gid_t orig_gid;
static int partial_line;
-static char cleanup_cmd[256];
+static int cleanup_ownership;
static struct timeval start_time;
static int ticks_per_sec;
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 FILE *metafile;
/*** Messages and exits ***/
-static void
-xsystem(const char *cmd)
-{
- int ret = system(cmd);
- if (ret < 0)
- die("system(\"%s\"): %m", cmd);
- if (!WIFEXITED(ret) || WEXITSTATUS(ret))
- die("system(\"%s\"): Exited with status %d", cmd, ret);
-}
-
static void NONRET
box_exit(int rc)
{
final_stats(&rus);
}
- if (rc < 2 && cleanup_cmd[0])
- xsystem(cleanup_cmd);
+ if (rc < 2 && cleanup_ownership)
+ {
+ chowntree("box", orig_uid, orig_gid);
+ }
meta_close();
exit(rc);
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 {
/*** 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)
int n = read(fd, buf, CG_BUFSIZE);
if (n < 0)
- die("Cannot read %s: %m", path);
+ {
+ 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')
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 (!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_id);
- 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);
- if (rmdir(cg_path) < 0)
- die("Cannot remove control group %s: %m", cg_path);
+ char path[256];
+ cg_makepath(path, sizeof(path), controller, "");
+
+ if (rmdir(path) < 0)
+ die("Cannot remove control group %s: %m", path);
+ }
}
/*** Disk quotas ***/
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]);
}
init(void)
{
msg("Preparing sandbox directory\n");
- xsystem("rm -rf box");
+ rmtree("box");
if (mkdir("box", 0700) < 0)
die("Cannot create box: %m");
if (chown("box", orig_uid, orig_gid) < 0)
die("Box directory not found, there isn't anything to clean up");
msg("Deleting sandbox directory\n");
- xsystem("rm -rf *");
- if (rmdir(box_dir) < 0)
- die("Cannot remove %s: %m", box_dir);
+ rmtree(box_dir);
cg_remove();
}
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);
- xsystem(cmd);
- snprintf(cleanup_cmd, sizeof(cleanup_cmd), "chown -R %d.%d box", orig_uid, orig_gid);
+ chowntree("box", box_uid, box_gid);
+ cleanup_ownership = 1;
if (pipe(error_pipes) < 0)
die("pipe: %m");
/*** 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\
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::q: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[] = {
{ "box-id", 1, NULL, 'b' },
- { "cg", 1, NULL, 'c' },
+ { "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 },
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();
+ 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);
case 'q':
sep = strchr(optarg, ',');
if (!sep)
- usage();
+ usage("Invalid quota specified: %s\n", optarg);
block_quota = atoi(optarg);
inode_quota = atoi(sep+1);
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();
{
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: