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/cred.h>
  16 #include <linux/err.h>
  17 #include <linux/acct.h>
  18 #include <linux/slab.h>
  19 #include <linux/proc_ns.h>
  20 #include <linux/reboot.h>
  21 #include <linux/export.h>
  22 #include <linux/sched/task.h>
  23 #include <linux/sched/signal.h>
  24
  25 struct pid_cache {
  26         int nr_ids;
  27         char name[16];
  28         struct kmem_cache *cachep;
  29         struct list_head list;
  30 };
  31
  32 static LIST_HEAD(pid_caches_lh);
  33 static DEFINE_MUTEX(pid_caches_mutex);
  34 static struct kmem_cache *pid_ns_cachep;
  35
  36 /*
  37  * creates the kmem cache to allocate pids from.
  38  * @nr_ids: the number of numerical ids this pid will have to carry
  39  */
  40
  41 static struct kmem_cache *create_pid_cachep(int nr_ids)
  42 {
  43         struct pid_cache *pcache;
  44         struct kmem_cache *cachep;
  45
  46         mutex_lock(&pid_caches_mutex);
  47         list_for_each_entry(pcache, &pid_caches_lh, list)
  48                 if (pcache->nr_ids == nr_ids)
  49                         goto out;
  50
  51         pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
  52         if (pcache == NULL)
  53                 goto err_alloc;
  54
  55         snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
  56         cachep = kmem_cache_create(pcache->name,
  57                         sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
  58                         0, SLAB_HWCACHE_ALIGN, NULL);
  59         if (cachep == NULL)
  60                 goto err_cachep;
  61
  62         pcache->nr_ids = nr_ids;
  63         pcache->cachep = cachep;
  64         list_add(&pcache->list, &pid_caches_lh);
  65 out:
  66         mutex_unlock(&pid_caches_mutex);
  67         return pcache->cachep;
  68
  69 err_cachep:
  70         kfree(pcache);
  71 err_alloc:
  72         mutex_unlock(&pid_caches_mutex);
  73         return NULL;
  74 }
  75
  76 static void proc_cleanup_work(struct work_struct *work)
  77 {
  78         struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
  79         pid_ns_release_proc(ns);
  80 }
  81
  82 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
  83 #define MAX_PID_NS_LEVEL 32
  84
  85 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
  86 {
  87         return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
  88 }
  89
  90 static void dec_pid_namespaces(struct ucounts *ucounts)
  91 {
  92         dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
  93 }
  94
  95 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
  96         struct pid_namespace *parent_pid_ns)
  97 {
  98         struct pid_namespace *ns;
  99         unsigned int level = parent_pid_ns->level + 1;
 100         struct ucounts *ucounts;
 101         int i;
 102         int err;
 103
 104         err = -ENOSPC;
 105         if (level > MAX_PID_NS_LEVEL)
 106                 goto out;
 107         ucounts = inc_pid_namespaces(user_ns);
 108         if (!ucounts)
 109                 goto out;
 110
 111         err = -ENOMEM;
 112         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
 113         if (ns == NULL)
 114                 goto out_dec;
 115
 116         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 117         if (!ns->pidmap[0].page)
 118                 goto out_free;
 119
 120         ns->pid_cachep = create_pid_cachep(level + 1);
 121         if (ns->pid_cachep == NULL)
 122                 goto out_free_map;
 123
 124         err = ns_alloc_inum(&ns->ns);
 125         if (err)
 126                 goto out_free_map;
 127         ns->ns.ops = &pidns_operations;
 128
 129         kref_init(&ns->kref);
 130         ns->level = level;
 131         ns->parent = get_pid_ns(parent_pid_ns);
 132         ns->user_ns = get_user_ns(user_ns);
 133         ns->ucounts = ucounts;
 134         ns->nr_hashed = PIDNS_HASH_ADDING;
 135         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 136
 137         set_bit(0, ns->pidmap[0].page);
 138         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 139
 140         for (i = 1; i < PIDMAP_ENTRIES; i++)
 141                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 142
 143         return ns;
 144
 145 out_free_map:
 146         kfree(ns->pidmap[0].page);
 147 out_free:
 148         kmem_cache_free(pid_ns_cachep, ns);
 149 out_dec:
 150         dec_pid_namespaces(ucounts);
 151 out:
 152         return ERR_PTR(err);
 153 }
 154
 155 static void delayed_free_pidns(struct rcu_head *p)
 156 {
 157         struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
 158
 159         dec_pid_namespaces(ns->ucounts);
 160         put_user_ns(ns->user_ns);
 161
 162         kmem_cache_free(pid_ns_cachep, ns);
 163 }
 164
 165 static void destroy_pid_namespace(struct pid_namespace *ns)
 166 {
 167         int i;
 168
 169         ns_free_inum(&ns->ns);
 170         for (i = 0; i < PIDMAP_ENTRIES; i++)
 171                 kfree(ns->pidmap[i].page);
 172         call_rcu(&ns->rcu, delayed_free_pidns);
 173 }
 174
 175 struct pid_namespace *copy_pid_ns(unsigned long flags,
 176         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 177 {
 178         if (!(flags & CLONE_NEWPID))
 179                 return get_pid_ns(old_ns);
 180         if (task_active_pid_ns(current) != old_ns)
 181                 return ERR_PTR(-EINVAL);
 182         return create_pid_namespace(user_ns, old_ns);
 183 }
 184
 185 static void free_pid_ns(struct kref *kref)
 186 {
 187         struct pid_namespace *ns;
 188
 189         ns = container_of(kref, struct pid_namespace, kref);
 190         destroy_pid_namespace(ns);
 191 }
 192
 193 void put_pid_ns(struct pid_namespace *ns)
 194 {
 195         struct pid_namespace *parent;
 196
 197         while (ns != &init_pid_ns) {
 198                 parent = ns->parent;
 199                 if (!kref_put(&ns->kref, free_pid_ns))
 200                         break;
 201                 ns = parent;
 202         }
 203 }
 204 EXPORT_SYMBOL_GPL(put_pid_ns);
 205
 206 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 207 {
 208         int nr;
 209         int rc;
 210         struct task_struct *task, *me = current;
 211         int init_pids = thread_group_leader(me) ? 1 : 2;
 212
 213         /* Don't allow any more processes into the pid namespace */
 214         disable_pid_allocation(pid_ns);
 215
 216         /*
 217          * Ignore SIGCHLD causing any terminated children to autoreap.
 218          * This speeds up the namespace shutdown, plus see the comment
 219          * below.
 220          */
 221         spin_lock_irq(&me->sighand->siglock);
 222         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 223         spin_unlock_irq(&me->sighand->siglock);
 224
 225         /*
 226          * The last thread in the cgroup-init thread group is terminating.
 227          * Find remaining pid_ts in the namespace, signal and wait for them
 228          * to exit.
 229          *
 230          * Note:  This signals each threads in the namespace - even those that
 231          *        belong to the same thread group, To avoid this, we would have
 232          *        to walk the entire tasklist looking a processes in this
 233          *        namespace, but that could be unnecessarily expensive if the
 234          *        pid namespace has just a few processes. Or we need to
 235          *        maintain a tasklist for each pid namespace.
 236          *
 237          */
 238         read_lock(&tasklist_lock);
 239         nr = next_pidmap(pid_ns, 1);
 240         while (nr > 0) {
 241                 rcu_read_lock();
 242
 243                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 244                 if (task && !__fatal_signal_pending(task))
 245                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 246
 247                 rcu_read_unlock();
 248
 249                 nr = next_pidmap(pid_ns, nr);
 250         }
 251         read_unlock(&tasklist_lock);
 252
 253         /*
 254          * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
 255          * sys_wait4() will also block until our children traced from the
 256          * parent namespace are detached and become EXIT_DEAD.
 257          */
 258         do {
 259                 clear_thread_flag(TIF_SIGPENDING);
 260                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 261         } while (rc != -ECHILD);
 262
 263         /*
 264          * sys_wait4() above can't reap the EXIT_DEAD children but we do not
 265          * really care, we could reparent them to the global init. We could
 266          * exit and reap ->child_reaper even if it is not the last thread in
 267          * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
 268          * pid_ns can not go away until proc_kill_sb() drops the reference.
 269          *
 270          * But this ns can also have other tasks injected by setns()+fork().
 271          * Again, ignoring the user visible semantics we do not really need
 272          * to wait until they are all reaped, but they can be reparented to
 273          * us and thus we need to ensure that pid->child_reaper stays valid
 274          * until they all go away. See free_pid()->wake_up_process().
 275          *
 276          * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
 277          * if reparented.
 278          */
 279         for (;;) {
 280                 set_current_state(TASK_UNINTERRUPTIBLE);
 281                 if (pid_ns->nr_hashed == init_pids)
 282                         break;
 283                 schedule();
 284         }
 285         __set_current_state(TASK_RUNNING);
 286
 287         if (pid_ns->reboot)
 288                 current->signal->group_exit_code = pid_ns->reboot;
 289
 290         acct_exit_ns(pid_ns);
 291         return;
 292 }
 293
 294 #ifdef CONFIG_CHECKPOINT_RESTORE
 295 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 296                 void __user *buffer, size_t *lenp, loff_t *ppos)
 297 {
 298         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 299         struct ctl_table tmp = *table;
 300
 301         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 302                 return -EPERM;
 303
 304         /*
 305          * Writing directly to ns' last_pid field is OK, since this field
 306          * is volatile in a living namespace anyway and a code writing to
 307          * it should synchronize its usage with external means.
 308          */
 309
 310         tmp.data = &pid_ns->last_pid;
 311         return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 312 }
 313
 314 extern int pid_max;
 315 static int zero = 0;
 316 static struct ctl_table pid_ns_ctl_table[] = {
 317         {
 318                 .procname = "ns_last_pid",
 319                 .maxlen = sizeof(int),
 320                 .mode = 0666, /* permissions are checked in the handler */
 321                 .proc_handler = pid_ns_ctl_handler,
 322                 .extra1 = &zero,
 323                 .extra2 = &pid_max,
 324         },
 325         { }
 326 };
 327 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 328 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 329
 330 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 331 {
 332         if (pid_ns == &init_pid_ns)
 333                 return 0;
 334
 335         switch (cmd) {
 336         case LINUX_REBOOT_CMD_RESTART2:
 337         case LINUX_REBOOT_CMD_RESTART:
 338                 pid_ns->reboot = SIGHUP;
 339                 break;
 340
 341         case LINUX_REBOOT_CMD_POWER_OFF:
 342         case LINUX_REBOOT_CMD_HALT:
 343                 pid_ns->reboot = SIGINT;
 344                 break;
 345         default:
 346                 return -EINVAL;
 347         }
 348
 349         read_lock(&tasklist_lock);
 350         force_sig(SIGKILL, pid_ns->child_reaper);
 351         read_unlock(&tasklist_lock);
 352
 353         do_exit(0);
 354
 355         /* Not reached */
 356         return 0;
 357 }
 358
 359 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
 360 {
 361         return container_of(ns, struct pid_namespace, ns);
 362 }
 363
 364 static struct ns_common *pidns_get(struct task_struct *task)
 365 {
 366         struct pid_namespace *ns;
 367
 368         rcu_read_lock();
 369         ns = task_active_pid_ns(task);
 370         if (ns)
 371                 get_pid_ns(ns);
 372         rcu_read_unlock();
 373
 374         return ns ? &ns->ns : NULL;
 375 }
 376
 377 static void pidns_put(struct ns_common *ns)
 378 {
 379         put_pid_ns(to_pid_ns(ns));
 380 }
 381
 382 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
 383 {
 384         struct pid_namespace *active = task_active_pid_ns(current);
 385         struct pid_namespace *ancestor, *new = to_pid_ns(ns);
 386
 387         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 388             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 389                 return -EPERM;
 390
 391         /*
 392          * Only allow entering the current active pid namespace
 393          * or a child of the current active pid namespace.
 394          *
 395          * This is required for fork to return a usable pid value and
 396          * this maintains the property that processes and their
 397          * children can not escape their current pid namespace.
 398          */
 399         if (new->level < active->level)
 400                 return -EINVAL;
 401
 402         ancestor = new;
 403         while (ancestor->level > active->level)
 404                 ancestor = ancestor->parent;
 405         if (ancestor != active)
 406                 return -EINVAL;
 407
 408         put_pid_ns(nsproxy->pid_ns_for_children);
 409         nsproxy->pid_ns_for_children = get_pid_ns(new);
 410         return 0;
 411 }
 412
 413 static struct ns_common *pidns_get_parent(struct ns_common *ns)
 414 {
 415         struct pid_namespace *active = task_active_pid_ns(current);
 416         struct pid_namespace *pid_ns, *p;
 417
 418         /* See if the parent is in the current namespace */
 419         pid_ns = p = to_pid_ns(ns)->parent;
 420         for (;;) {
 421                 if (!p)
 422                         return ERR_PTR(-EPERM);
 423                 if (p == active)
 424                         break;
 425                 p = p->parent;
 426         }
 427
 428         return &get_pid_ns(pid_ns)->ns;
 429 }
 430
 431 static struct user_namespace *pidns_owner(struct ns_common *ns)
 432 {
 433         return to_pid_ns(ns)->user_ns;
 434 }
 435
 436 const struct proc_ns_operations pidns_operations = {
 437         .name           = "pid",
 438         .type           = CLONE_NEWPID,
 439         .get            = pidns_get,
 440         .put            = pidns_put,
 441         .install        = pidns_install,
 442         .owner          = pidns_owner,
 443         .get_parent     = pidns_get_parent,
 444 };
 445
 446 static __init int pid_namespaces_init(void)
 447 {
 448         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 449
 450 #ifdef CONFIG_CHECKPOINT_RESTORE
 451         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 452 #endif
 453         return 0;
 454 }
 455
 456 __initcall(pid_namespaces_init);