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 struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
74 struct pid_namespace *parent_pid_ns)
76 struct pid_namespace *ns;
77 unsigned int level = parent_pid_ns->level + 1;
78 struct ucounts *ucounts;
82 if (!in_userns(parent_pid_ns->user_ns, user_ns))
86 if (level > MAX_PID_NS_LEVEL)
88 ucounts = inc_pid_namespaces(user_ns);
93 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
99 ns->pid_cachep = create_pid_cachep(level);
100 if (ns->pid_cachep == NULL)
103 err = ns_alloc_inum(&ns->ns);
106 ns->ns.ops = &pidns_operations;
108 refcount_set(&ns->ns.count, 1);
110 ns->parent = get_pid_ns(parent_pid_ns);
111 ns->user_ns = get_user_ns(user_ns);
112 ns->ucounts = ucounts;
113 ns->pid_allocated = PIDNS_ADDING;
114 #if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
115 ns->memfd_noexec_scope = pidns_memfd_noexec_scope(parent_pid_ns);
120 idr_destroy(&ns->idr);
121 kmem_cache_free(pid_ns_cachep, ns);
123 dec_pid_namespaces(ucounts);
128 static void delayed_free_pidns(struct rcu_head *p)
130 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
132 dec_pid_namespaces(ns->ucounts);
133 put_user_ns(ns->user_ns);
135 kmem_cache_free(pid_ns_cachep, ns);
138 static void destroy_pid_namespace(struct pid_namespace *ns)
140 ns_free_inum(&ns->ns);
142 idr_destroy(&ns->idr);
143 call_rcu(&ns->rcu, delayed_free_pidns);
146 struct pid_namespace *copy_pid_ns(unsigned long flags,
147 struct user_namespace *user_ns, struct pid_namespace *old_ns)
149 if (!(flags & CLONE_NEWPID))
150 return get_pid_ns(old_ns);
151 if (task_active_pid_ns(current) != old_ns)
152 return ERR_PTR(-EINVAL);
153 return create_pid_namespace(user_ns, old_ns);
156 void put_pid_ns(struct pid_namespace *ns)
158 struct pid_namespace *parent;
160 while (ns != &init_pid_ns) {
162 if (!refcount_dec_and_test(&ns->ns.count))
164 destroy_pid_namespace(ns);
168 EXPORT_SYMBOL_GPL(put_pid_ns);
170 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
174 struct task_struct *task, *me = current;
175 int init_pids = thread_group_leader(me) ? 1 : 2;
178 /* Don't allow any more processes into the pid namespace */
179 disable_pid_allocation(pid_ns);
182 * Ignore SIGCHLD causing any terminated children to autoreap.
183 * This speeds up the namespace shutdown, plus see the comment
186 spin_lock_irq(&me->sighand->siglock);
187 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
188 spin_unlock_irq(&me->sighand->siglock);
191 * The last thread in the cgroup-init thread group is terminating.
192 * Find remaining pid_ts in the namespace, signal and wait for them
195 * Note: This signals each threads in the namespace - even those that
196 * belong to the same thread group, To avoid this, we would have
197 * to walk the entire tasklist looking a processes in this
198 * namespace, but that could be unnecessarily expensive if the
199 * pid namespace has just a few processes. Or we need to
200 * maintain a tasklist for each pid namespace.
204 read_lock(&tasklist_lock);
206 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
207 task = pid_task(pid, PIDTYPE_PID);
208 if (task && !__fatal_signal_pending(task))
209 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
211 read_unlock(&tasklist_lock);
215 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
216 * kernel_wait4() will also block until our children traced from the
217 * parent namespace are detached and become EXIT_DEAD.
220 clear_thread_flag(TIF_SIGPENDING);
221 rc = kernel_wait4(-1, NULL, __WALL, NULL);
222 } while (rc != -ECHILD);
225 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
226 * process whose parents processes are outside of the pid
227 * namespace. Such processes are created with setns()+fork().
229 * If those EXIT_ZOMBIE processes are not reaped by their
230 * parents before their parents exit, they will be reparented
231 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
232 * stay valid until they all go away.
234 * The code relies on the pid_ns->child_reaper ignoring
235 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
236 * autoreaped if reparented.
238 * Semantically it is also desirable to wait for EXIT_ZOMBIE
239 * processes before allowing the child_reaper to be reaped, as
240 * that gives the invariant that when the init process of a
241 * pid namespace is reaped all of the processes in the pid
242 * namespace are gone.
244 * Once all of the other tasks are gone from the pid_namespace
245 * free_pid() will awaken this task.
248 set_current_state(TASK_INTERRUPTIBLE);
249 if (pid_ns->pid_allocated == init_pids)
252 * Release tasks_rcu_exit_srcu to avoid following deadlock:
254 * 1) TASK A unshare(CLONE_NEWPID)
255 * 2) TASK A fork() twice -> TASK B (child reaper for new ns)
257 * 3) TASK B exits, kills TASK C, waits for TASK A to reap it
258 * 4) TASK A calls synchronize_rcu_tasks()
259 * -> synchronize_srcu(tasks_rcu_exit_srcu)
262 * It is considered safe to release tasks_rcu_exit_srcu here
263 * because we assume the current task can not be concurrently
264 * reaped at this point.
266 exit_tasks_rcu_stop();
268 exit_tasks_rcu_start();
270 __set_current_state(TASK_RUNNING);
273 current->signal->group_exit_code = pid_ns->reboot;
275 acct_exit_ns(pid_ns);
279 #ifdef CONFIG_CHECKPOINT_RESTORE
280 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
281 void *buffer, size_t *lenp, loff_t *ppos)
283 struct pid_namespace *pid_ns = task_active_pid_ns(current);
284 struct ctl_table tmp = *table;
287 if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
290 next = idr_get_cursor(&pid_ns->idr) - 1;
293 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
295 idr_set_cursor(&pid_ns->idr, next + 1);
301 static struct ctl_table pid_ns_ctl_table[] = {
303 .procname = "ns_last_pid",
304 .maxlen = sizeof(int),
305 .mode = 0666, /* permissions are checked in the handler */
306 .proc_handler = pid_ns_ctl_handler,
307 .extra1 = SYSCTL_ZERO,
312 #endif /* CONFIG_CHECKPOINT_RESTORE */
314 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
316 if (pid_ns == &init_pid_ns)
320 case LINUX_REBOOT_CMD_RESTART2:
321 case LINUX_REBOOT_CMD_RESTART:
322 pid_ns->reboot = SIGHUP;
325 case LINUX_REBOOT_CMD_POWER_OFF:
326 case LINUX_REBOOT_CMD_HALT:
327 pid_ns->reboot = SIGINT;
333 read_lock(&tasklist_lock);
334 send_sig(SIGKILL, pid_ns->child_reaper, 1);
335 read_unlock(&tasklist_lock);
343 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
345 return container_of(ns, struct pid_namespace, ns);
348 static struct ns_common *pidns_get(struct task_struct *task)
350 struct pid_namespace *ns;
353 ns = task_active_pid_ns(task);
358 return ns ? &ns->ns : NULL;
361 static struct ns_common *pidns_for_children_get(struct task_struct *task)
363 struct pid_namespace *ns = NULL;
367 ns = task->nsproxy->pid_ns_for_children;
373 read_lock(&tasklist_lock);
374 if (!ns->child_reaper) {
378 read_unlock(&tasklist_lock);
381 return ns ? &ns->ns : NULL;
384 static void pidns_put(struct ns_common *ns)
386 put_pid_ns(to_pid_ns(ns));
389 static int pidns_install(struct nsset *nsset, struct ns_common *ns)
391 struct nsproxy *nsproxy = nsset->nsproxy;
392 struct pid_namespace *active = task_active_pid_ns(current);
393 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
395 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
396 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
400 * Only allow entering the current active pid namespace
401 * or a child of the current active pid namespace.
403 * This is required for fork to return a usable pid value and
404 * this maintains the property that processes and their
405 * children can not escape their current pid namespace.
407 if (new->level < active->level)
411 while (ancestor->level > active->level)
412 ancestor = ancestor->parent;
413 if (ancestor != active)
416 put_pid_ns(nsproxy->pid_ns_for_children);
417 nsproxy->pid_ns_for_children = get_pid_ns(new);
421 static struct ns_common *pidns_get_parent(struct ns_common *ns)
423 struct pid_namespace *active = task_active_pid_ns(current);
424 struct pid_namespace *pid_ns, *p;
426 /* See if the parent is in the current namespace */
427 pid_ns = p = to_pid_ns(ns)->parent;
430 return ERR_PTR(-EPERM);
436 return &get_pid_ns(pid_ns)->ns;
439 static struct user_namespace *pidns_owner(struct ns_common *ns)
441 return to_pid_ns(ns)->user_ns;
444 const struct proc_ns_operations pidns_operations = {
446 .type = CLONE_NEWPID,
449 .install = pidns_install,
450 .owner = pidns_owner,
451 .get_parent = pidns_get_parent,
454 const struct proc_ns_operations pidns_for_children_operations = {
455 .name = "pid_for_children",
456 .real_ns_name = "pid",
457 .type = CLONE_NEWPID,
458 .get = pidns_for_children_get,
460 .install = pidns_install,
461 .owner = pidns_owner,
462 .get_parent = pidns_get_parent,
465 static __init int pid_namespaces_init(void)
467 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC | SLAB_ACCOUNT);
469 #ifdef CONFIG_CHECKPOINT_RESTORE
470 register_sysctl_init("kernel", pid_ns_ctl_table);
473 register_pid_ns_sysctl_table_vm();
477 __initcall(pid_namespaces_init);