1 // SPDX-License-Identifier: GPL-2.0-only
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
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>
27 static DEFINE_MUTEX(pid_caches_mutex
);
28 static struct kmem_cache
*pid_ns_cachep
;
29 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
30 #define MAX_PID_NS_LEVEL 32
31 /* Write once array, filled from the beginning. */
32 static struct kmem_cache
*pid_cache
[MAX_PID_NS_LEVEL
];
35 * creates the kmem cache to allocate pids from.
36 * @level: pid namespace level
39 static struct kmem_cache
*create_pid_cachep(unsigned int level
)
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];
51 snprintf(name
, sizeof(name
), "pid_%u", level
+ 1);
52 len
= sizeof(struct pid
) + level
* sizeof(struct upid
);
53 mutex_lock(&pid_caches_mutex
);
54 /* Name collision forces to do allocation under mutex. */
56 *pkc
= kmem_cache_create(name
, len
, 0,
57 SLAB_HWCACHE_ALIGN
| SLAB_ACCOUNT
, 0);
58 mutex_unlock(&pid_caches_mutex
);
59 /* current can fail, but someone else can succeed. */
60 return READ_ONCE(*pkc
);
63 static void proc_cleanup_work(struct work_struct
*work
)
65 struct pid_namespace
*ns
= container_of(work
, struct pid_namespace
, proc_work
);
66 pid_ns_release_proc(ns
);
69 static struct ucounts
*inc_pid_namespaces(struct user_namespace
*ns
)
71 return inc_ucount(ns
, current_euid(), UCOUNT_PID_NAMESPACES
);
74 static void dec_pid_namespaces(struct ucounts
*ucounts
)
76 dec_ucount(ucounts
, UCOUNT_PID_NAMESPACES
);
79 static struct pid_namespace
*create_pid_namespace(struct user_namespace
*user_ns
,
80 struct pid_namespace
*parent_pid_ns
)
82 struct pid_namespace
*ns
;
83 unsigned int level
= parent_pid_ns
->level
+ 1;
84 struct ucounts
*ucounts
;
88 if (!in_userns(parent_pid_ns
->user_ns
, user_ns
))
92 if (level
> MAX_PID_NS_LEVEL
)
94 ucounts
= inc_pid_namespaces(user_ns
);
99 ns
= kmem_cache_zalloc(pid_ns_cachep
, GFP_KERNEL
);
105 ns
->pid_cachep
= create_pid_cachep(level
);
106 if (ns
->pid_cachep
== NULL
)
109 err
= ns_alloc_inum(&ns
->ns
);
112 ns
->ns
.ops
= &pidns_operations
;
114 kref_init(&ns
->kref
);
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
->proc_work
, proc_cleanup_work
);
125 idr_destroy(&ns
->idr
);
126 kmem_cache_free(pid_ns_cachep
, ns
);
128 dec_pid_namespaces(ucounts
);
133 static void delayed_free_pidns(struct rcu_head
*p
)
135 struct pid_namespace
*ns
= container_of(p
, struct pid_namespace
, rcu
);
137 dec_pid_namespaces(ns
->ucounts
);
138 put_user_ns(ns
->user_ns
);
140 kmem_cache_free(pid_ns_cachep
, ns
);
143 static void destroy_pid_namespace(struct pid_namespace
*ns
)
145 ns_free_inum(&ns
->ns
);
147 idr_destroy(&ns
->idr
);
148 call_rcu(&ns
->rcu
, delayed_free_pidns
);
151 struct pid_namespace
*copy_pid_ns(unsigned long flags
,
152 struct user_namespace
*user_ns
, struct pid_namespace
*old_ns
)
154 if (!(flags
& CLONE_NEWPID
))
155 return get_pid_ns(old_ns
);
156 if (task_active_pid_ns(current
) != old_ns
)
157 return ERR_PTR(-EINVAL
);
158 return create_pid_namespace(user_ns
, old_ns
);
161 static void free_pid_ns(struct kref
*kref
)
163 struct pid_namespace
*ns
;
165 ns
= container_of(kref
, struct pid_namespace
, kref
);
166 destroy_pid_namespace(ns
);
169 void put_pid_ns(struct pid_namespace
*ns
)
171 struct pid_namespace
*parent
;
173 while (ns
!= &init_pid_ns
) {
175 if (!kref_put(&ns
->kref
, free_pid_ns
))
180 EXPORT_SYMBOL_GPL(put_pid_ns
);
182 void zap_pid_ns_processes(struct pid_namespace
*pid_ns
)
186 struct task_struct
*task
, *me
= current
;
187 int init_pids
= thread_group_leader(me
) ? 1 : 2;
190 /* Don't allow any more processes into the pid namespace */
191 disable_pid_allocation(pid_ns
);
194 * Ignore SIGCHLD causing any terminated children to autoreap.
195 * This speeds up the namespace shutdown, plus see the comment
198 spin_lock_irq(&me
->sighand
->siglock
);
199 me
->sighand
->action
[SIGCHLD
- 1].sa
.sa_handler
= SIG_IGN
;
200 spin_unlock_irq(&me
->sighand
->siglock
);
203 * The last thread in the cgroup-init thread group is terminating.
204 * Find remaining pid_ts in the namespace, signal and wait for them
207 * Note: This signals each threads in the namespace - even those that
208 * belong to the same thread group, To avoid this, we would have
209 * to walk the entire tasklist looking a processes in this
210 * namespace, but that could be unnecessarily expensive if the
211 * pid namespace has just a few processes. Or we need to
212 * maintain a tasklist for each pid namespace.
216 read_lock(&tasklist_lock
);
218 idr_for_each_entry_continue(&pid_ns
->idr
, pid
, nr
) {
219 task
= pid_task(pid
, PIDTYPE_PID
);
220 if (task
&& !__fatal_signal_pending(task
))
221 group_send_sig_info(SIGKILL
, SEND_SIG_PRIV
, task
, PIDTYPE_MAX
);
223 read_unlock(&tasklist_lock
);
227 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
228 * kernel_wait4() will also block until our children traced from the
229 * parent namespace are detached and become EXIT_DEAD.
232 clear_thread_flag(TIF_SIGPENDING
);
233 rc
= kernel_wait4(-1, NULL
, __WALL
, NULL
);
234 } while (rc
!= -ECHILD
);
237 * kernel_wait4() above can't reap the EXIT_DEAD children but we do not
238 * really care, we could reparent them to the global init. We could
239 * exit and reap ->child_reaper even if it is not the last thread in
240 * this pid_ns, free_pid(pid_allocated == 0) calls proc_cleanup_work(),
241 * pid_ns can not go away until proc_kill_sb() drops the reference.
243 * But this ns can also have other tasks injected by setns()+fork().
244 * Again, ignoring the user visible semantics we do not really need
245 * to wait until they are all reaped, but they can be reparented to
246 * us and thus we need to ensure that pid->child_reaper stays valid
247 * until they all go away. See free_pid()->wake_up_process().
249 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
253 set_current_state(TASK_INTERRUPTIBLE
);
254 if (pid_ns
->pid_allocated
== init_pids
)
258 __set_current_state(TASK_RUNNING
);
261 current
->signal
->group_exit_code
= pid_ns
->reboot
;
263 acct_exit_ns(pid_ns
);
267 #ifdef CONFIG_CHECKPOINT_RESTORE
268 static int pid_ns_ctl_handler(struct ctl_table
*table
, int write
,
269 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
271 struct pid_namespace
*pid_ns
= task_active_pid_ns(current
);
272 struct ctl_table tmp
= *table
;
275 if (write
&& !ns_capable(pid_ns
->user_ns
, CAP_SYS_ADMIN
))
279 * Writing directly to ns' last_pid field is OK, since this field
280 * is volatile in a living namespace anyway and a code writing to
281 * it should synchronize its usage with external means.
284 next
= idr_get_cursor(&pid_ns
->idr
) - 1;
287 ret
= proc_dointvec_minmax(&tmp
, write
, buffer
, lenp
, ppos
);
289 idr_set_cursor(&pid_ns
->idr
, next
+ 1);
295 static struct ctl_table pid_ns_ctl_table
[] = {
297 .procname
= "ns_last_pid",
298 .maxlen
= sizeof(int),
299 .mode
= 0666, /* permissions are checked in the handler */
300 .proc_handler
= pid_ns_ctl_handler
,
301 .extra1
= SYSCTL_ZERO
,
306 static struct ctl_path kern_path
[] = { { .procname
= "kernel", }, { } };
307 #endif /* CONFIG_CHECKPOINT_RESTORE */
309 int reboot_pid_ns(struct pid_namespace
*pid_ns
, int cmd
)
311 if (pid_ns
== &init_pid_ns
)
315 case LINUX_REBOOT_CMD_RESTART2
:
316 case LINUX_REBOOT_CMD_RESTART
:
317 pid_ns
->reboot
= SIGHUP
;
320 case LINUX_REBOOT_CMD_POWER_OFF
:
321 case LINUX_REBOOT_CMD_HALT
:
322 pid_ns
->reboot
= SIGINT
;
328 read_lock(&tasklist_lock
);
329 send_sig(SIGKILL
, pid_ns
->child_reaper
, 1);
330 read_unlock(&tasklist_lock
);
338 static inline struct pid_namespace
*to_pid_ns(struct ns_common
*ns
)
340 return container_of(ns
, struct pid_namespace
, ns
);
343 static struct ns_common
*pidns_get(struct task_struct
*task
)
345 struct pid_namespace
*ns
;
348 ns
= task_active_pid_ns(task
);
353 return ns
? &ns
->ns
: NULL
;
356 static struct ns_common
*pidns_for_children_get(struct task_struct
*task
)
358 struct pid_namespace
*ns
= NULL
;
362 ns
= task
->nsproxy
->pid_ns_for_children
;
368 read_lock(&tasklist_lock
);
369 if (!ns
->child_reaper
) {
373 read_unlock(&tasklist_lock
);
376 return ns
? &ns
->ns
: NULL
;
379 static void pidns_put(struct ns_common
*ns
)
381 put_pid_ns(to_pid_ns(ns
));
384 static int pidns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
386 struct pid_namespace
*active
= task_active_pid_ns(current
);
387 struct pid_namespace
*ancestor
, *new = to_pid_ns(ns
);
389 if (!ns_capable(new->user_ns
, CAP_SYS_ADMIN
) ||
390 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
394 * Only allow entering the current active pid namespace
395 * or a child of the current active pid namespace.
397 * This is required for fork to return a usable pid value and
398 * this maintains the property that processes and their
399 * children can not escape their current pid namespace.
401 if (new->level
< active
->level
)
405 while (ancestor
->level
> active
->level
)
406 ancestor
= ancestor
->parent
;
407 if (ancestor
!= active
)
410 put_pid_ns(nsproxy
->pid_ns_for_children
);
411 nsproxy
->pid_ns_for_children
= get_pid_ns(new);
415 static struct ns_common
*pidns_get_parent(struct ns_common
*ns
)
417 struct pid_namespace
*active
= task_active_pid_ns(current
);
418 struct pid_namespace
*pid_ns
, *p
;
420 /* See if the parent is in the current namespace */
421 pid_ns
= p
= to_pid_ns(ns
)->parent
;
424 return ERR_PTR(-EPERM
);
430 return &get_pid_ns(pid_ns
)->ns
;
433 static struct user_namespace
*pidns_owner(struct ns_common
*ns
)
435 return to_pid_ns(ns
)->user_ns
;
438 const struct proc_ns_operations pidns_operations
= {
440 .type
= CLONE_NEWPID
,
443 .install
= pidns_install
,
444 .owner
= pidns_owner
,
445 .get_parent
= pidns_get_parent
,
448 const struct proc_ns_operations pidns_for_children_operations
= {
449 .name
= "pid_for_children",
450 .real_ns_name
= "pid",
451 .type
= CLONE_NEWPID
,
452 .get
= pidns_for_children_get
,
454 .install
= pidns_install
,
455 .owner
= pidns_owner
,
456 .get_parent
= pidns_get_parent
,
459 static __init
int pid_namespaces_init(void)
461 pid_ns_cachep
= KMEM_CACHE(pid_namespace
, SLAB_PANIC
);
463 #ifdef CONFIG_CHECKPOINT_RESTORE
464 register_sysctl_paths(kern_path
, pid_ns_ctl_table
);
469 __initcall(pid_namespaces_init
);