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 "pid_sysctl.h"
28 static DEFINE_MUTEX(pid_caches_mutex);
29 static struct kmem_cache *pid_ns_cachep;
30 /* Write once array, filled from the beginning. */
31 static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
34 * creates the kmem cache to allocate pids from.
35 * @level: pid namespace level
38 static struct kmem_cache *create_pid_cachep(unsigned int level)
40 /* Level 0 is init_pid_ns.pid_cachep */
41 struct kmem_cache **pkc = &pid_cache[level - 1];
42 struct kmem_cache *kc;
43 char name[4 + 10 + 1];
50 snprintf(name, sizeof(name), "pid_%u", level + 1);
51 len = struct_size_t(struct pid, numbers, level + 1);
52 mutex_lock(&pid_caches_mutex);
53 /* Name collision forces to do allocation under mutex. */
55 *pkc = kmem_cache_create(name, len, 0,
56 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
57 mutex_unlock(&pid_caches_mutex);
58 /* current can fail, but someone else can succeed. */
59 return READ_ONCE(*pkc);
62 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
64 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
67 static void dec_pid_namespaces(struct ucounts *ucounts)
69 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
72 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
73 struct pid_namespace *parent_pid_ns)
75 struct pid_namespace *ns;
76 unsigned int level = parent_pid_ns->level + 1;
77 struct ucounts *ucounts;
81 if (!in_userns(parent_pid_ns->user_ns, user_ns))
85 if (level > MAX_PID_NS_LEVEL)
87 ucounts = inc_pid_namespaces(user_ns);
92 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
98 ns->pid_cachep = create_pid_cachep(level);
99 if (ns->pid_cachep == NULL)
102 err = ns_alloc_inum(&ns->ns);
105 ns->ns.ops = &pidns_operations;
107 refcount_set(&ns->ns.count, 1);
109 ns->parent = get_pid_ns(parent_pid_ns);
110 ns->user_ns = get_user_ns(user_ns);
111 ns->ucounts = ucounts;
112 ns->pid_allocated = PIDNS_ADDING;
114 initialize_memfd_noexec_scope(ns);
119 idr_destroy(&ns->idr);
120 kmem_cache_free(pid_ns_cachep, ns);
122 dec_pid_namespaces(ucounts);
127 static void delayed_free_pidns(struct rcu_head *p)
129 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
131 dec_pid_namespaces(ns->ucounts);
132 put_user_ns(ns->user_ns);
134 kmem_cache_free(pid_ns_cachep, ns);
137 static void destroy_pid_namespace(struct pid_namespace *ns)
139 ns_free_inum(&ns->ns);
141 idr_destroy(&ns->idr);
142 call_rcu(&ns->rcu, delayed_free_pidns);
145 struct pid_namespace *copy_pid_ns(unsigned long flags,
146 struct user_namespace *user_ns, struct pid_namespace *old_ns)
148 if (!(flags & CLONE_NEWPID))
149 return get_pid_ns(old_ns);
150 if (task_active_pid_ns(current) != old_ns)
151 return ERR_PTR(-EINVAL);
152 return create_pid_namespace(user_ns, old_ns);
155 void put_pid_ns(struct pid_namespace *ns)
157 struct pid_namespace *parent;
159 while (ns != &init_pid_ns) {
161 if (!refcount_dec_and_test(&ns->ns.count))
163 destroy_pid_namespace(ns);
167 EXPORT_SYMBOL_GPL(put_pid_ns);
169 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
173 struct task_struct *task, *me = current;
174 int init_pids = thread_group_leader(me) ? 1 : 2;
177 /* Don't allow any more processes into the pid namespace */
178 disable_pid_allocation(pid_ns);
181 * Ignore SIGCHLD causing any terminated children to autoreap.
182 * This speeds up the namespace shutdown, plus see the comment
185 spin_lock_irq(&me->sighand->siglock);
186 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
187 spin_unlock_irq(&me->sighand->siglock);
190 * The last thread in the cgroup-init thread group is terminating.
191 * Find remaining pid_ts in the namespace, signal and wait for them
194 * Note: This signals each threads in the namespace - even those that
195 * belong to the same thread group, To avoid this, we would have
196 * to walk the entire tasklist looking a processes in this
197 * namespace, but that could be unnecessarily expensive if the
198 * pid namespace has just a few processes. Or we need to
199 * maintain a tasklist for each pid namespace.
203 read_lock(&tasklist_lock);
205 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
206 task = pid_task(pid, PIDTYPE_PID);
207 if (task && !__fatal_signal_pending(task))
208 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
210 read_unlock(&tasklist_lock);
214 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
215 * kernel_wait4() will also block until our children traced from the
216 * parent namespace are detached and become EXIT_DEAD.
219 clear_thread_flag(TIF_SIGPENDING);
220 rc = kernel_wait4(-1, NULL, __WALL, NULL);
221 } while (rc != -ECHILD);
224 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
225 * process whose parents processes are outside of the pid
226 * namespace. Such processes are created with setns()+fork().
228 * If those EXIT_ZOMBIE processes are not reaped by their
229 * parents before their parents exit, they will be reparented
230 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
231 * stay valid until they all go away.
233 * The code relies on the pid_ns->child_reaper ignoring
234 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
235 * autoreaped if reparented.
237 * Semantically it is also desirable to wait for EXIT_ZOMBIE
238 * processes before allowing the child_reaper to be reaped, as
239 * that gives the invariant that when the init process of a
240 * pid namespace is reaped all of the processes in the pid
241 * namespace are gone.
243 * Once all of the other tasks are gone from the pid_namespace
244 * free_pid() will awaken this task.
247 set_current_state(TASK_INTERRUPTIBLE);
248 if (pid_ns->pid_allocated == init_pids)
251 * Release tasks_rcu_exit_srcu to avoid following deadlock:
253 * 1) TASK A unshare(CLONE_NEWPID)
254 * 2) TASK A fork() twice -> TASK B (child reaper for new ns)
256 * 3) TASK B exits, kills TASK C, waits for TASK A to reap it
257 * 4) TASK A calls synchronize_rcu_tasks()
258 * -> synchronize_srcu(tasks_rcu_exit_srcu)
261 * It is considered safe to release tasks_rcu_exit_srcu here
262 * because we assume the current task can not be concurrently
263 * reaped at this point.
265 exit_tasks_rcu_stop();
267 exit_tasks_rcu_start();
269 __set_current_state(TASK_RUNNING);
272 current->signal->group_exit_code = pid_ns->reboot;
274 acct_exit_ns(pid_ns);
278 #ifdef CONFIG_CHECKPOINT_RESTORE
279 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
280 void *buffer, size_t *lenp, loff_t *ppos)
282 struct pid_namespace *pid_ns = task_active_pid_ns(current);
283 struct ctl_table tmp = *table;
286 if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
290 * Writing directly to ns' last_pid field is OK, since this field
291 * is volatile in a living namespace anyway and a code writing to
292 * it should synchronize its usage with external means.
295 next = idr_get_cursor(&pid_ns->idr) - 1;
298 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
300 idr_set_cursor(&pid_ns->idr, next + 1);
306 static 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,
317 #endif /* CONFIG_CHECKPOINT_RESTORE */
319 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
321 if (pid_ns == &init_pid_ns)
325 case LINUX_REBOOT_CMD_RESTART2:
326 case LINUX_REBOOT_CMD_RESTART:
327 pid_ns->reboot = SIGHUP;
330 case LINUX_REBOOT_CMD_POWER_OFF:
331 case LINUX_REBOOT_CMD_HALT:
332 pid_ns->reboot = SIGINT;
338 read_lock(&tasklist_lock);
339 send_sig(SIGKILL, pid_ns->child_reaper, 1);
340 read_unlock(&tasklist_lock);
348 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
350 return container_of(ns, struct pid_namespace, ns);
353 static struct ns_common *pidns_get(struct task_struct *task)
355 struct pid_namespace *ns;
358 ns = task_active_pid_ns(task);
363 return ns ? &ns->ns : NULL;
366 static struct ns_common *pidns_for_children_get(struct task_struct *task)
368 struct pid_namespace *ns = NULL;
372 ns = task->nsproxy->pid_ns_for_children;
378 read_lock(&tasklist_lock);
379 if (!ns->child_reaper) {
383 read_unlock(&tasklist_lock);
386 return ns ? &ns->ns : NULL;
389 static void pidns_put(struct ns_common *ns)
391 put_pid_ns(to_pid_ns(ns));
394 static int pidns_install(struct nsset *nsset, struct ns_common *ns)
396 struct nsproxy *nsproxy = nsset->nsproxy;
397 struct pid_namespace *active = task_active_pid_ns(current);
398 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
400 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
401 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
405 * Only allow entering the current active pid namespace
406 * or a child of the current active pid namespace.
408 * This is required for fork to return a usable pid value and
409 * this maintains the property that processes and their
410 * children can not escape their current pid namespace.
412 if (new->level < active->level)
416 while (ancestor->level > active->level)
417 ancestor = ancestor->parent;
418 if (ancestor != active)
421 put_pid_ns(nsproxy->pid_ns_for_children);
422 nsproxy->pid_ns_for_children = get_pid_ns(new);
426 static struct ns_common *pidns_get_parent(struct ns_common *ns)
428 struct pid_namespace *active = task_active_pid_ns(current);
429 struct pid_namespace *pid_ns, *p;
431 /* See if the parent is in the current namespace */
432 pid_ns = p = to_pid_ns(ns)->parent;
435 return ERR_PTR(-EPERM);
441 return &get_pid_ns(pid_ns)->ns;
444 static struct user_namespace *pidns_owner(struct ns_common *ns)
446 return to_pid_ns(ns)->user_ns;
449 const struct proc_ns_operations pidns_operations = {
451 .type = CLONE_NEWPID,
454 .install = pidns_install,
455 .owner = pidns_owner,
456 .get_parent = pidns_get_parent,
459 const struct proc_ns_operations pidns_for_children_operations = {
460 .name = "pid_for_children",
461 .real_ns_name = "pid",
462 .type = CLONE_NEWPID,
463 .get = pidns_for_children_get,
465 .install = pidns_install,
466 .owner = pidns_owner,
467 .get_parent = pidns_get_parent,
470 static __init int pid_namespaces_init(void)
472 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC | SLAB_ACCOUNT);
474 #ifdef CONFIG_CHECKPOINT_RESTORE
475 register_sysctl_init("kernel", pid_ns_ctl_table);
478 register_pid_ns_sysctl_table_vm();
482 __initcall(pid_namespaces_init);