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