1/*
2 * Pid namespaces
3 *
4 * Authors:
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
8 *
9 */
10
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/err.h>
16#include <linux/acct.h>
17#include <linux/slab.h>
18#include <linux/proc_ns.h>
19#include <linux/reboot.h>
20#include <linux/export.h>
21
22struct pid_cache {
23	int nr_ids;
24	char name[16];
25	struct kmem_cache *cachep;
26	struct list_head list;
27};
28
29static LIST_HEAD(pid_caches_lh);
30static DEFINE_MUTEX(pid_caches_mutex);
31static struct kmem_cache *pid_ns_cachep;
32
33/*
34 * creates the kmem cache to allocate pids from.
35 * @nr_ids: the number of numerical ids this pid will have to carry
36 */
37
38static struct kmem_cache *create_pid_cachep(int nr_ids)
39{
40	struct pid_cache *pcache;
41	struct kmem_cache *cachep;
42
43	mutex_lock(&pid_caches_mutex);
44	list_for_each_entry(pcache, &pid_caches_lh, list)
45		if (pcache->nr_ids == nr_ids)
46			goto out;
47
48	pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
49	if (pcache == NULL)
50		goto err_alloc;
51
52	snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
53	cachep = kmem_cache_create(pcache->name,
54			sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
55			0, SLAB_HWCACHE_ALIGN, NULL);
56	if (cachep == NULL)
57		goto err_cachep;
58
59	pcache->nr_ids = nr_ids;
60	pcache->cachep = cachep;
61	list_add(&pcache->list, &pid_caches_lh);
62out:
63	mutex_unlock(&pid_caches_mutex);
64	return pcache->cachep;
65
66err_cachep:
67	kfree(pcache);
68err_alloc:
69	mutex_unlock(&pid_caches_mutex);
70	return NULL;
71}
72
73static void proc_cleanup_work(struct work_struct *work)
74{
75	struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
76	pid_ns_release_proc(ns);
77}
78
79/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
80#define MAX_PID_NS_LEVEL 32
81
82static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
83	struct pid_namespace *parent_pid_ns)
84{
85	struct pid_namespace *ns;
86	unsigned int level = parent_pid_ns->level + 1;
87	int i;
88	int err;
89
90	if (level > MAX_PID_NS_LEVEL) {
91		err = -EINVAL;
92		goto out;
93	}
94
95	err = -ENOMEM;
96	ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
97	if (ns == NULL)
98		goto out;
99
100	ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
101	if (!ns->pidmap[0].page)
102		goto out_free;
103
104	ns->pid_cachep = create_pid_cachep(level + 1);
105	if (ns->pid_cachep == NULL)
106		goto out_free_map;
107
108	err = ns_alloc_inum(&ns->ns);
109	if (err)
110		goto out_free_map;
111	ns->ns.ops = &pidns_operations;
112
113	kref_init(&ns->kref);
114	ns->level = level;
115	ns->parent = get_pid_ns(parent_pid_ns);
116	ns->user_ns = get_user_ns(user_ns);
117	ns->nr_hashed = PIDNS_HASH_ADDING;
118	INIT_WORK(&ns->proc_work, proc_cleanup_work);
119
120	set_bit(0, ns->pidmap[0].page);
121	atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
122
123	for (i = 1; i < PIDMAP_ENTRIES; i++)
124		atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
125
126	return ns;
127
128out_free_map:
129	kfree(ns->pidmap[0].page);
130out_free:
131	kmem_cache_free(pid_ns_cachep, ns);
132out:
133	return ERR_PTR(err);
134}
135
136static void delayed_free_pidns(struct rcu_head *p)
137{
138	kmem_cache_free(pid_ns_cachep,
139			container_of(p, struct pid_namespace, rcu));
140}
141
142static void destroy_pid_namespace(struct pid_namespace *ns)
143{
144	int i;
145
146	ns_free_inum(&ns->ns);
147	for (i = 0; i < PIDMAP_ENTRIES; i++)
148		kfree(ns->pidmap[i].page);
149	put_user_ns(ns->user_ns);
150	call_rcu(&ns->rcu, delayed_free_pidns);
151}
152
153struct pid_namespace *copy_pid_ns(unsigned long flags,
154	struct user_namespace *user_ns, struct pid_namespace *old_ns)
155{
156	if (!(flags & CLONE_NEWPID))
157		return get_pid_ns(old_ns);
158	if (task_active_pid_ns(current) != old_ns)
159		return ERR_PTR(-EINVAL);
160	return create_pid_namespace(user_ns, old_ns);
161}
162
163static void free_pid_ns(struct kref *kref)
164{
165	struct pid_namespace *ns;
166
167	ns = container_of(kref, struct pid_namespace, kref);
168	destroy_pid_namespace(ns);
169}
170
171void put_pid_ns(struct pid_namespace *ns)
172{
173	struct pid_namespace *parent;
174
175	while (ns != &init_pid_ns) {
176		parent = ns->parent;
177		if (!kref_put(&ns->kref, free_pid_ns))
178			break;
179		ns = parent;
180	}
181}
182EXPORT_SYMBOL_GPL(put_pid_ns);
183
184void zap_pid_ns_processes(struct pid_namespace *pid_ns)
185{
186	int nr;
187	int rc;
188	struct task_struct *task, *me = current;
189	int init_pids = thread_group_leader(me) ? 1 : 2;
190
191	/* Don't allow any more processes into the pid namespace */
192	disable_pid_allocation(pid_ns);
193
194	/*
195	 * Ignore SIGCHLD causing any terminated children to autoreap.
196	 * This speeds up the namespace shutdown, plus see the comment
197	 * below.
198	 */
199	spin_lock_irq(&me->sighand->siglock);
200	me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
201	spin_unlock_irq(&me->sighand->siglock);
202
203	/*
204	 * The last thread in the cgroup-init thread group is terminating.
205	 * Find remaining pid_ts in the namespace, signal and wait for them
206	 * to exit.
207	 *
208	 * Note:  This signals each threads in the namespace - even those that
209	 * 	  belong to the same thread group, To avoid this, we would have
210	 * 	  to walk the entire tasklist looking a processes in this
211	 * 	  namespace, but that could be unnecessarily expensive if the
212	 * 	  pid namespace has just a few processes. Or we need to
213	 * 	  maintain a tasklist for each pid namespace.
214	 *
215	 */
216	read_lock(&tasklist_lock);
217	nr = next_pidmap(pid_ns, 1);
218	while (nr > 0) {
219		rcu_read_lock();
220
221		task = pid_task(find_vpid(nr), PIDTYPE_PID);
222		if (task && !__fatal_signal_pending(task))
223			send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
224
225		rcu_read_unlock();
226
227		nr = next_pidmap(pid_ns, nr);
228	}
229	read_unlock(&tasklist_lock);
230
231	/*
232	 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
233	 * sys_wait4() will also block until our children traced from the
234	 * parent namespace are detached and become EXIT_DEAD.
235	 */
236	do {
237		clear_thread_flag(TIF_SIGPENDING);
238		rc = sys_wait4(-1, NULL, __WALL, NULL);
239	} while (rc != -ECHILD);
240
241	/*
242	 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
243	 * really care, we could reparent them to the global init. We could
244	 * exit and reap ->child_reaper even if it is not the last thread in
245	 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
246	 * pid_ns can not go away until proc_kill_sb() drops the reference.
247	 *
248	 * But this ns can also have other tasks injected by setns()+fork().
249	 * Again, ignoring the user visible semantics we do not really need
250	 * to wait until they are all reaped, but they can be reparented to
251	 * us and thus we need to ensure that pid->child_reaper stays valid
252	 * until they all go away. See free_pid()->wake_up_process().
253	 *
254	 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
255	 * if reparented.
256	 */
257	for (;;) {
258		set_current_state(TASK_UNINTERRUPTIBLE);
259		if (pid_ns->nr_hashed == init_pids)
260			break;
261		schedule();
262	}
263	__set_current_state(TASK_RUNNING);
264
265	if (pid_ns->reboot)
266		current->signal->group_exit_code = pid_ns->reboot;
267
268	acct_exit_ns(pid_ns);
269	return;
270}
271
272#ifdef CONFIG_CHECKPOINT_RESTORE
273static int pid_ns_ctl_handler(struct ctl_table *table, int write,
274		void __user *buffer, size_t *lenp, loff_t *ppos)
275{
276	struct pid_namespace *pid_ns = task_active_pid_ns(current);
277	struct ctl_table tmp = *table;
278
279	if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
280		return -EPERM;
281
282	/*
283	 * Writing directly to ns' last_pid field is OK, since this field
284	 * is volatile in a living namespace anyway and a code writing to
285	 * it should synchronize its usage with external means.
286	 */
287
288	tmp.data = &pid_ns->last_pid;
289	return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
290}
291
292extern int pid_max;
293static int zero = 0;
294static struct ctl_table pid_ns_ctl_table[] = {
295	{
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,
300		.extra1 = &zero,
301		.extra2 = &pid_max,
302	},
303	{ }
304};
305static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
306#endif	/* CONFIG_CHECKPOINT_RESTORE */
307
308int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
309{
310	if (pid_ns == &init_pid_ns)
311		return 0;
312
313	switch (cmd) {
314	case LINUX_REBOOT_CMD_RESTART2:
315	case LINUX_REBOOT_CMD_RESTART:
316		pid_ns->reboot = SIGHUP;
317		break;
318
319	case LINUX_REBOOT_CMD_POWER_OFF:
320	case LINUX_REBOOT_CMD_HALT:
321		pid_ns->reboot = SIGINT;
322		break;
323	default:
324		return -EINVAL;
325	}
326
327	read_lock(&tasklist_lock);
328	force_sig(SIGKILL, pid_ns->child_reaper);
329	read_unlock(&tasklist_lock);
330
331	do_exit(0);
332
333	/* Not reached */
334	return 0;
335}
336
337static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
338{
339	return container_of(ns, struct pid_namespace, ns);
340}
341
342static struct ns_common *pidns_get(struct task_struct *task)
343{
344	struct pid_namespace *ns;
345
346	rcu_read_lock();
347	ns = task_active_pid_ns(task);
348	if (ns)
349		get_pid_ns(ns);
350	rcu_read_unlock();
351
352	return ns ? &ns->ns : NULL;
353}
354
355static void pidns_put(struct ns_common *ns)
356{
357	put_pid_ns(to_pid_ns(ns));
358}
359
360static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
361{
362	struct pid_namespace *active = task_active_pid_ns(current);
363	struct pid_namespace *ancestor, *new = to_pid_ns(ns);
364
365	if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
366	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
367		return -EPERM;
368
369	/*
370	 * Only allow entering the current active pid namespace
371	 * or a child of the current active pid namespace.
372	 *
373	 * This is required for fork to return a usable pid value and
374	 * this maintains the property that processes and their
375	 * children can not escape their current pid namespace.
376	 */
377	if (new->level < active->level)
378		return -EINVAL;
379
380	ancestor = new;
381	while (ancestor->level > active->level)
382		ancestor = ancestor->parent;
383	if (ancestor != active)
384		return -EINVAL;
385
386	put_pid_ns(nsproxy->pid_ns_for_children);
387	nsproxy->pid_ns_for_children = get_pid_ns(new);
388	return 0;
389}
390
391const struct proc_ns_operations pidns_operations = {
392	.name		= "pid",
393	.type		= CLONE_NEWPID,
394	.get		= pidns_get,
395	.put		= pidns_put,
396	.install	= pidns_install,
397};
398
399static __init int pid_namespaces_init(void)
400{
401	pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
402
403#ifdef CONFIG_CHECKPOINT_RESTORE
404	register_sysctl_paths(kern_path, pid_ns_ctl_table);
405#endif
406	return 0;
407}
408
409__initcall(pid_namespaces_init);
410