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
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/err.h>
16 #include <linux/acct.h>
17 #include <linux/slab.h>
18 #include <linux/proc_ns.h>
19 #include <linux/reboot.h>
20 #include <linux/export.h>
25 struct kmem_cache
*cachep
;
26 struct list_head list
;
29 static LIST_HEAD(pid_caches_lh
);
30 static DEFINE_MUTEX(pid_caches_mutex
);
31 static struct kmem_cache
*pid_ns_cachep
;
34 * creates the kmem cache to allocate pids from.
35 * @nr_ids: the number of numerical ids this pid will have to carry
38 static struct kmem_cache
*create_pid_cachep(int nr_ids
)
40 struct pid_cache
*pcache
;
41 struct kmem_cache
*cachep
;
43 mutex_lock(&pid_caches_mutex
);
44 list_for_each_entry(pcache
, &pid_caches_lh
, list
)
45 if (pcache
->nr_ids
== nr_ids
)
48 pcache
= kmalloc(sizeof(struct pid_cache
), GFP_KERNEL
);
52 snprintf(pcache
->name
, sizeof(pcache
->name
), "pid_%d", nr_ids
);
53 cachep
= kmem_cache_create(pcache
->name
,
54 sizeof(struct pid
) + (nr_ids
- 1) * sizeof(struct upid
),
55 0, SLAB_HWCACHE_ALIGN
, NULL
);
59 pcache
->nr_ids
= nr_ids
;
60 pcache
->cachep
= cachep
;
61 list_add(&pcache
->list
, &pid_caches_lh
);
63 mutex_unlock(&pid_caches_mutex
);
64 return pcache
->cachep
;
69 mutex_unlock(&pid_caches_mutex
);
73 static void proc_cleanup_work(struct work_struct
*work
)
75 struct pid_namespace
*ns
= container_of(work
, struct pid_namespace
, proc_work
);
76 pid_ns_release_proc(ns
);
79 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
80 #define MAX_PID_NS_LEVEL 32
82 static struct pid_namespace
*create_pid_namespace(struct user_namespace
*user_ns
,
83 struct pid_namespace
*parent_pid_ns
)
85 struct pid_namespace
*ns
;
86 unsigned int level
= parent_pid_ns
->level
+ 1;
90 if (level
> MAX_PID_NS_LEVEL
) {
96 ns
= kmem_cache_zalloc(pid_ns_cachep
, GFP_KERNEL
);
100 ns
->pidmap
[0].page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
101 if (!ns
->pidmap
[0].page
)
104 ns
->pid_cachep
= create_pid_cachep(level
+ 1);
105 if (ns
->pid_cachep
== NULL
)
108 err
= proc_alloc_inum(&ns
->proc_inum
);
112 kref_init(&ns
->kref
);
114 ns
->parent
= get_pid_ns(parent_pid_ns
);
115 ns
->user_ns
= get_user_ns(user_ns
);
116 ns
->nr_hashed
= PIDNS_HASH_ADDING
;
117 INIT_WORK(&ns
->proc_work
, proc_cleanup_work
);
119 set_bit(0, ns
->pidmap
[0].page
);
120 atomic_set(&ns
->pidmap
[0].nr_free
, BITS_PER_PAGE
- 1);
122 for (i
= 1; i
< PIDMAP_ENTRIES
; i
++)
123 atomic_set(&ns
->pidmap
[i
].nr_free
, BITS_PER_PAGE
);
128 kfree(ns
->pidmap
[0].page
);
130 kmem_cache_free(pid_ns_cachep
, ns
);
135 static void delayed_free_pidns(struct rcu_head
*p
)
137 kmem_cache_free(pid_ns_cachep
,
138 container_of(p
, struct pid_namespace
, rcu
));
141 static void destroy_pid_namespace(struct pid_namespace
*ns
)
145 proc_free_inum(ns
->proc_inum
);
146 for (i
= 0; i
< PIDMAP_ENTRIES
; i
++)
147 kfree(ns
->pidmap
[i
].page
);
148 put_user_ns(ns
->user_ns
);
149 call_rcu(&ns
->rcu
, delayed_free_pidns
);
152 struct pid_namespace
*copy_pid_ns(unsigned long flags
,
153 struct user_namespace
*user_ns
, struct pid_namespace
*old_ns
)
155 if (!(flags
& CLONE_NEWPID
))
156 return get_pid_ns(old_ns
);
157 if (task_active_pid_ns(current
) != old_ns
)
158 return ERR_PTR(-EINVAL
);
159 return create_pid_namespace(user_ns
, old_ns
);
162 static void free_pid_ns(struct kref
*kref
)
164 struct pid_namespace
*ns
;
166 ns
= container_of(kref
, struct pid_namespace
, kref
);
167 destroy_pid_namespace(ns
);
170 void put_pid_ns(struct pid_namespace
*ns
)
172 struct pid_namespace
*parent
;
174 while (ns
!= &init_pid_ns
) {
176 if (!kref_put(&ns
->kref
, free_pid_ns
))
181 EXPORT_SYMBOL_GPL(put_pid_ns
);
183 void zap_pid_ns_processes(struct pid_namespace
*pid_ns
)
187 struct task_struct
*task
, *me
= current
;
188 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.
215 read_lock(&tasklist_lock
);
216 nr
= next_pidmap(pid_ns
, 1);
220 task
= pid_task(find_vpid(nr
), PIDTYPE_PID
);
221 if (task
&& !__fatal_signal_pending(task
))
222 send_sig_info(SIGKILL
, SEND_SIG_FORCED
, task
);
226 nr
= next_pidmap(pid_ns
, nr
);
228 read_unlock(&tasklist_lock
);
231 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
232 * sys_wait4() will also block until our children traced from the
233 * parent namespace are detached and become EXIT_DEAD.
236 clear_thread_flag(TIF_SIGPENDING
);
237 rc
= sys_wait4(-1, NULL
, __WALL
, NULL
);
238 } while (rc
!= -ECHILD
);
241 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
242 * really care, we could reparent them to the global init. We could
243 * exit and reap ->child_reaper even if it is not the last thread in
244 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
245 * pid_ns can not go away until proc_kill_sb() drops the reference.
247 * But this ns can also have other tasks injected by setns()+fork().
248 * Again, ignoring the user visible semantics we do not really need
249 * to wait until they are all reaped, but they can be reparented to
250 * us and thus we need to ensure that pid->child_reaper stays valid
251 * until they all go away. See free_pid()->wake_up_process().
253 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
257 set_current_state(TASK_UNINTERRUPTIBLE
);
258 if (pid_ns
->nr_hashed
== init_pids
)
262 __set_current_state(TASK_RUNNING
);
265 current
->signal
->group_exit_code
= pid_ns
->reboot
;
267 acct_exit_ns(pid_ns
);
271 #ifdef CONFIG_CHECKPOINT_RESTORE
272 static int pid_ns_ctl_handler(struct ctl_table
*table
, int write
,
273 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
275 struct pid_namespace
*pid_ns
= task_active_pid_ns(current
);
276 struct ctl_table tmp
= *table
;
278 if (write
&& !ns_capable(pid_ns
->user_ns
, CAP_SYS_ADMIN
))
282 * Writing directly to ns' last_pid field is OK, since this field
283 * is volatile in a living namespace anyway and a code writing to
284 * it should synchronize its usage with external means.
287 tmp
.data
= &pid_ns
->last_pid
;
288 return proc_dointvec_minmax(&tmp
, write
, buffer
, lenp
, ppos
);
293 static struct ctl_table pid_ns_ctl_table
[] = {
295 .procname
= "ns_last_pid",
296 .maxlen
= sizeof(int),
297 .mode
= 0666, /* permissions are checked in the handler */
298 .proc_handler
= pid_ns_ctl_handler
,
304 static struct ctl_path kern_path
[] = { { .procname
= "kernel", }, { } };
305 #endif /* CONFIG_CHECKPOINT_RESTORE */
307 int reboot_pid_ns(struct pid_namespace
*pid_ns
, int cmd
)
309 if (pid_ns
== &init_pid_ns
)
313 case LINUX_REBOOT_CMD_RESTART2
:
314 case LINUX_REBOOT_CMD_RESTART
:
315 pid_ns
->reboot
= SIGHUP
;
318 case LINUX_REBOOT_CMD_POWER_OFF
:
319 case LINUX_REBOOT_CMD_HALT
:
320 pid_ns
->reboot
= SIGINT
;
326 read_lock(&tasklist_lock
);
327 force_sig(SIGKILL
, pid_ns
->child_reaper
);
328 read_unlock(&tasklist_lock
);
336 static void *pidns_get(struct task_struct
*task
)
338 struct pid_namespace
*ns
;
341 ns
= task_active_pid_ns(task
);
349 static void pidns_put(void *ns
)
354 static int pidns_install(struct nsproxy
*nsproxy
, void *ns
)
356 struct pid_namespace
*active
= task_active_pid_ns(current
);
357 struct pid_namespace
*ancestor
, *new = ns
;
359 if (!ns_capable(new->user_ns
, CAP_SYS_ADMIN
) ||
360 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
364 * Only allow entering the current active pid namespace
365 * or a child of the current active pid namespace.
367 * This is required for fork to return a usable pid value and
368 * this maintains the property that processes and their
369 * children can not escape their current pid namespace.
371 if (new->level
< active
->level
)
375 while (ancestor
->level
> active
->level
)
376 ancestor
= ancestor
->parent
;
377 if (ancestor
!= active
)
380 put_pid_ns(nsproxy
->pid_ns_for_children
);
381 nsproxy
->pid_ns_for_children
= get_pid_ns(new);
385 static unsigned int pidns_inum(void *ns
)
387 struct pid_namespace
*pid_ns
= ns
;
388 return pid_ns
->proc_inum
;
391 const struct proc_ns_operations pidns_operations
= {
393 .type
= CLONE_NEWPID
,
396 .install
= pidns_install
,
400 static __init
int pid_namespaces_init(void)
402 pid_ns_cachep
= KMEM_CACHE(pid_namespace
, SLAB_PANIC
);
404 #ifdef CONFIG_CHECKPOINT_RESTORE
405 register_sysctl_paths(kern_path
, pid_ns_ctl_table
);
410 __initcall(pid_namespaces_init
);