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_fs.h>
19 #include <linux/reboot.h>
20 #include <linux/export.h>
22 #define BITS_PER_PAGE (PAGE_SIZE*8)
27 struct kmem_cache *cachep;
28 struct list_head list;
31 static LIST_HEAD(pid_caches_lh);
32 static DEFINE_MUTEX(pid_caches_mutex);
33 static struct kmem_cache *pid_ns_cachep;
36 * creates the kmem cache to allocate pids from.
37 * @nr_ids: the number of numerical ids this pid will have to carry
40 static struct kmem_cache *create_pid_cachep(int nr_ids)
42 struct pid_cache *pcache;
43 struct kmem_cache *cachep;
45 mutex_lock(&pid_caches_mutex);
46 list_for_each_entry(pcache, &pid_caches_lh, list)
47 if (pcache->nr_ids == nr_ids)
50 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
54 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
55 cachep = kmem_cache_create(pcache->name,
56 sizeof(struct pid) + (nr_ids - 1) * sizeof(struct upid),
57 0, SLAB_HWCACHE_ALIGN, NULL);
61 pcache->nr_ids = nr_ids;
62 pcache->cachep = cachep;
63 list_add(&pcache->list, &pid_caches_lh);
65 mutex_unlock(&pid_caches_mutex);
66 return pcache->cachep;
71 mutex_unlock(&pid_caches_mutex);
75 /* MAX_PID_NS_LEVEL is needed for limiting size of 'struct pid' */
76 #define MAX_PID_NS_LEVEL 32
78 static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns,
79 struct pid_namespace *parent_pid_ns)
81 struct pid_namespace *ns;
82 unsigned int level = parent_pid_ns->level + 1;
86 if (level > MAX_PID_NS_LEVEL) {
92 ns = kmem_cache_zalloc(pid_ns_cachep, GFP_KERNEL);
96 ns->pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
97 if (!ns->pidmap[0].page)
100 ns->pid_cachep = create_pid_cachep(level + 1);
101 if (ns->pid_cachep == NULL)
104 kref_init(&ns->kref);
106 ns->parent = get_pid_ns(parent_pid_ns);
107 ns->user_ns = get_user_ns(user_ns);
109 set_bit(0, ns->pidmap[0].page);
110 atomic_set(&ns->pidmap[0].nr_free, BITS_PER_PAGE - 1);
112 for (i = 1; i < PIDMAP_ENTRIES; i++)
113 atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
115 err = pid_ns_prepare_proc(ns);
117 goto out_put_parent_pid_ns;
121 out_put_parent_pid_ns:
122 put_pid_ns(parent_pid_ns);
123 put_user_ns(user_ns);
125 kfree(ns->pidmap[0].page);
127 kmem_cache_free(pid_ns_cachep, ns);
132 static void destroy_pid_namespace(struct pid_namespace *ns)
136 for (i = 0; i < PIDMAP_ENTRIES; i++)
137 kfree(ns->pidmap[i].page);
138 put_user_ns(ns->user_ns);
139 kmem_cache_free(pid_ns_cachep, ns);
142 struct pid_namespace *copy_pid_ns(unsigned long flags,
143 struct user_namespace *user_ns, struct pid_namespace *old_ns)
145 if (!(flags & CLONE_NEWPID))
146 return get_pid_ns(old_ns);
147 if (flags & (CLONE_THREAD|CLONE_PARENT))
148 return ERR_PTR(-EINVAL);
149 return create_pid_namespace(user_ns, old_ns);
152 static void free_pid_ns(struct kref *kref)
154 struct pid_namespace *ns;
156 ns = container_of(kref, struct pid_namespace, kref);
157 destroy_pid_namespace(ns);
160 void put_pid_ns(struct pid_namespace *ns)
162 struct pid_namespace *parent;
164 while (ns != &init_pid_ns) {
166 if (!kref_put(&ns->kref, free_pid_ns))
171 EXPORT_SYMBOL_GPL(put_pid_ns);
173 void zap_pid_ns_processes(struct pid_namespace *pid_ns)
177 struct task_struct *task, *me = current;
179 /* Ignore SIGCHLD causing any terminated children to autoreap */
180 spin_lock_irq(&me->sighand->siglock);
181 me->sighand->action[SIGCHLD - 1].sa.sa_handler = SIG_IGN;
182 spin_unlock_irq(&me->sighand->siglock);
185 * The last thread in the cgroup-init thread group is terminating.
186 * Find remaining pid_ts in the namespace, signal and wait for them
189 * Note: This signals each threads in the namespace - even those that
190 * belong to the same thread group, To avoid this, we would have
191 * to walk the entire tasklist looking a processes in this
192 * namespace, but that could be unnecessarily expensive if the
193 * pid namespace has just a few processes. Or we need to
194 * maintain a tasklist for each pid namespace.
197 read_lock(&tasklist_lock);
198 nr = next_pidmap(pid_ns, 1);
202 task = pid_task(find_vpid(nr), PIDTYPE_PID);
203 if (task && !__fatal_signal_pending(task))
204 send_sig_info(SIGKILL, SEND_SIG_FORCED, task);
208 nr = next_pidmap(pid_ns, nr);
210 read_unlock(&tasklist_lock);
212 /* Firstly reap the EXIT_ZOMBIE children we may have. */
214 clear_thread_flag(TIF_SIGPENDING);
215 rc = sys_wait4(-1, NULL, __WALL, NULL);
216 } while (rc != -ECHILD);
219 * sys_wait4() above can't reap the TASK_DEAD children.
220 * Make sure they all go away, see __unhash_process().
223 bool need_wait = false;
225 read_lock(&tasklist_lock);
226 if (!list_empty(¤t->children)) {
227 __set_current_state(TASK_UNINTERRUPTIBLE);
230 read_unlock(&tasklist_lock);
238 current->signal->group_exit_code = pid_ns->reboot;
240 acct_exit_ns(pid_ns);
244 #ifdef CONFIG_CHECKPOINT_RESTORE
245 static int pid_ns_ctl_handler(struct ctl_table *table, int write,
246 void __user *buffer, size_t *lenp, loff_t *ppos)
248 struct pid_namespace *pid_ns = task_active_pid_ns(current);
249 struct ctl_table tmp = *table;
251 if (write && !ns_capable(pid_ns->user_ns, CAP_SYS_ADMIN))
255 * Writing directly to ns' last_pid field is OK, since this field
256 * is volatile in a living namespace anyway and a code writing to
257 * it should synchronize its usage with external means.
260 tmp.data = &pid_ns->last_pid;
261 return proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
266 static struct ctl_table pid_ns_ctl_table[] = {
268 .procname = "ns_last_pid",
269 .maxlen = sizeof(int),
270 .mode = 0666, /* permissions are checked in the handler */
271 .proc_handler = pid_ns_ctl_handler,
277 static struct ctl_path kern_path[] = { { .procname = "kernel", }, { } };
278 #endif /* CONFIG_CHECKPOINT_RESTORE */
280 int reboot_pid_ns(struct pid_namespace *pid_ns, int cmd)
282 if (pid_ns == &init_pid_ns)
286 case LINUX_REBOOT_CMD_RESTART2:
287 case LINUX_REBOOT_CMD_RESTART:
288 pid_ns->reboot = SIGHUP;
291 case LINUX_REBOOT_CMD_POWER_OFF:
292 case LINUX_REBOOT_CMD_HALT:
293 pid_ns->reboot = SIGINT;
299 read_lock(&tasklist_lock);
300 force_sig(SIGKILL, pid_ns->child_reaper);
301 read_unlock(&tasklist_lock);
309 static __init int pid_namespaces_init(void)
311 pid_ns_cachep = KMEM_CACHE(pid_namespace, SLAB_PANIC);
313 #ifdef CONFIG_CHECKPOINT_RESTORE
314 register_sysctl_paths(kern_path, pid_ns_ctl_table);
319 __initcall(pid_namespaces_init);