pid_namespace.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Pid namespaces
  4 *
  5 * Authors:
  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
  9 *
 10 */
 11
 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 <linux/nstree.h>
 27#include <uapi/linux/wait.h>
 28#include "pid_sysctl.h"
 29
 30static DEFINE_MUTEX(pid_caches_mutex);
 31static struct kmem_cache *pid_ns_cachep;
 32/* Write once array, filled from the beginning. */
 33static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
 34
 35/*
 36 * creates the kmem cache to allocate pids from.
 37 * @level: pid namespace level
 38 */
 39
 40static struct kmem_cache *create_pid_cachep(unsigned int level)
 41{
 42	/* Level 0 is init_pid_ns.pid_cachep */
 43	struct kmem_cache **pkc = &pid_cache[level - 1];
 44	struct kmem_cache *kc;
 45	char name[4 + 10 + 1];
 46	unsigned int len;
 47
 48	kc = READ_ONCE(*pkc);
 49	if (kc)
 50		return kc;
 51
 52	snprintf(name, sizeof(name), "pid_%u", level + 1);
 53	len = struct_size_t(struct pid, numbers, level + 1);
 54	mutex_lock(&pid_caches_mutex);
 55	/* Name collision forces to do allocation under mutex. */
 56	if (!*pkc)
 57		*pkc = kmem_cache_create(name, len, 0,
 58					 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
 59	mutex_unlock(&pid_caches_mutex);
 60	/* current can fail, but someone else can succeed. */
 61	return READ_ONCE(*pkc);
 62}
 63
 64static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
 65{
 66	return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
 67}
 68
 69static void dec_pid_namespaces(struct ucounts *ucounts)
 70{
 71	dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
 72}
 73
 74static void destroy_pid_namespace_work(struct work_struct *work);
 75
 76static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
 77	struct pid_namespace *parent_pid_ns)
 78{
 79	struct pid_namespace *ns;
 80	unsigned int level = parent_pid_ns->level + 1;
 81	struct ucounts *ucounts;
 82	int err;
 83
 84	err = -EINVAL;
 85	if (!in_userns(parent_pid_ns->user_ns, user_ns))
 86		goto out;
 87
 88	err = -ENOSPC;
 89	if (level > MAX_PID_NS_LEVEL)
 90		goto out;
 91	ucounts = inc_pid_namespaces(user_ns);
 92	if (!ucounts)
 93		goto out;
 94
 95	err = -ENOMEM;
 96	ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
 97	if (ns == NULL)
 98		goto out_dec;
 99
100	idr_init(&ns->idr);
101
102	ns->pid_cachep = create_pid_cachep(level);
103	if (ns->pid_cachep == NULL)
104		goto out_free_idr;
105
106	err = ns_common_init(ns);
107	if (err)
108		goto out_free_idr;
109
110	ns->pid_max = PID_MAX_LIMIT;
111	err = register_pidns_sysctls(ns);
112	if (err)
113		goto out_free_inum;
114
115	ns->level = level;
116	ns->parent = get_pid_ns(parent_pid_ns);
117	ns->user_ns = get_user_ns(user_ns);
118	ns->ucounts = ucounts;
119	ns->pid_allocated = PIDNS_ADDING;
120	INIT_WORK(&ns->work, destroy_pid_namespace_work);
121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
123	ns->memfd_noexec_scope = pidns_memfd_noexec_scope(parent_pid_ns);
124#endif
125
126	ns_tree_add(ns);
127	return ns;
128
129out_free_inum:
130	ns_common_free(ns);
131out_free_idr:
132	idr_destroy(&ns->idr);
133	kmem_cache_free(pid_ns_cachep, ns);
134out_dec:
135	dec_pid_namespaces(ucounts);
136out:
137	return ERR_PTR(err);
138}
139
140static void delayed_free_pidns(struct rcu_head *p)
141{
142	struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
143
144	dec_pid_namespaces(ns->ucounts);
145	put_user_ns(ns->user_ns);
146
147	kmem_cache_free(pid_ns_cachep, ns);
148}
149
150static void destroy_pid_namespace(struct pid_namespace *ns)
151{
152	ns_tree_remove(ns);
153	unregister_pidns_sysctls(ns);
154
155	ns_common_free(ns);
156
157	idr_destroy(&ns->idr);
158	call_rcu(&ns->rcu, delayed_free_pidns);
159}
160
161static void destroy_pid_namespace_work(struct work_struct *work)
162{
163	struct pid_namespace *ns =
164		container_of(work, struct pid_namespace, work);
165
166	do {
167		struct pid_namespace *parent;
168
169		parent = ns->parent;
170		destroy_pid_namespace(ns);
171		ns = parent;
172	} while (ns != &init_pid_ns && ns_ref_put(ns));
173}
174
175struct pid_namespace *copy_pid_ns(u64 flags,
176	struct user_namespace *user_ns, struct pid_namespace *old_ns)
177{
178	if (!(flags & CLONE_NEWPID))
179		return get_pid_ns(old_ns);
180	if (task_active_pid_ns(current) != old_ns)
181		return ERR_PTR(-EINVAL);
182	return create_pid_namespace(user_ns, old_ns);
183}
184
185void put_pid_ns(struct pid_namespace *ns)
186{
187	if (ns && ns_ref_put(ns))
188		schedule_work(&ns->work);
189}
190EXPORT_SYMBOL_GPL(put_pid_ns);
191
192void zap_pid_ns_processes(struct pid_namespace *pid_ns)
193{
194	int nr;
195	int rc;
196	struct task_struct *task, *me = current;
197	int init_pids = thread_group_leader(me) ? 1 : 2;
198	struct pid *pid;
199
200	/* Don't allow any more processes into the pid namespace */
201	disable_pid_allocation(pid_ns);
202
203	/*
204	 * Ignore SIGCHLD causing any terminated children to autoreap.
205	 * This speeds up the namespace shutdown, plus see the comment
206	 * below.
207	 */
208	spin_lock_irq(&me->sighand->siglock);
209	me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
210	spin_unlock_irq(&me->sighand->siglock);
211
212	/*
213	 * The last thread in the cgroup-init thread group is terminating.
214	 * Find remaining pid_ts in the namespace, signal and wait for them
215	 * to exit.
216	 *
217	 * Note:  This signals each threads in the namespace - even those that
218	 * 	  belong to the same thread group, To avoid this, we would have
219	 * 	  to walk the entire tasklist looking a processes in this
220	 * 	  namespace, but that could be unnecessarily expensive if the
221	 * 	  pid namespace has just a few processes. Or we need to
222	 * 	  maintain a tasklist for each pid namespace.
223	 *
224	 */
225	rcu_read_lock();
226	read_lock(&tasklist_lock);
227	nr = 2;
228	idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
229		task = pid_task(pid, PIDTYPE_PID);
230		if (task && !__fatal_signal_pending(task))
231			group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
232	}
233	read_unlock(&tasklist_lock);
234	rcu_read_unlock();
235
236	/*
237	 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
238	 * kernel_wait4() will also block until our children traced from the
239	 * parent namespace are detached and become EXIT_DEAD.
240	 */
241	do {
242		clear_thread_flag(TIF_SIGPENDING);
243		clear_thread_flag(TIF_NOTIFY_SIGNAL);
244		rc = kernel_wait4(-1, NULL, __WALL, NULL);
245	} while (rc != -ECHILD);
246
247	/*
248	 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
249	 * process whose parents processes are outside of the pid
250	 * namespace.  Such processes are created with setns()+fork().
251	 *
252	 * If those EXIT_ZOMBIE processes are not reaped by their
253	 * parents before their parents exit, they will be reparented
254	 * to pid_ns->child_reaper.  Thus pidns->child_reaper needs to
255	 * stay valid until they all go away.
256	 *
257	 * The code relies on the pid_ns->child_reaper ignoring
258	 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
259	 * autoreaped if reparented.
260	 *
261	 * Semantically it is also desirable to wait for EXIT_ZOMBIE
262	 * processes before allowing the child_reaper to be reaped, as
263	 * that gives the invariant that when the init process of a
264	 * pid namespace is reaped all of the processes in the pid
265	 * namespace are gone.
266	 *
267	 * Once all of the other tasks are gone from the pid_namespace
268	 * free_pid() will awaken this task.
269	 */
270	for (;;) {
271		set_current_state(TASK_INTERRUPTIBLE);
272		if (pid_ns->pid_allocated == init_pids)
273			break;
274		schedule();
275	}
276	__set_current_state(TASK_RUNNING);
277
278	if (pid_ns->reboot)
279		current->signal->group_exit_code = pid_ns->reboot;
280
281	acct_exit_ns(pid_ns);
282	return;
283}
284
285#ifdef CONFIG_CHECKPOINT_RESTORE
286static int pid_ns_ctl_handler(const struct ctl_table *table, int write,
287		void *buffer, size_t *lenp, loff_t *ppos)
288{
289	struct pid_namespace *pid_ns = task_active_pid_ns(current);
290	struct ctl_table tmp = *table;
291	int ret, next;
292
293	if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
294		return -EPERM;
295
296	next = idr_get_cursor(&pid_ns->idr) - 1;
297
298	tmp.data = &next;
299	tmp.extra2 = &pid_ns->pid_max;
300	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
301	if (!ret && write)
302		idr_set_cursor(&pid_ns->idr, next + 1);
303
304	return ret;
305}
306
307static const struct ctl_table pid_ns_ctl_table[] = {
308	{
309		.procname = "ns_last_pid",
310		.maxlen = sizeof(int),
311		.mode = 0666, /* permissions are checked in the handler */
312		.proc_handler = pid_ns_ctl_handler,
313		.extra1 = SYSCTL_ZERO,
314		.extra2 = &init_pid_ns.pid_max,
315	},
316};
317#endif	/* CONFIG_CHECKPOINT_RESTORE */
318
319int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
320{
321	if (pid_ns == &init_pid_ns)
322		return 0;
323
324	switch (cmd) {
325	case LINUX_REBOOT_CMD_RESTART2:
326	case LINUX_REBOOT_CMD_RESTART:
327		pid_ns->reboot = SIGHUP;
328		break;
329
330	case LINUX_REBOOT_CMD_POWER_OFF:
331	case LINUX_REBOOT_CMD_HALT:
332		pid_ns->reboot = SIGINT;
333		break;
334	default:
335		return -EINVAL;
336	}
337
338	read_lock(&tasklist_lock);
339	send_sig(SIGKILL, pid_ns->child_reaper, 1);
340	read_unlock(&tasklist_lock);
341
342	do_exit(0);
343
344	/* Not reached */
345	return 0;
346}
347
348static struct ns_common *pidns_get(struct task_struct *task)
349{
350	struct pid_namespace *ns;
351
352	rcu_read_lock();
353	ns = task_active_pid_ns(task);
354	if (ns)
355		get_pid_ns(ns);
356	rcu_read_unlock();
357
358	return ns ? &ns->ns : NULL;
359}
360
361static struct ns_common *pidns_for_children_get(struct task_struct *task)
362{
363	struct pid_namespace *ns = NULL;
364
365	task_lock(task);
366	if (task->nsproxy) {
367		ns = task->nsproxy->pid_ns_for_children;
368		get_pid_ns(ns);
369	}
370	task_unlock(task);
371
372	if (ns) {
373		read_lock(&tasklist_lock);
374		if (!ns->child_reaper) {
375			put_pid_ns(ns);
376			ns = NULL;
377		}
378		read_unlock(&tasklist_lock);
379	}
380
381	return ns ? &ns->ns : NULL;
382}
383
384static void pidns_put(struct ns_common *ns)
385{
386	put_pid_ns(to_pid_ns(ns));
387}
388
389bool pidns_is_ancestor(struct pid_namespace *child,
390		       struct pid_namespace *ancestor)
391{
392	struct pid_namespace *ns;
393
394	if (child->level < ancestor->level)
395		return false;
396	for (ns = child; ns->level > ancestor->level; ns = ns->parent)
397		;
398	return ns == ancestor;
399}
400
401static int pidns_install(struct nsset *nsset, struct ns_common *ns)
402{
403	struct nsproxy *nsproxy = nsset->nsproxy;
404	struct pid_namespace *active = task_active_pid_ns(current);
405	struct pid_namespace *new = to_pid_ns(ns);
406
407	if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
408	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
409		return -EPERM;
410
411	/*
412	 * Only allow entering the current active pid namespace
413	 * or a child of the current active pid namespace.
414	 *
415	 * This is required for fork to return a usable pid value and
416	 * this maintains the property that processes and their
417	 * children can not escape their current pid namespace.
418	 */
419	if (!pidns_is_ancestor(new, active))
420		return -EINVAL;
421
422	put_pid_ns(nsproxy->pid_ns_for_children);
423	nsproxy->pid_ns_for_children = get_pid_ns(new);
424	return 0;
425}
426
427static struct ns_common *pidns_get_parent(struct ns_common *ns)
428{
429	struct pid_namespace *active = task_active_pid_ns(current);
430	struct pid_namespace *pid_ns, *p;
431
432	/* See if the parent is in the current namespace */
433	pid_ns = p = to_pid_ns(ns)->parent;
434	for (;;) {
435		if (!p)
436			return ERR_PTR(-EPERM);
437		if (p == active)
438			break;
439		p = p->parent;
440	}
441
442	return &get_pid_ns(pid_ns)->ns;
443}
444
445static struct user_namespace *pidns_owner(struct ns_common *ns)
446{
447	return to_pid_ns(ns)->user_ns;
448}
449
450const struct proc_ns_operations pidns_operations = {
451	.name		= "pid",
452	.get		= pidns_get,
453	.put		= pidns_put,
454	.install	= pidns_install,
455	.owner		= pidns_owner,
456	.get_parent	= pidns_get_parent,
457};
458
459const struct proc_ns_operations pidns_for_children_operations = {
460	.name		= "pid_for_children",
461	.real_ns_name	= "pid",
462	.get		= pidns_for_children_get,
463	.put		= pidns_put,
464	.install	= pidns_install,
465	.owner		= pidns_owner,
466	.get_parent	= pidns_get_parent,
467};
468
469static __init int pid_namespaces_init(void)
470{
471	pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC | SLAB_ACCOUNT);
472
473#ifdef CONFIG_CHECKPOINT_RESTORE
474	register_sysctl_init("kernel", pid_ns_ctl_table);
475#endif
476
477	register_pid_ns_sysctl_table_vm();
478	ns_tree_add(&init_pid_ns);
479	return 0;
480}
481
482__initcall(pid_namespaces_init);