projects/linux·blob v4.13 · refs

490 lines · 11.8 KiB

1/*
2 * Pid namespaces
3 *
4 * Authors:
5 * (C) 2007 Pavel Emelyanov <[email protected]>, OpenVZ, SWsoft Inc.
6 * (C) 2007 Sukadev Bhattiprolu <[email protected]>, 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
25struct pid_cache {
26 int nr_ids;
27 char name[16];
28 struct kmem_cache *cachep;
29 struct list_head list;
30};
31
32static LIST_HEAD(pid_caches_lh);
33static DEFINE_MUTEX(pid_caches_mutex);
34static struct kmem_cache *pid_ns_cachep;
35
36/*
37 * creates the kmem cache to allocate pids from.
38 * @nr_ids: the number of numerical ids this pid will have to carry
39 */
40
41static struct kmem_cache *create_pid_cachep(int nr_ids)
42{
43 struct pid_cache *pcache;
44 struct kmem_cache *cachep;
45
46 mutex_lock(&pid_caches_mutex);
47 list_for_each_entry(pcache, &pid_caches_lh, list)
48 if (pcache->nr_ids == nr_ids)
49 goto out;
50
51 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
52 if (pcache == NULL)
53 goto err_alloc;
54
55 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
56 cachep = kmem_cache_create(pcache->name,
57 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
58 0, SLAB_HWCACHE_ALIGN, NULL);
59 if (cachep == NULL)
60 goto err_cachep;
61
62 pcache->nr_ids = nr_ids;
63 pcache->cachep = cachep;
64 list_add(&pcache->list, &pid_caches_lh);
65out:
66 mutex_unlock(&pid_caches_mutex);
67 return pcache->cachep;
68
69err_cachep:
70 kfree(pcache);
71err_alloc:
72 mutex_unlock(&pid_caches_mutex);
73 return NULL;
74}
75
76static void proc_cleanup_work(struct work_struct *work)
77{
78 struct pid_namespace *ns = container_of(work, struct pid_namespace, proc_work);
79 pid_ns_release_proc(ns);
80}
81
82/* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
83#define MAX_PID_NS_LEVEL 32
84
85static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
86{
87 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
88}
89
90static void dec_pid_namespaces(struct ucounts *ucounts)
91{
92 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
93}
94
95static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
96 struct pid_namespace *parent_pid_ns)
97{
98 struct pid_namespace *ns;
99 unsigned int level = parent_pid_ns->level + 1;
100 struct ucounts *ucounts;
101 int i;
102 int err;
103
104 err = -ENOSPC;
105 if (level > MAX_PID_NS_LEVEL)
106 goto out;
107 ucounts = inc_pid_namespaces(user_ns);
108 if (!ucounts)
109 goto out;
110
111 err = -ENOMEM;
112 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
113 if (ns == NULL)
114 goto out_dec;
115
116 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
117 if (!ns->pidmap[0].page)
118 goto out_free;
119
120 ns->pid_cachep = create_pid_cachep(level + 1);
121 if (ns->pid_cachep == NULL)
122 goto out_free_map;
123
124 err = ns_alloc_inum(&ns->ns);
125 if (err)
126 goto out_free_map;
127 ns->ns.ops = &pidns_operations;
128
129 kref_init(&ns->kref);
130 ns->level = level;
131 ns->parent = get_pid_ns(parent_pid_ns);
132 ns->user_ns = get_user_ns(user_ns);
133 ns->ucounts = ucounts;
134 ns->nr_hashed = PIDNS_HASH_ADDING;
135 INIT_WORK(&ns->proc_work, proc_cleanup_work);
136
137 set_bit(0, ns->pidmap[0].page);
138 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
139
140 for (i = 1; i < PIDMAP_ENTRIES; i++)
141 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
142
143 return ns;
144
145out_free_map:
146 kfree(ns->pidmap[0].page);
147out_free:
148 kmem_cache_free(pid_ns_cachep, ns);
149out_dec:
150 dec_pid_namespaces(ucounts);
151out:
152 return ERR_PTR(err);
153}
154
155static void delayed_free_pidns(struct rcu_head *p)
156{
157 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
158
159 dec_pid_namespaces(ns->ucounts);
160 put_user_ns(ns->user_ns);
161
162 kmem_cache_free(pid_ns_cachep, ns);
163}
164
165static void destroy_pid_namespace(struct pid_namespace *ns)
166{
167 int i;
168
169 ns_free_inum(&ns->ns);
170 for (i = 0; i < PIDMAP_ENTRIES; i++)
171 kfree(ns->pidmap[i].page);
172 call_rcu(&ns->rcu, delayed_free_pidns);
173}
174
175struct pid_namespace *copy_pid_ns(unsigned long 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
185static void free_pid_ns(struct kref *kref)
186{
187 struct pid_namespace *ns;
188
189 ns = container_of(kref, struct pid_namespace, kref);
190 destroy_pid_namespace(ns);
191}
192
193void put_pid_ns(struct pid_namespace *ns)
194{
195 struct pid_namespace *parent;
196
197 while (ns != &init_pid_ns) {
198 parent = ns->parent;
199 if (!kref_put(&ns->kref, free_pid_ns))
200 break;
201 ns = parent;
202 }
203}
204EXPORT_SYMBOL_GPL(put_pid_ns);
205
206void zap_pid_ns_processes(struct pid_namespace *pid_ns)
207{
208 int nr;
209 int rc;
210 struct task_struct *task, *me = current;
211 int init_pids = thread_group_leader(me) ? 1 : 2;
212
213 /* Don't allow any more processes into the pid namespace */
214 disable_pid_allocation(pid_ns);
215
216 /*
217 * Ignore SIGCHLD causing any terminated children to autoreap.
218 * This speeds up the namespace shutdown, plus see the comment
219 * below.
220 */
221 spin_lock_irq(&me->sighand->siglock);
222 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
223 spin_unlock_irq(&me->sighand->siglock);
224
225 /*
226 * The last thread in the cgroup-init thread group is terminating.
227 * Find remaining pid_ts in the namespace, signal and wait for them
228 * to exit.
229 *
230 * Note: This signals each threads in the namespace - even those that
231 * belong to the same thread group, To avoid this, we would have
232 * to walk the entire tasklist looking a processes in this
233 * namespace, but that could be unnecessarily expensive if the
234 * pid namespace has just a few processes. Or we need to
235 * maintain a tasklist for each pid namespace.
236 *
237 */
238 read_lock(&tasklist_lock);
239 nr = next_pidmap(pid_ns, 1);
240 while (nr > 0) {
241 rcu_read_lock();
242
243 task = pid_task(find_vpid(nr), PIDTYPE_PID);
244 if (task && !__fatal_signal_pending(task))
245 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
246
247 rcu_read_unlock();
248
249 nr = next_pidmap(pid_ns, nr);
250 }
251 read_unlock(&tasklist_lock);
252
253 /*
254 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
255 * sys_wait4() will also block until our children traced from the
256 * parent namespace are detached and become EXIT_DEAD.
257 */
258 do {
259 clear_thread_flag(TIF_SIGPENDING);
260 rc = sys_wait4(-1, NULL, __WALL, NULL);
261 } while (rc != -ECHILD);
262
263 /*
264 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
265 * really care, we could reparent them to the global init. We could
266 * exit and reap ->child_reaper even if it is not the last thread in
267 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
268 * pid_ns can not go away until proc_kill_sb() drops the reference.
269 *
270 * But this ns can also have other tasks injected by setns()+fork().
271 * Again, ignoring the user visible semantics we do not really need
272 * to wait until they are all reaped, but they can be reparented to
273 * us and thus we need to ensure that pid->child_reaper stays valid
274 * until they all go away. See free_pid()->wake_up_process().
275 *
276 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
277 * if reparented.
278 */
279 for (;;) {
280 set_current_state(TASK_INTERRUPTIBLE);
281 if (pid_ns->nr_hashed == init_pids)
282 break;
283 schedule();
284 }
285 __set_current_state(TASK_RUNNING);
286
287 if (pid_ns->reboot)
288 current->signal->group_exit_code = pid_ns->reboot;
289
290 acct_exit_ns(pid_ns);
291 return;
292}
293
294#ifdef CONFIG_CHECKPOINT_RESTORE
295static int pid_ns_ctl_handler(struct ctl_table *table, int write,
296 void __user *buffer, size_t *lenp, loff_t *ppos)
297{
298 struct pid_namespace *pid_ns = task_active_pid_ns(current);
299 struct ctl_table tmp = *table;
300
301 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
302 return -EPERM;
303
304 /*
305 * Writing directly to ns' last_pid field is OK, since this field
306 * is volatile in a living namespace anyway and a code writing to
307 * it should synchronize its usage with external means.
308 */
309
310 tmp.data = &pid_ns->last_pid;
311 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
312}
313
314extern int pid_max;
315static int zero = 0;
316static struct ctl_table pid_ns_ctl_table[] = {
317 {
318 .procname = "ns_last_pid",
319 .maxlen = sizeof(int),
320 .mode = 0666, /* permissions are checked in the handler */
321 .proc_handler = pid_ns_ctl_handler,
322 .extra1 = &zero,
323 .extra2 = &pid_max,
324 },
325 { }
326};
327static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
328#endif /* CONFIG_CHECKPOINT_RESTORE */
329
330int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
331{
332 if (pid_ns == &init_pid_ns)
333 return 0;
334
335 switch (cmd) {
336 case LINUX_REBOOT_CMD_RESTART2:
337 case LINUX_REBOOT_CMD_RESTART:
338 pid_ns->reboot = SIGHUP;
339 break;
340
341 case LINUX_REBOOT_CMD_POWER_OFF:
342 case LINUX_REBOOT_CMD_HALT:
343 pid_ns->reboot = SIGINT;
344 break;
345 default:
346 return -EINVAL;
347 }
348
349 read_lock(&tasklist_lock);
350 force_sig(SIGKILL, pid_ns->child_reaper);
351 read_unlock(&tasklist_lock);
352
353 do_exit(0);
354
355 /* Not reached */
356 return 0;
357}
358
359static inline struct pid_namespace *to_pid_ns(struct ns_common *ns)
360{
361 return container_of(ns, struct pid_namespace, ns);
362}
363
364static struct ns_common *pidns_get(struct task_struct *task)
365{
366 struct pid_namespace *ns;
367
368 rcu_read_lock();
369 ns = task_active_pid_ns(task);
370 if (ns)
371 get_pid_ns(ns);
372 rcu_read_unlock();
373
374 return ns ? &ns->ns : NULL;
375}
376
377static struct ns_common *pidns_for_children_get(struct task_struct *task)
378{
379 struct pid_namespace *ns = NULL;
380
381 task_lock(task);
382 if (task->nsproxy) {
383 ns = task->nsproxy->pid_ns_for_children;
384 get_pid_ns(ns);
385 }
386 task_unlock(task);
387
388 if (ns) {
389 read_lock(&tasklist_lock);
390 if (!ns->child_reaper) {
391 put_pid_ns(ns);
392 ns = NULL;
393 }
394 read_unlock(&tasklist_lock);
395 }
396
397 return ns ? &ns->ns : NULL;
398}
399
400static void pidns_put(struct ns_common *ns)
401{
402 put_pid_ns(to_pid_ns(ns));
403}
404
405static int pidns_install(struct nsproxy *nsproxy, struct ns_common *ns)
406{
407 struct pid_namespace *active = task_active_pid_ns(current);
408 struct pid_namespace *ancestor, *new = to_pid_ns(ns);
409
410 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
411 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
412 return -EPERM;
413
414 /*
415 * Only allow entering the current active pid namespace
416 * or a child of the current active pid namespace.
417 *
418 * This is required for fork to return a usable pid value and
419 * this maintains the property that processes and their
420 * children can not escape their current pid namespace.
421 */
422 if (new->level < active->level)
423 return -EINVAL;
424
425 ancestor = new;
426 while (ancestor->level > active->level)
427 ancestor = ancestor->parent;
428 if (ancestor != active)
429 return -EINVAL;
430
431 put_pid_ns(nsproxy->pid_ns_for_children);
432 nsproxy->pid_ns_for_children = get_pid_ns(new);
433 return 0;
434}
435
436static struct ns_common *pidns_get_parent(struct ns_common *ns)
437{
438 struct pid_namespace *active = task_active_pid_ns(current);
439 struct pid_namespace *pid_ns, *p;
440
441 /* See if the parent is in the current namespace */
442 pid_ns = p = to_pid_ns(ns)->parent;
443 for (;;) {
444 if (!p)
445 return ERR_PTR(-EPERM);
446 if (p == active)
447 break;
448 p = p->parent;
449 }
450
451 return &get_pid_ns(pid_ns)->ns;
452}
453
454static struct user_namespace *pidns_owner(struct ns_common *ns)
455{
456 return to_pid_ns(ns)->user_ns;
457}
458
459const struct proc_ns_operations pidns_operations = {
460 .name = "pid",
461 .type = CLONE_NEWPID,
462 .get = pidns_get,
463 .put = pidns_put,
464 .install = pidns_install,
465 .owner = pidns_owner,
466 .get_parent = pidns_get_parent,
467};
468
469const struct proc_ns_operations pidns_for_children_operations = {
470 .name = "pid_for_children",
471 .real_ns_name = "pid",
472 .type = CLONE_NEWPID,
473 .get = pidns_for_children_get,
474 .put = pidns_put,
475 .install = pidns_install,
476 .owner = pidns_owner,
477 .get_parent = pidns_get_parent,
478};
479
480static __init int pid_namespaces_init(void)
481{
482 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
483
484#ifdef CONFIG_CHECKPOINT_RESTORE
485 register_sysctl_paths(kern_path, pid_ns_ctl_table);
486#endif
487 return 0;
488}
489
490__initcall(pid_namespaces_init);