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/cred.h>
16#include <linux/err.h>
17#include <linux/acct.h>
18#include <linux/slab.h>
19#include <linux/proc_ns.h>
20#include <linux/reboot.h>
21#include <linux/export.h>
22#include <linux/sched/task.h>
23#include <linux/sched/signal.h>
24#include <linux/idr.h>
25
26static DEFINE_MUTEX(pid_caches_mutex);
27static struct kmem_cache *pid_ns_cachep;
28/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
29#define MAX_PID_NS_LEVEL 32
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(name, sizeof(name), "pid_%u", level + 1);
51 len = sizeof(struct pid) + level * sizeof(struct upid);
52 mutex_lock(&pid_caches_mutex);
53 /* Name collision forces to do allocation under mutex. */
54 if (!*pkc)
55 *pkc = kmem_cache_create(name, len, 0, SLAB_HWCACHE_ALIGN, 0);
56 mutex_unlock(&pid_caches_mutex);
57 /* current can fail, but someone else can succeed. */
58 return READ_ONCE(*pkc);
59}
60
61static void proc_cleanup_work(struct work_struct *work)
62{
63 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
64 pid_ns_release_proc(ns);
65}
66
67static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
68{
69 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
70}
71
72static void dec_pid_namespaces(struct ucounts *ucounts)
73{
74 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
75}
76
77static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
78 struct pid_namespace *parent_pid_ns)
79{
80 struct pid_namespace *ns;
81 unsigned int level = parent_pid_ns->level + 1;
82 struct ucounts *ucounts;
83 int err;
84
85 err = -EINVAL;
86 if (!in_userns(parent_pid_ns->user_ns, user_ns))
87 goto out;
88
89 err = -ENOSPC;
90 if (level > MAX_PID_NS_LEVEL)
91 goto out;
92 ucounts = inc_pid_namespaces(user_ns);
93 if (!ucounts)
94 goto out;
95
96 err = -ENOMEM;
97 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
98 if (ns == NULL)
99 goto out_dec;
100
101 idr_init(&ns->idr);
102
103 ns->pid_cachep = create_pid_cachep(level);
104 if (ns->pid_cachep == NULL)
105 goto out_free_idr;
106
107 err = ns_alloc_inum(&ns->ns);
108 if (err)
109 goto out_free_idr;
110 ns->ns.ops = &pidns_operations;
111
112 kref_init(&ns->kref);
113 ns->level = level;
114 ns->parent = get_pid_ns(parent_pid_ns);
115 ns->user_ns = get_user_ns(user_ns);
116 ns->ucounts = ucounts;
117 ns->pid_allocated = PIDNS_ADDING;
118 INIT_WORK(&ns->proc_work, proc_cleanup_work);
119
120 return ns;
121
122out_free_idr:
123 idr_destroy(&ns->idr);
124 kmem_cache_free(pid_ns_cachep, ns);
125out_dec:
126 dec_pid_namespaces(ucounts);
127out:
128 return ERR_PTR(err);
129}
130
131static void delayed_free_pidns(struct rcu_head *p)
132{
133 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
134
135 dec_pid_namespaces(ns->ucounts);
136 put_user_ns(ns->user_ns);
137
138 kmem_cache_free(pid_ns_cachep, ns);
139}
140
141static void destroy_pid_namespace(struct pid_namespace *ns)
142{
143 ns_free_inum(&ns->ns);
144
145 idr_destroy(&ns->idr);
146 call_rcu(&ns->rcu, delayed_free_pidns);
147}
148
149struct pid_namespace *copy_pid_ns(unsigned long flags,
150 struct user_namespace *user_ns, struct pid_namespace *old_ns)
151{
152 if (!(flags & CLONE_NEWPID))
153 return get_pid_ns(old_ns);
154 if (task_active_pid_ns(current) != old_ns)
155 return ERR_PTR(-EINVAL);
156 return create_pid_namespace(user_ns, old_ns);
157}
158
159static void free_pid_ns(struct kref *kref)
160{
161 struct pid_namespace *ns;
162
163 ns = container_of(kref, struct pid_namespace, kref);
164 destroy_pid_namespace(ns);
165}
166
167void put_pid_ns(struct pid_namespace *ns)
168{
169 struct pid_namespace *parent;
170
171 while (ns != &init_pid_ns) {
172 parent = ns->parent;
173 if (!kref_put(&ns->kref, free_pid_ns))
174 break;
175 ns = parent;
176 }
177}
178EXPORT_SYMBOL_GPL(put_pid_ns);
179
180void zap_pid_ns_processes(struct pid_namespace *pid_ns)
181{
182 int nr;
183 int rc;
184 struct task_struct *task, *me = current;
185 int init_pids = thread_group_leader(me) ? 1 : 2;
186 struct pid *pid;
187
188 /* Don't allow any more processes into the pid namespace */
189 disable_pid_allocation(pid_ns);
190
191 /*
192 * Ignore SIGCHLD causing any terminated children to autoreap.
193 * This speeds up the namespace shutdown, plus see the comment
194 * below.
195 */
196 spin_lock_irq(&me->sighand->siglock);
197 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
198 spin_unlock_irq(&me->sighand->siglock);
199
200 /*
201 * The last thread in the cgroup-init thread group is terminating.
202 * Find remaining pid_ts in the namespace, signal and wait for them
203 * to exit.
204 *
205 * Note: This signals each threads in the namespace - even those that
206 * belong to the same thread group, To avoid this, we would have
207 * to walk the entire tasklist looking a processes in this
208 * namespace, but that could be unnecessarily expensive if the
209 * pid namespace has just a few processes. Or we need to
210 * maintain a tasklist for each pid namespace.
211 *
212 */
213 rcu_read_lock();
214 read_lock(&tasklist_lock);
215 nr = 2;
216 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
217 task = pid_task(pid, PIDTYPE_PID);
218 if (task && !__fatal_signal_pending(task))
219 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
220 }
221 read_unlock(&tasklist_lock);
222 rcu_read_unlock();
223
224 /*
225 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
226 * kernel_wait4() will also block until our children traced from the
227 * parent namespace are detached and become EXIT_DEAD.
228 */
229 do {
230 clear_thread_flag(TIF_SIGPENDING);
231 rc = kernel_wait4(-1, NULL, __WALL, NULL);
232 } while (rc != -ECHILD);
233
234 /*
235 * kernel_wait4() above can't reap the EXIT_DEAD children but we do not
236 * really care, we could reparent them to the global init. We could
237 * exit and reap ->child_reaper even if it is not the last thread in
238 * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(),
239 * pid_ns can not go away until proc_kill_sb() drops the reference.
240 *
241 * But this ns can also have other tasks injected by setns()+fork().
242 * Again, ignoring the user visible semantics we do not really need
243 * to wait until they are all reaped, but they can be reparented to
244 * us and thus we need to ensure that pid->child_reaper stays valid
245 * until they all go away. See free_pid()->wake_up_process().
246 *
247 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
248 * if reparented.
249 */
250 for (;;) {
251 set_current_state(TASK_INTERRUPTIBLE);
252 if (pid_ns->pid_allocated == init_pids)
253 break;
254 schedule();
255 }
256 __set_current_state(TASK_RUNNING);
257
258 if (pid_ns->reboot)
259 current->signal->group_exit_code = pid_ns->reboot;
260
261 acct_exit_ns(pid_ns);
262 return;
263}
264
265#ifdef CONFIG_CHECKPOINT_RESTORE
266static int pid_ns_ctl_handler(struct ctl_table *table, int write,
267 void __user *buffer, size_t *lenp, loff_t *ppos)
268{
269 struct pid_namespace *pid_ns = task_active_pid_ns(current);
270 struct ctl_table tmp = *table;
271 int ret, next;
272
273 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
274 return -EPERM;
275
276 /*
277 * Writing directly to ns' last_pid field is OK, since this field
278 * is volatile in a living namespace anyway and a code writing to
279 * it should synchronize its usage with external means.
280 */
281
282 next = idr_get_cursor(&pid_ns->idr) - 1;
283
284 tmp.data = &next;
285 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
286 if (!ret && write)
287 idr_set_cursor(&pid_ns->idr, next + 1);
288
289 return ret;
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 struct ns_common *pidns_for_children_get(struct task_struct *task)
356{
357 struct pid_namespace *ns = NULL;
358
359 task_lock(task);
360 if (task->nsproxy) {
361 ns = task->nsproxy->pid_ns_for_children;
362 get_pid_ns(ns);
363 }
364 task_unlock(task);
365
366 if (ns) {
367 read_lock(&tasklist_lock);
368 if (!ns->child_reaper) {
369 put_pid_ns(ns);
370 ns = NULL;
371 }
372 read_unlock(&tasklist_lock);
373 }
374
375 return ns ? &ns->ns : NULL;
376}
377
378static void pidns_put(struct ns_common *ns)
379{
380 put_pid_ns(to_pid_ns(ns));
381}
382
383static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
384{
385 struct pid_namespace *active = task_active_pid_ns(current);
386 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
387
388 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
389 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
390 return -EPERM;
391
392 /*
393 * Only allow entering the current active pid namespace
394 * or a child of the current active pid namespace.
395 *
396 * This is required for fork to return a usable pid value and
397 * this maintains the property that processes and their
398 * children can not escape their current pid namespace.
399 */
400 if (new->level < active->level)
401 return -EINVAL;
402
403 ancestor = new;
404 while (ancestor->level > active->level)
405 ancestor = ancestor->parent;
406 if (ancestor != active)
407 return -EINVAL;
408
409 put_pid_ns(nsproxy->pid_ns_for_children);
410 nsproxy->pid_ns_for_children = get_pid_ns(new);
411 return 0;
412}
413
414static struct ns_common *pidns_get_parent(struct ns_common *ns)
415{
416 struct pid_namespace *active = task_active_pid_ns(current);
417 struct pid_namespace *pid_ns, *p;
418
419 /* See if the parent is in the current namespace */
420 pid_ns = p = to_pid_ns(ns)->parent;
421 for (;;) {
422 if (!p)
423 return ERR_PTR(-EPERM);
424 if (p == active)
425 break;
426 p = p->parent;
427 }
428
429 return &get_pid_ns(pid_ns)->ns;
430}
431
432static struct user_namespace *pidns_owner(struct ns_common *ns)
433{
434 return to_pid_ns(ns)->user_ns;
435}
436
437const struct proc_ns_operations pidns_operations = {
438 .name = "pid",
439 .type = CLONE_NEWPID,
440 .get = pidns_get,
441 .put = pidns_put,
442 .install = pidns_install,
443 .owner = pidns_owner,
444 .get_parent = pidns_get_parent,
445};
446
447const struct proc_ns_operations pidns_for_children_operations = {
448 .name = "pid_for_children",
449 .real_ns_name = "pid",
450 .type = CLONE_NEWPID,
451 .get = pidns_for_children_get,
452 .put = pidns_put,
453 .install = pidns_install,
454 .owner = pidns_owner,
455 .get_parent = pidns_get_parent,
456};
457
458static __init int pid_namespaces_init(void)
459{
460 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
461
462#ifdef CONFIG_CHECKPOINT_RESTORE
463 register_sysctl_paths(kern_path, pid_ns_ctl_table);
464#endif
465 return 0;
466}
467
468__initcall(pid_namespaces_init);
469