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/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>
26 struct kmem_cache
*cachep
;
27 struct list_head list
;
30 static LIST_HEAD(pid_caches_lh
);
31 static DEFINE_MUTEX(pid_caches_mutex
);
32 static struct kmem_cache
*pid_ns_cachep
;
35 * creates the kmem cache to allocate pids from.
36 * @nr_ids: the number of numerical ids this pid will have to carry
39 static struct kmem_cache
*create_pid_cachep(int nr_ids
)
41 struct pid_cache
*pcache
;
42 struct kmem_cache
*cachep
;
44 mutex_lock(&pid_caches_mutex
);
45 list_for_each_entry(pcache
, &pid_caches_lh
, list
)
46 if (pcache
->nr_ids
== nr_ids
)
49 pcache
= kmalloc(sizeof(struct pid_cache
), GFP_KERNEL
);
53 snprintf(pcache
->name
, sizeof(pcache
->name
), "pid_%d", nr_ids
);
54 cachep
= kmem_cache_create(pcache
->name
,
55 sizeof(struct pid
) + (nr_ids
- 1) * sizeof(struct upid
),
56 0, SLAB_HWCACHE_ALIGN
, NULL
);
60 pcache
->nr_ids
= nr_ids
;
61 pcache
->cachep
= cachep
;
62 list_add(&pcache
->list
, &pid_caches_lh
);
64 mutex_unlock(&pid_caches_mutex
);
65 return pcache
->cachep
;
70 mutex_unlock(&pid_caches_mutex
);
74 static void proc_cleanup_work(struct work_struct
*work
)
76 struct pid_namespace
*ns
= container_of(work
, struct pid_namespace
, proc_work
);
77 pid_ns_release_proc(ns
);
80 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
81 #define MAX_PID_NS_LEVEL 32
83 static struct ucounts
*inc_pid_namespaces(struct user_namespace
*ns
)
85 return inc_ucount(ns
, current_euid(), UCOUNT_PID_NAMESPACES
);
88 static void dec_pid_namespaces(struct ucounts
*ucounts
)
90 dec_ucount(ucounts
, UCOUNT_PID_NAMESPACES
);
93 static struct pid_namespace
*create_pid_namespace(struct user_namespace
*user_ns
,
94 struct pid_namespace
*parent_pid_ns
)
96 struct pid_namespace
*ns
;
97 unsigned int level
= parent_pid_ns
->level
+ 1;
98 struct ucounts
*ucounts
;
103 if (level
> MAX_PID_NS_LEVEL
)
105 ucounts
= inc_pid_namespaces(user_ns
);
110 ns
= kmem_cache_zalloc(pid_ns_cachep
, GFP_KERNEL
);
114 ns
->pidmap
[0].page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
115 if (!ns
->pidmap
[0].page
)
118 ns
->pid_cachep
= create_pid_cachep(level
+ 1);
119 if (ns
->pid_cachep
== NULL
)
122 err
= ns_alloc_inum(&ns
->ns
);
125 ns
->ns
.ops
= &pidns_operations
;
127 kref_init(&ns
->kref
);
129 ns
->parent
= get_pid_ns(parent_pid_ns
);
130 ns
->user_ns
= get_user_ns(user_ns
);
131 ns
->ucounts
= ucounts
;
132 ns
->nr_hashed
= PIDNS_HASH_ADDING
;
133 INIT_WORK(&ns
->proc_work
, proc_cleanup_work
);
135 set_bit(0, ns
->pidmap
[0].page
);
136 atomic_set(&ns
->pidmap
[0].nr_free
, BITS_PER_PAGE
- 1);
138 for (i
= 1; i
< PIDMAP_ENTRIES
; i
++)
139 atomic_set(&ns
->pidmap
[i
].nr_free
, BITS_PER_PAGE
);
144 kfree(ns
->pidmap
[0].page
);
146 kmem_cache_free(pid_ns_cachep
, ns
);
148 dec_pid_namespaces(ucounts
);
153 static void delayed_free_pidns(struct rcu_head
*p
)
155 struct pid_namespace
*ns
= container_of(p
, struct pid_namespace
, rcu
);
157 dec_pid_namespaces(ns
->ucounts
);
158 put_user_ns(ns
->user_ns
);
160 kmem_cache_free(pid_ns_cachep
, ns
);
163 static void destroy_pid_namespace(struct pid_namespace
*ns
)
167 ns_free_inum(&ns
->ns
);
168 for (i
= 0; i
< PIDMAP_ENTRIES
; i
++)
169 kfree(ns
->pidmap
[i
].page
);
170 call_rcu(&ns
->rcu
, delayed_free_pidns
);
173 struct pid_namespace
*copy_pid_ns(unsigned long flags
,
174 struct user_namespace
*user_ns
, struct pid_namespace
*old_ns
)
176 if (!(flags
& CLONE_NEWPID
))
177 return get_pid_ns(old_ns
);
178 if (task_active_pid_ns(current
) != old_ns
)
179 return ERR_PTR(-EINVAL
);
180 return create_pid_namespace(user_ns
, old_ns
);
183 static void free_pid_ns(struct kref
*kref
)
185 struct pid_namespace
*ns
;
187 ns
= container_of(kref
, struct pid_namespace
, kref
);
188 destroy_pid_namespace(ns
);
191 void put_pid_ns(struct pid_namespace
*ns
)
193 struct pid_namespace
*parent
;
195 while (ns
!= &init_pid_ns
) {
197 if (!kref_put(&ns
->kref
, free_pid_ns
))
202 EXPORT_SYMBOL_GPL(put_pid_ns
);
204 void zap_pid_ns_processes(struct pid_namespace
*pid_ns
)
208 struct task_struct
*task
, *me
= current
;
209 int init_pids
= thread_group_leader(me
) ? 1 : 2;
211 /* Don't allow any more processes into the pid namespace */
212 disable_pid_allocation(pid_ns
);
215 * Ignore SIGCHLD causing any terminated children to autoreap.
216 * This speeds up the namespace shutdown, plus see the comment
219 spin_lock_irq(&me
->sighand
->siglock
);
220 me
->sighand
->action
[SIGCHLD
- 1].sa
.sa_handler
= SIG_IGN
;
221 spin_unlock_irq(&me
->sighand
->siglock
);
224 * The last thread in the cgroup-init thread group is terminating.
225 * Find remaining pid_ts in the namespace, signal and wait for them
228 * Note: This signals each threads in the namespace - even those that
229 * belong to the same thread group, To avoid this, we would have
230 * to walk the entire tasklist looking a processes in this
231 * namespace, but that could be unnecessarily expensive if the
232 * pid namespace has just a few processes. Or we need to
233 * maintain a tasklist for each pid namespace.
236 read_lock(&tasklist_lock
);
237 nr
= next_pidmap(pid_ns
, 1);
241 task
= pid_task(find_vpid(nr
), PIDTYPE_PID
);
242 if (task
&& !__fatal_signal_pending(task
))
243 send_sig_info(SIGKILL
, SEND_SIG_FORCED
, task
);
247 nr
= next_pidmap(pid_ns
, nr
);
249 read_unlock(&tasklist_lock
);
252 * Reap the EXIT_ZOMBIE children we had before we ignored SIGCHLD.
253 * sys_wait4() will also block until our children traced from the
254 * parent namespace are detached and become EXIT_DEAD.
257 clear_thread_flag(TIF_SIGPENDING
);
258 rc
= sys_wait4(-1, NULL
, __WALL
, NULL
);
259 } while (rc
!= -ECHILD
);
262 * sys_wait4() above can't reap the EXIT_DEAD children but we do not
263 * really care, we could reparent them to the global init. We could
264 * exit and reap ->child_reaper even if it is not the last thread in
265 * this pid_ns, free_pid(nr_hashed == 0) calls proc_cleanup_work(),
266 * pid_ns can not go away until proc_kill_sb() drops the reference.
268 * But this ns can also have other tasks injected by setns()+fork().
269 * Again, ignoring the user visible semantics we do not really need
270 * to wait until they are all reaped, but they can be reparented to
271 * us and thus we need to ensure that pid->child_reaper stays valid
272 * until they all go away. See free_pid()->wake_up_process().
274 * We rely on ignored SIGCHLD, an injected zombie must be autoreaped
278 set_current_state(TASK_UNINTERRUPTIBLE
);
279 if (pid_ns
->nr_hashed
== init_pids
)
283 __set_current_state(TASK_RUNNING
);
286 current
->signal
->group_exit_code
= pid_ns
->reboot
;
288 acct_exit_ns(pid_ns
);
292 #ifdef CONFIG_CHECKPOINT_RESTORE
293 static int pid_ns_ctl_handler(struct ctl_table
*table
, int write
,
294 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
296 struct pid_namespace
*pid_ns
= task_active_pid_ns(current
);
297 struct ctl_table tmp
= *table
;
299 if (write
&& !ns_capable(pid_ns
->user_ns
, CAP_SYS_ADMIN
))
303 * Writing directly to ns' last_pid field is OK, since this field
304 * is volatile in a living namespace anyway and a code writing to
305 * it should synchronize its usage with external means.
308 tmp
.data
= &pid_ns
->last_pid
;
309 return proc_dointvec_minmax(&tmp
, write
, buffer
, lenp
, ppos
);
314 static struct ctl_table pid_ns_ctl_table
[] = {
316 .procname
= "ns_last_pid",
317 .maxlen
= sizeof(int),
318 .mode
= 0666, /* permissions are checked in the handler */
319 .proc_handler
= pid_ns_ctl_handler
,
325 static struct ctl_path kern_path
[] = { { .procname
= "kernel", }, { } };
326 #endif /* CONFIG_CHECKPOINT_RESTORE */
328 int reboot_pid_ns(struct pid_namespace
*pid_ns
, int cmd
)
330 if (pid_ns
== &init_pid_ns
)
334 case LINUX_REBOOT_CMD_RESTART2
:
335 case LINUX_REBOOT_CMD_RESTART
:
336 pid_ns
->reboot
= SIGHUP
;
339 case LINUX_REBOOT_CMD_POWER_OFF
:
340 case LINUX_REBOOT_CMD_HALT
:
341 pid_ns
->reboot
= SIGINT
;
347 read_lock(&tasklist_lock
);
348 force_sig(SIGKILL
, pid_ns
->child_reaper
);
349 read_unlock(&tasklist_lock
);
357 static inline struct pid_namespace
*to_pid_ns(struct ns_common
*ns
)
359 return container_of(ns
, struct pid_namespace
, ns
);
362 static struct ns_common
*pidns_get(struct task_struct
*task
)
364 struct pid_namespace
*ns
;
367 ns
= task_active_pid_ns(task
);
372 return ns
? &ns
->ns
: NULL
;
375 static void pidns_put(struct ns_common
*ns
)
377 put_pid_ns(to_pid_ns(ns
));
380 static int pidns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
382 struct pid_namespace
*active
= task_active_pid_ns(current
);
383 struct pid_namespace
*ancestor
, *new = to_pid_ns(ns
);
385 if (!ns_capable(new->user_ns
, CAP_SYS_ADMIN
) ||
386 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
390 * Only allow entering the current active pid namespace
391 * or a child of the current active pid namespace.
393 * This is required for fork to return a usable pid value and
394 * this maintains the property that processes and their
395 * children can not escape their current pid namespace.
397 if (new->level
< active
->level
)
401 while (ancestor
->level
> active
->level
)
402 ancestor
= ancestor
->parent
;
403 if (ancestor
!= active
)
406 put_pid_ns(nsproxy
->pid_ns_for_children
);
407 nsproxy
->pid_ns_for_children
= get_pid_ns(new);
411 static struct ns_common
*pidns_get_parent(struct ns_common
*ns
)
413 struct pid_namespace
*active
= task_active_pid_ns(current
);
414 struct pid_namespace
*pid_ns
, *p
;
416 /* See if the parent is in the current namespace */
417 pid_ns
= p
= to_pid_ns(ns
)->parent
;
420 return ERR_PTR(-EPERM
);
426 return &get_pid_ns(pid_ns
)->ns
;
429 static struct user_namespace
*pidns_owner(struct ns_common
*ns
)
431 return to_pid_ns(ns
)->user_ns
;
434 const struct proc_ns_operations pidns_operations
= {
436 .type
= CLONE_NEWPID
,
439 .install
= pidns_install
,
440 .owner
= pidns_owner
,
441 .get_parent
= pidns_get_parent
,
444 static __init
int pid_namespaces_init(void)
446 pid_ns_cachep
= KMEM_CACHE(pid_namespace
, SLAB_PANIC
);
448 #ifdef CONFIG_CHECKPOINT_RESTORE
449 register_sysctl_paths(kern_path
, pid_ns_ctl_table
);
454 __initcall(pid_namespaces_init
);