kernel/pid_namespace.c

   1 /*
   2  * Pid namespaces
   3  *
   4  * Authors:
   5  *    (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
   6  *    (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
   7  *     Many thanks to Oleg Nesterov for comments and help
   8  *
   9  */
  10
  11 #include <linux/pid.h>
  12 #include <linux/pid_namespace.h>
  13 #include <linux/user_namespace.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/err.h>
  16 #include <linux/acct.h>
  17 #include <linux/slab.h>
  18 #include <linux/proc_ns.h>
  19 #include <linux/reboot.h>
  20 #include <linux/export.h>
  21
  22 struct pid_cache {
  23         int nr_ids;
  24         char name[16];
  25         struct kmem_cache *cachep;
  26         struct list_head list;
  27 };
  28
  29 static LIST_HEAD(pid_caches_lh);
  30 static DEFINE_MUTEX(pid_caches_mutex);
  31 static struct kmem_cache *pid_ns_cachep;
  32
  33 /*
  34  * creates the kmem cache to allocate pids from.
  35  * @nr_ids: the number of numerical ids this pid will have to carry
  36  */
  37
  38 static struct kmem_cache *create_pid_cachep(int nr_ids)
  39 {
  40         struct pid_cache *pcache;
  41         struct kmem_cache *cachep;
  42
  43         mutex_lock(&pid_caches_mutex);
  44         list_for_each_entry(pcache, &pid_caches_lh, list)
  45                 if (pcache->nr_ids == nr_ids)
  46                         goto out;
  47
  48         pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
  49         if (pcache == NULL)
  50                 goto err_alloc;
  51
  52         snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
  53         cachep = kmem_cache_create(pcache->name,
  54                         sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
  55                         0, SLAB_HWCACHE_ALIGN, NULL);
  56         if (cachep == NULL)
  57                 goto err_cachep;
  58
  59         pcache->nr_ids = nr_ids;
  60         pcache->cachep = cachep;
  61         list_add(&pcache->list, &pid_caches_lh);
  62 out:
  63         mutex_unlock(&pid_caches_mutex);
  64         return pcache->cachep;
  65
  66 err_cachep:
  67         kfree(pcache);
  68 err_alloc:
  69         mutex_unlock(&pid_caches_mutex);
  70         return NULL;
  71 }
  72
  73 static void proc_cleanup_work(struct work_struct *work)
  74 {
  75         struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
  76         pid_ns_release_proc(ns);
  77 }
  78
  79 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
  80 #define MAX_PID_NS_LEVEL 32
  81
  82 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
  83         struct pid_namespace *parent_pid_ns)
  84 {
  85         struct pid_namespace *ns;
  86         unsigned int level = parent_pid_ns->level + 1;
  87         int i;
  88         int err;
  89
  90         if (level > MAX_PID_NS_LEVEL) {
  91                 err = -EINVAL;
  92                 goto out;
  93         }
  94
  95         err = -ENOMEM;
  96         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
  97         if (ns == NULL)
  98                 goto out;
  99
 100         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 101         if (!ns->pidmap[0].page)
 102                 goto out_free;
 103
 104         ns->pid_cachep = create_pid_cachep(level + 1);
 105         if (ns->pid_cachep == NULL)
 106                 goto out_free_map;
 107
 108         err = proc_alloc_inum(&ns->proc_inum);
 109         if (err)
 110                 goto out_free_map;
 111
 112         kref_init(&ns->kref);
 113         ns->level = level;
 114         ns->parent = get_pid_ns(parent_pid_ns);
 115         ns->user_ns = get_user_ns(user_ns);
 116         ns->nr_hashed = PIDNS_HASH_ADDING;
 117         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 118
 119         set_bit(0, ns->pidmap[0].page);
 120         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 121
 122         for (i = 1; i < PIDMAP_ENTRIES; i++)
 123                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 124
 125         return ns;
 126
 127 out_free_map:
 128         kfree(ns->pidmap[0].page);
 129 out_free:
 130         kmem_cache_free(pid_ns_cachep, ns);
 131 out:
 132         return ERR_PTR(err);
 133 }
 134
 135 static void delayed_free_pidns(struct rcu_head *p)
 136 {
 137         kmem_cache_free(pid_ns_cachep,
 138                         container_of(p, struct pid_namespace, rcu));
 139 }
 140
 141 static void destroy_pid_namespace(struct pid_namespace *ns)
 142 {
 143         int i;
 144
 145         proc_free_inum(ns->proc_inum);
 146         for (i = 0; i < PIDMAP_ENTRIES; i++)
 147                 kfree(ns->pidmap[i].page);
 148         put_user_ns(ns->user_ns);
 149         call_rcu(&ns->rcu, delayed_free_pidns);
 150 }
 151
 152 struct pid_namespace *copy_pid_ns(unsigned long flags,
 153         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 154 {
 155         if (!(flags & CLONE_NEWPID))
 156                 return get_pid_ns(old_ns);
 157         if (task_active_pid_ns(current) != old_ns)
 158                 return ERR_PTR(-EINVAL);
 159         return create_pid_namespace(user_ns, old_ns);
 160 }
 161
 162 static void free_pid_ns(struct kref *kref)
 163 {
 164         struct pid_namespace *ns;
 165
 166         ns = container_of(kref, struct pid_namespace, kref);
 167         destroy_pid_namespace(ns);
 168 }
 169
 170 void put_pid_ns(struct pid_namespace *ns)
 171 {
 172         struct pid_namespace *parent;
 173
 174         while (ns != &init_pid_ns) {
 175                 parent = ns->parent;
 176                 if (!kref_put(&ns->kref, free_pid_ns))
 177                         break;
 178                 ns = parent;
 179         }
 180 }
 181 EXPORT_SYMBOL_GPL(put_pid_ns);
 182
 183 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 184 {
 185         int nr;
 186         int rc;
 187         struct task_struct *task, *me = current;
 188         int init_pids = thread_group_leader(me) ? 1 : 2;
 189
 190         /* Don't allow any more processes into the pid namespace */
 191         disable_pid_allocation(pid_ns);
 192
 193         /*
 194          * Ignore SIGCHLD causing any terminated children to autoreap.
 195          * This speeds up the namespace shutdown, plus see the comment
 196          * below.
 197          */
 198         spin_lock_irq(&me->sighand->siglock);
 199         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 200         spin_unlock_irq(&me->sighand->siglock);
 201
 202         /*
 203          * The last thread in the cgroup-init thread group is terminating.
 204          * Find remaining pid_ts in the namespace, signal and wait for them
 205          * to exit.
 206          *
 207          * Note:  This signals each threads in the namespace - even those that
 208          *        belong to the same thread group, To avoid this, we would have
 209          *        to walk the entire tasklist looking a processes in this
 210          *        namespace, but that could be unnecessarily expensive if the
 211          *        pid namespace has just a few processes. Or we need to
 212          *        maintain a tasklist for each pid namespace.
 213          *
 214          */
 215         read_lock(&tasklist_lock);
 216         nr = next_pidmap(pid_ns, 1);
 217         while (nr > 0) {
 218                 rcu_read_lock();
 219
 220                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 221                 if (task && !__fatal_signal_pending(task))
 222                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 223
 224                 rcu_read_unlock();
 225
 226                 nr = next_pidmap(pid_ns, nr);
 227         }
 228         read_unlock(&tasklist_lock);
 229
 230         /*
 231          * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
 232          * sys_wait4() will also block until our children traced from the
 233          * parent namespace are detached and become EXIT_DEAD.
 234          */
 235         do {
 236                 clear_thread_flag(TIF_SIGPENDING);
 237                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 238         } while (rc != -ECHILD);
 239
 240         /*
 241          * sys_wait4() above can't reap the EXIT_DEAD children but we do not
 242          * really care, we could reparent them to the global init. We could
 243          * exit and reap ->child_reaper even if it is not the last thread in
 244          * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
 245          * pid_ns can not go away until proc_kill_sb() drops the reference.
 246          *
 247          * But this ns can also have other tasks injected by setns()+fork().
 248          * Again, ignoring the user visible semantics we do not really need
 249          * to wait until they are all reaped, but they can be reparented to
 250          * us and thus we need to ensure that pid->child_reaper stays valid
 251          * until they all go away. See free_pid()->wake_up_process().
 252          *
 253          * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
 254          * if reparented.
 255          */
 256         for (;;) {
 257                 set_current_state(TASK_UNINTERRUPTIBLE);
 258                 if (pid_ns->nr_hashed == init_pids)
 259                         break;
 260                 schedule();
 261         }
 262         __set_current_state(TASK_RUNNING);
 263
 264         if (pid_ns->reboot)
 265                 current->signal->group_exit_code = pid_ns->reboot;
 266
 267         acct_exit_ns(pid_ns);
 268         return;
 269 }
 270
 271 #ifdef CONFIG_CHECKPOINT_RESTORE
 272 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 273                 void __user *buffer, size_t *lenp, loff_t *ppos)
 274 {
 275         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 276         struct ctl_table tmp = *table;
 277
 278         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 279                 return -EPERM;
 280
 281         /*
 282          * Writing directly to ns' last_pid field is OK, since this field
 283          * is volatile in a living namespace anyway and a code writing to
 284          * it should synchronize its usage with external means.
 285          */
 286
 287         tmp.data = &pid_ns->last_pid;
 288         return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 289 }
 290
 291 extern int pid_max;
 292 static int zero = 0;
 293 static struct ctl_table pid_ns_ctl_table[] = {
 294         {
 295                 .procname = "ns_last_pid",
 296                 .maxlen = sizeof(int),
 297                 .mode = 0666, /* permissions are checked in the handler */
 298                 .proc_handler = pid_ns_ctl_handler,
 299                 .extra1 = &zero,
 300                 .extra2 = &pid_max,
 301         },
 302         { }
 303 };
 304 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 305 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 306
 307 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 308 {
 309         if (pid_ns == &init_pid_ns)
 310                 return 0;
 311
 312         switch (cmd) {
 313         case LINUX_REBOOT_CMD_RESTART2:
 314         case LINUX_REBOOT_CMD_RESTART:
 315                 pid_ns->reboot = SIGHUP;
 316                 break;
 317
 318         case LINUX_REBOOT_CMD_POWER_OFF:
 319         case LINUX_REBOOT_CMD_HALT:
 320                 pid_ns->reboot = SIGINT;
 321                 break;
 322         default:
 323                 return -EINVAL;
 324         }
 325
 326         read_lock(&tasklist_lock);
 327         force_sig(SIGKILL, pid_ns->child_reaper);
 328         read_unlock(&tasklist_lock);
 329
 330         do_exit(0);
 331
 332         /* Not reached */
 333         return 0;
 334 }
 335
 336 static void *pidns_get(struct task_struct *task)
 337 {
 338         struct pid_namespace *ns;
 339
 340         rcu_read_lock();
 341         ns = task_active_pid_ns(task);
 342         if (ns)
 343                 get_pid_ns(ns);
 344         rcu_read_unlock();
 345
 346         return ns;
 347 }
 348
 349 static void pidns_put(void *ns)
 350 {
 351         put_pid_ns(ns);
 352 }
 353
 354 static int pidns_install(struct nsproxy *nsproxy, void *ns)
 355 {
 356         struct pid_namespace *active = task_active_pid_ns(current);
 357         struct pid_namespace *ancestor, *new = ns;
 358
 359         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 360             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 361                 return -EPERM;
 362
 363         /*
 364          * Only allow entering the current active pid namespace
 365          * or a child of the current active pid namespace.
 366          *
 367          * This is required for fork to return a usable pid value and
 368          * this maintains the property that processes and their
 369          * children can not escape their current pid namespace.
 370          */
 371         if (new->level < active->level)
 372                 return -EINVAL;
 373
 374         ancestor = new;
 375         while (ancestor->level > active->level)
 376                 ancestor = ancestor->parent;
 377         if (ancestor != active)
 378                 return -EINVAL;
 379
 380         put_pid_ns(nsproxy->pid_ns_for_children);
 381         nsproxy->pid_ns_for_children = get_pid_ns(new);
 382         return 0;
 383 }
 384
 385 static unsigned int pidns_inum(void *ns)
 386 {
 387         struct pid_namespace *pid_ns = ns;
 388         return pid_ns->proc_inum;
 389 }
 390
 391 const struct proc_ns_operations pidns_operations = {
 392         .name           = "pid",
 393         .type           = CLONE_NEWPID,
 394         .get            = pidns_get,
 395         .put            = pidns_put,
 396         .install        = pidns_install,
 397         .inum           = pidns_inum,
 398 };
 399
 400 static __init int pid_namespaces_init(void)
 401 {
 402         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 403
 404 #ifdef CONFIG_CHECKPOINT_RESTORE
 405         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 406 #endif
 407         return 0;
 408 }
 409
 410 __initcall(pid_namespaces_init);