kernel/pid_namespace.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Pid namespaces
   4  *
   5  * Authors:
   6  *    (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
   7  *    (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
   8  *     Many thanks to Oleg Nesterov for comments and help
   9  *
  10  */
  11
  12 #include <linux/pid.h>
  13 #include <linux/pid_namespace.h>
  14 #include <linux/user_namespace.h>
  15 #include <linux/syscalls.h>
  16 #include <linux/cred.h>
  17 #include <linux/err.h>
  18 #include <linux/acct.h>
  19 #include <linux/slab.h>
  20 #include <linux/proc_ns.h>
  21 #include <linux/reboot.h>
  22 #include <linux/export.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/signal.h>
  25 #include <linux/idr.h>
  26
  27 static DEFINE_MUTEX(pid_caches_mutex);
  28 static struct kmem_cache *pid_ns_cachep;
  29 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
  30 #define MAX_PID_NS_LEVEL 32
  31 /* Write once array, filled from the beginning. */
  32 static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
  33
  34 /*
  35  * creates the kmem cache to allocate pids from.
  36  * @level: pid namespace level
  37  */
  38
  39 static struct kmem_cache *create_pid_cachep(unsigned int level)
  40 {
  41         /* Level 0 is init_pid_ns.pid_cachep */
  42         struct kmem_cache **pkc = &pid_cache[level - 1];
  43         struct kmem_cache *kc;
  44         char name[4 + 10 + 1];
  45         unsigned int len;
  46
  47         kc = READ_ONCE(*pkc);
  48         if (kc)
  49                 return kc;
  50
  51         snprintf(name, sizeof(name), "pid_%u", level + 1);
  52         len = sizeof(struct pid) + level * sizeof(struct upid);
  53         mutex_lock(&pid_caches_mutex);
  54         /* Name collision forces to do allocation under mutex. */
  55         if (!*pkc)
  56                 *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
  57         mutex_unlock(&pid_caches_mutex);
  58         /* current can fail, but someone else can succeed. */
  59         return READ_ONCE(*pkc);
  60 }
  61
  62 static void proc_cleanup_work(struct work_struct *work)
  63 {
  64         struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
  65         pid_ns_release_proc(ns);
  66 }
  67
  68 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
  69 {
  70         return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
  71 }
  72
  73 static void dec_pid_namespaces(struct ucounts *ucounts)
  74 {
  75         dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
  76 }
  77
  78 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
  79         struct pid_namespace *parent_pid_ns)
  80 {
  81         struct pid_namespace *ns;
  82         unsigned int level = parent_pid_ns->level + 1;
  83         struct ucounts *ucounts;
  84         int err;
  85
  86         err = -EINVAL;
  87         if (!in_userns(parent_pid_ns->user_ns, user_ns))
  88                 goto out;
  89
  90         err = -ENOSPC;
  91         if (level > MAX_PID_NS_LEVEL)
  92                 goto out;
  93         ucounts = inc_pid_namespaces(user_ns);
  94         if (!ucounts)
  95                 goto out;
  96
  97         err = -ENOMEM;
  98         ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
  99         if (ns == NULL)
 100                 goto out_dec;
 101
 102         idr_init(&ns->idr);
 103
 104         ns->pid_cachep = create_pid_cachep(level);
 105         if (ns->pid_cachep == NULL)
 106                 goto out_free_idr;
 107
 108         err = ns_alloc_inum(&ns->ns);
 109         if (err)
 110                 goto out_free_idr;
 111         ns->ns.ops = &pidns_operations;
 112
 113         kref_init(&ns->kref);
 114         ns->level = level;
 115         ns->parent = get_pid_ns(parent_pid_ns);
 116         ns->user_ns = get_user_ns(user_ns);
 117         ns->ucounts = ucounts;
 118         ns->pid_allocated = PIDNS_ADDING;
 119         INIT_WORK(&ns->proc_work, proc_cleanup_work);
 120
 121         return ns;
 122
 123 out_free_idr:
 124         idr_destroy(&ns->idr);
 125         kmem_cache_free(pid_ns_cachep, ns);
 126 out_dec:
 127         dec_pid_namespaces(ucounts);
 128 out:
 129         return ERR_PTR(err);
 130 }
 131
 132 static void delayed_free_pidns(struct rcu_head *p)
 133 {
 134         struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
 135
 136         dec_pid_namespaces(ns->ucounts);
 137         put_user_ns(ns->user_ns);
 138
 139         kmem_cache_free(pid_ns_cachep, ns);
 140 }
 141
 142 static void destroy_pid_namespace(struct pid_namespace *ns)
 143 {
 144         ns_free_inum(&ns->ns);
 145
 146         idr_destroy(&ns->idr);
 147         call_rcu(&ns->rcu, delayed_free_pidns);
 148 }
 149
 150 struct pid_namespace *copy_pid_ns(unsigned long flags,
 151         struct user_namespace *user_ns, struct pid_namespace *old_ns)
 152 {
 153         if (!(flags & CLONE_NEWPID))
 154                 return get_pid_ns(old_ns);
 155         if (task_active_pid_ns(current) != old_ns)
 156                 return ERR_PTR(-EINVAL);
 157         return create_pid_namespace(user_ns, old_ns);
 158 }
 159
 160 static void free_pid_ns(struct kref *kref)
 161 {
 162         struct pid_namespace *ns;
 163
 164         ns = container_of(kref, struct pid_namespace, kref);
 165         destroy_pid_namespace(ns);
 166 }
 167
 168 void put_pid_ns(struct pid_namespace *ns)
 169 {
 170         struct pid_namespace *parent;
 171
 172         while (ns != &init_pid_ns) {
 173                 parent = ns->parent;
 174                 if (!kref_put(&ns->kref, free_pid_ns))
 175                         break;
 176                 ns = parent;
 177         }
 178 }
 179 EXPORT_SYMBOL_GPL(put_pid_ns);
 180
 181 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 182 {
 183         int nr;
 184         int rc;
 185         struct task_struct *task, *me = current;
 186         int init_pids = thread_group_leader(me) ? 1 : 2;
 187         struct pid *pid;
 188
 189         /* Don't allow any more processes into the pid namespace */
 190         disable_pid_allocation(pid_ns);
 191
 192         /*
 193          * Ignore SIGCHLD causing any terminated children to autoreap.
 194          * This speeds up the namespace shutdown, plus see the comment
 195          * below.
 196          */
 197         spin_lock_irq(&me->sighand->siglock);
 198         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 199         spin_unlock_irq(&me->sighand->siglock);
 200
 201         /*
 202          * The last thread in the cgroup-init thread group is terminating.
 203          * Find remaining pid_ts in the namespace, signal and wait for them
 204          * to exit.
 205          *
 206          * Note:  This signals each threads in the namespace - even those that
 207          *        belong to the same thread group, To avoid this, we would have
 208          *        to walk the entire tasklist looking a processes in this
 209          *        namespace, but that could be unnecessarily expensive if the
 210          *        pid namespace has just a few processes. Or we need to
 211          *        maintain a tasklist for each pid namespace.
 212          *
 213          */
 214         rcu_read_lock();
 215         read_lock(&tasklist_lock);
 216         nr = 2;
 217         idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
 218                 task = pid_task(pid, PIDTYPE_PID);
 219                 if (task && !__fatal_signal_pending(task))
 220                         group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
 221         }
 222         read_unlock(&tasklist_lock);
 223         rcu_read_unlock();
 224
 225         /*
 226          * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
 227          * kernel_wait4() will also block until our children traced from the
 228          * parent namespace are detached and become EXIT_DEAD.
 229          */
 230         do {
 231                 clear_thread_flag(TIF_SIGPENDING);
 232                 rc = kernel_wait4(-1, NULL, __WALL, NULL);
 233         } while (rc != -ECHILD);
 234
 235         /*
 236          * kernel_wait4() above can't reap the EXIT_DEAD children but we do not
 237          * really care, we could reparent them to the global init. We could
 238          * exit and reap ->child_reaper even if it is not the last thread in
 239          * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(),
 240          * pid_ns can not go away until proc_kill_sb() drops the reference.
 241          *
 242          * But this ns can also have other tasks injected by setns()+fork().
 243          * Again, ignoring the user visible semantics we do not really need
 244          * to wait until they are all reaped, but they can be reparented to
 245          * us and thus we need to ensure that pid->child_reaper stays valid
 246          * until they all go away. See free_pid()->wake_up_process().
 247          *
 248          * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
 249          * if reparented.
 250          */
 251         for (;;) {
 252                 set_current_state(TASK_INTERRUPTIBLE);
 253                 if (pid_ns->pid_allocated == init_pids)
 254                         break;
 255                 schedule();
 256         }
 257         __set_current_state(TASK_RUNNING);
 258
 259         if (pid_ns->reboot)
 260                 current->signal->group_exit_code = pid_ns->reboot;
 261
 262         acct_exit_ns(pid_ns);
 263         return;
 264 }
 265
 266 #ifdef CONFIG_CHECKPOINT_RESTORE
 267 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 268                 void __user *buffer, size_t *lenp, loff_t *ppos)
 269 {
 270         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 271         struct ctl_table tmp = *table;
 272         int ret, next;
 273
 274         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 275                 return -EPERM;
 276
 277         /*
 278          * Writing directly to ns' last_pid field is OK, since this field
 279          * is volatile in a living namespace anyway and a code writing to
 280          * it should synchronize its usage with external means.
 281          */
 282
 283         next = idr_get_cursor(&pid_ns->idr) - 1;
 284
 285         tmp.data = &next;
 286         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 287         if (!ret && write)
 288                 idr_set_cursor(&pid_ns->idr, next + 1);
 289
 290         return ret;
 291 }
 292
 293 extern int pid_max;
 294 static int zero = 0;
 295 static struct ctl_table pid_ns_ctl_table[] = {
 296         {
 297                 .procname = "ns_last_pid",
 298                 .maxlen = sizeof(int),
 299                 .mode = 0666, /* permissions are checked in the handler */
 300                 .proc_handler = pid_ns_ctl_handler,
 301                 .extra1 = &zero,
 302                 .extra2 = &pid_max,
 303         },
 304         { }
 305 };
 306 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 307 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 308
 309 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 310 {
 311         if (pid_ns == &init_pid_ns)
 312                 return 0;
 313
 314         switch (cmd) {
 315         case LINUX_REBOOT_CMD_RESTART2:
 316         case LINUX_REBOOT_CMD_RESTART:
 317                 pid_ns->reboot = SIGHUP;
 318                 break;
 319
 320         case LINUX_REBOOT_CMD_POWER_OFF:
 321         case LINUX_REBOOT_CMD_HALT:
 322                 pid_ns->reboot = SIGINT;
 323                 break;
 324         default:
 325                 return -EINVAL;
 326         }
 327
 328         read_lock(&tasklist_lock);
 329         force_sig(SIGKILL, pid_ns->child_reaper);
 330         read_unlock(&tasklist_lock);
 331
 332         do_exit(0);
 333
 334         /* Not reached */
 335         return 0;
 336 }
 337
 338 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
 339 {
 340         return container_of(ns, struct pid_namespace, ns);
 341 }
 342
 343 static struct ns_common *pidns_get(struct task_struct *task)
 344 {
 345         struct pid_namespace *ns;
 346
 347         rcu_read_lock();
 348         ns = task_active_pid_ns(task);
 349         if (ns)
 350                 get_pid_ns(ns);
 351         rcu_read_unlock();
 352
 353         return ns ? &ns->ns : NULL;
 354 }
 355
 356 static struct ns_common *pidns_for_children_get(struct task_struct *task)
 357 {
 358         struct pid_namespace *ns = NULL;
 359
 360         task_lock(task);
 361         if (task->nsproxy) {
 362                 ns = task->nsproxy->pid_ns_for_children;
 363                 get_pid_ns(ns);
 364         }
 365         task_unlock(task);
 366
 367         if (ns) {
 368                 read_lock(&tasklist_lock);
 369                 if (!ns->child_reaper) {
 370                         put_pid_ns(ns);
 371                         ns = NULL;
 372                 }
 373                 read_unlock(&tasklist_lock);
 374         }
 375
 376         return ns ? &ns->ns : NULL;
 377 }
 378
 379 static void pidns_put(struct ns_common *ns)
 380 {
 381         put_pid_ns(to_pid_ns(ns));
 382 }
 383
 384 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
 385 {
 386         struct pid_namespace *active = task_active_pid_ns(current);
 387         struct pid_namespace *ancestor, *new = to_pid_ns(ns);
 388
 389         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 390             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 391                 return -EPERM;
 392
 393         /*
 394          * Only allow entering the current active pid namespace
 395          * or a child of the current active pid namespace.
 396          *
 397          * This is required for fork to return a usable pid value and
 398          * this maintains the property that processes and their
 399          * children can not escape their current pid namespace.
 400          */
 401         if (new->level < active->level)
 402                 return -EINVAL;
 403
 404         ancestor = new;
 405         while (ancestor->level > active->level)
 406                 ancestor = ancestor->parent;
 407         if (ancestor != active)
 408                 return -EINVAL;
 409
 410         put_pid_ns(nsproxy->pid_ns_for_children);
 411         nsproxy->pid_ns_for_children = get_pid_ns(new);
 412         return 0;
 413 }
 414
 415 static struct ns_common *pidns_get_parent(struct ns_common *ns)
 416 {
 417         struct pid_namespace *active = task_active_pid_ns(current);
 418         struct pid_namespace *pid_ns, *p;
 419
 420         /* See if the parent is in the current namespace */
 421         pid_ns = p = to_pid_ns(ns)->parent;
 422         for (;;) {
 423                 if (!p)
 424                         return ERR_PTR(-EPERM);
 425                 if (p == active)
 426                         break;
 427                 p = p->parent;
 428         }
 429
 430         return &get_pid_ns(pid_ns)->ns;
 431 }
 432
 433 static struct user_namespace *pidns_owner(struct ns_common *ns)
 434 {
 435         return to_pid_ns(ns)->user_ns;
 436 }
 437
 438 const struct proc_ns_operations pidns_operations = {
 439         .name           = "pid",
 440         .type           = CLONE_NEWPID,
 441         .get            = pidns_get,
 442         .put            = pidns_put,
 443         .install        = pidns_install,
 444         .owner          = pidns_owner,
 445         .get_parent     = pidns_get_parent,
 446 };
 447
 448 const struct proc_ns_operations pidns_for_children_operations = {
 449         .name           = "pid_for_children",
 450         .real_ns_name   = "pid",
 451         .type           = CLONE_NEWPID,
 452         .get            = pidns_for_children_get,
 453         .put            = pidns_put,
 454         .install        = pidns_install,
 455         .owner          = pidns_owner,
 456         .get_parent     = pidns_get_parent,
 457 };
 458
 459 static __init int pid_namespaces_init(void)
 460 {
 461         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 462
 463 #ifdef CONFIG_CHECKPOINT_RESTORE
 464         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 465 #endif
 466         return 0;
 467 }
 468
 469 __initcall(pid_namespaces_init);