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