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         kref_init(&ns->kref);
 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 static void free_pid_ns(struct kref *kref)
 152 {
 153         struct pid_namespace *ns;
 154
 155         ns = container_of(kref, struct pid_namespace, kref);
 156         destroy_pid_namespace(ns);
 157 }
 158
 159 void put_pid_ns(struct pid_namespace *ns)
 160 {
 161         struct pid_namespace *parent;
 162
 163         while (ns != &init_pid_ns) {
 164                 parent = ns->parent;
 165                 if (!kref_put(&ns->kref, free_pid_ns))
 166                         break;
 167                 ns = parent;
 168         }
 169 }
 170 EXPORT_SYMBOL_GPL(put_pid_ns);
 171
 172 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
 173 {
 174         int nr;
 175         int rc;
 176         struct task_struct *task, *me = current;
 177         int init_pids = thread_group_leader(me) ? 1 : 2;
 178         struct pid *pid;
 179
 180         /* Don't allow any more processes into the pid namespace */
 181         disable_pid_allocation(pid_ns);
 182
 183         /*
 184          * Ignore SIGCHLD causing any terminated children to autoreap.
 185          * This speeds up the namespace shutdown, plus see the comment
 186          * below.
 187          */
 188         spin_lock_irq(&me->sighand->siglock);
 189         me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
 190         spin_unlock_irq(&me->sighand->siglock);
 191
 192         /*
 193          * The last thread in the cgroup-init thread group is terminating.
 194          * Find remaining pid_ts in the namespace, signal and wait for them
 195          * to exit.
 196          *
 197          * Note:  This signals each threads in the namespace - even those that
 198          *        belong to the same thread group, To avoid this, we would have
 199          *        to walk the entire tasklist looking a processes in this
 200          *        namespace, but that could be unnecessarily expensive if the
 201          *        pid namespace has just a few processes. Or we need to
 202          *        maintain a tasklist for each pid namespace.
 203          *
 204          */
 205         rcu_read_lock();
 206         read_lock(&tasklist_lock);
 207         nr = 2;
 208         idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
 209                 task = pid_task(pid, PIDTYPE_PID);
 210                 if (task && !__fatal_signal_pending(task))
 211                         group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
 212         }
 213         read_unlock(&tasklist_lock);
 214         rcu_read_unlock();
 215
 216         /*
 217          * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
 218          * kernel_wait4() will also block until our children traced from the
 219          * parent namespace are detached and become EXIT_DEAD.
 220          */
 221         do {
 222                 clear_thread_flag(TIF_SIGPENDING);
 223                 rc = kernel_wait4(-1, NULL, __WALL, NULL);
 224         } while (rc != -ECHILD);
 225
 226         /*
 227          * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
 228          * process whose parents processes are outside of the pid
 229          * namespace.  Such processes are created with setns()+fork().
 230          *
 231          * If those EXIT_ZOMBIE processes are not reaped by their
 232          * parents before their parents exit, they will be reparented
 233          * to pid_ns->child_reaper.  Thus pidns->child_reaper needs to
 234          * stay valid until they all go away.
 235          *
 236          * The code relies on the the pid_ns->child_reaper ignoring
 237          * SIGCHILD to cause those EXIT_ZOMBIE processes to be
 238          * autoreaped if reparented.
 239          *
 240          * Semantically it is also desirable to wait for EXIT_ZOMBIE
 241          * processes before allowing the child_reaper to be reaped, as
 242          * that gives the invariant that when the init process of a
 243          * pid namespace is reaped all of the processes in the pid
 244          * namespace are gone.
 245          *
 246          * Once all of the other tasks are gone from the pid_namespace
 247          * free_pid() will awaken this task.
 248          */
 249         for (;;) {
 250                 set_current_state(TASK_INTERRUPTIBLE);
 251                 if (pid_ns->pid_allocated == init_pids)
 252                         break;
 253                 schedule();
 254         }
 255         __set_current_state(TASK_RUNNING);
 256
 257         if (pid_ns->reboot)
 258                 current->signal->group_exit_code = pid_ns->reboot;
 259
 260         acct_exit_ns(pid_ns);
 261         return;
 262 }
 263
 264 #ifdef CONFIG_CHECKPOINT_RESTORE
 265 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
 266                 void __user *buffer, size_t *lenp, loff_t *ppos)
 267 {
 268         struct pid_namespace *pid_ns = task_active_pid_ns(current);
 269         struct ctl_table tmp = *table;
 270         int ret, next;
 271
 272         if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
 273                 return -EPERM;
 274
 275         /*
 276          * Writing directly to ns' last_pid field is OK, since this field
 277          * is volatile in a living namespace anyway and a code writing to
 278          * it should synchronize its usage with external means.
 279          */
 280
 281         next = idr_get_cursor(&pid_ns->idr) - 1;
 282
 283         tmp.data = &next;
 284         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
 285         if (!ret && write)
 286                 idr_set_cursor(&pid_ns->idr, next + 1);
 287
 288         return ret;
 289 }
 290
 291 extern int pid_max;
 292 static struct ctl_table pid_ns_ctl_table[] = {
 293         {
 294                 .procname = "ns_last_pid",
 295                 .maxlen = sizeof(int),
 296                 .mode = 0666, /* permissions are checked in the handler */
 297                 .proc_handler = pid_ns_ctl_handler,
 298                 .extra1 = SYSCTL_ZERO,
 299                 .extra2 = &pid_max,
 300         },
 301         { }
 302 };
 303 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
 304 #endif  /* CONFIG_CHECKPOINT_RESTORE */
 305
 306 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
 307 {
 308         if (pid_ns == &init_pid_ns)
 309                 return 0;
 310
 311         switch (cmd) {
 312         case LINUX_REBOOT_CMD_RESTART2:
 313         case LINUX_REBOOT_CMD_RESTART:
 314                 pid_ns->reboot = SIGHUP;
 315                 break;
 316
 317         case LINUX_REBOOT_CMD_POWER_OFF:
 318         case LINUX_REBOOT_CMD_HALT:
 319                 pid_ns->reboot = SIGINT;
 320                 break;
 321         default:
 322                 return -EINVAL;
 323         }
 324
 325         read_lock(&tasklist_lock);
 326         send_sig(SIGKILL, pid_ns->child_reaper, 1);
 327         read_unlock(&tasklist_lock);
 328
 329         do_exit(0);
 330
 331         /* Not reached */
 332         return 0;
 333 }
 334
 335 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
 336 {
 337         return container_of(ns, struct pid_namespace, ns);
 338 }
 339
 340 static struct ns_common *pidns_get(struct task_struct *task)
 341 {
 342         struct pid_namespace *ns;
 343
 344         rcu_read_lock();
 345         ns = task_active_pid_ns(task);
 346         if (ns)
 347                 get_pid_ns(ns);
 348         rcu_read_unlock();
 349
 350         return ns ? &ns->ns : NULL;
 351 }
 352
 353 static struct ns_common *pidns_for_children_get(struct task_struct *task)
 354 {
 355         struct pid_namespace *ns = NULL;
 356
 357         task_lock(task);
 358         if (task->nsproxy) {
 359                 ns = task->nsproxy->pid_ns_for_children;
 360                 get_pid_ns(ns);
 361         }
 362         task_unlock(task);
 363
 364         if (ns) {
 365                 read_lock(&tasklist_lock);
 366                 if (!ns->child_reaper) {
 367                         put_pid_ns(ns);
 368                         ns = NULL;
 369                 }
 370                 read_unlock(&tasklist_lock);
 371         }
 372
 373         return ns ? &ns->ns : NULL;
 374 }
 375
 376 static void pidns_put(struct ns_common *ns)
 377 {
 378         put_pid_ns(to_pid_ns(ns));
 379 }
 380
 381 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
 382 {
 383         struct pid_namespace *active = task_active_pid_ns(current);
 384         struct pid_namespace *ancestor, *new = to_pid_ns(ns);
 385
 386         if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
 387             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
 388                 return -EPERM;
 389
 390         /*
 391          * Only allow entering the current active pid namespace
 392          * or a child of the current active pid namespace.
 393          *
 394          * This is required for fork to return a usable pid value and
 395          * this maintains the property that processes and their
 396          * children can not escape their current pid namespace.
 397          */
 398         if (new->level < active->level)
 399                 return -EINVAL;
 400
 401         ancestor = new;
 402         while (ancestor->level > active->level)
 403                 ancestor = ancestor->parent;
 404         if (ancestor != active)
 405                 return -EINVAL;
 406
 407         put_pid_ns(nsproxy->pid_ns_for_children);
 408         nsproxy->pid_ns_for_children = get_pid_ns(new);
 409         return 0;
 410 }
 411
 412 static struct ns_common *pidns_get_parent(struct ns_common *ns)
 413 {
 414         struct pid_namespace *active = task_active_pid_ns(current);
 415         struct pid_namespace *pid_ns, *p;
 416
 417         /* See if the parent is in the current namespace */
 418         pid_ns = p = to_pid_ns(ns)->parent;
 419         for (;;) {
 420                 if (!p)
 421                         return ERR_PTR(-EPERM);
 422                 if (p == active)
 423                         break;
 424                 p = p->parent;
 425         }
 426
 427         return &get_pid_ns(pid_ns)->ns;
 428 }
 429
 430 static struct user_namespace *pidns_owner(struct ns_common *ns)
 431 {
 432         return to_pid_ns(ns)->user_ns;
 433 }
 434
 435 const struct proc_ns_operations pidns_operations = {
 436         .name           = "pid",
 437         .type           = CLONE_NEWPID,
 438         .get            = pidns_get,
 439         .put            = pidns_put,
 440         .install        = pidns_install,
 441         .owner          = pidns_owner,
 442         .get_parent     = pidns_get_parent,
 443 };
 444
 445 const struct proc_ns_operations pidns_for_children_operations = {
 446         .name           = "pid_for_children",
 447         .real_ns_name   = "pid",
 448         .type           = CLONE_NEWPID,
 449         .get            = pidns_for_children_get,
 450         .put            = pidns_put,
 451         .install        = pidns_install,
 452         .owner          = pidns_owner,
 453         .get_parent     = pidns_get_parent,
 454 };
 455
 456 static __init int pid_namespaces_init(void)
 457 {
 458         pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
 459
 460 #ifdef CONFIG_CHECKPOINT_RESTORE
 461         register_sysctl_paths(kern_path, pid_ns_ctl_table);
 462 #endif
 463         return 0;
 464 }
 465
 466 __initcall(pid_namespaces_init);