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