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

source code of linux/kernel/pid_namespace.c