1 // SPDX-License-Identifier: GPL-2.0-only
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
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 #include <uapi/linux/wait.h>
27 #include "pid_sysctl.h"
29 static DEFINE_MUTEX(pid_caches_mutex);
30 static struct kmem_cache *pid_ns_cachep;
31 /* Write once array, filled from the beginning. */
32 static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
35 * creates the kmem cache to allocate pids from.
36 * @level: pid namespace level
39 static struct kmem_cache *create_pid_cachep(unsigned int level)
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];
51 snprintf(name, sizeof(name), "pid_%u", level + 1);
52 len = struct_size_t(struct pid, numbers, level + 1);
53 mutex_lock(&pid_caches_mutex);
54 /* Name collision forces to do allocation under mutex. */
56 *pkc = kmem_cache_create(name, len, 0,
57 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
58 mutex_unlock(&pid_caches_mutex);
59 /* current can fail, but someone else can succeed. */
60 return READ_ONCE(*pkc);
63 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
65 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
68 static void dec_pid_namespaces(struct ucounts *ucounts)
70 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
73 static void destroy_pid_namespace_work(struct work_struct *work);
75 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
76 struct pid_namespace *parent_pid_ns)
78 struct pid_namespace *ns;
79 unsigned int level = parent_pid_ns->level + 1;
80 struct ucounts *ucounts;
84 if (!in_userns(parent_pid_ns->user_ns, user_ns))
88 if (level > MAX_PID_NS_LEVEL)
90 ucounts = inc_pid_namespaces(user_ns);
95 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
101 ns->pid_cachep = create_pid_cachep(level);
102 if (ns->pid_cachep == NULL)
105 err = ns_alloc_inum(&ns->ns);
108 ns->ns.ops = &pidns_operations;
110 ns->pid_max = PID_MAX_LIMIT;
111 err = register_pidns_sysctls(ns);
115 refcount_set(&ns->ns.count, 1);
117 ns->parent = get_pid_ns(parent_pid_ns);
118 ns->user_ns = get_user_ns(user_ns);
119 ns->ucounts = ucounts;
120 ns->pid_allocated = PIDNS_ADDING;
121 INIT_WORK(&ns->work, destroy_pid_namespace_work);
123 #if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
124 ns->memfd_noexec_scope = pidns_memfd_noexec_scope(parent_pid_ns);
130 ns_free_inum(&ns->ns);
132 idr_destroy(&ns->idr);
133 kmem_cache_free(pid_ns_cachep, ns);
135 dec_pid_namespaces(ucounts);
140 static void delayed_free_pidns(struct rcu_head *p)
142 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
144 dec_pid_namespaces(ns->ucounts);
145 put_user_ns(ns->user_ns);
147 kmem_cache_free(pid_ns_cachep, ns);
150 static void destroy_pid_namespace(struct pid_namespace *ns)
152 unregister_pidns_sysctls(ns);
154 ns_free_inum(&ns->ns);
156 idr_destroy(&ns->idr);
157 call_rcu(&ns->rcu, delayed_free_pidns);
160 static void destroy_pid_namespace_work(struct work_struct *work)
162 struct pid_namespace *ns =
163 container_of(work, struct pid_namespace, work);
166 struct pid_namespace *parent;
169 destroy_pid_namespace(ns);
171 } while (ns != &init_pid_ns && refcount_dec_and_test(&ns->ns.count));
174 struct pid_namespace *copy_pid_ns(unsigned long flags,
175 struct user_namespace *user_ns, struct pid_namespace *old_ns)
177 if (!(flags & CLONE_NEWPID))
178 return get_pid_ns(old_ns);
179 if (task_active_pid_ns(current) != old_ns)
180 return ERR_PTR(-EINVAL);
181 return create_pid_namespace(user_ns, old_ns);
184 void put_pid_ns(struct pid_namespace *ns)
186 if (ns && ns != &init_pid_ns && refcount_dec_and_test(&ns->ns.count))
187 schedule_work(&ns->work);
189 EXPORT_SYMBOL_GPL(put_pid_ns);
191 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
195 struct task_struct *task, *me = current;
196 int init_pids = thread_group_leader(me) ? 1 : 2;
199 /* Don't allow any more processes into the pid namespace */
200 disable_pid_allocation(pid_ns);
203 * Ignore SIGCHLD causing any terminated children to autoreap.
204 * This speeds up the namespace shutdown, plus see the comment
207 spin_lock_irq(&me->sighand->siglock);
208 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
209 spin_unlock_irq(&me->sighand->siglock);
212 * The last thread in the cgroup-init thread group is terminating.
213 * Find remaining pid_ts in the namespace, signal and wait for them
216 * Note: This signals each threads in the namespace - even those that
217 * belong to the same thread group, To avoid this, we would have
218 * to walk the entire tasklist looking a processes in this
219 * namespace, but that could be unnecessarily expensive if the
220 * pid namespace has just a few processes. Or we need to
221 * maintain a tasklist for each pid namespace.
225 read_lock(&tasklist_lock);
227 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
228 task = pid_task(pid, PIDTYPE_PID);
229 if (task && !__fatal_signal_pending(task))
230 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
232 read_unlock(&tasklist_lock);
236 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
237 * kernel_wait4() will also block until our children traced from the
238 * parent namespace are detached and become EXIT_DEAD.
241 clear_thread_flag(TIF_SIGPENDING);
242 clear_thread_flag(TIF_NOTIFY_SIGNAL);
243 rc = kernel_wait4(-1, NULL, __WALL, NULL);
244 } while (rc != -ECHILD);
247 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
248 * process whose parents processes are outside of the pid
249 * namespace. Such processes are created with setns()+fork().
251 * If those EXIT_ZOMBIE processes are not reaped by their
252 * parents before their parents exit, they will be reparented
253 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
254 * stay valid until they all go away.
256 * The code relies on the pid_ns->child_reaper ignoring
257 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
258 * autoreaped if reparented.
260 * Semantically it is also desirable to wait for EXIT_ZOMBIE
261 * processes before allowing the child_reaper to be reaped, as
262 * that gives the invariant that when the init process of a
263 * pid namespace is reaped all of the processes in the pid
264 * namespace are gone.
266 * Once all of the other tasks are gone from the pid_namespace
267 * free_pid() will awaken this task.
270 set_current_state(TASK_INTERRUPTIBLE);
271 if (pid_ns->pid_allocated == init_pids)
275 __set_current_state(TASK_RUNNING);
278 current->signal->group_exit_code = pid_ns->reboot;
280 acct_exit_ns(pid_ns);
284 #ifdef CONFIG_CHECKPOINT_RESTORE
285 static int pid_ns_ctl_handler(const struct ctl_table *table, int write,
286 void *buffer, size_t *lenp, loff_t *ppos)
288 struct pid_namespace *pid_ns = task_active_pid_ns(current);
289 struct ctl_table tmp = *table;
292 if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
295 next = idr_get_cursor(&pid_ns->idr) - 1;
298 tmp.extra2 = &pid_ns->pid_max;
299 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
301 idr_set_cursor(&pid_ns->idr, next + 1);
306 static const struct ctl_table pid_ns_ctl_table[] = {
308 .procname = "ns_last_pid",
309 .maxlen = sizeof(int),
310 .mode = 0666, /* permissions are checked in the handler */
311 .proc_handler = pid_ns_ctl_handler,
312 .extra1 = SYSCTL_ZERO,
313 .extra2 = &init_pid_ns.pid_max,
316 #endif /* CONFIG_CHECKPOINT_RESTORE */
318 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
320 if (pid_ns == &init_pid_ns)
324 case LINUX_REBOOT_CMD_RESTART2:
325 case LINUX_REBOOT_CMD_RESTART:
326 pid_ns->reboot = SIGHUP;
329 case LINUX_REBOOT_CMD_POWER_OFF:
330 case LINUX_REBOOT_CMD_HALT:
331 pid_ns->reboot = SIGINT;
337 read_lock(&tasklist_lock);
338 send_sig(SIGKILL, pid_ns->child_reaper, 1);
339 read_unlock(&tasklist_lock);
347 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
349 return container_of(ns, struct pid_namespace, ns);
352 static struct ns_common *pidns_get(struct task_struct *task)
354 struct pid_namespace *ns;
357 ns = task_active_pid_ns(task);
362 return ns ? &ns->ns : NULL;
365 static struct ns_common *pidns_for_children_get(struct task_struct *task)
367 struct pid_namespace *ns = NULL;
371 ns = task->nsproxy->pid_ns_for_children;
377 read_lock(&tasklist_lock);
378 if (!ns->child_reaper) {
382 read_unlock(&tasklist_lock);
385 return ns ? &ns->ns : NULL;
388 static void pidns_put(struct ns_common *ns)
390 put_pid_ns(to_pid_ns(ns));
393 static int pidns_install(struct nsset *nsset, struct ns_common *ns)
395 struct nsproxy *nsproxy = nsset->nsproxy;
396 struct pid_namespace *active = task_active_pid_ns(current);
397 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
399 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
400 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
404 * Only allow entering the current active pid namespace
405 * or a child of the current active pid namespace.
407 * This is required for fork to return a usable pid value and
408 * this maintains the property that processes and their
409 * children can not escape their current pid namespace.
411 if (new->level < active->level)
415 while (ancestor->level > active->level)
416 ancestor = ancestor->parent;
417 if (ancestor != active)
420 put_pid_ns(nsproxy->pid_ns_for_children);
421 nsproxy->pid_ns_for_children = get_pid_ns(new);
425 static struct ns_common *pidns_get_parent(struct ns_common *ns)
427 struct pid_namespace *active = task_active_pid_ns(current);
428 struct pid_namespace *pid_ns, *p;
430 /* See if the parent is in the current namespace */
431 pid_ns = p = to_pid_ns(ns)->parent;
434 return ERR_PTR(-EPERM);
440 return &get_pid_ns(pid_ns)->ns;
443 static struct user_namespace *pidns_owner(struct ns_common *ns)
445 return to_pid_ns(ns)->user_ns;
448 const struct proc_ns_operations pidns_operations = {
450 .type = CLONE_NEWPID,
453 .install = pidns_install,
454 .owner = pidns_owner,
455 .get_parent = pidns_get_parent,
458 const struct proc_ns_operations pidns_for_children_operations = {
459 .name = "pid_for_children",
460 .real_ns_name = "pid",
461 .type = CLONE_NEWPID,
462 .get = pidns_for_children_get,
464 .install = pidns_install,
465 .owner = pidns_owner,
466 .get_parent = pidns_get_parent,
469 static __init int pid_namespaces_init(void)
471 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC | SLAB_ACCOUNT);
473 #ifdef CONFIG_CHECKPOINT_RESTORE
474 register_sysctl_init("kernel", pid_ns_ctl_table);
477 register_pid_ns_sysctl_table_vm();
481 __initcall(pid_namespaces_init);