5 * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
6 * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
7 * Many thanks to Oleg Nesterov for comments and help
11 #include <linux/pid.h>
12 #include <linux/pid_namespace.h>
13 #include <linux/user_namespace.h>
14 #include <linux/syscalls.h>
15 #include <linux/cred.h>
16 #include <linux/err.h>
17 #include <linux/acct.h>
18 #include <linux/slab.h>
19 #include <linux/proc_ns.h>
20 #include <linux/reboot.h>
21 #include <linux/export.h>
22 #include <linux/sched/task.h>
23 #include <linux/sched/signal.h>
24 #include <linux/idr.h>
26 static DEFINE_MUTEX(pid_caches_mutex);
27 static struct kmem_cache *pid_ns_cachep;
28 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
29 #define MAX_PID_NS_LEVEL 32
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 = sizeof(struct pid) + level * sizeof(struct upid);
52 mutex_lock(&pid_caches_mutex);
53 /* Name collision forces to do allocation under mutex. */
55 *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
56 mutex_unlock(&pid_caches_mutex);
57 /* current can fail, but someone else can succeed. */
58 return READ_ONCE(*pkc);
61 static void proc_cleanup_work(struct work_struct *work)
63 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
64 pid_ns_release_proc(ns);
67 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
69 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
72 static void dec_pid_namespaces(struct ucounts *ucounts)
74 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
77 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
78 struct pid_namespace *parent_pid_ns)
80 struct pid_namespace *ns;
81 unsigned int level = parent_pid_ns->level + 1;
82 struct ucounts *ucounts;
86 if (!in_userns(parent_pid_ns->user_ns, user_ns))
90 if (level > MAX_PID_NS_LEVEL)
92 ucounts = inc_pid_namespaces(user_ns);
97 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
103 ns->pid_cachep = create_pid_cachep(level);
104 if (ns->pid_cachep == NULL)
107 err = ns_alloc_inum(&ns->ns);
110 ns->ns.ops = &pidns_operations;
112 kref_init(&ns->kref);
114 ns->parent = get_pid_ns(parent_pid_ns);
115 ns->user_ns = get_user_ns(user_ns);
116 ns->ucounts = ucounts;
117 ns->pid_allocated = PIDNS_ADDING;
118 INIT_WORK(&ns->proc_work, proc_cleanup_work);
123 idr_destroy(&ns->idr);
124 kmem_cache_free(pid_ns_cachep, ns);
126 dec_pid_namespaces(ucounts);
131 static void delayed_free_pidns(struct rcu_head *p)
133 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
135 dec_pid_namespaces(ns->ucounts);
136 put_user_ns(ns->user_ns);
138 kmem_cache_free(pid_ns_cachep, ns);
141 static void destroy_pid_namespace(struct pid_namespace *ns)
143 ns_free_inum(&ns->ns);
145 idr_destroy(&ns->idr);
146 call_rcu(&ns->rcu, delayed_free_pidns);
149 struct pid_namespace *copy_pid_ns(unsigned long flags,
150 struct user_namespace *user_ns, struct pid_namespace *old_ns)
152 if (!(flags & CLONE_NEWPID))
153 return get_pid_ns(old_ns);
154 if (task_active_pid_ns(current) != old_ns)
155 return ERR_PTR(-EINVAL);
156 return create_pid_namespace(user_ns, old_ns);
159 static void free_pid_ns(struct kref *kref)
161 struct pid_namespace *ns;
163 ns = container_of(kref, struct pid_namespace, kref);
164 destroy_pid_namespace(ns);
167 void put_pid_ns(struct pid_namespace *ns)
169 struct pid_namespace *parent;
171 while (ns != &init_pid_ns) {
173 if (!kref_put(&ns->kref, free_pid_ns))
178 EXPORT_SYMBOL_GPL(put_pid_ns);
180 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
184 struct task_struct *task, *me = current;
185 int init_pids = thread_group_leader(me) ? 1 : 2;
188 /* Don't allow any more processes into the pid namespace */
189 disable_pid_allocation(pid_ns);
192 * Ignore SIGCHLD causing any terminated children to autoreap.
193 * This speeds up the namespace shutdown, plus see the comment
196 spin_lock_irq(&me->sighand->siglock);
197 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
198 spin_unlock_irq(&me->sighand->siglock);
201 * The last thread in the cgroup-init thread group is terminating.
202 * Find remaining pid_ts in the namespace, signal and wait for them
205 * Note: This signals each threads in the namespace - even those that
206 * belong to the same thread group, To avoid this, we would have
207 * to walk the entire tasklist looking a processes in this
208 * namespace, but that could be unnecessarily expensive if the
209 * pid namespace has just a few processes. Or we need to
210 * maintain a tasklist for each pid namespace.
214 read_lock(&tasklist_lock);
216 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
217 task = pid_task(pid, PIDTYPE_PID);
218 if (task && !__fatal_signal_pending(task))
219 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
221 read_unlock(&tasklist_lock);
225 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
226 * kernel_wait4() will also block until our children traced from the
227 * parent namespace are detached and become EXIT_DEAD.
230 clear_thread_flag(TIF_SIGPENDING);
231 rc = kernel_wait4(-1, NULL, __WALL, NULL);
232 } while (rc != -ECHILD);
235 * kernel_wait4() above can't reap the EXIT_DEAD children but we do not
236 * really care, we could reparent them to the global init. We could
237 * exit and reap ->child_reaper even if it is not the last thread in
238 * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(),
239 * pid_ns can not go away until proc_kill_sb() drops the reference.
241 * But this ns can also have other tasks injected by setns()+fork().
242 * Again, ignoring the user visible semantics we do not really need
243 * to wait until they are all reaped, but they can be reparented to
244 * us and thus we need to ensure that pid->child_reaper stays valid
245 * until they all go away. See free_pid()->wake_up_process().
247 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
251 set_current_state(TASK_INTERRUPTIBLE);
252 if (pid_ns->pid_allocated == init_pids)
256 __set_current_state(TASK_RUNNING);
259 current->signal->group_exit_code = pid_ns->reboot;
261 acct_exit_ns(pid_ns);
265 #ifdef CONFIG_CHECKPOINT_RESTORE
266 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
267 void __user *buffer, size_t *lenp, loff_t *ppos)
269 struct pid_namespace *pid_ns = task_active_pid_ns(current);
270 struct ctl_table tmp = *table;
273 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
277 * Writing directly to ns' last_pid field is OK, since this field
278 * is volatile in a living namespace anyway and a code writing to
279 * it should synchronize its usage with external means.
282 next = idr_get_cursor(&pid_ns->idr) - 1;
285 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
287 idr_set_cursor(&pid_ns->idr, next + 1);
294 static struct ctl_table pid_ns_ctl_table[] = {
296 .procname = "ns_last_pid",
297 .maxlen = sizeof(int),
298 .mode = 0666, /* permissions are checked in the handler */
299 .proc_handler = pid_ns_ctl_handler,
305 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
306 #endif /* CONFIG_CHECKPOINT_RESTORE */
308 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
310 if (pid_ns == &init_pid_ns)
314 case LINUX_REBOOT_CMD_RESTART2:
315 case LINUX_REBOOT_CMD_RESTART:
316 pid_ns->reboot = SIGHUP;
319 case LINUX_REBOOT_CMD_POWER_OFF:
320 case LINUX_REBOOT_CMD_HALT:
321 pid_ns->reboot = SIGINT;
327 read_lock(&tasklist_lock);
328 force_sig(SIGKILL, pid_ns->child_reaper);
329 read_unlock(&tasklist_lock);
337 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
339 return container_of(ns, struct pid_namespace, ns);
342 static struct ns_common *pidns_get(struct task_struct *task)
344 struct pid_namespace *ns;
347 ns = task_active_pid_ns(task);
352 return ns ? &ns->ns : NULL;
355 static struct ns_common *pidns_for_children_get(struct task_struct *task)
357 struct pid_namespace *ns = NULL;
361 ns = task->nsproxy->pid_ns_for_children;
367 read_lock(&tasklist_lock);
368 if (!ns->child_reaper) {
372 read_unlock(&tasklist_lock);
375 return ns ? &ns->ns : NULL;
378 static void pidns_put(struct ns_common *ns)
380 put_pid_ns(to_pid_ns(ns));
383 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
385 struct pid_namespace *active = task_active_pid_ns(current);
386 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
388 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
389 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
393 * Only allow entering the current active pid namespace
394 * or a child of the current active pid namespace.
396 * This is required for fork to return a usable pid value and
397 * this maintains the property that processes and their
398 * children can not escape their current pid namespace.
400 if (new->level < active->level)
404 while (ancestor->level > active->level)
405 ancestor = ancestor->parent;
406 if (ancestor != active)
409 put_pid_ns(nsproxy->pid_ns_for_children);
410 nsproxy->pid_ns_for_children = get_pid_ns(new);
414 static struct ns_common *pidns_get_parent(struct ns_common *ns)
416 struct pid_namespace *active = task_active_pid_ns(current);
417 struct pid_namespace *pid_ns, *p;
419 /* See if the parent is in the current namespace */
420 pid_ns = p = to_pid_ns(ns)->parent;
423 return ERR_PTR(-EPERM);
429 return &get_pid_ns(pid_ns)->ns;
432 static struct user_namespace *pidns_owner(struct ns_common *ns)
434 return to_pid_ns(ns)->user_ns;
437 const struct proc_ns_operations pidns_operations = {
439 .type = CLONE_NEWPID,
442 .install = pidns_install,
443 .owner = pidns_owner,
444 .get_parent = pidns_get_parent,
447 const struct proc_ns_operations pidns_for_children_operations = {
448 .name = "pid_for_children",
449 .real_ns_name = "pid",
450 .type = CLONE_NEWPID,
451 .get = pidns_for_children_get,
453 .install = pidns_install,
454 .owner = pidns_owner,
455 .get_parent = pidns_get_parent,
458 static __init int pid_namespaces_init(void)
460 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
462 #ifdef CONFIG_CHECKPOINT_RESTORE
463 register_sysctl_paths(kern_path, pid_ns_ctl_table);
468 __initcall(pid_namespaces_init);