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 = -EINVAL;
 105         if (!in_userns(parent_pid_ns->user_ns, user_ns))
 106                 goto out;
 107
 108         err = -ENOSPC;
 109         if (level > MAX_PID_NS_LEVEL)
 110                 goto out;
 111         ucounts = inc_pid_namespaces(user_ns);
 112         if (!ucounts)
 113                 goto out;
 114
 115         err = -ENOMEM;
 116         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
 117         if (ns == NULL)
 118                 goto out_dec;
 119
 120         ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 121         if (!ns->pidmap[0].page)
 122                 goto out_free;
 123
 124         ns->pid_cachep = create_pid_cachep(level + 1);
 125         if (ns->pid_cachep == NULL)
 126                 goto out_free_map;
 127
 128         err = ns_alloc_inum(&ns->ns);
 129         if (err)
 130                 goto out_free_map;
 131         ns->ns.ops = &pidns_operations;
 132
 133         kref_init(&ns->kref);
 134         ns->level = level;
 135         ns->parent = get_pid_ns(parent_pid_ns);
 136         ns->user_ns = get_user_ns(user_ns);
 137         ns->ucounts = ucounts;
 138         ns->nr_hashed = PIDNS_HASH_ADDING;
 139         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 140
 141         set_bit(0, ns->pidmap[0].page);
 142         atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
 143
 144         for (i = 1; i < PIDMAP_ENTRIES; i++)
 145                 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
 146
 147         return ns;
 148
 149 out_free_map:
 150         kfree(ns->pidmap[0].page);
 151 out_free:
 152         kmem_cache_free(pid_ns_cachep, ns);
 153 out_dec:
 154         dec_pid_namespaces(ucounts);
 155 out:
 156         return ERR_PTR(err);
 157 }
 158
 159 static void delayed_free_pidns(struct rcu_head *p)
 160 {
 161         struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
 162
 163         dec_pid_namespaces(ns->ucounts);
 164         put_user_ns(ns->user_ns);
 165
 166         kmem_cache_free(pid_ns_cachep, ns);
 167 }
 168
 169 static void destroy_pid_namespace(struct pid_namespace *ns)
 170 {
 171         int i;
 172
 173         ns_free_inum(&ns->ns);
 174         for (i = 0; i < PIDMAP_ENTRIES; i++)
 175                 kfree(ns->pidmap[i].page);
 176         call_rcu(&ns->rcu, delayed_free_pidns);
 177 }
 178
 179 struct pid_namespace *copy_pid_ns(unsigned long flags,
 180         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 181 {
 182         if (!(flags & CLONE_NEWPID))
 183                 return get_pid_ns(old_ns);
 184         if (task_active_pid_ns(current) != old_ns)
 185                 return ERR_PTR(-EINVAL);
 186         return create_pid_namespace(user_ns, old_ns);
 187 }
 188
 189 static void free_pid_ns(struct kref *kref)
 190 {
 191         struct pid_namespace *ns;
 192
 193         ns = container_of(kref, struct pid_namespace, kref);
 194         destroy_pid_namespace(ns);
 195 }
 196
 197 void put_pid_ns(struct pid_namespace *ns)
 198 {
 199         struct pid_namespace *parent;
 200
 201         while (ns != &init_pid_ns) {
 202                 parent = ns->parent;
 203                 if (!kref_put(&ns->kref, free_pid_ns))
 204                         break;
 205                 ns = parent;
 206         }
 207 }
 208 EXPORT_SYMBOL_GPL(put_pid_ns);
 209
 210 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 211 {
 212         int nr;
 213         int rc;
 214         struct task_struct *task, *me = current;
 215         int init_pids = thread_group_leader(me) ? 1 : 2;
 216
 217         /* Don't allow any more processes into the pid namespace */
 218         disable_pid_allocation(pid_ns);
 219
 220         /*
 221          * Ignore SIGCHLD causing any terminated children to autoreap.
 222          * This speeds up the namespace shutdown, plus see the comment
 223          * below.
 224          */
 225         spin_lock_irq(&me->sighand->siglock);
 226         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 227         spin_unlock_irq(&me->sighand->siglock);
 228
 229         /*
 230          * The last thread in the cgroup-init thread group is terminating.
 231          * Find remaining pid_ts in the namespace, signal and wait for them
 232          * to exit.
 233          *
 234          * Note:  This signals each threads in the namespace - even those that
 235          *        belong to the same thread group, To avoid this, we would have
 236          *        to walk the entire tasklist looking a processes in this
 237          *        namespace, but that could be unnecessarily expensive if the
 238          *        pid namespace has just a few processes. Or we need to
 239          *        maintain a tasklist for each pid namespace.
 240          *
 241          */
 242         read_lock(&tasklist_lock);
 243         nr = next_pidmap(pid_ns, 1);
 244         while (nr > 0) {
 245                 rcu_read_lock();
 246
 247                 task = pid_task(find_vpid(nr), PIDTYPE_PID);
 248                 if (task && !__fatal_signal_pending(task))
 249                         send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
 250
 251                 rcu_read_unlock();
 252
 253                 nr = next_pidmap(pid_ns, nr);
 254         }
 255         read_unlock(&tasklist_lock);
 256
 257         /*
 258          * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
 259          * sys_wait4() will also block until our children traced from the
 260          * parent namespace are detached and become EXIT_DEAD.
 261          */
 262         do {
 263                 clear_thread_flag(TIF_SIGPENDING);
 264                 rc = sys_wait4(-1, NULL, __WALL, NULL);
 265         } while (rc != -ECHILD);
 266
 267         /*
 268          * sys_wait4() above can't reap the EXIT_DEAD children but we do not
 269          * really care, we could reparent them to the global init. We could
 270          * exit and reap ->child_reaper even if it is not the last thread in
 271          * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
 272          * pid_ns can not go away until proc_kill_sb() drops the reference.
 273          *
 274          * But this ns can also have other tasks injected by setns()+fork().
 275          * Again, ignoring the user visible semantics we do not really need
 276          * to wait until they are all reaped, but they can be reparented to
 277          * us and thus we need to ensure that pid->child_reaper stays valid
 278          * until they all go away. See free_pid()->wake_up_process().
 279          *
 280          * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
 281          * if reparented.
 282          */
 283         for (;;) {
 284                 set_current_state(TASK_INTERRUPTIBLE);
 285                 if (pid_ns->nr_hashed == init_pids)
 286                         break;
 287                 schedule();
 288         }
 289         __set_current_state(TASK_RUNNING);
 290
 291         if (pid_ns->reboot)
 292                 current->signal->group_exit_code = pid_ns->reboot;
 293
 294         acct_exit_ns(pid_ns);
 295         return;
 296 }
 297
 298 #ifdef CONFIG_CHECKPOINT_RESTORE
 299 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 300                 void __user *buffer, size_t *lenp, loff_t *ppos)
 301 {
 302         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 303         struct ctl_table tmp = *table;
 304
 305         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 306                 return -EPERM;
 307
 308         /*
 309          * Writing directly to ns' last_pid field is OK, since this field
 310          * is volatile in a living namespace anyway and a code writing to
 311          * it should synchronize its usage with external means.
 312          */
 313
 314         tmp.data = &pid_ns->last_pid;
 315         return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 316 }
 317
 318 extern int pid_max;
 319 static int zero = 0;
 320 static struct ctl_table pid_ns_ctl_table[] = {
 321         {
 322                 .procname = "ns_last_pid",
 323                 .maxlen = sizeof(int),
 324                 .mode = 0666, /* permissions are checked in the handler */
 325                 .proc_handler = pid_ns_ctl_handler,
 326                 .extra1 = &zero,
 327                 .extra2 = &pid_max,
 328         },
 329         { }
 330 };
 331 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 332 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 333
 334 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 335 {
 336         if (pid_ns == &init_pid_ns)
 337                 return 0;
 338
 339         switch (cmd) {
 340         case LINUX_REBOOT_CMD_RESTART2:
 341         case LINUX_REBOOT_CMD_RESTART:
 342                 pid_ns->reboot = SIGHUP;
 343                 break;
 344
 345         case LINUX_REBOOT_CMD_POWER_OFF:
 346         case LINUX_REBOOT_CMD_HALT:
 347                 pid_ns->reboot = SIGINT;
 348                 break;
 349         default:
 350                 return -EINVAL;
 351         }
 352
 353         read_lock(&tasklist_lock);
 354         force_sig(SIGKILL, pid_ns->child_reaper);
 355         read_unlock(&tasklist_lock);
 356
 357         do_exit(0);
 358
 359         /* Not reached */
 360         return 0;
 361 }
 362
 363 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
 364 {
 365         return container_of(ns, struct pid_namespace, ns);
 366 }
 367
 368 static struct ns_common *pidns_get(struct task_struct *task)
 369 {
 370         struct pid_namespace *ns;
 371
 372         rcu_read_lock();
 373         ns = task_active_pid_ns(task);
 374         if (ns)
 375                 get_pid_ns(ns);
 376         rcu_read_unlock();
 377
 378         return ns ? &ns->ns : NULL;
 379 }
 380
 381 static struct ns_common *pidns_for_children_get(struct task_struct *task)
 382 {
 383         struct pid_namespace *ns = NULL;
 384
 385         task_lock(task);
 386         if (task->nsproxy) {
 387                 ns = task->nsproxy->pid_ns_for_children;
 388                 get_pid_ns(ns);
 389         }
 390         task_unlock(task);
 391
 392         if (ns) {
 393                 read_lock(&tasklist_lock);
 394                 if (!ns->child_reaper) {
 395                         put_pid_ns(ns);
 396                         ns = NULL;
 397                 }
 398                 read_unlock(&tasklist_lock);
 399         }
 400
 401         return ns ? &ns->ns : NULL;
 402 }
 403
 404 static void pidns_put(struct ns_common *ns)
 405 {
 406         put_pid_ns(to_pid_ns(ns));
 407 }
 408
 409 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
 410 {
 411         struct pid_namespace *active = task_active_pid_ns(current);
 412         struct pid_namespace *ancestor, *new = to_pid_ns(ns);
 413
 414         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 415             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 416                 return -EPERM;
 417
 418         /*
 419          * Only allow entering the current active pid namespace
 420          * or a child of the current active pid namespace.
 421          *
 422          * This is required for fork to return a usable pid value and
 423          * this maintains the property that processes and their
 424          * children can not escape their current pid namespace.
 425          */
 426         if (new->level < active->level)
 427                 return -EINVAL;
 428
 429         ancestor = new;
 430         while (ancestor->level > active->level)
 431                 ancestor = ancestor->parent;
 432         if (ancestor != active)
 433                 return -EINVAL;
 434
 435         put_pid_ns(nsproxy->pid_ns_for_children);
 436         nsproxy->pid_ns_for_children = get_pid_ns(new);
 437         return 0;
 438 }
 439
 440 static struct ns_common *pidns_get_parent(struct ns_common *ns)
 441 {
 442         struct pid_namespace *active = task_active_pid_ns(current);
 443         struct pid_namespace *pid_ns, *p;
 444
 445         /* See if the parent is in the current namespace */
 446         pid_ns = p = to_pid_ns(ns)->parent;
 447         for (;;) {
 448                 if (!p)
 449                         return ERR_PTR(-EPERM);
 450                 if (p == active)
 451                         break;
 452                 p = p->parent;
 453         }
 454
 455         return &get_pid_ns(pid_ns)->ns;
 456 }
 457
 458 static struct user_namespace *pidns_owner(struct ns_common *ns)
 459 {
 460         return to_pid_ns(ns)->user_ns;
 461 }
 462
 463 const struct proc_ns_operations pidns_operations = {
 464         .name           = "pid",
 465         .type           = CLONE_NEWPID,
 466         .get            = pidns_get,
 467         .put            = pidns_put,
 468         .install        = pidns_install,
 469         .owner          = pidns_owner,
 470         .get_parent     = pidns_get_parent,
 471 };
 472
 473 const struct proc_ns_operations pidns_for_children_operations = {
 474         .name           = "pid_for_children",
 475         .real_ns_name   = "pid",
 476         .type           = CLONE_NEWPID,
 477         .get            = pidns_for_children_get,
 478         .put            = pidns_put,
 479         .install        = pidns_install,
 480         .owner          = pidns_owner,
 481         .get_parent     = pidns_get_parent,
 482 };
 483
 484 static __init int pid_namespaces_init(void)
 485 {
 486         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 487
 488 #ifdef CONFIG_CHECKPOINT_RESTORE
 489         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 490 #endif
 491         return 0;
 492 }
 493
 494 __initcall(pid_namespaces_init);