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>
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];
33 * creates the kmem cache to allocate pids from.
34 * @level: pid namespace level
37 static struct kmem_cache *create_pid_cachep(unsigned int level)
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];
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. */
54 *pkc = kmem_cache_create(name, len, 0,
55 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, 0);
56 mutex_unlock(&pid_caches_mutex);
57 /* current can fail, but someone else can succeed. */
58 return READ_ONCE(*pkc);
61 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
63 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
66 static void dec_pid_namespaces(struct ucounts *ucounts)
68 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
71 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
72 struct pid_namespace *parent_pid_ns)
74 struct pid_namespace *ns;
75 unsigned int level = parent_pid_ns->level + 1;
76 struct ucounts *ucounts;
80 if (!in_userns(parent_pid_ns->user_ns, user_ns))
84 if (level > MAX_PID_NS_LEVEL)
86 ucounts = inc_pid_namespaces(user_ns);
91 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
97 ns->pid_cachep = create_pid_cachep(level);
98 if (ns->pid_cachep == NULL)
101 err = ns_alloc_inum(&ns->ns);
104 ns->ns.ops = &pidns_operations;
106 kref_init(&ns->kref);
108 ns->parent = get_pid_ns(parent_pid_ns);
109 ns->user_ns = get_user_ns(user_ns);
110 ns->ucounts = ucounts;
111 ns->pid_allocated = PIDNS_ADDING;
116 idr_destroy(&ns->idr);
117 kmem_cache_free(pid_ns_cachep, ns);
119 dec_pid_namespaces(ucounts);
124 static void delayed_free_pidns(struct rcu_head *p)
126 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
128 dec_pid_namespaces(ns->ucounts);
129 put_user_ns(ns->user_ns);
131 kmem_cache_free(pid_ns_cachep, ns);
134 static void destroy_pid_namespace(struct pid_namespace *ns)
136 ns_free_inum(&ns->ns);
138 idr_destroy(&ns->idr);
139 call_rcu(&ns->rcu, delayed_free_pidns);
142 struct pid_namespace *copy_pid_ns(unsigned long flags,
143 struct user_namespace *user_ns, struct pid_namespace *old_ns)
145 if (!(flags & CLONE_NEWPID))
146 return get_pid_ns(old_ns);
147 if (task_active_pid_ns(current) != old_ns)
148 return ERR_PTR(-EINVAL);
149 return create_pid_namespace(user_ns, old_ns);
152 static void free_pid_ns(struct kref *kref)
154 struct pid_namespace *ns;
156 ns = container_of(kref, struct pid_namespace, kref);
157 destroy_pid_namespace(ns);
160 void put_pid_ns(struct pid_namespace *ns)
162 struct pid_namespace *parent;
164 while (ns != &init_pid_ns) {
166 if (!kref_put(&ns->kref, free_pid_ns))
171 EXPORT_SYMBOL_GPL(put_pid_ns);
173 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
177 struct task_struct *task, *me = current;
178 int init_pids = thread_group_leader(me) ? 1 : 2;
181 /* Don't allow any more processes into the pid namespace */
182 disable_pid_allocation(pid_ns);
185 * Ignore SIGCHLD causing any terminated children to autoreap.
186 * This speeds up the namespace shutdown, plus see the comment
189 spin_lock_irq(&me->sighand->siglock);
190 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
191 spin_unlock_irq(&me->sighand->siglock);
194 * The last thread in the cgroup-init thread group is terminating.
195 * Find remaining pid_ts in the namespace, signal and wait for them
198 * Note: This signals each threads in the namespace - even those that
199 * belong to the same thread group, To avoid this, we would have
200 * to walk the entire tasklist looking a processes in this
201 * namespace, but that could be unnecessarily expensive if the
202 * pid namespace has just a few processes. Or we need to
203 * maintain a tasklist for each pid namespace.
207 read_lock(&tasklist_lock);
209 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
210 task = pid_task(pid, PIDTYPE_PID);
211 if (task && !__fatal_signal_pending(task))
212 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
214 read_unlock(&tasklist_lock);
218 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
219 * kernel_wait4() will also block until our children traced from the
220 * parent namespace are detached and become EXIT_DEAD.
223 clear_thread_flag(TIF_SIGPENDING);
224 rc = kernel_wait4(-1, NULL, __WALL, NULL);
225 } while (rc != -ECHILD);
228 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
229 * process whose parents processes are outside of the pid
230 * namespace. Such processes are created with setns()+fork().
232 * If those EXIT_ZOMBIE processes are not reaped by their
233 * parents before their parents exit, they will be reparented
234 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
235 * stay valid until they all go away.
237 * The code relies on the pid_ns->child_reaper ignoring
238 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
239 * autoreaped if reparented.
241 * Semantically it is also desirable to wait for EXIT_ZOMBIE
242 * processes before allowing the child_reaper to be reaped, as
243 * that gives the invariant that when the init process of a
244 * pid namespace is reaped all of the processes in the pid
245 * namespace are gone.
247 * Once all of the other tasks are gone from the pid_namespace
248 * free_pid() will awaken this task.
251 set_current_state(TASK_INTERRUPTIBLE);
252 if (pid_ns->pid_allocated == init_pids)
255 * Release tasks_rcu_exit_srcu to avoid following deadlock:
257 * 1) TASK A unshare(CLONE_NEWPID)
258 * 2) TASK A fork() twice -> TASK B (child reaper for new ns)
260 * 3) TASK B exits, kills TASK C, waits for TASK A to reap it
261 * 4) TASK A calls synchronize_rcu_tasks()
262 * -> synchronize_srcu(tasks_rcu_exit_srcu)
265 * It is considered safe to release tasks_rcu_exit_srcu here
266 * because we assume the current task can not be concurrently
267 * reaped at this point.
269 exit_tasks_rcu_stop();
271 exit_tasks_rcu_start();
273 __set_current_state(TASK_RUNNING);
276 current->signal->group_exit_code = pid_ns->reboot;
278 acct_exit_ns(pid_ns);
282 #ifdef CONFIG_CHECKPOINT_RESTORE
283 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
284 void *buffer, size_t *lenp, loff_t *ppos)
286 struct pid_namespace *pid_ns = task_active_pid_ns(current);
287 struct ctl_table tmp = *table;
290 if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
294 * Writing directly to ns' last_pid field is OK, since this field
295 * is volatile in a living namespace anyway and a code writing to
296 * it should synchronize its usage with external means.
299 next = idr_get_cursor(&pid_ns->idr) - 1;
302 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
304 idr_set_cursor(&pid_ns->idr, next + 1);
310 static struct ctl_table pid_ns_ctl_table[] = {
312 .procname = "ns_last_pid",
313 .maxlen = sizeof(int),
314 .mode = 0666, /* permissions are checked in the handler */
315 .proc_handler = pid_ns_ctl_handler,
316 .extra1 = SYSCTL_ZERO,
321 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
322 #endif /* CONFIG_CHECKPOINT_RESTORE */
324 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
326 if (pid_ns == &init_pid_ns)
330 case LINUX_REBOOT_CMD_RESTART2:
331 case LINUX_REBOOT_CMD_RESTART:
332 pid_ns->reboot = SIGHUP;
335 case LINUX_REBOOT_CMD_POWER_OFF:
336 case LINUX_REBOOT_CMD_HALT:
337 pid_ns->reboot = SIGINT;
343 read_lock(&tasklist_lock);
344 send_sig(SIGKILL, pid_ns->child_reaper, 1);
345 read_unlock(&tasklist_lock);
353 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
355 return container_of(ns, struct pid_namespace, ns);
358 static struct ns_common *pidns_get(struct task_struct *task)
360 struct pid_namespace *ns;
363 ns = task_active_pid_ns(task);
368 return ns ? &ns->ns : NULL;
371 static struct ns_common *pidns_for_children_get(struct task_struct *task)
373 struct pid_namespace *ns = NULL;
377 ns = task->nsproxy->pid_ns_for_children;
383 read_lock(&tasklist_lock);
384 if (!ns->child_reaper) {
388 read_unlock(&tasklist_lock);
391 return ns ? &ns->ns : NULL;
394 static void pidns_put(struct ns_common *ns)
396 put_pid_ns(to_pid_ns(ns));
399 static int pidns_install(struct nsset *nsset, struct ns_common *ns)
401 struct nsproxy *nsproxy = nsset->nsproxy;
402 struct pid_namespace *active = task_active_pid_ns(current);
403 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
405 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
406 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
410 * Only allow entering the current active pid namespace
411 * or a child of the current active pid namespace.
413 * This is required for fork to return a usable pid value and
414 * this maintains the property that processes and their
415 * children can not escape their current pid namespace.
417 if (new->level < active->level)
421 while (ancestor->level > active->level)
422 ancestor = ancestor->parent;
423 if (ancestor != active)
426 put_pid_ns(nsproxy->pid_ns_for_children);
427 nsproxy->pid_ns_for_children = get_pid_ns(new);
431 static struct ns_common *pidns_get_parent(struct ns_common *ns)
433 struct pid_namespace *active = task_active_pid_ns(current);
434 struct pid_namespace *pid_ns, *p;
436 /* See if the parent is in the current namespace */
437 pid_ns = p = to_pid_ns(ns)->parent;
440 return ERR_PTR(-EPERM);
446 return &get_pid_ns(pid_ns)->ns;
449 static struct user_namespace *pidns_owner(struct ns_common *ns)
451 return to_pid_ns(ns)->user_ns;
454 const struct proc_ns_operations pidns_operations = {
456 .type = CLONE_NEWPID,
459 .install = pidns_install,
460 .owner = pidns_owner,
461 .get_parent = pidns_get_parent,
464 const struct proc_ns_operations pidns_for_children_operations = {
465 .name = "pid_for_children",
466 .real_ns_name = "pid",
467 .type = CLONE_NEWPID,
468 .get = pidns_for_children_get,
470 .install = pidns_install,
471 .owner = pidns_owner,
472 .get_parent = pidns_get_parent,
475 static __init int pid_namespaces_init(void)
477 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
479 #ifdef CONFIG_CHECKPOINT_RESTORE
480 register_sysctl_paths(kern_path, pid_ns_ctl_table);
485 __initcall(pid_namespaces_init);