projects/linux·blob master · refs

474 lines · 11.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Pid namespaces
4 *
5 * Authors:
6 * (C) 2007 Pavel Emelyanov <[email protected]>, OpenVZ, SWsoft Inc.
7 * (C) 2007 Sukadev Bhattiprolu <[email protected]>, 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 <linux/nstree.h>
27#include <uapi/linux/wait.h>
28#include "pid_sysctl.h"
29
30static DEFINE_MUTEX(pid_caches_mutex);
31static struct kmem_cache *pid_ns_cachep;
32/* Write once array, filled from the beginning. */
33static struct kmem_cache *pid_cache[MAX_PID_NS_LEVEL];
34
35/*
36 * creates the kmem cache to allocate pids from.
37 * @level: pid namespace level
38 */
39
40static struct kmem_cache *create_pid_cachep(unsigned int level)
41{
42 /* Level 0 is init_pid_ns.pid_cachep */
43 struct kmem_cache **pkc = &pid_cache[level - 1];
44 struct kmem_cache *kc;
45 char name[4 + 10 + 1];
46 unsigned int len;
47
48 kc = READ_ONCE(*pkc);
49 if (kc)
50 return kc;
51
52 snprintf(name, sizeof(name), "pid_%u", level + 1);
53 len = struct_size_t(struct pid, numbers, level + 1);
54 mutex_lock(&pid_caches_mutex);
55 /* Name collision forces to do allocation under mutex. */
56 if (!*pkc)
57 *pkc = kmem_cache_create(name, len, 0,
58 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
59 mutex_unlock(&pid_caches_mutex);
60 /* current can fail, but someone else can succeed. */
61 return READ_ONCE(*pkc);
62}
63
64static struct ucounts *inc_pid_namespaces(struct user_namespace *ns)
65{
66 return inc_ucount(ns, current_euid(), UCOUNT_PID_NAMESPACES);
67}
68
69static void dec_pid_namespaces(struct ucounts *ucounts)
70{
71 dec_ucount(ucounts, UCOUNT_PID_NAMESPACES);
72}
73
74static void destroy_pid_namespace_work(struct work_struct *work);
75
76static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
77 struct pid_namespace *parent_pid_ns)
78{
79 struct pid_namespace *ns;
80 unsigned int level = parent_pid_ns->level + 1;
81 struct ucounts *ucounts;
82 int err;
83
84 err = -EINVAL;
85 if (!in_userns(parent_pid_ns->user_ns, user_ns))
86 goto out;
87
88 err = -ENOSPC;
89 if (level > MAX_PID_NS_LEVEL)
90 goto out;
91 ucounts = inc_pid_namespaces(user_ns);
92 if (!ucounts)
93 goto out;
94
95 err = -ENOMEM;
96 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
97 if (ns == NULL)
98 goto out_dec;
99
100 idr_init(&ns->idr);
101
102 ns->pid_cachep = create_pid_cachep(level);
103 if (ns->pid_cachep == NULL)
104 goto out_free_idr;
105
106 err = ns_common_init(ns);
107 if (err)
108 goto out_free_idr;
109
110 ns->pid_max = PID_MAX_LIMIT;
111 err = register_pidns_sysctls(ns);
112 if (err)
113 goto out_free_inum;
114
115 ns->level = level;
116 ns->parent = get_pid_ns(parent_pid_ns);
117 ns->user_ns = get_user_ns(user_ns);
118 ns->ucounts = ucounts;
119 ns->pid_allocated = PIDNS_ADDING;
120 INIT_WORK(&ns->work, destroy_pid_namespace_work);
121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_MEMFD_CREATE)
123 ns->memfd_noexec_scope = pidns_memfd_noexec_scope(parent_pid_ns);
124#endif
125
126 ns_tree_add(ns);
127 return ns;
128
129out_free_inum:
130 ns_common_free(ns);
131out_free_idr:
132 idr_destroy(&ns->idr);
133 kmem_cache_free(pid_ns_cachep, ns);
134out_dec:
135 dec_pid_namespaces(ucounts);
136out:
137 return ERR_PTR(err);
138}
139
140static void delayed_free_pidns(struct rcu_head *p)
141{
142 struct pid_namespace *ns = container_of(p, struct pid_namespace, rcu);
143
144 dec_pid_namespaces(ns->ucounts);
145 put_user_ns(ns->user_ns);
146
147 kmem_cache_free(pid_ns_cachep, ns);
148}
149
150static void destroy_pid_namespace(struct pid_namespace *ns)
151{
152 ns_tree_remove(ns);
153 unregister_pidns_sysctls(ns);
154
155 ns_common_free(ns);
156
157 idr_destroy(&ns->idr);
158 call_rcu(&ns->rcu, delayed_free_pidns);
159}
160
161static void destroy_pid_namespace_work(struct work_struct *work)
162{
163 struct pid_namespace *ns =
164 container_of(work, struct pid_namespace, work);
165
166 do {
167 struct pid_namespace *parent;
168
169 parent = ns->parent;
170 destroy_pid_namespace(ns);
171 ns = parent;
172 } while (ns != &init_pid_ns && ns_ref_put(ns));
173}
174
175struct pid_namespace *copy_pid_ns(u64 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
185void put_pid_ns(struct pid_namespace *ns)
186{
187 if (ns && ns_ref_put(ns))
188 schedule_work(&ns->work);
189}
190EXPORT_SYMBOL_GPL(put_pid_ns);
191
192void zap_pid_ns_processes(struct pid_namespace *pid_ns)
193{
194 int nr;
195 int rc;
196 struct task_struct *task, *me = current;
197 int init_pids = thread_group_leader(me) ? 1 : 2;
198 struct pid *pid;
199
200 /* Don't allow any more processes into the pid namespace */
201 disable_pid_allocation(pid_ns);
202
203 /*
204 * Ignore SIGCHLD causing any terminated children to autoreap.
205 * This speeds up the namespace shutdown, plus see the comment
206 * below.
207 */
208 spin_lock_irq(&me->sighand->siglock);
209 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
210 spin_unlock_irq(&me->sighand->siglock);
211
212 /*
213 * The last thread in the cgroup-init thread group is terminating.
214 * Find remaining pid_ts in the namespace, signal and wait for them
215 * to exit.
216 *
217 * Note: This signals each threads in the namespace - even those that
218 * belong to the same thread group, To avoid this, we would have
219 * to walk the entire tasklist looking a processes in this
220 * namespace, but that could be unnecessarily expensive if the
221 * pid namespace has just a few processes. Or we need to
222 * maintain a tasklist for each pid namespace.
223 *
224 */
225 rcu_read_lock();
226 read_lock(&tasklist_lock);
227 nr = 2;
228 idr_for_each_entry_continue(&pid_ns->idr, pid, nr) {
229 task = pid_task(pid, PIDTYPE_PID);
230 if (task && !__fatal_signal_pending(task))
231 group_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_MAX);
232 }
233 read_unlock(&tasklist_lock);
234 rcu_read_unlock();
235
236 /*
237 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
238 * kernel_wait4() will also block until our children traced from the
239 * parent namespace are detached and become EXIT_DEAD.
240 */
241 /* Task work must not busy-loop the reaper, see signal_pending(). */
242 guard(no_notify_signal)();
243 do {
244 clear_thread_flag(TIF_SIGPENDING);
245 rc = kernel_wait4(-1, NULL, __WALL, NULL);
246 } while (rc != -ECHILD);
247
248 /*
249 * kernel_wait4() misses EXIT_DEAD children, and EXIT_ZOMBIE
250 * process whose parents processes are outside of the pid
251 * namespace. Such processes are created with setns()+fork().
252 *
253 * If those EXIT_ZOMBIE processes are not reaped by their
254 * parents before their parents exit, they will be reparented
255 * to pid_ns->child_reaper. Thus pidns->child_reaper needs to
256 * stay valid until they all go away.
257 *
258 * The code relies on the pid_ns->child_reaper ignoring
259 * SIGCHILD to cause those EXIT_ZOMBIE processes to be
260 * autoreaped if reparented.
261 *
262 * Semantically it is also desirable to wait for EXIT_ZOMBIE
263 * processes before allowing the child_reaper to be reaped, as
264 * that gives the invariant that when the init process of a
265 * pid namespace is reaped all of the processes in the pid
266 * namespace are gone.
267 *
268 * Once all of the other tasks are gone from the pid_namespace
269 * free_pid() will awaken this task.
270 */
271 for (;;) {
272 set_current_state(TASK_IDLE | TASK_FREEZABLE);
273 if (pid_ns->pid_allocated == init_pids)
274 break;
275 schedule();
276 }
277 __set_current_state(TASK_RUNNING);
278
279 if (pid_ns->reboot)
280 current->signal->group_exit_code = pid_ns->reboot;
281
282 acct_exit_ns(pid_ns);
283 return;
284}
285
286#ifdef CONFIG_CHECKPOINT_RESTORE
287static int pid_ns_ctl_handler(const struct ctl_table *table, int write,
288 void *buffer, size_t *lenp, loff_t *ppos)
289{
290 struct pid_namespace *pid_ns = task_active_pid_ns(current);
291 struct ctl_table tmp = *table;
292 int ret, next;
293
294 if (write && !checkpoint_restore_ns_capable(pid_ns->user_ns))
295 return -EPERM;
296
297 next = idr_get_cursor(&pid_ns->idr) - 1;
298
299 tmp.data = &next;
300 tmp.extra2 = &pid_ns->pid_max;
301 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
302 if (!ret && write)
303 idr_set_cursor(&pid_ns->idr, next + 1);
304
305 return ret;
306}
307
308static const struct ctl_table pid_ns_ctl_table[] = {
309 {
310 .procname = "ns_last_pid",
311 .maxlen = sizeof(int),
312 .mode = 0666, /* permissions are checked in the handler */
313 .proc_handler = pid_ns_ctl_handler,
314 .extra1 = SYSCTL_ZERO,
315 .extra2 = &init_pid_ns.pid_max,
316 },
317};
318#endif /* CONFIG_CHECKPOINT_RESTORE */
319
320int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
321{
322 if (pid_ns == &init_pid_ns)
323 return 0;
324
325 switch (cmd) {
326 case LINUX_REBOOT_CMD_RESTART2:
327 case LINUX_REBOOT_CMD_RESTART:
328 pid_ns->reboot = SIGHUP;
329 break;
330
331 case LINUX_REBOOT_CMD_POWER_OFF:
332 case LINUX_REBOOT_CMD_HALT:
333 pid_ns->reboot = SIGINT;
334 break;
335 default:
336 return -EINVAL;
337 }
338
339 read_lock(&tasklist_lock);
340 send_sig(SIGKILL, pid_ns->child_reaper, 1);
341 read_unlock(&tasklist_lock);
342
343 do_exit(0);
344
345 /* Not reached */
346 return 0;
347}
348
349static struct ns_common *pidns_get(struct task_struct *task)
350{
351 struct pid_namespace *ns;
352
353 rcu_read_lock();
354 ns = task_active_pid_ns(task);
355 if (ns)
356 get_pid_ns(ns);
357 rcu_read_unlock();
358
359 return ns ? &ns->ns : NULL;
360}
361
362static struct ns_common *pidns_for_children_get(struct task_struct *task)
363{
364 struct pid_namespace *ns = NULL;
365
366 task_lock(task);
367 if (task->nsproxy) {
368 ns = task->nsproxy->pid_ns_for_children;
369 get_pid_ns(ns);
370 }
371 task_unlock(task);
372
373 return ns ? &ns->ns : NULL;
374}
375
376static void pidns_put(struct ns_common *ns)
377{
378 put_pid_ns(to_pid_ns(ns));
379}
380
381bool pidns_is_ancestor(struct pid_namespace *child,
382 struct pid_namespace *ancestor)
383{
384 struct pid_namespace *ns;
385
386 if (child->level < ancestor->level)
387 return false;
388 for (ns = child; ns->level > ancestor->level; ns = ns->parent)
389 ;
390 return ns == ancestor;
391}
392
393static int pidns_install(struct nsset *nsset, struct ns_common *ns)
394{
395 struct nsproxy *nsproxy = nsset->nsproxy;
396 struct pid_namespace *active = task_active_pid_ns(current);
397 struct pid_namespace *new = to_pid_ns(ns);
398
399 if (!ns_capable(new->user_ns, CAP_SYS_ADMIN) ||
400 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
401 return -EPERM;
402
403 /*
404 * Only allow entering the current active pid namespace
405 * or a child of the current active pid namespace.
406 *
407 * This is required for fork to return a usable pid value and
408 * this maintains the property that processes and their
409 * children can not escape their current pid namespace.
410 */
411 if (!pidns_is_ancestor(new, active))
412 return -EINVAL;
413
414 put_pid_ns(nsproxy->pid_ns_for_children);
415 nsproxy->pid_ns_for_children = get_pid_ns(new);
416 return 0;
417}
418
419static struct ns_common *pidns_get_parent(struct ns_common *ns)
420{
421 struct pid_namespace *active = task_active_pid_ns(current);
422 struct pid_namespace *pid_ns, *p;
423
424 /* See if the parent is in the current namespace */
425 pid_ns = p = to_pid_ns(ns)->parent;
426 for (;;) {
427 if (!p)
428 return ERR_PTR(-EPERM);
429 if (p == active)
430 break;
431 p = p->parent;
432 }
433
434 return &get_pid_ns(pid_ns)->ns;
435}
436
437static struct user_namespace *pidns_owner(struct ns_common *ns)
438{
439 return to_pid_ns(ns)->user_ns;
440}
441
442const struct proc_ns_operations pidns_operations = {
443 .name = "pid",
444 .get = pidns_get,
445 .put = pidns_put,
446 .install = pidns_install,
447 .owner = pidns_owner,
448 .get_parent = pidns_get_parent,
449};
450
451const struct proc_ns_operations pidns_for_children_operations = {
452 .name = "pid_for_children",
453 .real_ns_name = "pid",
454 .get = pidns_for_children_get,
455 .put = pidns_put,
456 .install = pidns_install,
457 .owner = pidns_owner,
458 .get_parent = pidns_get_parent,
459};
460
461static __init int pid_namespaces_init(void)
462{
463 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC | SLAB_ACCOUNT);
464
465#ifdef CONFIG_CHECKPOINT_RESTORE
466 register_sysctl_init("kernel", pid_ns_ctl_table);
467#endif
468
469 register_pid_ns_sysctl_table_vm();
470 ns_tree_add(&init_pid_ns);
471 return 0;
472}
473
474__initcall(pid_namespaces_init);