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>
27 struct kmem_cache *cachep;
28 struct list_head list;
31 static LIST_HEAD(pid_caches_lh);
32 static DEFINE_MUTEX(pid_caches_mutex);
33 static struct kmem_cache *pid_ns_cachep;
36 * creates the kmem cache to allocate pids from.
37 * @nr_ids: the number of numerical ids this pid will have to carry
40 static struct kmem_cache *create_pid_cachep(int nr_ids)
42 struct pid_cache *pcache;
43 struct kmem_cache *cachep;
45 mutex_lock(&pid_caches_mutex);
46 list_for_each_entry(pcache, &pid_caches_lh, list)
47 if (pcache->nr_ids == nr_ids)
50 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
54 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
55 cachep = kmem_cache_create(pcache->name,
56 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
57 0, SLAB_HWCACHE_ALIGN, NULL);
61 pcache->nr_ids = nr_ids;
62 pcache->cachep = cachep;
63 list_add(&pcache->list, &pid_caches_lh);
65 mutex_unlock(&pid_caches_mutex);
66 return pcache->cachep;
71 mutex_unlock(&pid_caches_mutex);
75 static void proc_cleanup_work(struct work_struct *work)
77 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
78 pid_ns_release_proc(ns);
81 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
82 #define MAX_PID_NS_LEVEL 32
84 static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
86 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
89 static void dec_pid_namespaces(struct ucounts *ucounts)
91 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
94 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
95 struct pid_namespace *parent_pid_ns)
97 struct pid_namespace *ns;
98 unsigned int level = parent_pid_ns->level + 1;
99 struct ucounts *ucounts;
104 if (level > MAX_PID_NS_LEVEL)
106 ucounts = inc_pid_namespaces(user_ns);
111 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
115 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
116 if (!ns->pidmap[0].page)
119 ns->pid_cachep = create_pid_cachep(level + 1);
120 if (ns->pid_cachep == NULL)
123 err = ns_alloc_inum(&ns->ns);
126 ns->ns.ops = &pidns_operations;
128 kref_init(&ns->kref);
130 ns->parent = get_pid_ns(parent_pid_ns);
131 ns->user_ns = get_user_ns(user_ns);
132 ns->ucounts = ucounts;
133 ns->nr_hashed = PIDNS_HASH_ADDING;
134 INIT_WORK(&ns->proc_work, proc_cleanup_work);
136 set_bit(0, ns->pidmap[0].page);
137 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
139 for (i = 1; i < PIDMAP_ENTRIES; i++)
140 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
145 kfree(ns->pidmap[0].page);
147 kmem_cache_free(pid_ns_cachep, ns);
149 dec_pid_namespaces(ucounts);
154 static void delayed_free_pidns(struct rcu_head *p)
156 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
158 dec_pid_namespaces(ns->ucounts);
159 put_user_ns(ns->user_ns);
161 kmem_cache_free(pid_ns_cachep, ns);
164 static void destroy_pid_namespace(struct pid_namespace *ns)
168 ns_free_inum(&ns->ns);
169 for (i = 0; i < PIDMAP_ENTRIES; i++)
170 kfree(ns->pidmap[i].page);
171 call_rcu(&ns->rcu, delayed_free_pidns);
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 static void free_pid_ns(struct kref *kref)
186 struct pid_namespace *ns;
188 ns = container_of(kref, struct pid_namespace, kref);
189 destroy_pid_namespace(ns);
192 void put_pid_ns(struct pid_namespace *ns)
194 struct pid_namespace *parent;
196 while (ns != &init_pid_ns) {
198 if (!kref_put(&ns->kref, free_pid_ns))
203 EXPORT_SYMBOL_GPL(put_pid_ns);
205 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
209 struct task_struct *task, *me = current;
210 int init_pids = thread_group_leader(me) ? 1 : 2;
212 /* Don't allow any more processes into the pid namespace */
213 disable_pid_allocation(pid_ns);
216 * Ignore SIGCHLD causing any terminated children to autoreap.
217 * This speeds up the namespace shutdown, plus see the comment
220 spin_lock_irq(&me->sighand->siglock);
221 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
222 spin_unlock_irq(&me->sighand->siglock);
225 * The last thread in the cgroup-init thread group is terminating.
226 * Find remaining pid_ts in the namespace, signal and wait for them
229 * Note: This signals each threads in the namespace - even those that
230 * belong to the same thread group, To avoid this, we would have
231 * to walk the entire tasklist looking a processes in this
232 * namespace, but that could be unnecessarily expensive if the
233 * pid namespace has just a few processes. Or we need to
234 * maintain a tasklist for each pid namespace.
237 read_lock(&tasklist_lock);
238 nr = next_pidmap(pid_ns, 1);
242 task = pid_task(find_vpid(nr), PIDTYPE_PID);
243 if (task && !__fatal_signal_pending(task))
244 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
248 nr = next_pidmap(pid_ns, nr);
250 read_unlock(&tasklist_lock);
253 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
254 * sys_wait4() will also block until our children traced from the
255 * parent namespace are detached and become EXIT_DEAD.
258 clear_thread_flag(TIF_SIGPENDING);
259 rc = sys_wait4(-1, NULL, __WALL, NULL);
260 } while (rc != -ECHILD);
263 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
264 * really care, we could reparent them to the global init. We could
265 * exit and reap ->child_reaper even if it is not the last thread in
266 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
267 * pid_ns can not go away until proc_kill_sb() drops the reference.
269 * But this ns can also have other tasks injected by setns()+fork().
270 * Again, ignoring the user visible semantics we do not really need
271 * to wait until they are all reaped, but they can be reparented to
272 * us and thus we need to ensure that pid->child_reaper stays valid
273 * until they all go away. See free_pid()->wake_up_process().
275 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
279 set_current_state(TASK_UNINTERRUPTIBLE);
280 if (pid_ns->nr_hashed == init_pids)
284 __set_current_state(TASK_RUNNING);
287 current->signal->group_exit_code = pid_ns->reboot;
289 acct_exit_ns(pid_ns);
293 #ifdef CONFIG_CHECKPOINT_RESTORE
294 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
295 void __user *buffer, size_t *lenp, loff_t *ppos)
297 struct pid_namespace *pid_ns = task_active_pid_ns(current);
298 struct ctl_table tmp = *table;
300 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
304 * Writing directly to ns' last_pid field is OK, since this field
305 * is volatile in a living namespace anyway and a code writing to
306 * it should synchronize its usage with external means.
309 tmp.data = &pid_ns->last_pid;
310 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
315 static struct ctl_table pid_ns_ctl_table[] = {
317 .procname = "ns_last_pid",
318 .maxlen = sizeof(int),
319 .mode = 0666, /* permissions are checked in the handler */
320 .proc_handler = pid_ns_ctl_handler,
326 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
327 #endif /* CONFIG_CHECKPOINT_RESTORE */
329 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
331 if (pid_ns == &init_pid_ns)
335 case LINUX_REBOOT_CMD_RESTART2:
336 case LINUX_REBOOT_CMD_RESTART:
337 pid_ns->reboot = SIGHUP;
340 case LINUX_REBOOT_CMD_POWER_OFF:
341 case LINUX_REBOOT_CMD_HALT:
342 pid_ns->reboot = SIGINT;
348 read_lock(&tasklist_lock);
349 force_sig(SIGKILL, pid_ns->child_reaper);
350 read_unlock(&tasklist_lock);
358 static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
360 return container_of(ns, struct pid_namespace, ns);
363 static struct ns_common *pidns_get(struct task_struct *task)
365 struct pid_namespace *ns;
368 ns = task_active_pid_ns(task);
373 return ns ? &ns->ns : NULL;
376 static void pidns_put(struct ns_common *ns)
378 put_pid_ns(to_pid_ns(ns));
381 static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
383 struct pid_namespace *active = task_active_pid_ns(current);
384 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
386 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
387 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
391 * Only allow entering the current active pid namespace
392 * or a child of the current active pid namespace.
394 * This is required for fork to return a usable pid value and
395 * this maintains the property that processes and their
396 * children can not escape their current pid namespace.
398 if (new->level < active->level)
402 while (ancestor->level > active->level)
403 ancestor = ancestor->parent;
404 if (ancestor != active)
407 put_pid_ns(nsproxy->pid_ns_for_children);
408 nsproxy->pid_ns_for_children = get_pid_ns(new);
412 static struct ns_common *pidns_get_parent(struct ns_common *ns)
414 struct pid_namespace *active = task_active_pid_ns(current);
415 struct pid_namespace *pid_ns, *p;
417 /* See if the parent is in the current namespace */
418 pid_ns = p = to_pid_ns(ns)->parent;
421 return ERR_PTR(-EPERM);
427 return &get_pid_ns(pid_ns)->ns;
430 static struct user_namespace *pidns_owner(struct ns_common *ns)
432 return to_pid_ns(ns)->user_ns;
435 const struct proc_ns_operations pidns_operations = {
437 .type = CLONE_NEWPID,
440 .install = pidns_install,
441 .owner = pidns_owner,
442 .get_parent = pidns_get_parent,
445 static __init int pid_namespaces_init(void)
447 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
449 #ifdef CONFIG_CHECKPOINT_RESTORE
450 register_sysctl_paths(kern_path, pid_ns_ctl_table);
455 __initcall(pid_namespaces_init);